Arm a resource manager
#include <unistd.h> #include <sys/iomsg.h> int ionotify ( int fd, int action, int flags, const struct sigevent *event );
- The file descriptor associated with the resource manager that you want to notify.
- The type of arming action to take; see Actions, below.
- The types of conditions that can be checked for notification; see Flags, below.
- A pointer to a sigevent structure that defines the event that you want the resource manager to send as a notification, or NULL to disarm a notification.
Use the -l c option to qcc to link against this library. This library is usually included automatically.
The ionotify() function arms the resource manager associated with fd to send the event notification event. The event is sent when a condition specified by a combination of action and flags occurs.
The flags argument specifies the types of conditions that can be checked for notification. Each resource manager maintains a different context for each notification condition. Only those notification bits specified are affected. In the following example, the second call to ionotify() doesn't affect the first, since it specifies a different notification:
ionotify( fd, _NOTIFY_ACTION_POLLARM, _NOTIFY_COND_INPUT, &event ); ionotify( fd, _NOTIFY_ACTION_POLLARM, _NOTIFY_COND_OUTPUT, &event );
The conditions specified by flags are:
- Out-of-band data is available. The definition of out-of-band data depends on the resource manager.
- There's room in the output buffer for more data. The amount of room available needed to satisfy this condition depends on the resource manager. Some resource managers may default to an empty output buffer, while others may choose some percentage of the buffer empty.
- There's input data available. The amount of data available defaults to 1. For a character device such as a serial port, this would be a character. For a POSIX message queue, it would be a message. Each resource manager selects an appropriate object.
- The conditions are defined with the following extended flags:
- _NOTIFY_CONDE_RDNORM — normal data is available; this is the same as _NOTIFY_COND_INPUT.
- _NOTIFY_CONDE_WRNORM — room for normal data; this is the same as _NOTIFY_COND_OUTPUT.
- _NOTIFY_CONDE_RDBAND — out-of-band data is available; this is the same as _NOTIFY_COND_OBAND.
- _NOTIFY_CONDE_PRI — priority data is available.
- _NOTIFY_CONDE_WRBAND — room for OOB data.
- _NOTIFY_CONDE_ERR — an error occurred on the device or stream.
- _NOTIFY_CONDE_HUP — the device has been disconnected.
- _NOTIFY_CONDE_NVAL — the file descriptor is invalid.
The method for changing the default number for _NOTIFY_COND_OUTPUT and _NOTIFY_COND_INPUT depends on the device. For example, character special devices can call readcond().
For resource managers that support both an edited and raw mode, the mode should be set to raw to ensure proper operation of ionotify().
The above flags are located in the top bits of flags. They are defined by _NOTIFY_COND_MASK.
In the case of an asynchronous notification using the passed event, such as a Neutrino pulse or queued realtime signal, the 32-bit value in event->sigev_value.sival_int is returned to you unmodified, unless you've selected the SI_NOTIFY code, in which case the top bits (defined by _NOTIFY_COND_MASK) are set to the active notifications. In this case, you should limit the sival_int to the mask defined by _NOTIFY_DATA_MASK.
For example, the Unix select() function specifies SI_NOTIFY and uses the allowable data bits of sival_int as a serial number.
The action argument specifies the type of arming action to take. When a condition is armed, the resource manager monitors it and, when met, delivers event using MsgDeliverEvent(). When an event is delivered, it's always disarmed except where noted below.
Note that for transition arming (as specified by an action of _NOTIFY_ACTION_TRANARM, only one notification of that type can be outstanding per device. When the transition arm fires, it's removed.
Each action is designed to support a specific notification type as follows:
- This action does a poll of the notification conditions specified
by flags. It never arms an event, and it cancels
all other asynchronous event notifications
set up by a previous call to ionotify(). This also allows it
to be used as a simple disarm call.
Returns active conditions as requested by flags.
- This action does a poll in the same way as
_NOTIFY_ACTION_POLL. However, if none of the
conditions specified in flags are present then each
condition specified in flags is armed. If any
condition is met, none of the conditions are
armed. The Unix select() function uses
ionotify() with this action.
Returns active conditions as requested by flags.
- This action arms for transitions of the notification conditions specified
by flags. A transition is defined as a
data transition from empty to nonempty
on input. Its use on output isn't defined. Note that if there
is data available when this call is used, a data transition
won't occur. To generate an event using this type of
must arm the event and then drain the input using a nonblocking
read. After this point, new input data causes the event to be delivered.
Since this arms for a transition, the return value is always zero.There's a limit of one notification per message queue or resource manager. If a notification was already armed for the specified resource, ionotify() indicates an error of EBUSY.
You can use the _NOTIFY_ACTION_POLLARM or _NOTIFY_ACTION_POLL action to generate events that are level- as opposed to transition-oriented.
When an action is armed in a resource manager, it remains armed until:
- A thread sets a new action (this disarms any current action and possibly arms a new action),
- The event is delivered and the action wasn't a continuous one,
- The thread closes the device.
Active conditions as requested by flags. In the case of a transition action, a zero is returned. If an error occurs, -1 is returned (errno is set).
- The connection indicated by fd doesn't exist, or fd is no longer connected to a channel.
- A notification was already armed for this resource; this function enforces a restriction of one per message queue or resource manager.
- A fault occurred when the kernel tried to access the buffers provided. This may have occurred on the receive or the reply.
- The call was interrupted by a signal.
- The resource manager couldn't allocate a notify entry to save the request.
- The requested action isn't supported by this resource manager.
- A fault occurred in a server's address space while accessing the server's message buffers. This may have occurred on the receive or the reply.
- A kernel timeout unblocked the call. See TimerTimeout().