FD_CLOEXEC bug fixes (filters) + support (actions). Avoid sockets descriptors leak.

[ncollignon]

• 001-fail2ban-server-socket-close-on-exec-no-leak.diff

Add code that marks server and client sockets with FD_CLOEXEC flags.
Avoid leaking file descriptors to processes spawned when handling
fail2ban actions (ex: iptables).

Unix sockets managed by fail2ban-server don't need to be passed to any
child process. Fail2ban already uses the FD_CLOEXEC flags in the filter
code.

This patch also avoids giving iptables access to fail2ban UNIX socket in
a SELinux environment (A sane SELinux policy should trigger an audit
event because "iptables" will be given read/write access to the fail2ban
control socket).

Some random references related to this bug:
http://sourceforge.net/tracker/?func=detail&atid=689044&aid=2086568&group_id=121032
http://www.redhat.com/archives/fedora-selinux-list/2009-June/msg00124.html
http://forums.fedoraforum.org/showthread.php?t=234230

• 002-fail2ban-filters-close-on-exec-typo-fix.diff

There is a typo in the fail2ban server/filter.py source code. The
FD_CLOEXEC is correctly set but additional random flags are also set.
It has no side-effect as long as the fd doesn't match a valid flag :)
"fcntl.fcntl(fd, fcntl.F_SETFD, fd | fcntl.FD_CLOEXEC)" <== the 3rd
parameter should be flags, not a file descriptor.

• 003-fail2ban-gamin-socket-close-on-exec-no-leak.diff

Add code that marks the Gamin monitor file descriptor with FD_CLOEXEC
flags. Avoid leaking file descriptors to processes spawned when handling
fail2ban actions (ex: iptables).

---

File descriptors in action process before patches:
dr-x------ 2 root root 0 .
dr-xr-xr-x 8 root root 0 ..
lr-x------ 1 root root 64 0 -> /dev/null <== OK
l-wx------ 1 root root 64 1 -> /tmp/test.log <== used by test action
lrwx------ 1 root root 64 2 -> /dev/null <== OK
lrwx------ 1 root root 64 3 -> socket:[116361] <== NOK (fail2ban.sock leak)
lr-x------ 1 root root 64 4 -> /proc/20090/fd <== used by test action
l-wx------ 1 root root 64 5 -> /var/log/fail2ban.log <== OK
lrwx------ 1 root root 64 6 -> socket:[115608] <== NOK (gamin sock leak)

File descriptors in action process after patches:
dr-x------ 2 root root 0 .
dr-xr-xr-x 8 root root 0 ..
lr-x------ 1 root root 64 0 -> /dev/null <== OK
l-wx------ 1 root root 64 1 -> /tmp/test.log <== used by test action
lrwx------ 1 root root 64 2 -> /dev/null <== OK
lr-x------ 1 root root 64 3 -> /proc/18284/fd <== used by test action
l-wx------ 1 root root 64 5 -> /var/log/fail2ban.log <== OK

[coveralls]

Coverage remained the same when pulling 39667ff on ncollignon:master into f5ad99b on fail2ban:master.

View Details

[yarikoptic]

Looks good to me -- thanks . Just to make sure -- are these patches "user tested" already? e.g. they were picked up from some packages of fail2ban (in Fedora?)

would you mind crafting a little unittest e.g. at least for the 001-fail2ban-server-socket-close-on-exec-no-leak.diff portion?