[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
bug#52835: [PATCH v4 0/4] Improve safety of start_child and piped-proces
bug#52835: [PATCH v4 0/4] Improve safety of start_child and piped-process.
Sat, 28 May 2022 14:46:30 +0200
retitle 52835 Improve safety of start_child and piped-process.
This time, it's another Guix bug  that prompted me to have a closer
look at piped-process and start_child, which don't seem to be very
multi-thread safe. I've ended up with a couple of improvements that
IMO would make all procedures relying on them more robust. Here's
roughly what I did:
* Fix the fd closing code that was bogus for unusual values for in,
out, err for start_child.
* Check for double closes and avoid them, so that we don't
accidentally close an fd that another thread could have opened.
* Remove some closing code in the child, since we're already
generically closing all fds.
* Add a pipe from the child to the parent that the former uses to
report its errno to the latter. This avoids the use of strerror and
printf in the child after forking, since they are not async-signal
safe. As a side effect, this lets piped-error raise the proper
system exception for the child errno, instead of returning the PID
of a process that hasn't exec'd successfully.
Josselin Poiret (4):
Fix child spawning closing standard fds prematurely.
Avoid double closes in piped-process.
Remove useless closing code in start_child.
Make start_child propagate the child errno to the parent.
configure.ac | 3 +-
libguile/posix.c | 187 ++++++++++++++++++++++++++++++++++-------------
2 files changed, 138 insertions(+), 52 deletions(-)
- bug#52835: [PATCH v4 0/4] Improve safety of start_child and piped-process.,
Josselin Poiret <=