aboutsummaryrefslogtreecommitdiff
path: root/fs/direct-io.c
blob: 44a360ca80464c78ace80bf547efe4b7b1f42d3a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
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
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
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
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
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
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
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
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
/*
 * fs/direct-io.c
 *
 * Copyright (C) 2002, Linus Torvalds.
 *
 * O_DIRECT
 *
 * 04Jul2002	Andrew Morton
 *		Initial version
 * 11Sep2002	janetinc@us.ibm.com
 * 		added readv/writev support.
 * 29Oct2002	Andrew Morton
 *		rewrote bio_add_page() support.
 * 30Oct2002	pbadari@us.ibm.com
 *		added support for non-aligned IO.
 * 06Nov2002	pbadari@us.ibm.com
 *		added asynchronous IO support.
 * 21Jul2003	nathans@sgi.com
 *		added IO completion notifier.
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/task_io_accounting_ops.h>
#include <linux/bio.h>
#include <linux/wait.h>
#include <linux/err.h>
#include <linux/blkdev.h>
#include <linux/buffer_head.h>
#include <linux/rwsem.h>
#include <linux/uio.h>
#include <linux/atomic.h>

/*
 * How many user pages to map in one call to get_user_pages().  This determines
 * the size of a structure on the stack.
 */
#define DIO_PAGES	64

/*
 * This code generally works in units of "dio_blocks".  A dio_block is
 * somewhere between the hard sector size and the filesystem block size.  it
 * is determined on a per-invocation basis.   When talking to the filesystem
 * we need to convert dio_blocks to fs_blocks by scaling the dio_block quantity
 * down by dio->blkfactor.  Similarly, fs-blocksize quantities are converted
 * to bio_block quantities by shifting left by blkfactor.
 *
 * If blkfactor is zero then the user's request was aligned to the filesystem's
 * blocksize.
 */

struct dio {
	/* BIO submission state */
	struct bio *bio;		/* bio under assembly */
	struct inode *inode;
	int rw;
	loff_t i_size;			/* i_size when submitted */
	int flags;			/* doesn't change */
	unsigned blkbits;		/* doesn't change */
	unsigned blkfactor;		/* When we're using an alignment which
					   is finer than the filesystem's soft
					   blocksize, this specifies how much
					   finer.  blkfactor=2 means 1/4-block
					   alignment.  Does not change */
	unsigned start_zero_done;	/* flag: sub-blocksize zeroing has
					   been performed at the start of a
					   write */
	int pages_in_io;		/* approximate total IO pages */
	size_t	size;			/* total request size (doesn't change)*/
	sector_t block_in_file;		/* Current offset into the underlying
					   file in dio_block units. */
	unsigned blocks_available;	/* At block_in_file.  changes */
	sector_t final_block_in_request;/* doesn't change */
	unsigned first_block_in_page;	/* doesn't change, Used only once */
	int boundary;			/* prev block is at a boundary */
	int reap_counter;		/* rate limit reaping */
	get_block_t *get_block;		/* block mapping function */
	dio_iodone_t *end_io;		/* IO completion function */
	dio_submit_t *submit_io;	/* IO submition function */
	loff_t logical_offset_in_bio;	/* current first logical block in bio */
	sector_t final_block_in_bio;	/* current final block in bio + 1 */
	sector_t next_block_for_io;	/* next block to be put under IO,
					   in dio_blocks units */
	struct buffer_head map_bh;	/* last get_block() result */

	/*
	 * Deferred addition of a page to the dio.  These variables are
	 * private to dio_send_cur_page(), submit_page_section() and
	 * dio_bio_add_page().
	 */
	struct page *cur_page;		/* The page */
	unsigned cur_page_offset;	/* Offset into it, in bytes */
	unsigned cur_page_len;		/* Nr of bytes at cur_page_offset */
	sector_t cur_page_block;	/* Where it starts */
	loff_t cur_page_fs_offset;	/* Offset in file */

	/* BIO completion state */
	spinlock_t bio_lock;		/* protects BIO fields below */
	unsigned long refcount;		/* direct_io_worker() and bios */
	struct bio *bio_list;		/* singly linked via bi_private */
	struct task_struct *waiter;	/* waiting task (NULL if none) */

	/* AIO related stuff */
	struct kiocb *iocb;		/* kiocb */
	int is_async;			/* is IO async ? */
	int io_error;			/* IO error in completion path */
	ssize_t result;                 /* IO result */

	/*
	 * Page fetching state. These variables belong to dio_refill_pages().
	 */
	int curr_page;			/* changes */
	int total_pages;		/* doesn't change */
	unsigned long curr_user_address;/* changes */

	/*
	 * Page queue.  These variables belong to dio_refill_pages() and
	 * dio_get_page().
	 */
	unsigned head;			/* next page to process */
	unsigned tail;			/* last valid page + 1 */
	int page_errors;		/* errno from get_user_pages() */

	/*
	 * pages[] (and any fields placed after it) are not zeroed out at
	 * allocation time.  Don't add new fields after pages[] unless you
	 * wish that they not be zeroed.
	 */
	struct page *pages[DIO_PAGES];	/* page buffer */
};

static void __inode_dio_wait(struct inode *inode)
{
	wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);

	do {
		prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
		if (atomic_read(&inode->i_dio_count))
			schedule();
	} while (atomic_read(&inode->i_dio_count));
	finish_wait(wq, &q.wait);
}

/**
 * inode_dio_wait - wait for outstanding DIO requests to finish
 * @inode: inode to wait for
 *
 * Waits for all pending direct I/O requests to finish so that we can
 * proceed with a truncate or equivalent operation.
 *
 * Must be called under a lock that serializes taking new references
 * to i_dio_count, usually by inode->i_mutex.
 */
void inode_dio_wait(struct inode *inode)
{
	if (atomic_read(&inode->i_dio_count))
		__inode_dio_wait(inode);
}
EXPORT_SYMBOL_GPL(inode_dio_wait);

/*
 * inode_dio_done - signal finish of a direct I/O requests
 * @inode: inode the direct I/O happens on
 *
 * This is called once we've finished processing a direct I/O request,
 * and is used to wake up callers waiting for direct I/O to be quiesced.
 */
void inode_dio_done(struct inode *inode)
{
	if (atomic_dec_and_test(&inode->i_dio_count))
		wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
}
EXPORT_SYMBOL_GPL(inode_dio_done);

/*
 * How many pages are in the queue?
 */
static inline unsigned dio_pages_present(struct dio *dio)
{
	return dio->tail - dio->head;
}

/*
 * Go grab and pin some userspace pages.   Typically we'll get 64 at a time.
 */
static int dio_refill_pages(struct dio *dio)
{
	int ret;
	int nr_pages;

	nr_pages = min(dio->total_pages - dio->curr_page, DIO_PAGES);
	ret = get_user_pages_fast(
		dio->curr_user_address,		/* Where from? */
		nr_pages,			/* How many pages? */
		dio->rw == READ,		/* Write to memory? */
		&dio->pages[0]);		/* Put results here */

	if (ret < 0 && dio->blocks_available && (dio->rw & WRITE)) {
		struct page *page = ZERO_PAGE(0);
		/*
		 * A memory fault, but the filesystem has some outstanding
		 * mapped blocks.  We need to use those blocks up to avoid
		 * leaking stale data in the file.
		 */
		if (dio->page_errors == 0)
			dio->page_errors = ret;
		page_cache_get(page);
		dio->pages[0] = page;
		dio->head = 0;
		dio->tail = 1;
		ret = 0;
		goto out;
	}

	if (ret >= 0) {
		dio->curr_user_address += ret * PAGE_SIZE;
		dio->curr_page += ret;
		dio->head = 0;
		dio->tail = ret;
		ret = 0;
	}
out:
	return ret;	
}

/*
 * Get another userspace page.  Returns an ERR_PTR on error.  Pages are
 * buffered inside the dio so that we can call get_user_pages() against a
 * decent number of pages, less frequently.  To provide nicer use of the
 * L1 cache.
 */
static struct page *dio_get_page(struct dio *dio)
{
	if (dio_pages_present(dio) == 0) {
		int ret;

		ret = dio_refill_pages(dio);
		if (ret)
			return ERR_PTR(ret);
		BUG_ON(dio_pages_present(dio) == 0);
	}
	return dio->pages[dio->head++];
}

/**
 * dio_complete() - called when all DIO BIO I/O has been completed
 * @offset: the byte offset in the file of the completed operation
 *
 * This releases locks as dictated by the locking type, lets interested parties
 * know that a DIO operation has completed, and calculates the resulting return
 * code for the operation.
 *
 * It lets the filesystem know if it registered an interest earlier via
 * get_block.  Pass the private field of the map buffer_head so that
 * filesystems can use it to hold additional state between get_block calls and
 * dio_complete.
 */
static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is_async)
{
	ssize_t transferred = 0;

	/*
	 * AIO submission can race with bio completion to get here while
	 * expecting to have the last io completed by bio completion.
	 * In that case -EIOCBQUEUED is in fact not an error we want
	 * to preserve through this call.
	 */
	if (ret == -EIOCBQUEUED)
		ret = 0;

	if (dio->result) {
		transferred = dio->result;

		/* Check for short read case */
		if ((dio->rw == READ) && ((offset + transferred) > dio->i_size))
			transferred = dio->i_size - offset;
	}

	if (ret == 0)
		ret = dio->page_errors;
	if (ret == 0)
		ret = dio->io_error;
	if (ret == 0)
		ret = transferred;

	if (dio->end_io && dio->result) {
		dio->end_io(dio->iocb, offset, transferred,
			    dio->map_bh.b_private, ret, is_async);
	} else {
		if (is_async)
			aio_complete(dio->iocb, ret, 0);
		inode_dio_done(dio->inode);
	}

	return ret;
}

static int dio_bio_complete(struct dio *dio, struct bio *bio);
/*
 * Asynchronous IO callback. 
 */
static void dio_bio_end_aio(struct bio *bio, int error)
{
	struct dio *dio = bio->bi_private;
	unsigned long remaining;
	unsigned long flags;

	/* cleanup the bio */
	dio_bio_complete(dio, bio);

	spin_lock_irqsave(&dio->bio_lock, flags);
	remaining = --dio->refcount;
	if (remaining == 1 && dio->waiter)
		wake_up_process(dio->waiter);
	spin_unlock_irqrestore(&dio->bio_lock, flags);

	if (remaining == 0) {
		dio_complete(dio, dio->iocb->ki_pos, 0, true);
		kfree(dio);
	}
}

/*
 * The BIO completion handler simply queues the BIO up for the process-context
 * handler.
 *
 * During I/O bi_private points at the dio.  After I/O, bi_private is used to
 * implement a singly-linked list of completed BIOs, at dio->bio_list.
 */
static void dio_bio_end_io(struct bio *bio, int error)
{
	struct dio *dio = bio->bi_private;
	unsigned long flags;

	spin_lock_irqsave(&dio->bio_lock, flags);
	bio->bi_private = dio->bio_list;
	dio->bio_list = bio;
	if (--dio->refcount == 1 && dio->waiter)
		wake_up_process(dio->waiter);
	spin_unlock_irqrestore(&dio->bio_lock, flags);
}

/**
 * dio_end_io - handle the end io action for the given bio
 * @bio: The direct io bio thats being completed
 * @error: Error if there was one
 *
 * This is meant to be called by any filesystem that uses their own dio_submit_t
 * so that the DIO specific endio actions are dealt with after the filesystem
 * has done it's completion work.
 */
void dio_end_io(struct bio *bio, int error)
{
	struct dio *dio = bio->bi_private;

	if (dio->is_async)
		dio_bio_end_aio(bio, error);
	else
		dio_bio_end_io(bio, error);
}
EXPORT_SYMBOL_GPL(dio_end_io);

static void
dio_bio_alloc(struct dio *dio, struct block_device *bdev,
		sector_t first_sector, int nr_vecs)
{
	struct bio *bio;

	/*
	 * bio_alloc() is guaranteed to return a bio when called with
	 * __GFP_WAIT and we request a valid number of vectors.
	 */
	bio = bio_alloc(GFP_KERNEL, nr_vecs);

	bio->bi_bdev = bdev;
	bio->bi_sector = first_sector;
	if (dio->is_async)
		bio->bi_end_io = dio_bio_end_aio;
	else
		bio->bi_end_io = dio_bio_end_io;

	dio->bio = bio;
	dio->logical_offset_in_bio = dio->cur_page_fs_offset;
}

/*
 * In the AIO read case we speculatively dirty the pages before starting IO.
 * During IO completion, any of these pages which happen to have been written
 * back will be redirtied by bio_check_pages_dirty().
 *
 * bios hold a dio reference between submit_bio and ->end_io.
 */
static void dio_bio_submit(struct dio *dio)
{
	struct bio *bio = dio->bio;
	unsigned long flags;

	bio->bi_private = dio;

	spin_lock_irqsave(&dio->bio_lock, flags);
	dio->refcount++;
	spin_unlock_irqrestore(&dio->bio_lock, flags);

	if (dio->is_async && dio->rw == READ)
		bio_set_pages_dirty(bio);

	if (dio->submit_io)
		dio->submit_io(dio->rw, bio, dio->inode,
			       dio->logical_offset_in_bio);
	else
		submit_bio(dio->rw, bio);

	dio->bio = NULL;
	dio->boundary = 0;
	dio->logical_offset_in_bio = 0;
}

/*
 * Release any resources in case of a failure
 */
static void dio_cleanup(struct dio *dio)
{
	while (dio_pages_present(dio))
		page_cache_release(dio_get_page(dio));
}

/*
 * Wait for the next BIO to complete.  Remove it and return it.  NULL is
 * returned once all BIOs have been completed.  This must only be called once
 * all bios have been issued so that dio->refcount can only decrease.  This
 * requires that that the caller hold a reference on the dio.
 */
static struct bio *dio_await_one(struct dio *dio)
{
	unsigned long flags;
	struct bio *bio = NULL;

	spin_lock_irqsave(&dio->bio_lock, flags);

	/*
	 * Wait as long as the list is empty and there are bios in flight.  bio
	 * completion drops the count, maybe adds to the list, and wakes while
	 * holding the bio_lock so we don't need set_current_state()'s barrier
	 * and can call it after testing our condition.
	 */
	while (dio->refcount > 1 && dio->bio_list == NULL) {
		__set_current_state(TASK_UNINTERRUPTIBLE);
		dio->waiter = current;
		spin_unlock_irqrestore(&dio->bio_lock, flags);
		io_schedule();
		/* wake up sets us TASK_RUNNING */
		spin_lock_irqsave(&dio->bio_lock, flags);
		dio->waiter = NULL;
	}
	if (dio->bio_list) {
		bio = dio->bio_list;
		dio->bio_list = bio->bi_private;
	}
	spin_unlock_irqrestore(&dio->bio_lock, flags);
	return bio;
}

/*
 * Process one completed BIO.  No locks are held.
 */
static int dio_bio_complete(struct dio *dio, struct bio *bio)
{
	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	struct bio_vec *bvec = bio->bi_io_vec;
	int page_no;

	if (!uptodate)
		dio->io_error = -EIO;

	if (dio->is_async && dio->rw == READ) {
		bio_check_pages_dirty(bio);	/* transfers ownership */
	} else {
		for (page_no = 0; page_no < bio->bi_vcnt; page_no++) {
			struct page *page = bvec[page_no].bv_page;

			if (dio->rw == READ && !PageCompound(page))
				set_page_dirty_lock(page);
			page_cache_release(page);
		}
		bio_put(bio);
	}
	return uptodate ? 0 : -EIO;
}

/*
 * Wait on and process all in-flight BIOs.  This must only be called once
 * all bios have been issued so that the refcount can only decrease.
 * This just waits for all bios to make it through dio_bio_complete.  IO
 * errors are propagated through dio->io_error and should be propagated via
 * dio_complete().
 */
static void dio_await_completion(struct dio *dio)
{
	struct bio *bio;
	do {
		bio = dio_await_one(dio);
		if (bio)
			dio_bio_complete(dio, bio);
	} while (bio);
}

/*
 * A really large O_DIRECT read or write can generate a lot of BIOs.  So
 * to keep the memory consumption sane we periodically reap any completed BIOs
 * during the BIO generation phase.
 *
 * This also helps to limit the peak amount of pinned userspace memory.
 */
static int dio_bio_reap(struct dio *dio)
{
	int ret = 0;

	if (dio->reap_counter++ >= 64) {
		while (dio->bio_list) {
			unsigned long flags;
			struct bio *bio;
			int ret2;

			spin_lock_irqsave(&dio->bio_lock, flags);
			bio = dio->bio_list;
			dio->bio_list = bio->bi_private;
			spin_unlock_irqrestore(&dio->bio_lock, flags);
			ret2 = dio_bio_complete(dio, bio);
			if (ret == 0)
				ret = ret2;
		}
		dio->reap_counter = 0;
	}
	return ret;
}

/*
 * Call into the fs to map some more disk blocks.  We record the current number
 * of available blocks at dio->blocks_available.  These are in units of the
 * fs blocksize, (1 << inode->i_blkbits).
 *
 * The fs is allowed to map lots of blocks at once.  If it wants to do that,
 * it uses the passed inode-relative block number as the file offset, as usual.
 *
 * get_block() is passed the number of i_blkbits-sized blocks which direct_io
 * has remaining to do.  The fs should not map more than this number of blocks.
 *
 * If the fs has mapped a lot of blocks, it should populate bh->b_size to
 * indicate how much contiguous disk space has been made available at
 * bh->b_blocknr.
 *
 * If *any* of the mapped blocks are new, then the fs must set buffer_new().
 * This isn't very efficient...
 *
 * In the case of filesystem holes: the fs may return an arbitrarily-large
 * hole by returning an appropriate value in b_size and by clearing
 * buffer_mapped().  However the direct-io code will only process holes one
 * block at a time - it will repeatedly call get_block() as it walks the hole.
 */
static int get_more_blocks(struct dio *dio)
{
	int ret;
	struct buffer_head *map_bh = &dio->map_bh;
	sector_t fs_startblk;	/* Into file, in filesystem-sized blocks */
	unsigned long fs_count;	/* Number of filesystem-sized blocks */
	unsigned long dio_count;/* Number of dio_block-sized blocks */
	unsigned long blkmask;
	int create;

	/*
	 * If there was a memory error and we've overwritten all the
	 * mapped blocks then we can now return that memory error
	 */
	ret = dio->page_errors;
	if (ret == 0) {
		BUG_ON(dio->block_in_file >= dio->final_block_in_request);
		fs_startblk = dio->block_in_file >> dio->blkfactor;
		dio_count = dio->final_block_in_request - dio->block_in_file;
		fs_count = dio_count >> dio->blkfactor;
		blkmask = (1 << dio->blkfactor) - 1;
		if (dio_count & blkmask)	
			fs_count++;

		map_bh->b_state = 0;
		map_bh->b_size = fs_count << dio->inode->i_blkbits;

		/*
		 * For writes inside i_size on a DIO_SKIP_HOLES filesystem we
		 * forbid block creations: only overwrites are permitted.
		 * We will return early to the caller once we see an
		 * unmapped buffer head returned, and the caller will fall
		 * back to buffered I/O.
		 *
		 * Otherwise the decision is left to the get_blocks method,
		 * which may decide to handle it or also return an unmapped
		 * buffer head.
		 */
		create = dio->rw & WRITE;
		if (dio->flags & DIO_SKIP_HOLES) {
			if (dio->block_in_file < (i_size_read(dio->inode) >>
							dio->blkbits))
				create = 0;
		}

		ret = (*dio->get_block)(dio->inode, fs_startblk,
						map_bh, create);
	}
	return ret;
}

/*
 * There is no bio.  Make one now.
 */
static int dio_new_bio(struct dio *dio, sector_t start_sector)
{
	sector_t sector;
	int ret, nr_pages;

	ret = dio_bio_reap(dio);
	if (ret)
		goto out;
	sector = start_sector << (dio->blkbits - 9);
	nr_pages = min(dio->pages_in_io, bio_get_nr_vecs(dio->map_bh.b_bdev));
	nr_pages = min(nr_pages, BIO_MAX_PAGES);
	BUG_ON(nr_pages <= 0);
	dio_bio_alloc(dio, dio->map_bh.b_bdev, sector, nr_pages);
	dio->boundary = 0;
out:
	return ret;
}

/*
 * Attempt to put the current chunk of 'cur_page' into the current BIO.  If
 * that was successful then update final_block_in_bio and take a ref against
 * the just-added page.
 *
 * Return zero on success.  Non-zero means the caller needs to start a new BIO.
 */
static int dio_bio_add_page(struct dio *dio)
{
	int ret;

	ret = bio_add_page(dio->bio, dio->cur_page,
			dio->cur_page_len, dio->cur_page_offset);
	if (ret == dio->cur_page_len) {
		/*
		 * Decrement count only, if we are done with this page
		 */
		if ((dio->cur_page_len + dio->cur_page_offset) == PAGE_SIZE)
			dio->pages_in_io--;
		page_cache_get(dio->cur_page);
		dio->final_block_in_bio = dio->cur_page_block +
			(dio->cur_page_len >> dio->blkbits);
		ret = 0;
	} else {
		ret = 1;
	}
	return ret;
}
		
/*
 * Put cur_page under IO.  The section of cur_page which is described by
 * cur_page_offset,cur_page_len is put into a BIO.  The section of cur_page
 * starts on-disk at cur_page_block.
 *
 * We take a ref against the page here (on behalf of its presence in the bio).
 *
 * The caller of this function is responsible for removing cur_page from the
 * dio, and for dropping the refcount which came from that presence.
 */
static int dio_send_cur_page(struct dio *dio)
{
	int ret = 0;

	if (dio->bio) {
		loff_t cur_offset = dio->cur_page_fs_offset;
		loff_t bio_next_offset = dio->logical_offset_in_bio +
			dio->bio->bi_size;

		/*
		 * See whether this new request is contiguous with the old.
		 *
		 * Btrfs cannot handle having logically non-contiguous requests
		 * submitted.  For example if you have
		 *
		 * Logical:  [0-4095][HOLE][8192-12287]
		 * Physical: [0-4095]      [4096-8191]
		 *
		 * We cannot submit those pages together as one BIO.  So if our
		 * current logical offset in the file does not equal what would
		 * be the next logical offset in the bio, submit the bio we
		 * have.
		 */
		if (dio->final_block_in_bio != dio->cur_page_block ||
		    cur_offset != bio_next_offset)
			dio_bio_submit(dio);
		/*
		 * Submit now if the underlying fs is about to perform a
		 * metadata read
		 */
		else if (dio->boundary)
			dio_bio_submit(dio);
	}

	if (dio->bio == NULL) {
		ret = dio_new_bio(dio, dio->cur_page_block);
		if (ret)
			goto out;
	}

	if (dio_bio_add_page(dio) != 0) {
		dio_bio_submit(dio);
		ret = dio_new_bio(dio, dio->cur_page_block);
		if (ret == 0) {
			ret = dio_bio_add_page(dio);
			BUG_ON(ret != 0);
		}
	}
out:
	return ret;
}

/*
 * An autonomous function to put a chunk of a page under deferred IO.
 *
 * The caller doesn't actually know (or care) whether this piece of page is in
 * a BIO, or is under IO or whatever.  We just take care of all possible 
 * situations here.  The separation between the logic of do_direct_IO() and
 * that of submit_page_section() is important for clarity.  Please don't break.
 *
 * The chunk of page starts on-disk at blocknr.
 *
 * We perform deferred IO, by recording the last-submitted page inside our
 * private part of the dio structure.  If possible, we just expand the IO
 * across that page here.
 *
 * If that doesn't work out then we put the old page into the bio and add this
 * page to the dio instead.
 */
static int
submit_page_section(struct dio *dio, struct page *page,
		unsigned offset, unsigned len, sector_t blocknr)
{
	int ret = 0;

	if (dio->rw & WRITE) {
		/*
		 * Read accounting is performed in submit_bio()
		 */
		task_io_account_write(len);
	}

	/*
	 * Can we just grow the current page's presence in the dio?
	 */
	if (	(dio->cur_page == page) &&
		(dio->cur_page_offset + dio->cur_page_len == offset) &&
		(dio->cur_page_block +
			(dio->cur_page_len >> dio->blkbits) == blocknr)) {
		dio->cur_page_len += len;

		/*
		 * If dio->boundary then we want to schedule the IO now to
		 * avoid metadata seeks.
		 */
		if (dio->boundary) {
			ret = dio_send_cur_page(dio);
			page_cache_release(dio->cur_page);
			dio->cur_page = NULL;
		}
		goto out;
	}

	/*
	 * If there's a deferred page already there then send it.
	 */
	if (dio->cur_page) {
		ret = dio_send_cur_page(dio);
		page_cache_release(dio->cur_page);
		dio->cur_page = NULL;
		if (ret)
			goto out;
	}

	page_cache_get(page);		/* It is in dio */
	dio->cur_page = page;
	dio->cur_page_offset = offset;
	dio->cur_page_len = len;
	dio->cur_page_block = blocknr;
	dio->cur_page_fs_offset = dio->block_in_file << dio->blkbits;
out:
	return ret;
}

/*
 * Clean any dirty buffers in the blockdev mapping which alias newly-created
 * file blocks.  Only called for S_ISREG files - blockdevs do not set
 * buffer_new
 */
static void clean_blockdev_aliases(struct dio *dio)
{
	unsigned i;
	unsigned nblocks;

	nblocks = dio->map_bh.b_size >> dio->inode->i_blkbits;

	for (i = 0; i < nblocks; i++) {
		unmap_underlying_metadata(dio->map_bh.b_bdev,
					dio->map_bh.b_blocknr + i);
	}
}

/*
 * If we are not writing the entire block and get_block() allocated
 * the block for us, we need to fill-in the unused portion of the
 * block with zeros. This happens only if user-buffer, fileoffset or
 * io length is not filesystem block-size multiple.
 *
 * `end' is zero if we're doing the start of the IO, 1 at the end of the
 * IO.
 */
static void dio_zero_block(struct dio *dio, int end)
{
	unsigned dio_blocks_per_fs_block;
	unsigned this_chunk_blocks;	/* In dio_blocks */
	unsigned this_chunk_bytes;
	struct page *page;

	dio->start_zero_done = 1;
	if (!dio->blkfactor || !buffer_new(&dio->map_bh))
		return;

	dio_blocks_per_fs_block = 1 << dio->blkfactor;
	this_chunk_blocks = dio->block_in_file & (dio_blocks_per_fs_block - 1);

	if (!this_chunk_blocks)
		return;

	/*
	 * We need to zero out part of an fs block.  It is either at the
	 * beginning or the end of the fs block.
	 */
	if (end) 
		this_chunk_blocks = dio_blocks_per_fs_block - this_chunk_blocks;

	this_chunk_bytes = this_chunk_blocks << dio->blkbits;

	page = ZERO_PAGE(0);
	if (submit_page_section(dio, page, 0, this_chunk_bytes, 
				dio->next_block_for_io))
		return;

	dio->next_block_for_io += this_chunk_blocks;
}

/*
 * Walk the user pages, and the file, mapping blocks to disk and generating
 * a sequence of (page,offset,len,block) mappings.  These mappings are injected
 * into submit_page_section(), which takes care of the next stage of submission
 *
 * Direct IO against a blockdev is different from a file.  Because we can
 * happily perform page-sized but 512-byte aligned IOs.  It is important that
 * blockdev IO be able to have fine alignment and large sizes.
 *
 * So what we do is to permit the ->get_block function to populate bh.b_size
 * with the size of IO which is permitted at this offset and this i_blkbits.
 *
 * For best results, the blockdev should be set up with 512-byte i_blkbits and
 * it should set b_size to PAGE_SIZE or more inside get_block().  This gives
 * fine alignment but still allows this function to work in PAGE_SIZE units.
 */
static int do_direct_IO(struct dio *dio)
{
	const unsigned blkbits = dio->blkbits;
	const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
	struct page *page;
	unsigned block_in_page;
	struct buffer_head *map_bh = &dio->map_bh;
	int ret = 0;

	/* The I/O can start at any block offset within the first page */
	block_in_page = dio->first_block_in_page;

	while (dio->block_in_file < dio->final_block_in_request) {
		page = dio_get_page(dio);
		if (IS_ERR(page)) {
			ret = PTR_ERR(page);
			goto out;
		}

		while (block_in_page < blocks_per_page) {
			unsigned offset_in_page = block_in_page << blkbits;
			unsigned this_chunk_bytes;	/* # of bytes mapped */
			unsigned this_chunk_blocks;	/* # of blocks */
			unsigned u;

			if (dio->blocks_available == 0) {
				/*
				 * Need to go and map some more disk
				 */
				unsigned long blkmask;
				unsigned long dio_remainder;

				ret = get_more_blocks(dio);
				if (ret) {
					page_cache_release(page);
					goto out;
				}
				if (!buffer_mapped(map_bh))
					goto do_holes;

				dio->blocks_available =
						map_bh->b_size >> dio->blkbits;
				dio->next_block_for_io =
					map_bh->b_blocknr << dio->blkfactor;
				if (buffer_new(map_bh))
					clean_blockdev_aliases(dio);

				if (!dio->blkfactor)
					goto do_holes;

				blkmask = (1 << dio->blkfactor) - 1;
				dio_remainder = (dio->block_in_file & blkmask);

				/*
				 * If we are at the start of IO and that IO
				 * starts partway into a fs-block,
				 * dio_remainder will be non-zero.  If the IO
				 * is a read then we can simply advance the IO
				 * cursor to the first block which is to be
				 * read.  But if the IO is a write and the
				 * block was newly allocated we cannot do that;
				 * the start of the fs block must be zeroed out
				 * on-disk
				 */
				if (!buffer_new(map_bh))
					dio->next_block_for_io += dio_remainder;
				dio->blocks_available -= dio_remainder;
			}
do_holes:
			/* Handle holes */
			if (!buffer_mapped(map_bh)) {
				loff_t i_size_aligned;

				/* AKPM: eargh, -ENOTBLK is a hack */
				if (dio->rw & WRITE) {
					page_cache_release(page);
					return -ENOTBLK;
				}

				/*
				 * Be sure to account for a partial block as the
				 * last block in the file
				 */
				i_size_aligned = ALIGN(i_size_read(dio->inode),
							1 << blkbits);
				if (dio->block_in_file >=
						i_size_aligned >> blkbits) {
					/* We hit eof */
					page_cache_release(page);
					goto out;
				}
				zero_user(page, block_in_page << blkbits,
						1 << blkbits);
				dio->block_in_file++;
				block_in_page++;
				goto next_block;
			}

			/*
			 * If we're performing IO which has an alignment which
			 * is finer than the underlying fs, go check to see if
			 * we must zero out the start of this block.
			 */
			if (unlikely(dio->blkfactor && !dio->start_zero_done))
				dio_zero_block(dio, 0);

			/*
			 * Work out, in this_chunk_blocks, how much disk we
			 * can add to this page
			 */
			this_chunk_blocks = dio->blocks_available;
			u = (PAGE_SIZE - offset_in_page) >> blkbits;
			if (this_chunk_blocks > u)
				this_chunk_blocks = u;
			u = dio->final_block_in_request - dio->block_in_file;
			if (this_chunk_blocks > u)
				this_chunk_blocks = u;
			this_chunk_bytes = this_chunk_blocks << blkbits;
			BUG_ON(this_chunk_bytes == 0);

			dio->boundary = buffer_boundary(map_bh);
			ret = submit_page_section(dio, page, offset_in_page,
				this_chunk_bytes, dio->next_block_for_io);
			if (ret) {
				page_cache_release(page);
				goto out;
			}
			dio->next_block_for_io += this_chunk_blocks;

			dio->block_in_file += this_chunk_blocks;
			block_in_page += this_chunk_blocks;
			dio->blocks_available -= this_chunk_blocks;
next_block:
			BUG_ON(dio->block_in_file > dio->final_block_in_request);
			if (dio->block_in_file == dio->final_block_in_request)
				break;
		}

		/* Drop the ref which was taken in get_user_pages() */
		page_cache_release(page);
		block_in_page = 0;
	}
out:
	return ret;
}

static ssize_t
direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode, 
	const struct iovec *iov, loff_t offset, unsigned long nr_segs, 
	unsigned blkbits, get_block_t get_block, dio_iodone_t end_io,
	dio_submit_t submit_io, struct dio *dio)
{
	unsigned long user_addr; 
	unsigned long flags;
	int seg;
	ssize_t ret = 0;
	ssize_t ret2;
	size_t bytes;

	dio->inode = inode;
	dio->rw = rw;
	dio->blkbits = blkbits;
	dio->blkfactor = inode->i_blkbits - blkbits;
	dio->block_in_file = offset >> blkbits;

	dio->get_block = get_block;
	dio->end_io = end_io;
	dio->submit_io = submit_io;
	dio->final_block_in_bio = -1;
	dio->next_block_for_io = -1;

	dio->iocb = iocb;
	dio->i_size = i_size_read(inode);

	spin_lock_init(&dio->bio_lock);
	dio->refcount = 1;

	/*
	 * In case of non-aligned buffers, we may need 2 more
	 * pages since we need to zero out first and last block.
	 */
	if (unlikely(dio->blkfactor))
		dio->pages_in_io = 2;

	for (seg = 0; seg < nr_segs; seg++) {
		user_addr = (unsigned long)iov[seg].iov_base;
		dio->pages_in_io +=
			((user_addr+iov[seg].iov_len +PAGE_SIZE-1)/PAGE_SIZE
				- user_addr/PAGE_SIZE);
	}

	for (seg = 0; seg < nr_segs; seg++) {
		user_addr = (unsigned long)iov[seg].iov_base;
		dio->size += bytes = iov[seg].iov_len;

		/* Index into the first page of the first block */
		dio->first_block_in_page = (user_addr & ~PAGE_MASK) >> blkbits;
		dio->final_block_in_request = dio->block_in_file +
						(bytes >> blkbits);
		/* Page fetching state */
		dio->head = 0;
		dio->tail = 0;
		dio->curr_page = 0;

		dio->total_pages = 0;
		if (user_addr & (PAGE_SIZE-1)) {
			dio->total_pages++;
			bytes -= PAGE_SIZE - (user_addr & (PAGE_SIZE - 1));
		}
		dio->total_pages += (bytes + PAGE_SIZE - 1) / PAGE_SIZE;
		dio->curr_user_address = user_addr;
	
		ret = do_direct_IO(dio);

		dio->result += iov[seg].iov_len -
			((dio->final_block_in_request - dio->block_in_file) <<
					blkbits);

		if (ret) {
			dio_cleanup(dio);
			break;
		}
	} /* end iovec loop */

	if (ret == -ENOTBLK) {
		/*
		 * The remaining part of the request will be
		 * be handled by buffered I/O when we return
		 */
		ret = 0;
	}
	/*
	 * There may be some unwritten disk at the end of a part-written
	 * fs-block-sized block.  Go zero that now.
	 */
	dio_zero_block(dio, 1);

	if (dio->cur_page) {
		ret2 = dio_send_cur_page(dio);
		if (ret == 0)
			ret = ret2;
		page_cache_release(dio->cur_page);
		dio->cur_page = NULL;
	}
	if (dio->bio)
		dio_bio_submit(dio);

	/*
	 * It is possible that, we return short IO due to end of file.
	 * In that case, we need to release all the pages we got hold on.
	 */
	dio_cleanup(dio);

	/*
	 * All block lookups have been performed. For READ requests
	 * we can let i_mutex go now that its achieved its purpose
	 * of protecting us from looking up uninitialized blocks.
	 */
	if (rw == READ && (dio->flags & DIO_LOCKING))
		mutex_unlock(&dio->inode->i_mutex);

	/*
	 * The only time we want to leave bios in flight is when a successful
	 * partial aio read or full aio write have been setup.  In that case
	 * bio completion will call aio_complete.  The only time it's safe to
	 * call aio_complete is when we return -EIOCBQUEUED, so we key on that.
	 * This had *better* be the only place that raises -EIOCBQUEUED.
	 */
	BUG_ON(ret == -EIOCBQUEUED);
	if (dio->is_async && ret == 0 && dio->result &&
	    ((rw & READ) || (dio->result == dio->size)))
		ret = -EIOCBQUEUED;

	if (ret != -EIOCBQUEUED)
		dio_await_completion(dio);

	/*
	 * Sync will always be dropping the final ref and completing the
	 * operation.  AIO can if it was a broken operation described above or
	 * in fact if all the bios race to complete before we get here.  In
	 * that case dio_complete() translates the EIOCBQUEUED into the proper
	 * return code that the caller will hand to aio_complete().
	 *
	 * This is managed by the bio_lock instead of being an atomic_t so that
	 * completion paths can drop their ref and use the remaining count to
	 * decide to wake the submission path atomically.
	 */
	spin_lock_irqsave(&dio->bio_lock, flags);
	ret2 = --dio->refcount;
	spin_unlock_irqrestore(&dio->bio_lock, flags);

	if (ret2 == 0) {
		ret = dio_complete(dio, offset, ret, false);
		kfree(dio);
	} else
		BUG_ON(ret != -EIOCBQUEUED);

	return ret;
}

/*
 * This is a library function for use by filesystem drivers.
 *
 * The locking rules are governed by the flags parameter:
 *  - if the flags value contains DIO_LOCKING we use a fancy locking
 *    scheme for dumb filesystems.
 *    For writes this function is called under i_mutex and returns with
 *    i_mutex held, for reads, i_mutex is not held on entry, but it is
 *    taken and dropped again before returning.
 *  - if the flags value does NOT contain DIO_LOCKING we don't use any
 *    internal locking but rather rely on the filesystem to synchronize
 *    direct I/O reads/writes versus each other and truncate.
 *
 * To help with locking against truncate we incremented the i_dio_count
 * counter before starting direct I/O, and decrement it once we are done.
 * Truncate can wait for it to reach zero to provide exclusion.  It is
 * expected that filesystem provide exclusion between new direct I/O
 * and truncates.  For DIO_LOCKING filesystems this is done by i_mutex,
 * but other filesystems need to take care of this on their own.
 */
ssize_t
__blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
	struct block_device *bdev, const struct iovec *iov, loff_t offset, 
	unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
	dio_submit_t submit_io,	int flags)
{
	int seg;
	size_t size;
	unsigned long addr;
	unsigned blkbits = inode->i_blkbits;
	unsigned bdev_blkbits = 0;
	unsigned blocksize_mask = (1 << blkbits) - 1;
	ssize_t retval = -EINVAL;
	loff_t end = offset;
	struct dio *dio;

	if (rw & WRITE)
		rw = WRITE_ODIRECT;

	if (bdev)
		bdev_blkbits = blksize_bits(bdev_logical_block_size(bdev));

	if (offset & blocksize_mask) {
		if (bdev)
			 blkbits = bdev_blkbits;
		blocksize_mask = (1 << blkbits) - 1;
		if (offset & blocksize_mask)
			goto out;
	}

	/* Check the memory alignment.  Blocks cannot straddle pages */
	for (seg = 0; seg < nr_segs; seg++) {
		addr = (unsigned long)iov[seg].iov_base;
		size = iov[seg].iov_len;
		end += size;
		if ((addr & blocksize_mask) || (size & blocksize_mask))  {
			if (bdev)
				 blkbits = bdev_blkbits;
			blocksize_mask = (1 << blkbits) - 1;
			if ((addr & blocksize_mask) || (size & blocksize_mask))  
				goto out;
		}
	}

	/* watch out for a 0 len io from a tricksy fs */
	if (rw == READ && end == offset)
		return 0;

	dio = kmalloc(sizeof(*dio), GFP_KERNEL);
	retval = -ENOMEM;
	if (!dio)
		goto out;
	/*
	 * Believe it or not, zeroing out the page array caused a .5%
	 * performance regression in a database benchmark.  So, we take
	 * care to only zero out what's needed.
	 */
	memset(dio, 0, offsetof(struct dio, pages));

	dio->flags = flags;
	if (dio->flags & DIO_LOCKING) {
		if (rw == READ) {
			struct address_space *mapping =
					iocb->ki_filp->f_mapping;

			/* will be released by direct_io_worker */
			mutex_lock(&inode->i_mutex);

			retval = filemap_write_and_wait_range(mapping, offset,
							      end - 1);
			if (retval) {
				mutex_unlock(&inode->i_mutex);
				kfree(dio);
				goto out;
			}
		}
	}

	/*
	 * Will be decremented at I/O completion time.
	 */
	atomic_inc(&inode->i_dio_count);

	/*
	 * For file extending writes updating i_size before data
	 * writeouts complete can expose uninitialized blocks. So
	 * even for AIO, we need to wait for i/o to complete before
	 * returning in this case.
	 */
	dio->is_async = !is_sync_kiocb(iocb) && !((rw & WRITE) &&
		(end > i_size_read(inode)));

	retval = direct_io_worker(rw, iocb, inode, iov, offset,
				nr_segs, blkbits, get_block, end_io,
				submit_io, dio);

out:
	return retval;
}
EXPORT_SYMBOL(__blockdev_direct_IO);