I did an experiment with one of our internal bpf programs.
The program has 1563 insns.
Without Edward's patch:
processed 13634 insns, stack depth 160
With Edward's patch:
processed 15807 insns, stack depth 160
So the number of processed insns regressed by roughly 16%.
Did anybody do any similar experiments to quantify the patch's
impact in verification performance (e.g., in terms of processed insns)?
On Tue, Jun 27, 2017 at 5:53 AM, Edward Cree via iovisor-dev
This series simplifies alignment tracking, generalises bounds tracking and
fixes some bounds-tracking bugs in the BPF verifier. Pointer arithmetic on
packet pointers, stack pointers, map value pointers and context pointers has
been unified, and bounds on these pointers are only checked when the pointer
Operations on pointers which destroy all relation to the original pointer
(such as multiplies and shifts) are disallowed if !env->allow_ptr_leaks,
otherwise they convert the pointer to an unknown scalar and feed it to the
normal scalar arithmetic handling.
Pointer types have been unified with the corresponding adjusted-pointer types
where those existed (e.g. PTR_TO_MAP_VALUE[_ADJ] or FRAME_PTR vs
PTR_TO_STACK); similarly, CONST_IMM and UNKNOWN_VALUE have been unified into
Pointer types (except CONST_PTR_TO_MAP, PTR_TO_MAP_VALUE_OR_NULL and
PTR_TO_PACKET_END, which do not allow arithmetic) have a 'fixed offset' and
a 'variable offset'; the former is used when e.g. adding an immediate or a
known-constant register, as long as it does not overflow. Otherwise the
latter is used, and any operation creating a new variable offset creates a
new 'id' (and, for PTR_TO_PACKET, clears the 'range').
SCALAR_VALUEs use the 'variable offset' fields to track the range of possible
values; the 'fixed offset' should never be set on a scalar.
As of patch 12/12, all tests of tools/testing/selftests/bpf/test_verifier
and tools/testing/selftests/bpf/test_align pass.
v3: added a few more tests; removed RFC tags.
v2: fixed nfp build, made test_align pass again and extended it with a few
new tests (though still need to add more).
Edward Cree (12):
selftests/bpf: add test for mixed signed and unsigned bounds checks
bpf/verifier: rework value tracking
nfp: change bpf verifier hooks to match new verifier data structures
bpf/verifier: track signed and unsigned min/max values
bpf/verifier: more concise register state logs for constant var_off
selftests/bpf: change test_verifier expectations
selftests/bpf: rewrite test_align
selftests/bpf: add a test to test_align
selftests/bpf: add test for bogus operations on pointers
selftests/bpf: don't try to access past MAX_PACKET_OFF in
selftests/bpf: add tests for subtraction & negative numbers
selftests/bpf: variable offset negative tests
drivers/net/ethernet/netronome/nfp/bpf/verifier.c | 24 +-
include/linux/bpf.h | 34 +-
include/linux/bpf_verifier.h | 56 +-
include/linux/tnum.h | 81 +
kernel/bpf/Makefile | 2 +-
kernel/bpf/tnum.c | 180 ++
kernel/bpf/verifier.c | 1943 ++++++++++++---------
tools/testing/selftests/bpf/test_align.c | 462 ++++-
tools/testing/selftests/bpf/test_verifier.c | 293 ++--
9 files changed, 2034 insertions(+), 1041 deletions(-)
create mode 100644 include/linux/tnum.h
create mode 100644 kernel/bpf/tnum.c
iovisor-dev mailing list