Skip to contentSkip to conversation

support SKF_NET_OFF and SKF_LL_OFF on skb frags

PATCH Applied Willem de Bruijn 3 participants 5 messages Apr 8, 2025 – Apr 10, 2025

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v3 cover support SKF_NET_OFF and SKF_LL_OFF on skb frags

From: Willem de Bruijn <[email protected]>

Address a longstanding issue that may lead to missed packets depending on system configuration.

Ensure that reading from packet contents works regardless of skb geometry, also when using the special SKF_.. negative offsets to offset from L2 or L3 header.

Patch 2 is the selftest for the fix.

v2->v3
- do not remove bpf_internal_load_pointer_neg_helper, because it is still used in the sparc32 JIT

v1->v2
- introduce bfp_skb_load_helper_convert_offset to avoid open coding
- selftest: add comment why early demux must be disabled

v2: https://lore.kernel.org/netdev/[email protected]/
v1: https://lore.kernel.org/netdev/[email protected]/

Willem de Bruijn (2):
bpf: support SKF_NET_OFF and SKF_LL_OFF on skb frags selftests/net: test sk_filter support for SKF_NET_OFF on frags

 net/core/filter.c                          |  80 ++++---
 tools/testing/selftests/net/.gitignore     |   1 +
 tools/testing/selftests/net/Makefile       |   2 +
 tools/testing/selftests/net/skf_net_off.c  | 244 +++++++++++++++++++++
 tools/testing/selftests/net/skf_net_off.sh |  30 +++
 5 files changed, 321 insertions(+), 36 deletions(-)
 create mode 100644 tools/testing/selftests/net/skf_net_off.c
 create mode 100755 tools/testing/selftests/net/skf_net_off.sh
Signature · 1 line
2.49.0.504.g3bcea36a83-goog
#1 lore raw

PATCH v3 1/2 bpf: support SKF_NET_OFF and SKF_LL_OFF on skb frags

From: Willem de Bruijn <[email protected]>

Classic BPF socket filters with SKB_NET_OFF and SKB_LL_OFF fail to read when these offsets extend into frags.

This has been observed with iwlwifi and reproduced with tun with IFF_NAPI_FRAGS. The below straightforward socket filter on UDP port, applied to a RAW socket, will silently miss matching packets.

    const int offset_proto = offsetof(struct ip6_hdr, ip6_nxt);
    const int offset_dport = sizeof(struct ip6_hdr) + offsetof(struct udphdr, dest);
    struct sock_filter filter_code[] = {
            BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, SKF_AD_OFF + SKF_AD_PKTTYPE),
            BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, PACKET_HOST, 0, 4),
            BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, SKF_NET_OFF + offset_proto),
            BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 2),
            BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, SKF_NET_OFF + offset_dport),

This is unexpected behavior. Socket filter programs should be consistent regardless of environment. Silent misses are particularly concerning as hard to detect.

Use skb_copy_bits for offsets outside linear, same as done for non-SKF_(LL|NET) offsets.

Offset is always positive after subtracting the reference threshold SKB_(LL|NET)_OFF, so is always >= skb_(mac|network)_offset. The sum of the two is an offset against skb->data, and may be negative, but it cannot point before skb->head, as skb_(mac|network)_offset would too.

This appears to go back to when frag support was introduced to sk_run_filter in linux-2.4.4, before the introduction of git.

The amount of code change and 8/16/32 bit duplication are unfortunate. But any attempt I made to be smarter saved very few LoC while complicating the code.

---

v2->v3
- do not remove bpf_internal_load_pointer_neg_helper, because it is still used in the sparc32 JIT
v1->v2
- introduce bfp_skb_load_helper_convert_offset to avoid open coding

---
 net/core/filter.c | 80 ++++++++++++++++++++++++++---------------------
 1 file changed, 44 insertions(+), 36 deletions(-)
net/core/filter.c +44 -36
index bc6828761a47..79cab4d78dc3 100644
@@ -218,24 +218,36 @@ BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 	return 0;
 }
 
+static int bpf_skb_load_helper_convert_offset(const struct sk_buff *skb, int offset)
+{
+	if (likely(offset >= 0))
+		return offset;
+
+	if (offset >= SKF_NET_OFF)
+		return offset - SKF_NET_OFF + skb_network_offset(skb);
+
+	if (offset >= SKF_LL_OFF && skb_mac_header_was_set(skb))
+		return offset - SKF_LL_OFF + skb_mac_offset(skb);
+
+	return INT_MIN;
+}
+
 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 	   data, int, headlen, int, offset)
 {
-	u8 tmp, *ptr;
+	u8 tmp;
 	const int len = sizeof(tmp);
 
-	if (offset >= 0) {
-		if (headlen - offset >= len)
-			return *(u8 *)(data + offset);
-		if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
-			return tmp;
-	} else {
-		ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
-		if (likely(ptr))
-			return *(u8 *)ptr;
-	}
+	offset = bpf_skb_load_helper_convert_offset(skb, offset);
+	if (offset == INT_MIN)
+		return -EFAULT;
 
-	return -EFAULT;
+	if (headlen - offset >= len)
+		return *(u8 *)(data + offset);
+	if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
+		return tmp;
+	else
+		return -EFAULT;
 }
 
 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
@@ -248,21 +260,19 @@ BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 	   data, int, headlen, int, offset)
 {
-	__be16 tmp, *ptr;
+	__be16 tmp;
 	const int len = sizeof(tmp);
 
-	if (offset >= 0) {
-		if (headlen - offset >= len)
-			return get_unaligned_be16(data + offset);
-		if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
-			return be16_to_cpu(tmp);
-	} else {
-		ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
-		if (likely(ptr))
-			return get_unaligned_be16(ptr);
-	}
+	offset = bpf_skb_load_helper_convert_offset(skb, offset);
+	if (offset == INT_MIN)
+		return -EFAULT;
 
-	return -EFAULT;
+	if (headlen - offset >= len)
+		return get_unaligned_be16(data + offset);
+	if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
+		return be16_to_cpu(tmp);
+	else
+		return -EFAULT;
 }
 
 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
@@ -275,21 +285,19 @@ BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 	   data, int, headlen, int, offset)
 {
-	__be32 tmp, *ptr;
+	__be32 tmp;
 	const int len = sizeof(tmp);
 
-	if (likely(offset >= 0)) {
-		if (headlen - offset >= len)
-			return get_unaligned_be32(data + offset);
-		if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
-			return be32_to_cpu(tmp);
-	} else {
-		ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
-		if (likely(ptr))
-			return get_unaligned_be32(ptr);
-	}
+	offset = bpf_skb_load_helper_convert_offset(skb, offset);
+	if (offset == INT_MIN)
+		return -EFAULT;
 
-	return -EFAULT;
+	if (headlen - offset >= len)
+		return get_unaligned_be32(data + offset);
+	if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
+		return be32_to_cpu(tmp);
+	else
+		return -EFAULT;
 }
 
 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
Signature · 1 line
2.49.0.504.g3bcea36a83-goog
Acked-by on v3 1/2
#2 lore raw

Re: [PATCH bpf v3 1/2] bpf: support SKF_NET_OFF and SKF_LL_OFF on skb frags

Quoted 45 lines · Willem de Bruijn

From: Willem de Bruijn <[email protected]>

Classic BPF socket filters with SKB_NET_OFF and SKB_LL_OFF fail to read when these offsets extend into frags.

This has been observed with iwlwifi and reproduced with tun with IFF_NAPI_FRAGS. The below straightforward socket filter on UDP port, applied to a RAW socket, will silently miss matching packets.

    const int offset_proto = offsetof(struct ip6_hdr, ip6_nxt);
    const int offset_dport = sizeof(struct ip6_hdr) + offsetof(struct udphdr, dest);
    struct sock_filter filter_code[] = {
            BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, SKF_AD_OFF + SKF_AD_PKTTYPE),
            BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, PACKET_HOST, 0, 4),
            BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, SKF_NET_OFF + offset_proto),
            BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 2),
            BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, SKF_NET_OFF + offset_dport),

This is unexpected behavior. Socket filter programs should be consistent regardless of environment. Silent misses are particularly concerning as hard to detect.

Use skb_copy_bits for offsets outside linear, same as done for non-SKF_(LL|NET) offsets.

Offset is always positive after subtracting the reference threshold SKB_(LL|NET)_OFF, so is always >= skb_(mac|network)_offset. The sum of the two is an offset against skb->data, and may be negative, but it cannot point before skb->head, as skb_(mac|network)_offset would too.

This appears to go back to when frag support was introduced to sk_run_filter in linux-2.4.4, before the introduction of git.

The amount of code change and 8/16/32 bit duplication are unfortunate. But any attempt I made to be smarter saved very few LoC while complicating the code.

Quoted 7 lines

---

v2->v3
- do not remove bpf_internal_load_pointer_neg_helper, because it is still used in the sparc32 JIT
v1->v2
- introduce bfp_skb_load_helper_convert_offset to avoid open coding

#3 lore raw

PATCH v3 2/2 selftests/net: test sk_filter support for SKF_NET_OFF on frags

In reply to Willem de Bruijn

From: Willem de Bruijn <[email protected]>

Verify that a classic BPF linux socket filter correctly matches packet contents. Including when accessing contents in an skb_frag.

1. Open a SOCK_RAW socket with a classic BPF filter on UDP dport 8000.
2. Open a tap device with IFF_NAPI_FRAGS to inject skbs with frags.
3. Send a packet for which the UDP header is in frag[0].
4. Receive this packet to demonstrate that the socket accepted it.

---

v1->v2
- add comment why early demux must be disabled

---
 tools/testing/selftests/net/.gitignore     |   1 +
 tools/testing/selftests/net/Makefile       |   2 +
 tools/testing/selftests/net/skf_net_off.c  | 244 +++++++++++++++++++++
 tools/testing/selftests/net/skf_net_off.sh |  30 +++
 4 files changed, 277 insertions(+)
 create mode 100644 tools/testing/selftests/net/skf_net_off.c
 create mode 100755 tools/testing/selftests/net/skf_net_off.sh
tools/testing/selftests/net/.gitignore +1 -0
index 679542f565a4..532bb732bc6d 100644
@@ -39,6 +39,7 @@ scm_rights
 sk_bind_sendto_listen
 sk_connect_zero_addr
 sk_so_peek_off
+skf_net_off
 socket
 so_incoming_cpu
 so_netns_cookie
tools/testing/selftests/net/Makefile +2 -0
index 6d718b478ed8..124078b56fa4 100644
@@ -106,6 +106,8 @@ TEST_PROGS += ipv6_route_update_soft_lockup.sh
 TEST_PROGS += busy_poll_test.sh
 TEST_GEN_PROGS += proc_net_pktgen
 TEST_PROGS += lwt_dst_cache_ref_loop.sh
+TEST_PROGS += skf_net_off.sh
+TEST_GEN_FILES += skf_net_off
 
 # YNL files, must be before "include ..lib.mk"
 YNL_GEN_FILES := busy_poller netlink-dumps
tools/testing/selftests/net/skf_net_off.c new file +244 -0

244 lines not on this page · raw message

tools/testing/selftests/net/skf_net_off.sh new file +30 -0
new file mode 100755
index 000000000000..5da5066fb465
@@ -0,0 +1,30 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+
+readonly NS="ns-$(mktemp -u XXXXXX)"
+
+cleanup() {
+	ip netns del $NS
+}
+
+ip netns add $NS
+trap cleanup EXIT
+
+ip -netns $NS link set lo up
+ip -netns $NS tuntap add name tap1 mode tap
+ip -netns $NS link set tap1 up
+ip -netns $NS link set dev tap1 addr 02:00:00:00:00:01
+ip -netns $NS -6 addr add fdab::1 peer fdab::2 dev tap1 nodad
+ip netns exec $NS ethtool -K tap1 gro off
+
+# disable early demux, else udp_v6_early_demux pulls udp header into linear
+ip netns exec $NS sysctl -w net.ipv4.ip_early_demux=0
+
+echo "no filter"
+ip netns exec $NS ./skf_net_off -i tap1
+
+echo "filter, linear skb (-f)"
+ip netns exec $NS ./skf_net_off -i tap1 -f
+
+echo "filter, fragmented skb (-f) (-F)"
+ip netns exec $NS ./skf_net_off -i tap1 -f -F
Signature · 1 line
2.49.0.504.g3bcea36a83-goog
#4 lore raw
In reply to Willem de Bruijn
This series was applied to bpf/bpf.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This series was applied to bpf/bpf.git (master)
by Alexei Starovoitov <[email protected]>:

Quoted 10 lines · On Tue, 8 Apr 2025 09:27:47 -0400 you wrote:

From: Willem de Bruijn <[email protected]>

Address a longstanding issue that may lead to missed packets depending on system configuration.

Ensure that reading from packet contents works regardless of skb geometry, also when using the special SKF_.. negative offsets to offset from L2 or L3 header.

[...]

Here is the summary with links:
- [bpf,v3,1/2] bpf: support SKF_NET_OFF and SKF_LL_OFF on skb frags

    https://git.kernel.org/bpf/bpf/c/d4bac0288a2b
  - [bpf,v3,2/2] selftests/net: test sk_filter support for SKF_NET_OFF on frags
    https://git.kernel.org/bpf/bpf/c/fcd7132cb1f9

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html