From: Ihor Solodrai <ihor.solodrai@linux.dev>
To: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>,
Andrii Nakryiko <andrii@kernel.org>, bpf <bpf@vger.kernel.org>,
Alexei Starovoitov <ast@kernel.org>,
Daniel Borkmann <daniel@iogearbox.net>,
Eduard <eddyz87@gmail.com>, Mykola Lysenko <mykolal@fb.com>,
Yonghong Song <yonghong.song@linux.dev>,
Kernel Team <kernel-team@meta.com>
Subject: Re: [PATCH bpf-next v3 3/3] selftests/bpf: add test cases with CONST_PTR_TO_MAP null checks
Date: Thu, 5 Jun 2025 16:40:31 -0700 [thread overview]
Message-ID: <8c24eae8-a932-4c1d-87ec-c5f8ef8fdf79@linux.dev> (raw)
In-Reply-To: <CAADnVQJ-sxOEdzy7OktZrTUDxk7Rw7H3zCt_u+iM987zTTDksw@mail.gmail.com>
On 6/5/25 10:00 AM, Alexei Starovoitov wrote:
> On Thu, Jun 5, 2025 at 9:42 AM Ihor Solodrai <ihor.solodrai@linux.dev> wrote:
>>
>> On 6/5/25 9:24 AM, Andrii Nakryiko wrote:
>>> On Wed, Jun 4, 2025 at 3:28 PM Ihor Solodrai <isolodrai@meta.com> wrote:
>>>>
>>>> [...]
>>>>
>>>> +SEC("socket")
>>>> +int map_ptr_is_never_null(void *ctx)
>>>> +{
>>>> + struct bpf_map *maps[2] = { 0 };
>>>> + struct bpf_map *map = NULL;
>>>> + int __attribute__((aligned(8))) key = 0;
>>>
>>> aligned(8) makes any difference?
>>
>> Yes. If not aligned (explicitly or by accident), verification fails
>> with "math between fp pointer and register with unbounded min value is
>> not allowed" at maps[key]. See the log below.
>
> It's not clear to me why "aligned" changes code gen,
> but let's not rely on it.
> Try 'unsigned int key' instead.
> Also note that &key part pessimizes the code.
> Do for (...; i < 2; i++) {u32 key = i; &key }
> instead.
I've tried changing the test like this:
@@ -144,12 +144,12 @@ int map_ptr_is_never_null(void *ctx)
{
struct bpf_map *maps[2] = { 0 };
struct bpf_map *map = NULL;
- int __attribute__((aligned(8))) key = 0;
- for (key = 0; key < 2; key++) {
+ for (int i = 0; i < 2; i++) {
+ __u32 key = i;
map = bpf_map_lookup_elem(&map_in_map, &key);
if (map)
- maps[key] = map;
+ maps[i] = map;
else
return 0;
}
This version passes verification independent of the first patch being
applied, which kinda defeats the purpose of the test. Verifier log
below:
0: R1=ctx() R10=fp0
; int map_ptr_is_never_null(void *ctx) @ verifier_map_in_map.c:143
0: (b4) w1 = 0 ; R1_w=0
; __u32 key = i; @ verifier_map_in_map.c:149
1: (63) *(u32 *)(r10 -4) = r1
mark_precise: frame0: last_idx 1 first_idx 0 subseq_idx -1
mark_precise: frame0: regs=r1 stack= before 0: (b4) w1 = 0
2: R1_w=0 R10=fp0 fp-8=0000????
2: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
3: (07) r2 += -4 ; R2_w=fp-4
; map = bpf_map_lookup_elem(&map_in_map, &key); @
verifier_map_in_map.c:150
4: (18) r1 = 0xff302cd6802e0a00 ;
R1_w=map_ptr(map=map_in_map,ks=4,vs=4)
6: (85) call bpf_map_lookup_elem#1 ;
R0_w=map_value_or_null(id=1,map=map_in_map,ks=4,vs=4)
; if (map) @ verifier_map_in_map.c:151
7: (15) if r0 == 0x0 goto pc+7 ; R0_w=map_ptr(ks=4,vs=4)
8: (b4) w1 = 1 ; R1_w=1
; __u32 key = i; @ verifier_map_in_map.c:149
9: (63) *(u32 *)(r10 -4) = r1 ; R1_w=1 R10=fp0 fp-8=mmmm????
10: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
11: (07) r2 += -4 ; R2_w=fp-4
; map = bpf_map_lookup_elem(&map_in_map, &key); @
verifier_map_in_map.c:150
12: (18) r1 = 0xff302cd6802e0a00 ;
R1_w=map_ptr(map=map_in_map,ks=4,vs=4)
14: (85) call bpf_map_lookup_elem#1 ;
R0=map_value_or_null(id=2,map=map_in_map,ks=4,vs=4)
; } @ verifier_map_in_map.c:164
15: (b4) w0 = 0 ; R0_w=0
16: (95) exit
If map[i] is changed back to map[key] like this:
for (int i = 0; i < 2; i++) {
__u32 key = i;
map = bpf_map_lookup_elem(&map_in_map, &key);
if (map)
maps[key] = map; /* change here */
else
return 0;
}
Verification fails with "invalid unbounded variable-offset write to
stack R2"
VERIFIER LOG:
=============
arg#0 reference type('UNKNOWN ') size cannot be determined: -22
0: R1=ctx() R10=fp0
; int map_ptr_is_never_null(void *ctx) @ verifier_map_in_map.c:143
0: (b7) r1 = 0 ; R1_w=0
; struct bpf_map *maps[2] = { 0 }; @ verifier_map_in_map.c:145
1: (7b) *(u64 *)(r10 -8) = r1 ; R1_w=0 R10=fp0 fp-8_w=0
2: (7b) *(u64 *)(r10 -16) = r1 ; R1_w=0 R10=fp0 fp-16_w=0
3: (b4) w1 = 0 ; R1_w=0
; __u32 key = i; @ verifier_map_in_map.c:149
4: (63) *(u32 *)(r10 -20) = r1 ; R1_w=0 R10=fp0 fp-24=0000????
5: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
6: (07) r2 += -20 ; R2_w=fp-20
; map = bpf_map_lookup_elem(&map_in_map, &key); @
verifier_map_in_map.c:150
7: (18) r1 = 0xff1f1a9b829ae400 ;
R1_w=map_ptr(map=map_in_map,ks=4,vs=4)
9: (85) call bpf_map_lookup_elem#1 ;
R0_w=map_value_or_null(id=1,map=map_in_map,ks=4,vs=4)
; if (map) @ verifier_map_in_map.c:151
10: (15) if r0 == 0x0 goto pc+24 ; R0_w=map_ptr(ks=4,vs=4)
; maps[key] = map; @ verifier_map_in_map.c:152
11: (61) r1 = *(u32 *)(r10 -20) ; R1_w=0 R10=fp0 fp-24=0000????
12: (67) r1 <<= 3 ; R1_w=0
13: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
14: (07) r2 += -16 ; R2_w=fp-16
15: (0f) r2 += r1 ; R1_w=0 R2_w=fp-16
16: (7b) *(u64 *)(r2 +0) = r0 ; R0_w=map_ptr(ks=4,vs=4)
R2_w=fp-16 fp-16_w=map_ptr(ks=4,vs=4)
17: (b4) w1 = 1 ; R1_w=1
; __u32 key = i; @ verifier_map_in_map.c:149
18: (63) *(u32 *)(r10 -20) = r1 ; R1_w=1 R10=fp0 fp-24=mmmm????
19: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
20: (07) r2 += -20 ; R2_w=fp-20
; map = bpf_map_lookup_elem(&map_in_map, &key); @
verifier_map_in_map.c:150
21: (18) r1 = 0xff1f1a9b829ae400 ;
R1_w=map_ptr(map=map_in_map,ks=4,vs=4)
23: (85) call bpf_map_lookup_elem#1 ;
R0=map_value_or_null(id=2,map=map_in_map,ks=4,vs=4)
; if (map) @ verifier_map_in_map.c:151
24: (15) if r0 == 0x0 goto pc+10 ; R0=map_ptr(ks=4,vs=4)
; maps[key] = map; @ verifier_map_in_map.c:152
25: (61) r1 = *(u32 *)(r10 -20) ;
R1_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
R10=fp0 fp-24=mmmm????
26: (67) r1 <<= 3 ;
R1_w=scalar(smin=0,smax=umax=0x7fffffff8,smax32=0x7ffffff8,umax32=0xfffffff8,var_off=(0x0;
0x7fffffff8))
27: (bf) r2 = r10 ; R2_w=fp0 R10=fp0
28: (07) r2 += -16 ; R2_w=fp-16
29: (0f) r2 += r1 ;
R1_w=scalar(smin=0,smax=umax=0x7fffffff8,smax32=0x7ffffff8,umax32=0xfffffff8,var_off=(0x0;
0x7fffffff8))
R2_w=fp(off=-16,smin=0,smax=umax=0x7fffffff8,smax32=0x7ffffff8,umax32=0xfffffff8,var_off=(0x0;
0x7fffffff8))
30: (7b) *(u64 *)(r2 +0) = r0
invalid unbounded variable-offset write to stack R2
... which can be "fixed" by aligning the local key variable
__u32 __attribute__((aligned(8))) key = i;
but that puts us back to square one.
It appears that alignment becomes a problem if the variable is used as
array index and also it's address is passed to a helper.
And if different vars are used for array access and map lookup, then
map_ptr nullness issue doesn't reproduce.
Any suggestions?
next prev parent reply other threads:[~2025-06-05 23:40 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-06-04 22:27 [PATCH bpf-next v3 1/3] bpf: make reg_not_null() true for CONST_PTR_TO_MAP Ihor Solodrai
2025-06-04 22:27 ` [PATCH bpf-next v3 2/3] selftests/bpf: add cmp_map_pointer_with_const test Ihor Solodrai
2025-06-04 22:41 ` Alexei Starovoitov
2025-06-05 3:04 ` Ihor Solodrai
2025-06-05 16:08 ` Alexei Starovoitov
2025-06-05 17:17 ` Ihor Solodrai
2025-06-05 17:42 ` Yonghong Song
2025-06-05 18:11 ` Alexei Starovoitov
2025-06-06 6:24 ` Yonghong Song
2025-06-05 16:20 ` Andrii Nakryiko
2025-06-05 16:30 ` Ihor Solodrai
2025-06-04 22:27 ` [PATCH bpf-next v3 3/3] selftests/bpf: add test cases with CONST_PTR_TO_MAP null checks Ihor Solodrai
2025-06-05 16:24 ` Andrii Nakryiko
2025-06-05 16:42 ` Ihor Solodrai
2025-06-05 17:00 ` Alexei Starovoitov
2025-06-05 23:40 ` Ihor Solodrai [this message]
2025-06-06 0:25 ` Alexei Starovoitov
2025-06-06 23:37 ` Ihor Solodrai
2025-06-06 23:52 ` Alexei Starovoitov
2025-06-07 14:07 ` Alexei Starovoitov
2025-06-05 16:27 ` [PATCH bpf-next v3 1/3] bpf: make reg_not_null() true for CONST_PTR_TO_MAP Andrii Nakryiko
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=8c24eae8-a932-4c1d-87ec-c5f8ef8fdf79@linux.dev \
--to=ihor.solodrai@linux.dev \
--cc=alexei.starovoitov@gmail.com \
--cc=andrii.nakryiko@gmail.com \
--cc=andrii@kernel.org \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=eddyz87@gmail.com \
--cc=kernel-team@meta.com \
--cc=mykolal@fb.com \
--cc=yonghong.song@linux.dev \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.