Using sed to get an env var from /proc/*/environ weirdness with \x00

Question

I'm trying to grovel through some other processes environment to get a specific env var.

So I've been trying a sed command like:

sed -n "s/\x00ENV_VAR_NAME=\([^\x00]*\)\x00/\1/p" /proc/pid/environ

But I'm getting as output the full environ file. If I replace the \1 with just a static string, I get that string plus the entire environ file:

sed -n "s/\x00ENV_VAR_NAME=\([^\x00]*\)\x00/BLAHBLAH/p" /proc/pid/environ

I should just be getting "BLAHBLAH" in the last example. This doesn't happen if I get rid of the null chars and use some other test data set.

This lead me to try transforming the \x00 to \x01's, which does seem to work:

cat /proc/pid/environ | tr '\000' '\001' | sed -n "s/\x01ENV_VAR_NAME=\([^\x01]*\)\x01/\1/p"

Am I missing something simple about sed here? Or should I just stick to this workaround?

Answer 1

A lot of programs written in C tend to fail with strings with embedded NULs as a NUL terminates a C-style string. Unless specially written to handle it.

I process /proc/*/environ on the command line with xargs:

xargs -n 1 -0 < /proc/pid/environ

This gives you one env var per line. Without a command, xargs just echos the argument. You can then easily use grep, sed, awk, etc on that by piping to it.

xargs -n 1 -0 < /proc/pid/environ | sed -n 's/^ENV_VAR_NAME=\(.*\)/\1/p'

I use this often enough that I have a shell function for it:

pidenv() 
{ 
    xargs -n 1 -0 < /proc/${1:-self}/environ
}

This gives you the environment of a specific pid, or self if no argument is supplied.