These are the only two calls to log.Fatal* reachable during suspended output. Call fmsg.BeforeExit here to catch that.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This removes the requirement to call fmsg.Exit on every exit path, and enables direct use of the "log" package. However, fmsg.BeforeExit is still encouraged when possible to catch exit on suspended output.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This interface is ugly and bug-prone. This change removes its write mode which has been obsoleted by CopyBind.
Signed-off-by: Ophestra <cat@gensokyo.uk>
These are significantly more efficient and less error-prone than mounting an external tmpfile. This should also reduce attack surface as the resulting files are private to its specific sandbox.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This method does not do anything and is not called from anywhere. It also does not make any sense as a public interface since the argument builder is no longer stateless.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Storing extra files as a slice requires the caller to allocate a large enough slice before initialising any file and never grow the slice.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Turns out the doc comment on os.File was lying about its methods being safe for concurrent use. The race detector picked up a data race from concurrent use of Fd and Close.
This change eliminates that by calling Fd in the prepare routine.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Helpers use extra files for various purposes. This provides a generic interface for implementing the fulfillment of these extra files without having to specifically handle them in the process creation code.
Signed-off-by: Ophestra <cat@gensokyo.uk>
In the current app implementation this gets dumped in the wait method after resuming output. Wait is never called in an early fault condition, so any error messages get lost.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Returning an error here puts exporter in an invalid state. The caller should guard against this condition instead.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This also does not require the libc tmpfile call.
BPF programs emitted by libseccomp seems to be deterministic. The tests would catch regressions as it verifies the program against known good output backed by manual testing.
Signed-off-by: Ophestra <cat@gensokyo.uk>
