darshan-posix.c 14.4 KB
Newer Older
1
2
3
4
5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6
#include "darshan-runtime-config.h"
7
8
9
10
11
12
13
14
15
16
17
18
19
20
#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>
21
#include <libgen.h>
22
#include <limits.h>
Philip Carns's avatar
Philip Carns committed
23
#include <aio.h>
24
25
#define __USE_GNU
#include <pthread.h>
26
27

#include "darshan.h"
28
#include "uthash.h"
29

30
#ifndef HAVE_OFF64_T
31
32
typedef int64_t off64_t;
#endif
33
34
35
#ifndef HAVE_AIOCB64
#define aiocb64 aiocb
#endif
36

37
38
/* TODO these go where ? */

39
40
41
42
43
#define DARSHAN_FORWARD_DECL(name,ret,args) \
  extern ret __real_ ## name args;

#define DARSHAN_DECL(__name) __wrap_ ## __name

44
45
#define MAP_OR_FAIL(func)

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
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
102
103
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
/* TODO: where do these file record structs go? (some needed for darshan-util) */
/* TODO: DARSHAN_* OR CP_* */

#define POSIX_MOD_NAME "POSIX"

enum darshan_posix_indices
{
    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,
    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,               /* number of times switched between read and write */
    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 */
    CP_MAX_READ_TIME_SIZE,
    CP_MAX_WRITE_TIME_SIZE,
    /* buckets */
    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,
    /* buckets */
    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,
    /* counters */
    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,             /* the four most frequently appearing access sizes */
    CP_ACCESS2_ACCESS,
    CP_ACCESS3_ACCESS,
    CP_ACCESS4_ACCESS,
    CP_ACCESS1_COUNT,              /* count of each of the most frequent access sizes */
    CP_ACCESS2_COUNT,
    CP_ACCESS3_COUNT,
    CP_ACCESS4_COUNT,
    CP_DEVICE,                     /* device id reported by stat */
    CP_SIZE_AT_OPEN,
    CP_FASTEST_RANK,
    CP_FASTEST_RANK_BYTES,
    CP_SLOWEST_RANK,
    CP_SLOWEST_RANK_BYTES,

    CP_NUM_INDICES,
};
129

130
/* floating point statistics */
131
enum darshan_f_posix_indices
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
{
    /* NOTE: adjust cp_normalize_timestamps() function if any TIMESTAMPS are
     * added or modified in this list
     */
    CP_F_OPEN_TIMESTAMP = 0,    /* 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_MAX_READ_TIME,
    CP_F_MAX_WRITE_TIME,
    /* Total I/O and meta time consumed by fastest and slowest ranks, 
     * reported in either MPI or POSIX time depending on how the file 
     * was accessed.
     */
    CP_F_FASTEST_RANK_TIME,     
    CP_F_SLOWEST_RANK_TIME,
    CP_F_VARIANCE_RANK_TIME,
    CP_F_VARIANCE_RANK_BYTES,

    CP_F_NUM_INDICES,
};

struct darshan_posix_file
{
161
    darshan_file_id f_id;
162
    int64_t rank;
163
164
165
166
    int64_t counters[CP_NUM_INDICES];
    double fcounters[CP_F_NUM_INDICES];
};

167
struct posix_runtime_file
168
169
{
    struct darshan_posix_file file_record;
170
    UT_hash_handle hlink;
171
};
172

173
struct posix_runtime_file_ref
174
{
175
176
177
178
179
180
181
182
    struct posix_runtime_file* file;
    int fd;
    UT_hash_handle hlink;
};

struct posix_runtime
{
    struct posix_runtime_file* file_array;
183
184
    int file_array_size;
    int file_count;
185
186
    struct posix_runtime_file* file_hash;
    struct posix_runtime_file_ref* fd_hash;
187
188
};

189
static struct posix_runtime *posix_runtime = NULL;
190
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
191
192
static int my_rank = -1;
static int darshan_mem_alignment = 1;
193

194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

209
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
210
DARSHAN_FORWARD_DECL(close, int, (int fd));
211

212
static void posix_runtime_initialize(void);
213
static void posix_runtime_finalize(void);
214

215
216
217
218
static struct posix_runtime_file* posix_file_by_name(const char *name);
static struct posix_runtime_file* posix_file_by_name_setfd(const char* name, int fd);
static void posix_file_close_fd(int fd);

219
220
221
static void posix_prepare_for_shutdown(void);
static void posix_get_output_data(void **buffer, int size);

222
223
#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)
224

225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
#define POSIX_SET(__file, __counter, __value) do {\
    (__file)->file_record.counters[__counter] = __value; \
} while(0)

#define POSIX_F_SET(__file, __counter, __value) do {\
    (__file)->file_record.fcounters[__counter] = __value; \
} while(0)

#define POSIX_INC(__file, __counter, __value) do {\
    (__file)->file_record.counters[__counter] += __value; \
} while(0)

#define POSIX_F_INC(__file, __counter, __value) do {\
    (__file)->file_record.fcounters[__counter] += __value; \
} while(0)

#define POSIX_F_INC_NO_OVERLAP(__file, __tm1, __tm2, __last, __counter) do { \
    if(__tm1 > __last) \
        POSIX_F_INC(__file, __counter, (__tm2-__tm1)); \
    else \
        POSIX_F_INC(__file, __counter, (__tm2 - __last)); \
    if(__tm2 > __last) \
        __last = __tm2; \
} while(0)

#define POSIX_VALUE(__file, __counter) \
    ((__file)->file_record.counters[__counter])

#define POSIX_F_VALUE(__file, __counter) \
    ((__file)->file_record.fcounters[__counter])

#define POSIX_MAX(__file, __counter, __value) do {\
    if((__file)->file_record.counters[__counter] < __value) \
    { \
        (__file)->file_record.counters[__counter] = __value; \
    } \
} while(0)

263
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
264
    struct posix_runtime_file* file; \
265
266
267
268
269
270
271
272
273
    char* exclude; \
    int tmp_index = 0; \
    if(__ret < 0) break; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
274
    file = posix_file_by_name_setfd(__path, __ret); \
275
    if(!file) break; \
276
    file->file_record.rank = my_rank; \
277
    if(__mode) \
278
        POSIX_SET(file, CP_MODE, __mode); \
279
    if(__stream_flag)\
280
        POSIX_INC(file, CP_POSIX_FOPENS, 1); \
281
    else \
282
283
284
285
        POSIX_INC(file, CP_POSIX_OPENS, 1); \
    if(POSIX_F_VALUE(file, CP_F_OPEN_TIMESTAMP) == 0) \
        POSIX_F_SET(file, CP_F_OPEN_TIMESTAMP, __tm1); \
} while(0)
286

287
int DARSHAN_DECL(open)(const char *path, int flags, ...)
288
289
290
291
292
{
    int mode = 0;
    int ret;
    double tm1, tm2;

293
294
    MAP_OR_FAIL(open);

295
    if(flags & O_CREAT) 
296
297
298
299
300
301
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

302
        tm1 = darshan_core_wtime();
303
        ret = __real_open(path, flags, mode);
304
        tm2 = darshan_core_wtime();
305
306
307
    }
    else
    {
308
        tm1 = darshan_core_wtime();
309
        ret = __real_open(path, flags);
310
        tm2 = darshan_core_wtime();
311
312
    }

313
314
315
    POSIX_LOCK();
    posix_runtime_initialize();

316
    POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
317

318
    POSIX_UNLOCK();
319
320
321
322

    return(ret);
}

323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
int DARSHAN_DECL(close)(int fd)
{
    struct darshan_file_runtime* file;
    int tmp_fd = fd;
    double tm1, tm2;
    int ret;

    MAP_OR_FAIL(close);

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

    POSIX_LOCK();
    posix_runtime_initialize();
338

339
    posix_file_close_fd(tmp_fd);
340
341
342
343
344
345
346

    POSIX_UNLOCK();    

    return(ret);
}

/* ***************************************************** */
347

348
static void posix_runtime_initialize()
349
{
350
351
352
353
354
355
356
357
358
    char *alignstr;
    int tmpval;
    int ret;
    int mem_limit;
    struct darshan_module_funcs posix_mod_fns =
    {
        .prepare_for_shutdown = &posix_prepare_for_shutdown,
        .get_output_data = &posix_get_output_data,
    };
359

360
    if(posix_runtime)
361
        return;
362

363
#if 0
364
365
366
367
368
    /* set the memory alignment according to config or environment variables */
    #if (__CP_MEM_ALIGNMENT < 1)
        #error Darshan must be configured with a positive value for --with-mem-align
    #endif
    alignstr = getenv("DARSHAN_MEMALIGN");
369
    if(alignstr)
370
371
372
373
374
375
376
377
    {
        ret = sscanf(alignstr, "%d", &tmpval);
        /* silently ignore if the env variable is set poorly */
        if(ret == 1 && tmpval > 0)
        {
            darshan_mem_alignment = tmpval;
        }
    }
378
    else
379
380
381
    {
        darshan_mem_alignment = __CP_MEM_ALIGNMENT;
    }
382

383
    /* avoid floating point errors on faulty input */
384
    if(darshan_mem_alignment < 1)
385
386
387
    {
        darshan_mem_alignment = 1;
    }
388
#endif
389

390
    posix_runtime = malloc(sizeof(*posix_runtime));
391
    if(!posix_runtime)
392
        return;
393
    memset(posix_runtime, 0, sizeof(*posix_runtime));
394

395
396
397
398
399
400
    /* register the posix module with darshan core */
    darshan_core_register_module(
        POSIX_MOD_NAME,
        &posix_mod_fns,
        &mem_limit);

401
    /* set maximum number of file records according to max memory limit */
402
403
    /* NOTE: maximum number of records is based on the size of a posix file record */
    posix_runtime->file_array_size = mem_limit / sizeof(struct darshan_posix_file);
404

405
    /* allocate array of runtime file records */
406
    posix_runtime->file_array = malloc(sizeof(struct posix_runtime_file) *
407
408
                                       posix_runtime->file_array_size);
    if(!posix_runtime->file_array)
409
    {
410
        posix_runtime->file_array_size = 0;
411
412
        return;
    }
413
    memset(posix_runtime->file_array, 0, sizeof(struct posix_runtime_file) *
414
           posix_runtime->file_array_size);
415

416
    return;
417
418
}

419
static struct posix_runtime_file* posix_file_by_name(const char *name)
420
{
421
    struct posix_runtime_file *file = NULL;
422
    char *newname = NULL;
423
    darshan_file_id file_id;
424

425
    if(!posix_runtime)
426
        return(NULL);
427

428
    newname = darshan_clean_file_path(name);
429
    if(!newname)
430
        newname = (char*)name;
431

432
433
434
435
436
    /* get a unique id for this file from darshan core */
    darshan_core_lookup_id(
        (void*)newname,
        strlen(newname),
        1,
437
        &file_id);
438

439
    /* search the hash table for this file record, and return if found */
440
441
    HASH_FIND(hlink, posix_runtime->file_hash, &file_id, sizeof(darshan_file_id), file);
    if(file)
442
    {
443
        if(newname != name)
444
            free(newname);
445
        return(file);
446
447
448
    }

    /* no existing record, assign a new file record from the global array */
449
450
    file = &posix_runtime->file_array[posix_runtime->file_count];
    file->file_record.f_id = file_id;
451

452
    /* add new record to file hash table */
453
    HASH_ADD(hlink, posix_runtime->file_hash, file_record.f_id, sizeof(darshan_file_id), file);
454

455
    posix_runtime->file_count++;
456
457

    if(newname != name)
458
        free(newname);
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
    return(file);
}

static struct posix_runtime_file* posix_file_by_name_setfd(const char* name, int fd)
{
    struct posix_runtime_file* file;
    struct posix_runtime_file_ref* ref;

    if(!posix_runtime)
        return(NULL);

    /* find file record by name first */
    file = posix_file_by_name(name);

    if(!file)
        return(NULL);

    /* search hash table for existing file ref for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
    if(ref)
    {
        /* we have a reference.  Make sure it points to the correct file
         * and return it
         */
        ref->file = file;
        return(file);
    }

    /* if we hit this point, then we don't have a reference for this fd
     * in the table yet.  Add it.
     */
    ref = malloc(sizeof(*ref));
    if(!ref)
        return(NULL);
    memset(ref, 0, sizeof(*ref));
    ref->file = file;
    ref->fd = fd;    

    HASH_ADD(hlink, posix_runtime->fd_hash, fd, sizeof(int), ref);

    return(file);
}

static void posix_file_close_fd(int fd)
{
    struct posix_runtime_file_ref *ref;

    if(!posix_runtime)
        return;

    /* search hash table for this fd */
    HASH_FIND(hlink, posix_runtime->fd_hash, &fd, sizeof(int), ref);
    if (ref)
    {
        /* we have a reference, delete it */
        HASH_DELETE(hlink, posix_runtime->fd_hash, ref);
        free(ref);
    }

    return;
519
520
}

521
522
/* ***************************************************** */

523
static void posix_prepare_for_shutdown()
524
525
{

526
    return;
527
528
}

529
static void posix_get_output_data(void **buffer, int size)
530
{
531
532

    return;
533
534
}

535
536
537
538
539
540
541
542
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ts=8 sts=4 sw=4 expandtab
 */