aboutsummaryrefslogtreecommitdiff
path: root/rt/aio_misc.c
blob: fb5fbf6985bafdace9eed89e9ef19c51c3c2ac96 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
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
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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
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
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
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
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
/* Handle general operations.
   Copyright (C) 1997, 1998, 1999 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public License as
   published by the Free Software Foundation; either version 2 of the
   License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public
   License along with the GNU C Library; see the file COPYING.LIB.  If not,
   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
   Boston, MA 02111-1307, USA.  */

#include <aio.h>
#include <errno.h>
#include <limits.h>
#include <pthread.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/stat.h>

#include "aio_misc.h"

/* Pool of request list entries.  */
static struct requestlist **pool;

/* Number of total and allocated pool entries.  */
static size_t pool_tab_size;
static size_t pool_size;

/* We implement a two dimensional array but allocate each row separately.
   The macro below determines how many entries should be used per row.
   It should better be a power of two.  */
#define ENTRIES_PER_ROW	16

/* The row table is incremented in units of this.  */
#define ROW_STEP	8

/* List of available entries.  */
static struct requestlist *freelist;

/* List of request waiting to be processed.  */
static struct requestlist *runlist;

/* Structure list of all currently processed requests.  */
static struct requestlist *requests;

/* Number of threads currently running.  */
static int nthreads;


/* These are the values used to optimize the use of AIO.  The user can
   overwrite them by using the `aio_init' function.  */
static struct aioinit optim =
{
  20,	/* int aio_threads;	Maximal number of threads.  */
  256,	/* int aio_num;		Number of expected simultanious requests. */
  0,
  0,
  0,
  0,
  { 0, }
};


/* Since the list is global we need a mutex protecting it.  */
pthread_mutex_t __aio_requests_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;


/* Functions to handle request list pool.  */
static struct requestlist *
get_elem (void)
{
  struct requestlist *result;

  if (freelist == NULL)
    {
      struct requestlist *new_row;
      size_t new_size;

      /* Compute new size.  */
      new_size = pool_size ? pool_size + ENTRIES_PER_ROW : optim.aio_num;

      if ((new_size / ENTRIES_PER_ROW) >= pool_tab_size)
	{
	  size_t new_tab_size = new_size / ENTRIES_PER_ROW;
	  struct requestlist **new_tab;

	  new_tab = (struct requestlist **)
	    realloc (pool, (new_tab_size * sizeof (struct requestlist *)));

	  if (new_tab == NULL)
	    return NULL;

	  pool_tab_size = new_tab_size;
	  pool = new_tab;
	}

      if (pool_size == 0)
	{
	  size_t cnt;

	  new_row = (struct requestlist *)
	    calloc (new_size, sizeof (struct requestlist));

	  if (new_row == NULL)
	    return NULL;

	  for (cnt = 0; cnt < new_size / ENTRIES_PER_ROW; ++cnt)
	    pool[cnt] = &new_row[cnt * ENTRIES_PER_ROW];
	}
      else
	{
	  /* Allocat one new row.  */
	  new_row = (struct requestlist *)
	    calloc (ENTRIES_PER_ROW, sizeof (struct requestlist));
	  if (new_row == NULL)
	    return NULL;

	  pool[new_size / ENTRIES_PER_ROW] = new_row;
	}

      /* Put all the new entries in the freelist.  */
      do
	{
	  new_row->next_prio = freelist;
	  freelist = new_row++;
	}
      while (++pool_size < new_size);
    }

  result = freelist;
  freelist = freelist->next_prio;

  return result;
}


void
internal_function
__aio_free_request (struct requestlist *elem)
{
  elem->running = no;
  elem->next_prio = freelist;
  freelist = elem;
}


struct requestlist *
internal_function
__aio_find_req (aiocb_union *elem)
{
  struct requestlist *runp = requests;
  int fildes = elem->aiocb.aio_fildes;

  while (runp != NULL && runp->aiocbp->aiocb.aio_fildes < fildes)
    runp = runp->next_fd;

  if (runp != NULL)
    {
      if (runp->aiocbp->aiocb.aio_fildes != fildes)
	runp = NULL;
      else
	while (runp != NULL && runp->aiocbp != elem)
	  runp = runp->next_prio;
    }

  return runp;
}


struct requestlist *
internal_function
__aio_find_req_fd (int fildes)
{
  struct requestlist *runp = requests;

  while (runp != NULL && runp->aiocbp->aiocb.aio_fildes < fildes)
    runp = runp->next_fd;

  return (runp != NULL && runp->aiocbp->aiocb.aio_fildes == fildes
	  ? runp : NULL);
}


/* The thread handler.  */
static void *handle_fildes_io (void *arg);


/* User optimization.  */
void
__aio_init (const struct aioinit *init)
{
  /* Get the mutex.  */
  pthread_mutex_lock (&__aio_requests_mutex);

  /* Only allow writing new values if the table is not yet allocated.  */
  if (pool == NULL)
    {
      optim.aio_threads = init->aio_threads < 1 ? 1 : init->aio_threads;
      optim.aio_num = (init->aio_num < ENTRIES_PER_ROW
		       ? ENTRIES_PER_ROW
		       : init->aio_num & ~ENTRIES_PER_ROW);
    }

  /* Release the mutex.  */
  pthread_mutex_unlock (&__aio_requests_mutex);
}
weak_alias (__aio_init, aio_init)


/* The main function of the async I/O handling.  It enqueues requests
   and if necessary starts and handles threads.  */
struct requestlist *
internal_function
__aio_enqueue_request (aiocb_union *aiocbp, int operation)
{
  int result = 0;
  int policy, prio;
  struct sched_param param;
  struct requestlist *last, *runp, *newp;
  int running = no;

  if (aiocbp->aiocb.aio_reqprio < 0
      || aiocbp->aiocb.aio_reqprio > AIO_PRIO_DELTA_MAX)
    {
      /* Invalid priority value.  */
      __set_errno (EINVAL);
      aiocbp->aiocb.__error_code = EINVAL;
      aiocbp->aiocb.__return_value = -1;
      return NULL;
    }

  /* Compute priority for this request.  */
  pthread_getschedparam (pthread_self (), &policy, &param);
  prio = param.sched_priority - aiocbp->aiocb.aio_reqprio;

  /* Get the mutex.  */
  pthread_mutex_lock (&__aio_requests_mutex);

  last = NULL;
  runp = requests;
  /* First look whether the current file descriptor is currently
     worked with.  */
  while (runp != NULL
	 && runp->aiocbp->aiocb.aio_fildes < aiocbp->aiocb.aio_fildes)
    {
      last = runp;
      runp = runp->next_fd;
    }

  /* Get a new element for the waiting list.  */
  newp = get_elem ();
  if (newp == NULL)
    {
      pthread_mutex_unlock (&__aio_requests_mutex);
      __set_errno (EAGAIN);
      return NULL;
    }
  newp->aiocbp = aiocbp;
  newp->waiting = NULL;

  aiocbp->aiocb.__abs_prio = prio;
  aiocbp->aiocb.__policy = policy;
  aiocbp->aiocb.aio_lio_opcode = operation;
  aiocbp->aiocb.__error_code = EINPROGRESS;
  aiocbp->aiocb.__return_value = 0;

  if (runp != NULL
      && runp->aiocbp->aiocb.aio_fildes == aiocbp->aiocb.aio_fildes)
    {
      /* The current file descriptor is worked on.  It makes no sense
	 to start another thread since this new thread would fight
	 with the running thread for the resources.  But we also cannot
	 say that the thread processing this desriptor shall immediately
	 after finishing the current job process this request if there
	 are other threads in the running queue which have a higher
	 priority.  */

      /* Simply enqueue it after the running one according to the
	 priority.  */
      while (runp->next_prio != NULL
	     && runp->next_prio->aiocbp->aiocb.__abs_prio >= prio)
	runp = runp->next_prio;

      newp->next_prio = runp->next_prio;
      runp->next_prio = newp;

      running = queued;
    }
  else
    {
      /* Enqueue this request for a new descriptor.  */
      if (last == NULL)
	{
	  newp->last_fd = NULL;
	  newp->next_fd = requests;
	  if (requests != NULL)
	    requests->last_fd = newp;
	  requests = newp;
	}
      else
	{
	  newp->next_fd = last->next_fd;
	  newp->last_fd = last;
	  last->next_fd = newp;
	  if (newp->next_fd != NULL)
	    newp->next_fd->last_fd = newp;
	}

      newp->next_prio = NULL;
    }

  if (running == no)
    {
      /* We try to create a new thread for this file descriptor.  The
	 function which gets called will handle all available requests
	 for this descriptor and when all are processed it will
	 terminate.

	 If no new thread can be created or if the specified limit of
	 threads for AIO is reached we queue the request.  */

      /* See if we can create a thread.  */
      if (nthreads < optim.aio_threads)
	{
	  pthread_t thid;
	  pthread_attr_t attr;

	  /* Make sure the thread is created detached.  */
	  pthread_attr_init (&attr);
	  pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);

	  /* Now try to start a thread.  */
	  if (pthread_create (&thid, &attr, handle_fildes_io, newp) == 0)
	    {
	      /* We managed to enqueue the request.  All errors which can
		 happen now can be recognized by calls to `aio_return' and
		 `aio_error'.  */
	      running = allocated;
	      ++nthreads;
	    }
	  else if (nthreads == 0)
	    /* We cannot create a thread in the moment and there is
	       also no thread running.  This is a problem.  `errno' is
	       set to EAGAIN if this is only a temporary problem.  */
	    result = -1;
	}
    }

  /* Enqueue the request in the run queue if it is not yet running.  */
  if (running < yes && result == 0)
    {
      if (runlist == NULL || runlist->aiocbp->aiocb.__abs_prio < prio)
	{
	  newp->next_run = runlist;
	  runlist = newp;
	}
      else
	{
	  runp = runlist;

	  while (runp->next_run != NULL
		 && runp->next_run->aiocbp->aiocb.__abs_prio >= prio)
	    runp = runp->next_run;

	  newp->next_run = runp->next_run;
	  runp->next_run = newp;
	}
    }

  if (result == 0)
    newp->running = running;
  else
    {
      /* Something went wrong.  */
      __aio_free_request (newp);
      newp = NULL;
    }

  /* Release the mutex.  */
  pthread_mutex_unlock (&__aio_requests_mutex);

  return newp;
}


static void *
handle_fildes_io (void *arg)
{
  pthread_t self = pthread_self ();
  struct sched_param param;
  struct requestlist *runp = (struct requestlist *) arg;
  aiocb_union *aiocbp;
  int policy;
  int fildes;

  pthread_getschedparam (self, &policy, &param);

  do
    {
      /* Update our variables.  */
      aiocbp = runp->aiocbp;
      fildes = aiocbp->aiocb.aio_fildes;

      /* Change the priority to the requested value (if necessary).  */
      if (aiocbp->aiocb.__abs_prio != param.sched_priority
	  || aiocbp->aiocb.__policy != policy)
	{
	  param.sched_priority = aiocbp->aiocb.__abs_prio;
	  policy = aiocbp->aiocb.__policy;
	  pthread_setschedparam (self, policy, &param);
	}

      /* Process request pointed to by RUNP.  We must not be disturbed
	 by signals.  */
      if ((aiocbp->aiocb.aio_lio_opcode & 127) == LIO_READ)
	{
	  if (aiocbp->aiocb.aio_lio_opcode & 128)
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (__pread64 (fildes,
					     (void *) aiocbp->aiocb64.aio_buf,
					     aiocbp->aiocb64.aio_nbytes,
					     aiocbp->aiocb64.aio_offset));
	  else
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (pread (fildes,
					 (void *) aiocbp->aiocb.aio_buf,
					 aiocbp->aiocb.aio_nbytes,
					 aiocbp->aiocb.aio_offset));

	  if (aiocbp->aiocb.__return_value == -1 && errno == ESPIPE)
	    /* The Linux kernel is different from others.  It returns
	       ESPIPE if using pread on a socket.  Other platforms
	       simply ignore the offset parameter and behave like
	       read.  */
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (read (fildes,
					(void *) aiocbp->aiocb64.aio_buf,
					aiocbp->aiocb64.aio_nbytes));
	}
      else if ((aiocbp->aiocb.aio_lio_opcode & 127) == LIO_WRITE)
	{
	  if (aiocbp->aiocb.aio_lio_opcode & 128)
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (__pwrite64 (fildes,
					      (const void *) aiocbp->aiocb64.aio_buf,
					      aiocbp->aiocb64.aio_nbytes,
					      aiocbp->aiocb64.aio_offset));
	  else
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (pwrite (fildes,
					  (const void *) aiocbp->aiocb.aio_buf,
					  aiocbp->aiocb.aio_nbytes,
					  aiocbp->aiocb.aio_offset));

	  if (aiocbp->aiocb.__return_value == -1 && errno == ESPIPE)
	    /* The Linux kernel is different from others.  It returns
	       ESPIPE if using pwrite on a socket.  Other platforms
	       simply ignore the offset parameter and behave like
	       write.  */
	    aiocbp->aiocb.__return_value =
	      TEMP_FAILURE_RETRY (write (fildes,
					 (void *) aiocbp->aiocb64.aio_buf,
					 aiocbp->aiocb64.aio_nbytes));
	}
      else if (aiocbp->aiocb.aio_lio_opcode == LIO_DSYNC)
	aiocbp->aiocb.__return_value = TEMP_FAILURE_RETRY (fdatasync (fildes));
      else if (aiocbp->aiocb.aio_lio_opcode == LIO_SYNC)
	aiocbp->aiocb.__return_value = TEMP_FAILURE_RETRY (fsync (fildes));
      else
	{
	  /* This is an invalid opcode.  */
	  aiocbp->aiocb.__return_value = -1;
	  __set_errno (EINVAL);
	}

      /* Get the mutex.  */
      pthread_mutex_lock (&__aio_requests_mutex);

      if (aiocbp->aiocb.__return_value == -1)
	aiocbp->aiocb.__error_code = errno;
      else
	aiocbp->aiocb.__error_code = 0;

      /* Send the signal to notify about finished processing of the
	 request.  */
      __aio_notify (runp);

      /* Now dequeue the current request.  */
      if (runp->next_prio == NULL)
	{
	  /* No outstanding request for this descriptor.  Remove this
	     descriptor from the list.  */
	  if (runp->next_fd != NULL)
	    runp->next_fd->last_fd = runp->last_fd;
	  if (runp->last_fd != NULL)
	    runp->last_fd->next_fd = runp->next_fd;
	  else
	    requests = runp->next_fd;
	}
      else
	{
	  runp->next_prio->last_fd = runp->last_fd;
	  runp->next_prio->next_fd = runp->next_fd;
	  runp->next_prio->running = yes;
	  if (runp->next_fd != NULL)
	    runp->next_fd->last_fd = runp->next_prio;
	  if (runp->last_fd != NULL)
	    runp->last_fd->next_fd = runp->next_prio;
	  else
	    requests = runp->next_prio;
	}

      /* Free the old element.  */
      __aio_free_request (runp);

      runp = runlist;
      if (runp != NULL)
	{
	  /* We must not run requests which are not marked `running'.  */
	  if (runp->running == yes)
	    runlist = runp->next_run;
	  else
	    {
	      struct requestlist *old;

	      do
		{
		  old = runp;
		  runp = runp->next_run;
		}
	      while (runp != NULL && runp->running != yes);

	      if (runp != NULL)
		old->next_run = runp->next_run;
	    }
	}

      /* If no request to work on we will stop the thread.  */
      if (runp == NULL)
	--nthreads;
      else
	runp->running = allocated;

      /* Release the mutex.  */
      pthread_mutex_unlock (&__aio_requests_mutex);
    }
  while (runp != NULL);

  pthread_exit (NULL);
}


/* Free allocated resources.  */
static void
__attribute__ ((unused))
free_res (void)
{
  size_t row;

  /* The first block of rows as specified in OPTIM is allocated in
     one chunk.  */
  free (pool[0]);

  for (row = optim.aio_num / ENTRIES_PER_ROW; row < pool_tab_size; ++row)
    free (pool[row]);

  free (pool);
}

text_set_element (__libc_subfreeres, free_res);