darshan-core.c 53.9 KB
Newer Older
1
/*
Shane Snyder's avatar
Shane Snyder committed
2 3 4
 * Copyright (C) 2015 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
5 6
 */

7
#define _XOPEN_SOURCE 500
8
#define _GNU_SOURCE
9

10 11 12 13 14 15
#include "darshan-runtime-config.h"

#include <stdio.h>
#ifdef HAVE_MNTENT_H
#include <mntent.h>
#endif
16 17 18 19 20
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <limits.h>
#include <pthread.h>
21
#include <fcntl.h>
22 23
#include <sys/types.h>
#include <sys/stat.h>
24
#include <sys/mman.h>
25
#include <sys/vfs.h>
26
#include <zlib.h>
27
#include <mpi.h>
28
#include <assert.h>
29

30
#include "uthash.h"
Shane Snyder's avatar
Shane Snyder committed
31
#include "darshan.h"
32
#include "darshan-core.h"
Shane Snyder's avatar
Shane Snyder committed
33
#include "darshan-dynamic.h"
34

35
extern char* __progname;
36
extern char* __progname_full;
37

38
/* internal variable delcarations */
39
static struct darshan_core_runtime *darshan_core = NULL;
40
static pthread_mutex_t darshan_core_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
41
static int my_rank = -1;
42
static int nprocs = -1;
43
static int darshan_mem_alignment = 1;
44

45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
/* paths prefixed with the following directories are not traced by darshan */
char* darshan_path_exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
#ifdef DARSHAN_BGQ
extern void bgq_runtime_initialize();
#endif

/* array of init functions for modules which need to be statically
 * initialized by darshan at startup time
 */
void (*mod_static_init_fns[])(void) =
{
#ifdef DARSHAN_BGQ
    &bgq_runtime_initialize,
#endif
    NULL
};

Shane Snyder's avatar
Shane Snyder committed
75 76 77
#define DARSHAN_CORE_LOCK() pthread_mutex_lock(&darshan_core_mutex)
#define DARSHAN_CORE_UNLOCK() pthread_mutex_unlock(&darshan_core_mutex)

78 79 80 81 82 83
/* FS mount information */
#define DARSHAN_MAX_MNTS 64
#define DARSHAN_MAX_MNT_PATH 256
#define DARSHAN_MAX_MNT_TYPE 32
struct mnt_data
{
84
    int block_size;
85 86 87 88 89 90
    char path[DARSHAN_MAX_MNT_PATH];
    char type[DARSHAN_MAX_MNT_TYPE];
};
static struct mnt_data mnt_data_array[DARSHAN_MAX_MNTS];
static int mnt_data_count = 0;

91 92
/* prototypes for internal helper functions */
static void darshan_log_record_hints_and_ver(
93
    struct darshan_core_runtime* core);
94 95
static void darshan_get_exe_and_mounts(
    struct darshan_core_runtime *core, int argc, char **argv);
96 97 98
static void darshan_add_record_hashref(
    struct darshan_core_runtime *core, char *name,
    darshan_record_id id, struct darshan_core_record_ref **ref);
99 100
static void darshan_block_size_from_path(
    const char *path, int *block_size);
101 102
static void darshan_get_logfile_name(
    char* logfile_name, int jobid, struct tm* start_tm);
103
static void darshan_get_shared_records(
104 105
    struct darshan_core_runtime *core, darshan_record_id **shared_recs,
    int *shared_rec_cnt);
106
static int darshan_log_open_all(
107
    char *logfile_name, MPI_File *log_fh);
108
static int darshan_deflate_buffer(
Shane Snyder's avatar
Shane Snyder committed
109 110
    void **pointers, int *lengths, int count, char *comp_buf,
    int *comp_buf_length);
111
static int darshan_log_write_record_hash(
112
    MPI_File log_fh, struct darshan_core_runtime *core,
113 114 115
    uint64_t *inout_off);
static int darshan_log_append_all(
    MPI_File log_fh, struct darshan_core_runtime *core, void *buf,
Shane Snyder's avatar
Shane Snyder committed
116
    int count, uint64_t *inout_off);
Shane Snyder's avatar
Shane Snyder committed
117 118
static void darshan_core_cleanup(
    struct darshan_core_runtime* core);
119

120 121
/* *********************************** */

Shane Snyder's avatar
Shane Snyder committed
122
void darshan_core_initialize(int argc, char **argv)
123
{
124
    struct darshan_core_runtime *init_core = NULL;
125 126
    int internal_timing_flag = 0;
    double init_start, init_time, init_max;
127 128 129
    int mmap_fd;
    int mmap_size;
    int sys_page_size;
130
    char *envstr;
131 132
    char *jobid_str;
    int jobid;
133 134
    int ret;
    int tmpval;
135
    int i;
136 137

    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
138
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
139 140 141 142

    if(getenv("DARSHAN_INTERNAL_TIMING"))
        internal_timing_flag = 1;

143
    if(internal_timing_flag)
144 145 146
        init_start = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* setup darshan runtime if darshan is enabled and hasn't been initialized already */
147
    if(!getenv("DARSHAN_DISABLE") && !darshan_core)
148
    {
149
        #if (__DARSHAN_MEM_ALIGNMENT < 1)
150 151
            #error Darshan must be configured with a positive value for --with-mem-align
        #endif
152
        envstr = getenv(DARSHAN_MEM_ALIGNMENT_OVERRIDE);
153 154 155 156 157 158 159 160 161 162 163
        if(envstr)
        {
            ret = sscanf(envstr, "%d", &tmpval);
            /* silently ignore if the env variable is set poorly */
            if(ret == 1 && tmpval > 0)
            {
                darshan_mem_alignment = tmpval;
            }
        }
        else
        {
164
            darshan_mem_alignment = __DARSHAN_MEM_ALIGNMENT;
165 166 167
        }

        /* avoid floating point errors on faulty input */
168
        if(darshan_mem_alignment < 1)
169 170 171
        {
            darshan_mem_alignment = 1;
        }
172

173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
        /* Use DARSHAN_JOBID_OVERRIDE for the env var for __DARSHAN_JOBID */
        envstr = getenv(DARSHAN_JOBID_OVERRIDE);
        if(!envstr)
        {
            envstr = __DARSHAN_JOBID;
        }

        /* find a job id */
        jobid_str = getenv(envstr);
        if(jobid_str)
        {
            /* in cobalt we can find it in env var */
            ret = sscanf(jobid_str, "%d", &jobid);
        }
        if(!jobid_str || ret != 1)
        {
            /* use pid as fall back */
            jobid = getpid();
        }

193 194 195
        /* allocate structure to track darshan core runtime information */
        init_core = malloc(sizeof(*init_core));
        if(init_core)
196
        {
197 198 199 200 201
            memset(init_core, 0, sizeof(*init_core));
            init_core->wtime_offset = DARSHAN_MPI_CALL(PMPI_Wtime)();

            sys_page_size = sysconf(_SC_PAGESIZE);
            assert(sys_page_size > 0);
202

203
            mmap_size = sizeof(struct darshan_header) + DARSHAN_JOB_RECORD_SIZE +
204
                + DARSHAN_RECORD_BUF_SIZE + DARSHAN_MOD_MEM_MAX;
205 206
            if(mmap_size % sys_page_size)
                mmap_size = ((mmap_size / sys_page_size) + 1) * sys_page_size;
207

208 209 210
            /* construct a unique temporary log file name for this process
             * to write mmap log data to
             */
211
            snprintf(init_core->mmap_log_name, PATH_MAX, "/tmp/darshan_job%d.%d",
212 213 214
                jobid, my_rank);

            /* create the temporary mmapped darshan log */
215
            mmap_fd = open(init_core->mmap_log_name, O_CREAT|O_RDWR|O_EXCL , 0644);
216
            if(mmap_fd < 0)
217
            {
218
                fprintf(stderr, "darshan library warning: "
219
                    "unable to create darshan log file %s\n", init_core->mmap_log_name);
220 221
                free(init_core);
                return;
222 223
            }

224
            /* TODO: what's more expensive? truncate or write zeros? perf test this call and later accesses */
225 226 227 228 229
            /* allocate the necessary space in the log file */
            ret = ftruncate(mmap_fd, mmap_size);
            if(ret < 0)
            {
                fprintf(stderr, "darshan library warning: "
230
                    "unable to allocate darshan log file %s\n", init_core->mmap_log_name);
231 232
                free(init_core);
                close(mmap_fd);
233
                unlink(init_core->mmap_log_name);
234 235 236 237 238
                return;
            }

            /* memory map buffers for getting at least some summary i/o data
             * into a log file if darshan does not shut down properly
239
             */
240 241
            void *mmap_p = mmap(NULL, mmap_size, PROT_WRITE, MAP_SHARED, mmap_fd, 0);
            if(mmap_p == MAP_FAILED)
242
            {
243
                fprintf(stderr, "darshan library warning: "
244
                    "unable to mmap darshan log file %s\n", init_core->mmap_log_name);
245 246
                free(init_core);
                close(mmap_fd);
247
                unlink(init_core->mmap_log_name);
248
                return;
249 250
            }

251 252 253
            /* close darshan log file (this does *not* unmap the log file) */
            close(mmap_fd);

254
            /* set the memory pointers for each log file region */
255
            init_core->log_hdr_p = (struct darshan_header *)mmap_p;
256
            init_core->log_job_p = (struct darshan_job *)
257
                ((char *)init_core->log_hdr_p + sizeof(struct darshan_header));
258
            init_core->log_exemnt_p = (char *)
259 260 261 262 263
                ((char *)init_core->log_job_p + sizeof(struct darshan_job));
            init_core->log_rec_p = (void *)
                ((char *)init_core->log_exemnt_p + DARSHAN_EXE_LEN + 1);
            init_core->log_mod_p = (void *)
                ((char *)init_core->log_rec_p + DARSHAN_RECORD_BUF_SIZE);
264 265 266 267 268

            /* set known header fields for the log file */
            strcpy(init_core->log_hdr_p->version_string, DARSHAN_LOG_VERSION);
            init_core->log_hdr_p->magic_nr = DARSHAN_MAGIC_NR;
            init_core->log_hdr_p->comp_type = DARSHAN_NO_COMP;
269 270
            init_core->log_hdr_p->rec_map.off =
                sizeof(struct darshan_header) + DARSHAN_JOB_RECORD_SIZE;
271

272 273 274 275 276
            /* set known job-level metadata fields for the log file */
            init_core->log_job_p->uid = getuid();
            init_core->log_job_p->start_time = time(NULL);
            init_core->log_job_p->nprocs = nprocs;
            init_core->log_job_p->jobid = (int64_t)jobid;
277 278 279 280 281 282

            /* if we are using any hints to write the log file, then record those
             * hints with the darshan job information
             */
            darshan_log_record_hints_and_ver(init_core);

283
            /* collect information about command line and mounted file systems */
284
            darshan_get_exe_and_mounts(init_core, argc, argv);
285

Shane Snyder's avatar
Shane Snyder committed
286
            /* bootstrap any modules with static initialization routines */
287 288 289 290 291 292 293 294
            i = 0;
            while(mod_static_init_fns[i])
            {
                (*mod_static_init_fns[i])();
                i++;
            }

            darshan_core = init_core;
295
        }
296 297
    }

298 299 300 301 302
    if(internal_timing_flag)
    {
        init_time = DARSHAN_MPI_CALL(PMPI_Wtime)() - init_start;
        DARSHAN_MPI_CALL(PMPI_Reduce)(&init_time, &init_max, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
303
        if(my_rank == 0)
304
        {
305 306
            fprintf(stderr, "#darshan:<op>\t<nprocs>\t<time>\n");
            fprintf(stderr, "darshan:init\t%d\t%f\n", nprocs, init_max);
307 308 309 310 311 312
        }
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
313
void darshan_core_shutdown()
314
{
315
    struct darshan_core_runtime *final_core;
316
    int internal_timing_flag = 0;
317 318 319 320
    struct tm *start_tm;
    time_t start_time_tmp;
    int64_t first_start_time;
    int64_t last_end_time;
321
    double start_log_time;
322
    double tm_end;
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342
    double open1, open2;
    double job1, job2;
    double rec1, rec2;
    double mod1[DARSHAN_MAX_MODS] = {0};
    double mod2[DARSHAN_MAX_MODS] = {0};
    double header1, header2;
    char *logfile_name;
    int local_mod_use[DARSHAN_MAX_MODS] = {0};
    int global_mod_use_count[DARSHAN_MAX_MODS] = {0};
    darshan_record_id *shared_recs;
    darshan_record_id *mod_shared_recs;
    int shared_rec_cnt = 0;
    int ret = 0;
    int all_ret = 0;
    int i;
    uint64_t gz_fp = 0;
    struct darshan_header out_header;
    struct darshan_job out_job;
    MPI_File log_fh;
    MPI_Status status;
343 344 345 346

    if(getenv("DARSHAN_INTERNAL_TIMING"))
        internal_timing_flag = 1;

347 348
    if(internal_timing_flag)
        start_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();
349

Shane Snyder's avatar
Shane Snyder committed
350
    /* disable darhan-core while we shutdown */
351
    DARSHAN_CORE_LOCK();
352
    if(!darshan_core)
353
    {
354
        DARSHAN_CORE_UNLOCK();
355 356
        return;
    }
357 358
    final_core = darshan_core;
    darshan_core = NULL;
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
    DARSHAN_CORE_UNLOCK();

    memcpy(&out_job, final_core->log_job_p, sizeof(struct darshan_job));

    /* XXX just copy mmap files somewhere else to avoid corruption */
    DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);
    if(my_rank == 0)
        system("cp /tmp/darshan* ~/Desktop");
    DARSHAN_MPI_CALL(PMPI_Barrier)(MPI_COMM_WORLD);

    /* indicate in the metadata field of the temporary darshan log file that
     * the darshan shutdown process was invoked on the data in the log. since
     * we have no way of knowing how far the shutdown process got, the data
     * could potentially be in inconsistent or corrupt state. in this case,
     * darshan currently does not generate a log file at all.
     */
    char *m = final_core->log_job_p->metadata + strlen(final_core->log_job_p->metadata);
    int meta_remain = DARSHAN_JOB_METADATA_LEN - strlen(final_core->log_job_p->metadata) - 1;
    snprintf(m, meta_remain, "darshan_shutdown=yes\n");
Shane Snyder's avatar
Shane Snyder committed
378 379 380 381 382

    for(i = 0; i < DARSHAN_MAX_MODS; i++)
    {
        if(final_core->mod_array[i])
        {
383
            local_mod_use[i] = 1;
384
            final_core->mod_array[i]->mod_funcs.begin_shutdown();
Shane Snyder's avatar
Shane Snyder committed
385 386
        }
    }
387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
        darshan_core_cleanup(final_core);
        return;
    }

    /* set the log file name on rank 0 */
    if(my_rank == 0)
    {
        /* use human readable start time format in log filename */
        start_time_tmp = final_core->log_job_p->start_time;
        start_tm = localtime(&start_time_tmp);

        darshan_get_logfile_name(logfile_name, final_core->log_job_p->jobid, start_tm);
    }

    /* broadcast log file name */
    DARSHAN_MPI_CALL(PMPI_Bcast)(logfile_name, PATH_MAX, MPI_CHAR, 0,
        MPI_COMM_WORLD);

    if(strlen(logfile_name) == 0)
    {
        /* failed to generate log file name */
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }
416

417
    final_core->log_job_p->end_time = time(NULL);
418

419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
    /* reduce to report first start time and last end time across all ranks
     * at rank 0
     */
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job_p->start_time, &first_start_time,
        1, MPI_LONG_LONG, MPI_MIN, 0, MPI_COMM_WORLD);
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job_p->end_time, &last_end_time,
        1, MPI_LONG_LONG, MPI_MAX, 0, MPI_COMM_WORLD);
    if(my_rank == 0)
    {
        out_job.start_time = first_start_time;
        out_job.end_time = last_end_time;
    }

    /* reduce the number of times a module was opened globally and bcast to everyone */
    DARSHAN_MPI_CALL(PMPI_Allreduce)(local_mod_use, global_mod_use_count,
        DARSHAN_MAX_MODS, MPI_INT, MPI_SUM, MPI_COMM_WORLD);

    /* get a list of records which are shared across all processes */
    darshan_get_shared_records(final_core, &shared_recs, &shared_rec_cnt);

    if(internal_timing_flag)
        open1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
    /* collectively open the darshan log file */
    ret = darshan_log_open_all(logfile_name, &log_fh);
    if(internal_timing_flag)
        open2 = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* error out if unable to open log file */
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
            fprintf(stderr, "darshan library warning: unable to open log file %s\n",
                logfile_name);
        }
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }

    if(internal_timing_flag)
        job1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
    /* rank 0 is responsible for writing the compressed darshan job information */
    if(my_rank == 0)
    {
        void *pointers[2] = {&out_job, final_core->log_exemnt_p};
        int lengths[2] = {sizeof(struct darshan_job), strlen(final_core->log_exemnt_p)};
        int comp_buf_sz = 0;

        /* compress the job info and the trailing mount/exe data */
        all_ret = darshan_deflate_buffer(pointers, lengths, 2,
            final_core->comp_buf, &comp_buf_sz);
        if(all_ret)
        {
            fprintf(stderr, "darshan library warning: unable to compress job data\n");
            unlink(logfile_name);
        }
        else
        {
            /* write the job information, preallocing space for the log header */
            gz_fp += sizeof(struct darshan_header);
            all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, gz_fp,
                final_core->comp_buf, comp_buf_sz, MPI_BYTE, &status);
            if(all_ret != MPI_SUCCESS)
            {
                fprintf(stderr,
                        "darshan library warning: unable to write job data to log file %s\n",
                        logfile_name);
                unlink(logfile_name);

            }
            gz_fp += comp_buf_sz;
        }
    }

    /* error out if unable to write job information */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&all_ret, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }
    if(internal_timing_flag)
        job2 = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* error out if unable to write job information */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&all_ret, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }
    if(internal_timing_flag)
        job2 = DARSHAN_MPI_CALL(PMPI_Wtime)();

    if(internal_timing_flag)
        rec1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
    /* write the record name->id hash to the log file */
    out_header.rec_map.off = gz_fp;
    ret = darshan_log_write_record_hash(log_fh, final_core, &gz_fp);
    out_header.rec_map.len = gz_fp - out_header.rec_map.off;

    /* error out if unable to write record hash */
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
            fprintf(stderr,
                "darshan library warning: unable to write record hash to log file %s\n",
                logfile_name);
            unlink(logfile_name);
        }
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }
    if(internal_timing_flag)
        rec2 = DARSHAN_MPI_CALL(PMPI_Wtime)();

    mod_shared_recs = malloc(shared_rec_cnt * sizeof(darshan_record_id));
    assert(mod_shared_recs);

    /* loop over globally used darshan modules and:
     *      - perform shared file reductions, if possible
     *      - get final output buffer
     *      - compress (zlib) provided output buffer
     *      - append compressed buffer to log file
     *      - add module index info (file offset/length) to log header
     *      - shutdown the module
     */
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
    {
        struct darshan_core_module* this_mod = final_core->mod_array[i];
        struct darshan_core_record_ref *ref = NULL;
        int mod_shared_rec_cnt = 0;
        void* mod_buf = NULL;
        int mod_buf_sz = 0;
        int j;

        if(global_mod_use_count[i] == 0)
        {
            if(my_rank == 0)
            {
                out_header.mod_map[i].off = 0;
                out_header.mod_map[i].len = 0;
            }
            continue;
        }

        if(internal_timing_flag)
            mod1[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();

        /* set the shared file list for this module */
        memset(mod_shared_recs, 0, shared_rec_cnt * sizeof(darshan_record_id));
        for(j = 0; j < shared_rec_cnt; j++)
        {
            HASH_FIND(hlink, final_core->rec_hash, &shared_recs[j],
                sizeof(darshan_record_id), ref);
            assert(ref);
            if(DARSHAN_MOD_FLAG_ISSET(ref->global_mod_flags, i))
            {
                mod_shared_recs[mod_shared_rec_cnt++] = shared_recs[j];
            }
        }

        /* if module is registered locally, get the corresponding output buffer
         * 
         * NOTE: this function can be used to run collective operations across
         * modules, if there are file records shared globally.
         */
        if(this_mod)
        {
            this_mod->mod_funcs.get_output_data(MPI_COMM_WORLD, mod_shared_recs,
                mod_shared_rec_cnt, &mod_buf, &mod_buf_sz);
        }

        /* append this module's data to the darshan log */
        out_header.mod_map[i].off = gz_fp;
        ret = darshan_log_append_all(log_fh, final_core, mod_buf, mod_buf_sz, &gz_fp);
        out_header.mod_map[i].len = gz_fp - out_header.mod_map[i].off;

        /* error out if the log append failed */
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
            MPI_LOR, MPI_COMM_WORLD);
        if(all_ret != 0)
        {
            if(my_rank == 0)
            {
                fprintf(stderr,
                    "darshan library warning: unable to write %s module data to log file %s\n",
                    darshan_module_names[i], logfile_name);
                unlink(logfile_name);
            }
            free(logfile_name);
            darshan_core_cleanup(final_core);
            return;
        }

        /* shutdown module if registered locally */
        if(this_mod)
        {
            this_mod->mod_funcs.shutdown();
        }
        if(internal_timing_flag)
            mod2[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
    }

    if(internal_timing_flag)
        header1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
    /* write out log header, after running 2 reduction on header variables:
     *  1) reduce 'partial_flag' variable to determine which modules ran out
     *     of memory for storing I/O data
     *  2) reduce 'mod_ver' array to determine which log format version each
     *     module used for this output log
     */
    DARSHAN_MPI_CALL(PMPI_Reduce)(&(final_core->log_hdr_p->partial_flag),
        &(out_header.partial_flag), 1, MPI_UINT32_T, MPI_BOR, 0, MPI_COMM_WORLD);
    if(my_rank == 0)
    {
        /* rank 0 is responsible for writing the log header */
        /* initialize the remaining header fields */
        strcpy(out_header.version_string, DARSHAN_LOG_VERSION);
        out_header.magic_nr = DARSHAN_MAGIC_NR;
        out_header.comp_type = DARSHAN_ZLIB_COMP;

        all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, 0, &out_header,
            sizeof(struct darshan_header), MPI_BYTE, &status);
        if(all_ret != MPI_SUCCESS)
        {
            fprintf(stderr, "darshan library warning: unable to write header to log file %s\n",
                    logfile_name);
            unlink(logfile_name);
        }
    }

    /* error out if unable to write log header */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&all_ret, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        free(logfile_name);
        darshan_core_cleanup(final_core);
        return;
    }
    if(internal_timing_flag)
        header2 = DARSHAN_MPI_CALL(PMPI_Wtime)();

    DARSHAN_MPI_CALL(PMPI_File_close)(&log_fh);

    /* if we got this far, there are no errors, so rename from *.darshan_partial
     * to *-<logwritetime>.darshan, which indicates that this log file is
     * complete and ready for analysis
     */
    if(my_rank == 0)
    {
        if(getenv("DARSHAN_LOGFILE"))
        {
#ifdef __DARSHAN_GROUP_READABLE_LOGS
            chmod(logfile_name, (S_IRUSR|S_IRGRP));
#else
            chmod(logfile_name, (S_IRUSR));
#endif
        }
        else
        {
            char* tmp_index;
            double end_log_time;
            char* new_logfile_name;

            new_logfile_name = malloc(PATH_MAX);
            if(new_logfile_name)
            {
                new_logfile_name[0] = '\0';
                end_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();
                strcat(new_logfile_name, logfile_name);
                tmp_index = strstr(new_logfile_name, ".darshan_partial");
                sprintf(tmp_index, "_%d.darshan", (int)(end_log_time-start_log_time+1));
                rename(logfile_name, new_logfile_name);
                /* set permissions on log file */
#ifdef __DARSHAN_GROUP_READABLE_LOGS
                chmod(new_logfile_name, (S_IRUSR|S_IRGRP));
#else
                chmod(new_logfile_name, (S_IRUSR));
#endif
                free(new_logfile_name);
            }
        }
    }

713 714
    /* remove the temporary mmap log files */
    unlink(final_core->mmap_log_name);
715 716

    free(logfile_name);
717
    darshan_core_cleanup(final_core);
718

719
    if(internal_timing_flag)
720
    {
721 722 723 724 725
        double open_tm, open_slowest;
        double header_tm, header_slowest;
        double job_tm, job_slowest;
        double rec_tm, rec_slowest;
        double mod_tm[DARSHAN_MAX_MODS], mod_slowest[DARSHAN_MAX_MODS];
726 727 728 729
        double all_tm, all_slowest;

        tm_end = DARSHAN_MPI_CALL(PMPI_Wtime)();

730 731 732 733
        open_tm = open2 - open1;
        header_tm = header2 - header1;
        job_tm = job2 - job1;
        rec_tm = rec2 - rec1;
734
        all_tm = tm_end - start_log_time;
735 736 737 738
        for(i = 0;i < DARSHAN_MAX_MODS; i++)
        {
            mod_tm[i] = mod2[i] - mod1[i];
        }
739

740 741 742 743 744 745 746 747
        DARSHAN_MPI_CALL(PMPI_Reduce)(&open_tm, &open_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&header_tm, &header_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&job_tm, &job_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&rec_tm, &rec_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
748 749
        DARSHAN_MPI_CALL(PMPI_Reduce)(&all_tm, &all_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
750 751
        DARSHAN_MPI_CALL(PMPI_Reduce)(mod_tm, mod_slowest, DARSHAN_MAX_MODS,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
752 753 754

        if(my_rank == 0)
        {
755
            fprintf(stderr, "#darshan:<op>\t<nprocs>\t<time>\n");
756 757 758 759 760 761 762 763 764 765
            fprintf(stderr, "darshan:log_open\t%d\t%f\n", nprocs, open_slowest);
            fprintf(stderr, "darshan:job_write\t%d\t%f\n", nprocs, job_slowest);
            fprintf(stderr, "darshan:hash_write\t%d\t%f\n", nprocs, rec_slowest);
            fprintf(stderr, "darshan:header_write\t%d\t%f\n", nprocs, header_slowest);
            for(i = 0; i < DARSHAN_MAX_MODS; i++)
            {
                if(global_mod_use_count[i])
                    fprintf(stderr, "darshan:%s_shutdown\t%d\t%f\n", darshan_module_names[i],
                        nprocs, mod_slowest[i]);
            }
766
            fprintf(stderr, "darshan:core_shutdown\t%d\t%f\n", nprocs, all_slowest);
767
        }
768
    }
769

770 771
    return;
}
772

Shane Snyder's avatar
Shane Snyder committed
773
/* *********************************** */
774

775
/* record any hints used to write the darshan log in the job data */
776
static void darshan_log_record_hints_and_ver(struct darshan_core_runtime* core)
777 778
{
    char* hints;
779
    char* job_hints;
780 781 782 783 784 785
    int meta_remain = 0;
    char* m;

    /* check environment variable to see if the default MPI file hints have
     * been overridden
     */
786
    hints = getenv(DARSHAN_LOG_HINTS_OVERRIDE);
787 788
    if(!hints)
    {
789
        hints = __DARSHAN_LOG_HINTS;
790 791 792 793 794
    }

    if(!hints || strlen(hints) < 1)
        return;

795 796
    job_hints = strdup(hints);
    if(!job_hints)
797 798 799
        return;

    meta_remain = DARSHAN_JOB_METADATA_LEN -
800
        strlen(core->log_job_p->metadata) - 1;
801 802
    if(meta_remain >= (strlen(PACKAGE_VERSION) + 9))
    {
803
        sprintf(core->log_job_p->metadata, "lib_ver=%s\n", PACKAGE_VERSION);
804 805
        meta_remain -= (strlen(PACKAGE_VERSION) + 9);
    }
806
    if(meta_remain >= (3 + strlen(job_hints)))
807
    {
808
        m = core->log_job_p->metadata + strlen(core->log_job_p->metadata);
809
        /* We have room to store the hints in the metadata portion of
810
         * the job structure.  We just prepend an h= to the hints list.  The
811 812 813
         * metadata parser will ignore = characters that appear in the value
         * portion of the metadata key/value pair.
         */
814
        sprintf(m, "h=%s\n", job_hints);
815
    }
816
    free(job_hints);
817 818 819 820

    return;
}

821 822 823 824 825 826 827 828 829 830 831 832 833 834
static int mnt_data_cmp(const void* a, const void* b)
{
    const struct mnt_data *d_a = (const struct mnt_data*)a;
    const struct mnt_data *d_b = (const struct mnt_data*)b;

    if(strlen(d_a->path) > strlen(d_b->path))
        return(-1);
    else if(strlen(d_a->path) < strlen(d_b->path))
        return(1);
    else
        return(0);
}

/* adds an entry to table of mounted file systems */
835
static void add_entry(char* buf, int* space_left, struct mntent *entry)
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859
{
    int ret;
    char tmp_mnt[256];
    struct statfs statfsbuf;

    strncpy(mnt_data_array[mnt_data_count].path, entry->mnt_dir,
        DARSHAN_MAX_MNT_PATH-1);
    strncpy(mnt_data_array[mnt_data_count].type, entry->mnt_type,
        DARSHAN_MAX_MNT_TYPE-1);
    /* NOTE: we now try to detect the preferred block size for each file 
     * system using fstatfs().  On Lustre we assume a size of 1 MiB 
     * because fstatfs() reports 4 KiB. 
     */
#ifndef LL_SUPER_MAGIC
#define LL_SUPER_MAGIC 0x0BD00BD0
#endif
    ret = statfs(entry->mnt_dir, &statfsbuf);
    if(ret == 0 && statfsbuf.f_type != LL_SUPER_MAGIC)
        mnt_data_array[mnt_data_count].block_size = statfsbuf.f_bsize;
    else if(ret == 0 && statfsbuf.f_type == LL_SUPER_MAGIC)
        mnt_data_array[mnt_data_count].block_size = 1024*1024;
    else
        mnt_data_array[mnt_data_count].block_size = 4096;

860
    /* store mount information with the job-level metadata in darshan log */
861
    ret = snprintf(tmp_mnt, 256, "\n%s\t%s",
862 863 864
        entry->mnt_type, entry->mnt_dir);
    if(ret < 256 && strlen(tmp_mnt) <= (*space_left))
    {
865
        strcat(buf, tmp_mnt);
866 867 868 869 870 871 872
        (*space_left) -= strlen(tmp_mnt);
    }

    mnt_data_count++;
    return;
}

873
/* darshan_get_exe_and_mounts()
874 875
 *
 * collects command line and list of mounted file systems into a string that
876
 * will be stored with the job-level metadata
877
 */
878
static void darshan_get_exe_and_mounts(struct darshan_core_runtime *core,
879
    int argc, char **argv)
880 881 882 883
{
    FILE* tab;
    struct mntent *entry;
    char* exclude;
884 885 886 887
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int space_left = DARSHAN_EXE_LEN;
    int i;
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908
    int tmp_index = 0;
    int skip = 0;

    /* skip these fs types */
    static char* fs_exclusions[] = {
        "tmpfs",
        "proc",
        "sysfs",
        "devpts",
        "binfmt_misc",
        "fusectl",
        "debugfs",
        "securityfs",
        "nfsd",
        "none",
        "rpc_pipefs",
        "hugetlbfs",
        "cgroup",
        NULL
    };

909 910 911
    /* record exe and arguments */
    for(i=0; i<argc; i++)
    {
912 913
        strncat(core->log_exemnt_p, argv[i], space_left);
        space_left = DARSHAN_EXE_LEN-strlen(core->log_exemnt_p);
914 915
        if(i < (argc-1))
        {
916 917
            strncat(core->log_exemnt_p, " ", space_left);
            space_left = DARSHAN_EXE_LEN-strlen(core->log_exemnt_p);
918 919 920 921 922 923 924 925
        }
    }

    /* if we don't see any arguments, then use glibc symbol to get
     * program name at least (this happens in fortran)
     */
    if(argc == 0)
    {
926 927 928 929
        strncat(core->log_exemnt_p, __progname_full, space_left);
        space_left = DARSHAN_EXE_LEN-strlen(core->log_exemnt_p);
        strncat(core->log_exemnt_p, " <unknown args>", space_left);
        space_left = DARSHAN_EXE_LEN-strlen(core->log_exemnt_p);
930 931 932 933 934 935
    }

    if(space_left == 0)
    {
        /* we ran out of room; mark that string was truncated */
        truncate_offset = DARSHAN_EXE_LEN - strlen(truncate_string);
936
        sprintf(&(core->log_exemnt_p[truncate_offset]), "%s",
937 938 939
            truncate_string);
    }

940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966
    /* we make two passes through mounted file systems; in the first pass we
     * grab any non-nfs mount points, then on the second pass we grab nfs
     * mount points
     */

    tab = setmntent("/etc/mtab", "r");
    if(!tab)
        return;
    /* loop through list of mounted file systems */
    while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
    {
        /* filter out excluded fs types */
        tmp_index = 0;
        skip = 0;
        while((exclude = fs_exclusions[tmp_index]))
        {
            if(!(strcmp(exclude, entry->mnt_type)))
            {
                skip =1;
                break;
            }
            tmp_index++;
        }

        if(skip || (strcmp(entry->mnt_type, "nfs") == 0))
            continue;

967
        add_entry(core->log_exemnt_p, &space_left, entry);
968 969 970 971 972 973 974 975 976 977 978 979
    }
    endmntent(tab);

    tab = setmntent("/etc/mtab", "r");
    if(!tab)
        return;
    /* loop through list of mounted file systems */
    while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
    {
        if(strcmp(entry->mnt_type, "nfs") != 0)
            continue;

980
        add_entry(core->log_exemnt_p, &space_left, entry);
981 982 983
    }
    endmntent(tab);

984
    /* sort mount points in order of longest path to shortest path.  This is
985 986 987 988 989 990 991
     * necessary so that if we try to match file paths to mount points later
     * we don't match on "/" every time.
     */
    qsort(mnt_data_array, mnt_data_count, sizeof(mnt_data_array[0]), mnt_data_cmp);
    return;
}

992 993 994 995 996
static void darshan_add_record_hashref(struct darshan_core_runtime *core,
    char *name, darshan_record_id id, struct darshan_core_record_ref **ref)
{
    int record_size = sizeof(darshan_record_id) + strlen(name) + 1;

997
    if((record_size + core->log_hdr_p->rec_map.len) > DARSHAN_RECORD_BUF_SIZE)
998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
        return;

    *ref = malloc(sizeof(**ref));
    if(*ref)
    {
        memset(*ref, 0, sizeof(**ref));

#if 0
        if(!mmap)
        {
            ref->rec.name = malloc(strlen(name) + 1);
        }
        else
#endif
        {
            /* store the rec id and full file path in record hash buffer */
1014
            void *tmp_p = (char *)core->log_rec_p + core->log_hdr_p->rec_map.len;
1015 1016 1017 1018 1019 1020
            *(darshan_record_id *)tmp_p = id;

            /* set the name pointer for this record to point to the
             * appropriate location in the record hash buffer
             */
            tmp_p = (char *)tmp_p + sizeof(darshan_record_id);
1021
            (*ref)->name = (char *)tmp_p;
1022 1023 1024
        }

        /* set record ref fields */
1025 1026 1027
        (*ref)->id = id;
        if((*ref)->name)
            strcpy((*ref)->name, name);
1028 1029

        /* TODO: look at HASH_ADD_KEYPTR, use same strategy (big contig pool) for non-mmap darshan */
1030
        HASH_ADD(hlink, core->rec_hash, id, sizeof(darshan_record_id), (*ref));
1031 1032 1033 1034 1035 1036 1037
        core->rec_hash_cnt++;
        core->log_hdr_p->rec_map.len += record_size;
    }

    return;
}

1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
static void darshan_block_size_from_path(const char *path, int *block_size)
{
    int i;
    *block_size = -1;

    for(i=0; i<mnt_data_count; i++)
    {
        if(!(strncmp(mnt_data_array[i].path, path, strlen(mnt_data_array[i].path))))
        {
            *block_size = mnt_data_array[i].block_size;
            return;
        }
    }

    return;
}

1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 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 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 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
/* construct the darshan log file name */
static void darshan_get_logfile_name(char* logfile_name, int jobid, struct tm* start_tm)
{
    char* user_logfile_name;
    char* logpath;
    char* logname_string;
    char* logpath_override = NULL;
#ifdef __DARSHAN_LOG_ENV
    char env_check[256];
    char* env_tok;
#endif
    uint64_t hlevel;
    char hname[HOST_NAME_MAX];
    uint64_t logmod;
    char cuser[L_cuserid] = {0};
    int ret;

    /* first, check if user specifies a complete logpath to use */
    user_logfile_name = getenv("DARSHAN_LOGFILE");
    if(user_logfile_name)
    {
        if(strlen(user_logfile_name) >= (PATH_MAX-1))
        {
            fprintf(stderr, "darshan library warning: user log file name too long.\n");
            logfile_name[0] = '\0';
        }
        else
        {
            strcpy(logfile_name, user_logfile_name);
        }
    }
    else
    {
        /* otherwise, generate the log path automatically */

        /* Use DARSHAN_LOG_PATH_OVERRIDE for the value or __DARSHAN_LOG_PATH */
        logpath = getenv(DARSHAN_LOG_PATH_OVERRIDE);
        if(!logpath)
        {
#ifdef __DARSHAN_LOG_PATH
            logpath = __DARSHAN_LOG_PATH;
#endif
        }

        /* get the username for this job.  In order we will try each of the
         * following until one of them succeeds:
         *
         * - cuserid()
         * - getenv("LOGNAME")
         * - snprintf(..., geteuid());
         *
         * Note that we do not use getpwuid() because it generally will not
         * work in statically compiled binaries.
         */

#ifndef DARSHAN_DISABLE_CUSERID
        cuserid(cuser);
#endif

        /* if cuserid() didn't work, then check the environment */
        if(strcmp(cuser, "") == 0)
        {
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }
        }

        /* if cuserid() and environment both fail, then fall back to uid */
        if(strcmp(cuser, "") == 0)
        {
            uid_t uid = geteuid();
            snprintf(cuser, sizeof(cuser), "%u", uid);
        }

        /* generate a random number to help differentiate the log */
        hlevel=DARSHAN_MPI_CALL(PMPI_Wtime)() * 1000000;
        (void)gethostname(hname, sizeof(hname));
        logmod = darshan_hash((void*)hname,strlen(hname),hlevel);

        /* see if darshan was configured using the --with-logpath-by-env
         * argument, which allows the user to specify an absolute path to
         * place logs via an env variable.
         */
#ifdef __DARSHAN_LOG_ENV
        /* just silently skip if the environment variable list is too big */
        if(strlen(__DARSHAN_LOG_ENV) < 256)
        {
            /* copy env variable list to a temporary buffer */
            strcpy(env_check, __DARSHAN_LOG_ENV);
            /* tokenize the comma-separated list */
            env_tok = strtok(env_check, ",");
            if(env_tok)
            {
                do
                {
                    /* check each env variable in order */
                    logpath_override = getenv(env_tok);
                    if(logpath_override)
                    {
                        /* stop as soon as we find a match */
                        break;
                    }
                }while((env_tok = strtok(NULL, ",")));
            }
        }
#endif

        if(logpath_override)
        {
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath_override,
                cuser, __progname, jobid,
                (start_tm->tm_mon+1),
                start_tm->tm_mday,
                (start_tm->tm_hour*60*60 + start_tm->tm_min*60 + start_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath_override, jobid);
            }
        }
        else if(logpath)
        {
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%d/%d/%d/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath, (start_tm->tm_year+1900),
                (start_tm->tm_mon+1), start_tm->tm_mday,
                cuser, __progname, jobid,
                (start_tm->tm_mon+1),
                start_tm->tm_mday,
                (start_tm->tm_hour*60*60 + start_tm->tm_min*60 + start_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath, jobid);
            }
        }
        else
        {
            logfile_name[0] = '\0';
        }
    }

    return;
}

1210
static void darshan_get_shared_records(struct darshan_core_runtime *core,
1211
    darshan_record_id **shared_recs, int *shared_rec_cnt)
1212
{
1213
    int i, j;
1214
    int tmp_cnt = core->rec_hash_cnt;
1215
    struct darshan_core_record_ref *tmp, *ref;
1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
    darshan_record_id *id_array;
    uint64_t *mod_flags;
    uint64_t *global_mod_flags;

    /* broadcast root's number of records to all other processes */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&tmp_cnt, 1, MPI_INT, 0, MPI_COMM_WORLD);

    /* use root record count to allocate data structures */
    id_array = malloc(tmp_cnt * sizeof(darshan_record_id));
    mod_flags = malloc(tmp_cnt * sizeof(uint64_t));
    global_mod_flags = malloc(tmp_cnt * sizeof(uint64_t));
    *shared_recs = malloc(tmp_cnt * sizeof(darshan_record_id));
    assert(id_array && mod_flags && global_mod_flags && *shared_recs);

    memset(mod_flags, 0, tmp_cnt * sizeof(uint64_t));
    memset(global_mod_flags, 0, tmp_cnt * sizeof(uint64_t));
    memset(*shared_recs, 0, tmp_cnt * sizeof(darshan_record_id));
1233 1234 1235 1236

    /* first, determine list of records root process has opened */
    if(my_rank == 0)
    {
1237
        i = 0;
1238
        HASH_ITER(hlink, core->rec_hash, ref, tmp)
1239
        {
1240
            id_array[i++] = ref->id;           
1241 1242 1243 1244
        }
    }

    /* broadcast root's list of records to all other processes */
1245
    DARSHAN_MPI_CALL(PMPI_Bcast)(id_array, (tmp_cnt * sizeof(darshan_record_id)),
1246 1247 1248
        MPI_BYTE, 0, MPI_COMM_WORLD);

    /* everyone looks to see if they opened the same records as root */