Re: [RFC] unix: fix use-after-free in unix_dgram_poll()/ 4.2.5

Olivier Mauras [off-list ref] writes:

[...]

I've encountered  issues with Jason's patch ported to 3.14.x which would break
openldap, rendering it unable to answer any query - Here's a strace of the
slapd process in this state http://pastebin.ca/3226383
Just ported Rainer's patch to 3.14 and so far I can't reproduce the issue -

I may be missing something here but the final state according to the
trace it that thread 775 of the process blocks in epoll_wait with a
descriptor set containing only a listening TCP socket (8) and waiting
for new connections. I don't think this can execute any code
changed by my patch and I'm fairly certain for this for Jason's, too:
Both are about AF_UNIX datagram sockets and the specific case where
either a write couldn't complete because the backlog of the receive
queue of the 1 side of a n:1 datagram socket arrangement was considered
too large or where a 'poll for write' check returned 'not writeable' for
the same reason.

Judging from the 2.4.42 sources, OpenLDAP doesn't use AF_UNIX datagram
sockets at all so it shouldn't ever be affected by any changes to the
code handling them.