With BPF constant blinding enabled, various tests exercising jump offset limits currently fail because size assumptions in the code no longer apply. The created BPF program will be expanded through BPF constant blinding, pushing jump target offsets over the limit. This small series fixes these by either avoiding the use of instructions with immediate values (patch 1) or accounting for the expansion (patch 3). Before: root@box:~# sysctl net.core.bpf_jit_harden=2 root@box:~# insmod test_bpf.ko insmod: ERROR: could not insert module test_bpf.ko: Invalid parameters root@box:~# dmesg | grep Summary [ 177.628099] test_bpf: Summary: 1000 PASSED, 49 FAILED, [988/988 JIT'ed] After: root@box:~# sysctl net.core.bpf_jit_harden=2 root@box:~# insmod test_bpf.ko root@box:~# dmesg | grep Summary [ 220.437597] test_bpf: Summary: 1049 PASSED, 0 FAILED, [1037/1037 JIT'ed] [ 220.477987] test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed] [ 220.480525] test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED Please apply! Thanks, Mathias Mathias Krause (3): bpf/tests: Make max jump tests constant blinding compatible bpf: Allow calling bpf_jit_blinding_enabled() with a NULL program bpf/tests: Make staggered jump tests constant blinding compatible include/linux/filter.h | 5 +++-- kernel/bpf/core.c | 3 +++ kernel/bpf/token.c | 3 +++ lib/test_bpf.c | 21 ++++++++++++++++++--- 4 files changed, 27 insertions(+), 5 deletions(-) -- 2.30.2
