Which PID is using a PORT inside a k8s pod without net tools - linux

Question

Sorry about the long question post, but I think it can be useful to others to learn how this works.
What I know:
On any linux host (not using docker container), I can look at /proc/net/tcp to extract information tcp socket related.
So, I can detect the ports in LISTEN state with:
cat /proc/net/tcp |
grep " 0A " |
sed 's/^[^:]*: \(..\)\(..\)\(..\)\(..\):\(....\).*/echo $((0x\4)).$((0x\3)).$((0x\2)).$((0x\1)):$((0x\5))/g' |
bash
Results:
0.0.0.0:111
10.174.109.1:53
127.0.0.53:53
0.0.0.0:22
127.0.0.1:631
0.0.0.0:8000
/proc/net/tcp gives UID, GID, unfortunately does not provides the PID. But returns the inode. That I can use to discover the PID using it as file descriptor.
So one way is to search /proc looking for the inode socket. It's slow, but works on host:
cat /proc/net/tcp |
grep " 0A " |
sed 's/^[^:]*: \(..\)\(..\)\(..\)\(..\):\(....\).\{72\}\([^ ]*\).*/echo $((0x\4)).$((0x\3)).$((0x\2)).$((0x\1)):$((0x\5))\\\t$(find \/proc\/ -type d -name fd 2>\/dev\/null \| while read f\; do ls -l $f 2>\/dev\/null \| grep -q \6 \&\& echo $f; done)/g' |
bash
output:
0.0.0.0:111 /proc/1/task/1/fd /proc/1/fd /proc/924/task/924/fd /proc/924/fd
10.174.109.1:53 /proc/23189/task/23189/fd /proc/23189/fd
127.0.0.53:53 /proc/923/task/923/fd /proc/923/fd
0.0.0.0:22 /proc/1194/task/1194/fd /proc/1194/fd
127.0.0.1:631 /proc/13921/task/13921/fd /proc/13921/fd
0.0.0.0:8000 /proc/23122/task/23122/fd /proc/23122/fd
Permission tip 1: You will only see what you have permission to look at.
Permission tip 2: fake root used in containers does not have access to all file descriptors in /proc/*/fd. You need to query it for each user.
If you run as normal user the results are:
0.0.0.0:111
10.174.109.1:53
127.0.0.53:53
0.0.0.0:22
127.0.0.1:631
0.0.0.0:8000 /proc/23122/task/23122/fd /proc/23122/fd
Using unshare to isolate environment it works as expected:
$ unshare -r --fork --pid unshare -r --fork --pid --mount-proc -n bash
# ps -fe
UID PID PPID C STIME TTY TIME CMD
root 1 0 2 07:19 pts/6 00:00:00 bash
root 100 1 0 07:19 pts/6 00:00:00 ps -fe
# netstat -ntpl
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
# python -m SimpleHTTPServer &
[1] 152
# Serving HTTP on 0.0.0.0 port 8000 ...
netstat -ntpl
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
tcp 0 0 0.0.0.0:8000 0.0.0.0:* LISTEN 152/python
# cat /proc/net/tcp |
> grep " 0A " |
> sed 's/^[^:]*: \(..\)\(..\)\(..\)\(..\):\(....\).\{72\}\([^ ]*\).*/echo $((0x\4)).$((0x\3)).$((0x\2)).$((0x\1)):$((0x\5))\\\t$(find \/proc\/ -type d -name fd 2>\/dev\/null \| while read f\; do ls -l $f 2>\/dev\/null \| grep -q \6 \&\& echo $f; done)/g' |
> bash
0.0.0.0:8000 /proc/152/task/152/fd /proc/152/fd
# ls -l /proc/152/fd
total 0
lrwx------ 1 root root 64 mai 25 07:20 0 -> /dev/pts/6
lrwx------ 1 root root 64 mai 25 07:20 1 -> /dev/pts/6
lrwx------ 1 root root 64 mai 25 07:20 2 -> /dev/pts/6
lrwx------ 1 root root 64 mai 25 07:20 3 -> 'socket:[52409024]'
lr-x------ 1 root root 64 mai 25 07:20 7 -> /dev/urandom
# cat /proc/net/tcp
sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode
0: 00000000:1F40 00000000:0000 0A 00000000:00000000 00:00000000 00000000 0 0 52409024 1 0000000000000000 100 0 0 10 0
Inside a docker container in my host, it seems to work in same way.
The problem:
I have a container inside a kubernetes pod running jitsi. Inside this container, I am unable to get the PID of the service listening the ports.
Nor after installing netstat:
root#jitsi-586cb55594-kfz6m:/# netstat -ntpl
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
tcp 0 0 0.0.0.0:5222 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:80 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:5269 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:8888 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:443 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:5280 0.0.0.0:* LISTEN -
tcp 0 0 0.0.0.0:5347 0.0.0.0:* LISTEN -
tcp6 0 0 :::5222 :::* LISTEN -
tcp6 0 0 :::5269 :::* LISTEN -
tcp6 0 0 :::5280 :::* LISTEN -
# ps -fe
UID PID PPID C STIME TTY TIME CMD
root 1 0 0 May22 ? 00:00:00 s6-svscan -t0 /var/run/s6/services
root 32 1 0 May22 ? 00:00:00 s6-supervise s6-fdholderd
root 199 1 0 May22 ? 00:00:00 s6-supervise jicofo
jicofo 203 199 0 May22 ? 00:04:17 java -Xmx3072m -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp -Dnet.java.sip.communicator.SC_HOME_DIR_LOCATION=/ -Dnet.java.sip.communicator.SC_HOME_DIR_NAME=config -Djava
root 5990 0 0 09:48 pts/2 00:00:00 bash
root 10926 5990 0 09:57 pts/2 00:00:00 ps -fe
Finally the Questions:
a) Why can't I read the file descriptors of the proccess listening port 5222 ?
root#jitsi-586cb55594-kfz6m:/# cat /proc/net/tcp | grep " 0A "
0: 00000000:1466 00000000:0000 0A 00000000:00000000 00:00000000 00000000 101 0 244887827 1 ffff9bd749145800 100 0 0 10 0
...
root#jitsi-586cb55594-kfz6m:/# echo $(( 0x1466 ))
5222
root#jitsi-586cb55594-kfz6m:/# ls -l /proc/*/fd/* 2>/dev/null | grep 244887827
root#jitsi-586cb55594-kfz6m:/# echo $?
1
root#jitsi-586cb55594-kfz6m:/# su - svc
svc#jitsi-586cb55594-kfz6m:~$ id -u
101
svc#jitsi-586cb55594-kfz6m:~$ ls -l /proc/*/fd/* 2>/dev/null | grep 244887827
svc#jitsi-586cb55594-kfz6m:~$ echo $?
1
b) There is another way to list inode and link it to a pid without searching /proc/*/fd ?
Update 1:
Based on Anton Kostenko tip, I looked to AppArmor. It's not the case because the server don't use AppArmor, but searching, took me to SELinux.
In a ubuntu machine where AppArmor is running, I got:
$ sudo apparmor_status | grep dock
docker-default
In the OKE(Oracle Kubernetes Engine, my case) node there is no AppArmor. I got SELinux instead:
$ man selinuxenabled | grep EXIT -A1
EXIT STATUS
It exits with status 0 if SELinux is enabled and 1 if it is not enabled.
$ selinuxenabled && echo $?
0
Now, I do believe that SELinux is blocking the /proc/*/fd listing from root inside the container. But I don't know yet how to unlock it.
References:
https://jvns.ca/blog/2016/10/10/what-even-is-a-container/

Answer 1

The issue is solved by adding the POSIX capability: CAP_SYS_PTRACE
I'm my case the container are under kubernetes orchestration.
this reference explains about kubectl and POSIX Capabilities
So I have
root#jitsi-55584f98bf-6cwpn:/# cat /proc/1/status | grep Cap
CapInh: 00000000a80425fb
CapPrm: 00000000a80425fb
CapEff: 00000000a80425fb
CapBnd: 00000000a80425fb
CapAmb: 0000000000000000
So I careful read the POSIX Capabilities Manual. But even adding CAP_SYS_ADMIN, the PID does not appear on netstat. So I tested all capabilities. CAP_SYS_PTRACE is The Chosen One
root#jitsi-65c6b5d4f7-r546h:/# cat /proc/1/status | grep Cap
CapInh: 00000000a80c25fb
CapPrm: 00000000a80c25fb
CapEff: 00000000a80c25fb
CapBnd: 00000000a80c25fb
CapAmb: 0000000000000000
So here my deployment spec change:
...
spec:
...
template:
...
spec:
...
containers:
...
securityContext:
capabilities:
add:
- SYS_PTRACE
...
Yet I don't know what security reasons selinux use to do it. But for now it's good enough for me.
References:
https://man7.org/linux/man-pages/man7/capabilities.7.html
https://kubernetes.io/docs/tasks/configure-pod-container/security-context/