PID namespaces isolate the process ID number space. The first process created in a new PID namespace will be PID 1.
jmuia@host$ sudo unshare --pid --fork
root@unshare# echo $$
1
However, if you ask ps it will tell you otherwise.
jmuia@host$ sudo unshare --pid --fork
root@unshare# ps aux
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 1 22.7 0.5 38000 6044 ? Ss 00:20 0:26 /sbin/init
root 2 0.0 0.0 0 0 ? S 00:20 0:00 [kthreadd]
...
root 2038 0.0 0.0 5996 736 pts/0 S 00:22 0:00 unshare --pid --fork
root 2039 3.0 0.4 21288 5052 pts/0 S 00:22 0:00 -bash
root 2053 0.0 0.3 36056 3236 pts/0 R+ 00:22 0:00 ps aux
On Linux, the ps command works by reading files in /proc, so it won’t be accurate if you’ve only entered a new PID namespace. The trick is to enter a new mount namespace and mount a new proc file system (since it inherits the caller’s mount list).
An interesting thing, though, is that you can’t interact with a host PID even though you can see it in /proc.
# htop running on host
jmuia@host$ ps aux | grep htop
jmuia 10934 1.4 0.0 27524 4880 pts/1 S+ 20:54 0:01 htop
# can still see htop in /proc from the PID namespace
jmuia@host$ sudo unshare --pid --fork
root@unshare# ps aux | grep htop
jmuia 10934 1.4 0.0 27524 4880 pts/1 S+ 20:54 0:02 htop
# but we can't kill it, because that PID doesn't exist in our namespace
root@unshare# kill -9 10934
-bash: kill: (10934) - No such process