SIGNAL(9) BSD Kernel Developer's Manual SIGNAL(9)NAME
signal, siginit, sigactsinit, sigactsunshare, sigactsfree, execsigs,
sigaction1, sigprocmask1, sigpending1, sigsuspend1, sigaltstack1,
pgsignal, kpgsignal, psignal, kpsignal, issignal, postsig, killproc,
sigexit, trapsignal, sendsig, sigcode, sigtramp — software signal facili‐
ties
SYNOPSIS
#include <sys/signal.h>
#include <sys/signalvar.h>
void
siginit(struct proc *p);
void
sigactsinit(struct proc *pp, int share);
void
sigactsunshare(struct proc *p);
void
sigactsfree(struct proc *p);
void
execsigs(struct proc *p);
int
sigaction1(struct lwp *l, int signum, const struct sigaction *nsa,
struct sigaction *osa, void *tramp, int vers);
int
sigprocmask1(struct lwp *l, int how, const sigset_t *nss, sigset_t *oss);
void
sigpending1(struct lwp *l, sigset_t *ss);
int
sigsuspend1(struct lwp *l, const sigset_t *ss);
int
sigaltstack1(struct lwp *l, const struct sigaltstack *nss,
struct sigaltstack *oss);
void
pgsignal(struct pgrp *pgrp, int signum, int checkctty);
void
kpgsignal(struct pgrp *pgrp, ksiginfo_t *ks, void *data, int checkctty);
void
psignal(struct proc *p, int signum);
void
kpsignal(struct proc *p, ksiginfo_t *ks, void *data);
int
issignal(struct lwp *l);
void
postsig(int signum);
void
killproc(struct proc *p, const char *why);
void
sigexit(struct lwp *l, int signum);
void
trapsignal(struct lwp *l, const ksiginfo_t *ks);
void
sendsig(const ksiginfo_t *ks, const sigset_t *mask);
DESCRIPTION
The system defines a set of signals that may be delivered to a process.
These functions implement the kernel portion of the signal facility.
Signal numbers used throughout the kernel signal facilities should always
be within the range of [1-NSIG].
Most of the kernel's signal infrastructure is implemented in machine-
independent code. Machine-dependent code provides support for invoking a
process's signal handler, restoring context when the signal handler
returns, generating signals when hardware traps occur, triggering the
delivery of signals when a process is about to return from the kernel to
userspace.
The signal state for a process is contained in struct sigctx. This
includes the list of signals with delivery pending, information about the
signal handler stack, the signal mask, and the address of the signal
trampoline.
The registered signal handlers for a process are recorded in struct
sigacts. This structure may be shared by multiple processes.
The kernel's signal facilities are implemented by the following func‐
tions:
siginit(p)
This function initializes the signal state of proc0 to the system
default. This signal state is then inherited by init(8) when it
is started by the kernel.
sigactsinit(pp, share)
This function creates an initial struct sigacts for the process
pp. If the share argument is non-zero, then pp shares the struct
sigacts by holding a reference. Otherwise, pp receives a new
struct sigacts which is copied from the parent.
sigactsunshare(p)
This function causes the process p to no longer share its struct
sigacts The current state of the signal actions is maintained in
the new copy.
sigactsfree(p)
This function decrements the reference count on the struct sigacts
of process p. If the reference count reaches zero, the struct
sigacts is freed.
execsigs(p)
This function is used to reset the signal state of the process p
to the system defaults when the process execs a new program image.
sigaction1(l, signum, nsa, osa, tramp, vers)
This function implements the sigaction(2) system call. The tramp
and vers arguments provide support for userspace signal trampo‐
lines. Trampoline version 0 is reserved for the legacy kernel-
provided signal trampoline; tramp must be NULL in this case. Oth‐
erwise, vers specifies the ABI of the trampoline specified by
tramp. The signal trampoline ABI is machine-dependent, and must
be coordinated with the sendsig() function.
sigprocmask1(l, how, nss, oss)
This function implements the sigprocmask(2) system call.
sigpending1(l, ss)
This function implements the sigpending(2) system call.
sigsuspend1(l, ss)
This function implements the sigsuspend(2) system call.
sigaltstack1(l, nss, oss)
This function implements the sigaltstack(2) system call.
pgsignal(pgrp, signum, checkctty)
This is a wrapper function for kpgsignal() which is described
below.
kpgsignal(pgrp, ks, data, checkctty)
Schedule the signal ks->ksi_signo to be delivered to all members
of the process group pgrp. If checkctty is non-zero, the signal
is only sent to processes which have a controlling terminal. The
data argument and the complete signal scheduling semantics are
described in the kpsignal() function below.
trapsignal(l, ks)
Sends the signal ks->ksi_signo caused by a hardware trap to the
current process.
psignal(p, signum)
This is a wrapper function for kpsignal() which is described
below.
kpsignal(p, ks, data)
Schedule the signal ks->ksi_signo to be delivered to the process
p. The data argument, if not NULL, points to the file descriptor
data that caused the signal to be generated in the SIGIO case.
With a few exceptions noted below, the target process signal dis‐
position is updated and is marked as runnable, so further handling
of the signal is done in the context of the target process after a
context switch; see issignal() below. Note that kpsignal() does
not by itself cause a context switch to happen.
The target process is not marked as runnable in the following
cases:
· The target process is sleeping uninterruptibly. The
signal will be noticed when the process returns from the
system call or trap.
· The target process is currently ignoring the signal.
· If a stop signal is sent to a sleeping process that
takes the default action (see sigaction(2)), the process
is stopped without awakening it.
· SIGCONT restarts a stopped process (or puts them back to
sleep) regardless of the signal action (e.g., blocked or
ignored).
If the target process is being traced, kpsignal() behaves as if
the target process were taking the default action for signum.
This allows the tracing process to be notified of the signal.
issignal(l)
This function determines which signal, if any, is to be posted to
the current process. A signal is to be posted if:
· The signal has a handler provided by the program image.
· The signal should cause the process to dump core and/or
terminate.
· The signal should interrupt the current system call.
Signals which cause the process to be stopped are handled within
issignal() directly.
issignal() should be called by machine-dependent code when return‐
ing to userspace from a system call or other trap or interrupt by
using the following code:
while (signum = CURSIG(curproc))
postsig(signum);
postsig(signum)
The postsig() function is used to invoke the action for the signal
signum in the current process. If the default action of a signal
is to terminate the process, and the signal does not have a regis‐
tered handler, the process exits using sigexit(), dumping a core
image if necessary.
killproc(p, why)
This function sends a SIGKILL signal to the specified process.
The message provided by why is sent to the system log and is also
displayed on the process's controlling terminal.
sigexit(l, signum)
This function forces the current process to exit with the signal
signum, generating a core file if appropriate. No checks are made
for masked or caught signals; the process always exits.
sendsig(ks, mask)
This function is provided by machine-dependent code, and is used
to invoke a signal handler for the current process. sendsig()
must prepare the registers and stack of the current process to
invoke the signal handler stored in the process's struct sigacts.
This may include switching to an alternate signal stack specified
by the process. The previous register, stack, and signal state
are stored in a ucontext_t, which is then copied out to the user's
stack.
The registers and stack must be set up to invoke the signal han‐
dler as follows:
(*handler)(int signum, siginfo_t *info, void *ctx)
where signum is the signal number, info contains additional signal
specific information when SA_SIGINFO is specified when setting up
the signal handler. ctx is the pointer to ucontext_t on the
user's stack. The registers and stack must also arrange for the
signal handler to return to the signal trampoline. The trampoline
is then used to return to the code which was executing when the
signal was delivered using the setcontext(2) system call.
For performance reasons, it is recommended that sendsig() arrange
for the signal handler to be invoked directly on architectures
where it is convenient to do so. In this case, the trampoline is
used only for the signal return path. If it is not feasible to
directly invoke the signal handler, the trampoline is also used to
invoke the handler, performing any final set up that was not pos‐
sible for sendsig() to perform.
sendsig() must invoke the signal trampoline with the correct ABI.
The ABI of the signal trampoline is specified on a per-signal
basis in the sigacts() structure for the process. Trampoline ver‐
sion 0 is reserved for the legacy kernel-provided, on-stack signal
trampoline. All other trampoline versions indicate a specific
trampoline ABI. This ABI is coordinated with machine-dependent
code in the system C library.
SIGNAL TRAMPOLINE
The signal trampoline is a special piece of code which provides support
for invoking the signal handlers for a process. The trampoline is used
to return from the signal handler back to the code which was executing
when the signal was delivered, and is also used to invoke the handler
itself on architectures where it is not feasible to have the kernel
invoke the handler directly.
In traditional UNIX systems, the signal trampoline, also referred to as
the “sigcode”, is provided by the kernel and copied to the top of the
user's stack when a new process is created or a new program image is
exec'd. Starting in NetBSD 2.0, the signal trampoline is provided by the
system C library. This allows for more flexibility when the signal
facility is extended, makes dealing with signals easier in debuggers,
such as gdb(1), and may also enhance system security by allowing the ker‐
nel to disallow execution of code on the stack.
The signal trampoline is specified on a per-signal basis. The correct
trampoline is selected automatically by the C library when a signal han‐
dler is registered by a process.
Signal trampolines have a special naming convention which enables debug‐
gers to determine the characteristics of the signal handler and its argu‐
ments. Trampoline functions are named like so:
__sigtramp_<flavor>_<version>
where:
⟨flavor⟩ The flavor of the signal handler. The following flavors are
valid:
sigcontext Specifies a traditional BSD-style (deprecated)
signal handler with the following signature:
void (*handler)(int signum,
int code,
struct sigcontext *scp);
siginfo Specifies a POSIX-style signal handler with the
following signature:
void (*handler)(int signum,
siginfo_t *si,
void *uc);
Note: sigcontext style signal handlers are dep‐
recated, and retained only for compatibility
with older binaries.
⟨version⟩ Specifies the ABI version of the signal trampoline. The tram‐
poline ABI is coordinated with the machine-dependent kernel
sendsig() function. The trampoline version needs to be unique
even across different trampoline flavors, in order to simplify
trampoline selection in the kernel.
The following is an example if a signal trampoline name which indicates
that the trampoline is used for traditional BSD-style signal handlers and
implements version 1 of the signal trampoline ABI:
__sigtramp_sigcontext_1
The current signal trampoline is:
__sigtramp_siginfo_2
SEE ALSOsigaction(2), signal(7), condvar(9)BSD April 29, 2010 BSD
