darshan-posix.c 46.3 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>
22 23
#define __USE_GNU
#include <pthread.h>
24 25 26

#include "darshan.h"

27
#ifndef HAVE_OFF64_T
28 29 30
typedef int64_t off64_t;
#endif

31 32
extern char* __progname_full;

33
#ifdef DARSHAN_PRELOAD
34 35 36
#define __USE_GNU
#include <dlfcn.h>
#include <stdlib.h>
37

38 39 40 41 42
#define DARSHAN_FORWARD_DECL(name,ret,args) \
  ret (*__real_ ## name)args = NULL;

#define DARSHAN_DECL(__name) __name

43 44 45 46 47 48 49 50 51 52 53 54 55
#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); \
       } \
    }


56
extern double (*__real_PMPI_Wtime)(void);
57

58 59 60 61 62 63 64
#else

#define DARSHAN_FORWARD_DECL(name,ret,args) \
  extern ret __real_ ## name args;

#define DARSHAN_DECL(__name) __wrap_ ## __name

65 66 67 68
#define MAP_OR_FAIL(func)

#define DARSHAN_MPI_CALL(func) func

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
#endif

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(open, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(open64, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(close, int, (int fd));
DARSHAN_FORWARD_DECL(write, ssize_t, (int fd, const void *buf, size_t count));
DARSHAN_FORWARD_DECL(read, ssize_t, (int fd, void *buf, size_t count));
DARSHAN_FORWARD_DECL(lseek, off_t, (int fd, off_t offset, int whence));
DARSHAN_FORWARD_DECL(lseek64, off64_t, (int fd, off64_t offset, int whence));
DARSHAN_FORWARD_DECL(pread, ssize_t, (int fd, 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(pwrite, ssize_t, (int fd, const void *buf, size_t count, off_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));
DARSHAN_FORWARD_DECL(__fxstat, int, (int vers, int fd, struct stat *buf));
DARSHAN_FORWARD_DECL(__fxstat64, int, (int vers, int fd, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__lxstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__lxstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__xstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__xstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(mmap, void*, (void *addr, size_t length, int prot, int flags, int fd, off_t offset));
DARSHAN_FORWARD_DECL(mmap64, void*, (void *addr, size_t length, int prot, int flags, int fd, off64_t offset));
DARSHAN_FORWARD_DECL(fopen, FILE*, (const char *path, const char *mode));
DARSHAN_FORWARD_DECL(fopen64, FILE*, (const char *path, const char *mode));
DARSHAN_FORWARD_DECL(fclose, int, (FILE *fp));
DARSHAN_FORWARD_DECL(fread, size_t, (void *ptr, size_t size, size_t nmemb, FILE *stream));
DARSHAN_FORWARD_DECL(fwrite, size_t, (const void *ptr, size_t size, size_t nmemb, FILE *stream));
DARSHAN_FORWARD_DECL(fseek, int, (FILE *stream, long offset, int whence));
DARSHAN_FORWARD_DECL(fsync, int, (int fd));
DARSHAN_FORWARD_DECL(fdatasync, int, (int fd));

104
pthread_mutex_t cp_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
105 106
struct darshan_job_runtime* darshan_global_job = NULL;
static int my_rank = -1;
107
static struct stat64 cp_stat_buf;
108
static int darshan_mem_alignment = 1;
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124

/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

125 126
static double posix_wtime(void);

127 128
static void cp_access_counter(struct darshan_file_runtime* file, ssize_t size,     enum cp_counter_type type);

Philip Carns's avatar
Philip Carns committed
129
#define CP_RECORD_WRITE(__ret, __fd, __count, __pwrite_flag, __pwrite_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
130
    size_t stride; \
Philip Carns's avatar
Philip Carns committed
131
    int64_t this_offset; \
132 133
    int64_t file_alignment; \
    struct darshan_file_runtime* file; \
134
    double __elapsed = __tm2-__tm1; \
135 136 137
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
Philip Carns's avatar
Philip Carns committed
138 139 140 141
    if(__pwrite_flag) \
        this_offset = __pwrite_offset; \
    else \
        this_offset = file->offset; \
142
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
Philip Carns's avatar
Philip Carns committed
143
    if(this_offset > file->last_byte_written) \
144
        CP_INC(file, CP_SEQ_WRITES, 1); \
Philip Carns's avatar
Philip Carns committed
145
    if(this_offset == (file->last_byte_written + 1)) \
146
        CP_INC(file, CP_CONSEC_WRITES, 1); \
Philip Carns's avatar
Philip Carns committed
147
    if(this_offset > 0 && this_offset > file->last_byte_written \
148
        && file->last_byte_written != 0) \
Philip Carns's avatar
Philip Carns committed
149
        stride = this_offset - file->last_byte_written - 1; \
150 151
    else \
        stride = 0; \
Philip Carns's avatar
Philip Carns committed
152 153 154
    file->last_byte_written = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
    CP_MAX(file, CP_MAX_BYTE_WRITTEN, (this_offset + __ret -1)); \
155 156 157 158 159 160 161 162 163
    CP_INC(file, CP_BYTES_WRITTEN, __ret); \
    if(__stream_flag) \
        CP_INC(file, CP_POSIX_FWRITES, 1); \
    else \
        CP_INC(file, CP_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); \
Philip Carns's avatar
Philip Carns committed
164
    if(file_alignment && (this_offset % file_alignment) != 0) \
165 166 167 168 169
        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; \
170
    CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (__elapsed)); \
171 172 173
    if(CP_F_VALUE(file, CP_F_WRITE_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_WRITE_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_WRITE_END_TIMESTAMP, __tm2); \
174 175 176
    if(CP_F_VALUE(file, CP_F_MAX_WRITE_TIME) < __elapsed){ \
        CP_F_SET(file, CP_F_MAX_WRITE_TIME, __elapsed); \
        CP_SET(file, CP_MAX_WRITE_TIME_SIZE, __ret); } \
177 178
} while(0)

179
#define CP_RECORD_READ(__ret, __fd, __count, __pread_flag, __pread_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
180
    size_t stride; \
181
    int64_t this_offset; \
182 183
    struct darshan_file_runtime* file; \
    int64_t file_alignment; \
184
    double __elapsed = __tm2-__tm1; \
185 186 187
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
188 189 190 191
    if(__pread_flag)\
        this_offset = __pread_offset; \
    else \
        this_offset = file->offset; \
192
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
193
    if(this_offset > file->last_byte_read) \
194
        CP_INC(file, CP_SEQ_READS, 1); \
195
    if(this_offset == (file->last_byte_read + 1)) \
196
        CP_INC(file, CP_CONSEC_READS, 1); \
197
    if(this_offset > 0 && this_offset > file->last_byte_read \
198
        && file->last_byte_read != 0) \
199
        stride = this_offset - file->last_byte_read - 1; \
200 201
    else \
        stride = 0; \
202 203 204
    file->last_byte_read = this_offset + __ret - 1; \
    CP_MAX(file, CP_MAX_BYTE_READ, (this_offset + __ret -1)); \
    file->offset = this_offset + __ret; \
205 206 207 208 209 210 211 212 213
    CP_INC(file, CP_BYTES_READ, __ret); \
    if(__stream_flag)\
        CP_INC(file, CP_POSIX_FREADS, 1); \
    else\
        CP_INC(file, CP_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); \
214
    if(file_alignment && (this_offset % file_alignment) != 0) \
215 216 217 218 219
        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; \
220
    CP_F_INC(file, CP_F_POSIX_READ_TIME, (__elapsed)); \
221 222 223
    if(CP_F_VALUE(file, CP_F_READ_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_READ_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_READ_END_TIMESTAMP, __tm2); \
224 225 226
    if(CP_F_VALUE(file, CP_F_MAX_READ_TIME) < __elapsed){ \
        CP_F_SET(file, CP_F_MAX_READ_TIME, __elapsed); \
        CP_SET(file, CP_MAX_READ_TIME_SIZE, __ret); } \
227 228
} while(0)

229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246
#define CP_LOOKUP_RECORD_STAT(__path, __statbuf, __tm1, __tm2) do { \
    char* exclude; \
    int tmp_index = 0; \
    struct darshan_file_runtime* file; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
    file = darshan_file_by_name(__path); \
    if (file) \
    { \
        CP_RECORD_STAT(file, __statbuf, __tm1, __tm2); \
    } \
} while(0)

    
247 248 249 250 251 252 253 254 255 256 257
#define CP_RECORD_STAT(__file, __statbuf, __tm1, __tm2) do { \
    if(!CP_VALUE((__file), CP_FILE_ALIGNMENT)){ \
        CP_SET((__file), CP_DEVICE, (__statbuf)->st_dev); \
        CP_SET((__file), CP_FILE_ALIGNMENT, (__statbuf)->st_blksize); \
        CP_SET((__file), CP_SIZE_AT_OPEN, (__statbuf)->st_size); \
    }\
    (__file)->log_file->rank = my_rank; \
    CP_F_INC(__file, CP_F_POSIX_META_TIME, (__tm2-__tm1)); \
    CP_INC(__file, CP_POSIX_STATS, 1); \
} while(0)

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275
static inline dev_t get_device(const char* path, struct stat64* statbuf)
{
    dev_t device = 0;

#ifdef __CP_ST_DEV_WORKAROUND
    struct stat64 dirstat;
    char* tmp_path = strdup(path);
    char* parent = dirname(tmp_path);
    if(parent && stat64(parent, &dirstat) == 0)
        device = dirstat.st_dev;
    if(tmp_path) free(tmp_path);
#else
    device = statbuf->st_dev;
#endif

    return(device);
}

276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292
#define CP_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
    struct darshan_file_runtime* file; \
    char* exclude; \
    int tmp_index = 0; \
    int hash_index; \
    if(__ret < 0) break; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
    file = darshan_file_by_name(__path); \
    if(!file) break; \
    if(file->fd != -1) break; /* TODO: handle multiple concurrent opens */ \
    file->fd = __ret; \
    if(!CP_VALUE(file, CP_FILE_ALIGNMENT)){ \
293 294
        if(stat64(__path, &cp_stat_buf) == 0) { \
            CP_SET(file, CP_DEVICE, get_device(__path, &cp_stat_buf)); \
295
            CP_SET(file, CP_FILE_ALIGNMENT, cp_stat_buf.st_blksize); \
296
            CP_SET(file, CP_SIZE_AT_OPEN, cp_stat_buf.st_size); \
297
        }\
298 299 300 301 302 303 304 305 306 307 308 309
    }\
    file->log_file->rank = my_rank; \
    if(__mode) \
        CP_SET(file, CP_MODE, __mode); \
    file->offset = 0; \
    file->last_byte_written = 0; \
    file->last_byte_read = 0; \
    if(__stream_flag)\
        CP_INC(file, CP_POSIX_FOPENS, 1); \
    else \
        CP_INC(file, CP_POSIX_OPENS, 1); \
    if(CP_F_VALUE(file, CP_F_OPEN_TIMESTAMP) == 0) \
310
        CP_F_SET(file, CP_F_OPEN_TIMESTAMP, posix_wtime()); \
311 312 313 314 315 316 317 318 319
    CP_F_INC(file, CP_F_POSIX_META_TIME, (__tm2-__tm1)); \
    hash_index = file->fd & CP_HASH_MASK; \
    file->fd_prev = NULL; \
    file->fd_next = darshan_global_job->fd_table[hash_index]; \
    if(file->fd_next) \
        file->fd_next->fd_prev = file; \
    darshan_global_job->fd_table[hash_index] = file; \
} while (0)

320
int DARSHAN_DECL(close)(int fd)
321 322 323 324 325 326 327
{
    struct darshan_file_runtime* file;
    int hash_index;
    int tmp_fd = fd;
    double tm1, tm2;
    int ret;

328 329
    MAP_OR_FAIL(close);

330 331 332 333 334 335 336 337 338 339 340
    tm1 = darshan_wtime();
    ret = __real_close(fd);
    tm2 = darshan_wtime();

    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->fd = -1;
        file->last_byte_written = 0;
        file->last_byte_read = 0;
341
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
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
        CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
        if(file->fd_prev == NULL)
        {
            /* head of fd hash table list */
            hash_index = tmp_fd & CP_HASH_MASK;
            darshan_global_job->fd_table[hash_index] = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = NULL;
        }
        else
        {
            if(file->fd_prev)
                file->fd_prev->fd_next = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = file->fd_prev;
        }
        file->fd_prev = NULL;
        file->fd_next = NULL;
        darshan_global_job->darshan_mru_file = file; /* in case we open it again */
    }
    CP_UNLOCK();

    return(ret);
}

367
int DARSHAN_DECL(fclose)(FILE *fp)
368 369 370 371 372 373 374
{
    struct darshan_file_runtime* file;
    int hash_index;
    int tmp_fd = fileno(fp);
    double tm1, tm2;
    int ret;

375 376
    MAP_OR_FAIL(fclose);

377 378 379 380 381 382 383 384 385 386 387
    tm1 = darshan_wtime();
    ret = __real_fclose(fp);
    tm2 = darshan_wtime();
    
    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->fd = -1;
        file->last_byte_written = 0;
        file->last_byte_read = 0;
388
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
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
        CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
        if(file->fd_prev == NULL)
        {
            /* head of fd hash table list */
            hash_index = tmp_fd & CP_HASH_MASK;
            darshan_global_job->fd_table[hash_index] = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = NULL;
        }
        else
        {
            if(file->fd_prev)
                file->fd_prev->fd_next = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = file->fd_prev;
        }
        file->fd_prev = NULL;
        file->fd_next = NULL;
        darshan_global_job->darshan_mru_file = file; /* in case we open it again */
    }
    CP_UNLOCK();

    return(ret);
}


415
int DARSHAN_DECL(fsync)(int fd)
416 417 418 419 420
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

421 422
    MAP_OR_FAIL(fsync);

423 424 425 426 427 428 429 430 431 432 433
    tm1 = darshan_wtime();
    ret = __real_fsync(fd);
    tm2 = darshan_wtime();

    if(ret < 0)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
434
        CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (tm2-tm1));
435 436 437 438 439 440 441
        CP_INC(file, CP_POSIX_FSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}

442
int DARSHAN_DECL(fdatasync)(int fd)
443 444 445 446 447
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

448 449
    MAP_OR_FAIL(fdatasync);

450 451 452 453 454 455 456 457 458 459
    tm1 = darshan_wtime();
    ret = __real_fdatasync(fd);
    tm2 = darshan_wtime();
    if(ret < 0)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
460
        CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (tm2-tm1));
461 462 463 464 465 466 467 468
        CP_INC(file, CP_POSIX_FDSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}


469
void* DARSHAN_DECL(mmap64)(void *addr, size_t length, int prot, int flags,
470 471 472 473 474
    int fd, off64_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

475 476
    MAP_OR_FAIL(mmap64);

477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
    ret = __real_mmap64(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_INC(file, CP_POSIX_MMAPS, 1);
    }
    CP_UNLOCK();

    return(ret);
}


493
void* DARSHAN_DECL(mmap)(void *addr, size_t length, int prot, int flags,
494 495 496 497 498
    int fd, off_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

499 500
    MAP_OR_FAIL(mmap);

501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
    ret = __real_mmap(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_INC(file, CP_POSIX_MMAPS, 1);
    }
    CP_UNLOCK();

    return(ret);
}

516
int DARSHAN_DECL(creat)(const char* path, mode_t mode)
517 518 519 520
{
    int ret;
    double tm1, tm2;

521 522
    MAP_OR_FAIL(creat);

523 524 525 526 527 528 529 530 531 532 533
    tm1 = darshan_wtime();
    ret = __real_creat(path, mode);
    tm2 = darshan_wtime();

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

534
int DARSHAN_DECL(creat64)(const char* path, mode_t mode)
535 536 537 538
{
    int ret;
    double tm1, tm2;

539 540
    MAP_OR_FAIL(creat64);

541 542 543 544 545 546 547 548 549 550 551
    tm1 = darshan_wtime();
    ret = __real_creat64(path, mode);
    tm2 = darshan_wtime();

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

552
int DARSHAN_DECL(open64)(const char* path, int flags, ...)
553 554 555 556 557
{
    int mode = 0;
    int ret;
    double tm1, tm2;

558 559
    MAP_OR_FAIL(open64);

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

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

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

585
int DARSHAN_DECL(open)(const char *path, int flags, ...)
586 587 588 589 590
{
    int mode = 0;
    int ret;
    double tm1, tm2;

591 592
    MAP_OR_FAIL(open);

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

        tm1 = darshan_wtime();
        ret = __real_open(path, flags, mode);
        tm2 = darshan_wtime();
    }
    else
    {
        tm1 = darshan_wtime();
        ret = __real_open(path, flags);
        tm2 = darshan_wtime();
    }

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

618
FILE* DARSHAN_DECL(fopen64)(const char *path, const char *mode)
619 620 621 622 623
{
    FILE* ret;
    int fd;
    double tm1, tm2;

624 625
    MAP_OR_FAIL(fopen64);

626 627 628 629 630 631 632 633 634 635 636 637 638 639 640
    tm1 = darshan_wtime();
    ret = __real_fopen64(path, mode);
    tm2 = darshan_wtime();
    if(ret == 0)
        fd = -1;
    else
        fd = fileno(ret);

    CP_LOCK();
    CP_RECORD_OPEN(fd, path, 0, 1, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

641
FILE* DARSHAN_DECL(fopen)(const char *path, const char *mode)
642 643 644 645 646
{
    FILE* ret;
    int fd;
    double tm1, tm2;

647 648
    MAP_OR_FAIL(fopen);

649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
    tm1 = darshan_wtime();
    ret = __real_fopen(path, mode);
    tm2 = darshan_wtime();
    if(ret == 0)
        fd = -1;
    else
        fd = fileno(ret);

    CP_LOCK();
    CP_RECORD_OPEN(fd, path, 0, 1, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

664
int DARSHAN_DECL(__xstat64)(int vers, const char *path, struct stat64 *buf)
665 666 667 668
{
    int ret;
    double tm1, tm2;

669 670
    MAP_OR_FAIL(__xstat64);

671 672 673 674 675 676 677
    tm1 = darshan_wtime();
    ret = __real___xstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
678
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
679 680 681 682 683
    CP_UNLOCK();

    return(ret);
}

684
int DARSHAN_DECL(__lxstat64)(int vers, const char *path, struct stat64 *buf)
685 686 687 688
{
    int ret;
    double tm1, tm2;

689 690
    MAP_OR_FAIL(__lxstat64);

691 692 693 694 695 696 697
    tm1 = darshan_wtime();
    ret = __real___lxstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
698
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
699 700 701 702 703
    CP_UNLOCK();

    return(ret);
}

704
int DARSHAN_DECL(__fxstat64)(int vers, int fd, struct stat64 *buf)
705 706 707 708 709
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

710 711
    MAP_OR_FAIL(__fxstat64);

712 713 714 715 716 717 718 719 720 721 722 723 724 725
    tm1 = darshan_wtime();
    ret = __real___fxstat64(vers, fd, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    /* skip logging if this was triggered internally */
    if(buf == &cp_stat_buf)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
726
        CP_RECORD_STAT(file, buf, tm1, tm2);
727 728 729 730 731 732 733
    }
    CP_UNLOCK();

    return(ret);
}


734
int DARSHAN_DECL(__xstat)(int vers, const char *path, struct stat *buf)
735 736 737 738
{
    int ret;
    double tm1, tm2;

739 740
    MAP_OR_FAIL(__xstat);

741 742 743 744 745 746 747
    tm1 = darshan_wtime();
    ret = __real___xstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
748
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
749 750 751 752 753
    CP_UNLOCK();

    return(ret);
}

754
int DARSHAN_DECL(__lxstat)(int vers, const char *path, struct stat *buf)
755 756 757 758
{
    int ret;
    double tm1, tm2;

759 760
    MAP_OR_FAIL(__lxstat);

761 762 763 764 765 766 767
    tm1 = darshan_wtime();
    ret = __real___lxstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
768
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
769 770 771 772 773
    CP_UNLOCK();

    return(ret);
}

774
int DARSHAN_DECL(__fxstat)(int vers, int fd, struct stat *buf)
775 776 777 778 779
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

780 781
    MAP_OR_FAIL(__fxstat);

782 783 784 785 786 787 788
    tm1 = darshan_wtime();
    ret = __real___fxstat(vers, fd, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    /* skip logging if this was triggered internally */
789
    if((void*)buf == (void*)&cp_stat_buf)
790 791 792 793 794 795
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
796
        CP_RECORD_STAT(file, buf, tm1, tm2);
797 798 799 800 801 802
    }
    CP_UNLOCK();

    return(ret);
}

803
ssize_t DARSHAN_DECL(pread64)(int fd, void *buf, size_t count, off64_t offset)
804 805 806 807 808
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

809 810
    MAP_OR_FAIL(pread64);

811
    if((unsigned long)buf % darshan_mem_alignment == 0)
812 813 814 815 816 817
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
818
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
819 820 821 822
    CP_UNLOCK();
    return(ret);
}

823
ssize_t DARSHAN_DECL(pread)(int fd, void *buf, size_t count, off_t offset)
824 825 826 827 828
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

829 830
    MAP_OR_FAIL(pread);

831
    if((unsigned long)buf % darshan_mem_alignment == 0)
832 833 834 835 836 837
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
838
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
839 840 841 842 843
    CP_UNLOCK();
    return(ret);
}


844
ssize_t DARSHAN_DECL(pwrite)(int fd, const void *buf, size_t count, off_t offset)
845 846 847 848 849
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

850 851
    MAP_OR_FAIL(pwrite);

852
    if((unsigned long)buf % darshan_mem_alignment == 0)
853 854 855 856 857 858
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
Philip Carns's avatar
Philip Carns committed
859
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
860 861 862 863
    CP_UNLOCK();
    return(ret);
}

864
ssize_t DARSHAN_DECL(pwrite64)(int fd, const void *buf, size_t count, off64_t offset)
865 866 867 868 869
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

870 871
    MAP_OR_FAIL(pwrite64);

872
    if((unsigned long)buf % darshan_mem_alignment == 0)
873 874 875 876 877 878
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
Philip Carns's avatar
Philip Carns committed
879
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
880 881 882 883
    CP_UNLOCK();
    return(ret);
}

884
ssize_t DARSHAN_DECL(readv)(int fd, const struct iovec *iov, int iovcnt)
885 886 887 888 889 890
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

891 892
    MAP_OR_FAIL(readv);

893 894
    for(i=0; i<iovcnt; i++)
    {
895
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
896 897 898 899 900 901 902
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_readv(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
903
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
904 905 906 907
    CP_UNLOCK();
    return(ret);
}

908
ssize_t DARSHAN_DECL(writev)(int fd, const struct iovec *iov, int iovcnt)
909 910 911 912 913 914
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

915 916
    MAP_OR_FAIL(writev);

917 918
    for(i=0; i<iovcnt; i++)
    {
919
        if(!((unsigned long)iov[i].iov_base % darshan_mem_alignment == 0))
920 921 922 923 924 925 926
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_writev(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
Philip Carns's avatar
Philip Carns committed
927
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
928 929 930 931
    CP_UNLOCK();
    return(ret);
}

932
size_t DARSHAN_DECL(fread)(void *ptr, size_t size, size_t nmemb, FILE *stream)
933 934 935 936 937
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

938 939
    MAP_OR_FAIL(fread);

940
    if((unsigned long)ptr % darshan_mem_alignment == 0)
941 942 943 944 945 946 947
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fread(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
948
        CP_RECORD_READ(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
949
    else
950
        CP_RECORD_READ(ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
951 952 953 954
    CP_UNLOCK();
    return(ret);
}

955
ssize_t DARSHAN_DECL(read)(int fd, void *buf, size_t count)
956 957 958 959 960
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

961 962
    MAP_OR_FAIL(read);

963
    if((unsigned long)buf % darshan_mem_alignment == 0)
964 965 966 967 968 969
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_read(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
970
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
971 972 973 974
    CP_UNLOCK();
    return(ret);
}

975
ssize_t DARSHAN_DECL(write)(int fd, const void *buf, size_t count)
976 977 978 979 980
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

981 982
    MAP_OR_FAIL(write);

983
    if((unsigned long)buf % darshan_mem_alignment == 0)
984 985 986 987 988 989
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_write(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
Philip Carns's avatar
Philip Carns committed
990
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
991 992 993 994
    CP_UNLOCK();
    return(ret);
}

995
size_t DARSHAN_DECL(fwrite)(const void *ptr, size_t size, size_t nmemb, FILE *stream)
996 997 998 999 1000
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

1001 1002
    MAP_OR_FAIL(fwrite);

1003
    if((unsigned long)ptr % darshan_mem_alignment == 0)
1004 1005 1006 1007 1008 1009 1010
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fwrite(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
Philip Carns's avatar
Philip Carns committed
1011
        CP_RECORD_WRITE(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
1012
    else
Philip Carns's avatar
Philip Carns committed
1013
        CP_RECORD_WRITE(ret, fileno(stream), 0, 0, 0, aligned_flag, 1, tm1, tm2);
1014 1015 1016 1017
    CP_UNLOCK();
    return(ret);
}

1018
off64_t DARSHAN_DECL(lseek64)(int fd, off64_t offset, int whence)
1019
{
Philip Carns's avatar
Philip Carns committed
1020
    off64_t ret;
1021 1022 1023
    struct darshan_file_runtime* file;
    double tm1, tm2;

1024 1025
    MAP_OR_FAIL(lseek64);

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
    tm1 = darshan_wtime();
    ret = __real_lseek64(fd, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fd);
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

1044
off_t DARSHAN_DECL(lseek)(int fd, off_t offset, int whence)
1045 1046 1047 1048 1049
{
    off_t ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

1050 1051
    MAP_OR_FAIL(lseek);

1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069
    tm1 = darshan_wtime();
    ret = __real_lseek(fd, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fd);
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

1070
int DARSHAN_DECL(fseek)(FILE *stream, long offset, int whence)
1071 1072 1073 1074 1075
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

1076 1077
    MAP_OR_FAIL(fseek);

1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
    tm1 = darshan_wtime();
    ret = __real_fseek(stream, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fileno(stream));
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_FSEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

void darshan_finalize(struct darshan_job_runtime* job)
{
    if(!job)
    {
        return;
    }

    free(job);
}

void darshan_initialize(int argc, char** argv,  int nprocs, int rank)
{
    int i;
    char* disable;
    char* disable_timing;
1111
    char* envstr;
1112 1113 1114
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
1115 1116
    int ret;
    int tmpval;
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131

    disable = getenv("DARSHAN_DISABLE");
    if(disable)
    {
        /* turn off tracing */
        return;
    }

    disable_timing = getenv("DARSHAN_DISABLE_TIMING");

    if(darshan_global_job != NULL)
    {
        return;
    }

1132 1133 1134
    #if (__CP_MEM_ALIGNMENT < 1)
        #error Darshan must be configured with a positive value for --with-mem-align
    #endif
1135 1136 1137
    envstr = getenv("DARSHAN_MEMALIGN");
    if (envstr)
    {
1138 1139 1140 1141 1142 1143
        ret = sscanf(envstr, "%d", &tmpval);
        /* silently ignore if the env variable is set poorly */
        if(ret == 1 && tmpval > 0)
        {
            darshan_mem_alignment = tmpval;
        }
1144 1145 1146 1147 1148 1149
    }
    else
    {
        darshan_mem_alignment = __CP_MEM_ALIGNMENT;
    }

1150 1151 1152 1153 1154 1155
    /* avoid floating point errors on faulty input */
    if (darshan_mem_alignment < 1)
    {
        darshan_mem_alignment = 1;
    }

1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
    /* allocate structure to track darshan_global_job information */
    darshan_global_job = malloc(sizeof(*darshan_global_job));
    if(!darshan_global_job)
    {
        return;
    }
    memset(darshan_global_job, 0, sizeof(*darshan_global_job));

    if(disable_timing)
    {
        darshan_global_job->flags |= CP_FLAG_NOTIMING;
    }

    /* set up file records */
    for(i=0; i<CP_MAX_FILES; i++)
    {
        darshan_global_job->file_runtime_array[i].log_file = 
            &darshan_global_job->file_array[i];
        darshan_global_job->file_runtime_array[i].fd = -1;
1175
        darshan_global_job->file_runtime_array[i].ncid = -1;
1176
        darshan_global_job->file_runtime_array[i].hid = -1;
1177 1178 1179 1180
        darshan_global_job->file_runtime_array[i].fh = MPI_FILE_NULL;
    }

    strcpy(darshan_global_job->log_job.version_string, CP_VERSION);
1181
    darshan_global_job->log_job.magic_nr = CP_MAGIC_NR;
1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256
    darshan_global_job->log_job.uid = getuid();
    darshan_global_job->log_job.start_time = time(NULL);
    darshan_global_job->log_job.nprocs = nprocs;
    my_rank = rank;

    /* record exe and arguments */
    for(i=0; i<argc; i++)
    {
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, argv[i], chars_left);
        if(i < (argc-1))
        {
            chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
            strncat(darshan_global_job->exe, " ", chars_left);
        }
    }

    /* if we don't see any arguments, then use glibc symbol to get
     * program name at least (this happens in fortran)
     */
    if(argc == 0)
    {
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, __progname_full, chars_left);
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, " <unknown args>", chars_left);
    }

    if(chars_left == 0)
    {
        /* we ran out of room; mark that string was truncated */
        truncate_offset = CP_EXE_LEN - strlen(truncate_string);
        sprintf(&darshan_global_job->exe[truncate_offset], "%s", 
            truncate_string);
    }
}

/* darshan_condense()
 *
 * collapses all file statistics into a single unified set of counters; used
 * when we have opened too many files to track independently
 */
void darshan_condense(void)
{
    struct darshan_file_runtime* base_file;
    struct darshan_file_runtime* iter_file;
    int i;
    int j;

    if(!darshan_global_job)
        return;

    base_file = &darshan_global_job->file_runtime_array[0];

    /* iterate through files */
    for(j=1; j<darshan_global_job->file_count; j++)
    {