-
Notifications
You must be signed in to change notification settings - Fork 149
/
Copy pathmf_iocache.cc
1710 lines (1492 loc) · 58.6 KB
/
mf_iocache.cc
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
/* Copyright (c) 2000, 2024, Oracle and/or its affiliates.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License, version 2.0,
as published by the Free Software Foundation.
This program is designed to work with certain software (including
but not limited to OpenSSL) that is licensed under separate terms,
as designated in a particular file or component or in included license
documentation. The authors of MySQL hereby grant you an additional
permission to link the program and your derivative works with the
separately licensed software that they have either included with
the program or referenced in the documentation.
Without limiting anything contained in the foregoing, this file,
which is part of C Driver for MySQL (Connector/C), is also subject to the
Universal FOSS Exception, version 1.0, a copy of which can be found at
http://oss.oracle.com/licenses/universal-foss-exception.
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 2.0, for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */
/**
@file mysys/mf_iocache.cc
Cashing of files with only does (sequential) read or writes of fixed-
length records. A read isn't allowed to go over file-length. A read is ok
if it ends at file-length and next read can try to read after file-length
(and get a EOF-error).
Possibly use of asyncronic io.
macros for read and writes for faster io.
Used instead of FILE when reading or writing whole files.
This code makes mf_rec_cache obsolete (currently only used by ISAM).
One can change info->pos_in_file to a higher value to skip bytes in the file
if also info->read_pos is set to info->read_end. If called through
open_cached_file(), then the temporary file will only be created if a write
exceeds the file buffer or if one calls my_b_flush_io_cache().
If one uses SEQ_READ_APPEND, then two buffers are allocated, one for
reading and another for writing. Reads are first done from disk and
then done from the write buffer. This is an efficient way to read
from a log file when one is writing to it at the same time.
For this to work, the file has to be opened in append mode!
Note that when one uses SEQ_READ_APPEND, one MUST write using
my_b_append ! This is needed because we need to lock the mutex
every time we access the write buffer.
TODO:
When one SEQ_READ_APPEND and we are reading and writing at the same time,
each time the write buffer gets full and it's written to disk, we will
always do a disk read to read a part of the buffer from disk to the
read buffer.
This should be fixed so that when we do a my_b_flush_io_cache() and
we have been reading the write buffer, we should transfer the rest of the
write buffer to the read buffer before we start to reuse it.
*/
#include <errno.h>
#include <fcntl.h>
#include <math.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <algorithm>
#include "m_string.h"
#include "my_byteorder.h"
#include "my_compiler.h"
#include "my_dbug.h"
#include "my_inttypes.h"
#include "my_io.h"
#include "my_macros.h"
#include "my_sys.h"
#include "my_systime.h"
#include "my_thread_local.h"
#include "mysql/components/services/bits/psi_bits.h"
#include "mysql/psi/mysql_cond.h"
#include "mysql/psi/mysql_file.h"
#include "mysql/psi/mysql_mutex.h"
#include "mysql/service_mysql_alloc.h"
#include "mysys/mysys_priv.h"
#include "thr_mutex.h"
PSI_file_key key_file_io_cache;
#ifndef NDEBUG
bool binlog_cache_temporary_file_is_encrypted = false;
#endif
#define lock_append_buffer(info) mysql_mutex_lock(&(info)->append_buffer_lock)
#define unlock_append_buffer(info) \
mysql_mutex_unlock(&(info)->append_buffer_lock)
#define IO_ROUND_UP(X) (((X) + IO_SIZE - 1) & ~(IO_SIZE - 1))
#define IO_ROUND_DN(X) ((X) & ~(IO_SIZE - 1))
/*
Setup internal pointers inside IO_CACHE
SYNOPSIS
setup_io_cache()
info IO_CACHE handler
NOTES
This is called on automatically on init or reinit of IO_CACHE
It must be called externally if one moves or copies an IO_CACHE
object.
*/
void setup_io_cache(IO_CACHE *info) {
/* Ensure that my_b_tell() and my_b_bytes_in_cache works */
if (info->type == WRITE_CACHE) {
info->current_pos = &info->write_pos;
info->current_end = &info->write_end;
} else {
info->current_pos = &info->read_pos;
info->current_end = &info->read_end;
}
}
static void init_functions(IO_CACHE *info) {
enum cache_type type = info->type;
switch (type) {
case READ_NET:
/*
Must be initialized by the caller. The problem is that
_my_b_net_read has to be defined in sql directory because of
the dependency on THD, and therefore cannot be visible to
programs that link against mysys but know nothing about THD, such
as myisamchk
*/
break;
case SEQ_READ_APPEND:
info->read_function = _my_b_seq_read;
info->write_function = nullptr; /* Force a core if used */
break;
default:
info->read_function = info->share ? _my_b_read_r : _my_b_read;
info->write_function = _my_b_write;
}
setup_io_cache(info);
}
/*
Initialize an IO_CACHE object
SYNOPSIS
init_io_cache_ext()
info cache handler to initialize
file File that should be associated to to the handler
If == -1 then real_open_cached_file()
will be called when it's time to open file.
cachesize Size of buffer to allocate for read/write
If == 0 then use my_default_record_cache_size
type Type of cache
seek_offset Where cache should start reading/writing
use_async_io Set to 1 of we should use async_io (if available)
cache_myflags Bitmap of different flags
MY_WME | MY_FAE | MY_NABP | MY_FNABP |
MY_DONT_CHECK_FILESIZE
file_key Instrumented file key for temporary cache file
RETURN
0 ok
# error
*/
int init_io_cache_ext(IO_CACHE *info, File file, size_t cachesize,
enum cache_type type, my_off_t seek_offset,
bool use_async_io, myf cache_myflags,
PSI_file_key file_key) {
size_t min_cache;
my_off_t pos;
my_off_t end_of_file = ~(my_off_t)0;
DBUG_TRACE;
DBUG_PRINT("enter", ("cache: %p type: %d pos: %ld", info, (int)type,
(ulong)seek_offset));
DBUG_EXECUTE_IF("simulate_init_io_cache_failure", return 1;);
info->file = file;
info->file_key = file_key;
info->type = TYPE_NOT_SET; /* Don't set it until mutex are created */
info->pos_in_file = seek_offset;
info->pre_close = info->pre_read = info->post_read = nullptr;
info->arg = nullptr;
info->alloced_buffer = false;
info->buffer = nullptr;
info->seek_not_done = false;
if (file >= 0) {
pos = mysql_file_tell(file, MYF(0));
if ((pos == (my_off_t)-1) && (my_errno() == ESPIPE)) {
/*
This kind of object doesn't support seek() or tell(). Don't set a
flag that will make us again try to seek() later and fail.
*/
info->seek_not_done = false;
/*
Additionally, if we're supposed to start somewhere other than the
the beginning of whatever this file is, then somebody made a bad
assumption.
*/
assert(seek_offset == 0);
} else
info->seek_not_done = (seek_offset != pos);
}
info->disk_writes = 0;
info->share = nullptr;
if (!cachesize && !(cachesize = my_default_record_cache_size))
return 1; /* No cache requested */
min_cache = use_async_io ? IO_SIZE * 4 : IO_SIZE * 2;
if (type == READ_CACHE ||
type == SEQ_READ_APPEND) { /* Assume file isn't growing */
if (!(cache_myflags & MY_DONT_CHECK_FILESIZE)) {
/* Calculate end of file to avoid allocating oversized buffers */
end_of_file = mysql_encryption_file_seek(info, 0L, MY_SEEK_END, MYF(0));
/* Need to reset seek_not_done now that we just did a seek. */
info->seek_not_done = !(end_of_file == seek_offset);
if (end_of_file < seek_offset) end_of_file = seek_offset;
/* Trim cache size if the file is very small */
if ((my_off_t)cachesize > end_of_file - seek_offset + IO_SIZE * 2 - 1) {
cachesize = (size_t)(end_of_file - seek_offset) + IO_SIZE * 2 - 1;
use_async_io = false; /* No need to use async */
}
}
}
cache_myflags &= ~MY_DONT_CHECK_FILESIZE;
if (type != READ_NET && type != WRITE_NET) {
/* Retry allocating memory in smaller blocks until we get one */
cachesize = ((cachesize + min_cache - 1) & ~(min_cache - 1));
for (;;) {
size_t buffer_block;
/*
Unset MY_WAIT_IF_FULL bit if it is set, to prevent conflict with
MY_ZEROFILL.
*/
myf flags = (myf)(cache_myflags & ~(MY_WME | MY_WAIT_IF_FULL));
if (cachesize < min_cache) cachesize = min_cache;
buffer_block = cachesize;
if (type == SEQ_READ_APPEND) buffer_block *= 2;
if (cachesize == min_cache) flags |= (myf)MY_WME;
if ((info->buffer = (uchar *)my_malloc(key_memory_IO_CACHE, buffer_block,
flags)) != nullptr) {
info->write_buffer = info->buffer;
if (type == SEQ_READ_APPEND)
info->write_buffer = info->buffer + cachesize;
info->alloced_buffer = true;
break; /* Enough memory found */
}
if (cachesize == min_cache) return 2; /* Can't alloc cache */
/* Try with less memory */
cachesize = (cachesize * 3 / 4 & ~(min_cache - 1));
}
}
DBUG_PRINT("info", ("init_io_cache: cachesize = %lu", (ulong)cachesize));
info->read_length = info->buffer_length = cachesize;
info->myflags = cache_myflags & ~(MY_NABP | MY_FNABP);
info->request_pos = info->read_pos = info->write_pos = info->buffer;
if (type == SEQ_READ_APPEND) {
info->append_read_pos = info->write_pos = info->write_buffer;
info->write_end = info->write_buffer + info->buffer_length;
mysql_mutex_init(key_IO_CACHE_append_buffer_lock, &info->append_buffer_lock,
MY_MUTEX_INIT_FAST);
}
#if defined(SAFE_MUTEX)
else {
/* Clear mutex so that safe_mutex will notice that it's not initialized */
new (&info->append_buffer_lock) mysql_mutex_t();
}
#endif
if (type == WRITE_CACHE)
info->write_end =
info->buffer + info->buffer_length - (seek_offset & (IO_SIZE - 1));
else
info->read_end = info->buffer; /* Nothing in cache */
/* End_of_file may be changed by user later */
info->end_of_file = end_of_file;
info->error = 0;
info->type = type;
init_functions(info);
return 0;
} /* init_io_cache_ext */
/*
Initialize an IO_CACHE object
SYNOPSIS
init_io_cache() - Wrapper for init_io_cache_ext()
NOTE
This function should be used if the IO_CACHE tempfile is not instrumented.
However, if a PSI key other than PSI_NOT_INSTRUMENTED is set explicitly,
that key will be used for instrumentation.
*/
int init_io_cache(IO_CACHE *info, File file, size_t cachesize,
enum cache_type type, my_off_t seek_offset, bool use_async_io,
myf cache_myflags) {
return init_io_cache_ext(
info, file, cachesize, type, seek_offset, use_async_io, cache_myflags,
info->file_key != PSI_NOT_INSTRUMENTED ? info->file_key
: key_file_io_cache);
}
/*
Use this to reset cache to re-start reading or to change the type
between READ_CACHE <-> WRITE_CACHE
If we are doing a reinit of a cache where we have the start of the file
in the cache, we are reusing this memory without flushing it to disk.
*/
bool reinit_io_cache(IO_CACHE *info, enum cache_type type, my_off_t seek_offset,
bool use_async_io [[maybe_unused]], bool clear_cache) {
DBUG_TRACE;
DBUG_PRINT("enter", ("cache: %p type: %d seek_offset: %lu clear_cache: %d",
info, type, (ulong)seek_offset, (int)clear_cache));
/* One can't do reinit with the following types */
assert(type != READ_NET && info->type != READ_NET && type != WRITE_NET &&
info->type != WRITE_NET && type != SEQ_READ_APPEND &&
info->type != SEQ_READ_APPEND);
/* If the whole file is in memory, avoid flushing to disk */
if (!clear_cache && seek_offset >= info->pos_in_file &&
seek_offset <= my_b_tell(info)) {
/* Reuse current buffer without flushing it to disk */
uchar *pos;
if (info->type == WRITE_CACHE && type == READ_CACHE) {
info->read_end = info->write_pos;
info->end_of_file = my_b_tell(info);
/*
Trigger a new seek only if we have a valid
file handle.
*/
info->seek_not_done = (info->file != -1);
} else if (type == WRITE_CACHE) {
if (info->type == READ_CACHE) {
info->write_end = info->write_buffer + info->buffer_length;
info->seek_not_done = true;
}
info->end_of_file = ~(my_off_t)0;
}
pos = info->request_pos + (seek_offset - info->pos_in_file);
if (type == WRITE_CACHE)
info->write_pos = pos;
else
info->read_pos = pos;
} else {
/*
If we change from WRITE_CACHE to READ_CACHE, assume that everything
after the current positions should be ignored
*/
if (info->type == WRITE_CACHE && type == READ_CACHE)
info->end_of_file = my_b_tell(info);
/* flush cache if we want to reuse it */
if (!clear_cache && my_b_flush_io_cache(info, 1)) return true;
info->pos_in_file = seek_offset;
/* Better to do always do a seek */
info->seek_not_done = true;
info->request_pos = info->read_pos = info->write_pos = info->buffer;
if (type == READ_CACHE) {
info->read_end = info->buffer; /* Nothing in cache */
} else {
info->write_end =
(info->buffer + info->buffer_length - (seek_offset & (IO_SIZE - 1)));
info->end_of_file = ~(my_off_t)0;
}
}
info->type = type;
info->error = 0;
init_functions(info);
if (DBUG_EVALUATE_IF("fault_injection_reinit_io_cache", true, false))
return true;
return false;
} /* reinit_io_cache */
/*
Read buffered.
SYNOPSIS
_my_b_read()
info IO_CACHE pointer
Buffer Buffer to retrieve count bytes from file
Count Number of bytes to read into Buffer
NOTE
This function is only called from the my_b_read() macro when there
isn't enough characters in the buffer to satisfy the request.
WARNING
When changing this function, be careful with handling file offsets
(end-of_file, pos_in_file). Do not cast them to possibly smaller
types than my_off_t unless you can be sure that their value fits.
Same applies to differences of file offsets.
When changing this function, check _my_b_read_r(). It might need the
same change.
RETURN
0 we succeeded in reading all data
1 Error: couldn't read requested characters. In this case:
If info->error == -1, we got a read error.
Otherwise info->error contains the number of bytes in Buffer.
*/
int _my_b_read(IO_CACHE *info, uchar *Buffer, size_t Count) {
size_t length, diff_length, left_length, max_length;
my_off_t pos_in_file;
DBUG_TRACE;
/* If the buffer is not empty yet, copy what is available. */
if ((left_length = (size_t)(info->read_end - info->read_pos))) {
assert(Count >= left_length); /* User is not using my_b_read() */
memcpy(Buffer, info->read_pos, left_length);
Buffer += left_length;
Count -= left_length;
}
/* pos_in_file always point on where info->buffer was read */
pos_in_file = info->pos_in_file + (size_t)(info->read_end - info->buffer);
/*
Whenever a function which operates on IO_CACHE flushes/writes
some part of the IO_CACHE to disk it will set the property
"seek_not_done" to indicate this to other functions operating
on the IO_CACHE.
*/
if (info->seek_not_done) {
if ((mysql_encryption_file_seek(info, pos_in_file, MY_SEEK_SET, MYF(0)) !=
MY_FILEPOS_ERROR)) {
/* No error, reset seek_not_done flag. */
info->seek_not_done = false;
} else {
/*
If the seek failed and the error number is ESPIPE, it is because
info->file is a pipe or socket or FIFO. We never should have tried
to seek on that. See Bugs#25807 and #22828 for more info.
*/
assert(my_errno() != ESPIPE);
info->error = -1;
return 1;
}
}
/*
Calculate, how much we are within a IO_SIZE block. Ideally this
should be zero.
*/
diff_length = (size_t)(pos_in_file & (IO_SIZE - 1));
/*
If more than a block plus the rest of the current block is wanted,
we do read directly, without filling the buffer.
*/
if (Count >=
(size_t)(IO_SIZE +
(IO_SIZE - diff_length))) { /* Fill first intern buffer */
size_t read_length;
if (info->end_of_file <= pos_in_file) {
/* End of file. Return, what we did copy from the buffer. */
info->error = (int)left_length;
return 1;
}
/*
Crop the wanted count to a multiple of IO_SIZE and subtract,
what we did already read from a block. That way, the read will
end aligned with a block.
*/
length = (Count & (size_t) ~(IO_SIZE - 1)) - diff_length;
if ((read_length = mysql_encryption_file_read(info, Buffer, length,
info->myflags)) != length) {
/*
If we didn't get, what we wanted, we either return -1 for a read
error, or (it's end of file), how much we got in total.
*/
info->error =
(read_length == (size_t)-1 ? -1 : (int)(read_length + left_length));
return 1;
}
Count -= length;
Buffer += length;
pos_in_file += length;
left_length += length;
diff_length = 0;
}
/*
At this point, we want less than one and a partial block.
We will read a full cache, minus the number of bytes, we are
within a block already. So we will reach new alignment.
*/
max_length = info->read_length - diff_length;
/* We will not read past end of file. */
if (info->type != READ_FIFO && max_length > (info->end_of_file - pos_in_file))
max_length = (size_t)(info->end_of_file - pos_in_file);
/*
If there is nothing left to read,
we either are done, or we failed to fulfill the request.
Otherwise, we read max_length into the cache.
*/
if (!max_length) {
if (Count) {
/* We couldn't fulfil the request. Return, how much we got. */
info->error = (int)left_length;
return 1;
}
length = 0; /* Didn't read any chars */
} else if ((length = mysql_encryption_file_read(
info, info->buffer, max_length, info->myflags)) < Count ||
length == (size_t)-1) {
/*
We got an read error, or less than requested (end of file).
If not a read error, copy, what we got.
*/
if (length != (size_t)-1) memcpy(Buffer, info->buffer, length);
info->pos_in_file = pos_in_file;
/* For a read error, return -1, otherwise, what we got in total. */
info->error = length == (size_t)-1 ? -1 : (int)(length + left_length);
info->read_pos = info->read_end = info->buffer;
return 1;
}
/*
Count is the remaining number of bytes requested.
length is the amount of data in the cache.
Read Count bytes from the cache.
*/
info->read_pos = info->buffer + Count;
info->read_end = info->buffer + length;
info->pos_in_file = pos_in_file;
memcpy(Buffer, info->buffer, Count);
return 0;
}
/*
Prepare IO_CACHE for shared use.
SYNOPSIS
init_io_cache_share()
read_cache A read cache. This will be copied for
every thread after setup.
cshare The share.
write_cache If non-NULL a write cache that is to be
synchronized with the read caches.
num_threads Number of threads sharing the cache
including the write thread if any.
DESCRIPTION
The shared cache is used so: One IO_CACHE is initialized with
init_io_cache(). This includes the allocation of a buffer. Then a
share is allocated and init_io_cache_share() is called with the io
cache and the share. Then the io cache is copied for each thread. So
every thread has its own copy of IO_CACHE. But the allocated buffer
is shared because cache->buffer is the same for all caches.
One thread reads data from the file into the buffer. All threads
read from the buffer, but every thread maintains its own set of
pointers into the buffer. When all threads have used up the buffer
contents, one of the threads reads the next block of data into the
buffer. To accomplish this, each thread enters the cache lock before
accessing the buffer. They wait in lock_io_cache() until all threads
joined the lock. The last thread entering the lock is in charge of
reading from file to buffer. It wakes all threads when done.
Synchronizing a write cache to the read caches works so: Whenever
the write buffer needs a flush, the write thread enters the lock and
waits for all other threads to enter the lock too. They do this when
they have used up the read buffer. When all threads are in the lock,
the write thread copies the write buffer to the read buffer and
wakes all threads.
share->running_threads is the number of threads not being in the
cache lock. When entering lock_io_cache() the number is decreased.
When the thread that fills the buffer enters unlock_io_cache() the
number is reset to the number of threads. The condition
running_threads == 0 means that all threads are in the lock. Bumping
up the number to the full count is non-intuitive. But increasing the
number by one for each thread that leaves the lock could lead to a
solo run of one thread. The last thread to join a lock reads from
file to buffer, wakes the other threads, processes the data in the
cache and enters the lock again. If no other thread left the lock
meanwhile, it would think it's the last one again and read the next
block...
The share has copies of 'error', 'buffer', 'read_end', and
'pos_in_file' from the thread that filled the buffer. We may not be
able to access this information directly from its cache because the
thread may be removed from the share before the variables could be
copied by all other threads. Or, if a write buffer is synchronized,
it would change its 'pos_in_file' after waking the other threads,
possibly before they could copy its value.
However, the 'buffer' variable in the share is for a synchronized
write cache. It needs to know where to put the data. Otherwise it
would need access to the read cache of one of the threads that is
not yet removed from the share.
RETURN
void
*/
void init_io_cache_share(IO_CACHE *read_cache, IO_CACHE_SHARE *cshare,
IO_CACHE *write_cache, uint num_threads) {
DBUG_TRACE;
DBUG_PRINT("io_cache_share", ("read_cache: %p share: %p "
"write_cache: %p threads: %u",
read_cache, cshare, write_cache, num_threads));
assert(num_threads > 1);
assert(read_cache->type == READ_CACHE);
assert(!write_cache || (write_cache->type == WRITE_CACHE));
mysql_mutex_init(key_IO_CACHE_SHARE_mutex, &cshare->mutex,
MY_MUTEX_INIT_FAST);
mysql_cond_init(key_IO_CACHE_SHARE_cond, &cshare->cond);
mysql_cond_init(key_IO_CACHE_SHARE_cond_writer, &cshare->cond_writer);
cshare->running_threads = num_threads;
cshare->total_threads = num_threads;
cshare->error = 0; /* Initialize. */
cshare->buffer = read_cache->buffer;
cshare->read_end = nullptr; /* See function comment of lock_io_cache(). */
cshare->pos_in_file = 0; /* See function comment of lock_io_cache(). */
cshare->source_cache = write_cache; /* Can be NULL. */
read_cache->share = cshare;
read_cache->read_function = _my_b_read_r;
read_cache->current_pos = nullptr;
read_cache->current_end = nullptr;
if (write_cache) write_cache->share = cshare;
}
/*
Remove a thread from shared access to IO_CACHE.
SYNOPSIS
remove_io_thread()
cache The IO_CACHE to be removed from the share.
NOTE
Every thread must do that on exit for not to deadlock other threads.
The last thread destroys the pthread resources.
A writer flushes its cache first.
RETURN
void
*/
void remove_io_thread(IO_CACHE *cache) {
IO_CACHE_SHARE *cshare = cache->share;
uint total;
DBUG_TRACE;
/* If the writer goes, it needs to flush the write cache. */
if (cache == cshare->source_cache) flush_io_cache(cache);
mysql_mutex_lock(&cshare->mutex);
DBUG_PRINT(
"io_cache_share",
("%s: %p", (cache == cshare->source_cache) ? "writer" : "reader", cache));
/* Remove from share. */
total = --cshare->total_threads;
DBUG_PRINT("io_cache_share", ("remaining threads: %u", total));
/* Detach from share. */
cache->share = nullptr;
/* If the writer goes, let the readers know. */
if (cache == cshare->source_cache) {
DBUG_PRINT("io_cache_share", ("writer leaves"));
cshare->source_cache = nullptr;
}
/* If all threads are waiting for me to join the lock, wake them. */
if (!--cshare->running_threads) {
DBUG_PRINT("io_cache_share", ("the last running thread leaves, wake all"));
mysql_cond_signal(&cshare->cond_writer);
mysql_cond_broadcast(&cshare->cond);
}
mysql_mutex_unlock(&cshare->mutex);
if (!total) {
DBUG_PRINT("io_cache_share", ("last thread removed, destroy share"));
mysql_cond_destroy(&cshare->cond_writer);
mysql_cond_destroy(&cshare->cond);
mysql_mutex_destroy(&cshare->mutex);
}
}
/*
Lock IO cache and wait for all other threads to join.
SYNOPSIS
lock_io_cache()
cache The cache of the thread entering the lock.
pos File position of the block to read.
Unused for the write thread.
DESCRIPTION
Wait for all threads to finish with the current buffer. We want
all threads to proceed in concert. The last thread to join
lock_io_cache() will read the block from file and all threads start
to use it. Then they will join again for reading the next block.
The waiting threads detect a fresh buffer by comparing
cshare->pos_in_file with the position they want to process next.
Since the first block may start at position 0, we take
cshare->read_end as an additional condition. This variable is
initialized to NULL and will be set after a block of data is written
to the buffer.
RETURN
1 OK, lock in place, go ahead and read.
0 OK, unlocked, another thread did the read.
*/
static int lock_io_cache(IO_CACHE *cache, my_off_t pos) {
IO_CACHE_SHARE *cshare = cache->share;
DBUG_TRACE;
/* Enter the lock. */
mysql_mutex_lock(&cshare->mutex);
cshare->running_threads--;
DBUG_PRINT("io_cache_share",
("%s: %p pos: %lu running: %u",
(cache == cshare->source_cache) ? "writer" : "reader", cache,
(ulong)pos, cshare->running_threads));
if (cshare->source_cache) {
/* A write cache is synchronized to the read caches. */
if (cache == cshare->source_cache) {
/* The writer waits until all readers are here. */
while (cshare->running_threads) {
DBUG_PRINT("io_cache_share", ("writer waits in lock"));
mysql_cond_wait(&cshare->cond_writer, &cshare->mutex);
}
DBUG_PRINT("io_cache_share", ("writer awoke, going to copy"));
/* Stay locked. Leave the lock later by unlock_io_cache(). */
return 1;
}
/* The last thread wakes the writer. */
if (!cshare->running_threads) {
DBUG_PRINT("io_cache_share", ("waking writer"));
mysql_cond_signal(&cshare->cond_writer);
}
/*
Readers wait until the data is copied from the writer. Another
reason to stop waiting is the removal of the write thread. If this
happens, we leave the lock with old data in the buffer.
*/
while ((!cshare->read_end || (cshare->pos_in_file < pos)) &&
cshare->source_cache) {
DBUG_PRINT("io_cache_share", ("reader waits in lock"));
mysql_cond_wait(&cshare->cond, &cshare->mutex);
}
/*
If the writer was removed from the share while this thread was
asleep, we need to simulate an EOF condition. The writer cannot
reset the share variables as they might still be in use by readers
of the last block. When we awake here then because the last
joining thread signalled us. If the writer is not the last, it
will not signal. So it is safe to clear the buffer here.
*/
if (!cshare->read_end || (cshare->pos_in_file < pos)) {
DBUG_PRINT("io_cache_share", ("reader found writer removed. EOF"));
cshare->read_end = cshare->buffer; /* Empty buffer. */
cshare->error = 0; /* EOF is not an error. */
}
} else {
/*
There are read caches only. The last thread arriving in
lock_io_cache() continues with a locked cache and reads the block.
*/
if (!cshare->running_threads) {
DBUG_PRINT("io_cache_share", ("last thread joined, going to read"));
/* Stay locked. Leave the lock later by unlock_io_cache(). */
return 1;
}
/*
All other threads wait until the requested block is read by the
last thread arriving. Another reason to stop waiting is the
removal of a thread. If this leads to all threads being in the
lock, we have to continue also. The first of the awaken threads
will then do the read.
*/
while ((!cshare->read_end || (cshare->pos_in_file < pos)) &&
cshare->running_threads) {
DBUG_PRINT("io_cache_share", ("reader waits in lock"));
mysql_cond_wait(&cshare->cond, &cshare->mutex);
}
/* If the block is not yet read, continue with a locked cache and read. */
if (!cshare->read_end || (cshare->pos_in_file < pos)) {
DBUG_PRINT("io_cache_share", ("reader awoke, going to read"));
/* Stay locked. Leave the lock later by unlock_io_cache(). */
return 1;
}
/* Another thread did read the block already. */
}
DBUG_PRINT(
"io_cache_share",
("reader awoke, going to process %u bytes",
(uint)(cshare->read_end ? (size_t)(cshare->read_end - cshare->buffer)
: 0)));
/*
Leave the lock. Do not call unlock_io_cache() later. The thread that
filled the buffer did this and marked all threads as running.
*/
mysql_mutex_unlock(&cshare->mutex);
return 0;
}
/*
Unlock IO cache.
SYNOPSIS
unlock_io_cache()
cache The cache of the thread leaving the lock.
NOTE
This is called by the thread that filled the buffer. It marks all
threads as running and awakes them. This must not be done by any
other thread.
Do not signal cond_writer. Either there is no writer or the writer
is the only one who can call this function.
The reason for resetting running_threads to total_threads before
waking all other threads is that it could be possible that this
thread is so fast with processing the buffer that it enters the lock
before even one other thread has left it. If every awoken thread
would increase running_threads by one, this thread could think that
he is again the last to join and would not wait for the other
threads to process the data.
RETURN
void
*/
static void unlock_io_cache(IO_CACHE *cache) {
IO_CACHE_SHARE *cshare = cache->share;
DBUG_TRACE;
DBUG_PRINT("io_cache_share",
("%s: %p pos: %lu running: %u",
(cache == cshare->source_cache) ? "writer" : "reader", cache,
(ulong)cshare->pos_in_file, cshare->total_threads));
cshare->running_threads = cshare->total_threads;
mysql_cond_broadcast(&cshare->cond);
mysql_mutex_unlock(&cshare->mutex);
}
/*
Read from IO_CACHE when it is shared between several threads.
SYNOPSIS
_my_b_read_r()
cache IO_CACHE pointer
Buffer Buffer to retrieve count bytes from file
Count Number of bytes to read into Buffer
NOTE
This function is only called from the my_b_read() macro when there
isn't enough characters in the buffer to satisfy the request.
IMPLEMENTATION
It works as follows: when a thread tries to read from a file (that
is, after using all the data from the (shared) buffer), it just
hangs on lock_io_cache(), waiting for other threads. When the very
last thread attempts a read, lock_io_cache() returns 1, the thread
does actual IO and unlock_io_cache(), which signals all the waiting
threads that data is in the buffer.
WARNING
When changing this function, be careful with handling file offsets
(end-of_file, pos_in_file). Do not cast them to possibly smaller
types than my_off_t unless you can be sure that their value fits.
Same applies to differences of file offsets. (Bug #11527)
When changing this function, check _my_b_read(). It might need the
same change.
RETURN
0 we succeeded in reading all data
1 Error: can't read requested characters
*/
int _my_b_read_r(IO_CACHE *cache, uchar *Buffer, size_t Count) {
my_off_t pos_in_file;
size_t length, diff_length, left_length;
IO_CACHE_SHARE *cshare = cache->share;
DBUG_TRACE;
if ((left_length = (size_t)(cache->read_end - cache->read_pos))) {
assert(Count >= left_length); /* User is not using my_b_read() */
memcpy(Buffer, cache->read_pos, left_length);
Buffer += left_length;
Count -= left_length;
}
while (Count) {
size_t cnt, len;
pos_in_file = cache->pos_in_file + (cache->read_end - cache->buffer);
diff_length = (size_t)(pos_in_file & (IO_SIZE - 1));
length = IO_ROUND_UP(Count + diff_length) - diff_length;
length = ((length <= cache->read_length)
? length + IO_ROUND_DN(cache->read_length - length)
: length - IO_ROUND_UP(length - cache->read_length));
if (cache->type != READ_FIFO &&
(length > (cache->end_of_file - pos_in_file)))
length = (size_t)(cache->end_of_file - pos_in_file);
if (length == 0) {
cache->error = (int)left_length;
return 1;
}
if (lock_io_cache(cache, pos_in_file)) {
/* With a synchronized write/read cache we won't come here... */
assert(!cshare->source_cache);
/*
... unless the writer has gone before this thread entered the
lock. Simulate EOF in this case. It can be distinguished by
cache->file.
*/
if (cache->file < 0)
len = 0;
else {
/*
Whenever a function which operates on IO_CACHE flushes/writes
some part of the IO_CACHE to disk it will set the property
"seek_not_done" to indicate this to other functions operating
on the IO_CACHE.
*/
if (cache->seek_not_done) {
if (mysql_encryption_file_seek(cache, pos_in_file, MY_SEEK_SET,
MYF(0)) == MY_FILEPOS_ERROR) {
cache->error = -1;
unlock_io_cache(cache);
return 1;
}
}
len = mysql_encryption_file_read(cache, cache->buffer, length,
cache->myflags);
}
DBUG_PRINT("io_cache_share", ("read %lu bytes", (ulong)len));
cache->read_end = cache->buffer + (len == (size_t)-1 ? 0 : len);
cache->error = (len == length ? 0 : (int)len);
cache->pos_in_file = pos_in_file;
/* Copy important values to the share. */
cshare->error = cache->error;
cshare->read_end = cache->read_end;
cshare->pos_in_file = pos_in_file;
/* Mark all threads as running and wake them. */
unlock_io_cache(cache);
} else {
/*
With a synchronized write/read cache readers always come here.
Copy important values from the share.
*/
cache->error = cshare->error;
cache->read_end = cshare->read_end;
cache->pos_in_file = cshare->pos_in_file;