bpf: Fix FIONREAD and copied_seq issues

A socket using sockmap has its own independent receive queue: ingress_msg. This queue may contain data from its own protocol stack or from other sockets.

The issue is that when reading from ingress_msg, we update tp->copied_seq by default. However, if the data is not from its own protocol stack, tcp->rcv_nxt is not increased. Later, if we convert this socket to a native socket, reading from this socket may fail because copied_seq might be significantly larger than rcv_nxt.

This fix also addresses the syzkaller-reported bug referenced in the Closes tag.

This patch marks the skmsg objects in ingress_msg. When reading, we update copied_seq only if the data is from its own protocol stack.

                                                     FD1:read()
                                                     --  FD1->copied_seq++
                                                         |  [read data]
                                                         |
                                [enqueue data]           v
                  [sockmap]     -> ingress to self ->  ingress_msg queue
FD1 native stack  ------>                                 ^
-- FD1->rcv_nxt++               -> redirect to other      | [enqueue data]
                                       |                  |

| ingress to FD1
v ^
... | [sockmap]
FD2 native stack

Closes https://syzkaller.appspot.com/bug?extid=06dbd397158ec0ea4983

Fixes 04919bed948dc ("tcp: Introduce tcp_read_skb()")

Reviewed-by Jakub Sitnicki <[email protected]>

Signed-off-by Jiayuan Chen <[email protected]>

---
 include/linux/skmsg.h |  2 ++
 net/core/skmsg.c      | 28 +++++++++++++++++++++++++---
 net/ipv4/tcp_bpf.c    |  5 +++--
 3 files changed, 30 insertions(+), 5 deletions(-)

include/linux/skmsg.h +2 -0

index 49847888c287..dfdc158ab88c 100644
@@ -141,6 +141,8 @@ int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 			     struct sk_msg *msg, u32 bytes);
 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
 		   int len, int flags);
+int __sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
+		     int len, int flags, int *copied_from_self);
 bool sk_msg_is_readable(struct sock *sk);
 
 static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)

net/core/skmsg.c +10 -3

index 2ac7731e1e0a..ca22ecdbf192 100644
@@ -409,22 +409,26 @@ int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 }
 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
 
-/* Receive sk_msg from psock->ingress_msg to @msg. */
-int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
-		   int len, int flags)
+int __sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
+		     int len, int flags, int *copied_from_self)
 {
 	struct iov_iter *iter = &msg->msg_iter;
 	int peek = flags & MSG_PEEK;
 	struct sk_msg *msg_rx;
 	int i, copied = 0;
+	bool from_self;
 
 	msg_rx = sk_psock_peek_msg(psock);
+	if (copied_from_self)
+		*copied_from_self = 0;
+
 	while (copied != len) {
 		struct scatterlist *sge;
 
 		if (unlikely(!msg_rx))
 			break;
 
+		from_self = msg_rx->sk == sk;
 		i = msg_rx->sg.start;
 		do {
 			struct page *page;

@@ -443,6 +447,9 @@ int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
 			}
 
 			copied += copy;
+			if (from_self && copied_from_self)
+				*copied_from_self += copy;
+
 			if (likely(!peek)) {
 				sge->offset += copy;
 				sge->length -= copy;

net/core/skmsg.c +14 -0

+
+/* Receive sk_msg from psock->ingress_msg to @msg. */
+int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
+		   int len, int flags)
+{
+	return __sk_msg_recvmsg(sk, psock, msg, len, flags, NULL);
+}
 EXPORT_SYMBOL_GPL(sk_msg_recvmsg);
 
 bool sk_msg_is_readable(struct sock *sk)

@@ -616,6 +631,12 @@ static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb
 	if (unlikely(!msg))
 		return -EAGAIN;
 	skb_set_owner_r(skb, sk);
+
+	/* This is used in tcp_bpf_recvmsg_parser() to determine whether the
+	 * data originates from the socket's own protocol stack. No need to
+	 * refcount sk because msg's lifetime is bound to sk via the ingress_msg.
+	 */
+	msg->sk = sk;
 	err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg, take_ref);
 	if (err < 0)
 		kfree(msg);

@@ -909,6 +930,7 @@ int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
 	sk_msg_compute_data_pointers(msg);
 	msg->sk = sk;
 	ret = bpf_prog_run_pin_on_cpu(prog, msg);
+	msg->sk = NULL;
 	ret = sk_psock_map_verd(ret, msg->sk_redir);
 	psock->apply_bytes = msg->apply_bytes;
 	if (ret == __SK_REDIRECT) {

net/ipv4/tcp_bpf.c +3 -2

index a268e1595b22..5c698fd7fbf8 100644
@@ -226,6 +226,7 @@ static int tcp_bpf_recvmsg_parser(struct sock *sk,
 	int peek = flags & MSG_PEEK;
 	struct sk_psock *psock;
 	struct tcp_sock *tcp;
+	int copied_from_self = 0;
 	int copied = 0;
 	u32 seq;
 

@@ -262,7 +263,7 @@ static int tcp_bpf_recvmsg_parser(struct sock *sk,
 	}
 
 msg_bytes_ready:
-	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
+	copied = __sk_msg_recvmsg(sk, psock, msg, len, flags, &copied_from_self);
 	/* The typical case for EFAULT is the socket was gracefully
 	 * shutdown with a FIN pkt. So check here the other case is
 	 * some error on copy_page_to_iter which would be unexpected.

@@ -277,7 +278,7 @@ static int tcp_bpf_recvmsg_parser(struct sock *sk,
 			goto out;
 		}
 	}
-	seq += copied;
+	seq += copied_from_self;
 	if (!copied) {
 		long timeo;
 		int data;

Quoted 34 lines · Jiayuan Chen

A socket using sockmap has its own independent receive queue: ingress_msg. This queue may contain data from its own protocol stack or from other sockets.

The issue is that when reading from ingress_msg, we update tp->copied_seq by default. However, if the data is not from its own protocol stack, tcp->rcv_nxt is not increased. Later, if we convert this socket to a native socket, reading from this socket may fail because copied_seq might be significantly larger than rcv_nxt.

This fix also addresses the syzkaller-reported bug referenced in the Closes tag.

This patch marks the skmsg objects in ingress_msg. When reading, we update copied_seq only if the data is from its own protocol stack.

                                                     FD1:read()
                                                     --  FD1->copied_seq++
                                                         |  [read data]
                                                         |
                                [enqueue data]           v
                  [sockmap]     -> ingress to self ->  ingress_msg queue
FD1 native stack  ------>                                 ^
-- FD1->rcv_nxt++               -> redirect to other      | [enqueue data]
                                       |                  |

| ingress to FD1
v ^
... | [sockmap]
FD2 native stack

Closes https://syzkaller.appspot.com/bug?extid=06dbd397158ec0ea4983

Fixes 04919bed948dc ("tcp: Introduce tcp_read_skb()")

Reviewed-by Jakub Sitnicki <[email protected]>

Signed-off-by Jiayuan Chen <[email protected]>

net/core/skmsg.c int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg, · line 497

Quoted 5 lines

net/core/skmsg.c +1 -0

@@ -487,6 +494,14 @@ int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
 out:
 	return copied;
 }
+EXPORT_SYMBOL_GPL(__sk_msg_recvmsg);

Nit: Sorry, I haven't caught that before. tcp_bpf is a built-in. We don't need to export this internal helper to modules.

We could probably push this without the 2/3 patch? If we are debating that patch still would be good to get this merged.

A socket using sockmap has its own independent receive queue: ingress_msg. This queue may contain data from its own protocol stack or from other sockets.

Therefore, for sockmap, relying solely on copied_seq and rcv_nxt to calculate FIONREAD is not enough.

This patch adds a new msg_tot_len field in the psock structure to record the data length in ingress_msg. Additionally, we implement new ioctl interfaces for TCP and UDP to intercept FIONREAD operations.

Unix and VSOCK sockets have similar issues, but fixing them is outside the scope of this patch as it would require more intrusive changes.

Previous work by John Fastabend made some efforts towards FIONREAD support:
commit e5c6de5fa025 ("bpf, sockmap: Incorrectly handling copied_seq") Although the current patch is based on the previous work by John Fastabend, it is acceptable for our Fixes tag to point to the same commit.

                                                      FD1:read()
                                                      --  FD1->copied_seq++
                                                          |  [read data]
                                                          |
                                   [enqueue data]         v
                  [sockmap]     -> ingress to self ->  ingress_msg queue
FD1 native stack  ------>                                 ^
-- FD1->rcv_nxt++               -> redirect to other      | [enqueue data]
                                       |                  |

| ingress to FD1
v ^
... | [sockmap]
FD2 native stack

Fixes 04919bed948dc ("tcp: Introduce tcp_read_skb()")

Signed-off-by Jiayuan Chen <[email protected]>

---
 include/linux/skmsg.h | 68 +++++++++++++++++++++++++++++++++++++++++--
 net/core/skmsg.c      |  3 ++
 net/ipv4/tcp_bpf.c    | 32 ++++++++++++++++++++
 net/ipv4/udp_bpf.c    | 23 ++++++++++++---
 4 files changed, 120 insertions(+), 6 deletions(-)

include/linux/skmsg.h +66 -2

index dfdc158ab88c..829b281d6c9c 100644
@@ -97,6 +97,8 @@ struct sk_psock {
 	struct sk_buff_head		ingress_skb;
 	struct list_head		ingress_msg;
 	spinlock_t			ingress_lock;
+	/** @msg_tot_len: Total bytes queued in ingress_msg list. */
+	u32				msg_tot_len;
 	unsigned long			state;
 	struct list_head		link;
 	spinlock_t			link_lock;

@@ -321,6 +323,27 @@ static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
 	kfree_skb(skb);
 }
 
+static inline u32 sk_psock_get_msg_len_nolock(struct sk_psock *psock)
+{
+	/* Used by ioctl to read msg_tot_len only; lock-free for performance */
+	return READ_ONCE(psock->msg_tot_len);
+}
+
+static inline void sk_psock_msg_len_add_locked(struct sk_psock *psock, int diff)
+{
+	/* Use WRITE_ONCE to ensure correct read in sk_psock_get_msg_len_nolock().
+	 * ingress_lock should be held to prevent concurrent updates to msg_tot_len
+	 */
+	WRITE_ONCE(psock->msg_tot_len, psock->msg_tot_len + diff);
+}
+
+static inline void sk_psock_msg_len_add(struct sk_psock *psock, int diff)
+{
+	spin_lock_bh(&psock->ingress_lock);
+	sk_psock_msg_len_add_locked(psock, diff);
+	spin_unlock_bh(&psock->ingress_lock);
+}
+
 static inline bool sk_psock_queue_msg(struct sk_psock *psock,
 				      struct sk_msg *msg)
 {

@@ -329,6 +352,7 @@ static inline bool sk_psock_queue_msg(struct sk_psock *psock,
 	spin_lock_bh(&psock->ingress_lock);
 	if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
 		list_add_tail(&msg->list, &psock->ingress_msg);
+		sk_psock_msg_len_add_locked(psock, msg->sg.size);
 		ret = true;
 	} else {
 		sk_msg_free(psock->sk, msg);

@@ -345,18 +369,25 @@ static inline struct sk_msg *sk_psock_dequeue_msg(struct sk_psock *psock)
 
 	spin_lock_bh(&psock->ingress_lock);
 	msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
-	if (msg)
+	if (msg) {
 		list_del(&msg->list);
+		sk_psock_msg_len_add_locked(psock, -msg->sg.size);
+	}
 	spin_unlock_bh(&psock->ingress_lock);
 	return msg;
 }
 
+static inline struct sk_msg *sk_psock_peek_msg_locked(struct sk_psock *psock)
+{
+	return list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
+}
+
 static inline struct sk_msg *sk_psock_peek_msg(struct sk_psock *psock)
 {
 	struct sk_msg *msg;
 
 	spin_lock_bh(&psock->ingress_lock);
-	msg = list_first_entry_or_null(&psock->ingress_msg, struct sk_msg, list);
+	msg = sk_psock_peek_msg_locked(psock);
 	spin_unlock_bh(&psock->ingress_lock);
 	return msg;
 }

@@ -523,6 +554,39 @@ static inline bool sk_psock_strp_enabled(struct sk_psock *psock)
 	return !!psock->saved_data_ready;
 }
 
+/* for tcp only, sk is locked */
+static inline ssize_t sk_psock_msg_inq(struct sock *sk)
+{
+	struct sk_psock *psock;
+	ssize_t inq = 0;
+
+	psock = sk_psock_get(sk);
+	if (likely(psock)) {
+		inq = sk_psock_get_msg_len_nolock(psock);
+		sk_psock_put(sk, psock);
+	}
+	return inq;
+}
+
+/* for udp only, sk is not locked */
+static inline ssize_t sk_msg_first_len(struct sock *sk)
+{
+	struct sk_psock *psock;
+	struct sk_msg *msg;
+	ssize_t inq = 0;
+
+	psock = sk_psock_get(sk);
+	if (likely(psock)) {
+		spin_lock_bh(&psock->ingress_lock);
+		msg = sk_psock_peek_msg_locked(psock);
+		if (msg)
+			inq = msg->sg.size;
+		spin_unlock_bh(&psock->ingress_lock);
+		sk_psock_put(sk, psock);
+	}
+	return inq;
+}
+
 #if IS_ENABLED(CONFIG_NET_SOCK_MSG)
 
 #define BPF_F_STRPARSER	(1UL << 1)

net/core/skmsg.c +3 -0

index ca22ecdbf192..dd1965bf8f6a 100644
@@ -458,6 +458,7 @@ int __sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg
 					atomic_sub(copy, &sk->sk_rmem_alloc);
 				}
 				msg_rx->sg.size -= copy;
+				sk_psock_msg_len_add(psock, -copy);
 
 				if (!sge->length) {
 					sk_msg_iter_var_next(i);

@@ -822,9 +823,11 @@ static void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
 		list_del(&msg->list);
 		if (!msg->skb)
 			atomic_sub(msg->sg.size, &psock->sk->sk_rmem_alloc);
+		sk_psock_msg_len_add(psock, -msg->sg.size);
 		sk_msg_free(psock->sk, msg);
 		kfree(msg);
 	}
+	WARN_ON_ONCE(psock->msg_tot_len);
 }
 
 static void __sk_psock_zap_ingress(struct sk_psock *psock)

net/ipv4/tcp_bpf.c +9 -0

index 5c698fd7fbf8..476101533349 100644
@@ -10,6 +10,7 @@
 
 #include <net/inet_common.h>
 #include <net/tls.h>
+#include <asm/ioctls.h>
 
 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
 {

@@ -332,6 +333,36 @@ static int tcp_bpf_recvmsg_parser(struct sock *sk,
 	return copied;
 }
 
+static int tcp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
+{
+	bool slow;
+
+	if (cmd != SIOCINQ)
+		return tcp_ioctl(sk, cmd, karg);
+
+	/* works similar as tcp_ioctl */

net/ipv4/tcp_bpf.c +13 -0

+	if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) {
+		unlock_sock_fast(sk, slow);
+		return tcp_ioctl(sk, cmd, karg);
+	}
+
+	*karg = sk_psock_msg_inq(sk);
+
+	unlock_sock_fast(sk, slow);
+
+	return 0;
+}
+
 static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 			   int flags, int *addr_len)
 {

@@ -610,6 +641,7 @@ static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
 	prot[TCP_BPF_BASE].close		= sock_map_close;
 	prot[TCP_BPF_BASE].recvmsg		= tcp_bpf_recvmsg;
 	prot[TCP_BPF_BASE].sock_is_readable	= sk_msg_is_readable;
+	prot[TCP_BPF_BASE].ioctl		= tcp_bpf_ioctl;
 
 	prot[TCP_BPF_TX]			= prot[TCP_BPF_BASE];
 	prot[TCP_BPF_TX].sendmsg		= tcp_bpf_sendmsg;

net/ipv4/udp_bpf.c +1 -0

index 0735d820e413..91233e37cd97 100644
@@ -5,6 +5,7 @@
 #include <net/sock.h>
 #include <net/udp.h>
 #include <net/inet_common.h>
+#include <asm/ioctls.h>
 
 #include "udp_impl.h"
 

@@ -111,12 +112,26 @@ enum {
 static DEFINE_SPINLOCK(udpv6_prot_lock);
 static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];

net/ipv4/udp_bpf.c +9 -4

+	*karg = sk_msg_first_len(sk);
+	return 0;
+}
+
 static void udp_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
 {
-	*prot        = *base;
-	prot->close  = sock_map_close;
-	prot->recvmsg = udp_bpf_recvmsg;
-	prot->sock_is_readable = sk_msg_is_readable;
+	*prot			= *base;
+	prot->close		= sock_map_close;
+	prot->recvmsg		= udp_bpf_recvmsg;
+	prot->sock_is_readable	= sk_msg_is_readable;
+	prot->ioctl		= udp_bpf_ioctl;
 }
 
 static void udp_bpf_check_v6_needs_rebuild(struct proto *ops)

On Tue, Jan 13, 2026 at 10:50 AM +08, Jiayuan Chen wrote:

Quoted 15 lines

net/ipv4/udp_bpf.c +1 -0

index 0735d820e413..91233e37cd97 100644
@@ -5,6 +5,7 @@
 #include <net/sock.h>
 #include <net/udp.h>
 #include <net/inet_common.h>
+#include <asm/ioctls.h>
 
 #include "udp_impl.h"
 

@@ -111,12 +112,26 @@ enum {
 static DEFINE_SPINLOCK(udpv6_prot_lock);
 static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];

net/ipv4/udp_bpf.c enum { · line 123

Quoted 10 lines

net/ipv4/udp_bpf.c +9 -0

 
+static int udp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
+{
+	if (cmd != SIOCINQ)
+		return udp_ioctl(sk, cmd, karg);
+
+	/* Since we don't hold a lock, sk_receive_queue may contain data.
+	 * BPF might only be processing this data at the moment. We only
+	 * care about the data in the ingress_msg here.
+	 */

I think we should strive for a design where FIONREAD does not go down after you add your socket to sockmap, if there was no recvmsg call in between. To show what I mean, I added this test that's currently failing for udp:

---8<---

tools/testing/selftests/bpf/prog_tests/sockmap_basic.c +64 -1

index 1f1289f5a8c2..123c96fcaef0 100644
@@ -1229,6 +1229,66 @@ static void test_sockmap_copied_seq(bool strp)
 	test_sockmap_pass_prog__destroy(skel);
 }
 
+/* Test FIONREAD when data exists in sk_receive_queue before sockmap insertion */
+static void test_sockmap_fionread_pre_insert(int sotype)
+{
+	int map, err, sent, recvd, zero = 0, avail = 0;
+	struct test_sockmap_pass_prog *skel = NULL;
+	int c = -1, p = -1;
+	char buf[10] = "0123456789", rcv[11];
+	struct bpf_program *prog;
+
+	skel = test_sockmap_pass_prog__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	prog = skel->progs.prog_skb_verdict;
+	map = bpf_map__fd(skel->maps.sock_map_rx);
+
+	err = bpf_prog_attach(bpf_program__fd(prog), map, BPF_SK_SKB_STREAM_VERDICT, 0);
+	if (!ASSERT_OK(err, "bpf_prog_attach verdict"))
+		goto end;
+
+	err = create_pair(AF_INET, sotype, &c, &p);
+	if (!ASSERT_OK(err, "create_pair"))
+		goto end;
+
+	/* Step 1: Send data BEFORE sockmap insertion - lands in sk_receive_queue */
+	sent = xsend(p, buf, sizeof(buf), 0);
+	if (!ASSERT_EQ(sent, sizeof(buf), "xsend pre-insert"))
+		goto end;
+
+	/* Step 2: Verify FIONREAD reports data in sk_receive_queue */
+	err = poll_read(c, IO_TIMEOUT_SEC);
+	if (!ASSERT_OK(err, "poll_read pre-insert"))
+		goto end;
+	err = ioctl(c, FIONREAD, &avail);
+	ASSERT_OK(err, "ioctl(FIONREAD) pre-insert error");
+	ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) pre-insert");
+
+	/* Step 3: Insert socket into sockmap */
+	err = bpf_map_update_elem(map, &zero, &c, BPF_ANY);
+	if (!ASSERT_OK(err, "bpf_map_update_elem(c)"))
+		goto end;
+
+	/* Step 4: FIONREAD should still report the data in sk_receive_queue */
+	err = ioctl(c, FIONREAD, &avail);
+	ASSERT_OK(err, "ioctl(FIONREAD) post-insert error");
+	ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) post-insert");
+
+	/* Verify we can still read the data */
+	recvd = recv_timeout(c, rcv, sizeof(rcv), MSG_DONTWAIT, IO_TIMEOUT_SEC);
+	ASSERT_EQ(recvd, sizeof(buf), "recv post-insert");
+	ASSERT_OK(memcmp(buf, rcv, recvd), "data mismatch");
+
+end:
+	if (c >= 0)
+		close(c);
+	if (p >= 0)
+		close(p);
+	test_sockmap_pass_prog__destroy(skel);
+}
+
 /* it is used to send data to via native stack and BPF redirecting */
 static void test_sockmap_multi_channels(int sotype)
 {

@@ -1373,4 +1433,8 @@ void test_sockmap_basic(void)
 		test_sockmap_multi_channels(SOCK_STREAM);
 	if (test__start_subtest("sockmap udp multi channels"))
 		test_sockmap_multi_channels(SOCK_DGRAM);
+	if (test__start_subtest("sockmap tcp fionread pre-insert"))
+		test_sockmap_fionread_pre_insert(SOCK_STREAM);
+	if (test__start_subtest("sockmap udp fionread pre-insert"))
+		test_sockmap_fionread_pre_insert(SOCK_DGRAM);
 }
--->8---

net/ipv4/udp_bpf.c enum { · line 127

Quoted 4 lines

net/ipv4/udp_bpf.c +4 -0

+	*karg = sk_msg_first_len(sk);
+	return 0;
+}
+

Quoted 3 lines · Jakub Sitnicki

On Tue, Jan 13, 2026 at 10:50 AM +08, Jiayuan Chen wrote:

[...]

Quoted 107 lines

Quoted 18 lines

net/ipv4/udp_bpf.c +0 -0

 index 0735d820e413..91233e37cd97 100644
 --- a/net/ipv4/udp_bpf.c
 +++ b/net/ipv4/udp_bpf.c
 @@ -5,6 +5,7 @@
 #include <net/sock.h>
 #include <net/udp.h>
 #include <net/inet_common.h>
 +#include <asm/ioctls.h>
 
 #include "udp_impl.h"
 
 @@ -111,12 +112,26 @@ enum {
 static DEFINE_SPINLOCK(udpv6_prot_lock);
 static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];

net/ipv4/udp_bpf.c

Quoted 10 lines

net/ipv4/udp_bpf.c +0 -0

 
 +static int udp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
 +{
 + if (cmd != SIOCINQ)
 + return udp_ioctl(sk, cmd, karg);
 +
 + /* Since we don't hold a lock, sk_receive_queue may contain data.
 + * BPF might only be processing this data at the moment. We only
 + * care about the data in the ingress_msg here.
 + */

I think we should strive for a design where FIONREAD does not go down after you add your socket to sockmap, if there was no recvmsg call in between. To show what I mean, I added this test that's currently failing for udp:

---8<---
diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c b/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c index 1f1289f5a8c2..123c96fcaef0 100644
--- a/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c
+++ b/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c
@@ -1229,6 +1229,66 @@ static void test_sockmap_copied_seq(bool strp) test_sockmap_pass_prog__destroy(skel);
}

+/* Test FIONREAD when data exists in sk_receive_queue before sockmap insertion */ +static void test_sockmap_fionread_pre_insert(int sotype)
+{
+ int map, err, sent, recvd, zero = 0, avail = 0;
+ struct test_sockmap_pass_prog *skel = NULL;
+ int c = -1, p = -1;
+ char buf[10] = "0123456789", rcv[11];
+ struct bpf_program *prog;
+
+ skel = test_sockmap_pass_prog__open_and_load();
+ if (!ASSERT_OK_PTR(skel, "open_and_load"))
+ return;
+
+ prog = skel->progs.prog_skb_verdict;
+ map = bpf_map__fd(skel->maps.sock_map_rx);
+
+ err = bpf_prog_attach(bpf_program__fd(prog), map, BPF_SK_SKB_STREAM_VERDICT, 0);
+ if (!ASSERT_OK(err, "bpf_prog_attach verdict"))
+ goto end;
+
+ err = create_pair(AF_INET, sotype, &c, &p);
+ if (!ASSERT_OK(err, "create_pair"))
+ goto end;
+
+ /* Step 1: Send data BEFORE sockmap insertion - lands in sk_receive_queue */
+ sent = xsend(p, buf, sizeof(buf), 0);
+ if (!ASSERT_EQ(sent, sizeof(buf), "xsend pre-insert"))
+ goto end;
+
+ /* Step 2: Verify FIONREAD reports data in sk_receive_queue */
+ err = poll_read(c, IO_TIMEOUT_SEC);
+ if (!ASSERT_OK(err, "poll_read pre-insert"))
+ goto end;
+ err = ioctl(c, FIONREAD, &avail);
+ ASSERT_OK(err, "ioctl(FIONREAD) pre-insert error");
+ ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) pre-insert");
+
+ /* Step 3: Insert socket into sockmap */
+ err = bpf_map_update_elem(map, &zero, &c, BPF_ANY);
+ if (!ASSERT_OK(err, "bpf_map_update_elem(c)"))
+ goto end;
+
+ /* Step 4: FIONREAD should still report the data in sk_receive_queue */
+ err = ioctl(c, FIONREAD, &avail);
+ ASSERT_OK(err, "ioctl(FIONREAD) post-insert error");
+ ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) post-insert");
+
+ /* Verify we can still read the data */
+ recvd = recv_timeout(c, rcv, sizeof(rcv), MSG_DONTWAIT, IO_TIMEOUT_SEC);
+ ASSERT_EQ(recvd, sizeof(buf), "recv post-insert");
+ ASSERT_OK(memcmp(buf, rcv, recvd), "data mismatch");
+
+end:
+ if (c >= 0)
+ close(c);
+ if (p >= 0)
+ close(p);
+ test_sockmap_pass_prog__destroy(skel);
+}
+
/* it is used to send data to via native stack and BPF redirecting */ static void test_sockmap_multi_channels(int sotype)
{
@@ -1373,4 +1433,8 @@ void test_sockmap_basic(void) test_sockmap_multi_channels(SOCK_STREAM);

Quoted 16 lines

if (test__start_subtest("sockmap udp multi channels")) test_sockmap_multi_channels(SOCK_DGRAM);
+ if (test__start_subtest("sockmap tcp fionread pre-insert"))
+ test_sockmap_fionread_pre_insert(SOCK_STREAM);
+ if (test__start_subtest("sockmap udp fionread pre-insert"))
+ test_sockmap_fionread_pre_insert(SOCK_DGRAM);
}
--->8---

Quoted 6 lines

+ *karg = sk_msg_first_len(sk);
+ return 0;
+}
+

I've been thinking about this some more and came to the conclusion that this udp_bpf_ioctl implementation is actually what we want, while tcp_bpf_ioctl *should not* be checking if the sk_receive_queue is non-empty.

Why? Because the verdict prog might redirect or drop the skbs from sk_receive_queue once it actually runs. The messages might never appear on the msg_ingress queue.

What I think we should be doing, in the end, is kicking the sk_receive_queue processing on bpf_map_update_elem, if there's data ready.

The API semantics I'm proposing is:

1. ioctl(FIONREAD) -> reports N bytes
2. bpf_map_update_elem(sk) -> socket inserted into sockmap
3. poll() for POLLIN -> wait for socket to be ready to read
5. ioctl(FIONREAD) -> report N bytes if verdict prog didn't redirect or drop it

We don't have to add the the queue kick on map update in this series.

If you decide to leave it for later, can I ask that you open an issue at our GH project [1]?

I don't want it to fall through the cracks. And I sometimes have people asking what they could help with in sockmap.

Thanks,
-jkbs

[1] https://github.com/sockmap-project/sockmap-project/issues

Quoted 157 lines · Jakub Sitnicki

On Tue, Jan 20, 2026 at 04:00 PM +01, Jakub Sitnicki wrote:

On Tue, Jan 13, 2026 at 10:50 AM +08, Jiayuan Chen wrote:

[...]

Quoted 17 lines

net/ipv4/udp_bpf.c +1 -0

index 0735d820e413..91233e37cd97 100644
@@ -5,6 +5,7 @@
 #include <net/sock.h>
 #include <net/udp.h>
 #include <net/inet_common.h>
+#include <asm/ioctls.h>
 
 #include "udp_impl.h"
 

@@ -111,12 +112,26 @@ enum {
 static DEFINE_SPINLOCK(udpv6_prot_lock);
 static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];
 

net/ipv4/udp_bpf.c enum { · line 124

Quoted 10 lines

net/ipv4/udp_bpf.c +9 -0

+static int udp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
+{
+ if (cmd != SIOCINQ)
+ return udp_ioctl(sk, cmd, karg);
+
+ /* Since we don't hold a lock, sk_receive_queue may contain data.
+ * BPF might only be processing this data at the moment. We only
+ * care about the data in the ingress_msg here.
+ */
 

I think we should strive for a design where FIONREAD does not go down after you add your socket to sockmap, if there was no recvmsg call in between. To show what I mean, I added this test that's currently failing for udp:

---8<---
diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c b/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c index 1f1289f5a8c2..123c96fcaef0 100644
--- a/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c
+++ b/tools/testing/selftests/bpf/prog_tests/sockmap_basic.c
@@ -1229,6 +1229,66 @@ static void test_sockmap_copied_seq(bool strp) test_sockmap_pass_prog__destroy(skel);
}

+/* Test FIONREAD when data exists in sk_receive_queue before sockmap insertion */ +static void test_sockmap_fionread_pre_insert(int sotype)
+{
+ int map, err, sent, recvd, zero = 0, avail = 0;
+ struct test_sockmap_pass_prog *skel = NULL;
+ int c = -1, p = -1;
+ char buf[10] = "0123456789", rcv[11];
+ struct bpf_program *prog;
+
+ skel = test_sockmap_pass_prog__open_and_load();
+ if (!ASSERT_OK_PTR(skel, "open_and_load"))
+ return;
+
+ prog = skel->progs.prog_skb_verdict;
+ map = bpf_map__fd(skel->maps.sock_map_rx);
+
+ err = bpf_prog_attach(bpf_program__fd(prog), map, BPF_SK_SKB_STREAM_VERDICT, 0);
+ if (!ASSERT_OK(err, "bpf_prog_attach verdict"))
+ goto end;
+
+ err = create_pair(AF_INET, sotype, &c, &p);
+ if (!ASSERT_OK(err, "create_pair"))
+ goto end;
+
+ /* Step 1: Send data BEFORE sockmap insertion - lands in sk_receive_queue */
+ sent = xsend(p, buf, sizeof(buf), 0);
+ if (!ASSERT_EQ(sent, sizeof(buf), "xsend pre-insert"))
+ goto end;
+
+ /* Step 2: Verify FIONREAD reports data in sk_receive_queue */
+ err = poll_read(c, IO_TIMEOUT_SEC);
+ if (!ASSERT_OK(err, "poll_read pre-insert"))
+ goto end;
+ err = ioctl(c, FIONREAD, &avail);
+ ASSERT_OK(err, "ioctl(FIONREAD) pre-insert error");
+ ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) pre-insert");
+
+ /* Step 3: Insert socket into sockmap */
+ err = bpf_map_update_elem(map, &zero, &c, BPF_ANY);
+ if (!ASSERT_OK(err, "bpf_map_update_elem(c)"))
+ goto end;
+
+ /* Step 4: FIONREAD should still report the data in sk_receive_queue */
+ err = ioctl(c, FIONREAD, &avail);
+ ASSERT_OK(err, "ioctl(FIONREAD) post-insert error");
+ ASSERT_EQ(avail, sizeof(buf), "ioctl(FIONREAD) post-insert");
+
+ /* Verify we can still read the data */
+ recvd = recv_timeout(c, rcv, sizeof(rcv), MSG_DONTWAIT, IO_TIMEOUT_SEC);
+ ASSERT_EQ(recvd, sizeof(buf), "recv post-insert");
+ ASSERT_OK(memcmp(buf, rcv, recvd), "data mismatch");
+
+end:
+ if (c >= 0)
+ close(c);
+ if (p >= 0)
+ close(p);
+ test_sockmap_pass_prog__destroy(skel);
+}
+
/* it is used to send data to via native stack and BPF redirecting */ static void test_sockmap_multi_channels(int sotype)
{
@@ -1373,4 +1433,8 @@ void test_sockmap_basic(void) test_sockmap_multi_channels(SOCK_STREAM);
if (test__start_subtest("sockmap udp multi channels")) test_sockmap_multi_channels(SOCK_DGRAM);
+ if (test__start_subtest("sockmap tcp fionread pre-insert"))
+ test_sockmap_fionread_pre_insert(SOCK_STREAM);
+ if (test__start_subtest("sockmap udp fionread pre-insert"))
+ test_sockmap_fionread_pre_insert(SOCK_DGRAM);
}
--->8---

net/ipv4/udp_bpf.c enum { · line 129

Quoted 6 lines

net/ipv4/udp_bpf.c +4 -0

+ *karg = sk_msg_first_len(sk);
+ return 0;
+}
+
 

I've been thinking about this some more and came to the conclusion that this udp_bpf_ioctl implementation is actually what we want, while tcp_bpf_ioctl *should not* be checking if the sk_receive_queue is non-empty.

Why? Because the verdict prog might redirect or drop the skbs from sk_receive_queue once it actually runs. The messages might never appear on the msg_ingress queue.

What I think we should be doing, in the end, is kicking the sk_receive_queue processing on bpf_map_update_elem, if there's data ready.

The API semantics I'm proposing is:

1. ioctl(FIONREAD) -> reports N bytes
2. bpf_map_update_elem(sk) -> socket inserted into sockmap
3. poll() for POLLIN -> wait for socket to be ready to read
5. ioctl(FIONREAD) -> report N bytes if verdict prog didn't redirect or drop it

We don't have to add the the queue kick on map update in this series.

If you decide to leave it for later, can I ask that you open an issue at our GH project [1]?

Quoted 8 lines

I don't want it to fall through the cracks. And I sometimes have people asking what they could help with in sockmap.

Thanks,
-jkbs

[1] https://github.com/sockmap-project/sockmap-project/issues

Hi Jakub,

Thanks for taking the time to think through this carefully. I agree with your analysis - reporting sk_receive_queue length is misleading since the verdict prog might redirect or drop those skbs.

There's no rush to merge this patch.

Since the kick queue on bpf_map_update_elem addresses a closely related issue, I think it makes sense to include it in this patchset for easier tracking rather than splitting it out.

I'll spend more time looking into this and come back with an updated version.

Thanks,
Jiayuan

January 21, 2026 at 20:55, "Jiayuan Chen" <[email protected] mailto:[email protected]?to=%22Jiayuan%20Chen%22%20%3Cjiayuan.chen%40linux.dev%3E

Quoted 45 lines

wrote:
January 21, 2026 at 17:36, "Jakub Sitnicki" <[email protected] mailto:[email protected]?to=%22Jakub%20Sitnicki%22%20%3Cjakub%40cloudflare.com%3E

Quoted 26 lines

I've been thinking about this some more and came to the conclusion that this udp_bpf_ioctl implementation is actually what we want, while tcp_bpf_ioctl *should not* be checking if the sk_receive_queue is non-empty.

Why? Because the verdict prog might redirect or drop the skbs from sk_receive_queue once it actually runs. The messages might never appear on the msg_ingress queue.

What I think we should be doing, in the end, is kicking the sk_receive_queue processing on bpf_map_update_elem, if there's data ready.

The API semantics I'm proposing is:

1. ioctl(FIONREAD) -> reports N bytes
2. bpf_map_update_elem(sk) -> socket inserted into sockmap
3. poll() for POLLIN -> wait for socket to be ready to read
5. ioctl(FIONREAD) -> report N bytes if verdict prog didn't redirect or drop it

We don't have to add the the queue kick on map update in this series.

If you decide to leave it for later, can I ask that you open an issue at our GH project [1]?

Quoted 8 lines

I don't want it to fall through the cracks. And I sometimes have people asking what they could help with in sockmap.

Thanks,
-jkbs

[1] https://github.com/sockmap-project/sockmap-project/issues

Hi Jakub,

Thanks for taking the time to think through this carefully. I agree with your analysis - reporting sk_receive_queue length is misleading since the verdict prog might redirect or drop those skbs.

There's no rush to merge this patch.

Quoted 9 lines

Since the kick queue on bpf_map_update_elem addresses a closely related issue, I think it makes sense to include it in this patchset for easier tracking rather than splitting it out.

I'll spend more time looking into this and come back with an updated version.

Thanks,
Jiayuan

Hi Jakub,

I've been thinking about this more, and I realize the problem is not as simple as it seems.

Regarding kicking the sk_receive_queue on bpf_map_update_elem: the BPF program may not be fully initialized at that point. For example, with a redirect program, the destination fd might not yet be inserted into the map. If we kick the data through the BPF program immediately, the redirect lookup would fail, leading to unexpected behavior (data being dropped or passed to the wrong socket).

I also considered triggering the kick in poll/select via sk_msg_is_readable(). However, this approach doesn't work for TCP because tcp_poll() -> tcp_stream_is_readable() -> tcp_epollin_ready() will return early when sk_receive_queue has data, before ever calling sk_is_readable().

In the next version, I'll address your other nits and remove the sk_receive_queue check from tcp_bpf_ioctl. I'll also open an issue on the GH project to track this problem so we can continue exploring better solutions.

tools/testing/selftests/bpf/prog_tests/sockmap_basic.c +61 -0

+/* it is used to send data to via native stack and BPF redirecting */
+static void test_sockmap_multi_channels(int sotype)
+{
+	int map, err, sent, recvd, zero = 0, one = 1, avail = 0;
+	struct test_sockmap_pass_prog *skel = NULL;
+	int c0 = -1, p0 = -1, c1 = -1, p1 = -1;
+	char buf[10] = "0123456789", rcv[11];
+	struct bpf_program *prog;
+
+	skel = test_sockmap_pass_prog__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	err = create_socket_pairs(AF_INET, sotype, &c0, &c1, &p0, &p1);
+	if (err)
+		goto end;
+
+	prog = skel->progs.prog_skb_verdict_ingress;
+	map = bpf_map__fd(skel->maps.sock_map_rx);
+
+	err = bpf_prog_attach(bpf_program__fd(prog), map, BPF_SK_SKB_STREAM_VERDICT, 0);
+	if (!ASSERT_OK(err, "bpf_prog_attach verdict"))
+		goto end;
+
+	err = bpf_map_update_elem(map, &zero, &p0, BPF_ANY);
+	if (!ASSERT_OK(err, "bpf_map_update_elem(p0)"))
+		goto end;
+
+	err = bpf_map_update_elem(map, &one, &p1, BPF_ANY);
+	if (!ASSERT_OK(err, "bpf_map_update_elem"))
+		goto end;
+
+	/* send data to p1 via native stack */
+	sent = xsend(c1, buf, 2, 0);
+	if (!ASSERT_EQ(sent, 2, "xsend(2)"))
+		goto end;
+
+	sleep(1);
+	err = ioctl(p1, FIONREAD, &avail);
+	ASSERT_OK(err, "ioctl(FIONREAD) partial call");
+	ASSERT_EQ(avail, 2, "ioctl(FIONREAD) partial return");
+
+	/* send data to p1 via bpf redirecting */
+	sent = xsend(c0, buf + 2, sizeof(buf) - 2, 0);
+	if (!ASSERT_EQ(sent, sizeof(buf) - 2, "xsend(remain-data)"))
+		goto end;
+
+	sleep(1);
+	err = ioctl(p1, FIONREAD, &avail);
+	ASSERT_OK(err, "ioctl(FIONREAD) full call");
+	ASSERT_EQ(avail, sotype == SOCK_DGRAM ? 2 : sizeof(buf), "ioctl(FIONREAD) full return");
+
+	recvd = recv_timeout(p1, rcv, sizeof(rcv), MSG_DONTWAIT, 1);
+	if (!ASSERT_EQ(recvd, sizeof(buf), "recv_timeout(p1)") ||
+	    !ASSERT_OK(memcmp(buf, rcv, recvd), "data mismatch"))
+		goto end;
+end:
+	if (c0 >= 0)
+		close(c0);
+	if (p0 >= 0)
+		close(p0);