darshan-logutils.c 45.7 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 _GNU_SOURCE
8
#include "darshan-util-config.h"
9 10
#include <stdio.h>
#include <string.h>
11
#include <assert.h>
12
#include <stdlib.h>
13
#include <unistd.h>
14
#include <inttypes.h>
15 16 17
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
18
#include <errno.h>
19

20 21
#include "darshan-logutils.h"

22 23 24
/* default input buffer size for decompression algorithm */
#define DARSHAN_DEF_COMP_BUF_SZ (1024*1024) /* 1 MiB */

25
/* special identifers for referring to header, job, and
26 27 28 29 30 31
 * record map regions of the darshan log file
 */
#define DARSHAN_HEADER_REGION_ID    (-3)
#define DARSHAN_JOB_REGION_ID       (-2)
#define DARSHAN_REC_MAP_REGION_ID   (-1)

32 33 34 35 36 37 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 65 66 67 68
struct darshan_dz_state
{
    /* (libz/bzip2) stream data structure for managing
     * compression and decompression state */
    void *strm;
    /* buffer for staging compressed data to/from log file */
    unsigned char *buf;
    /* size of staging buffer */
    int size;
    /* for reading logs, flag indicating end of log file region */
    int eor;
    /* the region we last tried reading/writing */
    int prev_reg_id;
};

/* internal fd data structure */
struct darshan_fd_int_state
{
    /* posix file descriptor for the log file */
    int fildes;
    /* file pointer position */
    int64_t pos;
    /* flag indicating whether log file was created (and written) */
    int creat_flag;
    /* compression type used on log file (libz or bzip2) */
    enum darshan_comp_type comp_type;
    /* log file path name */
    char logfile_path[PATH_MAX];
    /* pointer to exe & mount data in darshan job data structure */
    char *exe_mnt_data;
    /* whether previous file operations have failed */
    int err;

    /* compression/decompression state */
    struct darshan_dz_state dz;
};

69 70
static int darshan_log_getheader(darshan_fd fd);
static int darshan_log_putheader(darshan_fd fd);
71 72
static int darshan_log_seek(darshan_fd fd, off_t offset);
static int darshan_log_read(darshan_fd fd, void *buf, int len);
73
static int darshan_log_write(darshan_fd fd, void *buf, int len);
74 75
static int darshan_log_dzinit(struct darshan_fd_int_state *state);
static void darshan_log_dzdestroy(struct darshan_fd_int_state *state);
76 77 78 79 80
static int darshan_log_dzread(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_dzwrite(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_libz_read(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_libz_write(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_libz_flush(darshan_fd fd, int region_id);
81 82 83 84 85
#ifdef HAVE_LIBBZ2
static int darshan_log_bzip2_read(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_bzip2_write(darshan_fd fd, int region_id, void *buf, int len);
static int darshan_log_bzip2_flush(darshan_fd fd, int region_id);
#endif
86 87 88 89
static int darshan_log_dzload(darshan_fd fd, struct darshan_log_map map);
static int darshan_log_dzunload(darshan_fd fd, struct darshan_log_map *map_p);

/* TODO: check comments on functions to make sure they are right /cleanup */
90 91

/* TODO: can we make this s.t. we don't care about ordering (i.e., X macro it ) */
92
/* see gzip interface for ideas */
93 94
struct darshan_mod_logutil_funcs *mod_logutils[DARSHAN_MAX_MODS] =
{
95
    NULL,               /* NULL */
96
    &posix_logutils,    /* POSIX */
97 98 99
    &mpiio_logutils,    /* MPI-IO */
    &hdf5_logutils,     /* HDF5 */
    &pnetcdf_logutils,  /* PNETCDF */
100
    &bgq_logutils,      /* BG/Q */
101 102 103 104 105 106 107 108 109 110 111
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL
};
112 113 114

/* darshan_log_open()
 *
115
 * open an existing darshan log file for reading only
116
 *
117
 * returns file descriptor on success, NULL on failure
118
 */
119
darshan_fd darshan_log_open(const char *name)
120
{
121
    darshan_fd tmp_fd;
122
    int ret;
123

124
    /* allocate a darshan file descriptor */
125 126 127 128
    tmp_fd = malloc(sizeof(*tmp_fd));
    if(!tmp_fd)
        return(NULL);
    memset(tmp_fd, 0, sizeof(*tmp_fd));
129 130 131 132 133 134 135
    tmp_fd->state = malloc(sizeof(struct darshan_fd_int_state));
    if(!tmp_fd->state)
    {
        free(tmp_fd->state);
        return(NULL);
    }
    memset(tmp_fd->state, 0, sizeof(struct darshan_fd_int_state));
136

137
    /* open the log file in read mode */
138 139
    tmp_fd->state->fildes = open(name, O_RDONLY);
    if(tmp_fd->state->fildes < 0)
140
    {
141
        fprintf(stderr, "Error: failed to open darshan log file %s.\n", name);
142
        free(tmp_fd->state);
143 144 145
        free(tmp_fd);
        return(NULL);
    }
146
    strncpy(tmp_fd->state->logfile_path, name, PATH_MAX);
147 148 149 150 151

    /* read the header from the log file to init fd data structures */
    ret = darshan_log_getheader(tmp_fd);
    if(ret < 0)
    {
152 153
        close(tmp_fd->state->fildes);
        free(tmp_fd->state);
154
        free(tmp_fd);
155 156 157 158
        return(NULL);
    }

    /* initialize compression data structures */
159
    ret = darshan_log_dzinit(tmp_fd->state);
160 161 162
    if(ret < 0)
    {
        fprintf(stderr, "Error: failed to initialize decompression data structures.\n");
163 164
        close(tmp_fd->state->fildes);
        free(tmp_fd->state);
165 166
        free(tmp_fd);
        return(NULL);
167 168
    }

169 170 171 172 173 174 175
    return(tmp_fd);
}

/* darshan_log_create()
 *
 * create a darshan log file for writing with the given compression method
 *
176
 * returns file descriptor on success, NULL on failure
177 178 179 180
 */
darshan_fd darshan_log_create(const char *name, enum darshan_comp_type comp_type)
{
    darshan_fd tmp_fd;
181
    int ret;
182

183
    /* allocate a darshan file descriptor */
184 185 186 187
    tmp_fd = malloc(sizeof(*tmp_fd));
    if(!tmp_fd)
        return(NULL);
    memset(tmp_fd, 0, sizeof(*tmp_fd));
188 189 190 191 192 193 194
    tmp_fd->state = malloc(sizeof(struct darshan_fd_int_state));
    if(!tmp_fd->state)
    {
        free(tmp_fd);
        return(NULL);
    }
    memset(tmp_fd->state, 0, sizeof(struct darshan_fd_int_state));
195

196 197 198
    /* create the log for writing, making sure to not overwrite existing log */
    tmp_fd->state->fildes = creat(name, 0400);
    if(tmp_fd->state->fildes < 0)
199
    {
200
        fprintf(stderr, "Error: failed to open darshan log file %s.\n", name);
201
        free(tmp_fd->state);
202
        free(tmp_fd);
203
        return(NULL);
204
    }
205 206 207
    tmp_fd->state->creat_flag = 1;
    tmp_fd->state->comp_type = comp_type;
    strncpy(tmp_fd->state->logfile_path, name, PATH_MAX);
208

209 210 211 212 213
    /* position file pointer to prealloc space for the log file header
     * NOTE: the header is written at close time, after all internal data
     * structures have been properly set
     */
    ret = darshan_log_seek(tmp_fd, sizeof(struct darshan_header));
214
    if(ret < 0)
215 216
    {
        fprintf(stderr, "Error: unable to seek in darshan log file.\n");
217 218
        close(tmp_fd->state->fildes);
        free(tmp_fd->state);
219 220 221
        free(tmp_fd);
        unlink(name);
        return(NULL);
222 223
    }

224
    /* initialize compression data structures */
225
    ret = darshan_log_dzinit(tmp_fd->state);
226 227
    if(ret < 0)
    {
228
        fprintf(stderr, "Error: failed to initialize compression data structures.\n");
229 230
        close(tmp_fd->state->fildes);
        free(tmp_fd->state);
231 232 233
        free(tmp_fd);
        unlink(name);
        return(NULL);
234 235
    }

236
    return(tmp_fd);
237 238
}

239
/* darshan_log_getjob()
240 241
 *
 * read job level metadata from the darshan log file
242
 *
243
 * returns 0 on success, -1 on failure
244
 */
245
int darshan_log_getjob(darshan_fd fd, struct darshan_job *job)
246
{
247
    struct darshan_fd_int_state *state = fd->state;
248 249
    char job_buf[DARSHAN_JOB_RECORD_SIZE] = {0};
    int job_buf_sz = DARSHAN_JOB_RECORD_SIZE;
250
    int ret;
251

252
    assert(state);
253 254
    assert(fd->job_map.len > 0 && fd->job_map.off > 0);

255
    /* read the compressed job data from the log file */
256 257
    ret = darshan_log_dzread(fd, DARSHAN_JOB_REGION_ID, job_buf, job_buf_sz);
    if(ret <= sizeof(*job))
258
    {
259
        fprintf(stderr, "Error: failed to read darshan log file job data.\n");
260 261
        return(-1);
    }
262 263

    memcpy(job, job_buf, sizeof(*job));
264

265
    if(fd->swap_flag)
266
    {
267 268 269 270 271 272
        /* swap bytes if necessary */
        DARSHAN_BSWAP64(&job->uid);
        DARSHAN_BSWAP64(&job->start_time);
        DARSHAN_BSWAP64(&job->end_time);
        DARSHAN_BSWAP64(&job->nprocs);
        DARSHAN_BSWAP64(&job->jobid);
273 274
    }

275
    /* save trailing exe & mount information, so it can be retrieved later */
276 277 278
    if(!(state->exe_mnt_data))
        state->exe_mnt_data = malloc(DARSHAN_EXE_LEN+1);
    if(!(state->exe_mnt_data))
279
        return(-1);
280
    memcpy(state->exe_mnt_data, &job_buf[sizeof(*job)], DARSHAN_EXE_LEN+1);
281

282 283 284
    return(0);
}

285 286
/* darshan_log_putjob()
 *
287
 * write job level metadata to darshan log file
288 289 290 291 292
 *
 * returns 0 on success, -1 on failure
 */
int darshan_log_putjob(darshan_fd fd, struct darshan_job *job)
{
293
    struct darshan_fd_int_state *state = fd->state;
294 295 296 297
    struct darshan_job job_copy;
    int len;
    int ret;

298 299
    assert(state);

300 301 302 303 304 305 306 307 308 309 310 311 312 313
    memset(&job_copy, 0, sizeof(*job));
    memcpy(&job_copy, job, sizeof(*job));

    /* 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';
        }
    }

314 315 316
    /* write the compressed job data to log file */
    ret = darshan_log_dzwrite(fd, DARSHAN_JOB_REGION_ID, &job_copy, sizeof(*job));
    if(ret != sizeof(*job))
317
    {
318
        state->err = -1;
319 320 321 322 323 324 325 326 327 328 329 330 331
        fprintf(stderr, "Error: failed to write darshan log file job data.\n");
        return(-1);
    }

    return(0);
}

/* darshan_log_getexe()
 *
 * reads the application exe name from darshan log file
 * 
 * returns 0 on success, -1 on failure 
 */
332 333
int darshan_log_getexe(darshan_fd fd, char *buf)
{
334
    struct darshan_fd_int_state *state = fd->state;
335
    char *newline;
336
    int ret;
337

338 339
    assert(state);

340
    /* if the exe/mount data has not been saved yet, read in the job info */
341
    if(!(state->exe_mnt_data))
342
    {
343 344
        struct darshan_job job;
        ret = darshan_log_getjob(fd, &job);
345

346
        if(ret < 0 || !(state->exe_mnt_data))
347
            return(-1);
348
    }
349

350
    /* exe string is located before the first line break */
351
    newline = strchr(state->exe_mnt_data, '\n');
352 353

    /* copy over the exe string */
354
    if(newline)
355
        memcpy(buf, state->exe_mnt_data, (newline - state->exe_mnt_data));
356 357 358 359

    return (0);
}

360 361 362
/* darshan_log_putexe()
 *
 * wrties the application exe name to darshan log file
363
 * NOTE: this needs to be called immediately following put_job as it
364
 * expects the file pointer to be positioned immediately following
365
 * the darshan job information
366 367 368 369 370
 *
 * returns 0 on success, -1 on failure 
 */
int darshan_log_putexe(darshan_fd fd, char *buf)
{
371 372
    struct darshan_fd_int_state *state = fd->state;
    int len = strlen(buf);
373
    int ret;
374

375
    assert(fd->state);
376

377 378
    ret = darshan_log_dzwrite(fd, DARSHAN_JOB_REGION_ID, buf, len);
    if(ret != len)
379
    {
380
        state->err = -1;
381 382 383 384 385 386 387
        fprintf(stderr, "Error: failed to write exe string to darshan log file.\n");
        return(-1);
    }

    return(0);
}

388 389
/* darshan_log_getmounts()
 * 
390
 * retrieves mount table information from the log. Note that mnt_pts and
391
 * fs_types are arrays that will be allocated by the function and must be
392 393 394
 * freed by the caller. count will indicate the size of the arrays
 *
 * returns 0 on success, -1 on failure
395
 */
396
int darshan_log_getmounts(darshan_fd fd, char*** mnt_pts,
397 398
    char*** fs_types, int* count)
{
399
    struct darshan_fd_int_state *state = fd->state;
400 401
    char *pos;
    int array_index = 0;
402
    int ret;
403

404 405
    assert(state);

406
    /* if the exe/mount data has not been saved yet, read in the job info */
407
    if(!(state->exe_mnt_data))
408
    {
409 410
        struct darshan_job job;
        ret = darshan_log_getjob(fd, &job);
411

412
        if(ret < 0 || !(state->exe_mnt_data))
413
            return(-1);
414
    }
415

416
    /* count entries */
417
    *count = 0;
418
    pos = state->exe_mnt_data;
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
    while((pos = strchr(pos, '\n')) != NULL)
    {
        pos++;
        (*count)++;
    }

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

    /* allocate output arrays */
    *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)
     */
440
    while((pos = strrchr(state->exe_mnt_data, '\n')) != NULL)
441 442
    {
        /* overestimate string lengths */
443
        (*mnt_pts)[array_index] = malloc(DARSHAN_EXE_LEN);
444
        assert((*mnt_pts)[array_index]);
445
        (*fs_types)[array_index] = malloc(DARSHAN_EXE_LEN);
446 447
        assert((*fs_types)[array_index]);

448 449 450
        ret = sscanf(++pos, "%s\t%s", (*fs_types)[array_index],
            (*mnt_pts)[array_index]);
        if(ret != 2)
451
        {
452
            fprintf(stderr, "Error: poorly formatted mount table in darshan log file.\n");
453 454 455 456 457 458 459
            return(-1);
        }
        pos--;
        *pos = '\0';
        array_index++;
    }

460 461 462
    return(0);
}

463 464 465 466 467 468 469 470 471 472 473
/* darshan_log_putmounts()
 *
 * writes mount information to the darshan log file
 * NOTE: this function call should follow immediately after the call
 * to darshan_log_putexe(), as it assumes the darshan log file pointer
 * is pointing to the offset immediately following the exe string
 *
 * returns 0 on success, -1 on failure
 */
int darshan_log_putmounts(darshan_fd fd, char** mnt_pts, char** fs_types, int count)
{
474
    struct darshan_fd_int_state *state = fd->state;
475 476
    int i;
    char line[1024];
477 478 479
    char mnt_dat[DARSHAN_EXE_LEN] = {0};
    int mnt_dat_sz = 0;
    char *tmp;
480 481
    int ret;

482 483
    assert(state);

484
    /* write each mount entry to file */
485
    tmp = mnt_dat;
486 487 488
    for(i=count-1; i>=0; i--)
    {
        sprintf(line, "\n%s\t%s", fs_types[i], mnt_pts[i]);
489 490 491 492 493 494

        memcpy(tmp, line, strlen(line));
        tmp += strlen(line);
        mnt_dat_sz += strlen(line);
    }

495 496
    ret = darshan_log_dzwrite(fd, DARSHAN_JOB_REGION_ID, mnt_dat, mnt_dat_sz);
    if (ret != mnt_dat_sz)
497
    {
498
        state->err = -1;
499
        fprintf(stderr, "Error: failed to write darshan log mount data.\n");
500 501 502 503 504 505
        return(-1);
    }

    return(0);
}

506
/* darshan_log_gethash()
507
 *
508
 * read the hash of records from the darshan log file
509 510 511
 *
 * returns 0 on success, -1 on failure
 */
512
int darshan_log_gethash(darshan_fd fd, struct darshan_record_ref **hash)
513
{
514
    struct darshan_fd_int_state *state = fd->state;
515
    char *hash_buf;
516
    int hash_buf_sz;
517
    char *buf_ptr;
518
    darshan_record_id *rec_id_ptr;
519
    uint32_t *path_len_ptr, tmp_path_len;
520 521
    char *path_ptr;
    struct darshan_record_ref *ref;
522 523 524
    int read;
    int read_req_sz;
    int buf_remaining = 0;
525

526 527
    assert(state);

528 529 530 531 532 533 534
    /* just return if there is no record mapping data */
    if(fd->rec_map.len == 0)
    {
        *hash = NULL;
        return(0);
    }

535 536 537
    /* default to hash buffer twice as big as default compression buf */
    hash_buf = malloc(DARSHAN_DEF_COMP_BUF_SZ * 2);
    if(!hash_buf)
538
        return(-1);
539 540
    memset(hash_buf, 0, DARSHAN_DEF_COMP_BUF_SZ * 2);
    hash_buf_sz = DARSHAN_DEF_COMP_BUF_SZ * 2;
541

542
    do
543
    {
544 545
        /* read chunks of the darshan record id -> file name mapping from log file,
         * constructing a hash table in the process
546
         */
547 548 549 550
        read_req_sz = hash_buf_sz - buf_remaining;
        read = darshan_log_dzread(fd, DARSHAN_REC_MAP_REGION_ID,
            hash_buf + buf_remaining, read_req_sz);
        if(read < 0)
551
        {
552 553 554
            fprintf(stderr, "Error: failed to read record hash from darshan log file.\n");
            free(hash_buf);
            return(-1);
555 556
        }

557 558 559 560 561 562 563 564
        /* work through the hash buffer -- deserialize the mapping data and
         * add to the output hash table
         * NOTE: these mapping pairs are variable in length, so we have to be able
         * to handle incomplete mappings temporarily here
         */
        buf_ptr = hash_buf;
        buf_remaining += read;
        while(buf_remaining > (sizeof(darshan_record_id) + sizeof(uint32_t)))
565
        {
566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586
            /* see if we have enough buf space to read in the next full record */
            tmp_path_len = *(uint32_t *)(buf_ptr + sizeof(darshan_record_id));
            if(fd->swap_flag)
                DARSHAN_BSWAP32(&tmp_path_len);

            /* we need to read more before we continue deserializing */
            if(buf_remaining <
                (sizeof(darshan_record_id) + sizeof(uint32_t) + tmp_path_len))
                break;

            /* get pointers for each field of this darshan record */
            /* NOTE: darshan record hash serialization method: 
             *          ... darshan_record_id | (uint32_t) path_len | path ...
             */
            rec_id_ptr = (darshan_record_id *)buf_ptr;
            buf_ptr += sizeof(darshan_record_id);
            path_len_ptr = (uint32_t *)buf_ptr;
            buf_ptr += sizeof(uint32_t);
            path_ptr = (char *)buf_ptr;

            if(fd->swap_flag)
587
            {
588 589 590
                /* we need to sort out endianness issues before deserializing */
                DARSHAN_BSWAP64(rec_id_ptr);
                DARSHAN_BSWAP32(path_len_ptr);
591
            }
592 593 594

            HASH_FIND(hlink, *hash, rec_id_ptr, sizeof(darshan_record_id), ref);
            if(!ref)
595
            {
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616
                ref = malloc(sizeof(*ref));
                if(!ref)
                {
                    free(hash_buf);
                    return(-1);
                }
                ref->rec.name = malloc(*path_len_ptr + 1);
                if(!ref->rec.name)
                {
                    free(ref);
                    free(hash_buf);
                    return(-1);
                }

                /* set the fields for this record */
                ref->rec.id = *rec_id_ptr;
                memcpy(ref->rec.name, path_ptr, *path_len_ptr);
                ref->rec.name[*path_len_ptr] = '\0';

                /* add this record to the hash */
                HASH_ADD(hlink, *hash, rec.id, sizeof(darshan_record_id), ref);
617 618
            }

619 620 621
            buf_ptr += *path_len_ptr;
            buf_remaining -=
                (sizeof(darshan_record_id) + sizeof(uint32_t) + *path_len_ptr);
622
        }
623

624 625 626 627 628 629 630 631
        /* copy any leftover data to beginning of buffer to parse next */
        memcpy(hash_buf, buf_ptr, buf_remaining);

        /* we keep reading until we get a short read informing us we have
         * read all of the record hash
         */
    } while(read == read_req_sz);
    assert(buf_remaining == 0);
632

633
    free(hash_buf);
634 635 636
    return(0);
}

637 638 639
/* darshan_log_puthash()
 *
 * writes the hash table of records to the darshan log file
640 641 642
 * NOTE: this function call should follow immediately after the call
 * to darshan_log_putmounts(), as it assumes the darshan log file pointer
 * is pointing to the offset immediately following the mount information
643 644 645 646 647
 *
 * returns 0 on success, -1 on failure
 */
int darshan_log_puthash(darshan_fd fd, struct darshan_record_ref *hash)
{
648
    struct darshan_fd_int_state *state = fd->state;
649
    char *hash_buf;
650
    int hash_buf_sz;
651
    struct darshan_record_ref *ref, *tmp;
652 653 654
    char *buf_ptr;
    int path_len;
    int wrote;
655

656 657
    assert(state);

658 659
    /* allocate memory for largest possible hash record */
    hash_buf_sz = sizeof(darshan_record_id) + sizeof(uint32_t) + PATH_MAX;
660 661 662
    hash_buf = malloc(hash_buf_sz);
    if(!hash_buf)
        return(-1);
663
    memset(hash_buf, 0, hash_buf_sz);
664

665
    /* individually serialize each hash record and write to log file */
666 667
    HASH_ITER(hlink, hash, ref, tmp)
    {
668 669
        buf_ptr = hash_buf;
        path_len = strlen(ref->rec.name);
670

671
        /* the hash buffer has space to serialize this record
672 673 674
         * NOTE: darshan record hash serialization method: 
         *          ... darshan_record_id | (uint32_t) path_len | path ...
         */
675 676 677 678 679 680
        *((darshan_record_id *)buf_ptr) = ref->rec.id;
        buf_ptr += sizeof(darshan_record_id);
        *((uint32_t *)buf_ptr) = path_len;
        buf_ptr += sizeof(uint32_t);
        memcpy(buf_ptr, ref->rec.name, path_len);
        buf_ptr += path_len;
681

682 683 684 685 686
        /* write this hash entry to log file */
        wrote = darshan_log_dzwrite(fd, DARSHAN_REC_MAP_REGION_ID,
            hash_buf, (buf_ptr - hash_buf));
        if(wrote != (buf_ptr - hash_buf))
        {
687
            state->err = -1;
688 689 690 691
            fprintf(stderr, "Error: failed to write record hash to darshan log file.\n");
            free(hash_buf);
            return(-1);
        }
692 693 694 695 696 697 698 699
    }

    free(hash_buf);
    return(0);
}

/* darshan_log_getmod()
 *
700 701 702
 * get a chunk of module data from the darshan log file
 *
 * returns 0 on success, -1 on failure
703 704
 */
int darshan_log_getmod(darshan_fd fd, darshan_module_id mod_id,
705
    void *buf, int len)
706
{
707
    struct darshan_fd_int_state *state = fd->state;
708
    int ret;
709

710 711
    assert(state);

712
    if(mod_id < 0 || mod_id >= DARSHAN_MAX_MODS)
713
    {
714 715
        fprintf(stderr, "Error: invalid Darshan module id.\n");
        return(-1);
716 717
    }

718 719 720 721
    if(fd->mod_map[mod_id].len == 0)
        return(0); /* no data corresponding to this mod_id */

    /* read this module's data from the log file */
722 723
    ret = darshan_log_dzread(fd, mod_id, buf, len);
    if(ret < 0)
724
    {
725
        fprintf(stderr,
726
            "Error: failed to read module %s data from darshan log file.\n",
727
            darshan_module_names[mod_id]);
728 729 730
        return(-1);
    }

731
    return(ret);
732 733
}

734 735 736
/* darshan_log_putmod()
 *
 * write a chunk of module data to the darshan log file
Shane Snyder's avatar
Shane Snyder committed
737 738 739 740 741 742
 * NOTE: this function call should be called directly after the
 * put_hash() function, as it expects the file pointer to be
 * positioned directly past the record hash location. Also,
 * for a set of modules with data to write to file, this function
 * should be called in order of increasing module identifiers,
 * as the darshan log file format expects this ordering.
743 744 745 746 747 748
 *
 * returns 0 on success, -1 on failure
 */
int darshan_log_putmod(darshan_fd fd, darshan_module_id mod_id,
    void *mod_buf, int mod_buf_sz)
{
749
    struct darshan_fd_int_state *state = fd->state;
750 751
    int ret;

752 753
    assert(state);

Shane Snyder's avatar
Shane Snyder committed
754
    if(mod_id < 0 || mod_id >= DARSHAN_MAX_MODS)
755
    {
756
        state->err = -1;
Shane Snyder's avatar
Shane Snyder committed
757 758
        fprintf(stderr, "Error: invalid Darshan module id.\n");
        return(-1);
759 760 761
    }

    /* write the module chunk to the log file */
762 763
    ret = darshan_log_dzwrite(fd, mod_id, mod_buf, mod_buf_sz);
    if(ret != mod_buf_sz)
764
    {
765
        state->err = -1;
766 767 768 769 770 771 772 773 774
        fprintf(stderr,
            "Error: failed to write module %s data to darshan log file.\n",
            darshan_module_names[mod_id]);
        return(-1);
    }

    return(0);
}

775 776
/* darshan_log_close()
 *
777
 * close an open darshan file descriptor, freeing any resources
778 779 780 781
 *
 * returns 0 on success, -1 on failure
 */
void darshan_log_close(darshan_fd fd)
782
{
783
    struct darshan_fd_int_state *state = fd->state;
784 785
    int ret;

786 787 788 789
    assert(state);

    /* if the file was created for writing */
    if(state->creat_flag)
790 791
    {
        /* flush the last region of the log to file */
792
        switch(state->comp_type)
793 794
        {
            case DARSHAN_ZLIB_COMP:
795
                ret = darshan_log_libz_flush(fd, state->dz.prev_reg_id);
796 797
                if(ret == 0)
                    break;
798 799
#ifdef HAVE_LIBBZ2
            case DARSHAN_BZIP2_COMP:
800
                ret = darshan_log_bzip2_flush(fd, state->dz.prev_reg_id);
801 802 803
                if(ret == 0)
                    break;
#endif 
804 805
            default:
                /* if flush fails, remove the output log file */
806
                state->err = -1;
807 808 809 810 811
                fprintf(stderr, "Error: final flush to log file failed.\n");
                break;
        }

        /* if no errors flushing, write the log header before closing */
812
        if(state->err != -1)
813 814 815
        {
            ret = darshan_log_putheader(fd);
            if(ret < 0)
816
                state->err = -1;
817 818 819
        }
    }

820
    close(state->fildes);
821 822

    /* remove output log file if error writing to it */
823
    if((state->creat_flag) && (state->err == -1))
824 825
    {
        fprintf(stderr, "Unlinking darshan log file %s ...\n",
826 827
            state->logfile_path);
        unlink(state->logfile_path);
828
    }
829

830 831 832 833
    darshan_log_dzdestroy(state);
    if(state->exe_mnt_data)
        free(state->exe_mnt_data);
    free(state);
834
    free(fd);
835 836

    return;
837 838
}

839
/* **************************************************** */
840

841 842 843 844 845 846 847
/* read the header of the darshan log and set internal fd data structures
 * NOTE: this is the only portion of the darshan log that is uncompressed
 *
 * returns 0 on success, -1 on failure
 */
static int darshan_log_getheader(darshan_fd fd)
{
848
    struct darshan_fd_int_state *state = fd->state;
849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
    struct darshan_header header;
    int i;
    int ret;

    ret = darshan_log_seek(fd, 0);
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to seek in darshan log file.\n");
        return(-1);
    }

    /* read uncompressed header from log file */
    ret = darshan_log_read(fd, &header, sizeof(header));
    if(ret != sizeof(header))
    {
        fprintf(stderr, "Error: failed to read darshan log file header.\n");
        return(-1);
    }

    /* save the version string */
    strncpy(fd->version, header.version_string, 8);

    if(header.magic_nr == DARSHAN_MAGIC_NR)
    {
        /* no byte swapping needed, this file is in host format already */
        fd->swap_flag = 0;
    }
    else
    {
        /* try byte swapping */
        DARSHAN_BSWAP64(&(header.magic_nr));
        if(header.magic_nr == DARSHAN_MAGIC_NR)
        {
            fd->swap_flag = 1;

            /* swap the log map variables in the header */
            DARSHAN_BSWAP64(&(header.rec_map.off));
            DARSHAN_BSWAP64(&(header.rec_map.len));
            for(i = 0; i < DARSHAN_MAX_MODS; i++)
            {
                DARSHAN_BSWAP64(&(header.mod_map[i].off));
                DARSHAN_BSWAP64(&(header.mod_map[i].len));
            }
        }
        else
        {
            /* otherwise this file is just broken */
            fprintf(stderr, "Error: bad magic number in darshan log file.\n");
            return(-1);
        }
    }

901
    state->comp_type = header.comp_type;
902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917

    /* save the mapping of data within log file to this file descriptor */
    fd->job_map.off = sizeof(struct darshan_header);
    fd->job_map.len = header.rec_map.off - fd->job_map.off;
    memcpy(&fd->rec_map, &(header.rec_map), sizeof(struct darshan_log_map));
    memcpy(&fd->mod_map, &(header.mod_map), DARSHAN_MAX_MODS * sizeof(struct darshan_log_map));

    return(0);
}

/* write a darshan header to log file
 *
 * returns 0 on success, -1 on failure
 */
static int darshan_log_putheader(darshan_fd fd)
{
918
    struct darshan_fd_int_state *state = fd->state;
919 920 921 922 923 924 925 926 927 928 929 930 931
    struct darshan_header header;
    int ret;

    ret = darshan_log_seek(fd, 0);
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to seek in darshan log file.\n");
        return(-1);
    }

    memset(&header, 0, sizeof(header));
    strcpy(header.version_string, DARSHAN_LOG_VERSION);
    header.magic_nr = DARSHAN_MAGIC_NR;
932
    header.comp_type = state->comp_type;
933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948

    /* copy the mapping information to the header */
    memcpy(&header.rec_map, &fd->rec_map, sizeof(struct darshan_log_map));
    memcpy(&header.mod_map, &fd->mod_map, DARSHAN_MAX_MODS * sizeof(struct darshan_log_map));

    /* write header to file */
    ret = darshan_log_write(fd, &header, sizeof(header));
    if(ret != sizeof(header))
    {
        fprintf(stderr, "Error: failed to write Darshan log file header.\n");
        return(-1);
    }

    return(0);
}

949 950 951 952
/* return 0 on successful seek to offset, -1 on failure.
 */
static int darshan_log_seek(darshan_fd fd, off_t offset)
{
953
    struct darshan_fd_int_state *state = fd->state;
954
    off_t ret_off;
955

956
    if(state->pos == offset)
957 958
        return(0);

959
    ret_off = lseek(state->fildes, offset, SEEK_SET);
960
    if(ret_off == offset)
961
    {
962
        state->pos = offset;
963
        return(0);
964 965 966 967 968
    }

    return(-1);
}

969
/* return amount read on success, 0 on EOF, -1 on failure.
970
 */
971
static int darshan_log_read(darshan_fd fd, void* buf, int len)
972
{
973
    struct darshan_fd_int_state *state = fd->state;
974 975
    int ret;

976
    /* read data from the log file using the given map */
977
    ret = read(state->fildes, buf, len);
978
    if(ret > 0)
979
        state->pos += ret;
980

981
    return(ret);
982 983
}

984
/* return amount written on success, -1 on failure.
985
 */
986
static int darshan_log_write(darshan_fd fd, void* buf, int len)
987
{
988
    struct darshan_fd_int_state *state = fd->state;
989 990
    int ret;

991
    ret = write(state->fildes, buf, len);
992
    if(ret > 0)
993
        state->pos += ret;
994 995 996 997

    return(ret);
}

998
static int darshan_log_dzinit(struct darshan_fd_int_state *state)
999 1000 1001
{
    int ret;

1002
    /* initialize buffers for staging compressed data to/from log file */
1003 1004
    state->dz.buf = malloc(DARSHAN_DEF_COMP_BUF_SZ);
    if(state->dz.buf == NULL)
1005 1006
        return(-1);

1007
    state->dz.prev_reg_id = DARSHAN_HEADER_REGION_ID;
1008

1009
    switch(state->comp_type)
1010 1011
    {
        case DARSHAN_ZLIB_COMP:
1012 1013 1014 1015
        {
            z_stream *tmp_zstrm = malloc(sizeof(*tmp_zstrm));
            if(!tmp_zstrm)
            {
1016
                free(state->dz.buf);
1017 1018 1019 1020 1021 1022 1023 1024 1025
                return(-1);
            }
            tmp_zstrm->zalloc = Z_NULL;
            tmp_zstrm->zfree = Z_NULL;
            tmp_zstrm->opaque = Z_NULL;
            tmp_zstrm->avail_in = 0;
            tmp_zstrm->next_in = Z_NULL;

            /* TODO: worth using {inflate/deflate}Init2 ?? */
1026
            if(!(state->creat_flag))
1027 1028 1029 1030 1031 1032 1033 1034 1035
            {
                /* read only file, init inflate algorithm */
                ret = inflateInit(tmp_zstrm);
            }
            else
            {
                /* write only file, init deflate algorithm */
                ret = deflateInit(tmp_zstrm, Z_DEFAULT_COMPRESSION);
                tmp_zstrm->avail_out = DARSHAN_DEF_COMP_BUF_SZ;
1036
                tmp_zstrm->next_out = state->dz.buf;
1037 1038 1039 1040
            }
            if(ret != Z_OK)
            {
                free(tmp_zstrm);
1041
                free(state->dz.buf);
1042 1043
                return(-1);
            }
1044
            state->dz.strm = tmp_zstrm;
1045
            break;
1046
        }
1047 1048 1049 1050 1051 1052
#ifdef HAVE_LIBBZ2
        case DARSHAN_BZIP2_COMP:
        {
            bz_stream *tmp_bzstrm = malloc(sizeof(*tmp_bzstrm));
            if(!tmp_bzstrm)
            {
1053
                free(state->dz.buf);
1054 1055 1056 1057 1058 1059 1060 1061
                return(-1);
            }
            tmp_bzstrm->bzalloc = NULL;
            tmp_bzstrm->bzfree = NULL;
            tmp_bzstrm->opaque = NULL;
            tmp_bzstrm->avail_in = 0;
            tmp_bzstrm->next_in = Z_NULL;

1062
            if(state->creat_flag)
1063 1064 1065 1066 1067 1068 1069 1070 1071
            {
                /* read only file, init decompress algorithm */
                ret = BZ2_bzDecompressInit(tmp_bzstrm, 1, 0);
            }
            else
            {
                /* write only file, init compress algorithm */
                ret = BZ2_bzCompressInit(tmp_bzstrm, 9, 1, 30);
                tmp_bzstrm->avail_out = DARSHAN_DEF_COMP_BUF_SZ;
1072
                tmp_bzstrm->next_out = (char *)state->dz.buf;
1073 1074 1075 1076
            }
            if(ret != BZ_OK)
            {
                free(tmp_bzstrm);
1077
                free(state->dz.buf);
1078 1079
                return(-1);
            }
1080
            state->dz.strm = tmp_bzstrm;
1081 1082 1083
            break;
        }
#endif
1084 1085 1086 1087 1088 1089 1090 1091
        default:
            fprintf(stderr, "Error: invalid compression type.\n");
            return(-1);
    }

    return(0);
}

1092
static void darshan_log_dzdestroy(struct darshan_fd_int_state *state)
1093
{
1094
    switch(state->comp_type)
1095 1096
    {
        case DARSHAN_ZLIB_COMP:
1097 1098
            if(!(state->creat_flag))
                inflateEnd(state->dz.strm);
1099
            else
1100 1101
                deflateEnd(state->dz.strm);
            free(state->dz.strm);
1102
            break;
1103 1104
#ifdef HAVE_LIBBZ2
        case DARSHAN_BZIP2_COMP:
1105 1106
            if(!(state->creat_flag))
                BZ2_bzDecompressEnd(state->dz.strm);
1107
            else
1108 1109
                BZ2_bzCompressEnd(state->dz.strm);
            free(state->dz.strm);
1110 1111
            break;
#endif
1112 1113 1114 1115
        default:
            fprintf(stderr, "Error: invalid compression type.\n");
    }

1116
    free(state->dz.buf);
1117 1118 1119 1120 1121
    return;
}

static int darshan_log_dzread(darshan_fd fd, int region_id, void *buf, int len)
{
1122
    struct darshan_fd_int_state *state = fd->state;
1123 1124
    int ret;

1125
    switch(state->comp_type)
1126 1127 1128
    {
        case DARSHAN_ZLIB_COMP:
            ret = darshan_log_libz_read(fd, region_id, buf, len);
1129
            break;
1130 1131 1132 1133 1134
#ifdef HAVE_LIBBZ2
        case DARSHAN_BZIP2_COMP:
            ret = darshan_log_bzip2_read(fd, region_id, buf, len);
            break;
#endif
1135
        default:
1136
            fprintf(stderr, "Error: invalid compression type.\n");
Shane Snyder's avatar