Skip to contentSkip to conversation

selftests/bpf: Add tests to assert that netfilter progs cannot write to skb

PATCH Superseded Florian Westphal 2 participants 3 messages Sep 2

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 selftests/bpf: Add tests to assert that netfilter progs cannot write to skb

The netfilter framework is allergic to ip header changing after validation done by ip/ipv6 stack.

Assert that bpf netfilter program type exposes readonly access to the skb, either via direct packet access or via dynptr API.

Following additional tests are expected to be rejected by verifier:

1. alter skb->len.
2. alter skb->data.
3. prog calls bpf_dynptr_slice_rdwr.
4. alter location returned by dynptr API.

Add following test case for bpf runtime:
- alter skb data via bpf_dynptr_write()

Test checks via __retval() that bpf_dynptr_write() returned nonzero value.

---
 .../bpf/progs/verifier_netfilter_ctx.c        | 78 +++++++++++++++++++
 1 file changed, 78 insertions(+)
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +78 -0
index e2cbc5bda65e..7e304331409f 100644
@@ -113,4 +113,82 @@ int with_valid_ctx_access_test6(struct bpf_nf_ctx *ctx)
 	return th->dest == bpf_htons(22) ? NF_ACCEPT : NF_DROP;
 }
 
+SEC("netfilter")
+__description("netfilter test prog with skb write access")
+__failure __msg("only read is supported")
+int skb_len_write(struct bpf_nf_ctx *ctx)
+{
+	ctx->skb->len = 1;
+	return 1;
+}
+
+SEC("netfilter")
+__description("netfilter test prog with skb data write access")
+__failure __msg("cannot write into rdonly_untrusted_mem")
+int with_invalid_data_write_test4(struct bpf_nf_ctx *ctx)
+{
+	ctx->skb->data[0] = 0;
+	return 1;
+}
+
+SEC("netfilter")
+__description("netfilter test prog with bpf_dynptr_write")
+__success __failure_unpriv
+__retval(0)
+int with_dynptr_write(struct bpf_nf_ctx *ctx)
+{
+	struct __sk_buff *skb = (struct __sk_buff *)ctx->skb;
+	struct bpf_dynptr ptr;
+	u8 buffer[1] = {};
+
+	if (bpf_dynptr_from_skb(skb, 0, &ptr))
+		return 1;
+
+	if (bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0))
+		return 0; /* must always fail */
+
+	return 1;
+}
+
+SEC("netfilter")
+__description("netfilter test prog with bpf_dynptr_slice_rdwr")
+__failure __msg("the prog does not allow writes to packet data")
+int with_dynptr_rdwr(struct bpf_nf_ctx *ctx)
+{
+	struct __sk_buff *skb = (struct __sk_buff *)ctx->skb;
+	u8 buffer_iph[20] = {};
+	struct bpf_dynptr ptr;
+	struct iphdr *iph;
+
+	if (bpf_dynptr_from_skb(skb, 0, &ptr))
+		return 1;
+
+	iph = bpf_dynptr_slice_rdwr(&ptr, 0, buffer_iph, sizeof(buffer_iph));
+	if (!iph)
+		return 0;
+
+	return 1;
+}
+
+SEC("netfilter")
+__description("netfilter test prog with bpf dynptr_slice_write + write")
+__failure __msg("cannot write into rdonly_mem")
+int with_dynptr_store(struct bpf_nf_ctx *ctx)
+{
+	struct __sk_buff *skb = (struct __sk_buff *)ctx->skb;
+	u8 buffer_iph[20] = {};
+	struct bpf_dynptr ptr;
+	struct iphdr *iph;
+
+	if (bpf_dynptr_from_skb(skb, 0, &ptr))
+		return 1;
+
+	iph = bpf_dynptr_slice(&ptr, 0, buffer_iph, sizeof(buffer_iph));
+	if (!iph)
+		return 0;
+	iph->protocol = 42;
+
+	return 1;
+}
+
 char _license[] SEC("license") = "GPL";
Signature · 1 line
2.55.0
#1 lore raw
This isn't a bug, but would a name like skb_data_write read better here, given with_invalid_ctx_access_test4() already exists in this file and the other new progs carry no test number?
Quoted 11 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +3 -0
index e2cbc5bda65e..7e304331409f 100644
@@ -113,4 +113,82 @@ int with_valid_ctx_access_test6(struct bpf_nf_ctx *ctx)
 	return th->dest == bpf_htons(22) ? NF_ACCEPT : NF_DROP;
 }
 
+SEC("netfilter")
+__description("netfilter test prog with skb write access")
+__failure __msg("only read is supported")

tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c int with_valid_ctx_access_test6(struct bpf_nf_ctx *ctx) · line 128

Quoted 10 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +10 -0
+int skb_len_write(struct bpf_nf_ctx *ctx)
+{
+	ctx->skb->len = 1;
+	return 1;
+}
+
+SEC("netfilter")
+__description("netfilter test prog with skb data write access")
+__failure __msg("cannot write into rdonly_untrusted_mem")
+int with_invalid_data_write_test4(struct bpf_nf_ctx *ctx)

This isn't a bug, but would a name like skb_data_write read better here, given with_invalid_ctx_access_test4() already exists in this file and the other new progs carry no test number?

Quoted 4 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +4 -0
+{
+	ctx->skb->data[0] = 0;
+	return 1;
+}

[ ... ]

tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c int with_valid_ctx_access_test6(struct bpf_nf_ctx *ctx) · line 176

Quoted 4 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +4 -0
+SEC("netfilter")
+__description("netfilter test prog with bpf dynptr_slice_write + write")
+__failure __msg("cannot write into rdonly_mem")
+int with_dynptr_store(struct bpf_nf_ctx *ctx)

This isn't a bug, but should this description say bpf_dynptr_slice() (the helper the prog actually calls) rather than 'bpf dynptr_slice_write'? The neighbouring descriptions spell the callee exactly ('netfilter test prog with bpf_dynptr_write', 'netfilter test prog with bpf_dynptr_slice_rdwr'), and no helper named dynptr_slice_write exists.

Quoted 6 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +6 -0
+{
+	struct __sk_buff *skb = (struct __sk_buff *)ctx->skb;
+	u8 buffer_iph[20] = {};
+	struct bpf_dynptr ptr;
+	struct iphdr *iph;
+

tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c int with_valid_ctx_access_test6(struct bpf_nf_ctx *ctx) · line 192

Quoted 10 lines
tools/testing/selftests/bpf/progs/verifier_netfilter_ctx.c +10 -0
+	if (bpf_dynptr_from_skb(skb, 0, &ptr))
+		return 1;
+
+	iph = bpf_dynptr_slice(&ptr, 0, buffer_iph, sizeof(buffer_iph));
+	if (!iph)
+		return 0;
+	iph->protocol = 42;
+
+	return 1;
+}

Regarding the commit message:

Quoted 2 lines

Assert that bpf netfilter program type exposes readonly access to the skb, either via direct packet access or via dynptr API.

This isn't a bug, but since netfilter progs have no direct packet access, would it be clearer to describe this case as writing through ctx->skb (BTF walk) instead of 'direct packet access'? The test (ctx->skb->data[0] = 0) walks a BTF pointer, which is why the expected message is 'cannot write into rdonly_untrusted_mem' rather than a packet-access error.

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/33601668670

#2 lore raw
Quoted 8 lines · [email protected] <[email protected]> wrote:
Quoted 2 lines

Assert that bpf netfilter program type exposes readonly access to the skb, either via direct packet access or via dynptr API.

This isn't a bug, but since netfilter progs have no direct packet access, would it be clearer to describe this case as writing through ctx->skb (BTF walk) instead of 'direct packet access'? The test (ctx->skb->data[0] = 0) walks a BTF pointer, which is why the expected message is 'cannot write into rdonly_untrusted_mem' rather than a packet-access error.

I'll shorten this to:

"Assert that bpf netfilter programs do not allow skb write access."

I'll also follow the other suggested changes to function names.