Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 18 Aug 2022 13:40:36 +0300
From: Alexey Izbyshev <izbyshev@...ras.ru>
To: musl@...ts.openwall.com
Subject: Re: [PATCH] add close_range() syscall wrapper

On 2022-08-18 13:11, √Črico Nogueira wrote:
> On Wed Aug 17, 2022 at 9:35 PM -03, Rich Felker wrote:
>> On Wed, Aug 10, 2022 at 01:03:11PM +0000, Guilherme Janczak wrote:
>> > close_range() is a syscall present in FreeBSD 8.0 and Linux 5.9. glibc
>> > 2.34 added a wrapper.
>> > ---
>> 
>> The existence of this operation has been controversial, and it's
>> arguable that it should be excluded by policy not to support UB (it's
>> UB to close fds you don't own that might be used internally by the
>> implementation) though I'm not sure it really helps since folks who
>> want to use it will just make the syscall directly. We should probably
>> at least consider it for inclusion.
> 
> I remember an idea to implement fallback logic, in case the syscall is
> unavailable, had been mentioned. It would then avoid whatever fallback
> code the application tried to implement, which might not be as 
> relevant,
> now that opendir() can be called in a forked child. And I don't know if
> there's interest in implementing anything more complex at all.
> 
Glibc doesn't implement a fallback and explicitly says it in the manual. 
Using a different implementation in musl seems undesirable.

Note that CPython since 3.10 can use close_range() in fork/vfork child 
for subprocess.Popen(close_fds=True) (which is the default), so it 
expects close_range() to be async-signal-safe.

Alexey

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.