Re: [developer] Review - 5798 fexecve() needed per POSIX 2008

[Robert Mustacchi <rm@fingolfin.org>]

On 1/12/24 16:45, Cedric Blancher wrote:
> On Fri, 12 Jan 2024 at 15:06, Andy Fiddaman <illumos@fiddaman.net> wrote:
>>
>> Hello,
>>
>> I have a project coming up that would benefit from fexecve(2), and this
>> function is also part of POSIX 2008 so it .
>>
>> I've taken Garrett's work in
>>     http://cr.illumos.org/~webrev/gdamore/fexecve/
>> and put it up for review. It takes an  approach of using a consed up
>> /dev/fd path versus introducing a new system call such as execveat()
>> in order to reduce risk and knock on impact to auditing etc. This seems
>> like a reasonable trade off to me but I'm interested in views from others.
> 
> Unfortunately this is not going to work. The whole point of fexecve()
> is to make tools like shells more efficient, but relying on /dev/fd
> means that your shell will no longer work at boot time, or in a
> chroot, when /dev/fd is not available/mounted yet.

There is a bit more nuance here in the implementation. It does not rely
on /dev/fd being mounted or present in a chroot (though that is its own
question given what's happening in the kernel implementation). So all
the cases you want to work should if you actually look at the
implementation.

Robert