darshan-posix.c 27.8 KB
Newer Older
1
2
3
4
5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6
#include "darshan-runtime-config.h"
7
8
9
10
11
12
13
14
15
16
17
18
19
20
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdarg.h>
#include <string.h>
#include <time.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/uio.h>
#include <sys/mman.h>
#include <search.h>
#include <assert.h>
21
#include <libgen.h>
Philip Carns's avatar
Philip Carns committed
22
#include <aio.h>
23
#define __USE_GNU
24
#include <pthread.h>
25

26
#include "uthash.h"
27
#include "darshan.h"
28
#include "darshan-posix-log-format.h"
29

30
#ifndef HAVE_OFF64_T
31
32
typedef int64_t off64_t;
#endif
33
34
35
#ifndef HAVE_AIOCB64
#define aiocb64 aiocb
#endif
36

37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
#ifdef DARSHAN_PRELOAD
#define __USE_GNU
#include <dlfcn.h>
#include <stdlib.h>

#define DARSHAN_FORWARD_DECL(name,ret,args) \
  ret (*__real_ ## name)args = NULL;

#define DARSHAN_DECL(__name) __name

#define DARSHAN_MPI_CALL(func) __real_ ## func

#define MAP_OR_FAIL(func) \
    if (!(__real_ ## func)) \
    { \
        __real_ ## func = dlsym(RTLD_NEXT, #func); \
        if(!(__real_ ## func)) { \
           fprintf(stderr, "Darshan failed to map symbol: %s\n", #func); \
           exit(1); \
       } \
    }

#else
60

61
62
63
64
65
#define DARSHAN_FORWARD_DECL(name,ret,args) \
  extern ret __real_ ## name args;

#define DARSHAN_DECL(__name) __wrap_ ## __name

66
67
#define DARSHAN_MPI_CALL(func) func

68
69
#define MAP_OR_FAIL(func)

70
71
#endif

72
73
/* TODO: more libc, fgetc, etc etc etc. */

74
75
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(open64, int, (const char *path, int flags, ...));
76
77
78
79
80
81
82
83
84
85
DARSHAN_FORWARD_DECL(creat, int, (const char* path, mode_t mode));
DARSHAN_FORWARD_DECL(creat64, int, (const char* path, mode_t mode));
DARSHAN_FORWARD_DECL(read, ssize_t, (int fd, void *buf, size_t count));
DARSHAN_FORWARD_DECL(write, ssize_t, (int fd, const void *buf, size_t count));
DARSHAN_FORWARD_DECL(pread, ssize_t, (int fd, void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pwrite, ssize_t, (int fd, const void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pread64, ssize_t, (int fd, void *buf, size_t count, off64_t offset));
DARSHAN_FORWARD_DECL(pwrite64, ssize_t, (int fd, const void *buf, size_t count, off64_t offset));
DARSHAN_FORWARD_DECL(readv, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
DARSHAN_FORWARD_DECL(writev, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
86
87
DARSHAN_FORWARD_DECL(close, int, (int fd));

88
struct posix_file_runtime
89
{
90
    struct darshan_posix_file* file_record;
91
92
93
    double last_meta_end;
    double last_read_end;
    double last_write_end;
94
    UT_hash_handle hlink;
95
};
96

97
struct posix_file_runtime_ref
98
{
99
    struct posix_file_runtime* file;
100
101
102
103
104
105
    int fd;
    UT_hash_handle hlink;
};

struct posix_runtime
{
106
    struct posix_file_runtime* file_runtime_array;
107
    struct darshan_posix_file* file_record_array;
108
    int file_array_size;
109
    int file_array_ndx;
110
111
    struct posix_file_runtime* file_hash;
    struct posix_file_runtime_ref* fd_hash;
112
113
    void *red_buf;
    int shared_rec_count;
114
115
};

116
static struct posix_runtime *posix_runtime = NULL;
117
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
Shane Snyder's avatar
Shane Snyder committed
118
static int instrumentation_disabled = 0;
119
static int my_rank = -1;
120

121
/* TODO: I'm sure these should be applied on all modules */
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

137
static void posix_runtime_initialize(void);
138
139
140
static struct posix_file_runtime* posix_file_by_name(const char *name);
static struct posix_file_runtime* posix_file_by_name_setfd(const char* name, int fd);
static struct posix_file_runtime* posix_file_by_fd(int fd);
141
142
static void posix_file_close_fd(int fd);

Shane Snyder's avatar
Shane Snyder committed
143
static void posix_disable_instrumentation(void);
144
145
static void posix_prepare_for_reduction(darshan_record_id *shared_recs,
    int *shared_rec_count, void **send_buf, void **recv_buf, int *rec_size);
Shane Snyder's avatar
Shane Snyder committed
146
static void posix_reduce_records(void* infile_v, void* inoutfile_v,
147
148
    int *len, MPI_Datatype *datatype);
static void posix_get_output_data(void **buffer, int *size);
149
static void posix_shutdown(void);
150

151
152
#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)
153

154
#define POSIX_SET(__file, __counter, __value) do {\
155
    (__file)->file_record->counters[__counter] = __value; \
156
157
158
} while(0)

#define POSIX_F_SET(__file, __counter, __value) do {\
159
    (__file)->file_record->fcounters[__counter] = __value; \
160
161
162
} while(0)

#define POSIX_INC(__file, __counter, __value) do {\
163
    (__file)->file_record->counters[__counter] += __value; \
164
165
166
} while(0)

#define POSIX_F_INC(__file, __counter, __value) do {\
167
    (__file)->file_record->fcounters[__counter] += __value; \
168
169
170
171
172
173
174
175
176
177
178
179
} while(0)

#define POSIX_F_INC_NO_OVERLAP(__file, __tm1, __tm2, __last, __counter) do { \
    if(__tm1 > __last) \
        POSIX_F_INC(__file, __counter, (__tm2-__tm1)); \
    else \
        POSIX_F_INC(__file, __counter, (__tm2 - __last)); \
    if(__tm2 > __last) \
        __last = __tm2; \
} while(0)

#define POSIX_VALUE(__file, __counter) \
180
    ((__file)->file_record->counters[__counter])
181
182

#define POSIX_F_VALUE(__file, __counter) \
183
    ((__file)->file_record->fcounters[__counter])
184
185

#define POSIX_MAX(__file, __counter, __value) do {\
186
    if((__file)->file_record->counters[__counter] < __value) \
187
    { \
188
        (__file)->file_record->counters[__counter] = __value; \
189
190
191
    } \
} while(0)

192
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
193
    struct posix_file_runtime* file; \
194
195
196
197
198
199
200
201
202
    char* exclude; \
    int tmp_index = 0; \
    if(__ret < 0) break; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
203
    file = posix_file_by_name_setfd(__path, __ret); \
204
    if(!file) break; \
205
    /* CP_STAT_FILE(file, __path, __ret); */ \
206
    file->file_record->rank = my_rank; \
207
    if(__mode) \
208
209
210
211
        POSIX_SET(file, POSIX_MODE, __mode); \
    /* file->offset = 0; */ \
    /* file->last_byte_written = 0; */ \
    /* file->last_byte_read = 0; */ \
212
    if(__stream_flag)\
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
        POSIX_INC(file, POSIX_FOPENS, 1); \
    else \
        POSIX_INC(file, POSIX_OPENS, 1); \
    if(POSIX_F_VALUE(file, POSIX_F_OPEN_TIMESTAMP) == 0) \
        POSIX_F_SET(file, POSIX_F_OPEN_TIMESTAMP, __tm1); \
    POSIX_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_meta_end, POSIX_F_META_TIME); \
} while(0)

#define POSIX_RECORD_READ(__ret, __fd, __pread_flag, __pread_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
    /* size_t stride; \
    int64_t this_offset; */ \
    struct posix_file_runtime* file; \
    /* int64_t file_alignment; */ \
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
    file = posix_file_by_fd(__fd); \
    if(!file) break; \
    /* if(__pread_flag) \
        this_offset = __pread_offset; \
    else \
        this_offset = file->offset; \
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
    if(this_offset > file->last_byte_read) \
        CP_INC(file, CP_SEQ_READS, 1); \
    if(this_offset == (file->last_byte_read + 1)) \
        CP_INC(file, CP_CONSEC_READS, 1); \
    if(this_offset > 0 && this_offset > file->last_byte_read \
        && file->last_byte_read != 0) \
        stride = this_offset - file->last_byte_read - 1; \
    else \
        stride = 0; \
    file->last_byte_read = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
    CP_MAX(file, CP_MAX_BYTE_READ, (this_offset + __ret -1)); \
    CP_INC(file, CP_BYTES_READ, __ret); */\
    if(__stream_flag)\
        POSIX_INC(file, POSIX_FREADS, 1); \
    else\
        POSIX_INC(file, POSIX_READS, 1); \
    /* CP_BUCKET_INC(file, CP_SIZE_READ_0_100, __ret); \
    cp_access_counter(file, stride, CP_COUNTER_STRIDE); \
    if(!__aligned) \
        CP_INC(file, CP_MEM_NOT_ALIGNED, 1); \
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
        CP_INC(file, CP_FILE_NOT_ALIGNED, 1); \
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); \
    if(file->last_io_type == CP_WRITE) \
        CP_INC(file, CP_RW_SWITCHES, 1); \
    file->last_io_type = CP_READ; */ \
    POSIX_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_read_end, POSIX_F_READ_TIME);  \
    if(POSIX_F_VALUE(file, POSIX_F_READ_START_TIMESTAMP) == 0) \
        POSIX_F_SET(file, POSIX_F_READ_START_TIMESTAMP, __tm1); \
    POSIX_F_SET(file, POSIX_F_READ_END_TIMESTAMP, __tm2); \
    if(POSIX_F_VALUE(file, POSIX_F_MAX_READ_TIME) < __elapsed){ \
        POSIX_F_SET(file, POSIX_F_MAX_READ_TIME, __elapsed); \
        POSIX_SET(file, POSIX_MAX_READ_TIME_SIZE, __ret); } \
} while(0)

#define POSIX_RECORD_WRITE(__ret, __fd, __pwrite_flag, __pwrite_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
    /* size_t stride; \
    int64_t this_offset; */ \
    struct posix_file_runtime* file; \
    /* int64_t file_alignment; */ \
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
    file = posix_file_by_fd(__fd); \
    if(!file) break; \
    /* if(__pwrite_flag) \
        this_offset = __pwrite_offset; \
    else \
        this_offset = file->offset; \
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
    if(this_offset > file->last_byte_written) \
        CP_INC(file, CP_SEQ_WRITES, 1); \
    if(this_offset == (file->last_byte_written + 1)) \
        CP_INC(file, CP_CONSEC_WRITES, 1); \
    if(this_offset > 0 && this_offset > file->last_byte_written \
        && file->last_byte_written != 0) \
        stride = this_offset - file->last_byte_written - 1; \
292
    else \
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
        stride = 0; \
    file->last_byte_written = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
    CP_MAX(file, CP_MAX_BYTE_WRITTEN, (this_offset + __ret -1)); \
    CP_INC(file, CP_BYTES_WRITTEN, __ret); */ \
    if(__stream_flag) \
        POSIX_INC(file, POSIX_FWRITES, 1); \
    else \
        POSIX_INC(file, POSIX_WRITES, 1); \
    /* CP_BUCKET_INC(file, CP_SIZE_WRITE_0_100, __ret); \
    cp_access_counter(file, stride, CP_COUNTER_STRIDE); \
    if(!__aligned) \
        CP_INC(file, CP_MEM_NOT_ALIGNED, 1); \
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
        CP_INC(file, CP_FILE_NOT_ALIGNED, 1); \
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); \
    if(file->last_io_type == CP_READ) \
        CP_INC(file, CP_RW_SWITCHES, 1); \
    file->last_io_type = CP_WRITE; */ \
    POSIX_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_write_end, POSIX_F_WRITE_TIME); \
    if(POSIX_F_VALUE(file, POSIX_F_WRITE_START_TIMESTAMP) == 0) \
        POSIX_F_SET(file, POSIX_F_WRITE_START_TIMESTAMP, __tm1); \
    POSIX_F_SET(file, POSIX_F_WRITE_END_TIMESTAMP, __tm2); \
    if(POSIX_F_VALUE(file, POSIX_F_MAX_WRITE_TIME) < __elapsed){ \
        POSIX_F_SET(file, POSIX_F_MAX_WRITE_TIME, __elapsed); \
        POSIX_SET(file, POSIX_MAX_WRITE_TIME_SIZE, __ret); } \
319
} while(0)
320

321
int DARSHAN_DECL(open)(const char *path, int flags, ...)
322
323
324
325
326
{
    int mode = 0;
    int ret;
    double tm1, tm2;

327
328
    MAP_OR_FAIL(open);

329
    if(flags & O_CREAT) 
330
331
332
333
334
335
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

336
        tm1 = darshan_core_wtime();
337
        ret = __real_open(path, flags, mode);
338
        tm2 = darshan_core_wtime();
339
340
341
    }
    else
    {
342
        tm1 = darshan_core_wtime();
343
        ret = __real_open(path, flags);
344
        tm2 = darshan_core_wtime();
345
346
    }

347
348
    POSIX_LOCK();
    posix_runtime_initialize();
349
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
350
    POSIX_UNLOCK();
351
352
353
354

    return(ret);
}

Shane Snyder's avatar
Shane Snyder committed
355
356
357
358
359
360
int DARSHAN_DECL(open64)(const char *path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

361
    MAP_OR_FAIL(open64);
Shane Snyder's avatar
Shane Snyder committed
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382

    if(flags & O_CREAT)
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

        tm1 = darshan_core_wtime();
        ret = __real_open64(path, flags, mode);
        tm2 = darshan_core_wtime();
    }
    else
    {
        tm1 = darshan_core_wtime();
        ret = __real_open64(path, flags);
        tm2 = darshan_core_wtime();
    }

    POSIX_LOCK();
    posix_runtime_initialize();
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
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(creat)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat);

    tm1 = darshan_core_wtime();
    ret = __real_creat(path, mode);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(creat64)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat64);
Shane Snyder's avatar
Shane Snyder committed
414

415
416
417
418
419
420
    tm1 = darshan_core_wtime();
    ret = __real_creat64(path, mode);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
Shane Snyder's avatar
Shane Snyder committed
421
422
423
424
425
426
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

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
ssize_t DARSHAN_DECL(read)(int fd, void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(write);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_read(fd, buf, count);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_READ(ret, fd, 0, 0, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(write)(int fd, const void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(write);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_write(fd, buf, count);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_WRITE(ret, fd, 0, 0, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(pread)(int fd, void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_pread(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_READ(ret, fd, 1, offset, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(pwrite)(int fd, const void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_pwrite(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_WRITE(ret, fd, 1, offset, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(pread64)(int fd, void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread64);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_pread64(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_READ(ret, fd, 1, offset, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(pwrite64)(int fd, const void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite64);

    /* if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1; */

    tm1 = darshan_core_wtime();
    ret = __real_pwrite64(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
    POSIX_RECORD_WRITE(ret, fd, 1, offset, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(readv)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
    int aligned_flag = 0;
    /* int aligned_flag = 1; */
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(readv);
/*
    for(i=0; i<iovcnt; i++)
    {
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
            aligned_flag = 0;
    }*/

    tm1 = darshan_core_wtime();
    ret = __real_readv(fd, iov, iovcnt);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    POSIX_RECORD_READ(ret, fd, 0, 0, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(writev)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
    int aligned_flag = 0;
    /* int aligned_flag = 1; */
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(writev);
/*
    for(i=0; i<iovcnt; i++)
    {
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
            aligned_flag = 0;
    }*/

    tm1 = darshan_core_wtime();
    ret = __real_writev(fd, iov, iovcnt);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    POSIX_RECORD_WRITE(ret, fd, 0, 0, aligned_flag, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

613
614
int DARSHAN_DECL(close)(int fd)
{
615
    struct posix_file_runtime* file;
616
617
618
619
620
621
622
623
624
625
626
    double tm1, tm2;
    int ret;

    MAP_OR_FAIL(close);

    tm1 = darshan_core_wtime();
    ret = __real_close(fd);
    tm2 = darshan_core_wtime();

    POSIX_LOCK();
    posix_runtime_initialize();
Shane Snyder's avatar
Shane Snyder committed
627
628
629
    file = posix_file_by_fd(fd);
    if(file)
    {
630
631
632
633
        /* file->last_byte_written = 0; */
        /* file->last_byte_read = 0; */
        POSIX_F_SET(file, POSIX_F_CLOSE_TIMESTAMP, darshan_core_wtime());
        POSIX_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_meta_end, POSIX_F_META_TIME);
Shane Snyder's avatar
Shane Snyder committed
634
635
        posix_file_close_fd(fd);
    }
636
637
638
639
640
641
    POSIX_UNLOCK();    

    return(ret);
}

/* ***************************************************** */
642

643
static void posix_runtime_initialize()
644
{
645
646
647
    int mem_limit;
    struct darshan_module_funcs posix_mod_fns =
    {
Shane Snyder's avatar
Shane Snyder committed
648
        .disable_instrumentation = &posix_disable_instrumentation,
649
        .prepare_for_reduction = &posix_prepare_for_reduction,
Shane Snyder's avatar
Shane Snyder committed
650
        .reduce_records = &posix_reduce_records,
651
        .get_output_data = &posix_get_output_data,
652
        .shutdown = &posix_shutdown
653
    };
654

Shane Snyder's avatar
Shane Snyder committed
655
656
    /* don't do anything if already initialized or instrumenation is disabled */
    if(posix_runtime || instrumentation_disabled)
657
        return;
658

659
660
    /* register the posix module with darshan core */
    darshan_core_register_module(
661
        DARSHAN_POSIX_MOD,
662
663
664
665
666
667
668
669
670
671
672
673
674
675
        &posix_mod_fns,
        &mem_limit);

    /* return if no memory assigned by darshan core */
    if(mem_limit == 0)
        return;

    posix_runtime = malloc(sizeof(*posix_runtime));
    if(!posix_runtime)
        return;
    memset(posix_runtime, 0, sizeof(*posix_runtime));

    /* set maximum number of file records according to max memory limit */
    /* NOTE: maximum number of records is based on the size of a posix file record */
676
    /* TODO: should we base memory usage off file record or total runtime structure sizes? */
677
    posix_runtime->file_array_size = mem_limit / sizeof(struct darshan_posix_file);
678
    posix_runtime->file_array_ndx = 0;
679
680

    /* allocate array of runtime file records */
681
    posix_runtime->file_runtime_array = malloc(posix_runtime->file_array_size *
682
                                               sizeof(struct posix_file_runtime));
683
684
685
    posix_runtime->file_record_array = malloc(posix_runtime->file_array_size *
                                              sizeof(struct darshan_posix_file));
    if(!posix_runtime->file_runtime_array || !posix_runtime->file_record_array)
686
687
688
689
    {
        posix_runtime->file_array_size = 0;
        return;
    }
690
    memset(posix_runtime->file_runtime_array, 0, posix_runtime->file_array_size *
691
           sizeof(struct posix_file_runtime));
692
693
    memset(posix_runtime->file_record_array, 0, posix_runtime->file_array_size *
           sizeof(struct darshan_posix_file));
694

Shane Snyder's avatar
Shane Snyder committed
695
696
    /* TODO: can we move this out of here? perhaps register_module returns rank? */
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
697

698
699
    return;
}
700

701
static struct posix_file_runtime* posix_file_by_name(const char *name)
702
{
703
    struct posix_file_runtime *file = NULL;
704
    char *newname = NULL;
705
    darshan_record_id file_id;
706

Shane Snyder's avatar
Shane Snyder committed
707
    if(!posix_runtime || instrumentation_disabled)
708
        return(NULL);
709

710
    newname = darshan_clean_file_path(name);
711
    if(!newname)
712
        newname = (char*)name;
713

714
    /* get a unique id for this file from darshan core */
Shane Snyder's avatar
Shane Snyder committed
715
    darshan_core_register_record(
716
717
718
        (void*)newname,
        strlen(newname),
        1,
719
        DARSHAN_POSIX_MOD,
720
        &file_id);
721

722
    /* search the hash table for this file record, and return if found */
723
    HASH_FIND(hlink, posix_runtime->file_hash, &file_id, sizeof(darshan_record_id), file);
724
    if(file)
725
    {
726
        if(newname != name)
727
            free(newname);
728
        return(file);
729
730
    }

731
732
733
734
735
736
    if(posix_runtime->file_array_ndx < posix_runtime->file_array_size);
    {
        /* no existing record, assign a new file record from the global array */
        file = &(posix_runtime->file_runtime_array[posix_runtime->file_array_ndx]);
        file->file_record = &(posix_runtime->file_record_array[posix_runtime->file_array_ndx]);
        file->file_record->f_id = file_id;
737

738
        /* add new record to file hash table */
739
        HASH_ADD(hlink, posix_runtime->file_hash, file_record->f_id, sizeof(darshan_record_id), file);
740

741
742
        posix_runtime->file_array_ndx++;
    }
743
744

    if(newname != name)
745
        free(newname);
746
747
748
    return(file);
}

749
static struct posix_file_runtime* posix_file_by_name_setfd(const char* name, int fd)
750
{
751
752
    struct posix_file_runtime* file;
    struct posix_file_runtime_ref* ref;
753

Shane Snyder's avatar
Shane Snyder committed
754
    if(!posix_runtime || instrumentation_disabled)
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
        return(NULL);

    /* find file record by name first */
    file = posix_file_by_name(name);

    if(!file)
        return(NULL);

    /* search hash table for existing file ref for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
    if(ref)
    {
        /* we have a reference.  Make sure it points to the correct file
         * and return it
         */
        ref->file = file;
        return(file);
    }

    /* if we hit this point, then we don't have a reference for this fd
     * in the table yet.  Add it.
     */
    ref = malloc(sizeof(*ref));
    if(!ref)
        return(NULL);
    memset(ref, 0, sizeof(*ref));
Shane Snyder's avatar
Shane Snyder committed
781

782
783
784
785
786
787
788
    ref->file = file;
    ref->fd = fd;    
    HASH_ADD(hlink, posix_runtime->fd_hash, fd, sizeof(int), ref);

    return(file);
}

789
static struct posix_file_runtime* posix_file_by_fd(int fd)
Shane Snyder's avatar
Shane Snyder committed
790
{
791
    struct posix_file_runtime_ref* ref;
Shane Snyder's avatar
Shane Snyder committed
792
793
794
795
796
797
798
799
800
801
802
803

    if(!posix_runtime || instrumentation_disabled)
        return(NULL);

    /* search hash table for existing file ref for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
    if(ref)
        return(ref->file);

    return(NULL);
}

804
805
static void posix_file_close_fd(int fd)
{
806
    struct posix_file_runtime_ref* ref;
807

Shane Snyder's avatar
Shane Snyder committed
808
    if(!posix_runtime || instrumentation_disabled)
809
810
811
812
        return;

    /* search hash table for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
813
    if(ref)
814
815
816
817
818
819
820
    {
        /* we have a reference, delete it */
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }

    return;
821
822
}

823
824
825
826
827
828
829
830
831
832
833
834
835
static int posix_file_compare(const void* a, const void* b)
{
    const struct darshan_posix_file* f_a = a;
    const struct darshan_posix_file* f_b = b;

    if(f_a->rank < f_b->rank)
        return 1;
    if(f_a->rank > f_b->rank)
        return -1;

    return 0;
}

836
837
/* ***************************************************** */

Shane Snyder's avatar
Shane Snyder committed
838
static void posix_disable_instrumentation()
839
{
840
841
    assert(posix_runtime);

Shane Snyder's avatar
Shane Snyder committed
842
843
844
    POSIX_LOCK();
    instrumentation_disabled = 1;
    POSIX_UNLOCK();
845

Shane Snyder's avatar
Shane Snyder committed
846
847
848
    return;
}

849
850
851
852
853
854
855
static void posix_prepare_for_reduction(
    darshan_record_id *shared_recs,
    int *shared_rec_count,
    void **send_buf,
    void **recv_buf,
    int *rec_size)
{
856
    struct posix_file_runtime *file;
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
    int i;

    assert(posix_runtime);

    /* necessary initialization of shared records (e.g., change rank to -1) */
    for(i = 0; i < *shared_rec_count; i++)
    {
        HASH_FIND(hlink, posix_runtime->file_hash, &shared_recs[i],
            sizeof(darshan_record_id), file);
        assert(file);

        /* TODO: any initialization before reduction */
        file->file_record->rank = -1;
    }

    /* sort the array of files descending by rank so that we get all of the 
     * shared files (marked by rank -1) in a contiguous portion at end 
     * of the array
     */
    qsort(posix_runtime->file_record_array, posix_runtime->file_array_ndx,
        sizeof(struct darshan_posix_file), posix_file_compare);

    /* make *send_buf point to the shared files at the end of sorted array */
    *send_buf =
        &(posix_runtime->file_record_array[posix_runtime->file_array_ndx-(*shared_rec_count)]);

    /* allocate memory for the reduction output on rank 0 */
    if(my_rank == 0)
    {
        *recv_buf = malloc(*shared_rec_count * sizeof(struct darshan_posix_file));
        if(!(*recv_buf))
            return;
    }

    *rec_size = sizeof(struct darshan_posix_file);

Shane Snyder's avatar
Shane Snyder committed
893
    /* TODO: cleaner way to do this? */
894
895
896
897
898
899
900
    if(my_rank == 0)
        posix_runtime->red_buf = *recv_buf;
    posix_runtime->shared_rec_count = *shared_rec_count;

    return;
}

Shane Snyder's avatar
Shane Snyder committed
901
static void posix_reduce_records(
902
903
904
905
    void* infile_v,
    void* inoutfile_v,
    int *len,
    MPI_Datatype *datatype)
Shane Snyder's avatar
Shane Snyder committed
906
{
907
908
909
910
911
912
913
914
915
916
917
918
919
920
    struct darshan_posix_file tmp_file;
    struct darshan_posix_file *infile = infile_v;
    struct darshan_posix_file *inoutfile = inoutfile_v;
    int i;

    assert(posix_runtime);

    for(i = 0; i < *len; i++)
    {
        memset(&tmp_file, 0, sizeof(struct darshan_posix_file));

        tmp_file.f_id = infile->f_id;
        tmp_file.rank = -1;

921
922
        tmp_file.counters[POSIX_OPENS] = infile->counters[POSIX_OPENS] +
            inoutfile->counters[POSIX_OPENS];
923

924
925
926
        if((infile->fcounters[POSIX_F_OPEN_TIMESTAMP] > inoutfile->fcounters[POSIX_F_OPEN_TIMESTAMP]) &&
            (inoutfile->fcounters[POSIX_F_OPEN_TIMESTAMP] > 0))
            tmp_file.fcounters[POSIX_F_OPEN_TIMESTAMP] = inoutfile->fcounters[POSIX_F_OPEN_TIMESTAMP];
927
        else
928
            tmp_file.fcounters[POSIX_F_OPEN_TIMESTAMP] = infile->fcounters[POSIX_F_OPEN_TIMESTAMP];
929

930
931
        if(infile->fcounters[POSIX_F_CLOSE_TIMESTAMP] > inoutfile->fcounters[POSIX_F_CLOSE_TIMESTAMP])
            tmp_file.fcounters[POSIX_F_CLOSE_TIMESTAMP] = infile->fcounters[POSIX_F_CLOSE_TIMESTAMP];
932
        else
933
            tmp_file.fcounters[POSIX_F_CLOSE_TIMESTAMP] = inoutfile->fcounters[POSIX_F_CLOSE_TIMESTAMP];
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949

        /* update pointers */
        *inoutfile = tmp_file;
        inoutfile++;
        infile++;
    }

    return;
}

static void posix_get_output_data(
    void **buffer,
    int *size)
{
    assert(posix_runtime);

Shane Snyder's avatar
Shane Snyder committed
950
    /* TODO: cleaner way to do this? */
951
952
953
954
955
956
957
958
959
960
961
962
    /* clean up reduction state */
    if(my_rank == 0)
    {
        int tmp_ndx = posix_runtime->file_array_ndx - posix_runtime->shared_rec_count;
        memcpy(&(posix_runtime->file_record_array[tmp_ndx]), posix_runtime->red_buf,
            posix_runtime->shared_rec_count * sizeof(struct darshan_posix_file));
        free(posix_runtime->red_buf);
    }
    else
    {
        posix_runtime->file_array_ndx -= posix_runtime->shared_rec_count;
    }
963
964
965

    *buffer = (void *)(posix_runtime->file_record_array);
    *size = posix_runtime->file_array_ndx * sizeof(struct darshan_posix_file);
966

967
    return;
968
969
}

970
static void posix_shutdown()
971
{
972
    struct posix_file_runtime_ref *ref, *tmp;
973
974
975
976
977
978

    HASH_ITER(hlink, posix_runtime->fd_hash, ref, tmp)
    {
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }
979

980
    HASH_CLEAR(hlink, posix_runtime->file_hash); /* these entries are freed all at once below */
981
982
983
984
985

    free(posix_runtime->file_runtime_array);
    free(posix_runtime->file_record_array);
    free(posix_runtime);
    posix_runtime = NULL;
986
987

    return;
988
989
}

990
991
992
993
994
995
996
997
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ts=8 sts=4 sw=4 expandtab
 */