1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21#include <linux/kernel.h>
22#include <linux/syscalls.h>
23#include <linux/fs.h>
24#include <linux/mm.h>
25#include <linux/percpu.h>
26#include <linux/slab.h>
27#include <linux/capability.h>
28#include <linux/blkdev.h>
29#include <linux/file.h>
30#include <linux/quotaops.h>
31#include <linux/highmem.h>
32#include <linux/module.h>
33#include <linux/writeback.h>
34#include <linux/hash.h>
35#include <linux/suspend.h>
36#include <linux/buffer_head.h>
37#include <linux/task_io_accounting_ops.h>
38#include <linux/bio.h>
39#include <linux/notifier.h>
40#include <linux/cpu.h>
41#include <linux/bitops.h>
42#include <linux/mpage.h>
43#include <linux/bit_spinlock.h>
44
45static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
46
47#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
48
49inline void
50init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
51{
52 bh->b_end_io = handler;
53 bh->b_private = private;
54}
55
56static int sync_buffer(void *word)
57{
58 struct block_device *bd;
59 struct buffer_head *bh
60 = container_of(word, struct buffer_head, b_state);
61
62 smp_mb();
63 bd = bh->b_bdev;
64 if (bd)
65 blk_run_address_space(bd->bd_inode->i_mapping);
66 io_schedule();
67 return 0;
68}
69
70void fastcall __lock_buffer(struct buffer_head *bh)
71{
72 wait_on_bit_lock(&bh->b_state, BH_Lock, sync_buffer,
73 TASK_UNINTERRUPTIBLE);
74}
75EXPORT_SYMBOL(__lock_buffer);
76
77void fastcall unlock_buffer(struct buffer_head *bh)
78{
79 smp_mb__before_clear_bit();
80 clear_buffer_locked(bh);
81 smp_mb__after_clear_bit();
82 wake_up_bit(&bh->b_state, BH_Lock);
83}
84
85
86
87
88
89
90void __wait_on_buffer(struct buffer_head * bh)
91{
92 wait_on_bit(&bh->b_state, BH_Lock, sync_buffer, TASK_UNINTERRUPTIBLE);
93}
94
95static void
96__clear_page_buffers(struct page *page)
97{
98 ClearPagePrivate(page);
99 set_page_private(page, 0);
100 page_cache_release(page);
101}
102
103static void buffer_io_error(struct buffer_head *bh)
104{
105 char b[BDEVNAME_SIZE];
106
107 printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
108 bdevname(bh->b_bdev, b),
109 (unsigned long long)bh->b_blocknr);
110}
111
112
113
114
115
116
117
118
119
120static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
121{
122 if (uptodate) {
123 set_buffer_uptodate(bh);
124 } else {
125
126 clear_buffer_uptodate(bh);
127 }
128 unlock_buffer(bh);
129}
130
131
132
133
134
135void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
136{
137 __end_buffer_read_notouch(bh, uptodate);
138 put_bh(bh);
139}
140
141void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
142{
143 char b[BDEVNAME_SIZE];
144
145 if (uptodate) {
146 set_buffer_uptodate(bh);
147 } else {
148 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
149 buffer_io_error(bh);
150 printk(KERN_WARNING "lost page write due to "
151 "I/O error on %s\n",
152 bdevname(bh->b_bdev, b));
153 }
154 set_buffer_write_io_error(bh);
155 clear_buffer_uptodate(bh);
156 }
157 unlock_buffer(bh);
158 put_bh(bh);
159}
160
161
162
163
164
165int sync_blockdev(struct block_device *bdev)
166{
167 int ret = 0;
168
169 if (bdev)
170 ret = filemap_write_and_wait(bdev->bd_inode->i_mapping);
171 return ret;
172}
173EXPORT_SYMBOL(sync_blockdev);
174
175
176
177
178
179
180int fsync_bdev(struct block_device *bdev)
181{
182 struct super_block *sb = get_super(bdev);
183 if (sb) {
184 int res = fsync_super(sb);
185 drop_super(sb);
186 return res;
187 }
188 return sync_blockdev(bdev);
189}
190
191
192
193
194
195
196
197
198
199
200struct super_block *freeze_bdev(struct block_device *bdev)
201{
202 struct super_block *sb;
203
204 down(&bdev->bd_mount_sem);
205 sb = get_super(bdev);
206 if (sb && !(sb->s_flags & MS_RDONLY)) {
207 sb->s_frozen = SB_FREEZE_WRITE;
208 smp_wmb();
209
210 __fsync_super(sb);
211
212 sb->s_frozen = SB_FREEZE_TRANS;
213 smp_wmb();
214
215 sync_blockdev(sb->s_bdev);
216
217 if (sb->s_op->write_super_lockfs)
218 sb->s_op->write_super_lockfs(sb);
219 }
220
221 sync_blockdev(bdev);
222 return sb;
223}
224EXPORT_SYMBOL(freeze_bdev);
225
226
227
228
229
230
231
232
233void thaw_bdev(struct block_device *bdev, struct super_block *sb)
234{
235 if (sb) {
236 BUG_ON(sb->s_bdev != bdev);
237
238 if (sb->s_op->unlockfs)
239 sb->s_op->unlockfs(sb);
240 sb->s_frozen = SB_UNFROZEN;
241 smp_wmb();
242 wake_up(&sb->s_wait_unfrozen);
243 drop_super(sb);
244 }
245
246 up(&bdev->bd_mount_sem);
247}
248EXPORT_SYMBOL(thaw_bdev);
249
250
251
252
253
254
255
256
257
258
259
260
261static struct buffer_head *
262__find_get_block_slow(struct block_device *bdev, sector_t block)
263{
264 struct inode *bd_inode = bdev->bd_inode;
265 struct address_space *bd_mapping = bd_inode->i_mapping;
266 struct buffer_head *ret = NULL;
267 pgoff_t index;
268 struct buffer_head *bh;
269 struct buffer_head *head;
270 struct page *page;
271 int all_mapped = 1;
272
273 index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits);
274 page = find_get_page(bd_mapping, index);
275 if (!page)
276 goto out;
277
278 spin_lock(&bd_mapping->private_lock);
279 if (!page_has_buffers(page))
280 goto out_unlock;
281 head = page_buffers(page);
282 bh = head;
283 do {
284 if (bh->b_blocknr == block) {
285 ret = bh;
286 get_bh(bh);
287 goto out_unlock;
288 }
289 if (!buffer_mapped(bh))
290 all_mapped = 0;
291 bh = bh->b_this_page;
292 } while (bh != head);
293
294
295
296
297
298
299 if (all_mapped) {
300 printk("__find_get_block_slow() failed. "
301 "block=%llu, b_blocknr=%llu\n",
302 (unsigned long long)block,
303 (unsigned long long)bh->b_blocknr);
304 printk("b_state=0x%08lx, b_size=%zu\n",
305 bh->b_state, bh->b_size);
306 printk("device blocksize: %d\n", 1 << bd_inode->i_blkbits);
307 }
308out_unlock:
309 spin_unlock(&bd_mapping->private_lock);
310 page_cache_release(page);
311out:
312 return ret;
313}
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347void invalidate_bdev(struct block_device *bdev)
348{
349 struct address_space *mapping = bdev->bd_inode->i_mapping;
350
351 if (mapping->nrpages == 0)
352 return;
353
354 invalidate_bh_lrus();
355 invalidate_mapping_pages(mapping, 0, -1);
356}
357
358
359
360
361static void free_more_memory(void)
362{
363 struct zone **zones;
364 pg_data_t *pgdat;
365
366 wakeup_pdflush(1024);
367 yield();
368
369 for_each_online_pgdat(pgdat) {
370 zones = pgdat->node_zonelists[gfp_zone(GFP_NOFS)].zones;
371 if (*zones)
372 try_to_free_pages(zones, 0, GFP_NOFS);
373 }
374}
375
376
377
378
379
380static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
381{
382 unsigned long flags;
383 struct buffer_head *first;
384 struct buffer_head *tmp;
385 struct page *page;
386 int page_uptodate = 1;
387
388 BUG_ON(!buffer_async_read(bh));
389
390 page = bh->b_page;
391 if (uptodate) {
392 set_buffer_uptodate(bh);
393 } else {
394 clear_buffer_uptodate(bh);
395 if (printk_ratelimit())
396 buffer_io_error(bh);
397 SetPageError(page);
398 }
399
400
401
402
403
404
405 first = page_buffers(page);
406 local_irq_save(flags);
407 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
408 clear_buffer_async_read(bh);
409 unlock_buffer(bh);
410 tmp = bh;
411 do {
412 if (!buffer_uptodate(tmp))
413 page_uptodate = 0;
414 if (buffer_async_read(tmp)) {
415 BUG_ON(!buffer_locked(tmp));
416 goto still_busy;
417 }
418 tmp = tmp->b_this_page;
419 } while (tmp != bh);
420 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
421 local_irq_restore(flags);
422
423
424
425
426
427 if (page_uptodate && !PageError(page))
428 SetPageUptodate(page);
429 unlock_page(page);
430 return;
431
432still_busy:
433 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
434 local_irq_restore(flags);
435 return;
436}
437
438
439
440
441
442static void end_buffer_async_write(struct buffer_head *bh, int uptodate)
443{
444 char b[BDEVNAME_SIZE];
445 unsigned long flags;
446 struct buffer_head *first;
447 struct buffer_head *tmp;
448 struct page *page;
449
450 BUG_ON(!buffer_async_write(bh));
451
452 page = bh->b_page;
453 if (uptodate) {
454 set_buffer_uptodate(bh);
455 } else {
456 if (printk_ratelimit()) {
457 buffer_io_error(bh);
458 printk(KERN_WARNING "lost page write due to "
459 "I/O error on %s\n",
460 bdevname(bh->b_bdev, b));
461 }
462 set_bit(AS_EIO, &page->mapping->flags);
463 set_buffer_write_io_error(bh);
464 clear_buffer_uptodate(bh);
465 SetPageError(page);
466 }
467
468 first = page_buffers(page);
469 local_irq_save(flags);
470 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
471
472 clear_buffer_async_write(bh);
473 unlock_buffer(bh);
474 tmp = bh->b_this_page;
475 while (tmp != bh) {
476 if (buffer_async_write(tmp)) {
477 BUG_ON(!buffer_locked(tmp));
478 goto still_busy;
479 }
480 tmp = tmp->b_this_page;
481 }
482 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
483 local_irq_restore(flags);
484 end_page_writeback(page);
485 return;
486
487still_busy:
488 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
489 local_irq_restore(flags);
490 return;
491}
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514static void mark_buffer_async_read(struct buffer_head *bh)
515{
516 bh->b_end_io = end_buffer_async_read;
517 set_buffer_async_read(bh);
518}
519
520void mark_buffer_async_write(struct buffer_head *bh)
521{
522 bh->b_end_io = end_buffer_async_write;
523 set_buffer_async_write(bh);
524}
525EXPORT_SYMBOL(mark_buffer_async_write);
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580static inline void __remove_assoc_queue(struct buffer_head *bh)
581{
582 list_del_init(&bh->b_assoc_buffers);
583 WARN_ON(!bh->b_assoc_map);
584 if (buffer_write_io_error(bh))
585 set_bit(AS_EIO, &bh->b_assoc_map->flags);
586 bh->b_assoc_map = NULL;
587}
588
589int inode_has_buffers(struct inode *inode)
590{
591 return !list_empty(&inode->i_data.private_list);
592}
593
594
595
596
597
598
599
600
601
602
603
604static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
605{
606 struct buffer_head *bh;
607 struct list_head *p;
608 int err = 0;
609
610 spin_lock(lock);
611repeat:
612 list_for_each_prev(p, list) {
613 bh = BH_ENTRY(p);
614 if (buffer_locked(bh)) {
615 get_bh(bh);
616 spin_unlock(lock);
617 wait_on_buffer(bh);
618 if (!buffer_uptodate(bh))
619 err = -EIO;
620 brelse(bh);
621 spin_lock(lock);
622 goto repeat;
623 }
624 }
625 spin_unlock(lock);
626 return err;
627}
628
629
630
631
632
633
634
635
636
637
638
639
640
641int sync_mapping_buffers(struct address_space *mapping)
642{
643 struct address_space *buffer_mapping = mapping->assoc_mapping;
644
645 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
646 return 0;
647
648 return fsync_buffers_list(&buffer_mapping->private_lock,
649 &mapping->private_list);
650}
651EXPORT_SYMBOL(sync_mapping_buffers);
652
653
654
655
656
657
658
659void write_boundary_block(struct block_device *bdev,
660 sector_t bblock, unsigned blocksize)
661{
662 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
663 if (bh) {
664 if (buffer_dirty(bh))
665 ll_rw_block(WRITE, 1, &bh);
666 put_bh(bh);
667 }
668}
669
670void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
671{
672 struct address_space *mapping = inode->i_mapping;
673 struct address_space *buffer_mapping = bh->b_page->mapping;
674
675 mark_buffer_dirty(bh);
676 if (!mapping->assoc_mapping) {
677 mapping->assoc_mapping = buffer_mapping;
678 } else {
679 BUG_ON(mapping->assoc_mapping != buffer_mapping);
680 }
681 if (list_empty(&bh->b_assoc_buffers)) {
682 spin_lock(&buffer_mapping->private_lock);
683 list_move_tail(&bh->b_assoc_buffers,
684 &mapping->private_list);
685 bh->b_assoc_map = mapping;
686 spin_unlock(&buffer_mapping->private_lock);
687 }
688}
689EXPORT_SYMBOL(mark_buffer_dirty_inode);
690
691
692
693
694
695
696
697
698static int __set_page_dirty(struct page *page,
699 struct address_space *mapping, int warn)
700{
701 if (unlikely(!mapping))
702 return !TestSetPageDirty(page);
703
704 if (TestSetPageDirty(page))
705 return 0;
706
707 write_lock_irq(&mapping->tree_lock);
708 if (page->mapping) {
709 WARN_ON_ONCE(warn && !PageUptodate(page));
710
711 if (mapping_cap_account_dirty(mapping)) {
712 __inc_zone_page_state(page, NR_FILE_DIRTY);
713 __inc_bdi_stat(mapping->backing_dev_info,
714 BDI_RECLAIMABLE);
715 task_io_account_write(PAGE_CACHE_SIZE);
716 }
717 radix_tree_tag_set(&mapping->page_tree,
718 page_index(page), PAGECACHE_TAG_DIRTY);
719 }
720 write_unlock_irq(&mapping->tree_lock);
721 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
722
723 return 1;
724}
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751int __set_page_dirty_buffers(struct page *page)
752{
753 struct address_space *mapping = page_mapping(page);
754
755 if (unlikely(!mapping))
756 return !TestSetPageDirty(page);
757
758 spin_lock(&mapping->private_lock);
759 if (page_has_buffers(page)) {
760 struct buffer_head *head = page_buffers(page);
761 struct buffer_head *bh = head;
762
763 do {
764 set_buffer_dirty(bh);
765 bh = bh->b_this_page;
766 } while (bh != head);
767 }
768 spin_unlock(&mapping->private_lock);
769
770 return __set_page_dirty(page, mapping, 1);
771}
772EXPORT_SYMBOL(__set_page_dirty_buffers);
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
794{
795 struct buffer_head *bh;
796 struct list_head tmp;
797 int err = 0, err2;
798
799 INIT_LIST_HEAD(&tmp);
800
801 spin_lock(lock);
802 while (!list_empty(list)) {
803 bh = BH_ENTRY(list->next);
804 __remove_assoc_queue(bh);
805 if (buffer_dirty(bh) || buffer_locked(bh)) {
806 list_add(&bh->b_assoc_buffers, &tmp);
807 if (buffer_dirty(bh)) {
808 get_bh(bh);
809 spin_unlock(lock);
810
811
812
813
814
815
816 ll_rw_block(SWRITE, 1, &bh);
817 brelse(bh);
818 spin_lock(lock);
819 }
820 }
821 }
822
823 while (!list_empty(&tmp)) {
824 bh = BH_ENTRY(tmp.prev);
825 list_del_init(&bh->b_assoc_buffers);
826 get_bh(bh);
827 spin_unlock(lock);
828 wait_on_buffer(bh);
829 if (!buffer_uptodate(bh))
830 err = -EIO;
831 brelse(bh);
832 spin_lock(lock);
833 }
834
835 spin_unlock(lock);
836 err2 = osync_buffers_list(lock, list);
837 if (err)
838 return err;
839 else
840 return err2;
841}
842
843
844
845
846
847
848
849
850
851
852void invalidate_inode_buffers(struct inode *inode)
853{
854 if (inode_has_buffers(inode)) {
855 struct address_space *mapping = &inode->i_data;
856 struct list_head *list = &mapping->private_list;
857 struct address_space *buffer_mapping = mapping->assoc_mapping;
858
859 spin_lock(&buffer_mapping->private_lock);
860 while (!list_empty(list))
861 __remove_assoc_queue(BH_ENTRY(list->next));
862 spin_unlock(&buffer_mapping->private_lock);
863 }
864}
865
866
867
868
869
870
871
872int remove_inode_buffers(struct inode *inode)
873{
874 int ret = 1;
875
876 if (inode_has_buffers(inode)) {
877 struct address_space *mapping = &inode->i_data;
878 struct list_head *list = &mapping->private_list;
879 struct address_space *buffer_mapping = mapping->assoc_mapping;
880
881 spin_lock(&buffer_mapping->private_lock);
882 while (!list_empty(list)) {
883 struct buffer_head *bh = BH_ENTRY(list->next);
884 if (buffer_dirty(bh)) {
885 ret = 0;
886 break;
887 }
888 __remove_assoc_queue(bh);
889 }
890 spin_unlock(&buffer_mapping->private_lock);
891 }
892 return ret;
893}
894
895
896
897
898
899
900
901
902
903
904struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
905 int retry)
906{
907 struct buffer_head *bh, *head;
908 long offset;
909
910try_again:
911 head = NULL;
912 offset = PAGE_SIZE;
913 while ((offset -= size) >= 0) {
914 bh = alloc_buffer_head(GFP_NOFS);
915 if (!bh)
916 goto no_grow;
917
918 bh->b_bdev = NULL;
919 bh->b_this_page = head;
920 bh->b_blocknr = -1;
921 head = bh;
922
923 bh->b_state = 0;
924 atomic_set(&bh->b_count, 0);
925 bh->b_private = NULL;
926 bh->b_size = size;
927
928
929 set_bh_page(bh, page, offset);
930
931 init_buffer(bh, NULL, NULL);
932 }
933 return head;
934
935
936
937no_grow:
938 if (head) {
939 do {
940 bh = head;
941 head = head->b_this_page;
942 free_buffer_head(bh);
943 } while (head);
944 }
945
946
947
948
949
950
951
952 if (!retry)
953 return NULL;
954
955
956
957
958
959
960
961 free_more_memory();
962 goto try_again;
963}
964EXPORT_SYMBOL_GPL(alloc_page_buffers);
965
966static inline void
967link_dev_buffers(struct page *page, struct buffer_head *head)
968{
969 struct buffer_head *bh, *tail;
970
971 bh = head;
972 do {
973 tail = bh;
974 bh = bh->b_this_page;
975 } while (bh);
976 tail->b_this_page = head;
977 attach_page_buffers(page, head);
978}
979
980
981
982
983static void
984init_page_buffers(struct page *page, struct block_device *bdev,
985 sector_t block, int size)
986{
987 struct buffer_head *head = page_buffers(page);
988 struct buffer_head *bh = head;
989 int uptodate = PageUptodate(page);
990
991 do {
992 if (!buffer_mapped(bh)) {
993 init_buffer(bh, NULL, NULL);
994 bh->b_bdev = bdev;
995 bh->b_blocknr = block;
996 if (uptodate)
997 set_buffer_uptodate(bh);
998 set_buffer_mapped(bh);
999 }
1000 block++;
1001 bh = bh->b_this_page;
1002 } while (bh != head);
1003}
1004
1005
1006
1007
1008
1009
1010static struct page *
1011grow_dev_page(struct block_device *bdev, sector_t block,
1012 pgoff_t index, int size)
1013{
1014 struct inode *inode = bdev->bd_inode;
1015 struct page *page;
1016 struct buffer_head *bh;
1017
1018 page = find_or_create_page(inode->i_mapping, index,
1019 (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS)|__GFP_MOVABLE);
1020 if (!page)
1021 return NULL;
1022
1023 BUG_ON(!PageLocked(page));
1024
1025 if (page_has_buffers(page)) {
1026 bh = page_buffers(page);
1027 if (bh->b_size == size) {
1028 init_page_buffers(page, bdev, block, size);
1029 return page;
1030 }
1031 if (!try_to_free_buffers(page))
1032 goto failed;
1033 }
1034
1035
1036
1037
1038 bh = alloc_page_buffers(page, size, 0);
1039 if (!bh)
1040 goto failed;
1041
1042
1043
1044
1045
1046
1047 spin_lock(&inode->i_mapping->private_lock);
1048 link_dev_buffers(page, bh);
1049 init_page_buffers(page, bdev, block, size);
1050 spin_unlock(&inode->i_mapping->private_lock);
1051 return page;
1052
1053failed:
1054 BUG();
1055 unlock_page(page);
1056 page_cache_release(page);
1057 return NULL;
1058}
1059
1060
1061
1062
1063
1064static int
1065grow_buffers(struct block_device *bdev, sector_t block, int size)
1066{
1067 struct page *page;
1068 pgoff_t index;
1069 int sizebits;
1070
1071 sizebits = -1;
1072 do {
1073 sizebits++;
1074 } while ((size << sizebits) < PAGE_SIZE);
1075
1076 index = block >> sizebits;
1077
1078
1079
1080
1081
1082 if (unlikely(index != block >> sizebits)) {
1083 char b[BDEVNAME_SIZE];
1084
1085 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1086 "device %s\n",
1087 __FUNCTION__, (unsigned long long)block,
1088 bdevname(bdev, b));
1089 return -EIO;
1090 }
1091 block = index << sizebits;
1092
1093 page = grow_dev_page(bdev, block, index, size);
1094 if (!page)
1095 return 0;
1096 unlock_page(page);
1097 page_cache_release(page);
1098 return 1;
1099}
1100
1101static struct buffer_head *
1102__getblk_slow(struct block_device *bdev, sector_t block, int size)
1103{
1104
1105 if (unlikely(size & (bdev_hardsect_size(bdev)-1) ||
1106 (size < 512 || size > PAGE_SIZE))) {
1107 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1108 size);
1109 printk(KERN_ERR "hardsect size: %d\n",
1110 bdev_hardsect_size(bdev));
1111
1112 dump_stack();
1113 return NULL;
1114 }
1115
1116 for (;;) {
1117 struct buffer_head * bh;
1118 int ret;
1119
1120 bh = __find_get_block(bdev, block, size);
1121 if (bh)
1122 return bh;
1123
1124 ret = grow_buffers(bdev, block, size);
1125 if (ret < 0)
1126 return NULL;
1127 if (ret == 0)
1128 free_more_memory();
1129 }
1130}
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167void fastcall mark_buffer_dirty(struct buffer_head *bh)
1168{
1169 WARN_ON_ONCE(!buffer_uptodate(bh));
1170 if (!buffer_dirty(bh) && !test_set_buffer_dirty(bh))
1171 __set_page_dirty(bh->b_page, page_mapping(bh->b_page), 0);
1172}
1173
1174
1175
1176
1177
1178
1179
1180
1181void __brelse(struct buffer_head * buf)
1182{
1183 if (atomic_read(&buf->b_count)) {
1184 put_bh(buf);
1185 return;
1186 }
1187 printk(KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1188 WARN_ON(1);
1189}
1190
1191
1192
1193
1194
1195void __bforget(struct buffer_head *bh)
1196{
1197 clear_buffer_dirty(bh);
1198 if (!list_empty(&bh->b_assoc_buffers)) {
1199 struct address_space *buffer_mapping = bh->b_page->mapping;
1200
1201 spin_lock(&buffer_mapping->private_lock);
1202 list_del_init(&bh->b_assoc_buffers);
1203 bh->b_assoc_map = NULL;
1204 spin_unlock(&buffer_mapping->private_lock);
1205 }
1206 __brelse(bh);
1207}
1208
1209static struct buffer_head *__bread_slow(struct buffer_head *bh)
1210{
1211 lock_buffer(bh);
1212 if (buffer_uptodate(bh)) {
1213 unlock_buffer(bh);
1214 return bh;
1215 } else {
1216 get_bh(bh);
1217 bh->b_end_io = end_buffer_read_sync;
1218 submit_bh(READ, bh);
1219 wait_on_buffer(bh);
1220 if (buffer_uptodate(bh))
1221 return bh;
1222 }
1223 brelse(bh);
1224 return NULL;
1225}
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241#define BH_LRU_SIZE 8
1242
1243struct bh_lru {
1244 struct buffer_head *bhs[BH_LRU_SIZE];
1245};
1246
1247static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1248
1249#ifdef CONFIG_SMP
1250#define bh_lru_lock() local_irq_disable()
1251#define bh_lru_unlock() local_irq_enable()
1252#else
1253#define bh_lru_lock() preempt_disable()
1254#define bh_lru_unlock() preempt_enable()
1255#endif
1256
1257static inline void check_irqs_on(void)
1258{
1259#ifdef irqs_disabled
1260 BUG_ON(irqs_disabled());
1261#endif
1262}
1263
1264
1265
1266
1267static void bh_lru_install(struct buffer_head *bh)
1268{
1269 struct buffer_head *evictee = NULL;
1270 struct bh_lru *lru;
1271
1272 check_irqs_on();
1273 bh_lru_lock();
1274 lru = &__get_cpu_var(bh_lrus);
1275 if (lru->bhs[0] != bh) {
1276 struct buffer_head *bhs[BH_LRU_SIZE];
1277 int in;
1278 int out = 0;
1279
1280 get_bh(bh);
1281 bhs[out++] = bh;
1282 for (in = 0; in < BH_LRU_SIZE; in++) {
1283 struct buffer_head *bh2 = lru->bhs[in];
1284
1285 if (bh2 == bh) {
1286 __brelse(bh2);
1287 } else {
1288 if (out >= BH_LRU_SIZE) {
1289 BUG_ON(evictee != NULL);
1290 evictee = bh2;
1291 } else {
1292 bhs[out++] = bh2;
1293 }
1294 }
1295 }
1296 while (out < BH_LRU_SIZE)
1297 bhs[out++] = NULL;
1298 memcpy(lru->bhs, bhs, sizeof(bhs));
1299 }
1300 bh_lru_unlock();
1301
1302 if (evictee)
1303 __brelse(evictee);
1304}
1305
1306
1307
1308
1309static struct buffer_head *
1310lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1311{
1312 struct buffer_head *ret = NULL;
1313 struct bh_lru *lru;
1314 unsigned int i;
1315
1316 check_irqs_on();
1317 bh_lru_lock();
1318 lru = &__get_cpu_var(bh_lrus);
1319 for (i = 0; i < BH_LRU_SIZE; i++) {
1320 struct buffer_head *bh = lru->bhs[i];
1321
1322 if (bh && bh->b_bdev == bdev &&
1323 bh->b_blocknr == block && bh->b_size == size) {
1324 if (i) {
1325 while (i) {
1326 lru->bhs[i] = lru->bhs[i - 1];
1327 i--;
1328 }
1329 lru->bhs[0] = bh;
1330 }
1331 get_bh(bh);
1332 ret = bh;
1333 break;
1334 }
1335 }
1336 bh_lru_unlock();
1337 return ret;
1338}
1339
1340
1341
1342
1343
1344
1345struct buffer_head *
1346__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1347{
1348 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1349
1350 if (bh == NULL) {
1351 bh = __find_get_block_slow(bdev, block);
1352 if (bh)
1353 bh_lru_install(bh);
1354 }
1355 if (bh)
1356 touch_buffer(bh);
1357 return bh;
1358}
1359EXPORT_SYMBOL(__find_get_block);
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373struct buffer_head *
1374__getblk(struct block_device *bdev, sector_t block, unsigned size)
1375{
1376 struct buffer_head *bh = __find_get_block(bdev, block, size);
1377
1378 might_sleep();
1379 if (bh == NULL)
1380 bh = __getblk_slow(bdev, block, size);
1381 return bh;
1382}
1383EXPORT_SYMBOL(__getblk);
1384
1385
1386
1387
1388void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1389{
1390 struct buffer_head *bh = __getblk(bdev, block, size);
1391 if (likely(bh)) {
1392 ll_rw_block(READA, 1, &bh);
1393 brelse(bh);
1394 }
1395}
1396EXPORT_SYMBOL(__breadahead);
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407struct buffer_head *
1408__bread(struct block_device *bdev, sector_t block, unsigned size)
1409{
1410 struct buffer_head *bh = __getblk(bdev, block, size);
1411
1412 if (likely(bh) && !buffer_uptodate(bh))
1413 bh = __bread_slow(bh);
1414 return bh;
1415}
1416EXPORT_SYMBOL(__bread);
1417
1418
1419
1420
1421
1422
1423static void invalidate_bh_lru(void *arg)
1424{
1425 struct bh_lru *b = &get_cpu_var(bh_lrus);
1426 int i;
1427
1428 for (i = 0; i < BH_LRU_SIZE; i++) {
1429 brelse(b->bhs[i]);
1430 b->bhs[i] = NULL;
1431 }
1432 put_cpu_var(bh_lrus);
1433}
1434
1435void invalidate_bh_lrus(void)
1436{
1437 on_each_cpu(invalidate_bh_lru, NULL, 1, 1);
1438}
1439
1440void set_bh_page(struct buffer_head *bh,
1441 struct page *page, unsigned long offset)
1442{
1443 bh->b_page = page;
1444 BUG_ON(offset >= PAGE_SIZE);
1445 if (PageHighMem(page))
1446
1447
1448
1449 bh->b_data = (char *)(0 + offset);
1450 else
1451 bh->b_data = page_address(page) + offset;
1452}
1453EXPORT_SYMBOL(set_bh_page);
1454
1455
1456
1457
1458static void discard_buffer(struct buffer_head * bh)
1459{
1460 lock_buffer(bh);
1461 clear_buffer_dirty(bh);
1462 bh->b_bdev = NULL;
1463 clear_buffer_mapped(bh);
1464 clear_buffer_req(bh);
1465 clear_buffer_new(bh);
1466 clear_buffer_delay(bh);
1467 clear_buffer_unwritten(bh);
1468 unlock_buffer(bh);
1469}
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486void block_invalidatepage(struct page *page, unsigned long offset)
1487{
1488 struct buffer_head *head, *bh, *next;
1489 unsigned int curr_off = 0;
1490
1491 BUG_ON(!PageLocked(page));
1492 if (!page_has_buffers(page))
1493 goto out;
1494
1495 head = page_buffers(page);
1496 bh = head;
1497 do {
1498 unsigned int next_off = curr_off + bh->b_size;
1499 next = bh->b_this_page;
1500
1501
1502
1503
1504 if (offset <= curr_off)
1505 discard_buffer(bh);
1506 curr_off = next_off;
1507 bh = next;
1508 } while (bh != head);
1509
1510
1511
1512
1513
1514
1515 if (offset == 0)
1516 try_to_release_page(page, 0);
1517out:
1518 return;
1519}
1520EXPORT_SYMBOL(block_invalidatepage);
1521
1522
1523
1524
1525
1526
1527void create_empty_buffers(struct page *page,
1528 unsigned long blocksize, unsigned long b_state)
1529{
1530 struct buffer_head *bh, *head, *tail;
1531
1532 head = alloc_page_buffers(page, blocksize, 1);
1533 bh = head;
1534 do {
1535 bh->b_state |= b_state;
1536 tail = bh;
1537 bh = bh->b_this_page;
1538 } while (bh);
1539 tail->b_this_page = head;
1540
1541 spin_lock(&page->mapping->private_lock);
1542 if (PageUptodate(page) || PageDirty(page)) {
1543 bh = head;
1544 do {
1545 if (PageDirty(page))
1546 set_buffer_dirty(bh);
1547 if (PageUptodate(page))
1548 set_buffer_uptodate(bh);
1549 bh = bh->b_this_page;
1550 } while (bh != head);
1551 }
1552 attach_page_buffers(page, head);
1553 spin_unlock(&page->mapping->private_lock);
1554}
1555EXPORT_SYMBOL(create_empty_buffers);
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1574{
1575 struct buffer_head *old_bh;
1576
1577 might_sleep();
1578
1579 old_bh = __find_get_block_slow(bdev, block);
1580 if (old_bh) {
1581 clear_buffer_dirty(old_bh);
1582 wait_on_buffer(old_bh);
1583 clear_buffer_req(old_bh);
1584 __brelse(old_bh);
1585 }
1586}
1587EXPORT_SYMBOL(unmap_underlying_metadata);
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614static int __block_write_full_page(struct inode *inode, struct page *page,
1615 get_block_t *get_block, struct writeback_control *wbc)
1616{
1617 int err;
1618 sector_t block;
1619 sector_t last_block;
1620 struct buffer_head *bh, *head;
1621 const unsigned blocksize = 1 << inode->i_blkbits;
1622 int nr_underway = 0;
1623
1624 BUG_ON(!PageLocked(page));
1625
1626 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1627
1628 if (!page_has_buffers(page)) {
1629 create_empty_buffers(page, blocksize,
1630 (1 << BH_Dirty)|(1 << BH_Uptodate));
1631 }
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1644 head = page_buffers(page);
1645 bh = head;
1646
1647
1648
1649
1650
1651 do {
1652 if (block > last_block) {
1653
1654
1655
1656
1657
1658
1659
1660
1661 clear_buffer_dirty(bh);
1662 set_buffer_uptodate(bh);
1663 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1664 WARN_ON(bh->b_size != blocksize);
1665 err = get_block(inode, block, bh, 1);
1666 if (err)
1667 goto recover;
1668 if (buffer_new(bh)) {
1669
1670 clear_buffer_new(bh);
1671 unmap_underlying_metadata(bh->b_bdev,
1672 bh->b_blocknr);
1673 }
1674 }
1675 bh = bh->b_this_page;
1676 block++;
1677 } while (bh != head);
1678
1679 do {
1680 if (!buffer_mapped(bh))
1681 continue;
1682
1683
1684
1685
1686
1687
1688
1689 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1690 lock_buffer(bh);
1691 } else if (test_set_buffer_locked(bh)) {
1692 redirty_page_for_writepage(wbc, page);
1693 continue;
1694 }
1695 if (test_clear_buffer_dirty(bh)) {
1696 mark_buffer_async_write(bh);
1697 } else {
1698 unlock_buffer(bh);
1699 }
1700 } while ((bh = bh->b_this_page) != head);
1701
1702
1703
1704
1705
1706 BUG_ON(PageWriteback(page));
1707 set_page_writeback(page);
1708
1709 do {
1710 struct buffer_head *next = bh->b_this_page;
1711 if (buffer_async_write(bh)) {
1712 submit_bh(WRITE, bh);
1713 nr_underway++;
1714 }
1715 bh = next;
1716 } while (bh != head);
1717 unlock_page(page);
1718
1719 err = 0;
1720done:
1721 if (nr_underway == 0) {
1722
1723
1724
1725
1726
1727 end_page_writeback(page);
1728
1729
1730
1731
1732
1733 }
1734 return err;
1735
1736recover:
1737
1738
1739
1740
1741
1742
1743 bh = head;
1744
1745 do {
1746 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1747 lock_buffer(bh);
1748 mark_buffer_async_write(bh);
1749 } else {
1750
1751
1752
1753
1754 clear_buffer_dirty(bh);
1755 }
1756 } while ((bh = bh->b_this_page) != head);
1757 SetPageError(page);
1758 BUG_ON(PageWriteback(page));
1759 mapping_set_error(page->mapping, err);
1760 set_page_writeback(page);
1761 do {
1762 struct buffer_head *next = bh->b_this_page;
1763 if (buffer_async_write(bh)) {
1764 clear_buffer_dirty(bh);
1765 submit_bh(WRITE, bh);
1766 nr_underway++;
1767 }
1768 bh = next;
1769 } while (bh != head);
1770 unlock_page(page);
1771 goto done;
1772}
1773
1774
1775
1776
1777
1778
1779void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1780{
1781 unsigned int block_start, block_end;
1782 struct buffer_head *head, *bh;
1783
1784 BUG_ON(!PageLocked(page));
1785 if (!page_has_buffers(page))
1786 return;
1787
1788 bh = head = page_buffers(page);
1789 block_start = 0;
1790 do {
1791 block_end = block_start + bh->b_size;
1792
1793 if (buffer_new(bh)) {
1794 if (block_end > from && block_start < to) {
1795 if (!PageUptodate(page)) {
1796 unsigned start, size;
1797
1798 start = max(from, block_start);
1799 size = min(to, block_end) - start;
1800
1801 zero_user_page(page, start, size, KM_USER0);
1802 set_buffer_uptodate(bh);
1803 }
1804
1805 clear_buffer_new(bh);
1806 mark_buffer_dirty(bh);
1807 }
1808 }
1809
1810 block_start = block_end;
1811 bh = bh->b_this_page;
1812 } while (bh != head);
1813}
1814EXPORT_SYMBOL(page_zero_new_buffers);
1815
1816static int __block_prepare_write(struct inode *inode, struct page *page,
1817 unsigned from, unsigned to, get_block_t *get_block)
1818{
1819 unsigned block_start, block_end;
1820 sector_t block;
1821 int err = 0;
1822 unsigned blocksize, bbits;
1823 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1824
1825 BUG_ON(!PageLocked(page));
1826 BUG_ON(from > PAGE_CACHE_SIZE);
1827 BUG_ON(to > PAGE_CACHE_SIZE);
1828 BUG_ON(from > to);
1829
1830 blocksize = 1 << inode->i_blkbits;
1831 if (!page_has_buffers(page))
1832 create_empty_buffers(page, blocksize, 0);
1833 head = page_buffers(page);
1834
1835 bbits = inode->i_blkbits;
1836 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1837
1838 for(bh = head, block_start = 0; bh != head || !block_start;
1839 block++, block_start=block_end, bh = bh->b_this_page) {
1840 block_end = block_start + blocksize;
1841 if (block_end <= from || block_start >= to) {
1842 if (PageUptodate(page)) {
1843 if (!buffer_uptodate(bh))
1844 set_buffer_uptodate(bh);
1845 }
1846 continue;
1847 }
1848 if (buffer_new(bh))
1849 clear_buffer_new(bh);
1850 if (!buffer_mapped(bh)) {
1851 WARN_ON(bh->b_size != blocksize);
1852 err = get_block(inode, block, bh, 1);
1853 if (err)
1854 break;
1855 if (buffer_new(bh)) {
1856 unmap_underlying_metadata(bh->b_bdev,
1857 bh->b_blocknr);
1858 if (PageUptodate(page)) {
1859 clear_buffer_new(bh);
1860 set_buffer_uptodate(bh);
1861 mark_buffer_dirty(bh);
1862 continue;
1863 }
1864 if (block_end > to || block_start < from) {
1865 void *kaddr;
1866
1867 kaddr = kmap_atomic(page, KM_USER0);
1868 if (block_end > to)
1869 memset(kaddr+to, 0,
1870 block_end-to);
1871 if (block_start < from)
1872 memset(kaddr+block_start,
1873 0, from-block_start);
1874 flush_dcache_page(page);
1875 kunmap_atomic(kaddr, KM_USER0);
1876 }
1877 continue;
1878 }
1879 }
1880 if (PageUptodate(page)) {
1881 if (!buffer_uptodate(bh))
1882 set_buffer_uptodate(bh);
1883 continue;
1884 }
1885 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1886 !buffer_unwritten(bh) &&
1887 (block_start < from || block_end > to)) {
1888 ll_rw_block(READ, 1, &bh);
1889 *wait_bh++=bh;
1890 }
1891 }
1892
1893
1894
1895 while(wait_bh > wait) {
1896 wait_on_buffer(*--wait_bh);
1897 if (!buffer_uptodate(*wait_bh))
1898 err = -EIO;
1899 }
1900 if (unlikely(err))
1901 page_zero_new_buffers(page, from, to);
1902 return err;
1903}
1904
1905static int __block_commit_write(struct inode *inode, struct page *page,
1906 unsigned from, unsigned to)
1907{
1908 unsigned block_start, block_end;
1909 int partial = 0;
1910 unsigned blocksize;
1911 struct buffer_head *bh, *head;
1912
1913 blocksize = 1 << inode->i_blkbits;
1914
1915 for(bh = head = page_buffers(page), block_start = 0;
1916 bh != head || !block_start;
1917 block_start=block_end, bh = bh->b_this_page) {
1918 block_end = block_start + blocksize;
1919 if (block_end <= from || block_start >= to) {
1920 if (!buffer_uptodate(bh))
1921 partial = 1;
1922 } else {
1923 set_buffer_uptodate(bh);
1924 mark_buffer_dirty(bh);
1925 }
1926 clear_buffer_new(bh);
1927 }
1928
1929
1930
1931
1932
1933
1934
1935 if (!partial)
1936 SetPageUptodate(page);
1937 return 0;
1938}
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948int block_write_begin(struct file *file, struct address_space *mapping,
1949 loff_t pos, unsigned len, unsigned flags,
1950 struct page **pagep, void **fsdata,
1951 get_block_t *get_block)
1952{
1953 struct inode *inode = mapping->host;
1954 int status = 0;
1955 struct page *page;
1956 pgoff_t index;
1957 unsigned start, end;
1958 int ownpage = 0;
1959
1960 index = pos >> PAGE_CACHE_SHIFT;
1961 start = pos & (PAGE_CACHE_SIZE - 1);
1962 end = start + len;
1963
1964 page = *pagep;
1965 if (page == NULL) {
1966 ownpage = 1;
1967 page = __grab_cache_page(mapping, index);
1968 if (!page) {
1969 status = -ENOMEM;
1970 goto out;
1971 }
1972 *pagep = page;
1973 } else
1974 BUG_ON(!PageLocked(page));
1975
1976 status = __block_prepare_write(inode, page, start, end, get_block);
1977 if (unlikely(status)) {
1978 ClearPageUptodate(page);
1979
1980 if (ownpage) {
1981 unlock_page(page);
1982 page_cache_release(page);
1983 *pagep = NULL;
1984
1985
1986
1987
1988
1989
1990 if (pos + len > inode->i_size)
1991 vmtruncate(inode, inode->i_size);
1992 }
1993 goto out;
1994 }
1995
1996out:
1997 return status;
1998}
1999EXPORT_SYMBOL(block_write_begin);
2000
2001int block_write_end(struct file *file, struct address_space *mapping,
2002 loff_t pos, unsigned len, unsigned copied,
2003 struct page *page, void *fsdata)
2004{
2005 struct inode *inode = mapping->host;
2006 unsigned start;
2007
2008 start = pos & (PAGE_CACHE_SIZE - 1);
2009
2010 if (unlikely(copied < len)) {
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023 if (!PageUptodate(page))
2024 copied = 0;
2025
2026 page_zero_new_buffers(page, start+copied, start+len);
2027 }
2028 flush_dcache_page(page);
2029
2030
2031 __block_commit_write(inode, page, start, start+copied);
2032
2033 return copied;
2034}
2035EXPORT_SYMBOL(block_write_end);
2036
2037int generic_write_end(struct file *file, struct address_space *mapping,
2038 loff_t pos, unsigned len, unsigned copied,
2039 struct page *page, void *fsdata)
2040{
2041 struct inode *inode = mapping->host;
2042
2043 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2044
2045
2046
2047
2048
2049
2050
2051
2052 if (pos+copied > inode->i_size) {
2053 i_size_write(inode, pos+copied);
2054 mark_inode_dirty(inode);
2055 }
2056
2057 unlock_page(page);
2058 page_cache_release(page);
2059
2060 return copied;
2061}
2062EXPORT_SYMBOL(generic_write_end);
2063
2064
2065
2066
2067
2068
2069
2070
2071int block_read_full_page(struct page *page, get_block_t *get_block)
2072{
2073 struct inode *inode = page->mapping->host;
2074 sector_t iblock, lblock;
2075 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2076 unsigned int blocksize;
2077 int nr, i;
2078 int fully_mapped = 1;
2079
2080 BUG_ON(!PageLocked(page));
2081 blocksize = 1 << inode->i_blkbits;
2082 if (!page_has_buffers(page))
2083 create_empty_buffers(page, blocksize, 0);
2084 head = page_buffers(page);
2085
2086 iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2087 lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits;
2088 bh = head;
2089 nr = 0;
2090 i = 0;
2091
2092 do {
2093 if (buffer_uptodate(bh))
2094 continue;
2095
2096 if (!buffer_mapped(bh)) {
2097 int err = 0;
2098
2099 fully_mapped = 0;
2100 if (iblock < lblock) {
2101 WARN_ON(bh->b_size != blocksize);
2102 err = get_block(inode, iblock, bh, 0);
2103 if (err)
2104 SetPageError(page);
2105 }
2106 if (!buffer_mapped(bh)) {
2107 zero_user_page(page, i * blocksize, blocksize,
2108 KM_USER0);
2109 if (!err)
2110 set_buffer_uptodate(bh);
2111 continue;
2112 }
2113
2114
2115
2116
2117 if (buffer_uptodate(bh))
2118 continue;
2119 }
2120 arr[nr++] = bh;
2121 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2122
2123 if (fully_mapped)
2124 SetPageMappedToDisk(page);
2125
2126 if (!nr) {
2127
2128
2129
2130
2131 if (!PageError(page))
2132 SetPageUptodate(page);
2133 unlock_page(page);
2134 return 0;
2135 }
2136
2137
2138 for (i = 0; i < nr; i++) {
2139 bh = arr[i];
2140 lock_buffer(bh);
2141 mark_buffer_async_read(bh);
2142 }
2143
2144
2145
2146
2147
2148
2149 for (i = 0; i < nr; i++) {
2150 bh = arr[i];
2151 if (buffer_uptodate(bh))
2152 end_buffer_async_read(bh, 1);
2153 else
2154 submit_bh(READ, bh);
2155 }
2156 return 0;
2157}
2158
2159
2160
2161
2162
2163int generic_cont_expand_simple(struct inode *inode, loff_t size)
2164{
2165 struct address_space *mapping = inode->i_mapping;
2166 struct page *page;
2167 void *fsdata;
2168 unsigned long limit;
2169 int err;
2170
2171 err = -EFBIG;
2172 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2173 if (limit != RLIM_INFINITY && size > (loff_t)limit) {
2174 send_sig(SIGXFSZ, current, 0);
2175 goto out;
2176 }
2177 if (size > inode->i_sb->s_maxbytes)
2178 goto out;
2179
2180 err = pagecache_write_begin(NULL, mapping, size, 0,
2181 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2182 &page, &fsdata);
2183 if (err)
2184 goto out;
2185
2186 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2187 BUG_ON(err > 0);
2188
2189out:
2190 return err;
2191}
2192
2193int cont_expand_zero(struct file *file, struct address_space *mapping,
2194 loff_t pos, loff_t *bytes)
2195{
2196 struct inode *inode = mapping->host;
2197 unsigned blocksize = 1 << inode->i_blkbits;
2198 struct page *page;
2199 void *fsdata;
2200 pgoff_t index, curidx;
2201 loff_t curpos;
2202 unsigned zerofrom, offset, len;
2203 int err = 0;
2204
2205 index = pos >> PAGE_CACHE_SHIFT;
2206 offset = pos & ~PAGE_CACHE_MASK;
2207
2208 while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) {
2209 zerofrom = curpos & ~PAGE_CACHE_MASK;
2210 if (zerofrom & (blocksize-1)) {
2211 *bytes |= (blocksize-1);
2212 (*bytes)++;
2213 }
2214 len = PAGE_CACHE_SIZE - zerofrom;
2215
2216 err = pagecache_write_begin(file, mapping, curpos, len,
2217 AOP_FLAG_UNINTERRUPTIBLE,
2218 &page, &fsdata);
2219 if (err)
2220 goto out;
2221 zero_user_page(page, zerofrom, len, KM_USER0);
2222 err = pagecache_write_end(file, mapping, curpos, len, len,
2223 page, fsdata);
2224 if (err < 0)
2225 goto out;
2226 BUG_ON(err != len);
2227 err = 0;
2228 }
2229
2230
2231 if (index == curidx) {
2232 zerofrom = curpos & ~PAGE_CACHE_MASK;
2233
2234 if (offset <= zerofrom) {
2235 goto out;
2236 }
2237 if (zerofrom & (blocksize-1)) {
2238 *bytes |= (blocksize-1);
2239 (*bytes)++;
2240 }
2241 len = offset - zerofrom;
2242
2243 err = pagecache_write_begin(file, mapping, curpos, len,
2244 AOP_FLAG_UNINTERRUPTIBLE,
2245 &page, &fsdata);
2246 if (err)
2247 goto out;
2248 zero_user_page(page, zerofrom, len, KM_USER0);
2249 err = pagecache_write_end(file, mapping, curpos, len, len,
2250 page, fsdata);
2251 if (err < 0)
2252 goto out;
2253 BUG_ON(err != len);
2254 err = 0;
2255 }
2256out:
2257 return err;
2258}
2259
2260
2261
2262
2263
2264int cont_write_begin(struct file *file, struct address_space *mapping,
2265 loff_t pos, unsigned len, unsigned flags,
2266 struct page **pagep, void **fsdata,
2267 get_block_t *get_block, loff_t *bytes)
2268{
2269 struct inode *inode = mapping->host;
2270 unsigned blocksize = 1 << inode->i_blkbits;
2271 unsigned zerofrom;
2272 int err;
2273
2274 err = cont_expand_zero(file, mapping, pos, bytes);
2275 if (err)
2276 goto out;
2277
2278 zerofrom = *bytes & ~PAGE_CACHE_MASK;
2279 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2280 *bytes |= (blocksize-1);
2281 (*bytes)++;
2282 }
2283
2284 *pagep = NULL;
2285 err = block_write_begin(file, mapping, pos, len,
2286 flags, pagep, fsdata, get_block);
2287out:
2288 return err;
2289}
2290
2291int block_prepare_write(struct page *page, unsigned from, unsigned to,
2292 get_block_t *get_block)
2293{
2294 struct inode *inode = page->mapping->host;
2295 int err = __block_prepare_write(inode, page, from, to, get_block);
2296 if (err)
2297 ClearPageUptodate(page);
2298 return err;
2299}
2300
2301int block_commit_write(struct page *page, unsigned from, unsigned to)
2302{
2303 struct inode *inode = page->mapping->host;
2304 __block_commit_write(inode,page,from,to);
2305 return 0;
2306}
2307
2308int generic_commit_write(struct file *file, struct page *page,
2309 unsigned from, unsigned to)
2310{
2311 struct inode *inode = page->mapping->host;
2312 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2313 __block_commit_write(inode,page,from,to);
2314
2315
2316
2317
2318 if (pos > inode->i_size) {
2319 i_size_write(inode, pos);
2320 mark_inode_dirty(inode);
2321 }
2322 return 0;
2323}
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340int
2341block_page_mkwrite(struct vm_area_struct *vma, struct page *page,
2342 get_block_t get_block)
2343{
2344 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
2345 unsigned long end;
2346 loff_t size;
2347 int ret = -EINVAL;
2348
2349 lock_page(page);
2350 size = i_size_read(inode);
2351 if ((page->mapping != inode->i_mapping) ||
2352 (page_offset(page) > size)) {
2353
2354 goto out_unlock;
2355 }
2356
2357
2358 if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
2359 end = size & ~PAGE_CACHE_MASK;
2360 else
2361 end = PAGE_CACHE_SIZE;
2362
2363 ret = block_prepare_write(page, 0, end, get_block);
2364 if (!ret)
2365 ret = block_commit_write(page, 0, end);
2366
2367out_unlock:
2368 unlock_page(page);
2369 return ret;
2370}
2371
2372
2373
2374
2375
2376
2377static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2378{
2379 __end_buffer_read_notouch(bh, uptodate);
2380}
2381
2382
2383
2384
2385
2386
2387static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2388{
2389 struct buffer_head *bh;
2390
2391 BUG_ON(!PageLocked(page));
2392
2393 spin_lock(&page->mapping->private_lock);
2394 bh = head;
2395 do {
2396 if (PageDirty(page))
2397 set_buffer_dirty(bh);
2398 if (!bh->b_this_page)
2399 bh->b_this_page = head;
2400 bh = bh->b_this_page;
2401 } while (bh != head);
2402 attach_page_buffers(page, head);
2403 spin_unlock(&page->mapping->private_lock);
2404}
2405
2406
2407
2408
2409
2410int nobh_write_begin(struct file *file, struct address_space *mapping,
2411 loff_t pos, unsigned len, unsigned flags,
2412 struct page **pagep, void **fsdata,
2413 get_block_t *get_block)
2414{
2415 struct inode *inode = mapping->host;
2416 const unsigned blkbits = inode->i_blkbits;
2417 const unsigned blocksize = 1 << blkbits;
2418 struct buffer_head *head, *bh;
2419 struct page *page;
2420 pgoff_t index;
2421 unsigned from, to;
2422 unsigned block_in_page;
2423 unsigned block_start, block_end;
2424 sector_t block_in_file;
2425 char *kaddr;
2426 int nr_reads = 0;
2427 int ret = 0;
2428 int is_mapped_to_disk = 1;
2429
2430 index = pos >> PAGE_CACHE_SHIFT;
2431 from = pos & (PAGE_CACHE_SIZE - 1);
2432 to = from + len;
2433
2434 page = __grab_cache_page(mapping, index);
2435 if (!page)
2436 return -ENOMEM;
2437 *pagep = page;
2438 *fsdata = NULL;
2439
2440 if (page_has_buffers(page)) {
2441 unlock_page(page);
2442 page_cache_release(page);
2443 *pagep = NULL;
2444 return block_write_begin(file, mapping, pos, len, flags, pagep,
2445 fsdata, get_block);
2446 }
2447
2448 if (PageMappedToDisk(page))
2449 return 0;
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460 head = alloc_page_buffers(page, blocksize, 0);
2461 if (!head) {
2462 ret = -ENOMEM;
2463 goto out_release;
2464 }
2465
2466 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
2467
2468
2469
2470
2471
2472
2473 for (block_start = 0, block_in_page = 0, bh = head;
2474 block_start < PAGE_CACHE_SIZE;
2475 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2476 int create;
2477
2478 block_end = block_start + blocksize;
2479 bh->b_state = 0;
2480 create = 1;
2481 if (block_start >= to)
2482 create = 0;
2483 ret = get_block(inode, block_in_file + block_in_page,
2484 bh, create);
2485 if (ret)
2486 goto failed;
2487 if (!buffer_mapped(bh))
2488 is_mapped_to_disk = 0;
2489 if (buffer_new(bh))
2490 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2491 if (PageUptodate(page)) {
2492 set_buffer_uptodate(bh);
2493 continue;
2494 }
2495 if (buffer_new(bh) || !buffer_mapped(bh)) {
2496 kaddr = kmap_atomic(page, KM_USER0);
2497 if (block_start < from)
2498 memset(kaddr+block_start, 0, from-block_start);
2499 if (block_end > to)
2500 memset(kaddr + to, 0, block_end - to);
2501 flush_dcache_page(page);
2502 kunmap_atomic(kaddr, KM_USER0);
2503 continue;
2504 }
2505 if (buffer_uptodate(bh))
2506 continue;
2507 if (block_start < from || block_end > to) {
2508 lock_buffer(bh);
2509 bh->b_end_io = end_buffer_read_nobh;
2510 submit_bh(READ, bh);
2511 nr_reads++;
2512 }
2513 }
2514
2515 if (nr_reads) {
2516
2517
2518
2519
2520
2521 for (bh = head; bh; bh = bh->b_this_page) {
2522 wait_on_buffer(bh);
2523 if (!buffer_uptodate(bh))
2524 ret = -EIO;
2525 }
2526 if (ret)
2527 goto failed;
2528 }
2529
2530 if (is_mapped_to_disk)
2531 SetPageMappedToDisk(page);
2532
2533 *fsdata = head;
2534
2535 return 0;
2536
2537failed:
2538 BUG_ON(!ret);
2539
2540
2541
2542
2543
2544
2545
2546 attach_nobh_buffers(page, head);
2547 page_zero_new_buffers(page, from, to);
2548
2549out_release:
2550 unlock_page(page);
2551 page_cache_release(page);
2552 *pagep = NULL;
2553
2554 if (pos + len > inode->i_size)
2555 vmtruncate(inode, inode->i_size);
2556
2557 return ret;
2558}
2559EXPORT_SYMBOL(nobh_write_begin);
2560
2561int nobh_write_end(struct file *file, struct address_space *mapping,
2562 loff_t pos, unsigned len, unsigned copied,
2563 struct page *page, void *fsdata)
2564{
2565 struct inode *inode = page->mapping->host;
2566 struct buffer_head *head = fsdata;
2567 struct buffer_head *bh;
2568
2569 if (!PageMappedToDisk(page)) {
2570 if (unlikely(copied < len) && !page_has_buffers(page))
2571 attach_nobh_buffers(page, head);
2572 if (page_has_buffers(page))
2573 return generic_write_end(file, mapping, pos, len,
2574 copied, page, fsdata);
2575 }
2576
2577 SetPageUptodate(page);
2578 set_page_dirty(page);
2579 if (pos+copied > inode->i_size) {
2580 i_size_write(inode, pos+copied);
2581 mark_inode_dirty(inode);
2582 }
2583
2584 unlock_page(page);
2585 page_cache_release(page);
2586
2587 while (head) {
2588 bh = head;
2589 head = head->b_this_page;
2590 free_buffer_head(bh);
2591 }
2592
2593 return copied;
2594}
2595EXPORT_SYMBOL(nobh_write_end);
2596
2597
2598
2599
2600
2601
2602int nobh_writepage(struct page *page, get_block_t *get_block,
2603 struct writeback_control *wbc)
2604{
2605 struct inode * const inode = page->mapping->host;
2606 loff_t i_size = i_size_read(inode);
2607 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2608 unsigned offset;
2609 int ret;
2610
2611
2612 if (page->index < end_index)
2613 goto out;
2614
2615
2616 offset = i_size & (PAGE_CACHE_SIZE-1);
2617 if (page->index >= end_index+1 || !offset) {
2618
2619
2620
2621
2622
2623#if 0
2624
2625 if (page->mapping->a_ops->invalidatepage)
2626 page->mapping->a_ops->invalidatepage(page, offset);
2627#endif
2628 unlock_page(page);
2629 return 0;
2630 }
2631
2632
2633
2634
2635
2636
2637
2638
2639 zero_user_page(page, offset, PAGE_CACHE_SIZE - offset, KM_USER0);
2640out:
2641 ret = mpage_writepage(page, get_block, wbc);
2642 if (ret == -EAGAIN)
2643 ret = __block_write_full_page(inode, page, get_block, wbc);
2644 return ret;
2645}
2646EXPORT_SYMBOL(nobh_writepage);
2647
2648int nobh_truncate_page(struct address_space *mapping,
2649 loff_t from, get_block_t *get_block)
2650{
2651 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2652 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2653 unsigned blocksize;
2654 sector_t iblock;
2655 unsigned length, pos;
2656 struct inode *inode = mapping->host;
2657 struct page *page;
2658 struct buffer_head map_bh;
2659 int err;
2660
2661 blocksize = 1 << inode->i_blkbits;
2662 length = offset & (blocksize - 1);
2663
2664
2665 if (!length)
2666 return 0;
2667
2668 length = blocksize - length;
2669 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2670
2671 page = grab_cache_page(mapping, index);
2672 err = -ENOMEM;
2673 if (!page)
2674 goto out;
2675
2676 if (page_has_buffers(page)) {
2677has_buffers:
2678 unlock_page(page);
2679 page_cache_release(page);
2680 return block_truncate_page(mapping, from, get_block);
2681 }
2682
2683
2684 pos = blocksize;
2685 while (offset >= pos) {
2686 iblock++;
2687 pos += blocksize;
2688 }
2689
2690 err = get_block(inode, iblock, &map_bh, 0);
2691 if (err)
2692 goto unlock;
2693
2694 if (!buffer_mapped(&map_bh))
2695 goto unlock;
2696
2697
2698 if (!PageUptodate(page)) {
2699 err = mapping->a_ops->readpage(NULL, page);
2700 if (err) {
2701 page_cache_release(page);
2702 goto out;
2703 }
2704 lock_page(page);
2705 if (!PageUptodate(page)) {
2706 err = -EIO;
2707 goto unlock;
2708 }
2709 if (page_has_buffers(page))
2710 goto has_buffers;
2711 }
2712 zero_user_page(page, offset, length, KM_USER0);
2713 set_page_dirty(page);
2714 err = 0;
2715
2716unlock:
2717 unlock_page(page);
2718 page_cache_release(page);
2719out:
2720 return err;
2721}
2722EXPORT_SYMBOL(nobh_truncate_page);
2723
2724int block_truncate_page(struct address_space *mapping,
2725 loff_t from, get_block_t *get_block)
2726{
2727 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2728 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2729 unsigned blocksize;
2730 sector_t iblock;
2731 unsigned length, pos;
2732 struct inode *inode = mapping->host;
2733 struct page *page;
2734 struct buffer_head *bh;
2735 int err;
2736
2737 blocksize = 1 << inode->i_blkbits;
2738 length = offset & (blocksize - 1);
2739
2740
2741 if (!length)
2742 return 0;
2743
2744 length = blocksize - length;
2745 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2746
2747 page = grab_cache_page(mapping, index);
2748 err = -ENOMEM;
2749 if (!page)
2750 goto out;
2751
2752 if (!page_has_buffers(page))
2753 create_empty_buffers(page, blocksize, 0);
2754
2755
2756 bh = page_buffers(page);
2757 pos = blocksize;
2758 while (offset >= pos) {
2759 bh = bh->b_this_page;
2760 iblock++;
2761 pos += blocksize;
2762 }
2763
2764 err = 0;
2765 if (!buffer_mapped(bh)) {
2766 WARN_ON(bh->b_size != blocksize);
2767 err = get_block(inode, iblock, bh, 0);
2768 if (err)
2769 goto unlock;
2770
2771 if (!buffer_mapped(bh))
2772 goto unlock;
2773 }
2774
2775
2776 if (PageUptodate(page))
2777 set_buffer_uptodate(bh);
2778
2779 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2780 err = -EIO;
2781 ll_rw_block(READ, 1, &bh);
2782 wait_on_buffer(bh);
2783
2784 if (!buffer_uptodate(bh))
2785 goto unlock;
2786 }
2787
2788 zero_user_page(page, offset, length, KM_USER0);
2789 mark_buffer_dirty(bh);
2790 err = 0;
2791
2792unlock:
2793 unlock_page(page);
2794 page_cache_release(page);
2795out:
2796 return err;
2797}
2798
2799
2800
2801
2802int block_write_full_page(struct page *page, get_block_t *get_block,
2803 struct writeback_control *wbc)
2804{
2805 struct inode * const inode = page->mapping->host;
2806 loff_t i_size = i_size_read(inode);
2807 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2808 unsigned offset;
2809
2810
2811 if (page->index < end_index)
2812 return __block_write_full_page(inode, page, get_block, wbc);
2813
2814
2815 offset = i_size & (PAGE_CACHE_SIZE-1);
2816 if (page->index >= end_index+1 || !offset) {
2817
2818
2819
2820
2821
2822 do_invalidatepage(page, 0);
2823 unlock_page(page);
2824 return 0;
2825 }
2826
2827
2828
2829
2830
2831
2832
2833
2834 zero_user_page(page, offset, PAGE_CACHE_SIZE - offset, KM_USER0);
2835 return __block_write_full_page(inode, page, get_block, wbc);
2836}
2837
2838sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2839 get_block_t *get_block)
2840{
2841 struct buffer_head tmp;
2842 struct inode *inode = mapping->host;
2843 tmp.b_state = 0;
2844 tmp.b_blocknr = 0;
2845 tmp.b_size = 1 << inode->i_blkbits;
2846 get_block(inode, block, &tmp, 0);
2847 return tmp.b_blocknr;
2848}
2849
2850static void end_bio_bh_io_sync(struct bio *bio, int err)
2851{
2852 struct buffer_head *bh = bio->bi_private;
2853
2854 if (err == -EOPNOTSUPP) {
2855 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2856 set_bit(BH_Eopnotsupp, &bh->b_state);
2857 }
2858
2859 bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags));
2860 bio_put(bio);
2861}
2862
2863int submit_bh(int rw, struct buffer_head * bh)
2864{
2865 struct bio *bio;
2866 int ret = 0;
2867
2868 BUG_ON(!buffer_locked(bh));
2869 BUG_ON(!buffer_mapped(bh));
2870 BUG_ON(!bh->b_end_io);
2871
2872 if (buffer_ordered(bh) && (rw == WRITE))
2873 rw = WRITE_BARRIER;
2874
2875
2876
2877
2878
2879 if (test_set_buffer_req(bh) && (rw == WRITE || rw == WRITE_BARRIER))
2880 clear_buffer_write_io_error(bh);
2881
2882
2883
2884
2885
2886 bio = bio_alloc(GFP_NOIO, 1);
2887
2888 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
2889 bio->bi_bdev = bh->b_bdev;
2890 bio->bi_io_vec[0].bv_page = bh->b_page;
2891 bio->bi_io_vec[0].bv_len = bh->b_size;
2892 bio->bi_io_vec[0].bv_offset = bh_offset(bh);
2893
2894 bio->bi_vcnt = 1;
2895 bio->bi_idx = 0;
2896 bio->bi_size = bh->b_size;
2897
2898 bio->bi_end_io = end_bio_bh_io_sync;
2899 bio->bi_private = bh;
2900
2901 bio_get(bio);
2902 submit_bio(rw, bio);
2903
2904 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2905 ret = -EOPNOTSUPP;
2906
2907 bio_put(bio);
2908 return ret;
2909}
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937void ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
2938{
2939 int i;
2940
2941 for (i = 0; i < nr; i++) {
2942 struct buffer_head *bh = bhs[i];
2943
2944 if (rw == SWRITE)
2945 lock_buffer(bh);
2946 else if (test_set_buffer_locked(bh))
2947 continue;
2948
2949 if (rw == WRITE || rw == SWRITE) {
2950 if (test_clear_buffer_dirty(bh)) {
2951 bh->b_end_io = end_buffer_write_sync;
2952 get_bh(bh);
2953 submit_bh(WRITE, bh);
2954 continue;
2955 }
2956 } else {
2957 if (!buffer_uptodate(bh)) {
2958 bh->b_end_io = end_buffer_read_sync;
2959 get_bh(bh);
2960 submit_bh(rw, bh);
2961 continue;
2962 }
2963 }
2964 unlock_buffer(bh);
2965 }
2966}
2967
2968
2969
2970
2971
2972
2973int sync_dirty_buffer(struct buffer_head *bh)
2974{
2975 int ret = 0;
2976
2977 WARN_ON(atomic_read(&bh->b_count) < 1);
2978 lock_buffer(bh);
2979 if (test_clear_buffer_dirty(bh)) {
2980 get_bh(bh);
2981 bh->b_end_io = end_buffer_write_sync;
2982 ret = submit_bh(WRITE, bh);
2983 wait_on_buffer(bh);
2984 if (buffer_eopnotsupp(bh)) {
2985 clear_buffer_eopnotsupp(bh);
2986 ret = -EOPNOTSUPP;
2987 }
2988 if (!ret && !buffer_uptodate(bh))
2989 ret = -EIO;
2990 } else {
2991 unlock_buffer(bh);
2992 }
2993 return ret;
2994}
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016static inline int buffer_busy(struct buffer_head *bh)
3017{
3018 return atomic_read(&bh->b_count) |
3019 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3020}
3021
3022static int
3023drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3024{
3025 struct buffer_head *head = page_buffers(page);
3026 struct buffer_head *bh;
3027
3028 bh = head;
3029 do {
3030 if (buffer_write_io_error(bh) && page->mapping)
3031 set_bit(AS_EIO, &page->mapping->flags);
3032 if (buffer_busy(bh))
3033 goto failed;
3034 bh = bh->b_this_page;
3035 } while (bh != head);
3036
3037 do {
3038 struct buffer_head *next = bh->b_this_page;
3039
3040 if (!list_empty(&bh->b_assoc_buffers))
3041 __remove_assoc_queue(bh);
3042 bh = next;
3043 } while (bh != head);
3044 *buffers_to_free = head;
3045 __clear_page_buffers(page);
3046 return 1;
3047failed:
3048 return 0;
3049}
3050
3051int try_to_free_buffers(struct page *page)
3052{
3053 struct address_space * const mapping = page->mapping;
3054 struct buffer_head *buffers_to_free = NULL;
3055 int ret = 0;
3056
3057 BUG_ON(!PageLocked(page));
3058 if (PageWriteback(page))
3059 return 0;
3060
3061 if (mapping == NULL) {
3062 ret = drop_buffers(page, &buffers_to_free);
3063 goto out;
3064 }
3065
3066 spin_lock(&mapping->private_lock);
3067 ret = drop_buffers(page, &buffers_to_free);
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083 if (ret)
3084 cancel_dirty_page(page, PAGE_CACHE_SIZE);
3085 spin_unlock(&mapping->private_lock);
3086out:
3087 if (buffers_to_free) {
3088 struct buffer_head *bh = buffers_to_free;
3089
3090 do {
3091 struct buffer_head *next = bh->b_this_page;
3092 free_buffer_head(bh);
3093 bh = next;
3094 } while (bh != buffers_to_free);
3095 }
3096 return ret;
3097}
3098EXPORT_SYMBOL(try_to_free_buffers);
3099
3100void block_sync_page(struct page *page)
3101{
3102 struct address_space *mapping;
3103
3104 smp_mb();
3105 mapping = page_mapping(page);
3106 if (mapping)
3107 blk_run_backing_dev(mapping->backing_dev_info, page);
3108}
3109
3110
3111
3112
3113
3114
3115
3116
3117asmlinkage long sys_bdflush(int func, long data)
3118{
3119 static int msg_count;
3120
3121 if (!capable(CAP_SYS_ADMIN))
3122 return -EPERM;
3123
3124 if (msg_count < 5) {
3125 msg_count++;
3126 printk(KERN_INFO
3127 "warning: process `%s' used the obsolete bdflush"
3128 " system call\n", current->comm);
3129 printk(KERN_INFO "Fix your initscripts?\n");
3130 }
3131
3132 if (func == 1)
3133 do_exit(0);
3134 return 0;
3135}
3136
3137
3138
3139
3140static struct kmem_cache *bh_cachep;
3141
3142
3143
3144
3145
3146static int max_buffer_heads;
3147
3148int buffer_heads_over_limit;
3149
3150struct bh_accounting {
3151 int nr;
3152 int ratelimit;
3153};
3154
3155static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3156
3157static void recalc_bh_state(void)
3158{
3159 int i;
3160 int tot = 0;
3161
3162 if (__get_cpu_var(bh_accounting).ratelimit++ < 4096)
3163 return;
3164 __get_cpu_var(bh_accounting).ratelimit = 0;
3165 for_each_online_cpu(i)
3166 tot += per_cpu(bh_accounting, i).nr;
3167 buffer_heads_over_limit = (tot > max_buffer_heads);
3168}
3169
3170struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3171{
3172 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep,
3173 set_migrateflags(gfp_flags, __GFP_RECLAIMABLE));
3174 if (ret) {
3175 INIT_LIST_HEAD(&ret->b_assoc_buffers);
3176 get_cpu_var(bh_accounting).nr++;
3177 recalc_bh_state();
3178 put_cpu_var(bh_accounting);
3179 }
3180 return ret;
3181}
3182EXPORT_SYMBOL(alloc_buffer_head);
3183
3184void free_buffer_head(struct buffer_head *bh)
3185{
3186 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3187 kmem_cache_free(bh_cachep, bh);
3188 get_cpu_var(bh_accounting).nr--;
3189 recalc_bh_state();
3190 put_cpu_var(bh_accounting);
3191}
3192EXPORT_SYMBOL(free_buffer_head);
3193
3194static void buffer_exit_cpu(int cpu)
3195{
3196 int i;
3197 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3198
3199 for (i = 0; i < BH_LRU_SIZE; i++) {
3200 brelse(b->bhs[i]);
3201 b->bhs[i] = NULL;
3202 }
3203 get_cpu_var(bh_accounting).nr += per_cpu(bh_accounting, cpu).nr;
3204 per_cpu(bh_accounting, cpu).nr = 0;
3205 put_cpu_var(bh_accounting);
3206}
3207
3208static int buffer_cpu_notify(struct notifier_block *self,
3209 unsigned long action, void *hcpu)
3210{
3211 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
3212 buffer_exit_cpu((unsigned long)hcpu);
3213 return NOTIFY_OK;
3214}
3215
3216void __init buffer_init(void)
3217{
3218 int nrpages;
3219
3220 bh_cachep = KMEM_CACHE(buffer_head,
3221 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3222
3223
3224
3225
3226 nrpages = (nr_free_buffer_pages() * 10) / 100;
3227 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3228 hotcpu_notifier(buffer_cpu_notify, 0);
3229}
3230
3231EXPORT_SYMBOL(__bforget);
3232EXPORT_SYMBOL(__brelse);
3233EXPORT_SYMBOL(__wait_on_buffer);
3234EXPORT_SYMBOL(block_commit_write);
3235EXPORT_SYMBOL(block_prepare_write);
3236EXPORT_SYMBOL(block_page_mkwrite);
3237EXPORT_SYMBOL(block_read_full_page);
3238EXPORT_SYMBOL(block_sync_page);
3239EXPORT_SYMBOL(block_truncate_page);
3240EXPORT_SYMBOL(block_write_full_page);
3241EXPORT_SYMBOL(cont_write_begin);
3242EXPORT_SYMBOL(end_buffer_read_sync);
3243EXPORT_SYMBOL(end_buffer_write_sync);
3244EXPORT_SYMBOL(file_fsync);
3245EXPORT_SYMBOL(fsync_bdev);
3246EXPORT_SYMBOL(generic_block_bmap);
3247EXPORT_SYMBOL(generic_commit_write);
3248EXPORT_SYMBOL(generic_cont_expand_simple);
3249EXPORT_SYMBOL(init_buffer);
3250EXPORT_SYMBOL(invalidate_bdev);
3251EXPORT_SYMBOL(ll_rw_block);
3252EXPORT_SYMBOL(mark_buffer_dirty);
3253EXPORT_SYMBOL(submit_bh);
3254EXPORT_SYMBOL(sync_dirty_buffer);
3255EXPORT_SYMBOL(unlock_buffer);
3256