linux(4): Handle specific to Linux fstatat() error case

[dchagin]

Summary

Linux fstatat() returns ENOENT if pathname is an empty string and
AT_EMPTY_PATH was not specified in flags.

trivial fix, to get an opinon, is this needed for FreeBSD fstatat?

[kib]

Are you doing second copyinstr() over the same user path? This introduces the race.

What is the current FreeBSD reaction to the condition?

[dchagin]

In D41015#933552, @kib wrote:

Are you doing second copyinstr() over the same user path? This introduces the race.

yes, first in namei

What is the current FreeBSD reaction to the condition?

ENOTDIR as documented

[kib]

In D41015#933553, @dchagin wrote:

In D41015#933552, @kib wrote:

Are you doing second copyinstr() over the same user path? This introduces the race.

yes, first in namei

What is the current FreeBSD reaction to the condition?

ENOTDIR as documented

Look at NIRES_EMPTYPATH. If the flag is set and ENOTDIR is returned, translate it to ENOENT.

I assume we want to keep the FreeBSD error?

[mjg]

i would say unless there is a good reason to deviate, the error returned should be compatible with linux

[dchagin]

In D41015#933638, @mjg wrote:

i would say unless there is a good reason to deviate, the error returned should be compatible with linux

I don’t have a strong opinion, if we get a consensus, I’ll prepare the fix. Konstantin?

[kib]

Change the FreeBSD error.