md5: Accept "-" as alias for stdin.

[des]

Summary

PR: 276915
Reported by: Hannes Hauswedell <[e-mail]>

[des]

Note, this is a superset of @jhb's D43807

[allanjude]

I could have sworn it did this already.

Reviewed-by: allanjude

[des]

In D43870#1001037, @allanjude wrote:

I could have sworn it did this already.

You're probably thinking of D43532

[markj]

sbin/md5/md5.c

633

It's not obvious to me that we want to limit rights on stdin as well (we might want CAP_IOCTL as well, for instance). Do the tests all pass when the kern.trap_enotcap sysctl is set to 1?

[emaste]

sbin/md5/md5.c

633

Presumably it should also work if e.g. the user passes /dev/stdin as a file; I think it should be OK to apply the same &rights to files and stdin (even if more rights may be needed in the list)

[des]

sbin/md5/md5.c

633

Normally for stdin we'd call caph_limit_stdin() which is _more_ restrictive than what we're asking for here, so it's fine.

[markj]

sbin/md5/md5.c

633

I think that's backwards? caph_limit_stdin() is less restrictive, e.g., because it includes CAP_IOCTL.

In any case, I think this caph_rights_limit() call was useless before - we pass the rights to fileargs_open(), so they're applied automatically. We don't need to call it again here.

I would suggest removing this call outright and replacing the existing caph_limit_stdout/stderr() calls with a caph_limit_stdio() call. We're still left with inconsistent behaviour for the /dev/stdin case as emaste mentioned, but that's probably harmless.

[jhb]

I agree with Mark's comments on cap_rights_limit, though I think that's also an orthogonal fix.

[des]

sbin/md5/md5.c

633

See D43897