NetBSD-Bugs archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: kern/60975: NPF rules are almost never BPF JIT compiled since 2014
joe%eloisystems.com@localhost via gnats wrote:
> >Number: 60975
> >Category: kern
> >Synopsis: NPF rules are almost never BPF JIT compiled since 2014
> >Confidential: no
> >Severity: serious
> >Priority: high
> >Responsible: kern-bug-people
> >State: open
> >Class: sw-bug
> >Submitter-Id: net
> >Arrival-Date: Fri Nov 06 22:45:00 +0000 2026
> >Originator: Emmanuel Nyarko
> >Release: NetBSD 11
> >Organization:
> NetBSD
> >Environment:
> NetBSD 11.0_STABLE NetBSD 11.0_STABLE
> >Description:
>
> wrong size is passed to npf_bpf_compile. the compile function expects a instruction size count
> but the byte count of the nvlist blob is what is sent to BPF JIT.
>
> This almost always leads to RET instructions not falling on the last BPF code
> or random crashes rules count are more than the blob size.
The title is misleading. Even though a wrong size is passed, it's
harmless in stable releases because bpfjit doesn't read unreachable
code.
Below is a copy of my analysis sent to source-changes-d:
Alexander Nasonov wrote:
> In stable releases it reads past the end of allocations but it generates
> valid jitted code.
Further analysis below. Stable releases should be fine,
because I don't see reads of unreachable insns anywhere.
optimize_pass1 doesn't read unreachable insns:
for (i = 0; i < insn_count; i++) {
if (!SLIST_EMPTY(&insn_dat[i].bjumps))
unreachable = false;
insn_dat[i].unreachable = unreachable;
if (unreachable)
continue;
optimize_pass2 doesn't read unreachable insns, it only
computes addresses:
for (i = insn_count; i != 0; i--) {
pc = &insns[i-1];
pd = &insn_dat[i-1];
if (pd->unreachable)
continue;
optimize_pass3 doesn't read unreachable insns either:
for (i = 0; i < insn_count; i++) {
if (insn_dat[i].unreachable)
continue;
--
Alex
Home |
Main Index |
Thread Index |
Old Index