Re: [PATCH 0/5] NFSD: Apply export security when a compound crosses exports

[VeroFess <coderzeng@gmail.com>]

Chuck,

I tested the five patches on an x86-64 KASAN guest. They applied in
order, the changed NFSD objects built with W=1, and the full bzImage
built with Btrfs, TLS, and NFSv4.2 inter-server COPY enabled.
The wire tests covered NFSv4.0, 4.1, and 4.2 where applicable.

After a reserved-port OPEN, high-port AUTH_SYS compounds that used
LOOKUP before READ, COPY, or CLONE stopped at LOOKUP with WRONGSEC.
The destination was unchanged. Direct PUTFH returned WRONGSEC, while
low-port COPY and CLONE succeeded. SECINFO and SECINFO_NO_NAME still
succeeded on the high port. READDIR with rdattr_error omitted the
protected filehandle.

After switching the source export to root_squash, LOOKUP; READ,
RESTOREFH; READ, and saved-source COPY/CLONE returned ACCESS with the
old OPEN stateid. The all_squash UID/GID cases and same-export
COPY/CLONE passed with the expected data.

With krb5 or krb5i on a krb5p export, OPEN by filehandle returned
WRONGSEC at PUTFH; OPEN by name returned it at OPEN. krb5p worked from
a high port and over TLS. NFSv3 LOOKUP returned PERM from a high port
and succeeded from a reserved port. I saw no KASAN report or oops in
the guest log.

Tested-by: VeroFess <coderzeng@gmail.com>

VeroFess