bpftool/libbpf: Add probe for large INSN limit
Subsystems (5)
v2 cover bpftool/libbpf: Add probe for large INSN limit
This series implements a new BPF feature probe which checks for the commit c04c0d2b968a ("bpf: increase complexity limit and maximum program size"), which increases the maximum program size to 1M. It's based on the similar check in Cilium, although Cilium is already aiming to use bpftool checks and eventually drop all its custom checks.
Examples of outputs:
# bpftool feature probe [...] Scanning miscellaneous eBPF features... Large complexity limit and maximum program size (1M) is available
# bpftool feature probe macros [...] /*** eBPF misc features ***/ #define HAVE_HAVE_LARGE_INSN_LIMIT
# bpftool feature probe -j | jq '.["misc"]'
{
"have_large_insn_limit": true
}v1 -> v2:
- Test for 'BPF_MAXINSNS + 1' number of total insns.
- Remove info about the current 1M limit from probe's description.
Michal Rostecki (2):
libbpf: Add probe for large INSN limit
bpftool: Add misc secion and probe for large INSN limit
tools/bpf/bpftool/feature.c | 18 ++++++++++++++++++ tools/lib/bpf/libbpf.h | 1 + tools/lib/bpf/libbpf.map | 1 + tools/lib/bpf/libbpf_probes.c | 21 +++++++++++++++++++++ 4 files changed, 41 insertions(+)
Signature · 1 line
2.16.4
PATCH v2 1/2 libbpf: Add probe for large INSN limit
Introduce a new probe which checks whether kernel has large maximum program size (1M) which was increased in the following commit:
c04c0d2b968a ("bpf: increase complexity limit and maximum program size")
Based on the similar check in Cilium[0], authored by Daniel Borkmann.
[0] https://github.com/cilium/cilium/commit/657d0f585afd26232cfa5d4e70b6f64d2ea91596
--- tools/lib/bpf/libbpf.h | 1 + tools/lib/bpf/libbpf.map | 1 + tools/lib/bpf/libbpf_probes.c | 21 +++++++++++++++++++++ 3 files changed, 23 insertions(+)
tools/lib/bpf/libbpf.h +1 -0
@@ -521,6 +521,7 @@ LIBBPF_API bool bpf_probe_prog_type(enum bpf_prog_type prog_type, LIBBPF_API bool bpf_probe_map_type(enum bpf_map_type map_type, __u32 ifindex); LIBBPF_API bool bpf_probe_helper(enum bpf_func_id id,enum bpf_prog_type prog_type,__u32 ifindex); +LIBBPF_API bool bpf_probe_large_insn_limit(__u32 ifindex);/* * Get bpf_prog_info in continuous memory
tools/lib/bpf/libbpf.map +1 -0
@@ -219,6 +219,7 @@ LIBBPF_0.0.7 { bpf_object__detach_skeleton; bpf_object__load_skeleton; bpf_object__open_skeleton; + bpf_probe_large_insn_limit; bpf_prog_attach_xattr; bpf_program__attach; bpf_program__name;
tools/lib/bpf/libbpf_probes.c +21 -0
@@ -321,3 +321,24 @@ bool bpf_probe_helper(enum bpf_func_id id, enum bpf_prog_type prog_type,return res; } + +/* +* Probe for availability of kernel commit (5.3): +* +* c04c0d2b968a ("bpf: increase complexity limit and maximum program size") +*/ +bool bpf_probe_large_insn_limit (__u32 ifindex) +{ +struct bpf_insn insns[BPF_MAXINSNS +1 ]; +int i; + +for (i =0 ; i <BPF_MAXINSNS ; i++) + insns[i] =BPF_MOV64_IMM (BPF_REG_0 ,1 ); + insns[BPF_MAXINSNS ] =BPF_EXIT_INSN (); + + errno =0 ; +probe_load (BPF_PROG_TYPE_SCHED_CLS , insns,ARRAY_SIZE (insns),NULL ,0 , + ifindex); + +return errno !=E2BIG && errno !=EINVAL ; +}
Signature · 1 line
2.16.4
PATCH v2 2/2 bpftool: Add misc secion and probe for large INSN limit
In reply to Michal RosteckiIntroduce a new probe section (misc) for probes not related to concrete map types, program types, functions or kernel configuration. Introduce a probe for large INSN limit as the first one in that section.
--- tools/bpf/bpftool/feature.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
tools/bpf/bpftool/feature.c +18 -0
@@ -572,6 +572,18 @@ probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,printf ("\n" ); } +static void +probe_large_insn_limit (const char *define_prefix,__u32 ifindex) +{ +bool res; + + res =bpf_probe_large_insn_limit (ifindex); +print_bool_feature ("have_large_insn_limit" , +"Large complexity and program size limit" , +"HAVE_LARGE_INSN_LIMIT" , + res, define_prefix); +} +static int do_probe (int argc,char **argv) {enum probe_component target =COMPONENT_UNSPEC ;
@@ -724,6 +736,12 @@ static int do_probe(int argc, char **argv)probe_helpers_for_progtype (i, supported_types[i], define_prefix, ifindex); +print_end_then_start_section ("misc" , +"Scanning miscellaneous eBPF features..." , +"/*** eBPF misc features ***/" , + define_prefix); +probe_large_insn_limit (define_prefix, ifindex); + exit_close_json:if (json_output) {/* End current "section" of probes */
Signature · 1 line
2.16.4
Re: [PATCH bpf-next v2 2/2] bpftool: Add misc secion and probe for large INSN limit
Nit: typo in subject ("secion").
2020-01-07 14:03 UTC+0100 ~ Michal Rostecki <[email protected]>
Quoted 15 lines
Introduce a new probe section (misc) for probes not related to concrete map types, program types, functions or kernel configuration. Introduce a probe for large INSN limit as the first one in that section.
Signed-off-by Michal Rostecki <[email protected]>--- tools/bpf/bpftool/feature.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)tools/bpf/bpftool/feature.c +0
-0@@ -572,6 +572,18 @@ probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,printf ("\n" );
tools/bpf/bpftool/feature.c probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type, · line 582
Quoted 10 lines
tools/bpf/bpftool/feature.c +8
-0} +static void +probe_large_insn_limit (const char *define_prefix,__u32 ifindex) +{ +bool res; + + res =bpf_probe_large_insn_limit (ifindex); + print_bool_feature("have_large_insn_limit", + "Large complexity and program size limit",
I am not sure we should mention "complexity" here. Although it is related to program size in the kernel commit you describe, the probe that is run is only on instruction number. This can make a difference for offloaded programs: When you probe a device, if kernel has commit c04c0d2b968a and supports up to 1M instructions, but hardware supports no more than 4k instructions, you may still benefit from the new value for BPF_COMPLEXITY_LIMIT_INSNS for complexity, but not for the total number of available instructions. In that case the probe will fail, and the message on complexity would not be accurate.
Looks good otherwise, thanks Michal!
Quentin
Re: [PATCH bpf-next v2 2/2] bpftool: Add misc secion and probe for large INSN limit
Quoted 39 lines · Quentin Monnet
Nit: typo in subject ("secion").
2020-01-07 14:03 UTC+0100 ~ Michal Rostecki <[email protected]>
Quoted 15 lines
Introduce a new probe section (misc) for probes not related to concrete map types, program types, functions or kernel configuration. Introduce a probe for large INSN limit as the first one in that section.
Signed-off-by Michal Rostecki <[email protected]>--- tools/bpf/bpftool/feature.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)tools/bpf/bpftool/feature.c +0
-0@@ -572,6 +572,18 @@ probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,printf ("\n" );tools/bpf/bpftool/feature.c probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type, · line 582
Quoted 10 lines
tools/bpf/bpftool/feature.c +8
-0} +static void +probe_large_insn_limit (const char *define_prefix,__u32 ifindex) +{ +bool res; + + res =bpf_probe_large_insn_limit (ifindex); + print_bool_feature("have_large_insn_limit", + "Large complexity and program size limit",I am not sure we should mention "complexity" here. Although it is related to program size in the kernel commit you describe, the probe that is run is only on instruction number. This can make a difference for offloaded programs: When you probe a device, if kernel has commit
c04c0d2b968aand supports up to 1M instructions, but hardware supports no more than 4k instructions, you may still benefit from the new value forBPF_COMPLEXITY_LIMIT_INSNSfor complexity, but not for the total number of available instructions. In that case the probe will fail, and the message on complexity would not be accurate.
Quoted 3 lines
Looks good otherwise, thanks Michal!
Quentin
Thanks for review! Should I change the description just to "Large program size limit" or "Large instruction limit"?
Michal
Re: [PATCH bpf-next v2 2/2] bpftool: Add misc secion and probe for large INSN limit
2020-01-08 13:41 UTC+0000 ~ Michal Rostecki <[email protected]>
Quoted 40 lines · Quentin Monnet
Quoted 39 lines
Nit: typo in subject ("secion").
2020-01-07 14:03 UTC+0100 ~ Michal Rostecki <[email protected]>
Quoted 15 lines
Introduce a new probe section (misc) for probes not related to concrete map types, program types, functions or kernel configuration. Introduce a probe for large INSN limit as the first one in that section.
Signed-off-by Michal Rostecki <[email protected]>--- tools/bpf/bpftool/feature.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)tools/bpf/bpftool/feature.c +0
-0@@ -572,6 +572,18 @@ probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type,printf ("\n" );tools/bpf/bpftool/feature.c probe_helpers_for_progtype(enum bpf_prog_type prog_type, bool supported_type, · line 582
Quoted 10 lines
tools/bpf/bpftool/feature.c +8
-0} +static void +probe_large_insn_limit (const char *define_prefix,__u32 ifindex) +{ +bool res; + + res =bpf_probe_large_insn_limit (ifindex); + print_bool_feature("have_large_insn_limit", + "Large complexity and program size limit",I am not sure we should mention "complexity" here. Although it is related to program size in the kernel commit you describe, the probe that is run is only on instruction number. This can make a difference for offloaded programs: When you probe a device, if kernel has commit
c04c0d2b968aand supports up to 1M instructions, but hardware supports no more than 4k instructions, you may still benefit from the new value forBPF_COMPLEXITY_LIMIT_INSNSfor complexity, but not for the total number of available instructions. In that case the probe will fail, and the message on complexity would not be accurate.
Quoted 9 lines
Quoted 3 lines
Looks good otherwise, thanks Michal!
Quentin
Thanks for review! Should I change the description just to "Large program size limit" or "Large instruction limit"?
Michal
I don't really have a preference here, let's keep "program size"? Quentin