Signals are standardized messages sent to a running program to trigger specific behavior, such as quitting or error handling. They are a limited form of inter-process communication (IPC), typically used in Unix, Unix-like, and other POSIX-compliant operating systems.
A signal is an asynchronous notification sent to a process or to a specific thread within the same process to notify it of an event. Common uses of signals are to interrupt, suspend, terminate or kill a process. Signals originated in 1970s Bell Labs Unix and were later specified in the POSIX standard.
When a signal is sent, the operating system interrupts the target process's normal flow of execution to deliver the signal. Execution can be interrupted during any non-atomic instruction. If the process has previously registered a signal handler, that routine is executed. Otherwise, the default signal handler is executed.
Embedded programs may find signals useful for inter-process communications, as signals are notable for their algorithmic efficiency.
Signals are similar to interrupts, the difference being that interrupts are mediated by the CPU and handled by the kernel while signals are mediated by the kernel (possibly via system calls) and handled by individual processes. The kernel may pass an interrupt as a signal to the process that caused it (typical examples are SIGSEGV, SIGBUS, SIGILL and SIGFPE).
Contents
History
Version 1 Unix (1971) had separate system calls to catch interrupts, quits, and machine traps.
kill (a command to send a signal, by default SIGTERM) appeared in Version 2 (1972).
Version 4 (1973) combined all traps into one call, signal.
Version 5 (1974) could send arbitrary signals.
In Version 7 (1979) each numbered trap received a symbolic name.
Plan 9 from Bell Labs (late 80s) replaced signals with notes, which permit sending short, arbitrary strings.
Sending signals
The kill(2) system call sends a specified signal to a specified process, if permissions allow. Similarly, the kill(1) command allows a user to send signals to processes. The raise(3) library function sends the specified signal to the current process.
Exceptions such as division by zero, segmentation violation (SIGSEGV), and floating point exception (SIGFPE) will cause a core dump and terminate the program.
The kernel can generate signals to notify processes of events. For example, SIGPIPE will be generated when a process writes to a pipe which has been closed by the reader; by default, this causes the process to terminate, which is convenient when constructing shell pipelines.
Typing certain key combinations at the controlling terminal of a running process causes the system to send it certain signals:
Ctrl-C (in older Unixes, DEL) sends an INT signal ("interrupt", SIGINT); by default, this causes the process to terminate.
Ctrl-Z sends a TSTP signal ("terminal stop", SIGTSTP); by default, this causes the process to suspend execution.
Ctrl-\ sends a QUIT signal (SIGQUIT); by default, this causes the process to terminate and dump core.
Ctrl-T (not supported on all UNIXes) sends an INFO signal (SIGINFO); by default, and if supported by the command, this causes the operating system to show information about the running command.
These default key combinations with modern operating systems can be changed with the stty command.
Handling signals
Signal handlers can be installed with the signal(2) or sigaction(2) system call. If a signal handler is not installed for a particular signal, the default handler is used. Otherwise the signal is intercepted and the signal handler is invoked. The process can also specify two default behaviors, without creating a handler: ignore the signal (SIG_IGN) and use the default signal handler (SIG_DFL). There are two signals which cannot be intercepted and handled: SIGKILL and SIGSTOP.
Risks
Signal handling is vulnerable to race conditions. As signals are asynchronous, another signal (even of the same type) can be delivered to the process during execution of the signal handling routine.
The sigprocmask(2) call can be used to block and unblock delivery of signals. Blocked signals are not delivered to the process until unblocked. Signals that cannot be ignored (SIGKILL and SIGSTOP) cannot be blocked.
Signals can cause the interruption of a system call in progress, leaving it to the application to manage a non-transparent restart.
Signal handlers should be written in a way that does not result in any unwanted side-effects, e.g. errno alteration, signal mask alteration, signal disposition change, and other global process attribute changes. Use of non-reentrant functions, e.g., malloc or printf, inside signal handlers is also unsafe. In particular, the POSIX specification and the Linux man page signal (7) require that all system functions directly or indirectly called from a signal function are async-signal safe. The signal-safety(7) man page gives a list of such async-signal safe system functions (practically the system calls). Otherwise, it is an undefined behavior. It is suggested to simply set some volatile sig_atomic_t variable in a signal handler, and to test it elsewhere.
Signal handlers can instead put the signal into a queue and immediately return. The main thread will then continue "uninterrupted" until signals are taken from the queue, such as in an event loop. "Uninterrupted" here means that operations that block may return prematurely and must be resumed, as mentioned above. Signals should be processed from the queue on the main thread and not by worker pools, as that reintroduces the problem of asynchronicity. However, managing a queue is not possible in an async-signal safe way with only sig_atomic_t, as only single reads and writes to such variables are guaranteed to be atomic, not increments or (fetch-and)-decrements, as would be required for a queue. Thus, effectively, only one signal per handler can be queued safely with sig_atomic_t until it has been processed.
Relationship with hardware exceptions
A process's execution may result in the generation of a hardware exception, for instance, if the process attempts to divide by zero or incurs a page fault.
In Unix-like operating systems, this event automatically changes the processor context to start executing a kernel exception handler. In case of some exceptions, such as a page fault, the kernel has sufficient information to fully handle the event itself and resume the process's execution.
For other exceptions, however, the kernel cannot process them intelligently and it must defer the exception handling operation to the faulting process. This deferral is achieved via the signal mechanism, wherein the kernel sends to the process a signal corresponding to the current exception. For example, if a process attempted integer divide by zero on an x86 CPU, a divide error exception would be generated and cause the kernel to send the SIGFPE signal to the process.
Similarly, if the process attempted to access a memory address outside of its virtual address space, the kernel would notify the process of this violation via a SIGSEGV (segmentation violation signal). The exact mapping between signal names and exceptions is obviously dependent upon the CPU, since exception types differ between architectures.
POSIX signals
The list below documents the signals specified in the Single Unix Specification Version 5. All signals are defined as macro constants in the <signal.h> header file. The name of the macro constant consists of a "SIG" prefix followed by a mnemonic name for the signal.
A process can define how to handle incoming POSIX signals. If a process does not define a behaviour for a signal, then the default handler for that signal is being used. The table below lists some default actions for POSIX-compliant UNIX systems, such as FreeBSD, OpenBSD and Linux.
Portable number:
For most signals the corresponding signal number is implementation-defined. This column lists the numbers specified in the POSIX standard.
Actions explained:
Terminate – Abnormal termination of the process. The process is terminated with all the consequences of _exit() except that the status made available to wait() and waitpid() indicates abnormal termination by the specified signal.
Terminate (core dump) – Abnormal termination of the process. Additionally, implementation-defined abnormal termination actions, such as creation of a core file, may occur.
Ignore – Ignore the signal.
Stop – Stop (or suspend) the process.
Continue – Continue the process, if it is stopped; otherwise, ignore the signal.
SIGABRT and SIGIOT
The SIGABRT signal is sent to a process to tell it to abort, i.e. to terminate. The signal is usually initiated by the process itself when it calls abort() function of the C Standard Library, but it can be sent to the process from outside like any other signal.
Miscellaneous signals
The following signals are not specified in the POSIX specification. They are, however, sometimes used on various systems.
SIGEMT
The SIGEMT signal is sent to a process when an emulator trap occurs. While an emulator usually means software that executes other programs, in this case it means a program executed a supervisor call instruction (EMT was the instruction for this purpose on the DEC PDP-11 series of computers.)
SIGINFO
The SIGINFO signal is sent to a process when a status (info) request is received from the controlling terminal.
SIGPWR
The SIGPWR signal is sent to a process when the system experiences a power failure.
SIGLOST
The SIGLOST signal is sent to a process when a file lock is lost.
SIGSTKFLT
The SIGSTKFLT signal is sent to a process when the coprocessor experiences a stack fault (i.e. popping when the stack is empty or pushing when it is full). It is defined by, but not used on Linux, where a x87 coprocessor stack fault will generate SIGFPE instead.
SIGUNUSED
The SIGUNUSED signal is sent to a process when a system call with an unused system call number is made. It is synonymous with SIGSYS on most architectures.
SIGCLD
The SIGCLD signal is synonymous with SIGCHLD.
