x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 387 insertions, 0 deletions

diff --git a/net/rxrpc/oob.c b/net/rxrpc/oob.c
new file mode 100644
index 000000000..c80ee2487
--- /dev/null
+++ b/net/rxrpc/oob.c
@@ -0,0 +1,387 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Out of band message handling (e.g. challenge-response)
+ *
+ * Copyright (C) 2025 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/net.h>
+#include <linux/gfp.h>
+#include <linux/skbuff.h>
+#include <linux/export.h>
+#include <linux/sched/signal.h>
+#include <net/sock.h>
+#include <net/af_rxrpc.h>
+#include "ar-internal.h"
+
+enum rxrpc_oob_command {
+	RXRPC_OOB_CMD_UNSET,
+	RXRPC_OOB_CMD_RESPOND,
+} __mode(byte);
+
+struct rxrpc_oob_params {
+	u64			oob_id;		/* ID number of message if reply */
+	s32			abort_code;
+	enum rxrpc_oob_command	command;
+	bool			have_oob_id:1;
+};
+
+/*
+ * Post an out-of-band message for attention by the socket or kernel service
+ * associated with a reference call.
+ */
+bool rxrpc_notify_socket_oob(struct rxrpc_call *call, struct sk_buff *skb)
+{
+	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
+	struct rxrpc_sock *rx;
+	struct sock *sk;
+	bool queued = false;
+
+	rcu_read_lock();
+
+	rx = rcu_dereference(call->socket);
+	if (rx) {
+		sk = &rx->sk;
+		spin_lock_irq(&rx->recvmsg_lock);
+
+		if (sk->sk_state < RXRPC_CLOSE) {
+			skb->skb_mstamp_ns = rx->oob_id_counter++;
+			rxrpc_get_skb(skb, rxrpc_skb_get_post_oob);
+			skb_queue_tail(&rx->recvmsg_oobq, skb);
+			queued = true;
+
+			trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
+			if (rx->app_ops)
+				rx->app_ops->notify_oob(sk, skb);
+		}
+
+		spin_unlock_irq(&rx->recvmsg_lock);
+		if (queued && !rx->app_ops && !sock_flag(sk, SOCK_DEAD))
+			sk->sk_data_ready(sk);
+	}
+
+	rcu_read_unlock();
+	return queued;
+}
+
+/*
+ * Locate the OOB message to respond to by its ID.
+ */
+static struct sk_buff *rxrpc_find_pending_oob(struct rxrpc_sock *rx, u64 oob_id)
+{
+	struct rb_node *p;
+	struct sk_buff *skb;
+
+	p = rx->pending_oobq.rb_node;
+	while (p) {
+		skb = rb_entry(p, struct sk_buff, rbnode);
+
+		if (oob_id < skb->skb_mstamp_ns)
+			p = p->rb_left;
+		else if (oob_id > skb->skb_mstamp_ns)
+			p = p->rb_right;
+		else
+			return skb;
+	}
+
+	return NULL;
+}
+
+/*
+ * Add an OOB message into the pending-response set.  We always assign the next
+ * value from a 64-bit counter to the oob_id, so just assume we're always going
+ * to be on the right-hand edge of the tree and that the counter won't wrap.
+ * The tree is also given a ref to the message.
+ */
+void rxrpc_add_pending_oob(struct rxrpc_sock *rx, struct sk_buff *skb)
+{
+	struct rb_node **pp = &rx->pending_oobq.rb_node, *p = NULL;
+
+	while (*pp) {
+		p = *pp;
+		pp = &(*pp)->rb_right;
+	}
+
+	rb_link_node(&skb->rbnode, p, pp);
+	rb_insert_color(&skb->rbnode, &rx->pending_oobq);
+}
+
+/*
+ * Extract control messages from the sendmsg() control buffer.
+ */
+static int rxrpc_sendmsg_oob_cmsg(struct msghdr *msg, struct rxrpc_oob_params *p)
+{
+	struct cmsghdr *cmsg;
+	int len;
+
+	if (msg->msg_controllen == 0)
+		return -EINVAL;
+
+	for_each_cmsghdr(cmsg, msg) {
+		if (!CMSG_OK(msg, cmsg))
+			return -EINVAL;
+
+		len = cmsg->cmsg_len - sizeof(struct cmsghdr);
+		_debug("CMSG %d, %d, %d",
+		       cmsg->cmsg_level, cmsg->cmsg_type, len);
+
+		if (cmsg->cmsg_level != SOL_RXRPC)
+			continue;
+
+		switch (cmsg->cmsg_type) {
+		case RXRPC_OOB_ID:
+			if (len != sizeof(p->oob_id) || p->have_oob_id)
+				return -EINVAL;
+			memcpy(&p->oob_id, CMSG_DATA(cmsg), sizeof(p->oob_id));
+			p->have_oob_id = true;
+			break;
+		case RXRPC_RESPOND:
+			if (p->command != RXRPC_OOB_CMD_UNSET)
+				return -EINVAL;
+			p->command = RXRPC_OOB_CMD_RESPOND;
+			break;
+		case RXRPC_ABORT:
+			if (len != sizeof(p->abort_code) || p->abort_code)
+				return -EINVAL;
+			memcpy(&p->abort_code, CMSG_DATA(cmsg), sizeof(p->abort_code));
+			if (p->abort_code == 0)
+				return -EINVAL;
+			break;
+		case RXRPC_RESP_RXGK_APPDATA:
+			if (p->command != RXRPC_OOB_CMD_RESPOND)
+				return -EINVAL;
+			break;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	switch (p->command) {
+	case RXRPC_OOB_CMD_RESPOND:
+		if (!p->have_oob_id)
+			return -EBADSLT;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+/*
+ * Allow userspace to respond to an OOB using sendmsg().
+ */
+static int rxrpc_respond_to_oob(struct rxrpc_sock *rx,
+				struct rxrpc_oob_params *p,
+				struct msghdr *msg)
+{
+	struct rxrpc_connection *conn;
+	struct rxrpc_skb_priv *sp;
+	struct sk_buff *skb;
+	int ret;
+
+	skb = rxrpc_find_pending_oob(rx, p->oob_id);
+	if (skb)
+		rb_erase(&skb->rbnode, &rx->pending_oobq);
+	release_sock(&rx->sk);
+	if (!skb)
+		return -EBADSLT;
+
+	sp = rxrpc_skb(skb);
+
+	switch (p->command) {
+	case RXRPC_OOB_CMD_RESPOND:
+		ret = -EPROTO;
+		if (skb->mark != RXRPC_OOB_CHALLENGE)
+			break;
+		conn = sp->chall.conn;
+		ret = -EOPNOTSUPP;
+		if (!conn->security->sendmsg_respond_to_challenge)
+			break;
+		if (p->abort_code) {
+			rxrpc_abort_conn(conn, NULL, p->abort_code, -ECONNABORTED,
+					 rxrpc_abort_response_sendmsg);
+			ret = 0;
+		} else {
+			ret = conn->security->sendmsg_respond_to_challenge(skb, msg);
+		}
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	switch (skb->mark) {
+	case RXRPC_OOB_CHALLENGE:
+		rxrpc_put_connection(sp->chall.conn, rxrpc_conn_put_oob);
+		break;
+	}
+	rxrpc_free_skb(skb, rxrpc_skb_put_oob);
+	return ret;
+}
+
+/*
+ * Send an out-of-band message or respond to a received out-of-band message.
+ * - caller gives us the socket lock
+ * - the socket may be either a client socket or a server socket
+ */
+int rxrpc_sendmsg_oob(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
+{
+	struct rxrpc_oob_params p = {};
+	int ret;
+
+	_enter("");
+
+	ret = rxrpc_sendmsg_oob_cmsg(msg, &p);
+	if (ret < 0)
+		goto error_release_sock;
+
+	if (p.have_oob_id)
+		return rxrpc_respond_to_oob(rx, &p, msg);
+
+	release_sock(&rx->sk);
+
+	switch (p.command) {
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	_leave(" = %d", ret);
+	return ret;
+
+error_release_sock:
+	release_sock(&rx->sk);
+	return ret;
+}
+
+/**
+ * rxrpc_kernel_query_oob - Query the parameters of an out-of-band message
+ * @oob: The message to query
+ * @_peer: Where to return the peer record
+ * @_peer_appdata: The application data attached to a peer record
+ *
+ * Extract useful parameters from an out-of-band message.  The source peer
+ * parameters are returned through the argument list and the message type is
+ * returned.
+ *
+ * Return:
+ * * %RXRPC_OOB_CHALLENGE - Challenge wanting a response.
+ */
+enum rxrpc_oob_type rxrpc_kernel_query_oob(struct sk_buff *oob,
+					   struct rxrpc_peer **_peer,
+					   unsigned long *_peer_appdata)
+{
+	struct rxrpc_skb_priv *sp = rxrpc_skb(oob);
+	enum rxrpc_oob_type type = oob->mark;
+
+	switch (type) {
+	case RXRPC_OOB_CHALLENGE:
+		*_peer		= sp->chall.conn->peer;
+		*_peer_appdata	= sp->chall.conn->peer->app_data;
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		*_peer		= NULL;
+		*_peer_appdata	= 0;
+		break;
+	}
+
+	return type;
+}
+EXPORT_SYMBOL(rxrpc_kernel_query_oob);
+
+/**
+ * rxrpc_kernel_dequeue_oob - Dequeue and return the front OOB message
+ * @sock: The socket to query
+ * @_type: Where to return the message type
+ *
+ * Dequeue the front OOB message, if there is one, and return it and
+ * its type.
+ *
+ * Return: The sk_buff representing the OOB message or %NULL if the queue was
+ * empty.
+ */
+struct sk_buff *rxrpc_kernel_dequeue_oob(struct socket *sock,
+					 enum rxrpc_oob_type *_type)
+{
+	struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
+	struct sk_buff *oob;
+
+	oob = skb_dequeue(&rx->recvmsg_oobq);
+	if (oob)
+		*_type = oob->mark;
+	return oob;
+}
+EXPORT_SYMBOL(rxrpc_kernel_dequeue_oob);
+
+/**
+ * rxrpc_kernel_free_oob - Free an out-of-band message
+ * @oob: The OOB message to free
+ *
+ * Free an OOB message along with any resources it holds.
+ */
+void rxrpc_kernel_free_oob(struct sk_buff *oob)
+{
+	struct rxrpc_skb_priv *sp = rxrpc_skb(oob);
+
+	switch (oob->mark) {
+	case RXRPC_OOB_CHALLENGE:
+		rxrpc_put_connection(sp->chall.conn, rxrpc_conn_put_oob);
+		break;
+	}
+
+	rxrpc_free_skb(oob, rxrpc_skb_put_purge_oob);
+}
+EXPORT_SYMBOL(rxrpc_kernel_free_oob);
+
+/**
+ * rxrpc_kernel_query_challenge - Query the parameters of a challenge
+ * @challenge: The challenge to query
+ * @_peer: Where to return the peer record
+ * @_peer_appdata: The application data attached to a peer record
+ * @_service_id: Where to return the connection service ID
+ * @_security_index: Where to return the connection security index
+ *
+ * Extract useful parameters from a CHALLENGE message.
+ */
+void rxrpc_kernel_query_challenge(struct sk_buff *challenge,
+				  struct rxrpc_peer **_peer,
+				  unsigned long *_peer_appdata,
+				  u16 *_service_id, u8 *_security_index)
+{
+	struct rxrpc_skb_priv *sp = rxrpc_skb(challenge);
+
+	*_peer		= sp->chall.conn->peer;
+	*_peer_appdata	= sp->chall.conn->peer->app_data;
+	*_service_id	= sp->hdr.serviceId;
+	*_security_index = sp->hdr.securityIndex;
+}
+EXPORT_SYMBOL(rxrpc_kernel_query_challenge);
+
+/**
+ * rxrpc_kernel_reject_challenge - Allow a kernel service to reject a challenge
+ * @challenge: The challenge to be rejected
+ * @abort_code: The abort code to stick into the ABORT packet
+ * @error: Local error value
+ * @why: Indication as to why.
+ *
+ * Allow a kernel service to reject a challenge by aborting the connection if
+ * it's still in an abortable state.  The error is returned so this function
+ * can be used with a return statement.
+ *
+ * Return: The %error parameter.
+ */
+int rxrpc_kernel_reject_challenge(struct sk_buff *challenge, u32 abort_code,
+				  int error, enum rxrpc_abort_reason why)
+{
+	struct rxrpc_skb_priv *sp = rxrpc_skb(challenge);
+
+	_enter("{%x},%d,%d,%u", sp->hdr.serial, abort_code, error, why);
+
+	rxrpc_abort_conn(sp->chall.conn, NULL, abort_code, error, why);
+	return error;
+}
+EXPORT_SYMBOL(rxrpc_kernel_reject_challenge);