Why is CapEff all zeros in /proc/$PID/status

Question

I removed the setuid bit from the ping binary and added cap_net_raw+p instead as follows:

$ chmod 755 /bin/ping
$ setcap cap_net_raw+p /bin/ping

Then I ran ping in one terminal and checked the /proc/$PID/status of the running process from another:

$ ps aux | grep ping                                                                   
  user     5468  0.0  0.0  14948  1792 pts/20   S+   11:14   0:00 ping www.google.com
  user     5471  0.0  0.0  14224   896 pts/2    S+   11:14   0:00 grep --color=auto ping
$ cat /proc/5468/status | grep Cap
  CapInh: 0000000000000000
  CapPrm: 0000000000002000
  CapEff: 0000000000000000
  CapBnd: 0000003fffffffff
  CapAmb: 0000000000000000

If ping is currently running why is CapEff: 0000000000000000? Shouldn't cap_net_raw be in the effective set as well? Does /proc/$PID/status not reflect the current state of the thread?

Answer 1

A helpful person in #kernelnewbies on OFTC (irc) was kind enough to provide me with the answer.

ping sets cap_net_raw in the effective set, creates the socket, then drops cap_net_raw, as can been seen with strace:

$ strace -e socket,capset ping -c1 localhost
  capset({_LINUX_CAPABILITY_VERSION_3, 0}, {CAP_NET_RAW, CAP_NET_ADMIN|CAP_NET_RAW, 0}) = 0
  socket(PF_INET, SOCK_RAW, IPPROTO_ICMP) = 3
  capset({_LINUX_CAPABILITY_VERSION_3, 0}, {0, CAP_NET_ADMIN|CAP_NET_RAW, 0}) = 0

Once the socket is open, no more privileges are required to write to it.