Hi Bruno,
On 5/18/24 5:03 PM, Bruno Haible wrote:
> And it's a good idea to reference the bug report in a comment:
> @c https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=58266
> So that in the future it helps us to understand what the bug is
> about and how to reproduce it in newer versions of NetBSD.
The bug report got addressed today. In the following commit [1]:
PR/58266: Collin Funk: Fail if from == to, like FreeBSD and Linux. The test
is done in dup3 before any other tests so even if a bad descriptor it is
passed we will return EINVAL not EBADFD like Linux does.
and then a second to deal with syscall versioning [2].
Looking into their comment about EINVAL vs EBADF, they follow what
FreeBSD and OpenBSD do [3][4]. I think we may have the ordering wrong
in lib/dup3.c:
if (newfd < 0 || newfd >= getdtablesize () || fcntl (oldfd, F_GETFD) == -1)
{
errno = EBADF;
return -1;
}
if (newfd == oldfd)
{
errno = EINVAL;
return -1;
}
In the Linux kernel sources they are done in this order [5]:
if ((flags & ~O_CLOEXEC) != 0)
return -EINVAL;
if (unlikely(oldfd == newfd))
return -EINVAL;
if (newfd >= rlimit(RLIMIT_NOFILE))
return -EBADF;
Unless glibc wraps the syscall in it's own functions where the order
is changed. I'm not too sure how that stuff works but I couldn't seem
to find it.
Collin
[1]
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/kern/sys_descrip.c.diff?r1=1.48&r2=1.49&only_with_tag=MAIN
[2]
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/kern/sys_descrip.c.diff?r1=1.49&r2=1.50&only_with_tag=MAIN
[3] https://github.com/freebsd/freebsd-src/blob/main/lib/libc/gen/dup3.c
[4]
https://github.com/openbsd/src/blob/8b68e9dedc5b1a2ef9fae4d11f703c6d7be32352/sys/kern/kern_descrip.c#L328
[5]
https://github.com/torvalds/linux/blob/eb6a9339efeb6f3d2b5c86fdf2382cdc293eca2c/fs/file.c#L1354
