#include <sys/types.h>
#include <sys/asynch.h>

int aioread(int fildes, char *bufp, int bufs, off_t offset,
 int whence, aio_result_t *resultp);

int aiowrite(int fildes, const char *bufp, int bufs, off_t offset,
 int whence, aio_result_t *resultp);

The aioread() function initiates one asynchronous read(2) and returns control to the calling program. The read continues concurrently with other activity of the process. An attempt is made to read bufs bytes of data from the object referenced by the descriptor fildes into the buffer pointed to by bufp.

The aiowrite() function initiates one asynchronous write(2) and returns control to the calling program. The write continues concurrently with other activity of the process. An attempt is made to write bufs bytes of data from the buffer pointed to by bufp to the object referenced by the descriptor fildes.

On objects capable of seeking, the I/O operation starts at the position specified by whence and offset. These parameters have the same meaning as the corresponding parameters to the llseek(2) function. On objects not capable of seeking the I/O operation always start from the current position and the parameters whence and offset are ignored. The seek pointer for objects capable of seeking is not updated by aioread() or aiowrite(). Sequential asynchronous operations on these devices must be managed by the application using the whence and offset parameters.

The result of the asynchronous operation is stored in the structure pointed to by resultp:

int aio_return; /* return value of read() or write() */
int aio_errno; /* value of errno for read() or write() */

Upon completion of the operation both aio_return and aio_errno are set to reflect the result of the operation. Since AIO_INPROGRESS is not a value used by the system, the client can detect a change in state by initializing aio_return to this value.

The application-supplied buffer bufp should not be referenced by the application until after the operation has completed. While the operation is in progress, this buffer is in use by the operating system.

Notification of the completion of an asynchronous I/O operation can be obtained synchronously through the aiowait(3C) function, or asynchronously by installing a signal handler for the SIGIO signal. Asynchronous notification is accomplished by sending the process a SIGIO signal. If a signal handler is not installed for the SIGIO signal, asynchronous notification is disabled. The delivery of this instance of the SIGIO signal is reliable in that a signal delivered while the handler is executing is not lost. If the client ensures that aiowait() returns nothing (using a polling timeout) before returning from the signal handler, no asynchronous I/O notifications are lost. The aiowait() function is the only way to dequeue an asynchronous notification. The SIGIO signal can have several meanings simultaneously. For example, it can signify that a descriptor generated SIGIO and an asynchronous operation completed. Further, issuing an asynchronous request successfully guarantees that space exists to queue the completion notification.

The close(2), exit(2) and execve(2)) functions block until all pending asynchronous I/O operations can be canceled by the system.

It is an error to use the same result buffer in more than one outstanding request. These structures can be reused only after the system has completed the operation.

Upon successful completion, aioread() and aiowrite() return 0. Upon failure, aioread() and aiowrite() return -1 and set errno to indicate the error.

The aioread() and aiowrite() functions will fail if:

EAGAIN

EBADF

EFAULT

EINVAL

EINVAL

ENOMEM

The aioread() and aiowrite() functions have transitional interfaces for 64-bit file offsets. See lf64(5).

See attributes(5) for descriptions of the following attributes:

close(2), execve(2), exit(2), llseek(2), lseek(2), open(2), read(2), write(2), aiocancel(3C), aiowait(3C), sigvec(3UCB), attributes(5), lf64(5)