kill %jobspec tries to kill dead processes

kill %jobspec tries to kill dead processes

[Stephane Chazelas]

Hi,

kill %jobspec

usually avoids the problem when kill $pid may kill the wrong
process when it may have been reused. However, in:

sleep 1 | sleep 3 & sleep 2
kill %"sleep 1"

kill still tries to kill the process that was running sleep 1
even though the shell knows it has died:

$ strace -e kill zsh -c 'sleep 1 | sleep 3 & ps -f; sleep 2; printf "%s => %s\n" "${(@kv)jobstates}"; kill %"sleep 1"'
UID        PID  PPID  C STIME TTY          TIME CMD
chazelas  8175 14816  0 12:37 pts/4    00:00:00 strace -e kill zsh -c sleep 1 | sleep 3 & ps -f; sleep 2; printf "%s => %s\n" "${(@kv)jobstates}"; kill %"sleep 1"
chazelas  8177  8175  0 12:37 pts/4    00:00:00 zsh -c sleep 1 | sleep 3 & ps -f; sleep 2; printf "%s => %s\n" "${(@kv)jobstates}"; kill %"sleep 1"
chazelas  8178  8177  0 12:37 pts/4    00:00:00 sleep 1
chazelas  8179  8177  0 12:37 pts/4    00:00:00 sleep 3
chazelas  8180  8177  0 12:37 pts/4    00:00:00 ps -f
chazelas 14816 14812  0 Jan03 pts/4    00:00:01 /bin/zsh
--- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=8180, si_uid=1000, si_status=0, si_utime=0, si_stime=2} ---
--- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=8178, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
--- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=8181, si_uid=1000, si_status=0, si_utime=0, si_stime=0} ---
1 => running:+:8178=done:8179=running
kill(8178, SIGTERM)                     = -1 ESRCH (No such process)
kill(8179, SIGTERM)                     = 0
+++ exited with 0 +++

See how 8178 was known to be "done", but is still killed.

mksh has the same issue. bash is worse in that it doesn't
attempt to kill the sleep 3 process. ksh93 segfaults when I try
the same thing there.

-- 
Stephane

Re: kill %jobspec tries to kill dead processes

[Stephane Chazelas]

2018-01-05 12:46:02 +0000, Stephane Chazelas:
[...]
> sleep 1 | sleep 3 & sleep 2
> kill %"sleep 1"
> 
> kill still tries to kill the process that was running sleep 1
> even though the shell knows it has died:
[...]

I beleive the patch below should do it. I added an unrelated
comment about the "sig != 0". I guess we could skip the killing
altogether for sig == 0, but I left it in on the basis of "who
are we do deny the user's request even if it's pointless".

diff --git a/Src/signals.c b/Src/signals.c
index 94f379e..e06fc05 100644
--- a/Src/signals.c
+++ b/Src/signals.c
@@ -779,8 +779,17 @@ killjb(Job jn, int sig)
 	    return killpg(jn->gleader, sig);
     }
     for (pn = jn->procs; pn; pn = pn->next)
-        if ((err = kill(pn->pid, sig)) == -1 && errno != ESRCH && sig != 0)
-            return -1;
+	/*
+	 * Do not kill this job's process if it's already dead as its
+	 * pid could have been reused by the system.
+	 */
+	if (pn->status == SP_RUNNING || WIFSTOPPED(pn->status))
+	    /*
+	     * kill -0 on a job is pointless. We still call kill() for each process
+	     * in case the user cares about it but we ignore its outcome.
+	     */
+	    if ((err = kill(pn->pid, sig)) == -1 && errno != ESRCH && sig != 0)
+		return -1;
     return err;
 }