darshan-logutils.c 50.8 KB
Newer Older
1 2 3 4 5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6
#define _GNU_SOURCE
7
#include "darshan-util-config.h"
8 9
#include <stdio.h>
#include <string.h>
10
#include <assert.h>
11
#include <stdlib.h>
12
#include <unistd.h>
13
#include <inttypes.h>
14 15 16 17
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

18 19 20 21 22
#include <zlib.h>
#ifdef HAVE_LIBBZ2
#include <bzlib.h>
#endif

23 24
#include "darshan-logutils.h"

25 26 27 28 29 30 31 32 33 34 35 36
struct darshan_fd_s
{
    gzFile gzf;
#ifdef HAVE_LIBBZ2
    BZFILE* bzf;
#endif
    int64_t pos;
    char mode[2];
    int swap_flag;
    char version[10];
    int job_struct_size;
    char* name;
Philip Carns's avatar
Philip Carns committed
37
    int COMPAT_CP_EXE_LEN;
38 39
};

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
/* isn't there a clever c way to avoid this? */
char *darshan_names[] = {
    "CP_INDEP_OPENS",
    "CP_COLL_OPENS",               /* count of MPI collective opens */
    "CP_INDEP_READS",              /* count of independent MPI reads */
    "CP_INDEP_WRITES",             /* count of independent MPI writes */
    "CP_COLL_READS",               /* count of collective MPI reads */
    "CP_COLL_WRITES",              /* count of collective MPI writes */
    "CP_SPLIT_READS",              /* count of split collective MPI reads */
    "CP_SPLIT_WRITES",             /* count of split collective MPI writes */
    "CP_NB_READS",                 /* count of nonblocking MPI reads */
    "CP_NB_WRITES",                /* count of nonblocking MPI writes */
    "CP_SYNCS",                    /* count of MPI_File_sync */
    "CP_POSIX_READS",              /* count of posix reads */
    "CP_POSIX_WRITES",             /* count of posix writes */
    "CP_POSIX_OPENS",              /* count of posix opens */
    "CP_POSIX_SEEKS",              /* count of posix seeks */
    "CP_POSIX_STATS",              /* count of posix stat/lstat/fstats */
    "CP_POSIX_MMAPS",              /* count of posix mmaps */
    "CP_POSIX_FREADS",
    "CP_POSIX_FWRITES",
    "CP_POSIX_FOPENS",
    "CP_POSIX_FSEEKS",
    "CP_POSIX_FSYNCS",
    "CP_POSIX_FDSYNCS",
65 66 67
    "CP_INDEP_NC_OPENS",
    "CP_COLL_NC_OPENS",
    "CP_HDF5_OPENS",
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
    "CP_COMBINER_NAMED",           /* count of each MPI datatype category */
    "CP_COMBINER_DUP",
    "CP_COMBINER_CONTIGUOUS",
    "CP_COMBINER_VECTOR",
    "CP_COMBINER_HVECTOR_INTEGER",
    "CP_COMBINER_HVECTOR",
    "CP_COMBINER_INDEXED",
    "CP_COMBINER_HINDEXED_INTEGER",
    "CP_COMBINER_HINDEXED",
    "CP_COMBINER_INDEXED_BLOCK",
    "CP_COMBINER_STRUCT_INTEGER",
    "CP_COMBINER_STRUCT",
    "CP_COMBINER_SUBARRAY",
    "CP_COMBINER_DARRAY",
    "CP_COMBINER_F90_REAL",
    "CP_COMBINER_F90_COMPLEX",
    "CP_COMBINER_F90_INTEGER",
    "CP_COMBINER_RESIZED",
    "CP_HINTS",                     /* count of MPI hints used */
    "CP_VIEWS",                     /* count of MPI set view calls */
    "CP_MODE",                      /* mode of file */
    "CP_BYTES_READ",                /* total bytes read */
    "CP_BYTES_WRITTEN",             /* total bytes written */
    "CP_MAX_BYTE_READ",             /* highest offset byte read */
    "CP_MAX_BYTE_WRITTEN",          /* highest offset byte written */
    "CP_CONSEC_READS",              /* count of consecutive reads */
    "CP_CONSEC_WRITES",             /* count of consecutive writes */
    "CP_SEQ_READS",                 /* count of sequential reads */
    "CP_SEQ_WRITES",                /* count of sequential writes */
    "CP_RW_SWITCHES",
    "CP_MEM_NOT_ALIGNED",           /* count of accesses not mem aligned */
    "CP_MEM_ALIGNMENT",             /* mem alignment in bytes */
    "CP_FILE_NOT_ALIGNED",          /* count of accesses not file aligned */
    "CP_FILE_ALIGNMENT",            /* file alignment in bytes */
102 103
    "CP_MAX_READ_TIME_SIZE",
    "CP_MAX_WRITE_TIME_SIZE",
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
    "CP_SIZE_READ_0_100",           /* count of posix read size ranges */
    "CP_SIZE_READ_100_1K",
    "CP_SIZE_READ_1K_10K",
    "CP_SIZE_READ_10K_100K",
    "CP_SIZE_READ_100K_1M",
    "CP_SIZE_READ_1M_4M",
    "CP_SIZE_READ_4M_10M",
    "CP_SIZE_READ_10M_100M",
    "CP_SIZE_READ_100M_1G",
    "CP_SIZE_READ_1G_PLUS",
    "CP_SIZE_WRITE_0_100",          /* count of posix write size ranges */
    "CP_SIZE_WRITE_100_1K",
    "CP_SIZE_WRITE_1K_10K",
    "CP_SIZE_WRITE_10K_100K",
    "CP_SIZE_WRITE_100K_1M",
    "CP_SIZE_WRITE_1M_4M",
    "CP_SIZE_WRITE_4M_10M",
    "CP_SIZE_WRITE_10M_100M",
    "CP_SIZE_WRITE_100M_1G",
    "CP_SIZE_WRITE_1G_PLUS",
    "CP_SIZE_READ_AGG_0_100",       /* count of MPI read size ranges */
    "CP_SIZE_READ_AGG_100_1K",
    "CP_SIZE_READ_AGG_1K_10K",
    "CP_SIZE_READ_AGG_10K_100K",
    "CP_SIZE_READ_AGG_100K_1M",
    "CP_SIZE_READ_AGG_1M_4M",
    "CP_SIZE_READ_AGG_4M_10M",
    "CP_SIZE_READ_AGG_10M_100M",
    "CP_SIZE_READ_AGG_100M_1G",
    "CP_SIZE_READ_AGG_1G_PLUS",
    "CP_SIZE_WRITE_AGG_0_100",      /* count of MPI write size ranges */
    "CP_SIZE_WRITE_AGG_100_1K",
    "CP_SIZE_WRITE_AGG_1K_10K",
    "CP_SIZE_WRITE_AGG_10K_100K",
    "CP_SIZE_WRITE_AGG_100K_1M",
    "CP_SIZE_WRITE_AGG_1M_4M",
    "CP_SIZE_WRITE_AGG_4M_10M",
    "CP_SIZE_WRITE_AGG_10M_100M",
    "CP_SIZE_WRITE_AGG_100M_1G",
    "CP_SIZE_WRITE_AGG_1G_PLUS",
    "CP_EXTENT_READ_0_100",          /* count of MPI read extent ranges */
    "CP_EXTENT_READ_100_1K",
    "CP_EXTENT_READ_1K_10K",
    "CP_EXTENT_READ_10K_100K",
    "CP_EXTENT_READ_100K_1M",
    "CP_EXTENT_READ_1M_4M",
    "CP_EXTENT_READ_4M_10M",
    "CP_EXTENT_READ_10M_100M",
    "CP_EXTENT_READ_100M_1G",
    "CP_EXTENT_READ_1G_PLUS",
    "CP_EXTENT_WRITE_0_100",         /* count of MPI write extent ranges */
    "CP_EXTENT_WRITE_100_1K",
    "CP_EXTENT_WRITE_1K_10K",
    "CP_EXTENT_WRITE_10K_100K",
    "CP_EXTENT_WRITE_100K_1M",
    "CP_EXTENT_WRITE_1M_4M",
    "CP_EXTENT_WRITE_4M_10M",
    "CP_EXTENT_WRITE_10M_100M",
    "CP_EXTENT_WRITE_100M_1G",
    "CP_EXTENT_WRITE_1G_PLUS",
    "CP_STRIDE1_STRIDE",             /* the four most frequently appearing strides */
    "CP_STRIDE2_STRIDE",
    "CP_STRIDE3_STRIDE",
    "CP_STRIDE4_STRIDE",
    "CP_STRIDE1_COUNT",              /* count of each of the most frequent strides */
    "CP_STRIDE2_COUNT",
    "CP_STRIDE3_COUNT",
    "CP_STRIDE4_COUNT",
    "CP_ACCESS1_ACCESS",
    "CP_ACCESS2_ACCESS",
    "CP_ACCESS3_ACCESS",
    "CP_ACCESS4_ACCESS",
    "CP_ACCESS1_COUNT",
    "CP_ACCESS2_COUNT",
    "CP_ACCESS3_COUNT",
    "CP_ACCESS4_COUNT",
180
    "CP_DEVICE",
181
    "CP_SIZE_AT_OPEN",
182 183 184 185
    "CP_FASTEST_RANK",
    "CP_FASTEST_RANK_BYTES",
    "CP_SLOWEST_RANK",
    "CP_SLOWEST_RANK_BYTES",
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203

    "CP_NUM_INDICES"
};

/* isn't there a clever c way to avoid this? */
char *darshan_f_names[] = {
    "CP_F_OPEN_TIMESTAMP",        /* timestamp of first open */
    "CP_F_READ_START_TIMESTAMP",  /* timestamp of first read */
    "CP_F_WRITE_START_TIMESTAMP", /* timestamp of first write */
    "CP_F_CLOSE_TIMESTAMP",       /* timestamp of last close */
    "CP_F_READ_END_TIMESTAMP",    /* timestamp of last read */
    "CP_F_WRITE_END_TIMESTAMP",   /* timestamp of last write */
    "CP_F_POSIX_READ_TIME",       /* cumulative posix read time */
    "CP_F_POSIX_WRITE_TIME",      /* cumulative posix write time */
    "CP_F_POSIX_META_TIME",       /* cumulative posix meta time */
    "CP_F_MPI_META_TIME",         /* cumulative mpi-io metadata time */
    "CP_F_MPI_READ_TIME",         /* cumulative mpi-io read time */
    "CP_F_MPI_WRITE_TIME",        /* cumulative mpi-io write time */
204 205
    "CP_F_MAX_READ_TIME",
    "CP_F_MAX_WRITE_TIME",
206 207 208 209 210 211
    "CP_F_FASTEST_RANK_TIME",
    "CP_F_SLOWEST_RANK_TIME",
    "CP_F_VARIANCE_RANK_TIME",
    "CP_F_VARIANCE_RANK_BYTES",

    "CP_F_NUM_INDICES"
212 213
};

214 215 216 217 218 219 220
/* function pointers so that we can switch functions depending on what file
 * version is detected
 */
int (*getjob_internal)(darshan_fd file, struct darshan_job *job);
int (*getfile_internal)(darshan_fd fd, 
    struct darshan_job *job, 
    struct darshan_file *file);
221
#define JOB_SIZE_124 28
222
#define JOB_SIZE_200 56
223
#define JOB_SIZE_201 120
Philip Carns's avatar
Philip Carns committed
224 225
#define CP_JOB_RECORD_SIZE_200 1024
#define CP_JOB_RECORD_SIZE_1x 1024
226

227
/* internal routines for parsing different file versions */
228
static int getjob_internal_204(darshan_fd file, struct darshan_job *job);
229
static int getjob_internal_201(darshan_fd file, struct darshan_job *job);
230
static int getjob_internal_200(darshan_fd file, struct darshan_job *job);
Philip Carns's avatar
Philip Carns committed
231 232
static int getfile_internal_204(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file);
233 234 235 236 237 238 239 240 241
static int getfile_internal_200(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file);
static int getjob_internal_124(darshan_fd file, struct darshan_job *job);
static int getfile_internal_124(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file);
static int getfile_internal_122(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file);
static int getfile_internal_121(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file);
242 243
static int getfile_internal_1x(darshan_fd fd, struct darshan_job *job, 
    struct darshan_file *file, int n_counters, int n_fcounters);
244
static void shift_missing_1_24(struct darshan_file* file);
245 246
static void shift_missing_1_22(struct darshan_file* file);
static void shift_missing_1_21(struct darshan_file* file);
247

248 249
static int darshan_log_seek(darshan_fd fd, int64_t offset);
static int darshan_log_read(darshan_fd fd, void* buf, int len);
250
static int darshan_log_write(darshan_fd fd, void* buf, int len);
251 252
static const char* darshan_log_error(darshan_fd fd, int* errnum);

253
/* a rather crude API for accessing raw binary darshan files */
254
darshan_fd darshan_log_open(const char *name, const char* mode)
255
{
256 257 258 259 260 261
    int test_fd;
    uint8_t magic[2];
    int ret;
    int try_bz2 = 1;
    int len = strlen(name);

262 263 264 265
    /* we only allows "w" or "r" modes, nothing fancy */
    assert(strlen(mode) == 1);
    assert(mode[0] == 'r' || mode[0] == 'w');

266 267 268 269 270
    darshan_fd tmp_fd = malloc(sizeof(*tmp_fd));
    if(!tmp_fd)
        return(NULL);
    memset(tmp_fd, 0, sizeof(*tmp_fd));

271 272 273 274 275 276 277 278 279
    tmp_fd->mode[0] = mode[0];
    tmp_fd->mode[1] = mode[1];
    tmp_fd->name  = strdup(name);
    if(!tmp_fd->name)
    {
        free(tmp_fd);
        return(NULL);
    }

280 281 282 283 284 285 286 287
    if(strcmp(mode, "r") == 0)
    {
        /* Try to detect if existing file is a bzip2 file or not.  Both 
         * libbz2 and libz will fall back to normal I/O (without compression) 
         * automatically, so we need to do some detection manually up front 
         * in order to get a chance to try both compression formats.
         */
        test_fd = open(name, O_RDONLY);
288
        if(test_fd < 0)
289 290
        {
            perror("open");
291
            free(tmp_fd->name);
292 293 294 295 296 297 298 299
            free(tmp_fd);
            return(NULL);
        }
        ret = read(test_fd, &magic, 2);
        if(ret != 2)
        {
            fprintf(stderr, "Error: failed to read any data from %s.\n", 
                name);
300
            free(tmp_fd->name);
301 302 303 304 305 306 307 308 309 310 311
            free(tmp_fd);
            close(test_fd);
            return(NULL);
        }
        /* header magic for bz2 */
        if(magic[0] != 0x42 && magic[1] != 0x5A)
        {
            try_bz2 = 0;
        }
        close(test_fd);
    }
312

313 314 315 316 317 318
    if(strcmp(mode, "w") == 0)
    {
        /* TODO: is this the behavior that we want? */
        /* if we are writing a new file, go by the file extension to tell
         * whether to use bz2 or not?
         */
319
        if(len >= 3 && name[len-3] == 'b' && name[len-2] == 'z' && name[len-1] == '2')
320 321 322 323 324
            try_bz2 = 1;
        else
            try_bz2 = 0;
    }

325
#ifdef HAVE_LIBBZ2
326 327 328
    if(try_bz2)
    {
        tmp_fd->bzf = BZ2_bzopen(name, mode);
329
        if(!tmp_fd->bzf)
330
        {
331 332 333
            free(tmp_fd->name);
            free(tmp_fd);
            return(NULL);
334
        }
335
        return(tmp_fd);
336
    }
337 338 339 340 341 342 343
#else
    if(try_bz2)
    {
        fprintf(stderr, "Error: this Darshan build does not support bz2 files.\n");
        fprintf(stderr, "Error: please install libbz2-dev and reconfigure.\n");
        return(NULL);
    }
344 345
#endif

346
    tmp_fd->gzf = gzopen(name, mode);
347 348
    if(!tmp_fd->gzf)
    {
349
        free(tmp_fd->name);
350 351 352 353
        free(tmp_fd);
        tmp_fd = NULL;
    }
    return tmp_fd;
354 355
}

356
/* darshan_log_getjob()
357 358 359
 *
 * returns 0 on success, -1 on failure
 */
360
int darshan_log_getjob(darshan_fd file, struct darshan_job *job)
361 362
{
    int ret;
363
    char buffer[DARSHAN_JOB_METADATA_LEN];
364

365 366 367
    ret = darshan_log_seek(file, 0);
    if(ret < 0)
        return(ret);
368

369 370 371
    /* read version number first so we know how to digest the rest of the
     * file
     */
372
    ret = darshan_log_read(file, file->version, 10);
373
    if(ret < 10)
374
    {
375
        fprintf(stderr, "Error: invalid log file (failed to read version).\n");
376 377 378
        return(-1);
    }

Philip Carns's avatar
Philip Carns committed
379 380
    if(strcmp(file->version, "2.05") == 0)
    {
381
        getjob_internal = getjob_internal_204;
Philip Carns's avatar
Philip Carns committed
382 383 384 385 386
        getfile_internal = getfile_internal_204;
        file->job_struct_size = sizeof(*job);
        file->COMPAT_CP_EXE_LEN = CP_EXE_LEN;
    }
    else if(strcmp(file->version, "2.04") == 0)
Philip Carns's avatar
Philip Carns committed
387
    {
388
        getjob_internal = getjob_internal_204;
Philip Carns's avatar
Philip Carns committed
389 390 391 392 393
        getfile_internal = getfile_internal_204;
        file->job_struct_size = sizeof(*job);
        file->COMPAT_CP_EXE_LEN = CP_EXE_LEN;
    }
    else if(strcmp(file->version, "2.03") == 0)
394 395 396
    {
        getjob_internal = getjob_internal_201;
        getfile_internal = getfile_internal_200;
397
        file->job_struct_size = JOB_SIZE_201;
Philip Carns's avatar
Philip Carns committed
398
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_200-file->job_struct_size-1;
399 400
    }
    else if(strcmp(file->version, "2.02") == 0)
Philip Carns's avatar
Philip Carns committed
401 402 403
    {
        getjob_internal = getjob_internal_201;
        getfile_internal = getfile_internal_200;
404
        file->job_struct_size = JOB_SIZE_201;
Philip Carns's avatar
Philip Carns committed
405
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_200-file->job_struct_size-1;
Philip Carns's avatar
Philip Carns committed
406 407
    }
    else if(strcmp(file->version, "2.01") == 0)
408
    {
409
        getjob_internal = getjob_internal_201;
410
        getfile_internal = getfile_internal_200;
411
        file->job_struct_size = JOB_SIZE_201;
Philip Carns's avatar
Philip Carns committed
412
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_200-file->job_struct_size-1;
413
    }
414 415 416 417 418
    else if(strcmp(file->version, "2.00") == 0)
    {
        getjob_internal = getjob_internal_200;
        getfile_internal = getfile_internal_200;
        file->job_struct_size = JOB_SIZE_200;
Philip Carns's avatar
Philip Carns committed
419
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_200-file->job_struct_size-1;
420
    }
421
    else if(strcmp(file->version, "1.24") == 0)
422
    {
423 424
        getjob_internal = getjob_internal_124;
        getfile_internal = getfile_internal_124;
425
        file->job_struct_size = JOB_SIZE_124;
Philip Carns's avatar
Philip Carns committed
426
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_1x-file->job_struct_size-1;
427
    }
428
    else if(strcmp(file->version, "1.23") == 0)
429
    {
430 431 432
        /* same as 1.24, except that mnt points may be incorrect */
        getjob_internal = getjob_internal_124;
        getfile_internal = getfile_internal_124;
433
        file->job_struct_size = JOB_SIZE_124;
Philip Carns's avatar
Philip Carns committed
434
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_1x-file->job_struct_size-1;
435
    }
436
    else if(strcmp(file->version, "1.22") == 0)
437
    {
438 439
        getjob_internal = getjob_internal_124;
        getfile_internal = getfile_internal_122;
440
        file->job_struct_size = JOB_SIZE_124;
Philip Carns's avatar
Philip Carns committed
441
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_1x-file->job_struct_size-1;
442
    }
443
    else if(strcmp(file->version, "1.21") == 0)
444
    {
445
        getjob_internal = getjob_internal_124;
Philip Carns's avatar
Philip Carns committed
446
        getfile_internal = getfile_internal_121;
447
        file->job_struct_size = JOB_SIZE_124;
Philip Carns's avatar
Philip Carns committed
448
        file->COMPAT_CP_EXE_LEN = CP_JOB_RECORD_SIZE_1x-file->job_struct_size-1;
449 450
    }
    else
451
    {
452 453 454
        fprintf(stderr, "Error: incompatible darshan file.\n");
        fprintf(stderr, "Error: expected version %s, but got %s\n", 
                CP_VERSION, file->version);
455
        return(-1);
456 457
    }

458
    ret = getjob_internal(file, job);
459 460 461

    if (ret == 0)
    {
462
#ifdef HAVE_STRNDUP
463
        char *metadata = strndup(job->metadata, sizeof(job->metadata));
464
#else
Philip Carns's avatar
bug fix  
Philip Carns committed
465
        char *metadata = strdup(job->metadata);
466
#endif
467 468
        char *kv;
        char *key;
469
        char *value;
470 471
        char *save;

472 473 474
        for(kv=strtok_r(metadata, "\n", &save);
            kv != NULL;
            kv=strtok_r(NULL, "\n", &save))
475
        {
476 477 478 479 480 481 482 483 484 485 486 487
            /* NOTE: we intentionally only split on the first = character.
             * There may be additional = characters in the value portion
             * (for example, when storing mpi-io hints).
             */
            strcpy(buffer, kv);
            key = buffer;
            value = index(buffer, '=');
            if(!value)
                continue;
            /* convert = to a null terminator to split key and value */
            value[0] = '\0';
            value++;
488
            if (strcmp(key, "prev_ver") == 0)
489
            {
490
                strncpy(job->version_string, value, sizeof(job->version_string));
491
            }
492 493
        }
        free(metadata);
494 495
    }

496 497
    return(ret);
}
498

499 500 501 502 503 504 505 506
/* darshan_putjob()
 * write job header in gzfile
 *
 * return 0 on success, -1 on failure.
 */
int darshan_log_putjob(darshan_fd file, struct darshan_job *job)
{
    struct darshan_job job_copy;
507
    char    pv_str[64];
508
    int     ret;
509
    int len;
510

511 512
    ret = darshan_log_seek(file, 0);
    if(ret < 0)
513 514
        return(ret);

515 516
    memset(&job_copy, 0, sizeof(job_copy));
    memcpy(&job_copy, job, sizeof(job_copy));
517 518 519 520 521 522 523 524 525 526 527
    /* check for newline in existing metadata, add if needed */
    len = strlen(job_copy.metadata);
    if(len > 0 && len < DARSHAN_JOB_METADATA_LEN)
    {
        if(job_copy.metadata[len-1] != '\n')
        {
            job_copy.metadata[len] = '\n';
            job_copy.metadata[len+1] = '\0';
        }
    }

528
    sprintf(pv_str, "prev_ver=%s\n", job->version_string);
529
    sprintf(job_copy.version_string, "%s", CP_VERSION);
Philip Carns's avatar
Philip Carns committed
530 531 532 533
    if(strlen(job_copy.metadata) + strlen(pv_str) < DARSHAN_JOB_METADATA_LEN)
        strncat(job_copy.metadata, pv_str, strlen(pv_str));
    else
        sprintf(job_copy.metadata, "%s", pv_str);
534 535
    job_copy.magic_nr = CP_MAGIC_NR;

536
    ret = darshan_log_write(file, &job_copy, sizeof(job_copy));
537 538 539 540 541 542 543 544 545
    if (ret != sizeof(job_copy))
    {
        fprintf(stderr, "Error: failed to write job header: %d\n", ret);
        return(-1);
    }

    return(0);
}

546 547 548 549 550 551 552
/* darshan_log_getfile()
 *
 * return 1 if file record found, 0 on eof, and -1 on error
 */
int darshan_log_getfile(darshan_fd fd, struct darshan_job *job, struct darshan_file *file)
{
    int ret;
553

554 555 556
    ret = getfile_internal(fd, job, file);

    return(ret);
557 558
}

559 560 561 562 563 564 565 566
/* darshan_log_putfile()
 *
 * return 0 if file record written, -1 on error.
 */
int darshan_log_putfile(darshan_fd fd, struct darshan_job *job, struct darshan_file *file)
{
    int     ret;

567 568 569 570 571 572
    if(fd->pos < CP_JOB_RECORD_SIZE)
    {
        ret = darshan_log_seek(fd, CP_JOB_RECORD_SIZE);
        if(ret < 0)
            return(ret);
    }
573

574
    ret = darshan_log_write(fd, file, sizeof(*file));
575 576 577 578 579 580 581 582 583
    if (ret != sizeof(*file))
    {
        fprintf(stderr, "Error: writing file record failed: %d\n", ret);
        return(-1);
    }

    return(0);
}

584 585 586 587 588 589
/* darshan_log_getmounts()
 * 
 * retrieves mount table information from the log.  Note that devs, mnt_pts,
 * and fs_types are arrays that will be allocated by the function and must
 * be freed by the caller.  count will indicate the size of the arrays
 */
590
int darshan_log_getmounts(darshan_fd fd, int64_t** devs, char*** mnt_pts, char***
591
    fs_types, int* count)
592 593 594 595
{
    int ret;
    char* pos;
    int array_index = 0;
Philip Carns's avatar
Philip Carns committed
596
    char buf[fd->COMPAT_CP_EXE_LEN+1];
597

598 599 600
    ret = darshan_log_seek(fd, fd->job_struct_size);
    if(ret < 0)
        return(ret);
601

Philip Carns's avatar
Philip Carns committed
602 603
    ret = darshan_log_read(fd, buf, (fd->COMPAT_CP_EXE_LEN + 1));
    if (ret < (fd->COMPAT_CP_EXE_LEN + 1))
604
    {
605
        perror("darshan_log_read");
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624
        return(-1);
    }

    /* count entries */
    *count = 0;
    pos = buf;
    while((pos = strchr(pos, '\n')) != NULL)
    {
        pos++;
        (*count)++;
    }

    if(*count == 0)
    {
        /* no mount entries present */
        return(0);
    }

    /* allocate output arrays */
625
    *devs = malloc((*count)*sizeof(int64_t));
626 627 628 629 630 631 632 633 634 635 636 637
    assert(*devs);
    *mnt_pts = malloc((*count)*sizeof(char*));
    assert(*mnt_pts);
    *fs_types = malloc((*count)*sizeof(char*));
    assert(*fs_types);
    
    /* work backwards through the table and parse each line (except for
     * first, which holds command line information)
     */
    while((pos = strrchr(buf, '\n')) != NULL)
    {
        /* overestimate string lengths */
Philip Carns's avatar
Philip Carns committed
638
        (*mnt_pts)[array_index] = malloc(fd->COMPAT_CP_EXE_LEN);
639
        assert((*mnt_pts)[array_index]);
Philip Carns's avatar
Philip Carns committed
640
        (*fs_types)[array_index] = malloc(fd->COMPAT_CP_EXE_LEN);
641 642
        assert((*fs_types)[array_index]);
        
Philip Carns's avatar
Philip Carns committed
643
        ret = sscanf(++pos, "%" PRId64 "\t%s\t%s", &(*devs)[array_index],
644 645
            (*fs_types)[array_index], (*mnt_pts)[array_index]);

646 647 648 649 650 651 652 653 654
        if(ret != 3)
        {
            fprintf(stderr, "Error: poorly formatted mount table in log file.\n");
            return(-1);
        }
        pos--;
        *pos = '\0';
        array_index++;
    }
655

656 657 658
    return (0);
}

659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
/* darshan_log_putmounts
 *
 * encode mount information back into mtab format.
 *
 * returns 0 on success, -1 on failure.
 */
int darshan_log_putmounts(darshan_fd fd, int64_t* devs, char** mnt_pts, char** fs_types, int count)
{
    int     ret;
    char    line[1024];
    int     i;

    for(i=count-1; i>=0; i--)
    {
        sprintf(line, "\n%" PRId64 "\t%s\t%s",
                devs[i], fs_types[i], mnt_pts[i]);
675
        ret = darshan_log_write(fd, line, strlen(line));
676 677 678 679 680 681
        if (ret != strlen(line))
        {
            fprintf(stderr, "Error: failed to write mount entry: %d\n", ret);
            return(-1);
        }
    }
682 683 684 685 686 687 688 689

    /* seek ahead to end of exe region, will be zero filled */
    ret = darshan_log_seek(fd, CP_JOB_RECORD_SIZE);
    if (ret)
    {
        fprintf(stderr, "Error: forward seek failed: %d\n", CP_JOB_RECORD_SIZE);
    }

690 691
    return(0);
}
692

693
int darshan_log_getexe(darshan_fd fd, char *buf)
694 695
{
    int ret;
696
    char* newline;
697

698 699 700
    ret = darshan_log_seek(fd, fd->job_struct_size);
    if(ret < 0)
        return(ret);
701

Philip Carns's avatar
Philip Carns committed
702 703
    ret = darshan_log_read(fd, buf, (fd->COMPAT_CP_EXE_LEN + 1));
    if (ret < (fd->COMPAT_CP_EXE_LEN + 1))
704
    {
705
        perror("darshan_log_read");
706 707
        return(-1);
    }
708

709 710 711 712 713 714 715 716
    /* this call is only supposed to return the exe string, but starting in
     * log format 1.23 there could be a table of mount entry information
     * after the exe.  Look for newline character and truncate there.
     */
    newline = strchr(buf, '\n');
    if(newline)
        *newline = '\0';

717 718 719
    return (0);
}

720 721 722 723 724 725 726 727 728 729 730
/* darshan_log_putexe()
 *
 * Write the exe string to the log.
 *
 * return 0 on success, -1 on failure.
 */
int darshan_log_putexe(darshan_fd fd, char *buf)
{
    int     ret;
    int     len;

731 732
    ret = darshan_log_seek(fd, sizeof(struct darshan_job));
    if(ret < 0)
733 734 735 736
        return(ret);

    len = strlen(buf);

737
    ret = darshan_log_write(fd, buf, len);
738 739 740 741 742 743 744 745 746
    if (ret != len)
    {
        fprintf(stderr, "Error: failed to write exe info: %d\n", ret);
        ret = -1;
    }

    return(ret);
}

747
void darshan_log_close(darshan_fd file)
748
{
749 750 751 752 753 754 755 756
#ifdef HAVE_LIBBZ2
    if(file->bzf)
        BZ2_bzclose(file->bzf);
#endif

    if(file->gzf)
        gzclose(file->gzf);

757
    free(file->name);
758
    free(file);
759 760
}

761 762 763 764 765 766
/* darshan_log_print_version_warnings()
 *
 * Print summary of any problems with the detected log format
 */
void darshan_log_print_version_warnings(struct darshan_job *job)
{
Philip Carns's avatar
Philip Carns committed
767
    if(strcmp(job->version_string, "2.05") == 0)
768
    {
769
        /* current version */
770 771 772
        return;
    }

Philip Carns's avatar
Philip Carns committed
773 774 775 776 777 778 779
    if(strcmp(job->version_string, "2.04") == 0)
    {
        printf("# WARNING: version 2.04 log format has the following limitations:\n");
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
        return;
    }

Philip Carns's avatar
Philip Carns committed
780 781
    if(strcmp(job->version_string, "2.03") == 0)
    {
Philip Carns's avatar
Philip Carns committed
782
        /* no meaningful change to interpretation of log file, 2.03 just
Philip Carns's avatar
Philip Carns committed
783 784
         * increased the header space available for annotations.
         */
Philip Carns's avatar
Philip Carns committed
785 786
        printf("# WARNING: version 2.03 log format has the following limitations:\n");
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
Philip Carns's avatar
Philip Carns committed
787 788 789
        return;
    }

790 791 792 793
    if(strcmp(job->version_string, "2.02") == 0)
    {
        printf("# WARNING: version 2.01 log format has the following limitations:\n");
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
794
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
795 796 797
        return;
    }

Philip Carns's avatar
Philip Carns committed
798 799 800
    if(strcmp(job->version_string, "2.01") == 0)
    {
        printf("# WARNING: version 2.01 log format has the following limitations:\n");
801
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
802
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
803
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
Philip Carns's avatar
Philip Carns committed
804 805 806
        return;
    }

807 808
    if(strcmp(job->version_string, "2.00") == 0)
    {
Philip Carns's avatar
Philip Carns committed
809
        printf("# WARNING: version 2.00 log format has the following limitations:\n");
810
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
811
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
812
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
813 814
        return;
    }
815
 
816
    if(strcmp(job->version_string, "1.24") == 0)
817
    {
818 819 820 821 822 823 824 825 826
        printf("# WARNING: version 1.24 log format does not support the following parameters:\n");
        printf("#   CP_FASTEST_RANK\n");
        printf("#   CP_FASTEST_RANK_BYTES\n");
        printf("#   CP_SLOWEST_RANK\n");
        printf("#   CP_SLOWEST_RANK_BYTES\n");
        printf("#   CP_F_FASTEST_RANK_TIME\n");
        printf("#   CP_F_SLOWEST_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_BYTES\n");
Philip Carns's avatar
Philip Carns committed
827
        printf("# WARNING: version 1.24 log format has the following limitations:\n");
828
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
829 830
        printf("# - does not store the job id in the file.\n");
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
831
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
832 833
        return;
    }
834
    
835 836
    if(strcmp(job->version_string, "1.23") == 0)
    {
837 838 839 840 841 842 843 844 845
        printf("# WARNING: version 1.23 log format does not support the following parameters:\n");
        printf("#   CP_FASTEST_RANK\n");
        printf("#   CP_FASTEST_RANK_BYTES\n");
        printf("#   CP_SLOWEST_RANK\n");
        printf("#   CP_SLOWEST_RANK_BYTES\n");
        printf("#   CP_F_FASTEST_RANK_TIME\n");
        printf("#   CP_F_SLOWEST_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_BYTES\n");
Philip Carns's avatar
Philip Carns committed
846
        printf("# WARNING: version 1.23 log format has the following limitations:\n");
847
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
848 849 850
        printf("# - may have incorrect mount point mappings for files with rank > 0.\n");
        printf("# - does not store the job id in the file.\n");
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
851
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
852
        return;
853 854
    }

855 856 857 858
    if(strcmp(job->version_string, "1.22") == 0)
    {
        printf("# WARNING: version 1.22 log format does not support the following parameters:\n");
        printf("#   CP_DEVICE\n");
859
        printf("#   CP_SIZE_AT_OPEN\n");
860 861 862 863 864 865 866 867
        printf("#   CP_FASTEST_RANK\n");
        printf("#   CP_FASTEST_RANK_BYTES\n");
        printf("#   CP_SLOWEST_RANK\n");
        printf("#   CP_SLOWEST_RANK_BYTES\n");
        printf("#   CP_F_FASTEST_RANK_TIME\n");
        printf("#   CP_F_SLOWEST_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_BYTES\n");
Philip Carns's avatar
Philip Carns committed
868
        printf("# WARNING: version 1.22 log format has the following limitations:\n");
869
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
870 871 872 873
        printf("# - does not record mounted file systems, mount points, or fs types.\n");
        printf("# - attributes syncs to cumulative metadata time, rather than cumulative write time.\n");
        printf("# - does not store the job id in the file.\n");
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
874
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
875 876
        return;
    }
877 878 879 880 881 882 883 884 885

    if(strcmp(job->version_string, "1.21") == 0)
    {
        printf("# WARNING: version 1.21 log format does not support the following parameters:\n");
        printf("#   CP_INDEP_NC_OPENS\n");
        printf("#   CP_COLL_NC_OPENS\n");
        printf("#   CP_HDF5_OPENS\n");
        printf("#   CP_MAX_READ_TIME_SIZE\n");
        printf("#   CP_MAX_WRITE_TIME_SIZE\n");
886
        printf("#   CP_DEVICE\n");
887
        printf("#   CP_SIZE_AT_OPEN\n");
888 889
        printf("#   CP_F_MAX_READ_TIME\n");
        printf("#   CP_F_MAX_WRITE_TIME\n");
890 891 892 893 894 895 896 897
        printf("#   CP_FASTEST_RANK\n");
        printf("#   CP_FASTEST_RANK_BYTES\n");
        printf("#   CP_SLOWEST_RANK\n");
        printf("#   CP_SLOWEST_RANK_BYTES\n");
        printf("#   CP_F_FASTEST_RANK_TIME\n");
        printf("#   CP_F_SLOWEST_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_TIME\n");
        printf("#   CP_F_VARIANCE_RANK_BYTES\n");
Philip Carns's avatar
Philip Carns committed
898
        printf("# WARNING: version 1.21 log format has the following limitations:\n");
899
        printf("# - *_TIMESTAMP fields are not normalized relative to MPI_Init() time.\n");
Philip Carns's avatar
Philip Carns committed
900 901 902 903
        printf("# - does not record mounted file systems, mount points, or fs types.\n");
        printf("# - attributes syncs to cumulative metadata time, rather than cumulative write time.\n");
        printf("# - does not store the job id in the file.\n");
        printf("# - inaccurate statistics in some multi-threaded cases.\n");
Philip Carns's avatar
Philip Carns committed
904
        printf("# - CP_F_SLOWEST_RANK_TIME and CP_F_FASTEST_RANK_TIME only report elapsed time at the POSIX level.\n");
905 906 907 908 909 910 911 912
        return;
    }

    fprintf(stderr, "Error: version %s not supported by parser.\n",
        job->version_string);
    return;
}

913 914 915 916
/* shift_missing_1_21()
 *
 * translates indices to account for counters that weren't present in log
 * format 1.21
917
 */
918
/*******************************
Philip Carns's avatar
Philip Carns committed
919
 * version 1.21 to 2.00 differences 
920 921 922 923 924 925 926 927
 * - added:
 *   - CP_INDEP_NC_OPENS
 *   - CP_COLL_NC_OPENS
 *   - CP_HDF5_OPENS
 *   - CP_MAX_READ_TIME_SIZE
 *   - CP_MAX_WRITE_TIME_SIZE
 *   - CP_DEVICE
 *   - CP_SIZE_AT_OPEN
928 929 930 931
 *   - CP_FASTEST_RANK
 *   - CP_FASTEST_RANK_BYTES
 *   - CP_SLOWEST_RANK
 *   - CP_SLOWEST_RANK_BYTES
932 933
 *   - CP_F_MAX_READ_TIME
 *   - CP_F_MAX_WRITE_TIME
934 935 936 937
 *   - CP_F_FASTEST_RANK_TIME
 *   - CP_F_SLOWEST_RANK_TIME
 *   - CP_F_VARIANCE_RANK_TIME
 *   - CP_F_VARIANCE_RANK_BYTES
938
 * - changed params:
939 940 941
 *   - CP_FILE_RECORD_SIZE: 1184 to 1328
 *   - CP_NUM_INDICES: 133 to 144
 *   - CP_F_NUM_INDICES: 12 to 18
942
 */
943 944 945 946 947 948 949 950 951
static void shift_missing_1_21(struct darshan_file* file)
{
    int c_index = 0;
    int missing_counters[] = {
        CP_INDEP_NC_OPENS,
        CP_COLL_NC_OPENS,
        CP_HDF5_OPENS,
        CP_MAX_READ_TIME_SIZE,
        CP_MAX_WRITE_TIME_SIZE,
952
        CP_DEVICE,
953
        CP_SIZE_AT_OPEN,
954 955 956 957
        CP_FASTEST_RANK,
        CP_FASTEST_RANK_BYTES,
        CP_SLOWEST_RANK,
        CP_SLOWEST_RANK_BYTES,
958 959 960 961
        -1};
    int missing_f_counters[] = {
        CP_F_MAX_READ_TIME,
        CP_F_MAX_WRITE_TIME,
962 963 964 965
        CP_F_FASTEST_RANK_TIME,
        CP_F_SLOWEST_RANK_TIME,
        CP_F_VARIANCE_RANK_TIME,
        CP_F_VARIANCE_RANK_BYTES,
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
        -1};

    c_index = 0;
    while(missing_counters[c_index] != -1)
    {
        int missing_counter = missing_counters[c_index];
        c_index++;
        if(missing_counter < (CP_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->counters[missing_counter+1],
                &file->counters[missing_counter],
                (CP_NUM_INDICES-missing_counter-1)*sizeof(int64_t));
        }
        /* zero out missing counter */
        file->counters[missing_counter] = 0;
    }

    c_index = 0;
    while(missing_f_counters[c_index] != -1)
    {
        int missing_counter = missing_f_counters[c_index];
        c_index++;
        if(missing_counter < (CP_F_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->fcounters[missing_counter+1],
                &file->fcounters[missing_counter],
                (CP_F_NUM_INDICES-missing_counter-1)*sizeof(double));
        }
        /* zero out missing counter */
        file->fcounters[missing_counter] = 0;
    }

    return;
}
1002 1003 1004 1005 1006 1007

/* shift_missing_1_22()
 *
 * translates indices to account for counters that weren't present in log
 * format 1.22
 */
1008
/*******************************
Philip Carns's avatar
Philip Carns committed
1009
 * version 1.22 to 2.00 differences
1010 1011 1012 1013
 *
 * - added:
 *   - CP_DEVICE
 *   - CP_SIZE_AT_OPEN
1014 1015 1016 1017 1018 1019 1020 1021
 *   - CP_FASTEST_RANK
 *   - CP_FASTEST_RANK_BYTES
 *   - CP_SLOWEST_RANK
 *   - CP_SLOWEST_RANK_BYTES
 *   - CP_F_FASTEST_RANK_TIME
 *   - CP_F_SLOWEST_RANK_TIME
 *   - CP_F_VARIANCE_RANK_TIME
 *   - CP_F_VARIANCE_RANK_BYTES
1022
 * - changed params:
1023 1024 1025
 *   - CP_FILE_RECORD_SIZE: 1240 to 1328
 *   - CP_NUM_INDICES: 138 to 144
 *   - CP_F_NUM_INDICES: 14 to 18
1026
 */
1027 1028 1029 1030 1031
static void shift_missing_1_22(struct darshan_file* file)
{
    int c_index = 0;
    int missing_counters[] = {
        CP_DEVICE,
1032
        CP_SIZE_AT_OPEN,
1033 1034 1035 1036 1037 1038 1039 1040 1041 1042
        CP_FASTEST_RANK,
        CP_FASTEST_RANK_BYTES,
        CP_SLOWEST_RANK,
        CP_SLOWEST_RANK_BYTES,
        -1};
    int missing_f_counters[] = {
        CP_F_FASTEST_RANK_TIME,
        CP_F_SLOWEST_RANK_TIME,
        CP_F_VARIANCE_RANK_TIME,
        CP_F_VARIANCE_RANK_BYTES,
1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
        -1};

    c_index = 0;
    while(missing_counters[c_index] != -1)
    {
        int missing_counter = missing_counters[c_index];
        c_index++;
        if(missing_counter < (CP_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->counters[missing_counter+1],
                &file->counters[missing_counter],
                (CP_NUM_INDICES-missing_counter-1)*sizeof(int64_t));
        }
        /* zero out missing counter */
        file->counters[missing_counter] = 0;
    }

1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076
    c_index = 0;
    while(missing_f_counters[c_index] != -1)
    {
        int missing_counter = missing_f_counters[c_index];
        c_index++;
        if(missing_counter < (CP_F_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->fcounters[missing_counter+1],
                &file->fcounters[missing_counter],
                (CP_F_NUM_INDICES-missing_counter-1)*sizeof(double));
        }
        /* zero out missing counter */
        file->fcounters[missing_counter] = 0;
    }

1077 1078
    return;
}