sigvector(2)

NAME

sigvector — software signal facilities

SYNOPSIS

#include <signal.h> 

int sigvector( 
     int sig, 
     const struct sigvec *vec, 
     struct sigvec *ovec 
); 

DESCRIPTION

The system defines a set of signals that can be delivered to a process. The set of signals is defined in signal(5), along with the meaning and side effects of each signal. This manual entry, along with those for sigblock(2), sigsetmask(2), sigpause(3C), and sigspace(2), defines an alternate mechanism for handling these signals that ensures the delivery of signals and the integrity of signal handling procedures. The facilities described here should not be used in the same program as signal(2).

With the sigvector() interface, signal delivery resembles the occurrence of a hardware interrupt: the signal is blocked from further occurrence, the current process context is saved, and a new one is built. A process can specify a handler function to be invoked when a signal is delivered, or specify that a signal should be blocked or ignored. A process can also specify that a default action should be taken by the system when a signal occurs. It is possible to ensure a minimum amount of stack space for processing signals using sigspace() (see sigspace(2)).

All signals have the same priority. Signal routines execute with the signal that causes their invocation to be blocked, although other signals can yet occur. A global signal mask defines the set of signals currently blocked from delivery to a process. The signal mask for a process is initialized from that of its parent (normally 0). It can be changed with a sigblock(), sigsetmask(), or sigpause() call, or when a signal is delivered to the process.

A signal mask is represented as a long, with one bit representing each signal being blocked. The following macro defined in <signal.h> is used to convert a signal number to its corresponding bit in the mask:

#define sigmask(signo)     (1L << (signo-1)) 

When a signal condition arises for a process, the signal is added to a set of signals pending for the process. If the signal is not currently blocked by the process, it is delivered to the process. When a signal is delivered, the current state of the process is saved, a new signal mask is calculated (as described below), and the signal handler is invoked. The call to the handler is arranged so that if the signal handling routine returns normally, the process resumes execution in the same context as before the signal's delivery. If the process wishes to resume in a different context, it must arrange to restore the previous context itself.

When a signal is delivered to a process, a new signal mask is installed for the duration of the process' signal handler (or until a sigblock() or sigsetmask() call is made). This mask is formed by taking the current signal mask, computing the bit-wise inclusive OR with the value of vec.sv_mask (see below) from the most recent call to sigvector() for the signal to be delivered, and, unless the SV_RESETHAND flag is set (see below), setting the bit corresponding to the signal being delivered. When the user's signal handler returns normally, the original mask is restored.

sigvector() assigns a handler for the signal specified by sig. vec and ovec are pointers to sigvec structures that include the following elements:

void: (*sv_handler)();
long: sv_mask;
long: sv_flags;

If vec is non-zero, it specifies a handler routine (sv_handler), a mask (sv_mask) that the system should use when delivering the specified signal, and a set of flags (sv_flags) that modify the delivery of the signal. If ovec is non-zero, the previous handling information for the signal is returned to the user. If vec is zero, signal handling is unchanged. Thus, the call can be used to enquire about the current handling of a given signal. If vec and ovec point to the same structure, the value of vec is read prior to being overwritten.

The sv_flags field can be used to modify the receipt of signals. The following flag bits are defined:

SV_ONSTACK: Use the sigspace() allocated space.
SV_BSDSIG: Use the Berkeley signal semantics.
SV_RESETHAND: Use the semantics of signal(2).

If SV_ONSTACK is set, the system uses or permits the use of the space reserved for signal processing in the sigspace() system call.

If SV_BSDSIG is set, the signal is given the Berkeley semantics. The following signal is affected by this flag:

SIGCLD: In addition to being sent when a child process dies, the signal is also sent when any child's status changes from running to stopped. This would normally be used by a program such as csh (see csh(1)) when maintaining process groups under Berkeley job control.

If SV_RESETHAND is set, the signal handler is installed with the same semantics as a handler installed with signal(2). This affects signal mask set-up during the signal handler (see above) and whether the handler is reset after a signal is caught (see below).

If SV_RESETHAND is not set, once a signal handler is installed, it remains installed until another sigvector() call is made or an exec() system call is performed (see exec(2)). If SV_RESETHAND is set and the signal is not one of those marked "not reset when caught" under signal(5), the default action is reinstated when the signal is caught, prior to entering the signal-catching function. The "not reset when caught" distinction is not significant when sigvector() is called and SV_RESETHAND is not set.

The default action for a signal can be reinstated by setting sv_handler to SIG_DFL; this default usually results in termination of the process. If sv_handler is SIG_IGN the signal is usually subsequently ignored, and pending instances of the signal are discarded. The exact meaning of SIG_DFL and SIG_IGN for each signal is discussed in signal(5).

Certain system calls can be interrupted by a signal; all other system calls complete before the signal is serviced. The scp pointer described in signal(5) is never null if sigvector() is supported. scp points to a machine-dependent sigcontext structure. All implementations of this structure include the fields:

int: sc_syscall;
char: sc_syscall_action;

The value SYS_NOTSYSCALL for the sc_syscall field indicates that the signal is not interrupting a system call; any other value indicates which system call it is interrupting.

If a signal that is being caught occurs during a system call that can be interrupted, the signal handler is immediately invoked. If the signal handler exits normally, the value of the sc_syscall_action field is inspected; if the value is SIG_RETURN, the system call is aborted and the interrupted program continues past the call. The result of the interrupted call is -1 and errno is set to EINTR. If the value of the sc_syscall_action field is SIG_RESTART, the call is restarted. A call is restarted if, in the case of a read() or write() system call (see read(2) or write(2)), it had transferred no data. If some data had been transferred, the operation is considered to have completed with a partial transfer, and the sc_syscall value is SYS_NOTSYSCALL. Other values are undefined and reserved for future use.

Exiting the handler abnormally (such as with longjmp() — see setjmp(3C)) aborts the call, leaving the user responsible for the context of further execution. The value of scp->sc_syscall_action is ignored when the value of scp->sc_syscall is SYS_NOTSYSCALL. scp->sc_syscall_action is always initialized to SIG_RETURN before invocation of a signal handler. When an system call that can be interrupted is interrupted by multiple signals, if any signal handler returns a value of SIG_RETURN in scp->sc_syscall_action, all subsequent signal handlers are passed a value of SYS_NOTSYSCALL in scp->sc_syscall.

Note that calls to read(), write(), or ioctl() on fast devices (such as disks) cannot be interrupted, but I/O to a slow device (such as a printer) can be interrupted. Other system calls, such as those used for networking, also can be interrupted on some implementations. In these cases additional values can be specified for scp->sc_syscall. Programs that look at the values of scp->sc_syscall always should compare them to these symbolic constants; the numerical values represented by these constants might vary among implementations. System calls that can be interrupted and their corresponding values for scp->sc_syscall are listed below:

Call	sc_syscall value
read (slow devices)	`SYS_READ`
readv (slow devices)	`SYS_READV`
write (slow devices)	`SYS_WRITE`
writev (slow devices)	`SYS_WRITEV`
open (slow devices)	`SYS_OPEN`
ioctl (slow requests)	`SYS_IOCTL`
close (slow requests)	`SYS_CLOSE`
wait	`SYS_WAIT`
select	`SYS_SELECT`
pause	`SYS_PAUSE`
sigpause	`SYS_SIGPAUSE`
semop	`SYS_SEMOP`
msgsnd	`SYS_MSGSND`
msgrcv	`SYS_MSGRCV`

These system calls are not defined if the preprocessor macro _XPG2 is defined when <signal.h> is included. This is because the X/Open Portability Guide, Issue 2 specifies a different meaning for the symbol SYS_OPEN (see limits(5)).

After a fork() or vfork() system call, the child inherits all signals, the signal mask, and the reserved signal stack space.

exec(2) resets all caught signals to the default action; ignored signals remain ignored, the signal mask remains unchanged, and the reserved signal stack space is released.

The mask specified in vec is not allowed to block signals that cannot be ignored, as defined in signal(5). This is enforced silently by the system.

If sigvector() is called to catch SIGCLD in a process that currently has terminated (zombie) children, a SIGCLD signal is delivered to the calling process immediately, or as soon as SIGCLD is unblocked if it is currently blocked. Thus, in a process that spawns multiple children and catches SIGCLD, it is sometimes advisable to reinstall the handler for SIGCLD after each invocation in case there are multiple zombies present. This is true even though the handling of the signal is not reset by the system, as with signal(2), because deaths of multiple processes while SIGCLD is blocked in the handler result in delivery of only a single signal. Note that the function must reinstall itself after it has called wait() or wait3(). Otherwise the presence of the child that caused the original signal always causes another signal to be delivered.

RETURN VALUE

Upon successful completion, sigvector() returns 0; otherwise, it returns -1 and sets errno to indicate the reason.

ERRORS

sigvector() fails and no new signal handler is installed if any of the following conditions are encountered:

[EFAULT]: Either vec or ovec points to memory that is not a valid part of the process address space. Reliable detection of this error is implementation dependent.
[EINVAL]: sig is not a valid signal number.
[EINVAL]: An attempt was made to ignore or supply a handler for a signal that cannot be caught or ignored; see signal(5).

WARNINGS

Restarting a select(2) call can sometimes cause unexpected results. If the select() call has a timeout specified, the timeout is restarted with the call, ignoring any portion that had elapsed prior to interruption by the signal. Normally this simply extends the timeout and is not a problem. However, if a handler repeatedly catches signals, and the timeout specified to select() is longer than the time between those signals, restarting the select() call effectively renders the timeout infinite.

sigvector() should not be used in conjunction with the facilities described under sigset(3C).

APPLICATION USAGE

Threads Considerations

The signal disposition (such as catch/ignore/default) established by sigvector() is shared by all threads in the process. Each thread maintains its own blocked signal mask. For more information regarding signals and threads, refer to signal(5).

AUTHOR

sigvector() was developed by HP and the University of California, Berkeley.