Re: [PATCH bpf v10 14/14] bpf: sockmap, test progs verifier error with latest clang

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, May 22, 2023 at 07:56 PM -07, John Fastabend wrote:
> With a relatively recent clang (7090c10273119) and with this commit
> to fix warnings in selftests (c8ed668593972) that uses __sink(err)
> to resolve unused variables. We get the following verifier error.
>
> root@6e731a24b33a:/host/tools/testing/selftests/bpf# ./test_sockmap
> libbpf: prog 'bpf_sockmap': BPF program load failed: Permission denied
> libbpf: prog 'bpf_sockmap': -- BEGIN PROG LOAD LOG --
> 0: R1=ctx(off=0,imm=0) R10=fp0
> ; op = (int) skops->op;
> 0: (61) r2 = *(u32 *)(r1 +0)          ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; switch (op) {
> 1: (16) if w2 == 0x4 goto pc+5        ; R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> 2: (56) if w2 != 0x5 goto pc+15       ; R2_w=5
> ; lport = skops->local_port;
> 3: (61) r2 = *(u32 *)(r1 +68)         ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; if (lport == 10000) {
> 4: (56) if w2 != 0x2710 goto pc+13 18: R1=ctx(off=0,imm=0) R2=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R10=fp0
> ; __sink(err);
> 18: (bc) w1 = w0
> R0 !read_ok
> processed 18 insns (limit 1000000) max_states_per_insn 0 total_states 2 peak_states 2 mark_read 1
> -- END PROG LOAD LOG --
> libbpf: prog 'bpf_sockmap': failed to load: -13
> libbpf: failed to load object 'test_sockmap_kern.bpf.o'
> load_bpf_file: (-1) No such file or directory
> ERROR: (-1) load bpf failed
> libbpf: prog 'bpf_sockmap': BPF program load failed: Permission denied
> libbpf: prog 'bpf_sockmap': -- BEGIN PROG LOAD LOG --
> 0: R1=ctx(off=0,imm=0) R10=fp0
> ; op = (int) skops->op;
> 0: (61) r2 = *(u32 *)(r1 +0)          ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; switch (op) {
> 1: (16) if w2 == 0x4 goto pc+5        ; R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> 2: (56) if w2 != 0x5 goto pc+15       ; R2_w=5
> ; lport = skops->local_port;
> 3: (61) r2 = *(u32 *)(r1 +68)         ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; if (lport == 10000) {
> 4: (56) if w2 != 0x2710 goto pc+13 18: R1=ctx(off=0,imm=0) R2=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R10=fp0
> ; __sink(err);
> 18: (bc) w1 = w0
> R0 !read_ok
> processed 18 insns (limit 1000000) max_states_per_insn 0 total_states 2 peak_states 2 mark_read 1
> -- END PROG LOAD LOG --
> libbpf: prog 'bpf_sockmap': failed to load: -13
> libbpf: failed to load object 'test_sockhash_kern.bpf.o'
> load_bpf_file: (-1) No such file or directory
> ERROR: (-1) load bpf failed
> libbpf: prog 'bpf_sockmap': BPF program load failed: Permission denied
> libbpf: prog 'bpf_sockmap': -- BEGIN PROG LOAD LOG --
> 0: R1=ctx(off=0,imm=0) R10=fp0
> ; op = (int) skops->op;
> 0: (61) r2 = *(u32 *)(r1 +0)          ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; switch (op) {
> 1: (16) if w2 == 0x4 goto pc+5        ; R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> 2: (56) if w2 != 0x5 goto pc+15       ; R2_w=5
> ; lport = skops->local_port;
> 3: (61) r2 = *(u32 *)(r1 +68)         ; R1=ctx(off=0,imm=0) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
> ; if (lport == 10000) {
> 4: (56) if w2 != 0x2710 goto pc+13 18: R1=ctx(off=0,imm=0) R2=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R10=fp0
> ; __sink(err);
> 18: (bc) w1 = w0
> R0 !read_ok
> processed 18 insns (limit 1000000) max_states_per_insn 0 total_states 2 peak_states 2 mark_read 1
> -- END PROG LOAD LOG --
>
> To fix simply remove the err value because its not actually used anywhere
> in the testing. We can investigate the root cause later. Future patch should
> probably actually test the err value as well. Although if the map updates
> fail they will get caught eventually by userspace.
>
> Fixes: c8ed668593972 ("selftests/bpf: fix lots of silly mistakes pointed out by compiler")
> Signed-off-by: John Fastabend <john.fastabend@xxxxxxxxx>
> ---

I think the problem is that err is not initialized when we hit the default
switch branch. And __sink() generates a read.

Since we don't really consume the error value, this fix LGTM.

Reviewed-by: Jakub Sitnicki <jakub@xxxxxxxxxxxxxx>




[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux