[PATCH 13/23] linux: Add fchroot
Adhemerval Zanella Netto
adhemerval.zanella@linaro.org
Thu Nov 5 18:46:48 GMT 2026
On 05/11/26 14:26, Collin Funk wrote:
> Florian Weimer <fweimer@redhat.com> writes:
>
>> * Andreas Schwab:
>>
>>> On Nov 05 2026, Florian Weimer wrote:
>>>
>>>> * Adhemerval Zanella:
>>>>
>>>>> diff --git a/sysdeps/unix/sysv/linux/bits/unistd_ext.h b/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>>>> index a541b378d03..6597435ace7 100644
>>>>> --- a/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>>>> +++ b/sysdeps/unix/sysv/linux/bits/unistd_ext.h
>>>>> @@ -47,4 +47,10 @@ extern __pid_t gettid (void) __THROW;
>>>>> # define CLOSE_RANGE_CLOEXEC (1U << 2)
>>>>> #endif
>>>>>
>>>>> +/* Change the root directory of the calling process to the directory
>>>>> + referred to by the file descriptor FD. FD can also be FD_FAILFS_ROOT,
>>>>> + in which case all subsequent absolute file name lookups fail. FLAGS
>>>>> + must be zero. */
>>>>> +extern int fchroot (int __fd, unsigned int __flags) __THROW;
>>>>> +
>>>>> #endif /* __USE_GNU */
>>>>
>>>> CRIU defines _GNU_SOURCE, and it has:
>>>>
>>>> criu/files.c:#include <unistd.h>
>>>>
>>>> criu/files.c:static int fchroot(int fd)
>>>
>>> BSD also has fchroot without the flags.
>>
>> So does Hurd.
>
> And Solaris [1].
>
> I have been wanting this for the 'chroot' command from GNU coreutils for
> a while. If my memory serves me correctly, it is only available on
> NetBSD and not the other BSDs. For that reason, I didn't bother making
> use of it. I was not aware it was supported on Hurd, and only found out
> about Solaris recently.
It was added on FreeBSD 15 as well [1].
[1] https://man.freebsd.org/cgi/man.cgi?query=fchroot&apropos=0&sektion=0&manpath=FreeBSD+15.1-RELEASE+and+Ports.quarterly&format=html
>
>>> Should the one above be called fchroot2, or should the flags be
>>> dropped for now?
>>
>> It won't fix the fchroot build failure. It's surprisingly rare, though:
>>
>> <https://codesearch.debian.net/search?q=%28%5E%7Cint%5Cs*%29fchroot%5Cs*%5C%28&literal=0>
>>
>> I just got lucky, I didn't find it through Debian Code Search, just by
>> consulting the CRIU source code.
>>
>> I don't think we should aim for compatibility with Hurd/BSD if Linux has
>> something else. On the other hand, if POSIX standardizes fchroot in the
>> one-argument form, we'll have a problem.
>
> Yes. I was pleasantly suprised to see the system call added in Linux
> 7.3. But I find it a bit annoying that a different function signature
> than everyone else...
>
I think we can keep using the BSD/Solaris/Hurd prototype and add a fchroot2
with flags. Seems the lest disruptive way imho.
> Collin
>
> [1] https://docs.oracle.com/cd/E86824_01/html/E54765/fchroot-2.html
More information about the Libc-alpha
mailing list