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
|
/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 only,
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License version 2 for more details (a copy is included
* in the LICENSE file that accompanied this code).
*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
* http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
* CA 95054 USA or visit www.sun.com if you need additional information or
* have any questions.
*
* GPL HEADER END
*/
/*
* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
* Copyright (c) 2011, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/llite/rw.c
*
* Lustre Lite I/O page cache routines shared by different kernel revs
*/
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/unistd.h>
#include <linux/writeback.h>
#include <linux/uaccess.h>
#include <linux/fs.h>
#include <linux/pagemap.h>
/* current_is_kswapd() */
#include <linux/swap.h>
#define DEBUG_SUBSYSTEM S_LLITE
#include "../include/lustre_lite.h"
#include "../include/obd_cksum.h"
#include "llite_internal.h"
#include "../include/linux/lustre_compat25.h"
/**
* Finalizes cl-data before exiting typical address_space operation. Dual to
* ll_cl_init().
*/
void ll_cl_fini(struct ll_cl_context *lcc)
{
struct lu_env *env = lcc->lcc_env;
struct cl_io *io = lcc->lcc_io;
struct cl_page *page = lcc->lcc_page;
LASSERT(lcc->lcc_cookie == current);
LASSERT(env);
if (page) {
lu_ref_del(&page->cp_reference, "cl_io", io);
cl_page_put(env, page);
}
cl_env_put(env, &lcc->lcc_refcheck);
}
/**
* Initializes common cl-data at the typical address_space operation entry
* point.
*/
struct ll_cl_context *ll_cl_init(struct file *file, struct page *vmpage)
{
struct ll_cl_context *lcc;
struct lu_env *env;
struct cl_io *io;
struct cl_object *clob;
struct vvp_io *vio;
int refcheck;
int result = 0;
clob = ll_i2info(file_inode(file))->lli_clob;
LASSERT(clob);
env = cl_env_get(&refcheck);
if (IS_ERR(env))
return ERR_CAST(env);
lcc = &ll_env_info(env)->lti_io_ctx;
memset(lcc, 0, sizeof(*lcc));
lcc->lcc_env = env;
lcc->lcc_refcheck = refcheck;
lcc->lcc_cookie = current;
vio = vvp_env_io(env);
io = vio->vui_cl.cis_io;
lcc->lcc_io = io;
if (!io) {
struct inode *inode = file_inode(file);
CERROR("%s: " DFID " no active IO, please file a ticket.\n",
ll_get_fsname(inode->i_sb, NULL, 0),
PFID(ll_inode2fid(inode)));
dump_stack();
result = -EIO;
}
if (result == 0 && vmpage) {
struct cl_page *page;
LASSERT(io->ci_state == CIS_IO_GOING);
LASSERT(vio->vui_fd == LUSTRE_FPRIVATE(file));
page = cl_page_find(env, clob, vmpage->index, vmpage,
CPT_CACHEABLE);
if (!IS_ERR(page)) {
lcc->lcc_page = page;
lu_ref_add(&page->cp_reference, "cl_io", io);
result = 0;
} else
result = PTR_ERR(page);
}
if (result) {
ll_cl_fini(lcc);
lcc = ERR_PTR(result);
}
return lcc;
}
static void ll_ra_stats_inc_sbi(struct ll_sb_info *sbi, enum ra_stat which);
/**
* Get readahead pages from the filesystem readahead pool of the client for a
* thread.
*
* /param sbi superblock for filesystem readahead state ll_ra_info
* /param ria per-thread readahead state
* /param pages number of pages requested for readahead for the thread.
*
* WARNING: This algorithm is used to reduce contention on sbi->ll_lock.
* It should work well if the ra_max_pages is much greater than the single
* file's read-ahead window, and not too many threads contending for
* these readahead pages.
*
* TODO: There may be a 'global sync problem' if many threads are trying
* to get an ra budget that is larger than the remaining readahead pages
* and reach here at exactly the same time. They will compute /a ret to
* consume the remaining pages, but will fail at atomic_add_return() and
* get a zero ra window, although there is still ra space remaining. - Jay
*/
static unsigned long ll_ra_count_get(struct ll_sb_info *sbi,
struct ra_io_arg *ria,
unsigned long pages, unsigned long min)
{
struct ll_ra_info *ra = &sbi->ll_ra_info;
long ret;
/* If read-ahead pages left are less than 1M, do not do read-ahead,
* otherwise it will form small read RPC(< 1M), which hurt server
* performance a lot.
*/
ret = min(ra->ra_max_pages - atomic_read(&ra->ra_cur_pages), pages);
if (ret < 0 || ret < min_t(long, PTLRPC_MAX_BRW_PAGES, pages)) {
ret = 0;
goto out;
}
/* If the non-strided (ria_pages == 0) readahead window
* (ria_start + ret) has grown across an RPC boundary, then trim
* readahead size by the amount beyond the RPC so it ends on an
* RPC boundary. If the readahead window is already ending on
* an RPC boundary (beyond_rpc == 0), or smaller than a full
* RPC (beyond_rpc < ret) the readahead size is unchanged.
* The (beyond_rpc != 0) check is skipped since the conditional
* branch is more expensive than subtracting zero from the result.
*
* Strided read is left unaligned to avoid small fragments beyond
* the RPC boundary from needing an extra read RPC.
*/
if (ria->ria_pages == 0) {
long beyond_rpc = (ria->ria_start + ret) % PTLRPC_MAX_BRW_PAGES;
if (/* beyond_rpc != 0 && */ beyond_rpc < ret)
ret -= beyond_rpc;
}
if (atomic_add_return(ret, &ra->ra_cur_pages) > ra->ra_max_pages) {
atomic_sub(ret, &ra->ra_cur_pages);
ret = 0;
}
out:
if (ret < min) {
/* override ra limit for maximum performance */
atomic_add(min - ret, &ra->ra_cur_pages);
ret = min;
}
return ret;
}
void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len)
{
struct ll_ra_info *ra = &sbi->ll_ra_info;
atomic_sub(len, &ra->ra_cur_pages);
}
static void ll_ra_stats_inc_sbi(struct ll_sb_info *sbi, enum ra_stat which)
{
LASSERTF(which >= 0 && which < _NR_RA_STAT, "which: %u\n", which);
lprocfs_counter_incr(sbi->ll_ra_stats, which);
}
void ll_ra_stats_inc(struct inode *inode, enum ra_stat which)
{
struct ll_sb_info *sbi = ll_i2sbi(inode);
ll_ra_stats_inc_sbi(sbi, which);
}
#define RAS_CDEBUG(ras) \
CDEBUG(D_READA, \
"lrp %lu cr %lu cp %lu ws %lu wl %lu nra %lu r %lu ri %lu" \
"csr %lu sf %lu sp %lu sl %lu\n", \
ras->ras_last_readpage, ras->ras_consecutive_requests, \
ras->ras_consecutive_pages, ras->ras_window_start, \
ras->ras_window_len, ras->ras_next_readahead, \
ras->ras_requests, ras->ras_request_index, \
ras->ras_consecutive_stride_requests, ras->ras_stride_offset, \
ras->ras_stride_pages, ras->ras_stride_length)
static int index_in_window(unsigned long index, unsigned long point,
unsigned long before, unsigned long after)
{
unsigned long start = point - before, end = point + after;
if (start > point)
start = 0;
if (end < point)
end = ~0;
return start <= index && index <= end;
}
void ll_ras_enter(struct file *f)
{
struct ll_file_data *fd = LUSTRE_FPRIVATE(f);
struct ll_readahead_state *ras = &fd->fd_ras;
spin_lock(&ras->ras_lock);
ras->ras_requests++;
ras->ras_request_index = 0;
ras->ras_consecutive_requests++;
spin_unlock(&ras->ras_lock);
}
static int cl_read_ahead_page(const struct lu_env *env, struct cl_io *io,
struct cl_page_list *queue, struct cl_page *page,
struct cl_object *clob, pgoff_t *max_index)
{
struct page *vmpage = page->cp_vmpage;
struct vvp_page *vpg;
int rc;
rc = 0;
cl_page_assume(env, io, page);
lu_ref_add(&page->cp_reference, "ra", current);
vpg = cl2vvp_page(cl_object_page_slice(clob, page));
if (!vpg->vpg_defer_uptodate && !PageUptodate(vmpage)) {
CDEBUG(D_READA, "page index %lu, max_index: %lu\n",
vvp_index(vpg), *max_index);
if (*max_index == 0 || vvp_index(vpg) > *max_index)
rc = cl_page_is_under_lock(env, io, page, max_index);
if (rc == 0) {
vpg->vpg_defer_uptodate = 1;
vpg->vpg_ra_used = 0;
cl_page_list_add(queue, page);
rc = 1;
} else {
cl_page_discard(env, io, page);
rc = -ENOLCK;
}
} else {
/* skip completed pages */
cl_page_unassume(env, io, page);
}
lu_ref_del(&page->cp_reference, "ra", current);
cl_page_put(env, page);
return rc;
}
/**
* Initiates read-ahead of a page with given index.
*
* \retval +ve: page was added to \a queue.
*
* \retval -ENOLCK: there is no extent lock for this part of a file, stop
* read-ahead.
*
* \retval -ve, 0: page wasn't added to \a queue for other reason.
*/
static int ll_read_ahead_page(const struct lu_env *env, struct cl_io *io,
struct cl_page_list *queue,
pgoff_t index, pgoff_t *max_index)
{
struct cl_object *clob = io->ci_obj;
struct inode *inode = vvp_object_inode(clob);
struct page *vmpage;
struct cl_page *page;
enum ra_stat which = _NR_RA_STAT; /* keep gcc happy */
int rc = 0;
const char *msg = NULL;
vmpage = grab_cache_page_nowait(inode->i_mapping, index);
if (vmpage) {
/* Check if vmpage was truncated or reclaimed */
if (vmpage->mapping == inode->i_mapping) {
page = cl_page_find(env, clob, vmpage->index,
vmpage, CPT_CACHEABLE);
if (!IS_ERR(page)) {
rc = cl_read_ahead_page(env, io, queue,
page, clob, max_index);
if (rc == -ENOLCK) {
which = RA_STAT_FAILED_MATCH;
msg = "lock match failed";
}
} else {
which = RA_STAT_FAILED_GRAB_PAGE;
msg = "cl_page_find failed";
}
} else {
which = RA_STAT_WRONG_GRAB_PAGE;
msg = "g_c_p_n returned invalid page";
}
if (rc != 1)
unlock_page(vmpage);
page_cache_release(vmpage);
} else {
which = RA_STAT_FAILED_GRAB_PAGE;
msg = "g_c_p_n failed";
}
if (msg) {
ll_ra_stats_inc(inode, which);
CDEBUG(D_READA, "%s\n", msg);
}
return rc;
}
#define RIA_DEBUG(ria) \
CDEBUG(D_READA, "rs %lu re %lu ro %lu rl %lu rp %lu\n", \
ria->ria_start, ria->ria_end, ria->ria_stoff, ria->ria_length,\
ria->ria_pages)
/* Limit this to the blocksize instead of PTLRPC_BRW_MAX_SIZE, since we don't
* know what the actual RPC size is. If this needs to change, it makes more
* sense to tune the i_blkbits value for the file based on the OSTs it is
* striped over, rather than having a constant value for all files here.
*/
/* RAS_INCREASE_STEP should be (1UL << (inode->i_blkbits - PAGE_CACHE_SHIFT)).
* Temporarily set RAS_INCREASE_STEP to 1MB. After 4MB RPC is enabled
* by default, this should be adjusted corresponding with max_read_ahead_mb
* and max_read_ahead_per_file_mb otherwise the readahead budget can be used
* up quickly which will affect read performance significantly. See LU-2816
*/
#define RAS_INCREASE_STEP(inode) (ONE_MB_BRW_SIZE >> PAGE_CACHE_SHIFT)
static inline int stride_io_mode(struct ll_readahead_state *ras)
{
return ras->ras_consecutive_stride_requests > 1;
}
/* The function calculates how much pages will be read in
* [off, off + length], in such stride IO area,
* stride_offset = st_off, stride_length = st_len,
* stride_pages = st_pgs
*
* |------------------|*****|------------------|*****|------------|*****|....
* st_off
* |--- st_pgs ---|
* |----- st_len -----|
*
* How many pages it should read in such pattern
* |-------------------------------------------------------------|
* off
* |<------ length ------->|
*
* = |<----->| + |-------------------------------------| + |---|
* start_left st_pgs * i end_left
*/
static unsigned long
stride_pg_count(pgoff_t st_off, unsigned long st_len, unsigned long st_pgs,
unsigned long off, unsigned long length)
{
__u64 start = off > st_off ? off - st_off : 0;
__u64 end = off + length > st_off ? off + length - st_off : 0;
unsigned long start_left = 0;
unsigned long end_left = 0;
unsigned long pg_count;
if (st_len == 0 || length == 0 || end == 0)
return length;
start_left = do_div(start, st_len);
if (start_left < st_pgs)
start_left = st_pgs - start_left;
else
start_left = 0;
end_left = do_div(end, st_len);
if (end_left > st_pgs)
end_left = st_pgs;
CDEBUG(D_READA, "start %llu, end %llu start_left %lu end_left %lu\n",
start, end, start_left, end_left);
if (start == end)
pg_count = end_left - (st_pgs - start_left);
else
pg_count = start_left + st_pgs * (end - start - 1) + end_left;
CDEBUG(D_READA, "st_off %lu, st_len %lu st_pgs %lu off %lu length %lu pgcount %lu\n",
st_off, st_len, st_pgs, off, length, pg_count);
return pg_count;
}
static int ria_page_count(struct ra_io_arg *ria)
{
__u64 length = ria->ria_end >= ria->ria_start ?
ria->ria_end - ria->ria_start + 1 : 0;
return stride_pg_count(ria->ria_stoff, ria->ria_length,
ria->ria_pages, ria->ria_start,
length);
}
/*Check whether the index is in the defined ra-window */
static int ras_inside_ra_window(unsigned long idx, struct ra_io_arg *ria)
{
/* If ria_length == ria_pages, it means non-stride I/O mode,
* idx should always inside read-ahead window in this case
* For stride I/O mode, just check whether the idx is inside
* the ria_pages.
*/
return ria->ria_length == 0 || ria->ria_length == ria->ria_pages ||
(idx >= ria->ria_stoff && (idx - ria->ria_stoff) %
ria->ria_length < ria->ria_pages);
}
static int ll_read_ahead_pages(const struct lu_env *env,
struct cl_io *io, struct cl_page_list *queue,
struct ra_io_arg *ria,
unsigned long *reserved_pages,
unsigned long *ra_end)
{
int rc, count = 0;
bool stride_ria;
pgoff_t page_idx;
pgoff_t max_index = 0;
LASSERT(ria);
RIA_DEBUG(ria);
stride_ria = ria->ria_length > ria->ria_pages && ria->ria_pages > 0;
for (page_idx = ria->ria_start;
page_idx <= ria->ria_end && *reserved_pages > 0; page_idx++) {
if (ras_inside_ra_window(page_idx, ria)) {
/* If the page is inside the read-ahead window*/
rc = ll_read_ahead_page(env, io, queue,
page_idx, &max_index);
if (rc == 1) {
(*reserved_pages)--;
count++;
} else if (rc == -ENOLCK)
break;
} else if (stride_ria) {
/* If it is not in the read-ahead window, and it is
* read-ahead mode, then check whether it should skip
* the stride gap
*/
pgoff_t offset;
/* FIXME: This assertion only is valid when it is for
* forward read-ahead, it will be fixed when backward
* read-ahead is implemented
*/
LASSERTF(page_idx > ria->ria_stoff, "Invalid page_idx %lu rs %lu re %lu ro %lu rl %lu rp %lu\n",
page_idx,
ria->ria_start, ria->ria_end, ria->ria_stoff,
ria->ria_length, ria->ria_pages);
offset = page_idx - ria->ria_stoff;
offset = offset % (ria->ria_length);
if (offset > ria->ria_pages) {
page_idx += ria->ria_length - offset;
CDEBUG(D_READA, "i %lu skip %lu\n", page_idx,
ria->ria_length - offset);
continue;
}
}
}
*ra_end = page_idx;
return count;
}
int ll_readahead(const struct lu_env *env, struct cl_io *io,
struct cl_page_list *queue, struct ll_readahead_state *ras,
bool hit)
{
struct vvp_io *vio = vvp_env_io(env);
struct ll_thread_info *lti = ll_env_info(env);
struct cl_attr *attr = vvp_env_thread_attr(env);
unsigned long start = 0, end = 0, reserved;
unsigned long ra_end, len, mlen = 0;
struct inode *inode;
struct ra_io_arg *ria = <i->lti_ria;
struct cl_object *clob;
int ret = 0;
__u64 kms;
clob = io->ci_obj;
inode = vvp_object_inode(clob);
memset(ria, 0, sizeof(*ria));
cl_object_attr_lock(clob);
ret = cl_object_attr_get(env, clob, attr);
cl_object_attr_unlock(clob);
if (ret != 0)
return ret;
kms = attr->cat_kms;
if (kms == 0) {
ll_ra_stats_inc(inode, RA_STAT_ZERO_LEN);
return 0;
}
spin_lock(&ras->ras_lock);
/* Enlarge the RA window to encompass the full read */
if (vio->vui_ra_valid &&
ras->ras_window_start + ras->ras_window_len <
vio->vui_ra_start + vio->vui_ra_count) {
ras->ras_window_len = vio->vui_ra_start + vio->vui_ra_count -
ras->ras_window_start;
}
/* Reserve a part of the read-ahead window that we'll be issuing */
if (ras->ras_window_len) {
start = ras->ras_next_readahead;
end = ras->ras_window_start + ras->ras_window_len - 1;
}
if (end != 0) {
unsigned long rpc_boundary;
/*
* Align RA window to an optimal boundary.
*
* XXX This would be better to align to cl_max_pages_per_rpc
* instead of PTLRPC_MAX_BRW_PAGES, because the RPC size may
* be aligned to the RAID stripe size in the future and that
* is more important than the RPC size.
*/
/* Note: we only trim the RPC, instead of extending the RPC
* to the boundary, so to avoid reading too much pages during
* random reading.
*/
rpc_boundary = (end + 1) & (~(PTLRPC_MAX_BRW_PAGES - 1));
if (rpc_boundary > 0)
rpc_boundary--;
if (rpc_boundary > start)
end = rpc_boundary;
/* Truncate RA window to end of file */
end = min(end, (unsigned long)((kms - 1) >> PAGE_CACHE_SHIFT));
ras->ras_next_readahead = max(end, end + 1);
RAS_CDEBUG(ras);
}
ria->ria_start = start;
ria->ria_end = end;
/* If stride I/O mode is detected, get stride window*/
if (stride_io_mode(ras)) {
ria->ria_stoff = ras->ras_stride_offset;
ria->ria_length = ras->ras_stride_length;
ria->ria_pages = ras->ras_stride_pages;
}
spin_unlock(&ras->ras_lock);
if (end == 0) {
ll_ra_stats_inc(inode, RA_STAT_ZERO_WINDOW);
return 0;
}
len = ria_page_count(ria);
if (len == 0) {
ll_ra_stats_inc(inode, RA_STAT_ZERO_WINDOW);
return 0;
}
CDEBUG(D_READA, DFID ": ria: %lu/%lu, bead: %lu/%lu, hit: %d\n",
PFID(lu_object_fid(&clob->co_lu)),
ria->ria_start, ria->ria_end,
vio->vui_ra_valid ? vio->vui_ra_start : 0,
vio->vui_ra_valid ? vio->vui_ra_count : 0,
hit);
/* at least to extend the readahead window to cover current read */
if (!hit && vio->vui_ra_valid &&
vio->vui_ra_start + vio->vui_ra_count > ria->ria_start) {
/* to the end of current read window. */
mlen = vio->vui_ra_start + vio->vui_ra_count - ria->ria_start;
/* trim to RPC boundary */
start = ria->ria_start & (PTLRPC_MAX_BRW_PAGES - 1);
mlen = min(mlen, PTLRPC_MAX_BRW_PAGES - start);
}
reserved = ll_ra_count_get(ll_i2sbi(inode), ria, len, mlen);
if (reserved < len)
ll_ra_stats_inc(inode, RA_STAT_MAX_IN_FLIGHT);
CDEBUG(D_READA, "reserved pages %lu/%lu/%lu, ra_cur %d, ra_max %lu\n",
reserved, len, mlen,
atomic_read(&ll_i2sbi(inode)->ll_ra_info.ra_cur_pages),
ll_i2sbi(inode)->ll_ra_info.ra_max_pages);
ret = ll_read_ahead_pages(env, io, queue, ria, &reserved, &ra_end);
if (reserved != 0)
ll_ra_count_put(ll_i2sbi(inode), reserved);
if (ra_end == end + 1 && ra_end == (kms >> PAGE_CACHE_SHIFT))
ll_ra_stats_inc(inode, RA_STAT_EOF);
/* if we didn't get to the end of the region we reserved from
* the ras we need to go back and update the ras so that the
* next read-ahead tries from where we left off. we only do so
* if the region we failed to issue read-ahead on is still ahead
* of the app and behind the next index to start read-ahead from
*/
CDEBUG(D_READA, "ra_end %lu end %lu stride end %lu\n",
ra_end, end, ria->ria_end);
if (ra_end != end + 1) {
ll_ra_stats_inc(inode, RA_STAT_FAILED_REACH_END);
spin_lock(&ras->ras_lock);
if (ra_end < ras->ras_next_readahead &&
index_in_window(ra_end, ras->ras_window_start, 0,
ras->ras_window_len)) {
ras->ras_next_readahead = ra_end;
RAS_CDEBUG(ras);
}
spin_unlock(&ras->ras_lock);
}
return ret;
}
static void ras_set_start(struct inode *inode, struct ll_readahead_state *ras,
unsigned long index)
{
ras->ras_window_start = index & (~(RAS_INCREASE_STEP(inode) - 1));
}
/* called with the ras_lock held or from places where it doesn't matter */
static void ras_reset(struct inode *inode, struct ll_readahead_state *ras,
unsigned long index)
{
ras->ras_last_readpage = index;
ras->ras_consecutive_requests = 0;
ras->ras_consecutive_pages = 0;
ras->ras_window_len = 0;
ras_set_start(inode, ras, index);
ras->ras_next_readahead = max(ras->ras_window_start, index);
RAS_CDEBUG(ras);
}
/* called with the ras_lock held or from places where it doesn't matter */
static void ras_stride_reset(struct ll_readahead_state *ras)
{
ras->ras_consecutive_stride_requests = 0;
ras->ras_stride_length = 0;
ras->ras_stride_pages = 0;
RAS_CDEBUG(ras);
}
void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras)
{
spin_lock_init(&ras->ras_lock);
ras_reset(inode, ras, 0);
ras->ras_requests = 0;
}
/*
* Check whether the read request is in the stride window.
* If it is in the stride window, return 1, otherwise return 0.
*/
static int index_in_stride_window(struct ll_readahead_state *ras,
unsigned long index)
{
unsigned long stride_gap;
if (ras->ras_stride_length == 0 || ras->ras_stride_pages == 0 ||
ras->ras_stride_pages == ras->ras_stride_length)
return 0;
stride_gap = index - ras->ras_last_readpage - 1;
/* If it is contiguous read */
if (stride_gap == 0)
return ras->ras_consecutive_pages + 1 <= ras->ras_stride_pages;
/* Otherwise check the stride by itself */
return (ras->ras_stride_length - ras->ras_stride_pages) == stride_gap &&
ras->ras_consecutive_pages == ras->ras_stride_pages;
}
static void ras_update_stride_detector(struct ll_readahead_state *ras,
unsigned long index)
{
unsigned long stride_gap = index - ras->ras_last_readpage - 1;
if (!stride_io_mode(ras) && (stride_gap != 0 ||
ras->ras_consecutive_stride_requests == 0)) {
ras->ras_stride_pages = ras->ras_consecutive_pages;
ras->ras_stride_length = stride_gap+ras->ras_consecutive_pages;
}
LASSERT(ras->ras_request_index == 0);
LASSERT(ras->ras_consecutive_stride_requests == 0);
if (index <= ras->ras_last_readpage) {
/*Reset stride window for forward read*/
ras_stride_reset(ras);
return;
}
ras->ras_stride_pages = ras->ras_consecutive_pages;
ras->ras_stride_length = stride_gap+ras->ras_consecutive_pages;
RAS_CDEBUG(ras);
return;
}
/* Stride Read-ahead window will be increased inc_len according to
* stride I/O pattern
*/
static void ras_stride_increase_window(struct ll_readahead_state *ras,
struct ll_ra_info *ra,
unsigned long inc_len)
{
unsigned long left, step, window_len;
unsigned long stride_len;
LASSERT(ras->ras_stride_length > 0);
LASSERTF(ras->ras_window_start + ras->ras_window_len
>= ras->ras_stride_offset, "window_start %lu, window_len %lu stride_offset %lu\n",
ras->ras_window_start,
ras->ras_window_len, ras->ras_stride_offset);
stride_len = ras->ras_window_start + ras->ras_window_len -
ras->ras_stride_offset;
left = stride_len % ras->ras_stride_length;
window_len = ras->ras_window_len - left;
if (left < ras->ras_stride_pages)
left += inc_len;
else
left = ras->ras_stride_pages + inc_len;
LASSERT(ras->ras_stride_pages != 0);
step = left / ras->ras_stride_pages;
left %= ras->ras_stride_pages;
window_len += step * ras->ras_stride_length + left;
if (stride_pg_count(ras->ras_stride_offset, ras->ras_stride_length,
ras->ras_stride_pages, ras->ras_stride_offset,
window_len) <= ra->ra_max_pages_per_file)
ras->ras_window_len = window_len;
RAS_CDEBUG(ras);
}
static void ras_increase_window(struct inode *inode,
struct ll_readahead_state *ras,
struct ll_ra_info *ra)
{
/* The stretch of ra-window should be aligned with max rpc_size
* but current clio architecture does not support retrieve such
* information from lower layer. FIXME later
*/
if (stride_io_mode(ras))
ras_stride_increase_window(ras, ra, RAS_INCREASE_STEP(inode));
else
ras->ras_window_len = min(ras->ras_window_len +
RAS_INCREASE_STEP(inode),
ra->ra_max_pages_per_file);
}
void ras_update(struct ll_sb_info *sbi, struct inode *inode,
struct ll_readahead_state *ras, unsigned long index,
unsigned hit)
{
struct ll_ra_info *ra = &sbi->ll_ra_info;
int zero = 0, stride_detect = 0, ra_miss = 0;
spin_lock(&ras->ras_lock);
ll_ra_stats_inc_sbi(sbi, hit ? RA_STAT_HIT : RA_STAT_MISS);
/* reset the read-ahead window in two cases. First when the app seeks
* or reads to some other part of the file. Secondly if we get a
* read-ahead miss that we think we've previously issued. This can
* be a symptom of there being so many read-ahead pages that the VM is
* reclaiming it before we get to it.
*/
if (!index_in_window(index, ras->ras_last_readpage, 8, 8)) {
zero = 1;
ll_ra_stats_inc_sbi(sbi, RA_STAT_DISTANT_READPAGE);
} else if (!hit && ras->ras_window_len &&
index < ras->ras_next_readahead &&
index_in_window(index, ras->ras_window_start, 0,
ras->ras_window_len)) {
ra_miss = 1;
ll_ra_stats_inc_sbi(sbi, RA_STAT_MISS_IN_WINDOW);
}
/* On the second access to a file smaller than the tunable
* ra_max_read_ahead_whole_pages trigger RA on all pages in the
* file up to ra_max_pages_per_file. This is simply a best effort
* and only occurs once per open file. Normal RA behavior is reverted
* to for subsequent IO. The mmap case does not increment
* ras_requests and thus can never trigger this behavior.
*/
if (ras->ras_requests == 2 && !ras->ras_request_index) {
__u64 kms_pages;
kms_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
PAGE_CACHE_SHIFT;
CDEBUG(D_READA, "kmsp %llu mwp %lu mp %lu\n", kms_pages,
ra->ra_max_read_ahead_whole_pages, ra->ra_max_pages_per_file);
if (kms_pages &&
kms_pages <= ra->ra_max_read_ahead_whole_pages) {
ras->ras_window_start = 0;
ras->ras_last_readpage = 0;
ras->ras_next_readahead = 0;
ras->ras_window_len = min(ra->ra_max_pages_per_file,
ra->ra_max_read_ahead_whole_pages);
goto out_unlock;
}
}
if (zero) {
/* check whether it is in stride I/O mode*/
if (!index_in_stride_window(ras, index)) {
if (ras->ras_consecutive_stride_requests == 0 &&
ras->ras_request_index == 0) {
ras_update_stride_detector(ras, index);
ras->ras_consecutive_stride_requests++;
} else {
ras_stride_reset(ras);
}
ras_reset(inode, ras, index);
ras->ras_consecutive_pages++;
goto out_unlock;
} else {
ras->ras_consecutive_pages = 0;
ras->ras_consecutive_requests = 0;
if (++ras->ras_consecutive_stride_requests > 1)
stride_detect = 1;
RAS_CDEBUG(ras);
}
} else {
if (ra_miss) {
if (index_in_stride_window(ras, index) &&
stride_io_mode(ras)) {
/*If stride-RA hit cache miss, the stride dector
*will not be reset to avoid the overhead of
*redetecting read-ahead mode
*/
if (index != ras->ras_last_readpage + 1)
ras->ras_consecutive_pages = 0;
ras_reset(inode, ras, index);
RAS_CDEBUG(ras);
} else {
/* Reset both stride window and normal RA
* window
*/
ras_reset(inode, ras, index);
ras->ras_consecutive_pages++;
ras_stride_reset(ras);
goto out_unlock;
}
} else if (stride_io_mode(ras)) {
/* If this is contiguous read but in stride I/O mode
* currently, check whether stride step still is valid,
* if invalid, it will reset the stride ra window
*/
if (!index_in_stride_window(ras, index)) {
/* Shrink stride read-ahead window to be zero */
ras_stride_reset(ras);
ras->ras_window_len = 0;
ras->ras_next_readahead = index;
}
}
}
ras->ras_consecutive_pages++;
ras->ras_last_readpage = index;
ras_set_start(inode, ras, index);
if (stride_io_mode(ras)) {
/* Since stride readahead is sensitive to the offset
* of read-ahead, so we use original offset here,
* instead of ras_window_start, which is RPC aligned
*/
ras->ras_next_readahead = max(index, ras->ras_next_readahead);
} else {
if (ras->ras_next_readahead < ras->ras_window_start)
ras->ras_next_readahead = ras->ras_window_start;
if (!hit)
ras->ras_next_readahead = index + 1;
}
RAS_CDEBUG(ras);
/* Trigger RA in the mmap case where ras_consecutive_requests
* is not incremented and thus can't be used to trigger RA
*/
if (!ras->ras_window_len && ras->ras_consecutive_pages == 4) {
ras->ras_window_len = RAS_INCREASE_STEP(inode);
goto out_unlock;
}
/* Initially reset the stride window offset to next_readahead*/
if (ras->ras_consecutive_stride_requests == 2 && stride_detect) {
/**
* Once stride IO mode is detected, next_readahead should be
* reset to make sure next_readahead > stride offset
*/
ras->ras_next_readahead = max(index, ras->ras_next_readahead);
ras->ras_stride_offset = index;
ras->ras_window_len = RAS_INCREASE_STEP(inode);
}
/* The initial ras_window_len is set to the request size. To avoid
* uselessly reading and discarding pages for random IO the window is
* only increased once per consecutive request received. */
if ((ras->ras_consecutive_requests > 1 || stride_detect) &&
!ras->ras_request_index)
ras_increase_window(inode, ras, ra);
out_unlock:
RAS_CDEBUG(ras);
ras->ras_request_index++;
spin_unlock(&ras->ras_lock);
return;
}
int ll_writepage(struct page *vmpage, struct writeback_control *wbc)
{
struct inode *inode = vmpage->mapping->host;
struct ll_inode_info *lli = ll_i2info(inode);
struct lu_env *env;
struct cl_io *io;
struct cl_page *page;
struct cl_object *clob;
struct cl_env_nest nest;
bool redirtied = false;
bool unlocked = false;
int result;
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
LASSERT(ll_i2dtexp(inode));
env = cl_env_nested_get(&nest);
if (IS_ERR(env)) {
result = PTR_ERR(env);
goto out;
}
clob = ll_i2info(inode)->lli_clob;
LASSERT(clob);
io = vvp_env_thread_io(env);
io->ci_obj = clob;
io->ci_ignore_layout = 1;
result = cl_io_init(env, io, CIT_MISC, clob);
if (result == 0) {
page = cl_page_find(env, clob, vmpage->index,
vmpage, CPT_CACHEABLE);
if (!IS_ERR(page)) {
lu_ref_add(&page->cp_reference, "writepage",
current);
cl_page_assume(env, io, page);
result = cl_page_flush(env, io, page);
if (result != 0) {
/*
* Re-dirty page on error so it retries write,
* but not in case when IO has actually
* occurred and completed with an error.
*/
if (!PageError(vmpage)) {
redirty_page_for_writepage(wbc, vmpage);
result = 0;
redirtied = true;
}
}
cl_page_disown(env, io, page);
unlocked = true;
lu_ref_del(&page->cp_reference,
"writepage", current);
cl_page_put(env, page);
} else {
result = PTR_ERR(page);
}
}
cl_io_fini(env, io);
if (redirtied && wbc->sync_mode == WB_SYNC_ALL) {
loff_t offset = cl_offset(clob, vmpage->index);
/* Flush page failed because the extent is being written out.
* Wait for the write of extent to be finished to avoid
* breaking kernel which assumes ->writepage should mark
* PageWriteback or clean the page.
*/
result = cl_sync_file_range(inode, offset,
offset + PAGE_CACHE_SIZE - 1,
CL_FSYNC_LOCAL, 1);
if (result > 0) {
/* actually we may have written more than one page.
* decreasing this page because the caller will count
* it.
*/
wbc->nr_to_write -= result - 1;
result = 0;
}
}
cl_env_nested_put(&nest, env);
goto out;
out:
if (result < 0) {
if (!lli->lli_async_rc)
lli->lli_async_rc = result;
SetPageError(vmpage);
if (!unlocked)
unlock_page(vmpage);
}
return result;
}
int ll_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
struct inode *inode = mapping->host;
struct ll_sb_info *sbi = ll_i2sbi(inode);
loff_t start;
loff_t end;
enum cl_fsync_mode mode;
int range_whole = 0;
int result;
int ignore_layout = 0;
if (wbc->range_cyclic) {
start = mapping->writeback_index << PAGE_CACHE_SHIFT;
end = OBD_OBJECT_EOF;
} else {
start = wbc->range_start;
end = wbc->range_end;
if (end == LLONG_MAX) {
end = OBD_OBJECT_EOF;
range_whole = start == 0;
}
}
mode = CL_FSYNC_NONE;
if (wbc->sync_mode == WB_SYNC_ALL)
mode = CL_FSYNC_LOCAL;
if (sbi->ll_umounting)
/* if the mountpoint is being umounted, all pages have to be
* evicted to avoid hitting LBUG when truncate_inode_pages()
* is called later on.
*/
ignore_layout = 1;
result = cl_sync_file_range(inode, start, end, mode, ignore_layout);
if (result > 0) {
wbc->nr_to_write -= result;
result = 0;
}
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) {
if (end == OBD_OBJECT_EOF)
end = i_size_read(inode);
mapping->writeback_index = (end >> PAGE_CACHE_SHIFT) + 1;
}
return result;
}
int ll_readpage(struct file *file, struct page *vmpage)
{
struct ll_cl_context *lcc;
int result;
lcc = ll_cl_init(file, vmpage);
if (!IS_ERR(lcc)) {
struct lu_env *env = lcc->lcc_env;
struct cl_io *io = lcc->lcc_io;
struct cl_page *page = lcc->lcc_page;
LASSERT(page->cp_type == CPT_CACHEABLE);
if (likely(!PageUptodate(vmpage))) {
cl_page_assume(env, io, page);
result = cl_io_read_page(env, io, page);
} else {
/* Page from a non-object file. */
unlock_page(vmpage);
result = 0;
}
ll_cl_fini(lcc);
} else {
unlock_page(vmpage);
result = PTR_ERR(lcc);
}
return result;
}
int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
struct cl_page *page, enum cl_req_type crt)
{
struct cl_2queue *queue;
int result;
LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
queue = &io->ci_queue;
cl_2queue_init_page(queue, page);
result = cl_io_submit_sync(env, io, crt, queue, 0);
LASSERT(cl_page_is_owned(page, io));
if (crt == CRT_READ)
/*
* in CRT_WRITE case page is left locked even in case of
* error.
*/
cl_page_list_disown(env, io, &queue->c2_qin);
cl_2queue_fini(env, queue);
return result;
}
|