I know that in most cases, it is impossible to ignore all signals. If it was possible for an application to ignore all signals, how would you stop it? Since you wouldn't be able to use SIGKILL or SIGSTOP...Not even sure if this question makes sense...
SIGKILL will end the process without it having any say in the matter.
When SIGKILL is send to a process, the kernel will not relay the signal to the process and call a signal handler it specified. Instead the kernel will simply immediately stop and destroy the process.
So, SIGKILL will always work. There is nothing the process can do to prevent it. It won't get any time to execute any code or do any cleanup. This is why you would usually try to send a SIGTERM first to ask the process to come to an end on its own and follow with a SIGKILL after a while only if the process didn't honor the SIGTERM request.
For SIGSTOP the matter is similar.
I want use kill() to kill a program.
does linux kernel ensure that the program is killed before kill() returned?
If not, I have to check whether the program is killed already.
does linux kernel ensure that the program is killed before kill() returned?
No, kill merely sends a signal to a process or group of processes. Its successful return only tells you that the signal was sent.
The Linux man pages are straightforward about this, clarifying that a successful return means "at least one signal was sent".
So, this does not tell you whether the signal is blocked/held pending by the target process, was ignored (SIG_IGN), has taken its default action, has been handled, or accepted.
I understand there are catchable and uncatchable signals. However, it seems that both types of signals get sent to processes by the OS. What makes uncatchable signals uncatchable? Is it the signal handler that catches signals, and because no signal handler is written to handle an a particular signal (e.g. sigkill), that it becomes uncatchable? If that's true, can I conclude that it is possible to catch an uncatchable signal by writing a signal handler?
No, you can't catch the uncatchable signals as it is always caught by the default handler implemented by the kernel.
SIGKILL always terminates the process. Even if the process attempts to handle the SIGKILL signal by registering a handler for it, still the control would always land in the default SIGKILL handler which would terminate the program.
This is what happens when you try to shut down your system.
First, the system sends a SIGTERM signal to all the processes and waits for a while giving those processes a grace period. If it still doesn't stop even after the grace period, the system forcibly terminates all the process by using SIGKILL signal.
I have a test case where there are threads spawned using CLONE_THREAD option in clone() .Here if i want to kill a particular thread I suppose we should be using SYS_tgkill in systemcall(). But will the kill actually affect a thread if it is waiting in kernel space(say a futex_wait)?
I tried killing a thread created in the above manner.But when SIGKILL is sent to the same the whole process is getting killed.Am i missing something in using syscall(SYS_tgkill,pid,tid,9) ?
SIGKILL always kills the target process. There is no way around this; it's unblockable, unignorable, and uncatchable.
You could try sending another signal (like SIGUSR1 or SIGHUP or SIGRTMIN) and having a signal handler installed that calls pthread_exit (but note that this function is not async-signal-safe, so you must ensure that the signal handler did not interrupt another async-signal-unsafe function) or use cancellation (pthread_cancel) to stop the blocked thread.
This should work for normal blocking operations (like waiting for data from a pipe or socket), but it will not help you if the thread is in an uninterruptable sleep state (like trying to read from a badly scratched CD or failing hard disk).
On POSIX systems, termination signals usually have the following order (according to many MAN pages and the POSIX Spec):
SIGTERM - politely ask a process to terminate. It shall terminate gracefully, cleaning up all resources (files, sockets, child processes, etc.), deleting temporary files and so on.
SIGQUIT - more forceful request. It shall terminate ungraceful, still cleaning up resources that absolutely need cleanup, but maybe not delete temporary files, maybe write debug information somewhere; on some system also a core dump will be written (regardless if the signal is caught by the app or not).
SIGKILL - most forceful request. The process is not even asked to do anything, but the system will clean up the process, whether it like that or not. Most likely a core dump is written.
How does SIGINT fit into that picture? A CLI process is usually terminated by SIGINT when the user hits CRTL+C, however a background process can also be terminated by SIGINT using KILL utility. What I cannot see in the specs or the header files is if SIGINT is more or less forceful than SIGTERM or if there is any difference between SIGINT and SIGTERM at all.
UPDATE:
The best description of termination signals I found so far is in the GNU LibC Documentation. It explains very well that there is an intended difference between SIGTERM and SIGQUIT.
It says about SIGTERM:
It is the normal way to politely ask a program to terminate.
And it says about SIGQUIT:
[...] and produces a core dump when it terminates the process, just like a program error signal.
You can think of this as a program error condition “detected” by the user. [...]
Certain kinds of cleanups are best omitted in handling SIGQUIT. For example, if the program
creates temporary files, it should handle the other termination requests by deleting the temporary
files. But it is better for SIGQUIT not to delete them, so that the user can examine them in
conjunction with the core dump.
And SIGHUP is also explained well enough. SIGHUP is not really a termination signal, it just means the "connection" to the user has been lost, so the app cannot expect the user to read any further output (e.g. stdout/stderr output) and there is no input to expect from the user any longer. For most apps that mean they better quit. In theory an app could also decide that it goes into daemon mode when a SIGHUP is received and now runs as a background process, writing output to a configured log file. For most daemons already running in the background, SIGHUP usually means that they shall reexamine their configuration files, so you send it to background processes after editing config files.
However there is no useful explanation of SIGINT on this page, other than that it is sent by CRTL+C. Is there any reason why one would handle SIGINT in a different way than SIGTERM? If so what reason would this be and how would the handling be different?
SIGTERM and SIGKILL are intended for general purpose "terminate this process" requests. SIGTERM (by default) and SIGKILL (always) will cause process termination. SIGTERM may be caught by the process (e.g. so that it can do its own cleanup if it wants to), or even ignored completely; but SIGKILL cannot be caught or ignored.
SIGINT and SIGQUIT are intended specifically for requests from the terminal: particular input characters can be assigned to generate these signals (depending on the terminal control settings). The default action for SIGINT is the same sort of process termination as the default action for SIGTERM and the unchangeable action for SIGKILL; the default action for SIGQUIT is also process termination, but additional implementation-defined actions may occur, such as the generation of a core dump. Either can be caught or ignored by the process if required.
SIGHUP, as you say, is intended to indicate that the terminal connection has been lost, rather than to be a termination signal as such. But, again, the default action for SIGHUP (if the process does not catch or ignore it) is to terminate the process in the same way as SIGTERM etc. .
There is a table in the POSIX definitions for signal.h which lists the various signals and their default actions and purposes, and the General Terminal Interface chapter includes a lot more detail on the terminal-related signals.
man 7 signal
This is the convenient non-normative manpage of the Linux man-pages project that you often want to look at for Linux signal information.
Version 3.22 mentions interesting things such as:
The signals SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
and contains the table:
Signal Value Action Comment
----------------------------------------------------------------------
SIGHUP 1 Term Hangup detected on controlling terminal
or death of controlling process
SIGINT 2 Term Interrupt from keyboard
SIGQUIT 3 Core Quit from keyboard
SIGILL 4 Core Illegal Instruction
SIGABRT 6 Core Abort signal from abort(3)
SIGFPE 8 Core Floating point exception
SIGKILL 9 Term Kill signal
SIGSEGV 11 Core Invalid memory reference
SIGPIPE 13 Term Broken pipe: write to pipe with no
readers
SIGALRM 14 Term Timer signal from alarm(2)
SIGTERM 15 Term Termination signal
SIGUSR1 30,10,16 Term User-defined signal 1
SIGUSR2 31,12,17 Term User-defined signal 2
SIGCHLD 20,17,18 Ign Child stopped or terminated
SIGCONT 19,18,25 Cont Continue if stopped
SIGSTOP 17,19,23 Stop Stop process
SIGTSTP 18,20,24 Stop Stop typed at tty
SIGTTIN 21,21,26 Stop tty input for background process
SIGTTOU 22,22,27 Stop tty output for background process
which summarizes signal Action that distinguishes e.g. SIGQUIT from SIGQUIT, since SIGQUIT has action Core and SIGINT Term.
The actions are documented in the same document:
The entries in the "Action" column of the tables below specify the default disposition for each signal, as follows:
Term Default action is to terminate the process.
Ign Default action is to ignore the signal.
Core Default action is to terminate the process and dump core (see core(5)).
Stop Default action is to stop the process.
Cont Default action is to continue the process if it is currently stopped.
I cannot see any difference between SIGTERM and SIGINT from the point of view of the kernel since both have action Term and both can be caught. It seems that is just a "common usage convention distinction":
SIGINT is what happens when you do CTRL-C from the terminal
SIGTERM is the default signal sent by kill
Some signals are ANSI C and others not
A considerable difference is that:
SIGINT and SIGTERM are ANSI C, thus more portable
SIGQUIT and SIGKILL are not
They are described on section "7.14 Signal handling " of the C99 draft N1256:
SIGINT receipt of an interactive attention signal
SIGTERM a termination request sent to the program
which makes SIGINT a good candidate for an interactive Ctrl + C.
POSIX 7
POSIX 7 documents the signals with the signal.h header: https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/signal.h.html
This page also has the following table of interest which mentions some of the things we had already seen in man 7 signal:
Signal Default Action Description
SIGABRT A Process abort signal.
SIGALRM T Alarm clock.
SIGBUS A Access to an undefined portion of a memory object.
SIGCHLD I Child process terminated, stopped,
SIGCONT C Continue executing, if stopped.
SIGFPE A Erroneous arithmetic operation.
SIGHUP T Hangup.
SIGILL A Illegal instruction.
SIGINT T Terminal interrupt signal.
SIGKILL T Kill (cannot be caught or ignored).
SIGPIPE T Write on a pipe with no one to read it.
SIGQUIT A Terminal quit signal.
SIGSEGV A Invalid memory reference.
SIGSTOP S Stop executing (cannot be caught or ignored).
SIGTERM T Termination signal.
SIGTSTP S Terminal stop signal.
SIGTTIN S Background process attempting read.
SIGTTOU S Background process attempting write.
SIGUSR1 T User-defined signal 1.
SIGUSR2 T User-defined signal 2.
SIGTRAP A Trace/breakpoint trap.
SIGURG I High bandwidth data is available at a socket.
SIGXCPU A CPU time limit exceeded.
SIGXFSZ A File size limit exceeded.
BusyBox init
BusyBox's 1.29.2 default reboot command sends a SIGTERM to processes, sleeps for a second, and then sends SIGKILL. This seems to be a common convention across different distros.
When you shutdown a BusyBox system with:
reboot
it sends a signal to the init process.
Then, the init signal handler ends up calling:
static void run_shutdown_and_kill_processes(void)
{
/* Run everything to be run at "shutdown". This is done _prior_
* to killing everything, in case people wish to use scripts to
* shut things down gracefully... */
run_actions(SHUTDOWN);
message(L_CONSOLE | L_LOG, "The system is going down NOW!");
/* Send signals to every process _except_ pid 1 */
kill(-1, SIGTERM);
message(L_CONSOLE, "Sent SIG%s to all processes", "TERM");
sync();
sleep(1);
kill(-1, SIGKILL);
message(L_CONSOLE, "Sent SIG%s to all processes", "KILL");
sync();
/*sleep(1); - callers take care about making a pause */
}
which prints to the terminal:
The system is going down NOW!
Sent SIGTERM to all processes
Sent SIGKILL to all processes
Here is a minimal concrete example of that.
Signals sent by the kernel
SIGKILL:
OOM killer: What is RSS and VSZ in Linux memory management
As DarkDust noted many signals have the same results, but processes can attach different actions to them by distinguishing how each signal is generated. Looking at the FreeBSD kernel source code (kern_sig.c) I see that the two signals are handled in the same way, they terminate the process and are delivered to any thread.
SA_KILL|SA_PROC, /* SIGINT */
SA_KILL|SA_PROC, /* SIGTERM */
After a quick Google search for sigint vs sigterm, it looks like the only intended difference between the two is whether it was initiated by a keyboard shortcut or by an explicit call to kill.
As a result, you could, for example, intercept sigint and do something special with it, knowing that it was likely sent by a keyboard shortcut. Perhaps refresh the screen or something, instead of dying (not recommended, as people expect ^C to kill the program, just an example).
I also learned that ^\ should send sigquit, which I may start using myself. Looks very useful.
Using kill (both the system call and the utility) you can send almost any signal to any process, given you've got the permission. A process cannot distinguish how a signal came to life and who has sent it.
That being said, SIGINT really is meant to signal the Ctrl-C interruption, while SIGTERM is the general terminal signal. There is no concept of a signal being "more forceful", with the only exception that there are signals that cannot be blocked or handled (SIGKILL and SIGSTOP, according to the man page).
A signal can only be "more forceful" than another signal with respect to how a receiving process handles the signal (and what the default action for that signal is). For example, by default, both SIGTERM and SIGINT lead to termination. But if you ignore SIGTERM then it will not terminate your process, while SIGINT still will.
With the exception of a few signals, signal handlers can catch the various signals, or the default behavior upon receipt of a signal can be modified. See the signal(7) man page for details.