67

I am familiar with kill command , and most of the time we just use kill -9 to kill a process forcefully, there are many other signals that can be used with kill. But I wonder what are the use cases of pkill and killall, if there is already a kill command.

Do pkill and killall use the kill command in their implementation? I mean they are just wrappers over kill or they have their own implementation?

I would also like to know how pgrep command gets the process id from the process name.

Do all these commands use the same underlying system calls? Is there any difference from a performance point of view, which one is faster?

Milan
  • 153
  • 1
  • 6
Ijaz Ahmad
  • 7,146
  • 5
  • 32
  • 45
  • 7
    Two things: Why use `kill -9` by default? -15 (please stop) and -1 (modem has hung up, please CLEANLY closed yourself) are much more polite. Secondly. Beware of using killall on non-linux boxes. It might behave differently. (E.g. on solaris it kills all. NOT FILTERED on process names). – Hennes Dec 30 '15 at 16:49
  • 4
    Related: https://unix.stackexchange.com/questions/91527/whats-the-difference-between-pkill-and-killall – Byte Commander Jun 25 '19 at 09:28

3 Answers3

62

The kill command is a very simple wrapper to the kill system call, which knows only about process IDs (PIDs). pkill and killall are also wrappers to the kill system call, (actually, to the libc library which directly invokes the system call), but can determine the PIDs for you, based on things like, process name, owner of the process, session id, etc.

How pkill and killall work can be seen using ltrace or strace on them. On Linux, they both read through the /proc filesystem, and for each pid (directory) found, traverses the path in a way to identify a process by its name or other attributes. How this is done is technically speaking, kernel and system specific. In general, they read from /proc/<PID>/stat which contains the command name as the 2nd field. For pkill -f and pgrep examine the /cmdline entry for each PID's proc entry.

pkill and pgrep use the readproc system call, whereas killall does not. I couldn't say if there's a performance difference: you'll have to benchmark that on your own.

Otheus
  • 5,945
  • 1
  • 22
  • 53
  • `readproc` is not a system call. It's a libprocps library function. –  Sep 19 '20 at 14:33
8

A more practical answer:

  • kill will send a SIGTERM (terminate a process) signal to the OS for a given process id (e.g. 10341), by default this is SIGTERM. You can actually specify other signals too (e.g. -9 to force terminate)

  • killall will send SIGTERM to all processes matching a certain name exactly. It also comes wit other goodies such as subsetting by user.

  • pkill will send SIGTERM to all processes that match a given pattern. It uses pgrep under the hood.

fny
  • 373
  • 3
  • 7
3

kill and killall are tools which provide a way to kill a process. The first by its PID, the second by its name. pgrep (list) and pkill (kill by default) are tools which provide a way to send message to a process by its name or other attributes see: http://linux.die.net/man/1/pkill For more info about signals: http://linux.die.net/man/7/signal

dervishe
  • 465
  • 1
  • 5
  • 14
  • Do you mean `killall` is used to kill a process by its name? and it uses pgrep for this purpose? and `killall` also kills all the child processes ,? what signal `killall` uses by default? – Ijaz Ahmad Dec 30 '15 at 16:23
  • 3
    ...unless you're on Solaris, in which case `killall` will kill *all* processes you have the right to kill, so that if you're root, you're effectively rebooting the server. – Jenny D Dec 30 '15 at 16:25
  • 1
    yep: `killall chromium` will kill the chromium process, `pgrep chromium` will give you the PID list, `pkill chromium` will kill chromium. killall will send by default SIGTERM signal (as pkill) – dervishe Dec 30 '15 at 16:28