darshan-posix.c 57.3 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
#include <limits.h>
23
#include <aio.h>
24 25
#define __USE_GNU
#include <pthread.h>
26 27 28

#include "darshan.h"

29
#ifndef HAVE_OFF64_T
30 31
typedef int64_t off64_t;
#endif
32 33 34
#ifndef HAVE_AIOCB64
#define aiocb64 aiocb
#endif
35

36 37
extern char* __progname_full;

38
#ifdef DARSHAN_PRELOAD
39 40 41
#define __USE_GNU
#include <dlfcn.h>
#include <stdlib.h>
42

43 44 45 46 47
#define DARSHAN_FORWARD_DECL(name,ret,args) \
  ret (*__real_ ## name)args = NULL;

#define DARSHAN_DECL(__name) __name

48 49 50 51 52 53 54 55 56 57 58 59 60
#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); \
       } \
    }


61
extern double (*__real_PMPI_Wtime)(void);
62

63 64 65 66 67 68 69
#else

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

#define DARSHAN_DECL(__name) __wrap_ ## __name

70 71 72 73
#define MAP_OR_FAIL(func)

#define DARSHAN_MPI_CALL(func) func

74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107
#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));
108
DARSHAN_FORWARD_DECL(aio_read, int, (struct aiocb *aiocbp));
109
DARSHAN_FORWARD_DECL(aio_read64, int, (struct aiocb64 *aiocbp));
110
DARSHAN_FORWARD_DECL(aio_write, int, (struct aiocb *aiocbp));
111
DARSHAN_FORWARD_DECL(aio_write64, int, (struct aiocb64 *aiocbp));
112
DARSHAN_FORWARD_DECL(lio_listio, int, (int mode, struct aiocb *const aiocb_list[], int nitems, struct sigevent *sevp));
113
DARSHAN_FORWARD_DECL(lio_listio64, int, (int mode, struct aiocb64 *const aiocb_list[], int nitems, struct sigevent *sevp));
114
DARSHAN_FORWARD_DECL(aio_return, ssize_t, (struct aiocb *aiocbp));
115
DARSHAN_FORWARD_DECL(aio_return64, ssize_t, (struct aiocb64 *aiocbp));
116

117
pthread_mutex_t cp_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
118 119
struct darshan_job_runtime* darshan_global_job = NULL;
static int my_rank = -1;
120
static struct stat64 cp_stat_buf;
121
static int darshan_mem_alignment = 1;
122

123 124 125 126
/* struct to track information about aio operations in flight */
struct darshan_aio_tracker
{
    double tm1;
127
    void *aiocbp;
128 129 130
    struct darshan_aio_tracker* next;
};

131 132 133 134 135 136 137 138 139 140 141 142 143 144 145
/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

146 147
static double posix_wtime(void);

148 149
static void cp_access_counter(struct darshan_file_runtime* file, ssize_t size,     enum cp_counter_type type);

150 151 152 153 154 155 156 157
static struct darshan_file_ref* ref_by_handle(
    const void* handle,
    int handle_sz,
    enum darshan_handle_type handle_type);

static struct darshan_file_runtime* darshan_file_by_fd(int fd);
static void darshan_file_close_fd(int fd);
static struct darshan_file_runtime* darshan_file_by_name_setfd(const char* name, int fd);
158
static char* clean_path(const char* path);
159 160
static void darshan_aio_tracker_add(int fd, void *aiocbp);
static struct darshan_aio_tracker* darshan_aio_tracker_del(int fd, void *aiocbp);
161

162
#define CP_RECORD_WRITE(__ret, __fd, __count, __pwrite_flag, __pwrite_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
163
    size_t stride; \
164
    int64_t this_offset; \
165 166
    int64_t file_alignment; \
    struct darshan_file_runtime* file; \
167
    double __elapsed = __tm2-__tm1; \
168 169 170
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
171 172 173 174
    if(__pwrite_flag) \
        this_offset = __pwrite_offset; \
    else \
        this_offset = file->offset; \
175
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
176
    if(this_offset > file->last_byte_written) \
177
        CP_INC(file, CP_SEQ_WRITES, 1); \
178
    if(this_offset == (file->last_byte_written + 1)) \
179
        CP_INC(file, CP_CONSEC_WRITES, 1); \
180
    if(this_offset > 0 && this_offset > file->last_byte_written \
181
        && file->last_byte_written != 0) \
182
        stride = this_offset - file->last_byte_written - 1; \
183 184
    else \
        stride = 0; \
185 186 187
    file->last_byte_written = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
    CP_MAX(file, CP_MAX_BYTE_WRITTEN, (this_offset + __ret -1)); \
188 189 190 191 192 193 194 195 196
    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); \
197
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
198 199 200 201 202
        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; \
203
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_posix_write_end, CP_F_POSIX_WRITE_TIME); \
204 205 206
    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); \
207 208 209
    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); } \
210 211
} while(0)

212
#define CP_RECORD_READ(__ret, __fd, __count, __pread_flag, __pread_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
213
    size_t stride; \
214
    int64_t this_offset; \
215 216
    struct darshan_file_runtime* file; \
    int64_t file_alignment; \
217
    double __elapsed = __tm2-__tm1; \
218 219 220
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
221 222 223 224
    if(__pread_flag)\
        this_offset = __pread_offset; \
    else \
        this_offset = file->offset; \
225
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
226
    if(this_offset > file->last_byte_read) \
227
        CP_INC(file, CP_SEQ_READS, 1); \
228
    if(this_offset == (file->last_byte_read + 1)) \
229
        CP_INC(file, CP_CONSEC_READS, 1); \
230
    if(this_offset > 0 && this_offset > file->last_byte_read \
231
        && file->last_byte_read != 0) \
232
        stride = this_offset - file->last_byte_read - 1; \
233 234
    else \
        stride = 0; \
235 236 237
    file->last_byte_read = this_offset + __ret - 1; \
    CP_MAX(file, CP_MAX_BYTE_READ, (this_offset + __ret -1)); \
    file->offset = this_offset + __ret; \
238 239 240 241 242 243 244 245 246
    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); \
247
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
248 249 250 251 252
        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; \
253
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_posix_read_end, CP_F_POSIX_READ_TIME); \
254 255 256
    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); \
257 258 259
    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); } \
260 261
} while(0)

262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279
#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)

    
280
#define CP_RECORD_STAT(__file, __statbuf, __tm1, __tm2) do { \
281
    if(!CP_VALUE((__file), CP_POSIX_STATS) && !CP_VALUE((__file), CP_POSIX_OPENS)){ \
282 283 284 285
        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; \
286
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME); \
287 288 289
    CP_INC(__file, CP_POSIX_STATS, 1); \
} while(0)

290
#ifdef __CP_STAT_AT_OPEN
291
#define CP_STAT_FILE(_f, _p, _r) do { \
292
    if(!CP_VALUE((_f), CP_POSIX_STATS) && !CP_VALUE((_f), CP_POSIX_OPENS)){ \
293
        if(fstat64(_r, &cp_stat_buf) == 0) { \
294 295 296 297 298 299
            CP_SET(_f, CP_FILE_ALIGNMENT, cp_stat_buf.st_blksize); \
            CP_SET(_f, CP_SIZE_AT_OPEN, cp_stat_buf.st_size); \
        }\
    }\
}while(0)
#else
300
#define CP_STAT_FILE(_f, _p, _r) do { }while(0)
301 302
#endif

303 304 305 306 307 308 309 310 311 312 313
#define CP_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
    struct darshan_file_runtime* file; \
    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; \
314
    file = darshan_file_by_name_setfd(__path, __ret); \
315
    if(!file) break; \
316
    CP_STAT_FILE(file, __path, __ret); \
317 318 319 320 321 322 323 324 325 326 327
    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) \
328
        CP_F_SET(file, CP_F_OPEN_TIMESTAMP, posix_wtime()); \
329
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME); \
330 331
} while (0)

332
int DARSHAN_DECL(close)(int fd)
333 334 335 336 337 338
{
    struct darshan_file_runtime* file;
    int tmp_fd = fd;
    double tm1, tm2;
    int ret;

339 340
    MAP_OR_FAIL(close);

341 342 343 344 345 346 347 348 349 350
    tm1 = darshan_wtime();
    ret = __real_close(fd);
    tm2 = darshan_wtime();

    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->last_byte_written = 0;
        file->last_byte_read = 0;
351
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
352
        CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME);
353
        darshan_file_close_fd(tmp_fd);
354 355 356 357 358 359
    }
    CP_UNLOCK();

    return(ret);
}

360
int DARSHAN_DECL(fclose)(FILE *fp)
361 362 363 364 365 366
{
    struct darshan_file_runtime* file;
    int tmp_fd = fileno(fp);
    double tm1, tm2;
    int ret;

367 368
    MAP_OR_FAIL(fclose);

369 370 371 372 373 374 375 376 377 378
    tm1 = darshan_wtime();
    ret = __real_fclose(fp);
    tm2 = darshan_wtime();
    
    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->last_byte_written = 0;
        file->last_byte_read = 0;
379
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
380
        CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME);
381
        darshan_file_close_fd(tmp_fd);
382 383 384 385 386 387 388
    }
    CP_UNLOCK();

    return(ret);
}


389
int DARSHAN_DECL(fsync)(int fd)
390 391 392 393 394
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

395 396
    MAP_OR_FAIL(fsync);

397 398 399 400 401 402 403 404 405 406 407
    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)
    {
408
        CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_write_end, CP_F_POSIX_WRITE_TIME); \
409 410 411 412 413 414 415
        CP_INC(file, CP_POSIX_FSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}

416
int DARSHAN_DECL(fdatasync)(int fd)
417 418 419 420 421
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

422 423
    MAP_OR_FAIL(fdatasync);

424 425 426 427 428 429 430 431 432 433
    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)
    {
434
        CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_write_end, CP_F_POSIX_WRITE_TIME); \
435 436 437 438 439 440 441 442
        CP_INC(file, CP_POSIX_FDSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}


443
void* DARSHAN_DECL(mmap64)(void *addr, size_t length, int prot, int flags,
444 445 446 447 448
    int fd, off64_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

449 450
    MAP_OR_FAIL(mmap64);

451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
    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);
}


467
void* DARSHAN_DECL(mmap)(void *addr, size_t length, int prot, int flags,
468 469 470 471 472
    int fd, off_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

473 474
    MAP_OR_FAIL(mmap);

475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
    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);
}

490
int DARSHAN_DECL(creat)(const char* path, mode_t mode)
491 492 493 494
{
    int ret;
    double tm1, tm2;

495 496
    MAP_OR_FAIL(creat);

497 498 499 500 501 502 503 504 505 506 507
    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);
}

508
int DARSHAN_DECL(creat64)(const char* path, mode_t mode)
509 510 511 512
{
    int ret;
    double tm1, tm2;

513 514
    MAP_OR_FAIL(creat64);

515 516 517 518 519 520 521 522 523 524 525
    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);
}

526
int DARSHAN_DECL(open64)(const char* path, int flags, ...)
527 528 529 530 531
{
    int mode = 0;
    int ret;
    double tm1, tm2;

532 533
    MAP_OR_FAIL(open64);

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
    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);
}

559
int DARSHAN_DECL(open)(const char *path, int flags, ...)
560 561 562 563 564
{
    int mode = 0;
    int ret;
    double tm1, tm2;

565 566
    MAP_OR_FAIL(open);

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
    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);
}

592
FILE* DARSHAN_DECL(fopen64)(const char *path, const char *mode)
593 594 595 596 597
{
    FILE* ret;
    int fd;
    double tm1, tm2;

598 599
    MAP_OR_FAIL(fopen64);

600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
    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);
}

615
FILE* DARSHAN_DECL(fopen)(const char *path, const char *mode)
616 617 618 619 620
{
    FILE* ret;
    int fd;
    double tm1, tm2;

621 622
    MAP_OR_FAIL(fopen);

623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
    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);
}

638
int DARSHAN_DECL(__xstat64)(int vers, const char *path, struct stat64 *buf)
639 640 641 642
{
    int ret;
    double tm1, tm2;

643 644
    MAP_OR_FAIL(__xstat64);

645 646 647 648 649 650 651
    tm1 = darshan_wtime();
    ret = __real___xstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
652
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
653 654 655 656 657
    CP_UNLOCK();

    return(ret);
}

658
int DARSHAN_DECL(__lxstat64)(int vers, const char *path, struct stat64 *buf)
659 660 661 662
{
    int ret;
    double tm1, tm2;

663 664
    MAP_OR_FAIL(__lxstat64);

665 666 667 668 669 670 671
    tm1 = darshan_wtime();
    ret = __real___lxstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
672
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
673 674 675 676 677
    CP_UNLOCK();

    return(ret);
}

678
int DARSHAN_DECL(__fxstat64)(int vers, int fd, struct stat64 *buf)
679 680 681 682 683
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

684 685
    MAP_OR_FAIL(__fxstat64);

686 687 688 689 690 691 692 693 694 695 696 697 698 699
    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)
    {
700
        CP_RECORD_STAT(file, buf, tm1, tm2);
701 702 703 704 705 706 707
    }
    CP_UNLOCK();

    return(ret);
}


708
int DARSHAN_DECL(__xstat)(int vers, const char *path, struct stat *buf)
709 710 711 712
{
    int ret;
    double tm1, tm2;

713 714
    MAP_OR_FAIL(__xstat);

715 716 717 718 719 720 721
    tm1 = darshan_wtime();
    ret = __real___xstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
722
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
723 724 725 726 727
    CP_UNLOCK();

    return(ret);
}

728
int DARSHAN_DECL(__lxstat)(int vers, const char *path, struct stat *buf)
729 730 731 732
{
    int ret;
    double tm1, tm2;

733 734
    MAP_OR_FAIL(__lxstat);

735 736 737 738 739 740 741
    tm1 = darshan_wtime();
    ret = __real___lxstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
742
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
743 744 745 746 747
    CP_UNLOCK();

    return(ret);
}

748
int DARSHAN_DECL(__fxstat)(int vers, int fd, struct stat *buf)
749 750 751 752 753
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

754 755
    MAP_OR_FAIL(__fxstat);

756 757 758 759 760 761 762
    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 */
763
    if((void*)buf == (void*)&cp_stat_buf)
764 765 766 767 768 769
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
770
        CP_RECORD_STAT(file, buf, tm1, tm2);
771 772 773 774 775 776
    }
    CP_UNLOCK();

    return(ret);
}

777
ssize_t DARSHAN_DECL(pread64)(int fd, void *buf, size_t count, off64_t offset)
778 779 780 781 782
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

783 784
    MAP_OR_FAIL(pread64);

785
    if((unsigned long)buf % darshan_mem_alignment == 0)
786 787 788 789 790 791
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
792
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
793 794 795 796
    CP_UNLOCK();
    return(ret);
}

797
ssize_t DARSHAN_DECL(pread)(int fd, void *buf, size_t count, off_t offset)
798 799 800 801 802
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

803 804
    MAP_OR_FAIL(pread);

805
    if((unsigned long)buf % darshan_mem_alignment == 0)
806 807 808 809 810 811
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
812
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
813 814 815 816 817
    CP_UNLOCK();
    return(ret);
}


818
ssize_t DARSHAN_DECL(pwrite)(int fd, const void *buf, size_t count, off_t offset)
819 820 821 822 823
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

824 825
    MAP_OR_FAIL(pwrite);

826
    if((unsigned long)buf % darshan_mem_alignment == 0)
827 828 829 830 831 832
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
833
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
834 835 836 837
    CP_UNLOCK();
    return(ret);
}

838
ssize_t DARSHAN_DECL(pwrite64)(int fd, const void *buf, size_t count, off64_t offset)
839 840 841 842 843
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

844 845
    MAP_OR_FAIL(pwrite64);

846
    if((unsigned long)buf % darshan_mem_alignment == 0)
847 848 849 850 851 852
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
853
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
854 855 856 857
    CP_UNLOCK();
    return(ret);
}

858
ssize_t DARSHAN_DECL(readv)(int fd, const struct iovec *iov, int iovcnt)
859 860 861 862 863 864
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

865 866
    MAP_OR_FAIL(readv);

867 868
    for(i=0; i<iovcnt; i++)
    {
869
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
870 871 872 873 874 875 876
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_readv(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
877
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
878 879 880 881
    CP_UNLOCK();
    return(ret);
}

882
ssize_t DARSHAN_DECL(writev)(int fd, const struct iovec *iov, int iovcnt)
883 884 885 886 887 888
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

889 890
    MAP_OR_FAIL(writev);

891 892
    for(i=0; i<iovcnt; i++)
    {
893
        if(!((unsigned long)iov[i].iov_base % darshan_mem_alignment == 0))
894 895 896 897 898 899 900
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_writev(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
901
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
902 903 904 905
    CP_UNLOCK();
    return(ret);
}

906
size_t DARSHAN_DECL(fread)(void *ptr, size_t size, size_t nmemb, FILE *stream)
907 908 909 910 911
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

912 913
    MAP_OR_FAIL(fread);

914
    if((unsigned long)ptr % darshan_mem_alignment == 0)
915 916 917 918 919 920 921
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fread(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
922
        CP_RECORD_READ(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
923
    else
924
        CP_RECORD_READ(ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
925 926 927 928
    CP_UNLOCK();
    return(ret);
}

929
ssize_t DARSHAN_DECL(read)(int fd, void *buf, size_t count)
930 931 932 933 934
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

935 936
    MAP_OR_FAIL(read);

937
    if((unsigned long)buf % darshan_mem_alignment == 0)
938 939 940 941 942 943
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_read(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
944
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
945 946 947 948
    CP_UNLOCK();
    return(ret);
}

949
ssize_t DARSHAN_DECL(write)(int fd, const void *buf, size_t count)
950 951 952 953 954
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

955 956
    MAP_OR_FAIL(write);

957
    if((unsigned long)buf % darshan_mem_alignment == 0)
958 959 960 961 962 963
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_write(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
964
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
965 966 967 968
    CP_UNLOCK();
    return(ret);
}

969
size_t DARSHAN_DECL(fwrite)(const void *ptr, size_t size, size_t nmemb, FILE *stream)
970 971 972 973 974
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

975 976
    MAP_OR_FAIL(fwrite);

977
    if((unsigned long)ptr % darshan_mem_alignment == 0)
978 979 980 981 982 983 984
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fwrite(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
985
        CP_RECORD_WRITE(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
986
    else
987
        CP_RECORD_WRITE(ret, fileno(stream), 0, 0, 0, aligned_flag, 1, tm1, tm2);
988 989 990 991
    CP_UNLOCK();
    return(ret);
}

992
off64_t DARSHAN_DECL(lseek64)(int fd, off64_t offset, int whence)
993
{
994
    off64_t ret;
995 996 997
    struct darshan_file_runtime* file;
    double tm1, tm2;

998 999
    MAP_OR_FAIL(lseek64);

1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
    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;
1010
            CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME);
1011 1012 1013 1014 1015 1016 1017
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

1018
off_t DARSHAN_DECL(lseek)(int fd, off_t offset, int whence)
1019 1020 1021 1022 1023
{
    off_t ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

1024 1025
    MAP_OR_FAIL(lseek);

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
    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;
1036
            CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME);
1037 1038 1039 1040 1041 1042 1043
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

1044
ssize_t DARSHAN_DECL(aio_return64)(struct aiocb64 *aiocbp)
1045 1046
{
    int ret;
1047 1048 1049
    double tm2;
    struct darshan_aio_tracker *tmp;
    int aligned_flag = 0;
1050 1051 1052 1053

    MAP_OR_FAIL(aio_return64);

    ret = __real_aio_return64(aiocbp);
1054
    tm2 = darshan_wtime();
1055
    tmp = darshan_aio_tracker_del(aiocbp->aio_fildes, aiocbp);
1056 1057 1058 1059 1060 1061

    if(tmp)
    {
        if((unsigned long)aiocbp->aio_buf % darshan_mem_alignment == 0)
            aligned_flag = 1;
        CP_LOCK();
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
        if(aiocbp->aio_lio_opcode == LIO_WRITE)
        {
            CP_RECORD_WRITE(ret, aiocbp->aio_fildes, aiocbp->aio_nbytes,
                1, aiocbp->aio_offset, aligned_flag, 0, tmp->tm1, tm2);
        }
        if(aiocbp->aio_lio_opcode == LIO_READ)
        {
            CP_RECORD_READ(ret, aiocbp->aio_fildes, aiocbp->aio_nbytes,
                1, aiocbp->aio_offset, aligned_flag, 0, tmp->tm1, tm2);
        }
1072 1073 1074
        CP_UNLOCK();
        free(tmp);
    }
1075 1076 1077 1078 1079 1080 1081

    return(ret);
}

ssize_t DARSHAN_DECL(aio_return)(struct aiocb *aiocbp)
{
    int ret;
1082 1083 1084
    double tm2;
    struct darshan_aio_tracker *tmp;
    int aligned_flag = 0;
1085 1086 1087 1088

    MAP_OR_FAIL(aio_return);

    ret = __real_aio_return(aiocbp);
1089
    tm2 = darshan_wtime();
1090
    tmp = darshan_aio_tracker_del(aiocbp->aio_fildes, aiocbp);
1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101

    if(tmp)
    {
        if((unsigned long)aiocbp->aio_buf % darshan_mem_alignment == 0)
            aligned_flag = 1;
        CP_LOCK();
        CP_RECORD_WRITE(ret, aiocbp->aio_fildes, aiocbp->aio_nbytes,
            1, aiocbp->aio_offset, aligned_flag, 0, tmp->tm1, tm2);
        CP_UNLOCK();
        free(tmp);
    }
1102 1103 1104 1105 1106 1107 1108 1109

    return(ret);
}

int DARSHAN_DECL(lio_listio)(int mode, struct aiocb *const aiocb_list[],
    int nitems, struct sigevent *sevp)
{
    int ret;
1110
    int i;
1111 1112 1113 1114

    MAP_OR_FAIL(lio_listio);

    ret = __real_lio_listio(mode, aiocb_list, nitems, sevp);
1115 1116 1117 1118
    if(ret == 0)
    {
        for(i=0; i<nitems; i++)
        {
1119
            darshan_aio_tracker_add(aiocb_list[i]->aio_fildes, aiocb_list[i]);        
1120 1121
        }
    }
1122 1123 1124 1125

    return(ret);
}

1126
int DARSHAN_DECL(lio_listio64)(int mode, struct aiocb64 *const aiocb_list[],
1127 1128 1129
    int nitems, struct sigevent *sevp)
{
    int ret;
1130
    int i;
1131

1132
    MAP_OR_FAIL(lio_listio);
1133

1134
    ret = __real_lio_listio64(mode, aiocb_list, nitems, sevp);
1135 1136 1137 1138
    if(ret == 0)
    {
        for(i=0; i<nitems; i++)
        {
1139
            darshan_aio_tracker_add(aiocb_list[i]->aio_fildes, aiocb_list[i]);        
1140 1141
        }
    }
1142 1143 1144 1145

    return(ret);
}

1146
int DARSHAN_DECL(aio_write64)(struct aiocb64 *aiocbp)
1147 1148 1149 1150 1151 1152
{
    int ret;

    MAP_OR_FAIL(aio_write64);

    ret = __real_aio_write64(aiocbp);
1153
    if(ret == 0)
1154
        darshan_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165

    return(ret);
}

int DARSHAN_DECL(aio_write)(struct aiocb *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_write);

    ret = __real_aio_write(aiocbp);
1166
    if(ret == 0)
1167
        darshan_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1168 1169 1170 1171

    return(ret);
}

1172
int DARSHAN_DECL(aio_read64)(struct aiocb64 *aiocbp)
1173 1174 1175 1176 1177 1178
{
    int ret;

    MAP_OR_FAIL(aio_read64);

    ret = __real_aio_read64(aiocbp);
1179
    if(ret == 0)
1180
        darshan_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191

    return(ret);
}

int DARSHAN_DECL(aio_read)(struct aiocb *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_read);

    ret = __real_aio_read(aiocbp);
1192
    if(ret == 0)
1193
        darshan_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1194 1195 1196 1197

    return(ret);
}

1198
int DARSHAN_DECL(fseek)(FILE *stream, long offset, int whence)
1199 1200 1201 1202 1203
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

1204 1205
    MAP_OR_FAIL(fseek);

1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
    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;
1216
            CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, CP_F_POSIX_META_TIME);
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
            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;
1239
    char* envstr;
1240 1241 1242
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
1243 1244
    int ret;
    int tmpval;
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259

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

    disable_timing = getenv("DARSHAN_DISABLE_TIMING");

    if(darshan_global_job != NULL)
    {
        return;
    }

1260 1261 1262
    #if (__CP_MEM_ALIGNMENT < 1)
        #error Darshan must be configured with a positive value for --with-mem-align
    #endif
1263 1264 1265
    envstr = getenv("DARSHAN_MEMALIGN");
    if (envstr)
    {
1266 1267 1268 1269 1270 1271
        ret = sscanf(envstr, "%d", &tmpval);
        /* silently ignore if the env variable is set poorly */
        if(ret == 1 && tmpval > 0)
        {
            darshan_mem_alignment = tmpval;
        }
1272 1273 1274 1275 1276 1277
    }
    else
    {
        darshan_mem_alignment = __CP_MEM_ALIGNMENT;
    }

1278 1279 1280 1281 1282 1283
    /* avoid floating point errors on faulty input */
    if (darshan_mem_alignment < 1)
    {
        darshan_mem_alignment = 1;
    }

1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304
    /* 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];
    }

    strcpy(darshan_global_job->log_job.version_string, CP_VERSION);
1305
    darshan_global_job->log_job.magic_nr = CP_MAGIC_NR;
1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
    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);
    }
1341 1342 1343 1344 1345 1346 1347 1348

    /* collect information about command line and 
     * mounted file systems 
     */
    darshan_global_job->trailing_data = 
        darshan_get_exe_and_mounts(darshan_global_job);

    return;
1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388
}

/* 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++)
    {
        iter_file = &darshan_global_job->file_runtime_array[j];

        /* iterate through records */
        for(i=0; i<CP_NUM_INDICES; i++)
        {
            switch(i)
            {
                /* NOTE: several fields cease to make sense if the records
                 * have been condensed.  Just let them get summed anyway.
                 */
                /* TODO: double check this */

                /* keep up with global maxes in case they are helpful */
                case CP_MAX_BYTE_READ:
                case CP_MAX_BYTE_WRITTEN:
                    CP_MAX(base_file, i, CP_VALUE(iter_file, i));
                    break;

1389 1390 1391 1392 1393 1394 1395
                /* do nothing with these; they are handled in the floating
                 * point loop 
                 */
                case CP_MAX_WRITE_TIME_SIZE:
                case CP_MAX_READ_TIME_SIZE:
                    break;

1396 1397
                /* pick one */
                case CP_DEVICE:
1398
                case CP_SIZE_AT_OPEN:
1399 1400 1401
                    CP_SET(base_file, i, CP_VALUE(iter_file, i));
                    break;

1402 1403 1404 1405 1406 1407 1408 1409
                /* most records can simply be added */
                default:
                    CP_INC(base_file, i, CP_VALUE(iter_file, i));
                    break;
            }
        }
        for(i=0; i<CP_F_NUM_INDICES; i++)
        {
1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
            switch(i)
            {
                case CP_F_MAX_WRITE_TIME:
                    if(CP_F_VALUE(iter_file, i) > CP_F_VALUE(base_file, i))
                    {
                        CP_F_SET(base_file, i, CP_F_VALUE(iter_file, i));
                        CP_SET(base_file, CP_MAX_WRITE_TIME_SIZE, 
                            CP_VALUE(iter_file, CP_MAX_WRITE_TIME_SIZE));
                    }
                    break;
                case CP_F_MAX_READ_TIME:
                    if(CP_F_VALUE(iter_file, i) > CP_F_VALUE(base_file, i))
                    {
                        CP_F_SET(base_file, i, CP_F_VALUE(iter_file, i));
                        CP_SET(base_file, CP_MAX_READ_TIME_SIZE, 
                            CP_VALUE(iter_file, CP_MAX_READ_TIME_SIZE));
                    }
                    break;
                default:
                    CP_F_SET(base_file, i, CP_F_VALUE(iter_file, i) + CP_F_VALUE(base_file, i));
                    break;
            }
1432
        }
1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447

        if(base_file->aio_list_tail)
        {
            /* base has an aio list already; add on to it */
            assert(base_file->aio_list_head);
            base_file->aio_list_tail->next = iter_file->aio_list_head;
            if(iter_file->aio_list_tail)
                base_file->aio_list_tail = iter_file->aio_list_tail;
        }
        else
        {
            /* take on list from iter */
            base_file->aio_list_head = iter_file->aio_list_head;
            base_file->aio_list_tail = iter_file->aio_list_tail;
        }
1448 1449 1450 1451 1452 1453 1454 1455 1456
    }
    
    base_file->log_file->hash = 0;
    
    darshan_global_job->flags |= CP_FLAG_CONDENSED;
    darshan_global_job->file_count = 1;

    /* clear hash tables for safety */
    memset(darshan_global_job->name_table, 0, CP_HASH_SIZE*sizeof(struct darshan_file_runtime*));
1457
    memset(darshan_global_job->handle_table, 0, CP_HASH_SIZE*sizeof(*darshan_global_job->handle_table));
1458 1459 1460 1461
    
    return;
}

1462
static struct darshan_file_runtime* darshan_file_by_name_setfd(const char* name, int fd)
1463 1464 1465
{
    struct darshan_file_runtime* tmp_file;

1466
    tmp_file = darshan_file_by_name_sethandle(name, &fd, sizeof(fd), DARSHAN_FD);
1467 1468 1469
    return(tmp_file);
}

1470
static void darshan_file_close_fd(int fd)
1471
{
1472 1473 1474
    darshan_file_closehandle(&fd, sizeof(fd), DARSHAN_FD);
    return;
}
1475

1476 1477 1478
static struct darshan_file_runtime* darshan_file_by_fd(int fd)
{
    struct darshan_file_runtime* tmp_file;
1479

1480 1481 1482
    tmp_file = darshan_file_by_handle(&fd, sizeof(fd), DARSHAN_FD);
    
    return(tmp_file);
1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611
}

static int access_comparison(const void* a_p, const void* b_p)
{
    const struct cp_access_counter* a = a_p;
    const struct cp_access_counter* b = b_p;

    if(a->size < b->size)
        return(-1);
    if(a->size > b->size)
        return(1);
    return(0);
}

/* cp_access_counter()
 *
 * records the occurance of a particular access size for a file,
 * current implementation uses glibc red black tree
 */
static void cp_access_counter(struct darshan_file_runtime* file, ssize_t size, enum cp_counter_type type)
{
    struct cp_access_counter* counter;
    struct cp_access_counter* found;
    void* tmp;
    void** root;
    int* count;
    struct cp_access_counter tmp_counter;

    /* don't count sizes or strides of 0 */
    if(size == 0)
        return;
    
    switch(type)
    {
        case CP_COUNTER_ACCESS:
            root = &file->access_root;
            count = &file->access_count;
            break;
        case CP_COUNTER_STRIDE:
            root = &file->stride_root;
            count = &file->stride_count;
            break;
        default:
            return;
    }

    /* check to see if this size is already recorded */
    tmp_counter.size = size;
    tmp_counter.freq = 1;
    tmp = tfind(&tmp_counter, root, access_comparison);
    if(tmp)
    {
        found = *(struct cp_access_counter**)tmp;
        found->freq++;
        return;
    }

    /* we can add a new one as long as we haven't hit the limit */
    if(*count < CP_MAX_ACCESS_COUNT_RUNTIME)
    {
        counter = malloc(sizeof(*counter));
        if(!counter)
        {
            return;
        }

        counter->size = size;
        counter->freq = 1;

        tmp = tsearch(counter, root, access_comparison);
        found = *(struct cp_access_counter**)tmp;
        /* if we get a new answer out here we are in trouble; this was
         * already checked with the tfind()
         */
        assert(found == counter);

        (*count)++;
    }

    return;
}

void darshan_shutdown_bench(int argc, char** argv, int rank, int nprocs)
{
    int* fd_array;
    int64_t* size_array;
    int i;
    int nfiles;
    char path[256];
    int iters;
    
    /* combinations to build:
     * - 1 unique file per proc
     * - 1 shared file per proc
     * - 1024 unique file per proc
     * - 1024 shared per proc
     */

    srand(rank);
    fd_array = malloc(sizeof(int)*CP_MAX_FILES);
    size_array = malloc(sizeof(int64_t)*CP_MAX_ACCESS_COUNT_RUNTIME);

    assert(fd_array&&size_array);

    for(i=0; i<CP_MAX_FILES; i++)
        fd_array[i] = i;
    for(i=0; i<CP_MAX_ACCESS_COUNT_RUNTIME; i++)
        size_array[i] = rand();

    /* clear out existing stuff */
    darshan_walk_file_accesses(darshan_global_job);
    darshan_finalize(darshan_global_job);
    darshan_global_job = NULL;

    /***********************************************************/
    /* reset darshan to start clean */
    darshan_initialize(argc, argv, nprocs, rank);

    /* populate one unique file per proc */
    nfiles = 1;
    iters = 1;
    for(i=0; i<nfiles; i++)
    {
        sprintf(path, "%d-%d", i, rank);
        CP_RECORD_OPEN(fd_array[i], path, 777, 0, 0, 0);
    }

    for(i=0; i<iters; i++)
    {
1612
        CP_RECORD_WRITE(size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], fd_array[i%nfiles], size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], 0, 0, 1, 0, 1, 2);
1613 1614 1615 1616
    }

    if(rank == 0)
        printf("# 1 unique file per proc\n");
1617
    //DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);
1618 1619 1620 1621 1622
    darshan_shutdown(1);
    darshan_global_job = NULL;

    /***********************************************************/
    /* reset darshan to start clean */
1623
    sleep(1);
1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636
    darshan_initialize(argc, argv, nprocs, rank);

    /* populate one shared file per proc */
    nfiles = 1;
    iters = 1;
    for(i=0; i<nfiles; i++)
    {
        sprintf(path, "%d", i);
        CP_RECORD_OPEN(fd_array[i], path, 777, 0, 0, 0);
    }

    for(i=0; i<iters; i++)
    {
1637
        CP_RECORD_WRITE(size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], fd_array[i%nfiles], size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], 0, 0, 1, 0, 1, 2);
1638 1639 1640 1641
    }

    if(rank == 0)
        printf("# 1 shared file across procs\n");
1642
    //DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);
1643 1644 1645 1646 1647
    darshan_shutdown(1);
    darshan_global_job = NULL;

    /***********************************************************/
    /* reset darshan to start clean */
1648
    sleep(1);
1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
    darshan_initialize(argc, argv, nprocs, rank);

    /* populate 1024 unique file per proc */
    nfiles = 1024;
    iters = 1024;
    for(i=0; i<nfiles; i++)
    {
        sprintf(path, "%d-%d", i, rank);
        CP_RECORD_OPEN(fd_array[i], path, 777, 0, 0, 0);
    }

    for(i=0; i<iters; i++)
    {
1662
        CP_RECORD_WRITE(size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], fd_array[i%nfiles], size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], 0, 0, 1, 0, 1, 2);
1663 1664 1665 1666
    }

    if(rank == 0)
        printf("# 1024 unique files per proc\n");
1667
    //DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);
1668 1669 1670 1671 1672
    darshan_shutdown(1);
    darshan_global_job = NULL;

    /***********************************************************/
    /* reset darshan to start clean */
1673
    sleep(1);
1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
    darshan_initialize(argc, argv, nprocs, rank);

    /* populate 1024 shared file per proc */
    nfiles = 1024;
    iters = 1024;
    for(i=0; i<nfiles; i++)
    {
        sprintf(path, "%d", i);
        CP_RECORD_OPEN(fd_array[i], path, 777, 0, 0, 0);
    }

    for(i=0; i<iters; i++)
    {
1687
        CP_RECORD_WRITE(size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], fd_array[i%nfiles], size_array[(i/nfiles)%CP_MAX_ACCESS_COUNT_RUNTIME], 0, 0, 1, 0, 1, 2);
1688 1689 1690 1691
    }

    if(rank == 0)
        printf("# 1024 shared files across procs\n");
1692
    //DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);
1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746
    darshan_shutdown(1);
    darshan_global_job = NULL;

    darshan_initialize(argc, argv, nprocs, rank);

    free(fd_array);
    free(size_array);

    return;
}

void darshan_search_bench(int argc, char** argv, int iters)
{
    int* fd_array;
    int64_t* size_array;
    int i,j;
    int skip = 32;
    int nfiles;
    char path[256];
    double tm1, tm2;
    
    fd_array = malloc(sizeof(int)*CP_MAX_FILES);
    size_array = malloc(sizeof(int64_t)*CP_MAX_ACCESS_COUNT_RUNTIME);

    assert(fd_array&&size_array);

    for(i=0; i<CP_MAX_FILES; i++)
        fd_array[i] = i;
    for(i=0; i<CP_MAX_ACCESS_COUNT_RUNTIME; i++)
        size_array[i] = rand();

    printf("#<iters>\t<numfiles>\t<numsizes>\t<total time>\t<per iter>\n");

    for(j=0; j<2; j++)
    {
        /* warm up */
        /* reset darshan to start clean */
        darshan_walk_file_accesses(darshan_global_job);
        darshan_finalize(darshan_global_job);
        darshan_global_job = NULL;
        darshan_initialize(argc, argv, 1, 0);

        nfiles = 1;
        /* populate entries for each file */
        for(i=0; i<nfiles; i++)
        {
            sprintf(path, "%d",