darshan-posix.c 31.3 KB
Newer Older
1
/*
Shane Snyder's avatar
Shane Snyder committed
2 3 4
 * Copyright (C) 2015 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
5 6
 */

7
#include "darshan-runtime-config.h"
8 9 10 11 12 13 14 15 16 17 18 19 20 21
#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>
22
#include <libgen.h>
23
#include <aio.h>
24
#define __USE_GNU
25
#include <pthread.h>
26

27
#include "uthash.h"
28
#include "darshan.h"
29
#include "darshan-posix-log-format.h"
30

31
#ifndef HAVE_OFF64_T
32 33
typedef int64_t off64_t;
#endif
34 35 36
#ifndef HAVE_AIOCB64
#define aiocb64 aiocb
#endif
37

38 39
/* TODO: more libc, fgetc, etc etc etc. */

40 41
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(open64, int, (const char *path, int flags, ...));
42 43 44 45 46 47 48 49 50 51
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));
52 53
DARSHAN_FORWARD_DECL(close, int, (int fd));

54 55 56 57 58 59 60 61 62 63 64 65 66 67
static void posix_runtime_initialize(void);
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);
static void posix_file_close_fd(int fd);

static void posix_disable_instrumentation(void);
static void posix_prepare_for_reduction(darshan_record_id *shared_recs,
    int *shared_rec_count, void **send_buf, void **recv_buf, int *rec_size);
static void posix_record_reduction_op(void* infile_v, void* inoutfile_v,
    int *len, MPI_Datatype *datatype);
static void posix_get_output_data(void **buffer, int *size);
static void posix_shutdown(void);

68 69 70 71 72 73 74 75 76 77 78 79 80
enum posix_io_type
{
    POSIX_READ = 1,
    POSIX_WRITE = 2,
};

/* The posix_file_runtime structure maintains necessary runtime metadata
 * for the POSIX file record (darshan_posix_file structure, defined in
 * darshan-posix-log-format.h) pointed to by 'file_record'. This metadata
 * assists with the instrumenting of specific statistics in the file record.
 * 'hlink' is a hash table link structure used to add/remove this record
 * from the hash table of POSIX file records for this process. 
 */
81
struct posix_file_runtime
82
{
83
    struct darshan_posix_file* file_record;
84 85 86 87
    int64_t offset;
    int64_t last_byte_read;
    int64_t last_byte_written;
    enum posix_io_type last_io_type;
88 89 90
    double last_meta_end;
    double last_read_end;
    double last_write_end;
91
    UT_hash_handle hlink;
92
};
93

94 95 96 97 98 99 100 101 102 103 104 105
/* The posix_file_runtime_ref structure is used to associate a POSIX
 * file descriptor with an already existing POSIX file record. This is
 * necessary as many POSIX I/O functions take only an input file descriptor,
 * but POSIX file records are indexed by their full file paths (i.e., darshan
 * record identifiers for POSIX files are created by hashing the file path).
 * In other words, this structure is necessary as it allows us to look up a
 * file record either by a pathname (posix_file_runtime) or by POSIX file
 * descriptor (posix_file_runtime_ref), depending on which parameters are
 * available. This structure includes another hash table link, since separate
 * hashes are maintained for posix_file_runtime structures and posix_file_runtime_ref
 * structures.
 */
106
struct posix_file_runtime_ref
107
{
108
    struct posix_file_runtime* file;
109 110 111 112
    int fd;
    UT_hash_handle hlink;
};

113 114 115 116
/* The posix_runtime structure maintains necessary state for storing
 * POSIX file records and for coordinating with darshan-core at 
 * shutdown time.
 */
117 118
struct posix_runtime
{
119
    struct posix_file_runtime* file_runtime_array;
120
    struct darshan_posix_file* file_record_array;
121
    int file_array_size;
122
    int file_array_ndx;
123 124
    struct posix_file_runtime* file_hash;
    struct posix_file_runtime_ref* fd_hash;
125 126
    void *red_buf;
    int shared_rec_count;
127 128
};

129
static struct posix_runtime *posix_runtime = NULL;
130
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
Shane Snyder's avatar
Shane Snyder committed
131
static int instrumentation_disabled = 0;
132
static int my_rank = -1;
133 134 135

#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)
136

137
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
138
    struct posix_file_runtime* file; \
139 140 141
    char* exclude; \
    int tmp_index = 0; \
    if(__ret < 0) break; \
142
    while((exclude = darshan_path_exclusions[tmp_index])) { \
143 144 145 146 147
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
148
    file = posix_file_by_name_setfd(__path, __ret); \
149
    if(!file) break; \
150
    /* CP_STAT_FILE(file, __path, __ret); */ \
151
    file->file_record->rank = my_rank; \
152
    if(__mode) \
153 154 155 156
        DARSHAN_COUNTER_SET(file->file_record, POSIX_MODE, __mode); \
    file->offset = 0; \
    file->last_byte_written = 0; \
    file->last_byte_read = 0; \
157
    if(__stream_flag)\
158
        DARSHAN_COUNTER_INC(file->file_record, POSIX_FOPENS, 1); \
159
    else \
160 161 162 163
        DARSHAN_COUNTER_INC(file->file_record, POSIX_OPENS, 1); \
    if(DARSHAN_COUNTER_F_VALUE(file->file_record, POSIX_F_OPEN_TIMESTAMP) == 0) \
        DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_OPEN_TIMESTAMP, __tm1); \
    DARSHAN_COUNTER_F_INC_NO_OVERLAP(file->file_record, __tm1, __tm2, file->last_meta_end, POSIX_F_META_TIME); \
164 165 166
} while(0)

#define POSIX_RECORD_READ(__ret, __fd, __pread_flag, __pread_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
167 168
    /* size_t stride; */\
    int64_t this_offset; \
169 170 171 172 173 174
    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; \
175
    if(__pread_flag) \
176 177 178
        this_offset = __pread_offset; \
    else \
        this_offset = file->offset; \
179
    /* file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); */\
180
    if(this_offset > file->last_byte_read) \
181
        DARSHAN_COUNTER_INC(file->file_record, POSIX_SEQ_READS, 1); \
182
    if(this_offset == (file->last_byte_read + 1)) \
183 184
        DARSHAN_COUNTER_INC(file->file_record, POSIX_CONSEC_READS, 1); \
    /* if(this_offset > 0 && this_offset > file->last_byte_read \
185 186 187
        && file->last_byte_read != 0) \
        stride = this_offset - file->last_byte_read - 1; \
    else \
188
        stride = 0; */\
189 190
    file->last_byte_read = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
191 192
    DARSHAN_COUNTER_MAX(file->file_record, POSIX_MAX_BYTE_READ, (this_offset + __ret - 1)); \
    DARSHAN_COUNTER_INC(file->file_record, POSIX_BYTES_READ, __ret); \
193
    if(__stream_flag)\
194
        DARSHAN_COUNTER_INC(file->file_record, POSIX_FREADS, 1); \
195
    else\
196
        DARSHAN_COUNTER_INC(file->file_record, POSIX_READS, 1); \
197 198 199 200 201 202
    /* 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); \
203 204 205 206 207 208 209 210 211 212 213
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); */\
    if(file->last_io_type == POSIX_WRITE) \
        DARSHAN_COUNTER_INC(file->file_record, POSIX_RW_SWITCHES, 1); \
    file->last_io_type = POSIX_READ; \
    DARSHAN_COUNTER_F_INC_NO_OVERLAP(file->file_record, __tm1, __tm2, file->last_read_end, POSIX_F_READ_TIME); \
    if(DARSHAN_COUNTER_F_VALUE(file->file_record, POSIX_F_READ_START_TIMESTAMP) == 0) \
        DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_READ_START_TIMESTAMP, __tm1); \
    DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_READ_END_TIMESTAMP, __tm2); \
    if(DARSHAN_COUNTER_F_VALUE(file->file_record, POSIX_F_MAX_READ_TIME) < __elapsed){ \
        DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_MAX_READ_TIME, __elapsed); \
        DARSHAN_COUNTER_SET(file->file_record, POSIX_MAX_READ_TIME_SIZE, __ret); } \
214 215 216
} while(0)

#define POSIX_RECORD_WRITE(__ret, __fd, __pwrite_flag, __pwrite_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
217 218
    /* size_t stride; */\
    int64_t this_offset; \
219 220 221 222 223 224
    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; \
225
    if(__pwrite_flag) \
226 227 228
        this_offset = __pwrite_offset; \
    else \
        this_offset = file->offset; \
229
    /* file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); */\
230
    if(this_offset > file->last_byte_written) \
231
        DARSHAN_COUNTER_INC(file->file_record, POSIX_SEQ_WRITES, 1); \
232
    if(this_offset == (file->last_byte_written + 1)) \
233 234
        DARSHAN_COUNTER_INC(file->file_record, POSIX_CONSEC_WRITES, 1); \
    /* if(this_offset > 0 && this_offset > file->last_byte_written \
235 236
        && file->last_byte_written != 0) \
        stride = this_offset - file->last_byte_written - 1; \
237
    else \
238
        stride = 0; */\
239 240
    file->last_byte_written = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
241 242
    DARSHAN_COUNTER_MAX(file->file_record, POSIX_MAX_BYTE_WRITTEN, (this_offset + __ret - 1)); \
    DARSHAN_COUNTER_INC(file->file_record, POSIX_BYTES_WRITTEN, __ret); \
243
    if(__stream_flag) \
244
        DARSHAN_COUNTER_INC(file->file_record, POSIX_FWRITES, 1); \
245
    else \
246
        DARSHAN_COUNTER_INC(file->file_record, POSIX_WRITES, 1); \
247 248 249 250 251 252
    /* 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); \
253 254 255 256 257 258 259 260 261 262 263
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); */\
    if(file->last_io_type == POSIX_READ) \
        DARSHAN_COUNTER_INC(file->file_record, POSIX_RW_SWITCHES, 1); \
    file->last_io_type = POSIX_WRITE; \
    DARSHAN_COUNTER_F_INC_NO_OVERLAP(file->file_record, __tm1, __tm2, file->last_write_end, POSIX_F_WRITE_TIME); \
    if(DARSHAN_COUNTER_F_VALUE(file->file_record, POSIX_F_WRITE_START_TIMESTAMP) == 0) \
        DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_WRITE_START_TIMESTAMP, __tm1); \
    DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_WRITE_END_TIMESTAMP, __tm2); \
    if(DARSHAN_COUNTER_F_VALUE(file->file_record, POSIX_F_MAX_WRITE_TIME) < __elapsed){ \
        DARSHAN_COUNTER_F_SET(file->file_record, POSIX_F_MAX_WRITE_TIME, __elapsed); \
        DARSHAN_COUNTER_SET(file->file_record, POSIX_MAX_WRITE_TIME_SIZE, __ret); } \
264
} while(0)
265

266 267 268 269
/**********************************************************
 *      Wrappers for POSIX I/O functions of interest      * 
 **********************************************************/

270
int DARSHAN_DECL(open)(const char *path, int flags, ...)
271 272 273 274 275
{
    int mode = 0;
    int ret;
    double tm1, tm2;

276 277
    MAP_OR_FAIL(open);

278
    if(flags & O_CREAT) 
279 280 281 282 283 284
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

285
        tm1 = darshan_core_wtime();
286
        ret = __real_open(path, flags, mode);
287
        tm2 = darshan_core_wtime();
288 289 290
    }
    else
    {
291
        tm1 = darshan_core_wtime();
292
        ret = __real_open(path, flags);
293
        tm2 = darshan_core_wtime();
294 295
    }

296 297
    POSIX_LOCK();
    posix_runtime_initialize();
298
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
299
    POSIX_UNLOCK();
300 301 302 303

    return(ret);
}

Shane Snyder's avatar
Shane Snyder committed
304 305 306 307 308 309
int DARSHAN_DECL(open64)(const char *path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

310
    MAP_OR_FAIL(open64);
Shane Snyder's avatar
Shane Snyder committed
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331

    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();
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
    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
363

364 365 366 367 368 369
    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
370 371 372 373 374 375
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    POSIX_UNLOCK();

    return(ret);
}

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
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);
}

562 563
int DARSHAN_DECL(close)(int fd)
{
564
    struct posix_file_runtime* file;
565 566 567 568 569 570 571 572 573 574 575
    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
576 577 578
    file = posix_file_by_fd(fd);
    if(file)
    {
579 580 581 582 583 584
        file->last_byte_written = 0;
        file->last_byte_read = 0;
        DARSHAN_COUNTER_F_SET(file->file_record,
            POSIX_F_CLOSE_TIMESTAMP, darshan_core_wtime());
        DARSHAN_COUNTER_F_INC_NO_OVERLAP(file->file_record,
             tm1, tm2, file->last_meta_end, POSIX_F_META_TIME);
Shane Snyder's avatar
Shane Snyder committed
585 586
        posix_file_close_fd(fd);
    }
587 588 589 590 591
    POSIX_UNLOCK();    

    return(ret);
}

592 593 594
/**********************************************************
 * Internal functions for manipulating POSIX module state *
 **********************************************************/
595

596
/* initialize internal POSIX module data structures and register with darshan-core */
597
static void posix_runtime_initialize()
598
{
599 600 601
    int mem_limit;
    struct darshan_module_funcs posix_mod_fns =
    {
Shane Snyder's avatar
Shane Snyder committed
602
        .disable_instrumentation = &posix_disable_instrumentation,
603
        .prepare_for_reduction = &posix_prepare_for_reduction,
604
        .record_reduction_op = &posix_record_reduction_op,
605
        .get_output_data = &posix_get_output_data,
606
        .shutdown = &posix_shutdown
607
    };
608

Shane Snyder's avatar
Shane Snyder committed
609 610
    /* don't do anything if already initialized or instrumenation is disabled */
    if(posix_runtime || instrumentation_disabled)
611
        return;
612

613 614
    /* register the posix module with darshan core */
    darshan_core_register_module(
615
        DARSHAN_POSIX_MOD,
616 617 618 619 620 621 622 623 624 625 626 627 628 629
        &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 */
630
    /* TODO: should we base memory usage off file record or total runtime structure sizes? */
631
    posix_runtime->file_array_size = mem_limit / sizeof(struct darshan_posix_file);
632
    posix_runtime->file_array_ndx = 0;
633 634

    /* allocate array of runtime file records */
635
    posix_runtime->file_runtime_array = malloc(posix_runtime->file_array_size *
636
                                               sizeof(struct posix_file_runtime));
637 638 639
    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)
640 641 642 643
    {
        posix_runtime->file_array_size = 0;
        return;
    }
644
    memset(posix_runtime->file_runtime_array, 0, posix_runtime->file_array_size *
645
           sizeof(struct posix_file_runtime));
646 647
    memset(posix_runtime->file_record_array, 0, posix_runtime->file_array_size *
           sizeof(struct darshan_posix_file));
648

Shane Snyder's avatar
Shane Snyder committed
649 650
    /* TODO: can we move this out of here? perhaps register_module returns rank? */
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
651

652 653
    return;
}
654

655
/* get a POSIX file record for the given file path */
656
static struct posix_file_runtime* posix_file_by_name(const char *name)
657
{
658
    struct posix_file_runtime *file = NULL;
659
    char *newname = NULL;
660
    darshan_record_id file_id;
661

Shane Snyder's avatar
Shane Snyder committed
662
    if(!posix_runtime || instrumentation_disabled)
663
        return(NULL);
664

665
    newname = darshan_clean_file_path(name);
666
    if(!newname)
667
        newname = (char*)name;
668

669
    /* get a unique id for this file from darshan core */
Shane Snyder's avatar
Shane Snyder committed
670
    darshan_core_register_record(
671 672 673
        (void*)newname,
        strlen(newname),
        1,
674
        DARSHAN_POSIX_MOD,
675
        &file_id);
676

677
    /* search the hash table for this file record, and return if found */
678
    HASH_FIND(hlink, posix_runtime->file_hash, &file_id, sizeof(darshan_record_id), file);
679
    if(file)
680
    {
681
        if(newname != name)
682
            free(newname);
683
        return(file);
684 685
    }

686 687 688 689 690 691
    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;
692

693
        /* add new record to file hash table */
694
        HASH_ADD(hlink, posix_runtime->file_hash, file_record->f_id, sizeof(darshan_record_id), file);
695

696 697
        posix_runtime->file_array_ndx++;
    }
698 699

    if(newname != name)
700
        free(newname);
701 702 703
    return(file);
}

704 705 706
/* get a POSIX file record for the given file path, and also create a
 * reference structure using the returned file descriptor
 */
707
static struct posix_file_runtime* posix_file_by_name_setfd(const char* name, int fd)
708
{
709 710
    struct posix_file_runtime* file;
    struct posix_file_runtime_ref* ref;
711

Shane Snyder's avatar
Shane Snyder committed
712
    if(!posix_runtime || instrumentation_disabled)
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
        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
739

740 741 742 743 744 745 746
    ref->file = file;
    ref->fd = fd;    
    HASH_ADD(hlink, posix_runtime->fd_hash, fd, sizeof(int), ref);

    return(file);
}

747
/* get a POSIX file record for the given file descriptor */
748
static struct posix_file_runtime* posix_file_by_fd(int fd)
Shane Snyder's avatar
Shane Snyder committed
749
{
750
    struct posix_file_runtime_ref* ref;
Shane Snyder's avatar
Shane Snyder committed
751 752 753 754 755 756 757 758 759 760 761 762

    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);
}

763
/* free up reference data structures for the given file descriptor */
764 765
static void posix_file_close_fd(int fd)
{
766
    struct posix_file_runtime_ref* ref;
767

Shane Snyder's avatar
Shane Snyder committed
768
    if(!posix_runtime || instrumentation_disabled)
769 770 771 772
        return;

    /* search hash table for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
773
    if(ref)
774 775 776 777 778 779 780
    {
        /* we have a reference, delete it */
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }

    return;
781 782
}

783
/* compare function for sorting file records by descending rank */
784 785 786 787 788 789 790 791 792 793 794 795 796
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;
}

797 798 799
/************************************************************************
 * Functions exported by this module for coordinating with darshan-core *
 ************************************************************************/
800

Shane Snyder's avatar
Shane Snyder committed
801
static void posix_disable_instrumentation()
802
{
803 804
    assert(posix_runtime);

Shane Snyder's avatar
Shane Snyder committed
805 806 807
    POSIX_LOCK();
    instrumentation_disabled = 1;
    POSIX_UNLOCK();
808

Shane Snyder's avatar
Shane Snyder committed
809 810 811
    return;
}

812 813 814 815 816 817 818
static void posix_prepare_for_reduction(
    darshan_record_id *shared_recs,
    int *shared_rec_count,
    void **send_buf,
    void **recv_buf,
    int *rec_size)
{
819
    struct posix_file_runtime *file;
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
    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
856
    /* TODO: cleaner way to do this? */
857 858 859 860 861 862 863
    if(my_rank == 0)
        posix_runtime->red_buf = *recv_buf;
    posix_runtime->shared_rec_count = *shared_rec_count;

    return;
}

864
static void posix_record_reduction_op(
865 866 867 868
    void* infile_v,
    void* inoutfile_v,
    int *len,
    MPI_Datatype *datatype)
Shane Snyder's avatar
Shane Snyder committed
869
{
870 871 872 873
    struct darshan_posix_file tmp_file;
    struct darshan_posix_file *infile = infile_v;
    struct darshan_posix_file *inoutfile = inoutfile_v;
    int i;
874
    int j;
875 876 877

    assert(posix_runtime);

878
    for(i=0; i<*len; i++)
879 880 881 882 883 884
    {
        memset(&tmp_file, 0, sizeof(struct darshan_posix_file));

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

885 886 887 888 889 890 891 892
        /* sum */
        for(j=POSIX_OPENS; j<=POSIX_FWRITES; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + inoutfile->counters[j];
        }

        tmp_file.counters[POSIX_MODE] = infile->counters[POSIX_MODE];

893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914
        /* sum */
        for(j=POSIX_BYTES_READ; j<=POSIX_BYTES_WRITTEN; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + inoutfile->counters[j];
        }

        /* max */
        for(j=POSIX_MAX_BYTE_READ; j<=POSIX_MAX_BYTE_WRITTEN; j++)
        {
            tmp_file.counters[j] = (
                (infile->counters[j] > inoutfile->counters[j]) ?
                infile->counters[j] :
                inoutfile->counters[j]);
        }

        /* sum */
        for(j=POSIX_CONSEC_READS; j<=POSIX_RW_SWITCHES; j++)
        {
            tmp_file.counters[j] = infile->counters[j] +
                inoutfile->counters[j];
        }

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
        /* min non-zero (if available) value */
        for(j=POSIX_F_OPEN_TIMESTAMP; j<=POSIX_F_WRITE_START_TIMESTAMP; j++)
        {
            if(infile->fcounters[j] > inoutfile->fcounters[j] && inoutfile->fcounters[j] > 0)
                tmp_file.fcounters[j] = inoutfile->fcounters[j];
            else
                tmp_file.fcounters[j] = infile->fcounters[j];
        }

        /* max */
        for(j=POSIX_F_READ_END_TIMESTAMP; j<=POSIX_F_CLOSE_TIMESTAMP; j++)
        {
            if(infile->fcounters[j] > inoutfile->fcounters[j])
                tmp_file.fcounters[j] = infile->fcounters[j];
            else
                tmp_file.fcounters[j] = inoutfile->fcounters[j];
        }

        /* sum */
        for(j=POSIX_F_READ_TIME; j<=POSIX_F_META_TIME; j++)
        {
            tmp_file.fcounters[j] = infile->fcounters[j] + inoutfile->fcounters[j];
        }

        /* max (special case) */
        if(infile->fcounters[POSIX_F_MAX_READ_TIME] >
            inoutfile->fcounters[POSIX_F_MAX_READ_TIME])
        {
            tmp_file.fcounters[POSIX_F_MAX_READ_TIME] =
                infile->fcounters[POSIX_F_MAX_READ_TIME];
            tmp_file.counters[POSIX_MAX_READ_TIME_SIZE] =
                infile->counters[POSIX_MAX_READ_TIME_SIZE];
        }
948
        else
949 950 951 952 953 954 955 956 957 958 959 960 961 962 963
        {
            tmp_file.fcounters[POSIX_F_MAX_READ_TIME] =
                inoutfile->fcounters[POSIX_F_MAX_READ_TIME];
            tmp_file.counters[POSIX_MAX_READ_TIME_SIZE] =
                inoutfile->counters[POSIX_MAX_READ_TIME_SIZE];
        }

        if(infile->fcounters[POSIX_F_MAX_WRITE_TIME] >
            inoutfile->fcounters[POSIX_F_MAX_WRITE_TIME])
        {
            tmp_file.fcounters[POSIX_F_MAX_WRITE_TIME] =
                infile->fcounters[POSIX_F_MAX_WRITE_TIME];
            tmp_file.counters[POSIX_MAX_WRITE_TIME_SIZE] =
                infile->counters[POSIX_MAX_WRITE_TIME_SIZE];
        }
964
        else
965 966 967 968 969 970
        {
            tmp_file.fcounters[POSIX_F_MAX_WRITE_TIME] =
                inoutfile->fcounters[POSIX_F_MAX_WRITE_TIME];
            tmp_file.counters[POSIX_MAX_WRITE_TIME_SIZE] =
                inoutfile->counters[POSIX_MAX_WRITE_TIME_SIZE];
        }
971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986

        /* 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
987
    /* TODO: cleaner way to do this? */
988 989 990 991 992 993 994 995 996 997 998 999
    /* 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;
    }
1000 1001 1002

    *buffer = (void *)(posix_runtime->file_record_array);
    *size = posix_runtime->file_array_ndx * sizeof(struct darshan_posix_file);
1003

1004
    return;
1005 1006
}

1007
static void posix_shutdown()
1008
{
1009
    struct posix_file_runtime_ref *ref, *tmp;
1010 1011 1012 1013 1014 1015

    HASH_ITER(hlink, posix_runtime->fd_hash, ref, tmp)
    {
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }
1016

1017
    HASH_CLEAR(hlink, posix_runtime->file_hash); /* these entries are freed all at once below */
1018 1019 1020 1021 1022

    free(posix_runtime->file_runtime_array);
    free(posix_runtime->file_record_array);
    free(posix_runtime);
    posix_runtime = NULL;
1023 1024

    return;
1025 1026
}

1027 1028 1029 1030 1031 1032 1033 1034
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ts=8 sts=4 sw=4 expandtab
 */