Implement POSIX *at() Functions

[bonefish]

POSIX.1-2008 introduces standard functions to operate on FD + relative path pairs:

• faccessat()
• fchmodat()
• fchownat()
• fstatat()
• linkat()
• mkdirat()
• mkfifoat()
• mknodat()
• openat()
• readlinkat()
• renameat()
• symlinkat()
• unlinkat()
• utimensat()

The new open modes O_SEARCH AND O_EXEC should also be implemented.

[bonefish]

Implemented openat() and readlinkat() in hrev33976 and hrev33978 respectively.

[bonefish]

Implemented fstatat() in hrev33987.

[bonefish]

Implemented POSIX.1-2008 functions unlinkat(), symlinkat(), mkdirat(), and
utimensat() in hrev34010.

[axeld]

Implemented the remaining functions in hrev34288 - just noticed that linkat() is still missing, though, will do that next.

[axeld]

Implemented linkat() in hrev34289.

[bonefish]

Nice so far, but what about O_SEARCH and O_EXEC. Reopen the ticket or create a new one?

[axeld]

Is this related to the *at() functions? If not, I would prefer a new ticket for those.

[bonefish]

Replying to axeld:

Is this related to the *at() functions? If not, I would prefer a new ticket for those.

Er, have you read the specification of any of the *at() functions? O_SEARCH definitely is related. O_EXEC not so much, unless you consider fexecve() kind of belonging to the lot -- it could be somewhat tricky to implement, anyway, so we might want to track it separately.

[axeld]

According to the specification, O_SEARCH can also be passed to open(). That's why I asked :-)

[bonefish]

Replying to axeld:

According to the specification, O_SEARCH can also be passed to open(). That's why I asked :-)

Yep, it can be passed to open() and openat(), but it affects all *at() functions. It's mostly about permission checking, but not supporting this mode calls for ported packages that check for the *at() functions existence and implicitly assume O_SEARCH to exist as well. I'm actually almost surprised that none of the ported code in the repository does that. This kind of problem is why I had to introduce the B_ENABLE_INCOMPLETE_POSIX_AT_SUPPORT in the first place.

[axeld]

I guess it's good enough for the time being, but that should definitely be fixed before the next release :-)