darshan-core.c 51.3 KB
Newer Older
1
/*
Shane Snyder's avatar
Shane Snyder committed
2 3 4
 * Copyright (C) 2015 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
5 6
 */

7
#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 22 23
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/vfs.h>
24
#include <zlib.h>
25
#include <mpi.h>
26
#include <assert.h>
27

28
#include "uthash.h"
Shane Snyder's avatar
Shane Snyder committed
29
#include "darshan.h"
30
#include "darshan-core.h"
Shane Snyder's avatar
Shane Snyder committed
31
#include "darshan-dynamic.h"
32

33
extern char* __progname;
34
extern char* __progname_full;
35

36
/* internal variable delcarations */
37
static struct darshan_core_runtime *darshan_core = NULL;
38
static pthread_mutex_t darshan_core_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
39
static int my_rank = -1;
40
static int nprocs = -1;
41
static int darshan_mem_alignment = 1;
42

43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
/* 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
};

Shane Snyder's avatar
Shane Snyder committed
58 59 60
#define DARSHAN_CORE_LOCK() pthread_mutex_lock(&darshan_core_mutex)
#define DARSHAN_CORE_UNLOCK() pthread_mutex_unlock(&darshan_core_mutex)

61 62 63 64 65 66
/* FS mount information */
#define DARSHAN_MAX_MNTS 64
#define DARSHAN_MAX_MNT_PATH 256
#define DARSHAN_MAX_MNT_TYPE 32
struct mnt_data
{
67
    int block_size;
68 69 70 71 72 73
    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;

74 75 76 77
/* prototypes for internal helper functions */
static void darshan_get_logfile_name(
    char* logfile_name, int jobid, struct tm* start_tm);
static void darshan_log_record_hints_and_ver(
78 79 80 81 82 83
    struct darshan_core_runtime* core);
static void darshan_get_exe_and_mounts_root(
    struct darshan_core_runtime *core, char* trailing_data,
    int space_left);
static char* darshan_get_exe_and_mounts(
    struct darshan_core_runtime *core);
84 85
static void darshan_block_size_from_path(
    const char *path, int *block_size);
86
static void darshan_get_shared_records(
87
    struct darshan_core_runtime *core, darshan_record_id *shared_recs);
88
static int darshan_log_open_all(
89
    char *logfile_name, MPI_File *log_fh);
90 91 92
static int darshan_deflate_buffer(
    void **pointers, int *lengths, int count, int nocomp_flag,
    char *comp_buf, int *comp_length);
93
static int darshan_log_write_record_hash(
94
    MPI_File log_fh, struct darshan_core_runtime *core,
95 96 97 98
    uint64_t *inout_off);
static int darshan_log_append_all(
    MPI_File log_fh, struct darshan_core_runtime *core, void *buf,
    int count, uint64_t *inout_off, uint64_t *agg_uncomp_sz);
Shane Snyder's avatar
Shane Snyder committed
99 100
static void darshan_core_cleanup(
    struct darshan_core_runtime* core);
101

102 103
/* *********************************** */

Shane Snyder's avatar
Shane Snyder committed
104
void darshan_core_initialize(int argc, char **argv)
105 106 107 108
{
    int i;
    int internal_timing_flag = 0;
    double init_start, init_time, init_max;
109
    char *envstr;
110 111 112
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
113 114
    int ret;
    int tmpval;
115 116

    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
117
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
118 119 120 121

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

122
    if(internal_timing_flag)
123 124 125
        init_start = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* setup darshan runtime if darshan is enabled and hasn't been initialized already */
126
    if(!getenv("DARSHAN_DISABLE") && !darshan_core)
127
    {
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
        #if (__CP_MEM_ALIGNMENT < 1)
            #error Darshan must be configured with a positive value for --with-mem-align
        #endif
        envstr = getenv("DARSHAN_MEMALIGN");
        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
        {
            darshan_mem_alignment = __CP_MEM_ALIGNMENT;
        }

        /* avoid floating point errors on faulty input */
        if (darshan_mem_alignment < 1)
        {
            darshan_mem_alignment = 1;
        }
151 152 153 154

        /* allocate structure to track darshan_core_runtime information */
        darshan_core = malloc(sizeof(*darshan_core));
        if(darshan_core)
155
        {
156
            memset(darshan_core, 0, sizeof(*darshan_core));
157

158 159 160 161
            darshan_core->log_job.uid = getuid();
            darshan_core->log_job.start_time = time(NULL);
            darshan_core->log_job.nprocs = nprocs;
            darshan_core->wtime_offset = DARSHAN_MPI_CALL(PMPI_Wtime)();
162 163

            /* record exe and arguments */
164
            for(i=0; i<argc; i++)
165
            {
166
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
167 168
                strncat(darshan_core->exe, argv[i], chars_left);
                if(i < (argc-1))
169
                {
170
                    chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
171
                    strncat(darshan_core->exe, " ", chars_left);
172 173 174 175 176 177 178 179
                }
            }

            /* if we don't see any arguments, then use glibc symbol to get
             * program name at least (this happens in fortran)
             */
            if(argc == 0)
            {
180
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
181
                strncat(darshan_core->exe, __progname_full, chars_left);
182
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
183
                strncat(darshan_core->exe, " <unknown args>", chars_left);
184 185 186 187 188
            }

            if(chars_left == 0)
            {
                /* we ran out of room; mark that string was truncated */
189
                truncate_offset = DARSHAN_EXE_LEN - strlen(truncate_string);
190
                sprintf(&darshan_core->exe[truncate_offset], "%s",
191 192
                    truncate_string);
            }
193 194 195

            /* collect information about command line and mounted file systems */
            darshan_core->trailing_data = darshan_get_exe_and_mounts(darshan_core);
196 197 198 199 200 201 202 203
        }
    }

    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);
204
        if(my_rank == 0)
205 206 207 208 209 210 211 212 213
        {
            printf("#darshan:<op>\t<nprocs>\t<time>\n");
            printf("darshan:init\t%d\t%f\n", nprocs, init_max);
        }
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
214
void darshan_core_shutdown()
215
{
216
    int i;
217
    char *logfile_name;
218
    struct darshan_core_runtime *final_core;
219
    int internal_timing_flag = 0;
220 221
    char *envjobid;
    char *jobid_str;
222
    int jobid;
223
    struct tm *start_tm;
224
    time_t start_time_tmp;
225 226
    int ret = 0;
    int all_ret = 0;
227 228
    int64_t first_start_time;
    int64_t last_end_time;
229 230
    int local_mod_use[DARSHAN_MAX_MODS] = {0};
    int global_mod_use_count[DARSHAN_MAX_MODS] = {0};
231
    darshan_record_id shared_recs[DARSHAN_CORE_MAX_RECORDS] = {0};
232
    double start_log_time;
233 234 235 236 237 238 239
    double open1, open2;
    double job1, job2;
    double rec1, rec2;
    double mod1[DARSHAN_MAX_MODS] = {0};
    double mod2[DARSHAN_MAX_MODS] = {0};
    double header1, header2;
    double tm_end;
240 241
    uint64_t gz_fp = 0;
    uint64_t tmp_off = 0;
242 243
    MPI_File log_fh;
    MPI_Status status;
244 245 246 247

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

248 249
    start_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();

Shane Snyder's avatar
Shane Snyder committed
250
    /* disable darhan-core while we shutdown */
251
    DARSHAN_CORE_LOCK();
252
    if(!darshan_core)
253
    {
254
        DARSHAN_CORE_UNLOCK();
255 256
        return;
    }
257 258
    final_core = darshan_core;
    darshan_core = NULL;
Shane Snyder's avatar
Shane Snyder committed
259

260
    /* we also need to set which modules were registered on this process and
Shane Snyder's avatar
Shane Snyder committed
261 262 263 264 265 266 267 268 269 270
     * disable tracing within those modules while we shutdown
     */
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
    {
        if(final_core->mod_array[i])
        {
            local_mod_use[i] = 1;
            final_core->mod_array[i]->mod_funcs.disable_instrumentation();
        }
    }
271
    DARSHAN_CORE_UNLOCK();
272 273 274 275

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
276
        darshan_core_cleanup(final_core);
277 278 279
        return;
    }

280
    /* set darshan job id/metadata and constuct log file name on rank 0 */
281
    if(my_rank == 0)
282 283 284
    {
        /* Use CP_JOBID_OVERRIDE for the env var or CP_JOBID */
        envjobid = getenv(CP_JOBID_OVERRIDE);
285
        if(!envjobid)
286 287 288 289
        {
            envjobid = CP_JOBID;
        }

290
        /* find a job id */
291 292 293 294 295 296 297 298 299 300 301 302
        jobid_str = getenv(envjobid);
        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();
        }

303
        final_core->log_job.jobid = (int64_t)jobid;
304

305
        /* if we are using any hints to write the log file, then record those
306
         * hints with the darshan job information
307
         */
308
        darshan_log_record_hints_and_ver(final_core);
309

310
        /* use human readable start time format in log filename */
311
        start_time_tmp = final_core->log_job.start_time;
312
        start_tm = localtime(&start_time_tmp);
313

314 315
        /* construct log file name */
        darshan_get_logfile_name(logfile_name, jobid, start_tm);
316 317 318 319 320 321 322 323 324
    }

    /* 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 */
325
        free(logfile_name);
326
        darshan_core_cleanup(final_core);
327 328 329
        return;
    }

330
    final_core->log_job.end_time = time(NULL);
331

332 333 334
    /* reduce to report first start time and last end time across all ranks
     * at rank 0
     */
335 336
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job.start_time, &first_start_time, 1, MPI_LONG_LONG, MPI_MIN, 0, MPI_COMM_WORLD);
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job.end_time, &last_end_time, 1, MPI_LONG_LONG, MPI_MAX, 0, MPI_COMM_WORLD);
337 338
    if(my_rank == 0)
    {
339 340
        final_core->log_job.start_time = first_start_time;
        final_core->log_job.end_time = last_end_time;
341
    }
342

343 344 345
    /* 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);

346
    /* get a list of records which are shared across all processes */
347
    darshan_get_shared_records(final_core, shared_recs);
348

349 350
    if(internal_timing_flag)
        open1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
351
    /* collectively open the darshan log file */
352
    ret = darshan_log_open_all(logfile_name, &log_fh);
353 354
    if(internal_timing_flag)
        open2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
355 356 357 358 359 360 361 362

    /* 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)
        {
363 364
            fprintf(stderr, "darshan library warning: unable to open log file %s\n",
                logfile_name);
365 366 367
            unlink(logfile_name);
        }
        free(logfile_name);
368
        darshan_core_cleanup(final_core);
369 370 371
        return;
    }

372 373
    if(internal_timing_flag)
        job1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
374
    /* rank 0 is responsible for writing the compressed darshan job information */
Shane Snyder's avatar
Shane Snyder committed
375
    if(my_rank == 0)
376
    {
377 378 379
        void *pointers[2] = {&final_core->log_job, final_core->trailing_data};
        int lengths[2] = {sizeof(struct darshan_job), DARSHAN_EXE_LEN+1};
        int comp_buf_sz = 0;
380

381
        /* compress the job info and the trailing mount/exe data */
382
        all_ret = darshan_deflate_buffer(pointers, lengths, 2, 0,
383 384
            final_core->comp_buf, &comp_buf_sz);
        if(all_ret)
385
        {
386
            fprintf(stderr, "darshan library warning: unable to compress job data\n");
387
            unlink(logfile_name);
388
        }
389 390 391
        else
        {
            /* write the job information, preallocing space for the log header */
392 393 394
            gz_fp += sizeof(struct darshan_header) + 23; /* gzip headers/trailers ... */
            all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, gz_fp,
                final_core->comp_buf, comp_buf_sz, MPI_BYTE, &status);
395 396 397 398 399
            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);
Shane Snyder's avatar
Shane Snyder committed
400
                
401
            }
Shane Snyder's avatar
Shane Snyder committed
402

403 404
            /* set the beginning offset of record hash, which follows job info just written */
            gz_fp += comp_buf_sz;
405
        }
406 407
    }

408 409 410 411 412
    /* 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);
413
        darshan_core_cleanup(final_core);
414 415
        return;
    }
416 417
    if(internal_timing_flag)
        job2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
418

419 420
    if(internal_timing_flag)
        rec1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
421
    /* write the record name->id hash to the log file */
422 423
    ret = darshan_log_write_record_hash(log_fh, final_core, &gz_fp);
    tmp_off = final_core->log_header.rec_map.off + final_core->log_header.rec_map.len;
424

425
    /* error out if unable to write record hash */
426 427 428 429 430 431
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
432
            fprintf(stderr, "darshan library warning: unable to write record hash to log file %s\n",
433
                logfile_name);
434
            unlink(logfile_name);
435 436
        }
        free(logfile_name);
437
        darshan_core_cleanup(final_core);
438 439
        return;
    }
Shane Snyder's avatar
Shane Snyder committed
440 441
    if(internal_timing_flag)
        rec2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
442 443

    /* loop over globally used darshan modules and:
444
     *      - perform shared file reductions, if possible
445
     *      - get final output buffer
446
     *      - compress (zlib) provided output buffer
Shane Snyder's avatar
Shane Snyder committed
447
     *      - append compressed buffer to log file
448 449
     *      - add module index info (file offset/length) to log header
     *      - shutdown the module
450
     */
451
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
452
    {
453
        struct darshan_core_module* this_mod = final_core->mod_array[i];
454 455
        darshan_record_id mod_shared_recs[DARSHAN_CORE_MAX_RECORDS];
        struct darshan_core_record_ref *ref = NULL;
456
        void* mod_buf = NULL;
457
        int mod_buf_sz = 0;
458
        int j;
459

460
        if(global_mod_use_count[i] == 0)
461 462
        {
            if(my_rank == 0)
463 464 465 466
            {
                final_core->log_header.mod_map[i].off = 0;
                final_core->log_header.mod_map[i].len = 0;
            }
467
            continue;
468
        }
469 470 471 472
 
        if(internal_timing_flag)
            mod1[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();   
        /* if all processes used this module, prepare to do a shared file reduction */
Shane Snyder's avatar
Shane Snyder committed
473
        if(global_mod_use_count[i] == nprocs)
474
        {
475 476 477 478 479 480 481 482 483 484 485 486 487
            int shared_rec_count = 0;
            int rec_sz = 0;
            void *red_send_buf = NULL, *red_recv_buf = NULL;
            MPI_Datatype red_type;
            MPI_Op red_op;

            /* set the shared file list for this module */
            memset(mod_shared_recs, 0, DARSHAN_CORE_MAX_RECORDS * sizeof(darshan_record_id));
            for(j = 0; j < DARSHAN_CORE_MAX_RECORDS && shared_recs[j] != 0; j++)
            {
                HASH_FIND(hlink, final_core->rec_hash, &shared_recs[j],
                    sizeof(darshan_record_id), ref);
                assert(ref);
Shane Snyder's avatar
Shane Snyder committed
488
                if(DARSHAN_CORE_MOD_ISSET(ref->global_mod_flags, i))
489 490 491 492 493 494
                {
                    mod_shared_recs[shared_rec_count++] = shared_recs[j];
                }
            }

            /* if there are globally shared files, do a shared file reduction */
495 496
            if(shared_rec_count && this_mod->mod_funcs.prepare_for_reduction &&
               this_mod->mod_funcs.record_reduction_op)
497 498 499 500 501 502 503 504 505 506 507 508 509
            {
                this_mod->mod_funcs.prepare_for_reduction(mod_shared_recs, &shared_rec_count,
                    &red_send_buf, &red_recv_buf, &rec_sz);

                if(shared_rec_count)
                {
                    /* construct a datatype for a file record.  This is serving no purpose
                     * except to make sure we can do a reduction on proper boundaries
                     */
                    DARSHAN_MPI_CALL(PMPI_Type_contiguous)(rec_sz, MPI_BYTE, &red_type);
                    DARSHAN_MPI_CALL(PMPI_Type_commit)(&red_type);

                    /* register a reduction operator for this module */
510
                    DARSHAN_MPI_CALL(PMPI_Op_create)(this_mod->mod_funcs.record_reduction_op,
511 512 513 514 515 516 517 518 519 520
                        1, &red_op);

                    /* reduce shared file records for this module */
                    DARSHAN_MPI_CALL(PMPI_Reduce)(red_send_buf, red_recv_buf,
                        shared_rec_count, red_type, red_op, 0, MPI_COMM_WORLD);

                    DARSHAN_MPI_CALL(PMPI_Type_free)(&red_type);
                    DARSHAN_MPI_CALL(PMPI_Op_free)(&red_op);
                }
            }
521
        }
522

523
        /* if module is registered locally, get the corresponding output buffer */
524
        if(this_mod)
525 526
        {
            /* get output buffer from module */
527
            this_mod->mod_funcs.get_output_data(&mod_buf, &mod_buf_sz);
528 529
        }

530
        final_core->log_header.mod_map[i].off = tmp_off;
531

532 533 534 535
        /* append this module's data to the darshan log */
        ret = darshan_log_append_all(log_fh, final_core, mod_buf, mod_buf_sz,
            &gz_fp, &(final_core->log_header.mod_map[i].len));
        tmp_off += final_core->log_header.mod_map[i].len;
536

537
        /* error out if the log append failed */
538 539 540
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
            MPI_LOR, MPI_COMM_WORLD);
        if(all_ret != 0)
541
        {
542 543 544 545 546 547 548 549
            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);
550
            darshan_core_cleanup(final_core);
551
            return;
552 553 554
        }

        /* shutdown module if registered locally */
555
        if(this_mod)
556 557 558
        {
            this_mod->mod_funcs.shutdown();
        }
559 560
        if(internal_timing_flag)
            mod2[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
561 562
    }

563 564
    if(internal_timing_flag)
        header1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
565
    /* rank 0 is responsible for writing the log header */
566 567
    if(my_rank == 0)
    {
568 569 570
        void *header_buf = &(final_core->log_header);
        int header_buf_sz = sizeof(struct darshan_header);
        int comp_buf_sz = 0;
571

572 573 574 575 576 577 578 579 580 581 582 583 584 585
        /* initialize the remaining header fields */
        strcpy(final_core->log_header.version_string, DARSHAN_LOG_VERSION);
        final_core->log_header.magic_nr = DARSHAN_MAGIC_NR;

        /* deflate the header */
        /* NOTE: the header is not actually compressed because space for it must
         *       be preallocated before writing. i.e., the "compressed" header
         *       must be constant sized, sizeof(struct darshan_header) + 23.
         *       it is still necessary to deflate the header or the resulting log
         *       file will not be a valid gzip file.
         */
        all_ret = darshan_deflate_buffer((void **)&header_buf, &header_buf_sz, 1, 1,
            final_core->comp_buf, &comp_buf_sz);
        if(all_ret)
586
        {
587
            fprintf(stderr, "darshan library warning: unable to compress header\n");
588
            unlink(logfile_name);
589
        }
590 591 592 593 594 595 596 597 598 599 600
        else
        {
            all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, 0, final_core->comp_buf,
                comp_buf_sz, 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);
            }
        }
601 602
    }

603 604 605 606 607
    /* 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);
608
        darshan_core_cleanup(final_core);
609 610
        return;
    }
611 612
    if(internal_timing_flag)
        header2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
613

614 615 616 617 618 619
    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.gz, which indicates that this log file is
     * complete and ready for analysis
     */
620 621
    if(my_rank == 0)
    {
Shane Snyder's avatar
Shane Snyder committed
622
        if(getenv("DARSHAN_LOGFILE"))
623
        {
624
#ifdef __CP_GROUP_READABLE_LOGS
Shane Snyder's avatar
Shane Snyder committed
625
            chmod(logfile_name, (S_IRUSR|S_IRGRP));
626
#else
Shane Snyder's avatar
Shane Snyder committed
627
            chmod(logfile_name, (S_IRUSR));
628
#endif
Shane Snyder's avatar
Shane Snyder committed
629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
        }
        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.gz", (int)(end_log_time-start_log_time+1));
                rename(logfile_name, new_logfile_name);
                /* set permissions on log file */
#ifdef __CP_GROUP_READABLE_LOGS
                chmod(new_logfile_name, (S_IRUSR|S_IRGRP));
#else
                chmod(new_logfile_name, (S_IRUSR));
#endif
                free(new_logfile_name);
            }
653
        }
654
    }
655

656
    free(logfile_name);
657
    darshan_core_cleanup(final_core);
658

659
    if(internal_timing_flag)
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
        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];
        double all_tm, all_slowest;

        tm_end = DARSHAN_MPI_CALL(PMPI_Wtime)();

        open_tm = open2 - open1;
        header_tm = header2 - header1;
        job_tm = job2 - job1;
        rec_tm = rec2 - rec1;
        all_tm = tm_end - start_log_time;
        for(i = 0;i < DARSHAN_MAX_MODS; i++)
        {
            mod_tm[i] = mod2[i] - mod1[i];
        }

        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);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&all_tm, &all_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(mod_tm, mod_slowest, DARSHAN_MAX_MODS,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);

        if(my_rank == 0)
        {
            printf("#darshan:<op>\t<nprocs>\t<time>\n");
            printf("darshan:log_open\t%d\t%f\n", nprocs, open_slowest);
            printf("darshan:job_write\t%d\t%f\n", nprocs, job_slowest);
            printf("darshan:hash_write\t%d\t%f\n", nprocs, rec_slowest);
Shane Snyder's avatar
Shane Snyder committed
699
            printf("darshan:header_write\t%d\t%f\n", nprocs, header_slowest);
700 701 702 703
            for(i = 0; i < DARSHAN_MAX_MODS; i++)
            {
                if(global_mod_use_count[i])
                    printf("darshan:%s_shutdown\t%d\t%f\n", darshan_module_names[i],
Shane Snyder's avatar
Shane Snyder committed
704
                        nprocs, mod_slowest[i]);
705 706 707
            }
            printf("darshan:core_shutdown\t%d\t%f\n", nprocs, all_slowest);
        }
708 709 710 711
    }
    
    return;
}
712

Shane Snyder's avatar
Shane Snyder committed
713
/* *********************************** */
714

715
/* construct the darshan log file name */
716
static void darshan_get_logfile_name(char* logfile_name, int jobid, struct tm* start_tm)
717
{
Shane Snyder's avatar
Shane Snyder committed
718
    char* user_logfile_name;
719 720 721 722 723 724 725 726 727 728 729 730 731
    char* logpath;
    char* logname_string;
    char* logpath_override = NULL;
#ifdef __CP_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;

Shane Snyder's avatar
Shane Snyder committed
732 733 734 735 736 737 738 739 740 741 742 743 744 745 746
    /* 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
747
    {
Shane Snyder's avatar
Shane Snyder committed
748 749 750 751 752 753
        /* otherwise, generate the log path automatically */

        /* Use CP_LOG_PATH_OVERRIDE for the value or __CP_LOG_PATH */
        logpath = getenv(CP_LOG_PATH_OVERRIDE);
        if(!logpath)
        {
754
#ifdef __CP_LOG_PATH
Shane Snyder's avatar
Shane Snyder committed
755
            logpath = __CP_LOG_PATH;
756
#endif
Shane Snyder's avatar
Shane Snyder committed
757
        }
758

Shane Snyder's avatar
Shane Snyder committed
759 760 761 762 763 764 765 766 767 768
        /* 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.
         */
769 770

#ifndef DARSHAN_DISABLE_CUSERID
Shane Snyder's avatar
Shane Snyder committed
771
        cuserid(cuser);
772 773
#endif

Shane Snyder's avatar
Shane Snyder committed
774 775
        /* if cuserid() didn't work, then check the environment */
        if(strcmp(cuser, "") == 0)
776
        {
Shane Snyder's avatar
Shane Snyder committed
777 778 779 780 781
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }
782 783
        }

Shane Snyder's avatar
Shane Snyder committed
784 785 786 787 788 789
        /* 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);
        }
790

Shane Snyder's avatar
Shane Snyder committed
791 792 793 794
        /* 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);
795

Shane Snyder's avatar
Shane Snyder committed
796 797 798 799
        /* 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.
         */
800
#ifdef __CP_LOG_ENV
Shane Snyder's avatar
Shane Snyder committed
801 802
        /* just silently skip if the environment variable list is too big */
        if(strlen(__CP_LOG_ENV) < 256)
803
        {
Shane Snyder's avatar
Shane Snyder committed
804 805 806 807 808
            /* copy env variable list to a temporary buffer */
            strcpy(env_check, __CP_LOG_ENV);
            /* tokenize the comma-separated list */
            env_tok = strtok(env_check, ",");
            if(env_tok)
809
            {
Shane Snyder's avatar
Shane Snyder committed
810
                do
811
                {
Shane Snyder's avatar
Shane Snyder committed
812 813 814 815 816 817 818 819 820
                    /* 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, ",")));
            }
821 822 823
        }
#endif

Shane Snyder's avatar
Shane Snyder committed
824
        if(logpath_override)
825
        {
Shane Snyder's avatar
Shane Snyder committed
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840
            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);
            }
841
        }
Shane Snyder's avatar
Shane Snyder committed
842
        else if(logpath)
843
        {
Shane Snyder's avatar
Shane Snyder committed
844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863
            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';
864 865 866 867
        }
    }

    return;
868 869
}

870
/* record any hints used to write the darshan log in the log header */
871
static void darshan_log_record_hints_and_ver(struct darshan_core_runtime* core)
872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894
{
    char* hints;
    char* header_hints;
    int meta_remain = 0;
    char* m;

    /* check environment variable to see if the default MPI file hints have
     * been overridden
     */
    hints = getenv(CP_LOG_HINTS_OVERRIDE);
    if(!hints)
    {
        hints = __CP_LOG_HINTS;
    }

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

    header_hints = strdup(hints);
    if(!header_hints)
        return;

    meta_remain = DARSHAN_JOB_METADATA_LEN -
895
        strlen(core->log_job.metadata) - 1;
896 897
    if(meta_remain >= (strlen(PACKAGE_VERSION) + 9))
    {
898
        sprintf(core->log_job.metadata, "lib_ver=%s\n", PACKAGE_VERSION);
899 900 901 902
        meta_remain -= (strlen(PACKAGE_VERSION) + 9);
    }
    if(meta_remain >= (3 + strlen(header_hints)))
    {
903
        m = core->log_job.metadata + strlen(core->log_job.metadata);
904 905 906 907 908 909 910 911 912 913 914 915
        /* We have room to store the hints in the metadata portion of
         * the job header.  We just prepend an h= to the hints list.  The
         * metadata parser will ignore = characters that appear in the value
         * portion of the metadata key/value pair.
         */
        sprintf(m, "h=%s\n", header_hints);
    }
    free(header_hints);

    return;
}

916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
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 */
static void add_entry(char* trailing_data, int* space_left, struct mntent *entry)
{
    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;

    /* store mount information for use in header of darshan log */
956
    ret = snprintf(tmp_mnt, 256, "\n%s\t%s",
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001
        entry->mnt_type, entry->mnt_dir);
    if(ret < 256 && strlen(tmp_mnt) <= (*space_left))
    {
        strcat(trailing_data, tmp_mnt);
        (*space_left) -= strlen(tmp_mnt);
    }

    mnt_data_count++;
    return;
}

/* darshan_get_exe_and_mounts_root()
 *
 * collects command line and list of mounted file systems into a string that
 * will be stored with the job header
 */
static void darshan_get_exe_and_mounts_root(struct darshan_core_runtime *core,
    char* trailing_data, int space_left)
{
    FILE* tab;
    struct mntent *entry;
    char* exclude;
    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
    };

    /* length of exe has already been safety checked in darshan-posix.c */
    strcat(trailing_data, core->exe);
1002
    space_left = DARSHAN_EXE_LEN - strlen(trailing_data);
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013