| Message ID | 20250313172127.1098195-1-luis.gerhorst@fau.de |
|---|---|
| Headers | show |
| Series | bpf: Mitigate Spectre v1 using barriers | expand |
On Thu, 2025-03-13 at 18:21 +0100, Luis Gerhorst wrote: > This improves the expressiveness of unprivileged BPF by inserting > speculation barriers instead of rejecting the programs. > > The approach was previously presented at LPC'24 [1] and RAID'24 [2]. > > To mitigate the Spectre v1 (PHT) vulnerability, the kernel rejects > potentially-dangerous unprivileged BPF programs as of > commit 9183671af6db ("bpf: Fix leakage under speculation on mispredicted > branches"). In [2], we have analyzed 364 object files from open source > projects (Linux Samples and Selftests, BCC, Loxilb, Cilium, libbpf > Examples, Parca, and Prevail) and found that this affects 31% to 54% of > programs. > > To resolve this in the majority of cases this patchset adds a fall-back > for mitigating Spectre v1 using speculation barriers. The kernel still > optimistically attempts to verify all speculative paths but uses > speculation barriers against v1 when unsafe behavior is detected. This > allows for more programs to be accepted without disabling the BPF > Spectre mitigations (e.g., by setting cpu_mitigations_off()). > > In [1] we have measured the overhead of this approach relative to having > mitigations off and including the upstream Spectre v4 mitigations. For > event tracing and stack-sampling profilers, we found that mitigations > increase BPF program execution time by 0% to 62%. For the Loxilb network > load balancer, we have measured a 14% slowdown in SCTP performance but > no significant slowdown for TCP. This overhead only applies to programs > that were previously rejected. > > I reran the expressiveness-evaluation with v6.14 and made sure the main > results still match those from [1] and [2] (which used v6.5). > > Main design decisions are: > > * Do not use separate bytecode insns for v1 and v4 barriers. This > simplifies the verifier significantly and has the only downside that > performance on PowerPC is not as high as it could be. > > * Allow archs to still disable v1/v4 mitigations separately by setting > bpf_jit_bypass_spec_v1/v4(). This has the benefit that archs can > benefit from improved BPF expressiveness / performance if they are not > vulnerable (e.g., ARM64 for v4 in the kernel). > > * Do not remove the empty BPF_NOSPEC implementation for backends for > which it is unknown whether they are vulnerable to Spectre v1. [...] I think it would be good to have some tests checking that nospec instructions are inserted in expected locations. Could you please take look at use of __xlated tag in e.g. tools/testing/selftests/bpf/progs/verifier_sdiv.c ? [...]
Eduard Zingerman <eddyz87@gmail.com> writes: > I think it would be good to have some tests checking that nospec > instructions are inserted in expected locations. > Could you please take look at use of __xlated tag in e.g. > tools/testing/selftests/bpf/progs/verifier_sdiv.c ? That looks very promising, I will look into it for v2. Thanks for the pointer. I guess it might be worth it to add __xlated to at least on test per nospec-related code path. If there are other rewrites at play that will make it harder to adapt the tests when the other rewrite is ever changed, but it might also help in catching interactions between the other rewrites and the nospec. Also, thanks for the review of patches 2 and 3.
This improves the expressiveness of unprivileged BPF by inserting speculation barriers instead of rejecting the programs. The approach was previously presented at LPC'24 [1] and RAID'24 [2]. To mitigate the Spectre v1 (PHT) vulnerability, the kernel rejects potentially-dangerous unprivileged BPF programs as of commit 9183671af6db ("bpf: Fix leakage under speculation on mispredicted branches"). In [2], we have analyzed 364 object files from open source projects (Linux Samples and Selftests, BCC, Loxilb, Cilium, libbpf Examples, Parca, and Prevail) and found that this affects 31% to 54% of programs. To resolve this in the majority of cases this patchset adds a fall-back for mitigating Spectre v1 using speculation barriers. The kernel still optimistically attempts to verify all speculative paths but uses speculation barriers against v1 when unsafe behavior is detected. This allows for more programs to be accepted without disabling the BPF Spectre mitigations (e.g., by setting cpu_mitigations_off()). In [1] we have measured the overhead of this approach relative to having mitigations off and including the upstream Spectre v4 mitigations. For event tracing and stack-sampling profilers, we found that mitigations increase BPF program execution time by 0% to 62%. For the Loxilb network load balancer, we have measured a 14% slowdown in SCTP performance but no significant slowdown for TCP. This overhead only applies to programs that were previously rejected. I reran the expressiveness-evaluation with v6.14 and made sure the main results still match those from [1] and [2] (which used v6.5). Main design decisions are: * Do not use separate bytecode insns for v1 and v4 barriers. This simplifies the verifier significantly and has the only downside that performance on PowerPC is not as high as it could be. * Allow archs to still disable v1/v4 mitigations separately by setting bpf_jit_bypass_spec_v1/v4(). This has the benefit that archs can benefit from improved BPF expressiveness / performance if they are not vulnerable (e.g., ARM64 for v4 in the kernel). * Do not remove the empty BPF_NOSPEC implementation for backends for which it is unknown whether they are vulnerable to Spectre v1. [1] https://lpc.events/event/18/contributions/1954/ ("Mitigating Spectre-PHT using Speculation Barriers in Linux eBPF") [2] https://arxiv.org/pdf/2405.00078 ("VeriFence: Lightweight and Precise Spectre Defenses for Untrusted Linux Kernel Extensions") Changes: * RFC -> v1: - rebase to bpf-next-250313 - tests: mark expected successes/new errors - add bpt_jit_bypass_spec_v1/v4() to avoid #ifdef in bpf_bypass_spec_v1/v4() - ensure that nospec with v1-support is implemented for archs for which GCC supports speculation barriers, except for MIPS - arm64: emit speculation barrier - powerpc: change nospec to include v1 barrier - discuss potential security (archs that do not impl. BPF nospec) and performance (only PowerPC) regressions RFC: https://lore.kernel.org/bpf/20250224203619.594724-1-luis.gerhorst@fau.de/ Luis Gerhorst (11): bpf: Move insn if/else into do_check_insn() bpf: Return -EFAULT on misconfigurations bpf: Return -EFAULT on internal errors bpf, arm64, powerpc: Add bpf_jit_bypass_spec_v1/v4() bpf, arm64, powerpc: Change nospec to include v1 barrier bpf: Rename sanitize_stack_spill to nospec_result bpf: Fall back to nospec for Spectre v1 bpf: Allow nospec-protected var-offset stack access bpf: Return PTR_ERR from push_stack() bpf: Fall back to nospec for sanitization-failures bpf: Fall back to nospec for spec path verification arch/arm64/net/bpf_jit.h | 5 + arch/arm64/net/bpf_jit_comp.c | 28 +- arch/powerpc/net/bpf_jit_comp64.c | 79 +- include/linux/bpf.h | 11 +- include/linux/bpf_verifier.h | 3 +- include/linux/filter.h | 2 +- kernel/bpf/core.c | 32 +- kernel/bpf/verifier.c | 723 ++++++++++-------- .../selftests/bpf/progs/verifier_and.c | 3 +- .../selftests/bpf/progs/verifier_bounds.c | 35 +- .../bpf/progs/verifier_bounds_deduction.c | 43 +- .../selftests/bpf/progs/verifier_map_ptr.c | 12 +- .../selftests/bpf/progs/verifier_movsx.c | 6 +- .../selftests/bpf/progs/verifier_unpriv.c | 3 +- .../bpf/progs/verifier_value_ptr_arith.c | 50 +- .../selftests/bpf/verifier/dead_code.c | 3 +- tools/testing/selftests/bpf/verifier/jmp32.c | 33 +- tools/testing/selftests/bpf/verifier/jset.c | 10 +- 18 files changed, 630 insertions(+), 451 deletions(-) base-commit: 46d38f489ef02175dcff1e03a849c226eb0729a6