Bug 15496

On Wed, 15 Dec 1999 15:48:18 GMT, m.seaman@inpharmatica.co.uk wrote:

> 	killall(1) reads procfs /proc/${pid}/status entries to find
> the process names the user wants to kill.

Can we get the process names anywhere else?

Ciao,
Sheldon.

Sheldon Hearn wrote:
> 
> On Wed, 15 Dec 1999 15:48:18 GMT, m.seaman@inpharmatica.co.uk wrote:
> 
> >       killall(1) reads procfs /proc/${pid}/status entries to find
> > the process names the user wants to kill.
> 
> Can we get the process names anywhere else?

Hmmm... The only other prospect in /proc is the 'cmdline' entry, but that has
the same 16 character limitation as the status entry.

st-pancras:/tmp:% ./12345678901234567 & 
[1] 79007
st-pancras:/tmp:% cd /proc/79007
st-pancras:/proc/79007:% cat cmdline 
1234567890123456st-pancras:/proc/79007:% 

Failing that, we could parse the output of /bin/ps -ax --- easily doable from
perl, but dependent on processes not mangling argv[0] in unforeseen ways (ps
-axc has the same 16 character limitation as above).

Otherwise, we'd be looking at reading kvm the way ps(1) does, which is
not much fun from perl: it would probably be necessary to implement a
FreeBSD::KVM module using the perl '.xs' code thingy to generate an interface
to the kvm functions.  Plus we would have to use suidperl to run the script
setgid to kmem, which doesn't strike me as particularly desirable.

Hmmm... here's a patch to killall(1) to use ps(1) output as well as reading
/proc:  ps(1) output is used if the given program-name-to-kill is longer than
16 characters, or a match could not otherwise be made, to distinguish between
similarly named processes around 16 characters long. eg:

st-pancras:~:% /tmp/1234567890123456 & 
[1] 79288
st-pancras:~:% /tmp/12345678901234567 &
[2] 79289
st-pancras:~:% ./killall.pl -v 12345678901234567
kill -TERM 79289
[2]  - Terminated                    /tmp/12345678901234567
st-pancras:~:% ./killall.pl -v 1234567890123456
kill -TERM 79288
[1]  + Terminated                    /tmp/1234567890123456


 diff -u killall.pl.orig killall.pl 
--- killall.pl.orig  Tue Aug 31 10:08:27 1999
+++ killall.pl     Fri Dec 17 15:19:39 1999
@@ -30,11 +30,13 @@
 
 
 $ENV{'PATH'} = '/bin:/usr/bin'; # security
+$pscmd = '/bin/ps -ax';                # list all processes
 $procfs = '/proc';
 $signal = 'SIGTERM';           # default signal for kill
 $debug = 0;
 $match = 0;                    # 0 match exactly program name
 $show = 0;                     # do nothings
+%procs = ();                   # Lookup processes by PID
 
 # see /sys/miscfs/procfs/procfs_status.c
 $PROC_NAME =  0;
@@ -68,6 +70,23 @@
 undef %pidu;
 undef %thiskill;
 
+open(PS, "$pscmd|") || die "$pscmd $!\n";
+while (<PS>) {
+    my ($pid, $proc);
+    chomp;
+
+    # Processes can alter their argv[0], which will be reflected in
+    # ps(1) output, so this is by no means reliable.  Typically login
+    # shells will have `-' pre-pended.  System processes like swapper
+    # are usually in (brackets) and then there's sendmail and nfsd that
+    # use argv[0] for status messages.
+
+    if (($pid, $proc) = m/^\s*(\d+)\s+\S+\s+\S+\s+\S+\s+(\S+)/) {
+         ($procs{$pid} = $proc) =~ s@.*/@@;    # Strip any leading path
+    }
+}
+close(PS);
+
 foreach (sort{$a <=> $b} grep(/^[0-9]/, readdir(PROCFS))) {
     $status = "$procfs/$_/status";
     $pid = $_;
@@ -85,8 +104,12 @@
            ($programMatch = $program) =~ s/(\W)/\\$1/g if $match;
 
             # match program name
-           if ($proc[$PROC_NAME] eq $program ||
-               ($match && $proc[$PROC_NAME] =~ /$programMatch/i)) {
+           if ( (length $program <= 16 && 
+                 ($proc[$PROC_NAME] eq $program ||
+                  ($match && $proc[$PROC_NAME] =~ /$programMatch/i))) ||
+                (defined $procs{$pid} && 
+                 ($procs{$pid} eq $program ||
+                  ($match && $procs{$pid} =~ /$programMatch/i)))) {
 
                # id test
                if ($proc[$PROC_EUID] eq $id || # effective uid


-- 
           Certe, Toto, sentio nos in Kansate non iam adesse.

   Dr. Matthew Seaman, Inpharmatica Ltd, 60 Charlotte St, London, W1P 2AX
            Tel: +44 171 631 4644 x229  Fax: +44 171 631 4844

Since killall(1) has recently transmogrified into a C program, I guess this PR
might be closable.  However, the same bug still seems to be present:

st-pancras:/tmp:# uname -a
FreeBSD st-pancras.inpharmatica.co.uk 4.1.1-STABLE FreeBSD 4.1.1-STABLE #0:
Tue Sep 26 12:32:33 BST 2000    
root@st-pancras.inpharmatica.co.uk:/export/i386/obj/export/src/sys/ST-PANCRAS 
i386
st-pancras:/tmp:# file /usr/bin/killall
/usr/bin/killall: ELF 32-bit LSB executable, Intel 80386, version 1 (FreeBSD),
dynamically linked, stripped
st-pancras:/tmp:# ./123456789012345 & 
[1] 249
st-pancras:/tmp:# ./1234567890123456 &
[2] 250
st-pancras:/tmp:# ./12345678901234567 &
[3] 251
st-pancras:/tmp:# head /proc/{249,250,251}/status 
==> /proc/249/status <==
123456789012345 249 227 249 227 5,0 ctty 969973928,166376 0,0 0,6293 nanslp 0
0 0,0,0,2,3,4,5,20,31,994,998 -

==> /proc/250/status <==
1234567890123456 250 227 250 227 5,0 ctty 969973930,750074 0,0 0,6484 nanslp 0
0 0,0,0,2,3,4,5,20,31,994,998 -

==> /proc/251/status <==
1234567890123456 251 227 251 227 5,0 ctty 969973934,657696 0,0 0,6582 nanslp 0
0 0,0,0,2,3,4,5,20,31,994,998 -
st-pancras:/tmp:# killall 12345678901234567 
No matching processes were found
st-pancras:/tmp:# killall 1234567890123456
[3]  - Terminated                    ./12345678901234567
[2]  - Terminated                    ./1234567890123456

	Cheers,

	Matthew

-- 
           Certe, Toto, sentio nos in Kansate non iam adesse.

   Dr. Matthew Seaman, Inpharmatica Ltd, 60 Charlotte St, London, W1T 2NU
            Tel: +44 20 7631 4644 x229  Fax: +44 20 7631 4844

Comment 4 dd 2001-06-18 23:30:44 UTC

State Changed
From-To: open->closed

killall doesn't use procfs anymore.