fix process leakage in container delete when use share pid namespace

[lifubang]

If container b joins container a's pid namespace, once delete container b, it may cause leakage of process in containerb.

For example:

Use busybox image as rootfs, with start arg ["sleep", "100000"]:

For container a: use new pid namesapce without path: "namespaces": [{"type": "pid"}

root@demo:/opt/busybox# ../runc/runc run -d a
root@demo:/opt/busybox# ../runc/runc list
ID          PID         STATUS      BUNDLE                                            CREATED                          OWNER
a           3162        running     /opt/busybox   2019-04-08T11:34:02.449638033Z   root
root@demo:/opt/busybox# ../runc/runc ps a
UID        PID  PPID  C STIME TTY          TIME CMD
root      3162     1  0 19:34 ?        00:00:00 sleep 100000

For container b: use new pid namesapce with path: "namespaces": [{"type": "pid", "path": "/proc/3162/ns/pid"}

root@demo:/opt/busybox# ../runc/runc run -d b
root@demo:/opt/busybox# ../runc/runc list
ID          PID         STATUS      BUNDLE                                            CREATED                          OWNER
a           3162        running     /opt/busybox   2019-04-08T11:34:02.449638033Z   root
b           3581        running     /opt/busybox   2019-04-08T11:35:05.241568752Z   root
root@demo:/opt/busybox# ../runc/runc exec -d b sleep 20000
root@demo:/opt/busybox# ../runc/runc ps b
UID        PID  PPID  C STIME TTY          TIME CMD
root      3581     1  0 19:35 ?        00:00:00 sleep 100000
root      6256     1  0 19:42 ?        00:00:00 sleep 20000
root@demo:/opt/busybox# ../runc/runc kill b 9
root@demo:/opt/busybox# ../runc/runc list
ID          PID         STATUS      BUNDLE                                            CREATED                          OWNER
a           3162        running     /opt/busybox   2019-04-08T11:34:02.449638033Z   root
b           0           stopped     /opt/busybox   2019-04-08T11:35:05.241568752Z   root
root@demo:/opt/busybox# ../runc/runc delete b
root@demo:/opt/busybox# ps -ef | grep 6256
root      6256     1  0 19:42 ?        00:00:00 sleep 20000

And container b's cgroup path is not deleted.

Signed-off-by: Lifubang [email protected]

[lifubang]

There are 2 ways to fix this problem:

1. use signalAllProcesses when destroy the container;
2. throw the error to runc delete method.

For this PR, I choose the first way, if the maintainers feel the second way more better, I'll change the code.

[kolyshkin]

@lifubang can you please amend the commit message with the info about the repro (i.e. what you have in this PR description)?

Otherwise LGTM

[kolyshkin]

@cyphar @AkihiroSuda PTAL

[lifubang]

can you please amend the commit message with the info about the repro

updated

[fahedouch]

seems to be supported here. this should be closed cc @kolyshkin

[kolyshkin]

Yes, this was independently fixed by https://github.com/opencontainers/runc/pull/2085