Signals and Exit Codes

Unix Signals

Linux signal numbers vary slightly by architecture; the numbers below are for x86_64 Linux. Names are POSIX. The -l flag of kill prints the local mapping.

Number	Name	Default	Meaning
1	SIGHUP	Term	terminal hangup; daemons commonly reload
2	SIGINT	Term	interrupt (Ctrl-C)
3	SIGQUIT	Core	quit + core dump (Ctrl-\)
4	SIGILL	Core	illegal instruction
5	SIGTRAP	Core	trace / breakpoint trap
6	SIGABRT	Core	abort()
7	SIGBUS	Core	bus error (memory access)
8	SIGFPE	Core	floating-point exception
9	SIGKILL	Term	forced termination, cannot be caught
10	SIGUSR1	Term	user-defined 1
11	SIGSEGV	Core	segmentation fault (invalid memory)
12	SIGUSR2	Term	user-defined 2
13	SIGPIPE	Term	write to closed pipe / socket
14	SIGALRM	Term	alarm timer expired
15	SIGTERM	Term	graceful termination request (default for kill)
16	SIGSTKFLT	Term	stack fault (rare)
17	SIGCHLD	Ign	child stopped or terminated
18	SIGCONT	Cont	continue (resume)
19	SIGSTOP	Stop	stop (cannot be caught)
20	SIGTSTP	Stop	terminal stop (Ctrl-Z)
21	SIGTTIN	Stop	background process tried to read tty
22	SIGTTOU	Stop	background process tried to write tty
23	SIGURG	Ign	out-of-band socket data
24	SIGXCPU	Core	CPU time limit exceeded
25	SIGXFSZ	Core	file size limit exceeded
26	SIGVTALRM	Term	virtual alarm
27	SIGPROF	Term	profiling alarm
28	SIGWINCH	Ign	terminal window size changed
29	SIGIO / SIGPOLL	Term	async I/O
30	SIGPWR	Term	power failure
31	SIGSYS	Core	bad system call

Default actions: Term = terminate; Core = terminate + core dump; Stop = pause; Cont = resume; Ign = ignored.

The Two You Can’t Catch

Two signals the kernel handles itself, with no opportunity for the target process to install a handler. SIGKILL terminates; SIGSTOP suspends. Every other signal can be caught, blocked, or ignored, which is why these two are the always-works escape hatch.

• SIGKILL (9), forced termination. The kernel kills the process; no chance to clean up.

• SIGSTOP (19), forced pause. SIGCONT resumes.

Every other signal can be caught, blocked, or ignored.

Real-Time Signals

Linux also defines real-time signals 32-64 (SIGRTMIN .. SIGRTMAX). These are queued (multiple deliveries are not coalesced the way standard signals are) and used by libpthread internally and by some applications for IPC.

Sending Signals

The shell-side primitives. kill sends to a PID, pkill / killall send by name, and trap installs a handler in the current shell so a script can clean up before the signal would otherwise terminate it.

$ kill -TERM PID
$ kill -9 PID
$ kill -HUP PID
$ pkill -USR1 myproc
$ killall -SIGTERM myproc

$ trap 'echo got SIGINT; cleanup; exit 130' INT
$ trap 'cleanup' EXIT

Common Daemon Conventions

What well-behaved daemons do when each signal arrives. SIGHUP for config reload is the most-cited convention; SIGUSR1 / SIGUSR2 are application-defined hooks (log rotation, graceful restart); SIGTERM is the polite shutdown request before SIGKILL.

Signal	Common effect
SIGHUP	reload configuration
SIGUSR1	rotate logs / dump stats
SIGUSR2	graceful binary upgrade (nginx, unicorn)
SIGTERM	graceful shutdown
SIGINT	interrupt (often same as SIGTERM)
SIGQUIT	graceful drain (nginx) or core-dump (most processes)
SIGKILL	last resort

Unix Exit Codes

A process’s exit status is an 8-bit integer. By convention:

Code	Meaning
0	success
1	general / catch-all error
2	misuse of shell builtins
64	EX_USAGE command-line usage error
65	EX_DATAERR input data error
66	EX_NOINPUT cannot open input
67	EX_NOUSER user does not exist
68	EX_NOHOST host does not exist
69	EX_UNAVAILABLE service unavailable
70	EX_SOFTWARE internal software error
71	EX_OSERR operating-system error
72	EX_OSFILE system file missing
73	EX_CANTCREAT cannot create output
74	EX_IOERR IO error
75	EX_TEMPFAIL temporary failure; retry safe
76	EX_PROTOCOL remote protocol error
77	EX_NOPERM permission denied
78	EX_CONFIG configuration error
126	command found but not executable
127	command not found
128	invalid argument to exit
128 + N	terminated by signal N

Examples:

• 137, killed by SIGKILL (128 + 9). Often OOM killer or kill -9.

• 143, terminated by SIGTERM (128 + 15). Typical orderly shutdown.

• 130, interrupted by SIGINT (128 + 2). User pressed Ctrl-C.

The 64-78 range comes from sysexits.h, BSD convention; not universally used, but worth knowing.

$? and Pipelines

$? carries the exit status of the last command in the foreground. In a pipeline, $? is the last stage by default; ${PIPESTATUS[@]} exposes every stage; set -o pipefail makes the pipeline exit non-zero if any stage failed.

$ cmd
$ echo "$?"

$ a | b | c
$ echo "${PIPESTATUS[@]}"

$ set -o pipefail

Timeout Conventions

Exit codes timeout(1) uses to distinguish “the command ran but took too long” from “the command itself failed.” 124 is the common one; 125-127 are the failure modes; 137 is “we had to SIGKILL it” (128+9).

Code	Source
124	timeout, command timed out
125	timeout, internal error
126	timeout, command found but not executable
127	timeout, command not found
137	usually timeout --signal=KILL (128+9)

What to Use in Your Programs

A short rule for picking exit codes when you write something new. 0 / 1 / 2 cover most cases; reach for sysexits.h numbers only if your program participates in that BSD convention; reserve custom codes for cases where callers genuinely need to branch on the failure type.

• 0 for success.

• 1 for general failure.

• 2 for usage / syntax errors (argparse does this in Python).

• Numbers from sysexits.h if your program participates in the BSD convention.

• Custom non-overlapping codes only when callers genuinely need to distinguish.