linux.git/tools/testing/selftests/bpf, branch v5.9-rc2

bpf: selftests: global_funcs: Check err_str before strstr

2020-08-20T21:31:14+00:00

The error path in libbpf.c:load_program() has calls to pr_warn()
which ends up for global_funcs tests to
test_global_funcs.c:libbpf_debug_print().

For the tests with no struct test_def::err_str initialized with a
string, it causes call of strstr() with NULL as the second argument
and it segfaults.

Fix it by calling strstr() only for non-NULL err_str.

Signed-off-by: Yauheni Kaliuta 
Signed-off-by: Alexei Starovoitov 
Acked-by: Yonghong Song 
Link: https://lore.kernel.org/bpf/20200820115843.39454-1-yauheni.kaliuta@redhat.com

selftests/bpf: Remove test_align leftovers

2020-08-20T21:25:25+00:00

Calling generic selftests "make install" fails as rsync expects all
files from TEST_GEN_PROGS to be present. The binary is not generated
anymore (commit 3b09d27cc93d) so we can safely remove it from there
and also from gitignore.

Fixes: 3b09d27cc93d ("selftests/bpf: Move test_align under test_progs")
Signed-off-by: Veronika Kabatova 
Signed-off-by: Alexei Starovoitov 
Acked-by: Jesper Dangaard Brouer 
Link: https://lore.kernel.org/bpf/20200819160710.1345956-1-vkabatov@redhat.com

selftests/bpf: Make test_varlen work with 32-bit user-space arch

2020-08-13T23:45:41+00:00

Despite bpftool generating data section memory layout that will work for
32-bit architectures on user-space side, BPF programs should be careful to not
use ambiguous types like `long`, which have different size in 32-bit and
64-bit environments. Fix that in test by using __u64 explicitly, which is
a recommended approach anyway.

Signed-off-by: Andrii Nakryiko 
Signed-off-by: Alexei Starovoitov 
Link: https://lore.kernel.org/bpf/20200813204945.1020225-10-andriin@fb.com

selftests/bpf: Correct various core_reloc 64-bit assumptions

2020-08-13T23:45:41+00:00

Ensure that types are memory layout- and field alignment-compatible regardless
of 32/64-bitness mix of libbpf and BPF architecture.

Signed-off-by: Andrii Nakryiko 
Signed-off-by: Alexei Starovoitov 
Link: https://lore.kernel.org/bpf/20200813204945.1020225-8-andriin@fb.com

selftests/bpf: Fix btf_dump test cases on 32-bit arches

2020-08-13T23:45:41+00:00

Fix btf_dump test cases by hard-coding BPF's pointer size of 8 bytes for cases
where it's impossible to deterimne the pointer size (no long type in BTF). In
cases where it's known, validate libbpf correctly determines it as 8.

Signed-off-by: Andrii Nakryiko 
Signed-off-by: Alexei Starovoitov 
Link: https://lore.kernel.org/bpf/20200813204945.1020225-6-andriin@fb.com

selftest/bpf: Fix compilation warnings in 32-bit mode

2020-08-13T23:45:41+00:00

Fix compilation warnings emitted when compiling selftests for 32-bit platform
(x86 in my case).

Signed-off-by: Andrii Nakryiko 
Signed-off-by: Alexei Starovoitov 
Link: https://lore.kernel.org/bpf/20200813204945.1020225-3-andriin@fb.com

bpf, selftests: Add tests to sock_ops for loading sk

2020-08-13T20:40:43+00:00

Add tests to directly accesse sock_ops sk field. Then use it to
ensure a bad pointer access will fault if something goes wrong.
We do three tests:

The first test ensures when we read sock_ops sk pointer into the
same register that we don't fault as described earlier. Here r9
is chosen as the temp register.  The xlated code is,

  36: (7b) *(u64 *)(r1 +32) = r9
  37: (61) r9 = *(u32 *)(r1 +28)
  38: (15) if r9 == 0x0 goto pc+3
  39: (79) r9 = *(u64 *)(r1 +32)
  40: (79) r1 = *(u64 *)(r1 +0)
  41: (05) goto pc+1
  42: (79) r9 = *(u64 *)(r1 +32)

The second test ensures the temp register selection does not collide
with in-use register r9. Shown here r8 is chosen because r9 is the
sock_ops pointer. The xlated code is as follows,

  46: (7b) *(u64 *)(r9 +32) = r8
  47: (61) r8 = *(u32 *)(r9 +28)
  48: (15) if r8 == 0x0 goto pc+3
  49: (79) r8 = *(u64 *)(r9 +32)
  50: (79) r9 = *(u64 *)(r9 +0)
  51: (05) goto pc+1
  52: (79) r8 = *(u64 *)(r9 +32)

And finally, ensure we didn't break the base case where dst_reg does
not equal the source register,

  56: (61) r2 = *(u32 *)(r1 +28)
  57: (15) if r2 == 0x0 goto pc+1
  58: (79) r2 = *(u64 *)(r1 +0)

Notice it takes us an extra four instructions when src reg is the
same as dst reg. One to save the reg, two to restore depending on
the branch taken and a goto to jump over the second restore.

Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Acked-by: Song Liu 
Acked-by: Martin KaFai Lau 
Link: https://lore.kernel.org/bpf/159718355325.4728.4163036953345999636.stgit@john-Precision-5820-Tower

bpf, selftests: Add tests for sock_ops load with r9, r8.r7 registers

2020-08-13T20:40:43+00:00

Loads in sock_ops case when using high registers requires extra logic to
ensure the correct temporary value is used. We need to ensure the temp
register does not use either the src_reg or dst_reg. Lets add an asm
test to force the logic is triggered.

The xlated code is here,

  30: (7b) *(u64 *)(r9 +32) = r7
  31: (61) r7 = *(u32 *)(r9 +28)
  32: (15) if r7 == 0x0 goto pc+2
  33: (79) r7 = *(u64 *)(r9 +0)
  34: (63) *(u32 *)(r7 +916) = r8
  35: (79) r7 = *(u64 *)(r9 +32)

Notice r9 and r8 are not used for temp registers and r7 is chosen.

Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Acked-by: Song Liu 
Acked-by: Martin KaFai Lau 
Link: https://lore.kernel.org/bpf/159718353345.4728.8805043614257933227.stgit@john-Precision-5820-Tower

bpf, selftests: Add tests for ctx access in sock_ops with single register

2020-08-13T20:40:43+00:00

To verify fix ("bpf: sock_ops ctx access may stomp registers in corner case")
we want to force compiler to generate the following code when accessing a
field with BPF_TCP_SOCK_GET_COMMON,

     r1 = *(u32 *)(r1 + 96) // r1 is skops ptr

Rather than depend on clang to do this we add the test with inline asm to
the tcpbpf test. This saves us from having to create another runner and
ensures that if we break this again test_tcpbpf will crash.

With above code we get the xlated code,

  11: (7b) *(u64 *)(r1 +32) = r9
  12: (61) r9 = *(u32 *)(r1 +28)
  13: (15) if r9 == 0x0 goto pc+4
  14: (79) r9 = *(u64 *)(r1 +32)
  15: (79) r1 = *(u64 *)(r1 +0)
  16: (61) r1 = *(u32 *)(r1 +2348)
  17: (05) goto pc+1
  18: (79) r9 = *(u64 *)(r1 +32)

We also add the normal case where src_reg != dst_reg so we can compare
code generation easily from llvm-objdump and ensure that case continues
to work correctly. The normal code is xlated to,

  20: (b7) r1 = 0
  21: (61) r1 = *(u32 *)(r3 +28)
  22: (15) if r1 == 0x0 goto pc+2
  23: (79) r1 = *(u64 *)(r3 +0)
  24: (61) r1 = *(u32 *)(r1 +2348)

Where the temp variable is not used.

Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Acked-by: Song Liu 
Acked-by: Martin KaFai Lau 
Link: https://lore.kernel.org/bpf/159718351457.4728.3295119261717842496.stgit@john-Precision-5820-Tower

selftests/bpf: Fix v4_to_v6 in sk_lookup

2020-08-11T13:36:51+00:00

I'm getting some garbage in bytes 8 and 9 when doing conversion
from sockaddr_in to sockaddr_in6 (leftover from AF_INET?). Let's
explicitly clear the higher bytes.

Fixes: 0ab5539f8584 ("selftests/bpf: Tests for BPF_SK_LOOKUP attach point")
Signed-off-by: Stanislav Fomichev 
Signed-off-by: Daniel Borkmann 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20200807223846.4190917-1-sdf@google.com