open.2: describe O_RESOLVE_BENEATH errors correctly

[val_packett.cool]

Summary

The behavior is the same as in capability mode, it does not actually
return EINVAL for absolute lookups:

openat(AT_FDCWD,"/tmp/test",O_RDONLY|O_DIRECTORY,00) = 3 (0x3)
openat(3,"../../",O_RDONLY|0x800000,00)          ERR#93 'Capabilities insufficient'
openat(3,"/etc/passwd",O_RDONLY|0x800000,00)     ERR#93 'Capabilities insufficient'

Fixes: 1f305be43 ("Document {O,AT}_RESOLVE_BENEATH...")
Sponsored by: https://www.patreon.com/valpackett

[pauamma_gundo.com]

lib/libc/sys/open.2

602–628

I'm not sure this is as clear as the previous version. "A, or B, and C or D" leaves me unsure whether the "C or D" restriction applies to B only, or to both A and B. I don't think the previous version had that ambiguity.

[val_packett.cool]

Hopefully more clear this way?

[pauamma_gundo.com]

In D38675#880230, @val_packett.cool wrote:

Hopefully more clear this way?

It helps. So (I think) does removing some of the commas. What do others think?

lib/libc/sys/open.2

602–628

[kib]

The common practice in the error section of the man pages is to list each cause separately, repeating the error number (symbol). So your change would be better done in opposite direction, IMO: splitting the 'absolute path', 'dotdot', and 'dotdot' leading outside the root.

Note that allowance for dotdot presence is controlled by a sysctl knob.

[val_packett.cool]

Yeah, makes sense, let's try like this

[val_packett.cool]

Though, hm, various AT_RESOLVE_BENEATH documentation does collapse it all into one item:

https://cgit.freebsd.org/src/tree/lib/libc/sys/chmod.2?id=ef6fcc5e2b0714c859d2e4ba23a55b1fd12f8a4e#n312

[pauamma_gundo.com]

Minor nits, fixable on commit.

lib/libc/sys/open.2

31

Remember to bump.

611

For tense consistency with "is"

625–628

633