darshan-posix.c 17.2 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 <aio.h>
23
#define __USE_GNU
24
#include <pthread.h>
25

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

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

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

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

#define DARSHAN_DECL(__name) __name

#define DARSHAN_MPI_CALL(func) __real_ ## func

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

#else
60

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

#define DARSHAN_DECL(__name) __wrap_ ## __name

66 67
#define DARSHAN_MPI_CALL(func) func

68 69
#define MAP_OR_FAIL(func)

70 71 72 73 74 75
#endif

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));

76
struct posix_runtime_file
77
{
78
    struct darshan_posix_file* file_record;
79
    UT_hash_handle hlink;
80
};
81

82
struct posix_runtime_file_ref
83
{
84 85 86 87 88 89 90
    struct posix_runtime_file* file;
    int fd;
    UT_hash_handle hlink;
};

struct posix_runtime
{
91 92
    struct posix_runtime_file* file_runtime_array;
    struct darshan_posix_file* file_record_array;
93
    int file_array_size;
94
    int file_array_ndx;
95 96
    struct posix_runtime_file* file_hash;
    struct posix_runtime_file_ref* fd_hash;
97 98
    void *red_buf;
    int shared_rec_count;
99 100
};

101
static struct posix_runtime *posix_runtime = NULL;
102
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
Shane Snyder's avatar
Shane Snyder committed
103
static int instrumentation_disabled = 0;
104
static int my_rank = -1;
105

106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

121
static void posix_runtime_initialize(void);
122 123
static struct posix_runtime_file* posix_file_by_name(const char *name);
static struct posix_runtime_file* posix_file_by_name_setfd(const char* name, int fd);
Shane Snyder's avatar
Shane Snyder committed
124
static struct posix_runtime_file* posix_file_by_fd(int fd);
125 126
static void posix_file_close_fd(int fd);

Shane Snyder's avatar
Shane Snyder committed
127
static void posix_disable_instrumentation(void);
128 129 130 131 132
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_reduce_record(void* infile_v, void* inoutfile_v,
    int *len, MPI_Datatype *datatype);
static void posix_get_output_data(void **buffer, int *size);
133
static void posix_shutdown(void);
134

135 136
#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)
137

138
#define POSIX_SET(__file, __counter, __value) do {\
139
    (__file)->file_record->counters[__counter] = __value; \
140 141 142
} while(0)

#define POSIX_F_SET(__file, __counter, __value) do {\
143
    (__file)->file_record->fcounters[__counter] = __value; \
144 145 146
} while(0)

#define POSIX_INC(__file, __counter, __value) do {\
147
    (__file)->file_record->counters[__counter] += __value; \
148 149 150
} while(0)

#define POSIX_F_INC(__file, __counter, __value) do {\
151
    (__file)->file_record->fcounters[__counter] += __value; \
152 153 154 155 156 157 158 159 160 161 162 163
} while(0)

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

#define POSIX_VALUE(__file, __counter) \
164
    ((__file)->file_record->counters[__counter])
165 166

#define POSIX_F_VALUE(__file, __counter) \
167
    ((__file)->file_record->fcounters[__counter])
168 169

#define POSIX_MAX(__file, __counter, __value) do {\
170
    if((__file)->file_record->counters[__counter] < __value) \
171
    { \
172
        (__file)->file_record->counters[__counter] = __value; \
173 174 175
    } \
} while(0)

176
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
177
    struct posix_runtime_file* file; \
178 179 180 181 182 183 184 185 186
    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; \
187
    file = posix_file_by_name_setfd(__path, __ret); \
188
    if(!file) break; \
189
    file->file_record->rank = my_rank; \
190
    if(__mode) \
191
        POSIX_SET(file, CP_MODE, __mode); \
192
    if(__stream_flag)\
193
        POSIX_INC(file, CP_POSIX_FOPENS, 1); \
194
    else \
195 196 197 198
        POSIX_INC(file, CP_POSIX_OPENS, 1); \
    if(POSIX_F_VALUE(file, CP_F_OPEN_TIMESTAMP) == 0) \
        POSIX_F_SET(file, CP_F_OPEN_TIMESTAMP, __tm1); \
} while(0)
199

200
int DARSHAN_DECL(open)(const char *path, int flags, ...)
201 202 203 204 205
{
    int mode = 0;
    int ret;
    double tm1, tm2;

206 207
    MAP_OR_FAIL(open);

208
    if(flags & O_CREAT) 
209 210 211 212 213 214
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

215
        tm1 = darshan_core_wtime();
216
        ret = __real_open(path, flags, mode);
217
        tm2 = darshan_core_wtime();
218 219 220
    }
    else
    {
221
        tm1 = darshan_core_wtime();
222
        ret = __real_open(path, flags);
223
        tm2 = darshan_core_wtime();
224 225
    }

226 227 228
    POSIX_LOCK();
    posix_runtime_initialize();

229
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
230

231
    POSIX_UNLOCK();
232 233 234 235

    return(ret);
}

Shane Snyder's avatar
Shane Snyder committed
236 237 238 239 240 241
int DARSHAN_DECL(open64)(const char *path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

242
    MAP_OR_FAIL(open64);
Shane Snyder's avatar
Shane Snyder committed
243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

    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();

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

    return(ret);
}

271 272
int DARSHAN_DECL(close)(int fd)
{
Shane Snyder's avatar
Shane Snyder committed
273
    struct posix_runtime_file* file;
274 275 276 277 278 279 280 281 282 283 284
    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();
285

Shane Snyder's avatar
Shane Snyder committed
286 287 288 289 290 291
    file = posix_file_by_fd(fd);
    if(file)
    {
        POSIX_F_SET(file, CP_F_CLOSE_TIMESTAMP, darshan_core_wtime());
        posix_file_close_fd(fd);
    }
292 293 294 295 296 297
    POSIX_UNLOCK();    

    return(ret);
}

/* ***************************************************** */
298

299
static void posix_runtime_initialize()
300
{
301 302 303 304 305 306
    char *alignstr;
    int tmpval;
    int ret;
    int mem_limit;
    struct darshan_module_funcs posix_mod_fns =
    {
Shane Snyder's avatar
Shane Snyder committed
307
        .disable_instrumentation = &posix_disable_instrumentation,
308 309
        .prepare_for_reduction = &posix_prepare_for_reduction,
        .reduce_record = &posix_reduce_record,
310
        .get_output_data = &posix_get_output_data,
311
        .shutdown = &posix_shutdown
312
    };
313

Shane Snyder's avatar
Shane Snyder committed
314 315
    /* don't do anything if already initialized or instrumenation is disabled */
    if(posix_runtime || instrumentation_disabled)
316
        return;
317

318 319
    /* register the posix module with darshan core */
    darshan_core_register_module(
320
        DARSHAN_POSIX_MOD,
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
        &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 */
    posix_runtime->file_array_size = mem_limit / sizeof(struct darshan_posix_file);
336
    posix_runtime->file_array_ndx = 0;
337 338

    /* allocate array of runtime file records */
339 340 341 342 343
    posix_runtime->file_runtime_array = malloc(posix_runtime->file_array_size *
                                               sizeof(struct posix_runtime_file));
    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)
344 345 346 347
    {
        posix_runtime->file_array_size = 0;
        return;
    }
348 349 350 351
    memset(posix_runtime->file_runtime_array, 0, posix_runtime->file_array_size *
           sizeof(struct posix_runtime_file));
    memset(posix_runtime->file_record_array, 0, posix_runtime->file_array_size *
           sizeof(struct darshan_posix_file));
352

Shane Snyder's avatar
Shane Snyder committed
353
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank); /* TODO: can we move this out of here? */
354

355 356
    return;
}
357

358
static struct posix_runtime_file* posix_file_by_name(const char *name)
359
{
360
    struct posix_runtime_file *file = NULL;
361
    char *newname = NULL;
362
    darshan_record_id file_id;
363

Shane Snyder's avatar
Shane Snyder committed
364
    if(!posix_runtime || instrumentation_disabled)
365
        return(NULL);
366

367
    newname = darshan_clean_file_path(name);
368
    if(!newname)
369
        newname = (char*)name;
370

371
    /* get a unique id for this file from darshan core */
372
    darshan_core_lookup_record_id(
373 374 375
        (void*)newname,
        strlen(newname),
        1,
376
        DARSHAN_POSIX_MOD,
377
        &file_id);
378

379
    /* search the hash table for this file record, and return if found */
380
    HASH_FIND(hlink, posix_runtime->file_hash, &file_id, sizeof(darshan_record_id), file);
381
    if(file)
382
    {
383
        if(newname != name)
384
            free(newname);
385
        return(file);
386 387
    }

388 389 390 391 392 393
    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;
394

395
        /* add new record to file hash table */
396
        HASH_ADD(hlink, posix_runtime->file_hash, file_record->f_id, sizeof(darshan_record_id), file);
397

398 399
        posix_runtime->file_array_ndx++;
    }
400 401

    if(newname != name)
402
        free(newname);
403 404 405 406 407 408 409 410
    return(file);
}

static struct posix_runtime_file* posix_file_by_name_setfd(const char* name, int fd)
{
    struct posix_runtime_file* file;
    struct posix_runtime_file_ref* ref;

Shane Snyder's avatar
Shane Snyder committed
411
    if(!posix_runtime || instrumentation_disabled)
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
        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
438

439 440 441 442 443 444 445
    ref->file = file;
    ref->fd = fd;    
    HASH_ADD(hlink, posix_runtime->fd_hash, fd, sizeof(int), ref);

    return(file);
}

Shane Snyder's avatar
Shane Snyder committed
446 447 448 449 450 451 452 453 454 455 456 457 458 459 460
static struct posix_runtime_file* posix_file_by_fd(int fd)
{
    struct posix_runtime_file_ref* ref;

    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);
}

461 462
static void posix_file_close_fd(int fd)
{
463
    struct posix_runtime_file_ref* ref;
464

Shane Snyder's avatar
Shane Snyder committed
465
    if(!posix_runtime || instrumentation_disabled)
466 467 468 469
        return;

    /* search hash table for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
470
    if(ref)
471 472 473 474 475 476 477
    {
        /* we have a reference, delete it */
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }

    return;
478 479
}

480 481 482 483 484 485 486 487 488 489 490 491 492
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;
}

493 494
/* ***************************************************** */

Shane Snyder's avatar
Shane Snyder committed
495
static void posix_disable_instrumentation()
496
{
497 498
    assert(posix_runtime);

Shane Snyder's avatar
Shane Snyder committed
499 500 501
    POSIX_LOCK();
    instrumentation_disabled = 1;
    POSIX_UNLOCK();
502

Shane Snyder's avatar
Shane Snyder committed
503 504 505
    return;
}

506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
static void posix_prepare_for_reduction(
    darshan_record_id *shared_recs,
    int *shared_rec_count,
    void **send_buf,
    void **recv_buf,
    int *rec_size)
{
    struct posix_runtime_file *file;
    struct darshan_posix_file *tmp_array;
    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);

    /* TODO: HACK-Y -- how can we do this in a cleaner way?? */
    if(my_rank == 0)
        posix_runtime->red_buf = *recv_buf;
    posix_runtime->shared_rec_count = *shared_rec_count;

    return;
}

static void posix_reduce_record(
    void* infile_v,
    void* inoutfile_v,
    int *len,
    MPI_Datatype *datatype)
Shane Snyder's avatar
Shane Snyder committed
564
{
565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620
    struct darshan_posix_file tmp_file;
    struct darshan_posix_file *infile = infile_v;
    struct darshan_posix_file *inoutfile = inoutfile_v;
    int i;

    assert(posix_runtime);

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

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

        tmp_file.counters[CP_POSIX_OPENS] = infile->counters[CP_POSIX_OPENS] +
            inoutfile->counters[CP_POSIX_OPENS];

        if((infile->fcounters[CP_F_OPEN_TIMESTAMP] > inoutfile->fcounters[CP_F_OPEN_TIMESTAMP]) &&
            (inoutfile->fcounters[CP_F_OPEN_TIMESTAMP] > 0))
            tmp_file.fcounters[CP_F_OPEN_TIMESTAMP] = inoutfile->fcounters[CP_F_OPEN_TIMESTAMP];
        else
            tmp_file.fcounters[CP_F_OPEN_TIMESTAMP] = infile->fcounters[CP_F_OPEN_TIMESTAMP];

        if(infile->fcounters[CP_F_CLOSE_TIMESTAMP] > inoutfile->fcounters[CP_F_CLOSE_TIMESTAMP])
            tmp_file.fcounters[CP_F_CLOSE_TIMESTAMP] = infile->fcounters[CP_F_CLOSE_TIMESTAMP];
        else
            tmp_file.fcounters[CP_F_CLOSE_TIMESTAMP] = inoutfile->fcounters[CP_F_CLOSE_TIMESTAMP];

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

    return;
}

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

    /* TODO: HACK-Y -- how can we do this in a cleaner way?? */
    /* 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;
    }
621 622 623

    *buffer = (void *)(posix_runtime->file_record_array);
    *size = posix_runtime->file_array_ndx * sizeof(struct darshan_posix_file);
624

625
    return;
626 627
}

628
static void posix_shutdown()
629
{
630 631 632 633 634 635 636
    struct posix_runtime_file_ref *ref, *tmp;

    HASH_ITER(hlink, posix_runtime->fd_hash, ref, tmp)
    {
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }
637

638
    HASH_CLEAR(hlink, posix_runtime->file_hash); /* these entries are freed all at once below */
639 640 641 642 643

    free(posix_runtime->file_runtime_array);
    free(posix_runtime->file_record_array);
    free(posix_runtime);
    posix_runtime = NULL;
644 645

    return;
646 647
}

648 649 650 651 652 653 654 655
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ts=8 sts=4 sw=4 expandtab
 */