man7/aio.7

   1 '\" t
   2 .\" Copyright (c) 2010 by Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  16 .\" responsibility for errors or omissions, or for damages resulting from
  17 .\" the use of the information contained herein.  The author(s) may not
  18 .\" have taken the same level of care in the production of this manual,
  19 .\" which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .TH AIO 7 2015-05-07 "Linux" "Linux Programmer's Manual"
  27 .SH NAME
  28 aio \- POSIX asynchronous I/O overview
  29 .SH DESCRIPTION
  30 The POSIX asynchronous I/O (AIO) interface allows applications
  31 to initiate one or more I/O operations that are performed
  32 asynchronously (i.e., in the background).
  33 The application can elect to be notified of completion of
  34 the I/O operation in a variety of ways:
  35 by delivery of a signal, by instantiation of a thread,
  36 or no notification at all.
  37
  38 The POSIX AIO interface consists of the following functions:
  39 .TP 16
  40 .BR aio_read (3)
  41 Enqueue a read request.
  42 This is the asynchronous analog of
  43 .BR read (2).
  44 .TP
  45 .BR aio_write (3)
  46 Enqueue a write request.
  47 This is the asynchronous analog of
  48 .BR write (2).
  49 .TP
  50 .BR aio_fsync (3)
  51 Enqueue a sync request for the I/O operations on a file descriptor.
  52 This is the asynchronous analog of
  53 .BR fsync (2)
  54 and
  55 .BR fdatasync (2).
  56 .TP
  57 .BR aio_error (3)
  58 Obtain the error status of an enqueued I/O request.
  59 .TP
  60 .BR aio_return (3)
  61 Obtain the return status of a completed I/O request.
  62 .TP
  63 .BR aio_suspend (3)
  64 Suspend the caller until one or more of a specified set of
  65 I/O requests completes.
  66 .TP
  67 .BR aio_cancel (3)
  68 Attempt to cancel outstanding I/O requests on a specified
  69 file descriptor.
  70 .TP
  71 .BR lio_listio (3)
  72 Enqueue multiple I/O requests using a single function call.
  73 .PP
  74 The
  75 .I aiocb
  76 ("asynchronous I/O control block") structure defines
  77 parameters that control an I/O operation.
  78 An argument of this type is employed with all of the functions listed above.
  79 This structure has the following form:
  80 .PP
  81 .in +4n
  82 .nf
  83 #include <aiocb.h>
  84
  85 struct aiocb {
  86     /* The order of these fields is implementation-dependent */
  87
  88     int             aio_fildes;     /* File descriptor */
  89     off_t           aio_offset;     /* File offset */
  90     volatile void  *aio_buf;        /* Location of buffer */
  91     size_t          aio_nbytes;     /* Length of transfer */
  92     int             aio_reqprio;    /* Request priority */
  93     struct sigevent aio_sigevent;   /* Notification method */
  94     int             aio_lio_opcode; /* Operation to be performed;
  95                                        lio_listio() only */
  96
  97     /* Various implementation-internal fields not shown */
  98 };
  99
 100 /* Operation codes for \(aqaio_lio_opcode\(aq: */
 101
 102 enum { LIO_READ, LIO_WRITE, LIO_NOP };
 103
 104 .fi
 105 .in
 106 The fields of this structure are as follows:
 107 .TP 16
 108 .I aio_filedes
 109 The file descriptor on which the I/O operation is to be performed.
 110 .TP
 111 .I aio_offset
 112 This is the file offset at which the I/O operation is to be performed.
 113 .TP
 114 .I aio_buf
 115 This is the buffer used to transfer data for a read or write operation.
 116 .TP
 117 .I aio_nbytes
 118 This is the size of the buffer pointed to by
 119 .IR aio_buf .
 120 .TP
 121 .I aio_reqprio
 122 This field specifies a value that is subtracted
 123 from the calling thread's real-time priority in order to
 124 determine the priority for execution of this I/O request (see
 125 .BR pthread_setschedparam (3)).
 126 The specified value must be between 0 and the value returned by
 127 .IR sysconf(_SC_AIO_PRIO_DELTA_MAX) .
 128 This field is ignored for file synchronization operations.
 129 .TP
 130 .I aio_sigevent
 131 This field is a structure that specifies how the caller is
 132 to be notified when the asynchronous I/O operation completes.
 133 Possible values for
 134 .IR aio_sigevent.sigev_notify
 135 are
 136 .BR SIGEV_NONE ,
 137 .BR SIGEV_SIGNAL ,
 138 and
 139 .BR SIGEV_THREAD .
 140 See
 141 .BR sigevent (7)
 142 for further details.
 143 .TP
 144 .I aio_lio_opcode
 145 The type of operation to be performed; used only for
 146 .BR lio_listio (3).
 147 .PP
 148 In addition to the standard functions listed above,
 149 the GNU C library provides the following extension to the POSIX AIO API:
 150 .TP 16
 151 .BR aio_init (3)
 152 Set parameters for tuning the behavior of the glibc POSIX AIO implementation.
 153 .SH ERRORS
 154 .TP
 155 .B EINVAL
 156 The
 157 .I aio_reqprio
 158 field of the
 159 .I aiocb
 160 structure was less than 0,
 161 or was greater than the limit returned by the call
 162 .IR sysconf(_SC_AIO_PRIO_DELTA_MAX) .
 163 .SH VERSIONS
 164 The POSIX AIO interfaces are provided by glibc since version 2.1.
 165 .SH CONFORMING TO
 166 POSIX.1-2001, POSIX.1-2008.
 167 .SH NOTES
 168 It is a good idea to zero out the control block buffer before use (see
 169 .BR memset (3)).
 170 The control block buffer and the buffer pointed to by
 171 .I aio_buf
 172 must not be changed while the I/O operation is in progress.
 173 These buffers must remain valid until the I/O operation completes.
 174
 175 Simultaneous asynchronous read or write operations using the same
 176 .I aiocb
 177 structure yield undefined results.
 178
 179 The current Linux POSIX AIO implementation is provided in user space by glibc.
 180 This has a number of limitations, most notably that maintaining multiple
 181 threads to perform I/O operations is expensive and scales poorly.
 182 Work has been in progress for some time on a kernel
 183 state-machine-based implementation of asynchronous I/O
 184 (see
 185 .BR io_submit (2),
 186 .BR io_setup (2),
 187 .BR io_cancel (2),
 188 .BR io_destroy (2),
 189 .BR io_getevents (2)),
 190 but this implementation hasn't yet matured to the point where
 191 the POSIX AIO implementation can be completely
 192 reimplemented using the kernel system calls.
 193 .\" http://lse.sourceforge.net/io/aio.html
 194 .\" http://lse.sourceforge.net/io/aionotes.txt
 195 .\" http://lwn.net/Articles/148755/
 196 .SH EXAMPLE
 197 The program below opens each of the files named in its command-line
 198 arguments and queues a request on the resulting file descriptor using
 199 .BR aio_read (3).
 200 The program then loops,
 201 periodically monitoring each of the I/O operations
 202 that is still in progress using
 203 .BR aio_error (3).
 204 Each of the I/O requests is set up to provide notification by delivery
 205 of a signal.
 206 After all I/O requests have completed,
 207 the program retrieves their status using
 208 .BR aio_return (3).
 209
 210 The
 211 .B SIGQUIT
 212 signal (generated by typing control-\\) causes the program to request
 213 cancellation of each of the outstanding requests using
 214 .BR aio_cancel (3).
 215
 216 Here is an example of what we might see when running this program.
 217 In this example, the program queues two requests to standard input,
 218 and these are satisfied by two lines of input containing
 219 "abc" and "x".
 220
 221 .in +4n
 222 .nf
 223 $ \fB./a.out /dev/stdin /dev/stdin\fP
 224 opened /dev/stdin on descriptor 3
 225 opened /dev/stdin on descriptor 4
 226 aio_error():
 227     for request 0 (descriptor 3): In progress
 228     for request 1 (descriptor 4): In progress
 229 \fBabc\fP
 230 I/O completion signal received
 231 aio_error():
 232     for request 0 (descriptor 3): I/O succeeded
 233     for request 1 (descriptor 4): In progress
 234 aio_error():
 235     for request 1 (descriptor 4): In progress
 236 \fBx\fP
 237 I/O completion signal received
 238 aio_error():
 239     for request 1 (descriptor 4): I/O succeeded
 240 All I/O requests completed
 241 aio_return():
 242     for request 0 (descriptor 3): 4
 243     for request 1 (descriptor 4): 2
 244 .fi
 245 .in
 246 .SS Program source
 247 \&
 248 .nf
 249 #include <stdlib.h>
 250 #include <unistd.h>
 251 #include <stdio.h>
 252 #include <errno.h>
 253 #include <aio.h>
 254 #include <signal.h>
 255
 256 #define BUF_SIZE 20     /* Size of buffers for read operations */
 257
 258 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); } while (0)
 259
 260 #define errMsg(msg)  do { perror(msg); } while (0)
 261
 262 struct ioRequest {      /* Application\-defined structure for tracking
 263                            I/O requests */
 264     int           reqNum;
 265     int           status;
 266     struct aiocb *aiocbp;
 267 };
 268
 269 static volatile sig_atomic_t gotSIGQUIT = 0;
 270                         /* On delivery of SIGQUIT, we attempt to
 271                            cancel all outstanding I/O requests */
 272
 273 static void             /* Handler for SIGQUIT */
 274 quitHandler(int sig)
 275 {
 276     gotSIGQUIT = 1;
 277 }
 278
 279 #define IO_SIGNAL SIGUSR1   /* Signal used to notify I/O completion */
 280
 281 static void                 /* Handler for I/O completion signal */
 282 aioSigHandler(int sig, siginfo_t *si, void *ucontext)
 283 {
 284     write(STDOUT_FILENO, "I/O completion signal received\\n", 31);
 285
 286     /* The corresponding ioRequest structure would be available as
 287            struct ioRequest *ioReq = si\->si_value.sival_ptr;
 288        and the file descriptor would then be available via
 289            ioReq\->aiocbp\->aio_fildes */
 290 }
 291
 292 int
 293 main(int argc, char *argv[])
 294 {
 295     struct ioRequest *ioList;
 296     struct aiocb *aiocbList;
 297     struct sigaction sa;
 298     int s, j;
 299     int numReqs;        /* Total number of queued I/O requests */
 300     int openReqs;       /* Number of I/O requests still in progress */
 301
 302     if (argc < 2) {
 303         fprintf(stderr, "Usage: %s <pathname> <pathname>...\\n",
 304                 argv[0]);
 305         exit(EXIT_FAILURE);
 306     }
 307
 308     numReqs = argc \- 1;
 309
 310     /* Allocate our arrays */
 311
 312     ioList = calloc(numReqs, sizeof(struct ioRequest));
 313     if (ioList == NULL)
 314         errExit("calloc");
 315
 316     aiocbList = calloc(numReqs, sizeof(struct aiocb));
 317     if (aiocbList == NULL)
 318         errExit("calloc");
 319
 320     /* Establish handlers for SIGQUIT and the I/O completion signal */
 321
 322     sa.sa_flags = SA_RESTART;
 323     sigemptyset(&sa.sa_mask);
 324
 325     sa.sa_handler = quitHandler;
 326     if (sigaction(SIGQUIT, &sa, NULL) == \-1)
 327         errExit("sigaction");
 328
 329     sa.sa_flags = SA_RESTART | SA_SIGINFO;
 330     sa.sa_sigaction = aioSigHandler;
 331     if (sigaction(IO_SIGNAL, &sa, NULL) == \-1)
 332         errExit("sigaction");
 333
 334     /* Open each file specified on the command line, and queue
 335        a read request on the resulting file descriptor */
 336
 337     for (j = 0; j < numReqs; j++) {
 338         ioList[j].reqNum = j;
 339         ioList[j].status = EINPROGRESS;
 340         ioList[j].aiocbp = &aiocbList[j];
 341
 342         ioList[j].aiocbp\->aio_fildes = open(argv[j + 1], O_RDONLY);
 343         if (ioList[j].aiocbp\->aio_fildes == \-1)
 344             errExit("open");
 345         printf("opened %s on descriptor %d\\n", argv[j + 1],
 346                 ioList[j].aiocbp\->aio_fildes);
 347
 348         ioList[j].aiocbp\->aio_buf = malloc(BUF_SIZE);
 349         if (ioList[j].aiocbp\->aio_buf == NULL)
 350             errExit("malloc");
 351
 352         ioList[j].aiocbp\->aio_nbytes = BUF_SIZE;
 353         ioList[j].aiocbp\->aio_reqprio = 0;
 354         ioList[j].aiocbp\->aio_offset = 0;
 355         ioList[j].aiocbp\->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 356         ioList[j].aiocbp\->aio_sigevent.sigev_signo = IO_SIGNAL;
 357         ioList[j].aiocbp\->aio_sigevent.sigev_value.sival_ptr =
 358                                 &ioList[j];
 359
 360         s = aio_read(ioList[j].aiocbp);
 361         if (s == \-1)
 362             errExit("aio_read");
 363     }
 364
 365     openReqs = numReqs;
 366
 367     /* Loop, monitoring status of I/O requests */
 368
 369     while (openReqs > 0) {
 370         sleep(3);       /* Delay between each monitoring step */
 371
 372         if (gotSIGQUIT) {
 373
 374             /* On receipt of SIGQUIT, attempt to cancel each of the
 375                outstanding I/O requests, and display status returned
 376                from the cancellation requests */
 377
 378             printf("got SIGQUIT; canceling I/O requests: \\n");
 379
 380             for (j = 0; j < numReqs; j++) {
 381                 if (ioList[j].status == EINPROGRESS) {
 382                     printf("    Request %d on descriptor %d:", j,
 383                             ioList[j].aiocbp\->aio_fildes);
 384                     s = aio_cancel(ioList[j].aiocbp\->aio_fildes,
 385                             ioList[j].aiocbp);
 386                     if (s == AIO_CANCELED)
 387                         printf("I/O canceled\\n");
 388                     else if (s == AIO_NOTCANCELED)
 389                             printf("I/O not canceled\\n");
 390                     else if (s == AIO_ALLDONE)
 391                         printf("I/O all done\\n");
 392                     else
 393                         errMsg("aio_cancel");
 394                 }
 395             }
 396
 397             gotSIGQUIT = 0;
 398         }
 399
 400         /* Check the status of each I/O request that is still
 401            in progress */
 402
 403         printf("aio_error():\\n");
 404         for (j = 0; j < numReqs; j++) {
 405             if (ioList[j].status == EINPROGRESS) {
 406                 printf("    for request %d (descriptor %d): ",
 407                         j, ioList[j].aiocbp\->aio_fildes);
 408                 ioList[j].status = aio_error(ioList[j].aiocbp);
 409
 410                 switch (ioList[j].status) {
 411                 case 0:
 412                     printf("I/O succeeded\\n");
 413                     break;
 414                 case EINPROGRESS:
 415                     printf("In progress\\n");
 416                     break;
 417                 case ECANCELED:
 418                     printf("Canceled\\n");
 419                     break;
 420                 default:
 421                     errMsg("aio_error");
 422                     break;
 423                 }
 424
 425                 if (ioList[j].status != EINPROGRESS)
 426                     openReqs\-\-;
 427             }
 428         }
 429     }
 430
 431     printf("All I/O requests completed\\n");
 432
 433     /* Check status return of all I/O requests */
 434
 435     printf("aio_return():\\n");
 436     for (j = 0; j < numReqs; j++) {
 437         ssize_t s;
 438
 439         s = aio_return(ioList[j].aiocbp);
 440         printf("    for request %d (descriptor %d): %zd\\n",
 441                 j, ioList[j].aiocbp\->aio_fildes, s);
 442     }
 443
 444     exit(EXIT_SUCCESS);
 445 }
 446 .fi
 447 .SH SEE ALSO
 448 .ad l
 449 .nh
 450 .BR io_cancel (2),
 451 .BR io_destroy (2),
 452 .BR io_getevents (2),
 453 .BR io_setup (2),
 454 .BR io_submit (2),
 455 .BR aio_cancel (3),
 456 .BR aio_error (3),
 457 .BR aio_init (3),
 458 .BR aio_read (3),
 459 .BR aio_return (3),
 460 .BR aio_write (3),
 461 .BR lio_listio (3)
 462
 463 "Asynchronous I/O Support in Linux 2.5",
 464 Bhattacharya, Pratt, Pulavarty, and Morgan,
 465 Proceedings of the Linux Symposium, 2003,
 466 .UR https://www.kernel.org/doc/ols/2003/ols2003-pages-351-366.pdf
 467 .UE