bpf: add per-insn complexity limit
authorAlexei Starovoitov <ast@kernel.org>
Tue, 4 Dec 2018 06:46:06 +0000 (22:46 -0800)
committerDaniel Borkmann <daniel@iogearbox.net>
Tue, 4 Dec 2018 16:22:02 +0000 (17:22 +0100)
commitceefbc96fa5c5b975d87bf8e89ba8416f6b764d9
tree46ba4c3e98ffb4eb565dfce9b0bb2fda5ef35ead
parent4f7b3e82589e0de723780198ec7983e427144c0a
bpf: add per-insn complexity limit

malicious bpf program may try to force the verifier to remember
a lot of distinct verifier states.
Put a limit to number of per-insn 'struct bpf_verifier_state'.
Note that hitting the limit doesn't reject the program.
It potentially makes the verifier do more steps to analyze the program.
It means that malicious programs will hit BPF_COMPLEXITY_LIMIT_INSNS sooner
instead of spending cpu time walking long link list.

The limit of BPF_COMPLEXITY_LIMIT_STATES==64 affects cilium progs
with slight increase in number of "steps" it takes to successfully verify
the programs:
                       before    after
bpf_lb-DLB_L3.o         1940      1940
bpf_lb-DLB_L4.o         3089      3089
bpf_lb-DUNKNOWN.o       1065      1065
bpf_lxc-DDROP_ALL.o     28052  |  28162
bpf_lxc-DUNKNOWN.o      35487  |  35541
bpf_netdev.o            10864     10864
bpf_overlay.o           6643      6643
bpf_lcx_jit.o           38437     38437

But it also makes malicious program to be rejected in 0.4 seconds vs 6.5
Hence apply this limit to unprivileged programs only.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
kernel/bpf/verifier.c