bpf: fix nullness propagation for reg to reg comparisons
Subsystems (4)
PATCH v2 1/2 bpf: fix nullness propagation for reg to reg comparisons
After befae75856ab, the verifier would propagate null information after JEQ/JNE, e.g., if two pointers, one is maybe_null and the other is not, the former would be marked as non-null in eq path. However, as comment "PTR_TO_BTF_ID points to a kernel struct that does not need to be null checked by the BPF program ... The verifier must keep this in mind and can make no assumptions about null or non-null when doing branch ...". If one pointer is maybe_null and the other is PTR_TO_BTF, the former is incorrectly marked non-null. The following BPF prog can trigger a null-ptr-deref, also see this report for more details[1]:
0: (18) r1 = map_fd ; R1_w=map_ptr(ks=4, vs=4) 2: (79) r6 = *(u64 *)(r1 +8) ; R6_w=bpf_map->inner_map_data ; R6 is PTR_TO_BTF_ID ; equals to null at runtime 3: (bf) r2 = r10 4: (07) r2 += -4 5: (62) *(u32 *)(r2 +0) = 0 6: (85) call bpf_map_lookup_elem#1 ; R0_w=map_value_or_null 7: (1d) if r6 == r0 goto pc+1 8: (95) exit ; from 7 to 9: R0=map_value R6=ptr_bpf_map 9: (61) r0 = *(u32 *)(r0 +0) ; null-ptr-deref 10: (95) exit
So, make the verifier propagate nullness information for reg to reg comparisons only if neither reg is PTR_TO_BTF_ID.
[1] https://lore.kernel.org/bpf/CACkBjsaFJwjC5oiw-1KXvcazywodwXo4zGYsRHwbr2gSG9WcSw@mail.gmail.com/T/#u
--- kernel/bpf/verifier.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
kernel/bpf/verifier.c +8 -1
@@ -11822,10 +11822,17 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env,* register B - not null * for JNE A, B, ... - A is not null in the false branch; * for JEQ A, B, ... - A is not null in the true branch. +* +* Since PTR_TO_BTF_ID points to a kernel struct that does +* not need to be null checked by the BPF program, i.e., +* could be null even without PTR_MAYBE_NULL marking, so +* only propagate nullness when neither reg is that type. */ if (!is_jmp32 &&BPF_SRC (insn->code) ==BPF_X &&__is_pointer_value (false, src_reg) &&__is_pointer_value (false, dst_reg) && -type_may_be_null (src_reg->type) !=type_may_be_null (dst_reg->type)) { +type_may_be_null (src_reg->type) !=type_may_be_null (dst_reg->type) && +base_type (src_reg->type) !=PTR_TO_BTF_ID && +base_type (dst_reg->type) !=PTR_TO_BTF_ID ) { eq_branch_regs =NULL ;switch (opcode) {case BPF_JEQ :
base-commit: ef3911a3e4d65d2db617366e79517b896045a6e2
Signature · 1 line
2.38.1
PATCH v2 2/2 selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
Verify that nullness information is not porpagated in the branches of register to register JEQ and JNE operations if one of them is PTR_TO_BTF_ID.
--- .../bpf/verifier/jeq_infer_not_null.c | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+)
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +22 -0
@@ -172,3 +172,25 @@ .prog_type = BPF_PROG_TYPE_XDP, .result = ACCEPT, }, +{ + "jne/jeq infer not null, PTR_TO_MAP_OR_NULL unchanged with PTR_TO_BTF_ID reg", + .insns = { +BPF_MOV64_REG (BPF_REG_2 ,BPF_REG_10 ), +BPF_ALU64_IMM (BPF_ADD ,BPF_REG_2 ,-8 ), +BPF_ST_MEM (BPF_DW ,BPF_REG_2 ,0 ,0 ), +BPF_LD_MAP_FD (BPF_REG_1 ,0 ), +/* r6 = bpf_map->inner_map_meta; */ +BPF_LDX_MEM (BPF_DW ,BPF_REG_6 ,BPF_REG_1 ,8 ), +/* r0 = map_lookup_elem(r1, r2); */ +BPF_EMIT_CALL (BPF_FUNC_map_lookup_elem), +/* if (r0 == r6) read *r0; */ +BPF_JMP_REG (BPF_JEQ ,BPF_REG_6 ,BPF_REG_0 ,1 ), +BPF_EXIT_INSN (), +BPF_LDX_MEM (BPF_W ,BPF_REG_0 ,BPF_REG_0 ,0 ), + BPF_EXIT_INSN(), + }, + .fixup_map_hash_8b = { 3 }, + .prog_type = BPF_PROG_TYPE_XDP, + .result = REJECT, + .errstr = "R0 invalid mem access 'map_value_or_null'", +},
Signature · 1 line
2.38.1
Martin KaFai Lau
reviewerRe: [PATCH bpf-next v2 2/2] selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
Quoted 3 lines · Hao Sun
Verify that nullness information is not porpagated in the branches of register to register JEQ and JNE operations if one of them is
PTR_TO_BTF_ID.
Thanks for the fix and test.
Quoted 14 lines
Signed-off-by Hao Sun <[email protected]>Acked-by Yonghong Song <[email protected]>--- .../bpf/verifier/jeq_infer_not_null.c | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+)tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +0
-0@@ -172,3 +172,25 @@ .prog_type = BPF_PROG_TYPE_XDP, .result = ACCEPT,
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c line 183
Quoted 10 lines
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +9
-0}, +{ + "jne/jeq infer not null, PTR_TO_MAP_OR_NULL unchanged with PTR_TO_BTF_ID reg", + .insns = { +BPF_MOV64_REG (BPF_REG_2 ,BPF_REG_10 ), +BPF_ALU64_IMM (BPF_ADD ,BPF_REG_2 ,-8 ), +BPF_ST_MEM (BPF_DW ,BPF_REG_2 ,0 ,0 ), +BPF_LD_MAP_FD (BPF_REG_1 ,0 ), +/* r6 = bpf_map->inner_map_meta; */ + BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_1, 8),
This bpf_map->inner_map_meta requires CO-RE. It works now but could be fragile in different platform and in the future bpf_map changes. Take a look at the map_ptr_kern.c which uses "__attribute__((preserve_access_index))" at the "struct bpf_map".
Please translate this verifer test into a proper bpf prog in C code such that it can use the CO-RE in libbpf. It should run under test_progs instead of test_verifier. The bpf prog can include the "vmlinux.h" to get the "__attribute__((preserve_access_index))" for free. Take a look at https://lore.kernel.org/all/[email protected]/ which has example on how to check verifier message in test_progs.
Quoted 13 lines
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +13
-0+/* r0 = map_lookup_elem(r1, r2); */ +BPF_EMIT_CALL (BPF_FUNC_map_lookup_elem), +/* if (r0 == r6) read *r0; */ + BPF_JMP_REG(BPF_JEQ, BPF_REG_6, BPF_REG_0, 1), +BPF_EXIT_INSN (), + BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_0, 0), +BPF_EXIT_INSN (), + }, + .fixup_map_hash_8b = { 3 }, + .prog_type = BPF_PROG_TYPE_XDP, + .result = REJECT, + .errstr = "R0 invalid mem access 'map_value_or_null'", +},
Re: [PATCH bpf-next v2 2/2] selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
Quoted 36 lines · Martin KaFai Lau
Quoted 3 lines · Hao Sun
Verify that nullness information is not porpagated in the branches of register to register JEQ and JNE operations if one of them is
PTR_TO_BTF_ID.Thanks for the fix and test.
Quoted 12 lines
Signed-off-by Hao Sun <[email protected]>Acked-by Yonghong Song <[email protected]>--- .../bpf/verifier/jeq_infer_not_null.c | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+)tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +0
-0@@ -172,3 +172,25 @@ .prog_type = BPF_PROG_TYPE_XDP, .result = ACCEPT,tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c line 183
Quoted 10 lines
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +9
-0}, +{ + "jne/jeq infer not null, PTR_TO_MAP_OR_NULL unchanged with PTR_TO_BTF_ID reg", + .insns = { +BPF_MOV64_REG (BPF_REG_2 ,BPF_REG_10 ), +BPF_ALU64_IMM (BPF_ADD ,BPF_REG_2 ,-8 ), +BPF_ST_MEM (BPF_DW ,BPF_REG_2 ,0 ,0 ), +BPF_LD_MAP_FD (BPF_REG_1 ,0 ), +/* r6 = bpf_map->inner_map_meta; */ + BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_1, 8),This
bpf_map->inner_map_metarequires CO-RE. It works now but could be fragile in different platform and in the futurebpf_mapchanges. Take a look at themap_ptr_kern.cwhich uses "__attribute__((preserve_access_index))" at the "structbpf_map".
Quoted 2 lines
Please translate this verifer test into a proper bpf prog in C code such that it can use the CO-RE in libbpf. It should run under
test_progsinstead oftest_verifier. The bpf prog can include the "vmlinux.h" to get the "__attribute__((preserve_access_index))" for free. Take a look at https://lore.kernel.org/all/[email protected]/ which has example on how to check verifier message intest_progs.
Sure, thanks for the hint, will send patch v3 soon.
Thanks
Hao
Re: [PATCH bpf-next v2 2/2] selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
In reply to Martin KaFai LauQuoted 36 lines · Martin KaFai Lau
Quoted 3 lines · Hao Sun
Verify that nullness information is not porpagated in the branches of register to register JEQ and JNE operations if one of them is
PTR_TO_BTF_ID.Thanks for the fix and test.
Quoted 12 lines
Signed-off-by Hao Sun <[email protected]>Acked-by Yonghong Song <[email protected]>--- .../bpf/verifier/jeq_infer_not_null.c | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+)tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +0
-0@@ -172,3 +172,25 @@ .prog_type = BPF_PROG_TYPE_XDP, .result = ACCEPT,tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c line 183
Quoted 10 lines
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c +9
-0}, +{ + "jne/jeq infer not null, PTR_TO_MAP_OR_NULL unchanged with PTR_TO_BTF_ID reg", + .insns = { +BPF_MOV64_REG (BPF_REG_2 ,BPF_REG_10 ), +BPF_ALU64_IMM (BPF_ADD ,BPF_REG_2 ,-8 ), +BPF_ST_MEM (BPF_DW ,BPF_REG_2 ,0 ,0 ), +BPF_LD_MAP_FD (BPF_REG_1 ,0 ), +/* r6 = bpf_map->inner_map_meta; */ + BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_1, 8),This
bpf_map->inner_map_metarequires CO-RE. It works now but could be fragile in different platform and in the futurebpf_mapchanges. Take a look at themap_ptr_kern.cwhich uses "__attribute__((preserve_access_index))" at the "structbpf_map".
Quoted 2 lines
Please translate this verifer test into a proper bpf prog in C code such that it can use the CO-RE in libbpf. It should run under
test_progsinstead oftest_verifier. The bpf prog can include the "vmlinux.h" to get the "__attribute__((preserve_access_index))" for free. Take a look at https://lore.kernel.org/all/[email protected]/ which has example on how to check verifier message intest_progs.
Hi,
I’ve tried something like the bellow, but soon realized that this won’t work because once compiler figures out inner_map equals to val, it can choose either reg to write into in the following path, meaning that this program can be rejected due to writing into read-only PTR_TO_BTF_ID reg, and this makes the test useless.
Essentially, we want two regs, one points to PTR_TO_BTD_ID, one points to MAP_VALUR_OR_NULL, then compare them and deref map val. It’s hard to implement this in C level because compilers decide which reg to use but not us, maybe we can just drop this test.
thoughts?
+struct {
+ __uint(type, BPF_MAP_TYPE_HASH);
+ __uint(max_entries, 1);
+ __type(key, u64);
+ __type(value, u64);
+} m_hash SEC(".maps");
+
+SEC("?raw_tp")
+__failure __msg("invalid mem access 'map_value_or_null") +int jeq_infer_not_null_ptr_to_btfid(void *ctx) +{
+ struct bpf_map *map = (struct bpf_map *)&m_hash;
+ struct bpf_map *inner_map = map->inner_map_meta;
+ u64 key = 0, ret = 0, *val;
+
+ val = bpf_map_lookup_elem(map, &key);
+ /* Do not mark ptr as non-null if one of them is
+ * PTR_TO_BTF_ID, reject because of invalid access
+ * to map value.
+ */
+ if (val == inner_map)
+ ret = *val;
+
+ return ret;
+}
Martin KaFai Lau
reviewerRe: [PATCH bpf-next v2 2/2] selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
Quoted 7 lines · Hao Sun
Hi,
I’ve tried something like the bellow, but soon realized that this won’t work because once compiler figures out
inner_mapequals toval, it can choose either reg to write into in the following path, meaning that this program can be rejected due to writing into read-onlyPTR_TO_BTF_IDreg, and this makes the test useless.
hmm... I read the above a few times but I still don't quite get it. In particular, '...can be rejected due to writing into read-only PTR_TO_BTF_ID reg...'. Where is it writing into a read-only PTR_TO_BTF_ID reg in the following bpf prog? Did I overlook something?
Quoted 3 lines
Essentially, we want two regs, one points to
PTR_TO_BTD_ID, one points toMAP_VALUR_OR_NULL, then compare them and deref map val.
If I read this request correctly, I guess the compiler has changed 'ret = *val' to 'ret = *inner_map'? Thus, the verifier did not reject because it deref a PTR_TO_BTF_ID?
Quoted 2 lines
It’s hard to implement this in C level because compilers decide which reg to use but not us, maybe we can just drop this test.
Have you tried inline assembly. Something like this (untested):
asm volatile (
"r8 = %[val];\n"
"r9 = %[inner_map];\n"
"if r8 != r9 goto +1;\n"
"%[ret] = *(u64 *)(r8 +0);\n"
:[ret] "+r"(ret)
: [inner_map] "r"(inner_map), [val] "r"(val)
:"r8", "r9");Please attach the verifier output in the future. It will be easier to understand.
Quoted 28 lines
thoughts?
(quoted diff) +25
-0+struct { + __uint(type, BPF_MAP_TYPE_HASH); + __uint(max_entries, 1); + __type(key, u64); + __type(value, u64); +} m_hash SEC(".maps"); + +SEC("?raw_tp") +__failure __msg("invalid mem access 'map_value_or_null") +int jeq_infer_not_null_ptr_to_btfid(void *ctx) +{ + struct bpf_map *map = (struct bpf_map *)&m_hash; + struct bpf_map *inner_map = map->inner_map_meta; + u64 key = 0, ret = 0, *val; + + val = bpf_map_lookup_elem(map, &key); + /* Do not mark ptr as non-null if one of them is + * PTR_TO_BTF_ID, reject because of invalid access + * to map value. + */ + if (val == inner_map) + ret = *val; + + return ret; +}
Re: [PATCH bpf-next v2 2/2] selftests/bpf: check null propagation only neither reg is PTR_TO_BTF_ID
Quoted 16 lines · Martin KaFai Lau <[email protected]> 于2022年12月22日周四 05:21写道:
Quoted 7 lines · Hao Sun
Hi,
I’ve tried something like the bellow, but soon realized that this won’t work because once compiler figures out
inner_mapequals toval, it can choose either reg to write into in the following path, meaning that this program can be rejected due to writing into read-onlyPTR_TO_BTF_IDreg, and this makes the test useless.hmm... I read the above a few times but I still don't quite get it. In particular, '...can be rejected due to writing into read-only
PTR_TO_BTF_IDreg...'. Where is it writing into a read-onlyPTR_TO_BTF_IDreg in the following bpf prog? Did I overlook something?Quoted 1 line
Quoted 7 lines
Quoted 2 lines
Essentially, we want two regs, one points to
PTR_TO_BTD_ID, one points toMAP_VALUR_OR_NULL, then compare them and deref map val.If I read this request correctly, I guess the compiler has changed 'ret = *val' to 'ret =
*inner_map'? Thus, the verifier did not reject because it deref aPTR_TO_BTF_ID?
Yes, and if we do "*val = 0", it's rejected due to writing to read-only PTR_TO_BTF_ID reg.
Quoted 5 lines
Quoted 2 lines
It’s hard to implement this in C level because compilers decide which reg to use but not us, maybe we can just drop this test.
Have you tried inline assembly. Something like this (untested):
Quoted 9 lines
asm volatile (
"r8 = %[val];\n"
"r9 = %[inner_map];\n"
"if r8 != r9 goto +1;\n"
"%[ret] = *(u64 *)(r8 +0);\n"
:[ret] "+r"(ret)
: [inner_map] "r"(inner_map), [val] "r"(val)
:"r8", "r9");This would work, didn't realize that I can inline BPF insns this way. Thanks!
Quoted 2 lines
Please attach the verifier output in the future. It will be easier to understand.
Will do the next time.