darshan-posix.c 59.1 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 8 9
#define _XOPEN_SOURCE 500
#define _GNU_SOURCE

10
#include "darshan-runtime-config.h"
11 12 13 14 15 16 17 18 19 20 21 22 23 24
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdarg.h>
#include <string.h>
#include <time.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/uio.h>
#include <sys/mman.h>
#include <search.h>
#include <assert.h>
25
#include <libgen.h>
26
#include <aio.h>
27
#include <pthread.h>
28

29
#include "utlist.h"
30
#include "darshan.h"
31
#include "darshan-dynamic.h"
32

33
#ifndef HAVE_OFF64_T
34 35
typedef int64_t off64_t;
#endif
36 37 38
#ifndef HAVE_AIOCB64
#define aiocb64 aiocb
#endif
39

40

41 42
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(open64, int, (const char *path, int flags, ...));
43 44
DARSHAN_FORWARD_DECL(creat, int, (const char* path, mode_t mode));
DARSHAN_FORWARD_DECL(creat64, int, (const char* path, mode_t mode));
45 46 47 48
DARSHAN_FORWARD_DECL(mkstemp, int, (char *template));
DARSHAN_FORWARD_DECL(mkostemp, int, (char *template, int flags));
DARSHAN_FORWARD_DECL(mkstemps, int, (char *template, int suffixlen));
DARSHAN_FORWARD_DECL(mkostemps, int, (char *template, int suffixlen, int flags));
49 50 51 52 53 54 55 56
DARSHAN_FORWARD_DECL(read, ssize_t, (int fd, void *buf, size_t count));
DARSHAN_FORWARD_DECL(write, ssize_t, (int fd, const void *buf, size_t count));
DARSHAN_FORWARD_DECL(pread, ssize_t, (int fd, void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pwrite, ssize_t, (int fd, const void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pread64, ssize_t, (int fd, void *buf, size_t count, off64_t offset));
DARSHAN_FORWARD_DECL(pwrite64, ssize_t, (int fd, const void *buf, size_t count, off64_t offset));
DARSHAN_FORWARD_DECL(readv, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
DARSHAN_FORWARD_DECL(writev, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
57 58
DARSHAN_FORWARD_DECL(lseek, off_t, (int fd, off_t offset, int whence));
DARSHAN_FORWARD_DECL(lseek64, off64_t, (int fd, off64_t offset, int whence));
59 60 61 62 63 64
DARSHAN_FORWARD_DECL(__xstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__xstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__lxstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__lxstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__fxstat, int, (int vers, int fd, struct stat *buf));
DARSHAN_FORWARD_DECL(__fxstat64, int, (int vers, int fd, struct stat64 *buf));
65
#ifdef DARSHAN_WRAP_MMAP
Shane Snyder's avatar
Shane Snyder committed
66 67
DARSHAN_FORWARD_DECL(mmap, void*, (void *addr, size_t length, int prot, int flags, int fd, off_t offset));
DARSHAN_FORWARD_DECL(mmap64, void*, (void *addr, size_t length, int prot, int flags, int fd, off64_t offset));
68
#endif /* DARSHAN_WRAP_MMAP */
69 70
DARSHAN_FORWARD_DECL(fsync, int, (int fd));
DARSHAN_FORWARD_DECL(fdatasync, int, (int fd));
71
DARSHAN_FORWARD_DECL(close, int, (int fd));
72 73 74 75 76 77 78 79
DARSHAN_FORWARD_DECL(aio_read, int, (struct aiocb *aiocbp));
DARSHAN_FORWARD_DECL(aio_write, int, (struct aiocb *aiocbp));
DARSHAN_FORWARD_DECL(aio_read64, int, (struct aiocb64 *aiocbp));
DARSHAN_FORWARD_DECL(aio_write64, int, (struct aiocb64 *aiocbp));
DARSHAN_FORWARD_DECL(aio_return, ssize_t, (struct aiocb *aiocbp));
DARSHAN_FORWARD_DECL(aio_return64, ssize_t, (struct aiocb64 *aiocbp));
DARSHAN_FORWARD_DECL(lio_listio, int, (int mode, struct aiocb *const aiocb_list[], int nitems, struct sigevent *sevp));
DARSHAN_FORWARD_DECL(lio_listio64, int, (int mode, struct aiocb64 *const aiocb_list[], int nitems, struct sigevent *sevp));
80

81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98
/* The posix_file_record_ref structure maintains necessary runtime metadata
 * for the POSIX file record (darshan_posix_file structure, defined in
 * darshan-posix-log-format.h) pointed to by 'file_rec'. This metadata
 * assists with the instrumenting of specific statistics in the file record.
 *
 * RATIONALE: the POSIX module needs to track some stateful, volatile 
 * information about each open file (like the current file offset, most recent 
 * access time, etc.) to aid in instrumentation, but this information can't be
 * stored in the darshan_posix_file struct because we don't want it to appear in
 * the final darshan log file.  We therefore associate a posix_file_record_ref
 * struct with each darshan_posix_file struct in order to track this information
 * (i.e., the mapping between posix_file_record_ref structs to darshan_posix_file
 * structs is one-to-one).
 *
 * NOTE: we use the 'darshan_record_ref' interface (in darshan-common) to
 * associate different types of handles with this posix_file_record_ref struct.
 * This allows us to index this struct (and the underlying file record) by using
 * either the corresponding Darshan record identifier (derived from the filename)
99 100
 * or by a generated file descriptor, for instance. Note that, while there should
 * only be a single Darshan record identifier that indexes a posix_file_record_ref,
101 102
 * there could be multiple open file descriptors that index it.
 */
103
struct posix_file_record_ref
104
{
105
    struct darshan_posix_file *file_rec;
106 107 108
    int64_t offset;
    int64_t last_byte_read;
    int64_t last_byte_written;
Shane Snyder's avatar
Shane Snyder committed
109
    enum darshan_io_type last_io_type;
110 111 112
    double last_meta_end;
    double last_read_end;
    double last_write_end;
113
    void *access_root;
114
    int access_count;
115
    void *stride_root;
116
    int stride_count;
117
    struct posix_aio_tracker* aio_list;
118
    int fs_type; /* same as darshan_fs_info->fs_type */
119 120
};

121 122 123 124
/* The posix_runtime structure maintains necessary state for storing
 * POSIX file records and for coordinating with darshan-core at 
 * shutdown time.
 */
125 126
struct posix_runtime
{
127 128 129
    void *rec_id_hash;
    void *fd_hash;
    int file_rec_count;
130 131
};

132 133 134 135 136
/* struct to track information about aio operations in flight */
struct posix_aio_tracker
{
    double tm1;
    void *aiocbp;
137
    struct posix_aio_tracker *next;
138
};
139

140 141 142
static void posix_runtime_initialize(
    void);
static struct posix_file_record_ref *posix_track_new_file_record(
143
    darshan_record_id rec_id, const char *path);
144 145 146 147
static void posix_aio_tracker_add(
    int fd, void *aiocbp);
static struct posix_aio_tracker* posix_aio_tracker_del(
    int fd, void *aiocbp);
148 149
static void posix_finalize_file_records(
    void *rec_ref_p);
150 151 152 153 154
static void posix_record_reduction_op(
    void* infile_v, void* inoutfile_v, int *len, MPI_Datatype *datatype);
static void posix_shared_record_variance(
    MPI_Comm mod_comm, struct darshan_posix_file *inrec_array,
    struct darshan_posix_file *outrec_array, int shared_rec_count);
155
static void posix_cleanup_runtime(
156
    void);
157 158

static void posix_shutdown(
159
    MPI_Comm mod_comm, darshan_record_id *shared_recs,
160
    int shared_rec_count, void **posix_buf, int *posix_buf_sz);
161

162 163 164 165 166
/* extern DXT function defs */
extern void dxt_posix_write(darshan_record_id rec_id, int64_t offset,
    int64_t length, double start_time, double end_time);
extern void dxt_posix_read(darshan_record_id rec_id, int64_t offset,
    int64_t length, double start_time, double end_time);
167

168 169 170 171 172
static struct posix_runtime *posix_runtime = NULL;
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
static int instrumentation_disabled = 0;
static int my_rank = -1;
static int darshan_mem_alignment = 1;
173
static int enable_dxt_io_trace = 0;
174

175 176 177
#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)

178 179
#define POSIX_PRE_RECORD() do { \
    POSIX_LOCK(); \
180
    if(!instrumentation_disabled) { \
181 182 183
        if(!posix_runtime) { \
            posix_runtime_initialize(); \
        } \
184
        if(posix_runtime) break; \
185
    } \
186 187
    POSIX_UNLOCK(); \
    return(ret); \
188 189 190 191 192
} while(0)

#define POSIX_POST_RECORD() do { \
    POSIX_UNLOCK(); \
} while(0)
193

194
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __tm1, __tm2) do { \
195
    darshan_record_id rec_id; \
196 197
    struct posix_file_record_ref *rec_ref; \
    char *newpath; \
198
    if(__ret < 0) break; \
199 200 201 202 203 204 205
    newpath = darshan_clean_file_path(__path); \
    if(!newpath) newpath = (char *)__path; \
    if(darshan_core_excluded_path(newpath)) { \
        if(newpath != __path) free(newpath); \
        break; \
    } \
    rec_id = darshan_core_gen_record_id(newpath); \
206
    rec_ref = darshan_lookup_record_ref(posix_runtime->rec_id_hash, &rec_id, sizeof(darshan_record_id)); \
207
    if(!rec_ref) rec_ref = posix_track_new_file_record(rec_id, newpath); \
208
    if(!rec_ref) { \
209 210
        if(newpath != __path) free(newpath); \
        break; \
211
    } \
212
    if(__mode) \
213 214 215 216
        rec_ref->file_rec->counters[POSIX_MODE] = __mode; \
    rec_ref->offset = 0; \
    rec_ref->last_byte_written = 0; \
    rec_ref->last_byte_read = 0; \
217
    rec_ref->file_rec->counters[POSIX_OPENS] += 1; \
218 219 220 221
    if(rec_ref->file_rec->fcounters[POSIX_F_OPEN_START_TIMESTAMP] == 0 || \
     rec_ref->file_rec->fcounters[POSIX_F_OPEN_START_TIMESTAMP] > __tm1) \
        rec_ref->file_rec->fcounters[POSIX_F_OPEN_START_TIMESTAMP] = __tm1; \
    rec_ref->file_rec->fcounters[POSIX_F_OPEN_END_TIMESTAMP] = __tm2; \
222 223
    DARSHAN_TIMER_INC_NO_OVERLAP(rec_ref->file_rec->fcounters[POSIX_F_META_TIME], \
        __tm1, __tm2, rec_ref->last_meta_end); \
224
    darshan_add_record_ref(&(posix_runtime->fd_hash), &__ret, sizeof(int), rec_ref); \
225
    darshan_instrument_fs_data(rec_ref->fs_type, newpath, __ret); \
226
    if(newpath != __path) free(newpath); \
227 228
} while(0)

229
#define POSIX_RECORD_READ(__ret, __fd, __pread_flag, __pread_offset, __aligned, __tm1, __tm2) do { \
230
    struct posix_file_record_ref* rec_ref; \
231
    size_t stride; \
232
    int64_t this_offset; \
233
    int64_t file_alignment; \
234 235
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
236 237
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &(__fd), sizeof(int)); \
    if(!rec_ref) break; \
238
    if(__pread_flag) \
239 240
        this_offset = __pread_offset; \
    else \
241
        this_offset = rec_ref->offset; \
242
    /* DXT to record detailed read tracing information */ \
243 244
    if(enable_dxt_io_trace) { \
        dxt_posix_read(rec_ref->file_rec->base_rec.id, this_offset, __ret, __tm1, __tm2); \
245
    } \
246 247 248 249 250 251 252
    if(this_offset > rec_ref->last_byte_read) \
        rec_ref->file_rec->counters[POSIX_SEQ_READS] += 1;  \
    if(this_offset == (rec_ref->last_byte_read + 1)) \
        rec_ref->file_rec->counters[POSIX_CONSEC_READS] += 1;  \
    if(this_offset > 0 && this_offset > rec_ref->last_byte_read \
        && rec_ref->last_byte_read != 0) \
        stride = this_offset - rec_ref->last_byte_read - 1; \
253
    else \
254
        stride = 0; \
255 256 257 258 259
    rec_ref->last_byte_read = this_offset + __ret - 1; \
    rec_ref->offset = this_offset + __ret; \
    if(rec_ref->file_rec->counters[POSIX_MAX_BYTE_READ] < (this_offset + __ret - 1)) \
        rec_ref->file_rec->counters[POSIX_MAX_BYTE_READ] = (this_offset + __ret - 1); \
    rec_ref->file_rec->counters[POSIX_BYTES_READ] += __ret; \
260
    rec_ref->file_rec->counters[POSIX_READS] += 1; \
261 262 263 264 265 266 267
    DARSHAN_BUCKET_INC(&(rec_ref->file_rec->counters[POSIX_SIZE_READ_0_100]), __ret); \
    darshan_common_val_counter(&rec_ref->access_root, &rec_ref->access_count, __ret, \
        &(rec_ref->file_rec->counters[POSIX_ACCESS1_ACCESS]), \
        &(rec_ref->file_rec->counters[POSIX_ACCESS1_COUNT])); \
    darshan_common_val_counter(&rec_ref->stride_root, &rec_ref->stride_count, stride, \
        &(rec_ref->file_rec->counters[POSIX_STRIDE1_STRIDE]), \
        &(rec_ref->file_rec->counters[POSIX_STRIDE1_COUNT])); \
268
    if(!__aligned) \
269 270
        rec_ref->file_rec->counters[POSIX_MEM_NOT_ALIGNED] += 1; \
    file_alignment = rec_ref->file_rec->counters[POSIX_FILE_ALIGNMENT]; \
271
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
272 273 274 275
        rec_ref->file_rec->counters[POSIX_FILE_NOT_ALIGNED] += 1; \
    if(rec_ref->last_io_type == DARSHAN_IO_WRITE) \
        rec_ref->file_rec->counters[POSIX_RW_SWITCHES] += 1; \
    rec_ref->last_io_type = DARSHAN_IO_READ; \
276 277
    if(rec_ref->file_rec->fcounters[POSIX_F_READ_START_TIMESTAMP] == 0 || \
     rec_ref->file_rec->fcounters[POSIX_F_READ_START_TIMESTAMP] > __tm1) \
278 279 280 281 282 283 284
        rec_ref->file_rec->fcounters[POSIX_F_READ_START_TIMESTAMP] = __tm1; \
    rec_ref->file_rec->fcounters[POSIX_F_READ_END_TIMESTAMP] = __tm2; \
    if(rec_ref->file_rec->fcounters[POSIX_F_MAX_READ_TIME] < __elapsed) { \
        rec_ref->file_rec->fcounters[POSIX_F_MAX_READ_TIME] = __elapsed; \
        rec_ref->file_rec->counters[POSIX_MAX_READ_TIME_SIZE] = __ret; } \
    DARSHAN_TIMER_INC_NO_OVERLAP(rec_ref->file_rec->fcounters[POSIX_F_READ_TIME], \
        __tm1, __tm2, rec_ref->last_read_end); \
285 286
} while(0)

287
#define POSIX_RECORD_WRITE(__ret, __fd, __pwrite_flag, __pwrite_offset, __aligned, __tm1, __tm2) do { \
288
    struct posix_file_record_ref* rec_ref; \
289
    size_t stride; \
290
    int64_t this_offset; \
291
    int64_t file_alignment; \
292 293
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
294 295
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &__fd, sizeof(int)); \
    if(!rec_ref) break; \
296
    if(__pwrite_flag) \
297 298
        this_offset = __pwrite_offset; \
    else \
299
        this_offset = rec_ref->offset; \
300
    /* DXT to record detailed write tracing information */ \
301 302
    if(enable_dxt_io_trace) { \
        dxt_posix_write(rec_ref->file_rec->base_rec.id, this_offset, __ret, __tm1, __tm2); \
303
    } \
304 305 306 307 308 309 310
    if(this_offset > rec_ref->last_byte_written) \
        rec_ref->file_rec->counters[POSIX_SEQ_WRITES] += 1; \
    if(this_offset == (rec_ref->last_byte_written + 1)) \
        rec_ref->file_rec->counters[POSIX_CONSEC_WRITES] += 1; \
    if(this_offset > 0 && this_offset > rec_ref->last_byte_written \
        && rec_ref->last_byte_written != 0) \
        stride = this_offset - rec_ref->last_byte_written - 1; \
311
    else \
312
        stride = 0; \
313 314 315 316 317
    rec_ref->last_byte_written = this_offset + __ret - 1; \
    rec_ref->offset = this_offset + __ret; \
    if(rec_ref->file_rec->counters[POSIX_MAX_BYTE_WRITTEN] < (this_offset + __ret - 1)) \
        rec_ref->file_rec->counters[POSIX_MAX_BYTE_WRITTEN] = (this_offset + __ret - 1); \
    rec_ref->file_rec->counters[POSIX_BYTES_WRITTEN] += __ret; \
318
    rec_ref->file_rec->counters[POSIX_WRITES] += 1; \
319 320 321 322 323 324 325
    DARSHAN_BUCKET_INC(&(rec_ref->file_rec->counters[POSIX_SIZE_WRITE_0_100]), __ret); \
    darshan_common_val_counter(&rec_ref->access_root, &rec_ref->access_count, __ret, \
        &(rec_ref->file_rec->counters[POSIX_ACCESS1_ACCESS]), \
        &(rec_ref->file_rec->counters[POSIX_ACCESS1_COUNT])); \
    darshan_common_val_counter(&rec_ref->stride_root, &rec_ref->stride_count, stride, \
        &(rec_ref->file_rec->counters[POSIX_STRIDE1_STRIDE]), \
        &(rec_ref->file_rec->counters[POSIX_STRIDE1_COUNT])); \
326
    if(!__aligned) \
327 328
        rec_ref->file_rec->counters[POSIX_MEM_NOT_ALIGNED] += 1; \
    file_alignment = rec_ref->file_rec->counters[POSIX_FILE_ALIGNMENT]; \
329
    if(file_alignment > 0 && (this_offset % file_alignment) != 0) \
330 331 332 333
        rec_ref->file_rec->counters[POSIX_FILE_NOT_ALIGNED] += 1; \
    if(rec_ref->last_io_type == DARSHAN_IO_READ) \
        rec_ref->file_rec->counters[POSIX_RW_SWITCHES] += 1; \
    rec_ref->last_io_type = DARSHAN_IO_WRITE; \
334 335
    if(rec_ref->file_rec->fcounters[POSIX_F_WRITE_START_TIMESTAMP] == 0 || \
     rec_ref->file_rec->fcounters[POSIX_F_WRITE_START_TIMESTAMP] > __tm1) \
336 337 338 339 340 341 342
        rec_ref->file_rec->fcounters[POSIX_F_WRITE_START_TIMESTAMP] = __tm1; \
    rec_ref->file_rec->fcounters[POSIX_F_WRITE_END_TIMESTAMP] = __tm2; \
    if(rec_ref->file_rec->fcounters[POSIX_F_MAX_WRITE_TIME] < __elapsed) { \
        rec_ref->file_rec->fcounters[POSIX_F_MAX_WRITE_TIME] = __elapsed; \
        rec_ref->file_rec->counters[POSIX_MAX_WRITE_TIME_SIZE] = __ret; } \
    DARSHAN_TIMER_INC_NO_OVERLAP(rec_ref->file_rec->fcounters[POSIX_F_WRITE_TIME], \
        __tm1, __tm2, rec_ref->last_write_end); \
343
} while(0)
344

345
#define POSIX_LOOKUP_RECORD_STAT(__path, __statbuf, __tm1, __tm2) do { \
346
    darshan_record_id rec_id; \
347 348 349 350 351 352
    struct posix_file_record_ref* rec_ref; \
    char *newpath = darshan_clean_file_path(__path); \
    if(!newpath) newpath = (char *)__path; \
    if(darshan_core_excluded_path(newpath)) { \
        if(newpath != __path) free(newpath); \
        break; \
353
    } \
354 355
    rec_id = darshan_core_gen_record_id(newpath); \
    rec_ref = darshan_lookup_record_ref(posix_runtime->rec_id_hash, &rec_id, sizeof(darshan_record_id)); \
356
    if(!rec_ref) rec_ref = posix_track_new_file_record(rec_id, newpath); \
357
    if(newpath != __path) free(newpath); \
358 359
    if(rec_ref) { \
        POSIX_RECORD_STAT(rec_ref, __statbuf, __tm1, __tm2); \
360 361 362
    } \
} while(0)

363 364 365 366
#define POSIX_RECORD_STAT(__rec_ref, __statbuf, __tm1, __tm2) do { \
    (__rec_ref)->file_rec->counters[POSIX_STATS] += 1; \
    DARSHAN_TIMER_INC_NO_OVERLAP((__rec_ref)->file_rec->fcounters[POSIX_F_META_TIME], \
        __tm1, __tm2, (__rec_ref)->last_meta_end); \
367 368
} while(0)

369

370 371 372 373
/**********************************************************
 *      Wrappers for POSIX I/O functions of interest      * 
 **********************************************************/

374
int DARSHAN_DECL(open)(const char *path, int flags, ...)
375 376 377 378 379
{
    int mode = 0;
    int ret;
    double tm1, tm2;

380 381
    MAP_OR_FAIL(open);

382
    if(flags & O_CREAT) 
383 384 385 386 387 388
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

389
        tm1 = darshan_core_wtime();
390
        ret = __real_open(path, flags, mode);
391
        tm2 = darshan_core_wtime();
392 393 394
    }
    else
    {
395
        tm1 = darshan_core_wtime();
396
        ret = __real_open(path, flags);
397
        tm2 = darshan_core_wtime();
398 399
    }

400
    POSIX_PRE_RECORD();
401
    POSIX_RECORD_OPEN(ret, path, mode, tm1, tm2);
402
    POSIX_POST_RECORD();
403 404 405 406

    return(ret);
}

Shane Snyder's avatar
Shane Snyder committed
407 408 409 410 411 412
int DARSHAN_DECL(open64)(const char *path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

413
    MAP_OR_FAIL(open64);
Shane Snyder's avatar
Shane Snyder committed
414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432

    if(flags & O_CREAT)
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

        tm1 = darshan_core_wtime();
        ret = __real_open64(path, flags, mode);
        tm2 = darshan_core_wtime();
    }
    else
    {
        tm1 = darshan_core_wtime();
        ret = __real_open64(path, flags);
        tm2 = darshan_core_wtime();
    }

433
    POSIX_PRE_RECORD();
434
    POSIX_RECORD_OPEN(ret, path, mode, tm1, tm2);
435
    POSIX_POST_RECORD();
436 437 438 439 440 441 442 443 444 445 446 447 448 449 450

    return(ret);
}

int DARSHAN_DECL(creat)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat);

    tm1 = darshan_core_wtime();
    ret = __real_creat(path, mode);
    tm2 = darshan_core_wtime();

451
    POSIX_PRE_RECORD();
452
    POSIX_RECORD_OPEN(ret, path, mode, tm1, tm2);
453
    POSIX_POST_RECORD();
454 455 456 457 458 459 460 461 462 463

    return(ret);
}

int DARSHAN_DECL(creat64)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat64);
Shane Snyder's avatar
Shane Snyder committed
464

465 466 467 468
    tm1 = darshan_core_wtime();
    ret = __real_creat64(path, mode);
    tm2 = darshan_core_wtime();

469
    POSIX_PRE_RECORD();
470
    POSIX_RECORD_OPEN(ret, path, mode, tm1, tm2);
471
    POSIX_POST_RECORD();
472 473 474 475

    return(ret);
}

476 477 478 479 480 481 482 483 484 485 486
int DARSHAN_DECL(mkstemp)(char* template)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(mkstemp);

    tm1 = darshan_core_wtime();
    ret = __real_mkstemp(template);
    tm2 = darshan_core_wtime();

487
    POSIX_PRE_RECORD();
488
    POSIX_RECORD_OPEN(ret, template, 0, tm1, tm2);
489
    POSIX_POST_RECORD();
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504

    return(ret);
}

int DARSHAN_DECL(mkostemp)(char* template, int flags)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(mkostemp);

    tm1 = darshan_core_wtime();
    ret = __real_mkostemp(template, flags);
    tm2 = darshan_core_wtime();

505
    POSIX_PRE_RECORD();
506
    POSIX_RECORD_OPEN(ret, template, 0, tm1, tm2);
507
    POSIX_POST_RECORD();
508 509 510 511 512 513 514 515 516 517 518 519 520 521 522

    return(ret);
}

int DARSHAN_DECL(mkstemps)(char* template, int suffixlen)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(mkstemps);

    tm1 = darshan_core_wtime();
    ret = __real_mkstemps(template, suffixlen);
    tm2 = darshan_core_wtime();

523
    POSIX_PRE_RECORD();
524
    POSIX_RECORD_OPEN(ret, template, 0, tm1, tm2);
525
    POSIX_POST_RECORD();
526 527 528 529 530 531 532 533 534 535 536 537 538 539 540

    return(ret);
}

int DARSHAN_DECL(mkostemps)(char* template, int suffixlen, int flags)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(mkostemps);

    tm1 = darshan_core_wtime();
    ret = __real_mkostemps(template, suffixlen, flags);
    tm2 = darshan_core_wtime();

541
    POSIX_PRE_RECORD();
542
    POSIX_RECORD_OPEN(ret, template, 0, tm1, tm2);
543
    POSIX_POST_RECORD();
544 545 546 547

    return(ret);
}

548 549 550 551 552 553
ssize_t DARSHAN_DECL(read)(int fd, void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

554
    MAP_OR_FAIL(read);
555

556
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
557 558 559 560 561

    tm1 = darshan_core_wtime();
    ret = __real_read(fd, buf, count);
    tm2 = darshan_core_wtime();

562
    POSIX_PRE_RECORD();
563
    POSIX_RECORD_READ(ret, fd, 0, 0, aligned_flag, tm1, tm2);
564
    POSIX_POST_RECORD();
565 566 567 568 569 570 571 572 573 574 575 576

    return(ret);
}

ssize_t DARSHAN_DECL(write)(int fd, const void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(write);

577
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
578 579 580 581 582

    tm1 = darshan_core_wtime();
    ret = __real_write(fd, buf, count);
    tm2 = darshan_core_wtime();

583
    POSIX_PRE_RECORD();
584
    POSIX_RECORD_WRITE(ret, fd, 0, 0, aligned_flag, tm1, tm2);
585
    POSIX_POST_RECORD();
586 587 588 589 590 591 592 593 594 595 596 597

    return(ret);
}

ssize_t DARSHAN_DECL(pread)(int fd, void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread);

598
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
599 600 601 602 603

    tm1 = darshan_core_wtime();
    ret = __real_pread(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

604
    POSIX_PRE_RECORD();
605
    POSIX_RECORD_READ(ret, fd, 1, offset, aligned_flag, tm1, tm2);
606
    POSIX_POST_RECORD();
607 608 609 610 611 612 613 614 615 616 617 618

    return(ret);
}

ssize_t DARSHAN_DECL(pwrite)(int fd, const void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite);

619
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
620 621 622 623 624

    tm1 = darshan_core_wtime();
    ret = __real_pwrite(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

625
    POSIX_PRE_RECORD();
626
    POSIX_RECORD_WRITE(ret, fd, 1, offset, aligned_flag, tm1, tm2);
627
    POSIX_POST_RECORD();
628 629 630 631 632 633 634 635 636 637 638 639

    return(ret);
}

ssize_t DARSHAN_DECL(pread64)(int fd, void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread64);

640
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
641 642 643 644 645

    tm1 = darshan_core_wtime();
    ret = __real_pread64(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

646
    POSIX_PRE_RECORD();
647
    POSIX_RECORD_READ(ret, fd, 1, offset, aligned_flag, tm1, tm2);
648
    POSIX_POST_RECORD();
649 650 651 652 653 654 655 656 657 658 659 660

    return(ret);
}

ssize_t DARSHAN_DECL(pwrite64)(int fd, const void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite64);

661
    if((unsigned long)buf % darshan_mem_alignment == 0) aligned_flag = 1;
662 663 664 665 666

    tm1 = darshan_core_wtime();
    ret = __real_pwrite64(fd, buf, count, offset);
    tm2 = darshan_core_wtime();

667
    POSIX_PRE_RECORD();
668
    POSIX_RECORD_WRITE(ret, fd, 1, offset, aligned_flag, tm1, tm2);
669
    POSIX_POST_RECORD();
670 671 672 673 674 675 676

    return(ret);
}

ssize_t DARSHAN_DECL(readv)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
677
    int aligned_flag = 1;
678 679 680 681
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(readv);
682

683 684 685 686
    for(i=0; i<iovcnt; i++)
    {
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
            aligned_flag = 0;
687
    }
688 689 690 691 692

    tm1 = darshan_core_wtime();
    ret = __real_readv(fd, iov, iovcnt);
    tm2 = darshan_core_wtime();

693
    POSIX_PRE_RECORD();
694
    POSIX_RECORD_READ(ret, fd, 0, 0, aligned_flag, tm1, tm2);
695
    POSIX_POST_RECORD();
696 697 698 699 700 701 702

    return(ret);
}

ssize_t DARSHAN_DECL(writev)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
703
    int aligned_flag = 1;
704 705 706 707
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(writev);
708

709 710 711 712
    for(i=0; i<iovcnt; i++)
    {
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
            aligned_flag = 0;
713
    }
714 715 716 717 718

    tm1 = darshan_core_wtime();
    ret = __real_writev(fd, iov, iovcnt);
    tm2 = darshan_core_wtime();

719
    POSIX_PRE_RECORD();
720
    POSIX_RECORD_WRITE(ret, fd, 0, 0, aligned_flag, tm1, tm2);
721
    POSIX_POST_RECORD();
722 723 724 725 726 727 728

    return(ret);
}

off_t DARSHAN_DECL(lseek)(int fd, off_t offset, int whence)
{
    off_t ret;
729
    struct posix_file_record_ref *rec_ref;
730 731 732 733 734 735 736 737 738 739
    double tm1, tm2;

    MAP_OR_FAIL(lseek);

    tm1 = darshan_core_wtime();
    ret = __real_lseek(fd, offset, whence);
    tm2 = darshan_core_wtime();

    if(ret >= 0)
    {
740
        POSIX_PRE_RECORD();
741 742
        rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
        if(rec_ref)
743
        {
744
            rec_ref->offset = ret;
745
            DARSHAN_TIMER_INC_NO_OVERLAP(
746 747 748
                rec_ref->file_rec->fcounters[POSIX_F_META_TIME],
                tm1, tm2, rec_ref->last_meta_end);
            rec_ref->file_rec->counters[POSIX_SEEKS] += 1;
749
        }
750
        POSIX_POST_RECORD();
751 752 753 754 755 756 757 758
    }

    return(ret);
}

off_t DARSHAN_DECL(lseek64)(int fd, off_t offset, int whence)
{
    off_t ret;
759
    struct posix_file_record_ref *rec_ref;
760 761 762 763 764 765 766 767 768 769
    double tm1, tm2;

    MAP_OR_FAIL(lseek64);

    tm1 = darshan_core_wtime();
    ret = __real_lseek64(fd, offset, whence);
    tm2 = darshan_core_wtime();

    if(ret >= 0)
    {
770
        POSIX_PRE_RECORD();
771 772
        rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
        if(rec_ref)
773
        {
774
            rec_ref->offset = ret;
775
            DARSHAN_TIMER_INC_NO_OVERLAP(
776 777 778
                rec_ref->file_rec->fcounters[POSIX_F_META_TIME],
                tm1, tm2, rec_ref->last_meta_end);
            rec_ref->file_rec->counters[POSIX_SEEKS] += 1;
779
        }
780
        POSIX_POST_RECORD();
781 782 783 784 785
    }

    return(ret);
}

786 787 788 789 790 791 792 793 794 795 796 797 798 799
int DARSHAN_DECL(__xstat)(int vers, const char *path, struct stat *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__xstat);

    tm1 = darshan_core_wtime();
    ret = __real___xstat(vers, path, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

800
    POSIX_PRE_RECORD();
801
    POSIX_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
802
    POSIX_POST_RECORD();
803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820

    return(ret);
}

int DARSHAN_DECL(__xstat64)(int vers, const char *path, struct stat64 *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__xstat64);

    tm1 = darshan_core_wtime();
    ret = __real___xstat64(vers, path, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

821
    POSIX_PRE_RECORD();
822
    POSIX_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
823
    POSIX_POST_RECORD();
824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841

    return(ret);
}

int DARSHAN_DECL(__lxstat)(int vers, const char *path, struct stat *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__lxstat);

    tm1 = darshan_core_wtime();
    ret = __real___lxstat(vers, path, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

842
    POSIX_PRE_RECORD();
843
    POSIX_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
844
    POSIX_POST_RECORD();
845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862

    return(ret);
}

int DARSHAN_DECL(__lxstat64)(int vers, const char *path, struct stat64 *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__lxstat64);

    tm1 = darshan_core_wtime();
    ret = __real___lxstat64(vers, path, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

863
    POSIX_PRE_RECORD();
864
    POSIX_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
865
    POSIX_POST_RECORD();
866 867 868 869 870 871 872

    return(ret);
}

int DARSHAN_DECL(__fxstat)(int vers, int fd, struct stat *buf)
{
    int ret;
873
    struct posix_file_record_ref *rec_ref;
874 875 876 877 878 879 880 881 882 883 884
    double tm1, tm2;

    MAP_OR_FAIL(__fxstat);

    tm1 = darshan_core_wtime();
    ret = __real___fxstat(vers, fd, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

885
    POSIX_PRE_RECORD();
886 887
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
888
    {
889
        POSIX_RECORD_STAT(rec_ref, buf, tm1, tm2);
890
    }
891
    POSIX_POST_RECORD();
892 893 894 895 896 897 898

    return(ret);
}

int DARSHAN_DECL(__fxstat64)(int vers, int fd, struct stat64 *buf)
{
    int ret;
899
    struct posix_file_record_ref *rec_ref;
900 901 902 903 904 905 906 907 908 909 910
    double tm1, tm2;

    MAP_OR_FAIL(__fxstat64);

    tm1 = darshan_core_wtime();
    ret = __real___fxstat64(vers, fd, buf);
    tm2 = darshan_core_wtime();

    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

911
    POSIX_PRE_RECORD();
912 913
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
914
    {
915
        POSIX_RECORD_STAT(rec_ref, buf, tm1, tm2);
916
    }
917
    POSIX_POST_RECORD();
918 919 920 921

    return(ret);
}

922
#ifdef DARSHAN_WRAP_MMAP
Shane Snyder's avatar
Shane Snyder committed
923 924 925 926
void* DARSHAN_DECL(mmap)(void *addr, size_t length, int prot, int flags,
    int fd, off_t offset)
{
    void* ret;
927
    struct posix_file_record_ref *rec_ref;
Shane Snyder's avatar
Shane Snyder committed
928 929 930

    MAP_OR_FAIL(mmap);

931 932 933 934 935 936 937 938
    if(fd < 0 || (flags & MAP_ANONYMOUS))
    {
        /* mmap is not associated with a backing file; skip all Darshan
         * characterization attempts.
         */
        return(__real_mmap(addr, length, prot, flags, fd, offset));
    }

Shane Snyder's avatar
Shane Snyder committed
939 940 941 942
    ret = __real_mmap(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

943
    POSIX_PRE_RECORD();
944 945
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
Shane Snyder's avatar
Shane Snyder committed
946
    {
947
        rec_ref->file_rec->counters[POSIX_MMAPS] += 1;
Shane Snyder's avatar
Shane Snyder committed
948
    }
949
    POSIX_POST_RECORD();
Shane Snyder's avatar
Shane Snyder committed
950 951 952

    return(ret);
}
953
#endif /* DARSHAN_WRAP_MMAP */
Shane Snyder's avatar
Shane Snyder committed
954

955
#ifdef DARSHAN_WRAP_MMAP
Shane Snyder's avatar
Shane Snyder committed
956 957 958 959
void* DARSHAN_DECL(mmap64)(void *addr, size_t length, int prot, int flags,
    int fd, off64_t offset)
{
    void* ret;
960
    struct posix_file_record_ref *rec_ref;
Shane Snyder's avatar
Shane Snyder committed
961 962 963

    MAP_OR_FAIL(mmap64);

964 965 966 967 968 969 970 971
    if(fd < 0 || (flags & MAP_ANONYMOUS))
    {
        /* mmap is not associated with a backing file; skip all Darshan
         * characterization attempts.
         */
        return(__real_mmap64(addr, length, prot, flags, fd, offset));
    }

Shane Snyder's avatar
Shane Snyder committed
972 973 974 975
    ret = __real_mmap64(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

976
    POSIX_PRE_RECORD();
977 978
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
Shane Snyder's avatar
Shane Snyder committed
979
    {
980
        rec_ref->file_rec->counters[POSIX_MMAPS] += 1;
Shane Snyder's avatar
Shane Snyder committed
981
    }
982
    POSIX_POST_RECORD();
Shane Snyder's avatar
Shane Snyder committed
983 984 985

    return(ret);
}
986
#endif /* DARSHAN_WRAP_MMAP */
Shane Snyder's avatar
Shane Snyder committed
987

988 989 990
int DARSHAN_DECL(fsync)(int fd)
{
    int ret;
991
    struct posix_file_record_ref *rec_ref;
992 993 994 995 996 997 998 999 1000 1001 1002
    double tm1, tm2;

    MAP_OR_FAIL(fsync);

    tm1 = darshan_core_wtime();
    ret = __real_fsync(fd);
    tm2 = darshan_core_wtime();

    if(ret < 0)
        return(ret);

1003
    POSIX_PRE_RECORD();
1004 1005
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
1006
    {
1007
        DARSHAN_TIMER_INC_NO_OVERLAP(
1008 1009 1010
            rec_ref->file_rec->fcounters[POSIX_F_WRITE_TIME],
            tm1, tm2, rec_ref->last_write_end);
        rec_ref->file_rec->counters[POSIX_FSYNCS] += 1;
1011
    }
1012
    POSIX_POST_RECORD();
1013 1014 1015 1016 1017 1018 1019

    return(ret);
}

int DARSHAN_DECL(fdatasync)(int fd)
{
    int ret;
1020
    struct posix_file_record_ref *rec_ref;
1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031
    double tm1, tm2;

    MAP_OR_FAIL(fdatasync);

    tm1 = darshan_core_wtime();
    ret = __real_fdatasync(fd);
    tm2 = darshan_core_wtime();

    if(ret < 0)
        return(ret);

1032
    POSIX_PRE_RECORD();
1033 1034
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
1035
    {
1036
        DARSHAN_TIMER_INC_NO_OVERLAP(
1037 1038 1039
            rec_ref->file_rec->fcounters[POSIX_F_WRITE_TIME],
            tm1, tm2, rec_ref->last_write_end);
        rec_ref->file_rec->counters[POSIX_FDSYNCS] += 1;
1040
    }
1041
    POSIX_POST_RECORD();
1042 1043 1044 1045

    return(ret);
}

1046 1047 1048
int DARSHAN_DECL(close)(int fd)
{
    int ret;
1049 1050
    struct posix_file_record_ref *rec_ref;
    double tm1, tm2;
1051 1052 1053 1054 1055 1056 1057

    MAP_OR_FAIL(close);

    tm1 = darshan_core_wtime();
    ret = __real_close(fd);
    tm2 = darshan_core_wtime();

1058
    POSIX_PRE_RECORD();
1059 1060
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
Shane Snyder's avatar
Shane Snyder committed
1061
    {
1062 1063
        rec_ref->last_byte_written = 0;
        rec_ref->last_byte_read = 0;
1064 1065 1066 1067
        if(rec_ref->file_rec->fcounters[POSIX_F_CLOSE_START_TIMESTAMP] == 0 ||
         rec_ref->file_rec->fcounters[POSIX_F_CLOSE_START_TIMESTAMP] > tm1)
           rec_ref->file_rec->fcounters[POSIX_F_CLOSE_START_TIMESTAMP] = tm1;
        rec_ref->file_rec->fcounters[POSIX_F_CLOSE_END_TIMESTAMP] = tm2;
1068
        DARSHAN_TIMER_INC_NO_OVERLAP(
1069 1070 1071
            rec_ref->file_rec->fcounters[POSIX_F_META_TIME],
            tm1, tm2, rec_ref->last_meta_end);
        darshan_delete_record_ref(&(posix_runtime->fd_hash), &fd, sizeof(int));
Shane Snyder's avatar
Shane Snyder committed
1072
    }
1073
    POSIX_POST_RECORD();
1074 1075 1076 1077

    return(ret);
}

1078 1079 1080 1081 1082 1083 1084 1085 1086
int DARSHAN_DECL(aio_read)(struct aiocb *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_read);

    ret = __real_aio_read(aiocbp);
    if(ret == 0)
    {
1087
        POSIX_PRE_RECORD();
1088
        posix_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1089
        POSIX_POST_RECORD();
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
    }

    return(ret);
}

int DARSHAN_DECL(aio_write)(struct aiocb *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_write);

    ret = __real_aio_write(aiocbp);
    if(ret == 0)
    {
1104
        POSIX_PRE_RECORD();
1105
        posix_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1106
        POSIX_POST_RECORD();
1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
    }

    return(ret);
}

int DARSHAN_DECL(aio_read64)(struct aiocb64 *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_read64);

    ret = __real_aio_read64(aiocbp);
    if(ret == 0)
    {
1121
        POSIX_PRE_RECORD();
1122
        posix_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1123
        POSIX_POST_RECORD();
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137
    }

    return(ret);
}

int DARSHAN_DECL(aio_write64)(struct aiocb64 *aiocbp)
{
    int ret;

    MAP_OR_FAIL(aio_write64);

    ret = __real_aio_write64(aiocbp);
    if(ret == 0)
    {
1138
        POSIX_PRE_RECORD();
1139
        posix_aio_tracker_add(aiocbp->aio_fildes, aiocbp);
1140
        POSIX_POST_RECORD();
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
    }

    return(ret);
}

ssize_t DARSHAN_DECL(aio_return)(struct aiocb *aiocbp)
{
    int ret;
    double tm2;
    struct posix_aio_tracker *tmp;
    int aligned_flag = 0;

    MAP_OR_FAIL(aio_return);

    ret = __real_aio_return(aiocbp);
    tm2 = darshan_core_wtime();

1158
    POSIX_PRE_RECORD();
1159
    tmp = posix_aio_tracker_del(aiocbp->aio_fildes, aiocbp);
1160
    if(tmp)
1161 1162 1163 1164 1165 1166
    {
        if((unsigned long)aiocbp->aio_buf % darshan_mem_alignment == 0)
            aligned_flag = 1;
        if(aiocbp->aio_lio_opcode == LIO_WRITE)
        {
            POSIX_RECORD_WRITE(ret, aiocbp->aio_fildes,
1167
                1, aiocbp->aio_offset, aligned_flag,
1168 1169 1170 1171 1172
                tmp->tm1, tm2);
        }
        else if(aiocbp->aio_lio_opcode == LIO_READ)
        {
            POSIX_RECORD_READ(ret, aiocbp->aio_fildes,
1173
                1, aiocbp->aio_offset, aligned_flag,
1174 1175 1176 1177
                tmp->tm1, tm2);
        }
        free(tmp);
    }
1178
    POSIX_POST_RECORD();
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194

    return(ret);
}

ssize_t DARSHAN_DECL(aio_return64)(struct aiocb64 *aiocbp)
{
    int ret;
    double tm2;
    struct posix_aio_tracker *tmp;
    int aligned_flag = 0;

    MAP_OR_FAIL(aio_return64);

    ret = __real_aio_return64(aiocbp);
    tm2 = darshan_core_wtime();

1195
    POSIX_PRE_RECORD();
1196
    tmp = posix_aio_tracker_del(aiocbp->aio_fildes, aiocbp);
1197
    if(tmp)
1198 1199 1200 1201 1202 1203
    {
        if((unsigned long)aiocbp->aio_buf % darshan_mem_alignment == 0)
            aligned_flag = 1;
        if(aiocbp->aio_lio_opcode == LIO_WRITE)
        {
            POSIX_RECORD_WRITE(ret, aiocbp->aio_fildes,
1204
                1, aiocbp->aio_offset, aligned_flag,
1205 1206 1207 1208 1209
                tmp->tm1, tm2);
        }
        else if(aiocbp->aio_lio_opcode == LIO_READ)
        {
            POSIX_RECORD_READ(ret, aiocbp->aio_fildes,
1210
                1, aiocbp->aio_offset, aligned_flag,
1211 1212 1213 1214
                tmp->tm1, tm2);
        }
        free(tmp);
    }
1215
    POSIX_POST_RECORD();
1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230

    return(ret);
}

int DARSHAN_DECL(lio_listio)(int mode, struct aiocb *const aiocb_list[],
    int nitems, struct sigevent *sevp)
{
    int ret;
    int i;

    MAP_OR_FAIL(lio_listio);

    ret = __real_lio_listio(mode, aiocb_list, nitems, sevp);
    if(ret == 0)
    {
1231
        POSIX_PRE_RECORD();
1232 1233 1234 1235
        for(i = 0; i < nitems; i++)
        {
            posix_aio_tracker_add(aiocb_list[i]->aio_fildes, aiocb_list[i]);
        }
1236
        POSIX_POST_RECORD();
1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
    }

    return(ret);
}

int DARSHAN_DECL(lio_listio64)(int mode, struct aiocb64 *const aiocb_list[],
    int nitems, struct sigevent *sevp)
{
    int ret;
    int i;

    MAP_OR_FAIL(lio_listio64);

    ret = __real_lio_listio64(mode, aiocb_list, nitems, sevp);
    if(ret == 0)
    {
1253
        POSIX_PRE_RECORD();
1254 1255 1256 1257
        for(i = 0; i < nitems; i++)
        {
            posix_aio_tracker_add(aiocb_list[i]->aio_fildes, aiocb_list[i]);
        }
1258
        POSIX_POST_RECORD();
1259 1260 1261 1262 1263
    }

    return(ret);
}

1264 1265 1266
/**********************************************************
 * Internal functions for manipulating POSIX module state *
 **********************************************************/
1267

1268
/* initialize internal POSIX module data structures and register with darshan-core */
1269
static void posix_runtime_initialize()
1270
{
1271
    int psx_buf_size;
1272

1273
    /* try and store a default number of records for this module */
1274 1275
    psx_buf_size = DARSHAN_DEF_MOD_REC_COUNT * sizeof(struct darshan_posix_file);

1276
    /* register the POSIX module with darshan core */
1277
    darshan_core_register_module(
1278
        DARSHAN_POSIX_MOD,
1279
        &posix_shutdown,
1280
        &psx_buf_size,
1281
        &my_rank,
1282
        &darshan_mem_alignment);
1283

1284 1285
    /* return if darshan-core does not provide enough module memory */
    if(psx_buf_size < sizeof(struct darshan_posix_file))
1286 1287
    {
        darshan_core_unregister_module(DARSHAN_POSIX_MOD);
1288
        return;
1289
    }
1290 1291 1292

    posix_runtime = malloc(sizeof(*posix_runtime));
    if(!posix_runtime)
1293 1294
    {
        darshan_core_unregister_module(DARSHAN_POSIX_MOD);
1295
        return;
1296
    }
1297 1298
    memset(posix_runtime, 0, sizeof(*posix_runtime));

1299
    /* check if DXT (Darshan extended tracing) should be enabled */
1300
    if (getenv("DXT_ENABLE_IO_TRACE")) {
1301 1302 1303
        enable_dxt_io_trace = 1;
    }

1304 1305
    return;
}
1306

1307
static struct posix_file_record_ref *posix_track_new_file_record(
1308
    darshan_record_id rec_id, const char *path)
1309
{
1310
    struct darshan_posix_file *file_rec = NULL;
1311
    struct posix_file_record_ref *rec_ref = NULL;
1312
    struct darshan_fs_info fs_info;
1313
    int ret;
1314

1315 1316
    rec_ref = malloc(sizeof(*rec_ref));
    if(!rec_ref)
1317
        return(NULL);
1318
    memset(rec_ref, 0, sizeof(*rec_ref));
1319

1320 1321 1322 1323
    /* add a reference to this file record based on record id */
    ret = darshan_add_record_ref(&(posix_runtime->rec_id_hash), &rec_id,
        sizeof(darshan_record_id), rec_ref);
    if(ret == 0)
1324
    {
1325
        free(rec_ref);
1326 1327 1328
        return(NULL);
    }

1329 1330 1331
    /* register the actual file record with darshan-core so it is persisted
     * in the log file
     */
1332 1333 1334 1335 1336
    file_rec = darshan_core_register_record(
        rec_id,
        path,
        DARSHAN_POSIX_MOD,
        sizeof(struct darshan_posix_file),
Shane Snyder's avatar
Shane Snyder committed
1337
        &fs_info);
1338

1339
    if(!file_rec)
1340
    {
1341 1342 1343 1344
        darshan_delete_record_ref(&(posix_runtime->rec_id_hash),
            &rec_id, sizeof(darshan_record_id));
        free(rec_ref);
        return(NULL);
1345 1346
    }

1347 1348 1349 1350
    /* registering this file record was successful, so initialize some fields */
    file_rec->base_rec.id = rec_id;
    file_rec->base_rec.rank = my_rank;
    file_rec->counters[POSIX_MEM_ALIGNMENT] = darshan_mem_alignment;
Shane Snyder's avatar
Shane Snyder committed
1351
    file_rec->counters[POSIX_FILE_ALIGNMENT] = fs_info.block_size;
1352 1353 1354 1355
#ifndef DARSHAN_WRAP_MMAP
    /* set invalid value here if MMAP instrumentation is disabled */
    file_rec->counters[POSIX_MMAPS] = -1;
#endif /* undefined DARSHAN_WRAP_MMAP */
1356
    rec_ref->fs_type = fs_info.fs_type;
1357
    rec_ref->file_rec = file_rec;
1358 1359
    posix_runtime->file_rec_count++;

1360 1361 1362
    return(rec_ref);
}

1363
/* finds the tracker structure for a given aio operation, removes it from
1364
 * the associated linked list for this file record, and returns a pointer.  
1365 1366 1367 1368 1369 1370
 *
 * returns NULL if aio operation not found
 */
static struct posix_aio_tracker* posix_aio_tracker_del(int fd, void *aiocbp)
{
    struct posix_aio_tracker *tracker = NULL, *iter, *tmp;
1371
    struct posix_file_record_ref *rec_ref;
1372

1373 1374
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
1375
    {
1376
        LL_FOREACH_SAFE(rec_ref->aio_list, iter, tmp)
1377
        {
1378
            if(iter->aiocbp == aiocbp)
1379
            {
1380
                LL_DELETE(rec_ref->aio_list, iter);
1381 1382 1383 1384 1385 1386 1387 1388 1389
                tracker = iter;
                break;
            }
        }
    }

    return(tracker);
}

1390 1391 1392 1393
/* adds a tracker for the given aio operation */
static void posix_aio_tracker_add(int fd, void *aiocbp)
{
    struct posix_aio_tracker* tracker;
1394
    struct posix_file_record_ref *rec_ref;
1395

1396 1397
    rec_ref = darshan_lookup_record_ref(posix_runtime->fd_hash, &fd, sizeof(int));
    if(rec_ref)
1398 1399
    {
        tracker = malloc(sizeof(*tracker));
1400
        if(tracker)
1401 1402 1403
        {
            tracker->tm1 = darshan_core_wtime();
            tracker->aiocbp = aiocbp;
1404
            LL_PREPEND(rec_ref->aio_list, tracker);
1405 1406 1407 1408 1409 1410
        }
    }

    return;
}