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, 424 insertions, 0 deletions

diff --git a/drivers/net/wireguard/allowedips.c b/drivers/net/wireguard/allowedips.c
new file mode 100644
index 000000000..5ece9acad
--- /dev/null
+++ b/drivers/net/wireguard/allowedips.c
@@ -0,0 +1,424 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
+ */
+
+#include "allowedips.h"
+#include "peer.h"
+
+enum { MAX_ALLOWEDIPS_DEPTH = 129 };
+
+static struct kmem_cache *node_cache;
+
+static void swap_endian(u8 *dst, const u8 *src, u8 bits)
+{
+	if (bits == 32) {
+		*(u32 *)dst = be32_to_cpu(*(const __be32 *)src);
+	} else if (bits == 128) {
+		((u64 *)dst)[0] = get_unaligned_be64(src);
+		((u64 *)dst)[1] = get_unaligned_be64(src + 8);
+	}
+}
+
+static void copy_and_assign_cidr(struct allowedips_node *node, const u8 *src,
+				 u8 cidr, u8 bits)
+{
+	node->cidr = cidr;
+	node->bit_at_a = cidr / 8U;
+#ifdef __LITTLE_ENDIAN
+	node->bit_at_a ^= (bits / 8U - 1U) % 8U;
+#endif
+	node->bit_at_b = 7U - (cidr % 8U);
+	node->bitlen = bits;
+	memcpy(node->bits, src, bits / 8U);
+}
+
+static inline u8 choose(struct allowedips_node *node, const u8 *key)
+{
+	return (key[node->bit_at_a] >> node->bit_at_b) & 1;
+}
+
+static void push_rcu(struct allowedips_node **stack,
+		     struct allowedips_node __rcu *p, unsigned int *len)
+{
+	if (rcu_access_pointer(p)) {
+		if (WARN_ON(IS_ENABLED(DEBUG) && *len >= MAX_ALLOWEDIPS_DEPTH))
+			return;
+		stack[(*len)++] = rcu_dereference_raw(p);
+	}
+}
+
+static void root_free_rcu(struct rcu_head *rcu)
+{
+	struct allowedips_node *node, *stack[MAX_ALLOWEDIPS_DEPTH] = {
+		container_of(rcu, struct allowedips_node, rcu) };
+	unsigned int len = 1;
+
+	while (len > 0 && (node = stack[--len])) {
+		push_rcu(stack, node->bit[0], &len);
+		push_rcu(stack, node->bit[1], &len);
+		kmem_cache_free(node_cache, node);
+	}
+}
+
+static void root_remove_peer_lists(struct allowedips_node *root)
+{
+	struct allowedips_node *node, *stack[MAX_ALLOWEDIPS_DEPTH] = { root };
+	unsigned int len = 1;
+
+	while (len > 0 && (node = stack[--len])) {
+		push_rcu(stack, node->bit[0], &len);
+		push_rcu(stack, node->bit[1], &len);
+		if (rcu_access_pointer(node->peer))
+			list_del(&node->peer_list);
+	}
+}
+
+static unsigned int fls128(u64 a, u64 b)
+{
+	return a ? fls64(a) + 64U : fls64(b);
+}
+
+static u8 common_bits(const struct allowedips_node *node, const u8 *key,
+		      u8 bits)
+{
+	if (bits == 32)
+		return 32U - fls(*(const u32 *)node->bits ^ *(const u32 *)key);
+	else if (bits == 128)
+		return 128U - fls128(
+			*(const u64 *)&node->bits[0] ^ *(const u64 *)&key[0],
+			*(const u64 *)&node->bits[8] ^ *(const u64 *)&key[8]);
+	return 0;
+}
+
+static bool prefix_matches(const struct allowedips_node *node, const u8 *key,
+			   u8 bits)
+{
+	/* This could be much faster if it actually just compared the common
+	 * bits properly, by precomputing a mask bswap(~0 << (32 - cidr)), and
+	 * the rest, but it turns out that common_bits is already super fast on
+	 * modern processors, even taking into account the unfortunate bswap.
+	 * So, we just inline it like this instead.
+	 */
+	return common_bits(node, key, bits) >= node->cidr;
+}
+
+static struct allowedips_node *find_node(struct allowedips_node *trie, u8 bits,
+					 const u8 *key)
+{
+	struct allowedips_node *node = trie, *found = NULL;
+
+	while (node && prefix_matches(node, key, bits)) {
+		if (rcu_access_pointer(node->peer))
+			found = node;
+		if (node->cidr == bits)
+			break;
+		node = rcu_dereference_bh(node->bit[choose(node, key)]);
+	}
+	return found;
+}
+
+/* Returns a strong reference to a peer */
+static struct wg_peer *lookup(struct allowedips_node __rcu *root, u8 bits,
+			      const void *be_ip)
+{
+	/* Aligned so it can be passed to fls/fls64 */
+	u8 ip[16] __aligned(__alignof(u64));
+	struct allowedips_node *node;
+	struct wg_peer *peer = NULL;
+
+	swap_endian(ip, be_ip, bits);
+
+	rcu_read_lock_bh();
+retry:
+	node = find_node(rcu_dereference_bh(root), bits, ip);
+	if (node) {
+		peer = wg_peer_get_maybe_zero(rcu_dereference_bh(node->peer));
+		if (!peer)
+			goto retry;
+	}
+	rcu_read_unlock_bh();
+	return peer;
+}
+
+static bool node_placement(struct allowedips_node __rcu *trie, const u8 *key,
+			   u8 cidr, u8 bits, struct allowedips_node **rnode,
+			   struct mutex *lock)
+{
+	struct allowedips_node *node = rcu_dereference_protected(trie, lockdep_is_held(lock));
+	struct allowedips_node *parent = NULL;
+	bool exact = false;
+
+	while (node && node->cidr <= cidr && prefix_matches(node, key, bits)) {
+		parent = node;
+		if (parent->cidr == cidr) {
+			exact = true;
+			break;
+		}
+		node = rcu_dereference_protected(parent->bit[choose(parent, key)], lockdep_is_held(lock));
+	}
+	*rnode = parent;
+	return exact;
+}
+
+static inline void connect_node(struct allowedips_node __rcu **parent, u8 bit, struct allowedips_node *node)
+{
+	node->parent_bit_packed = (unsigned long)parent | bit;
+	rcu_assign_pointer(*parent, node);
+}
+
+static inline void choose_and_connect_node(struct allowedips_node *parent, struct allowedips_node *node)
+{
+	u8 bit = choose(parent, node->bits);
+	connect_node(&parent->bit[bit], bit, node);
+}
+
+static int add(struct allowedips_node __rcu **trie, u8 bits, const u8 *key,
+	       u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	struct allowedips_node *node, *parent, *down, *newnode;
+
+	if (unlikely(cidr > bits || !peer))
+		return -EINVAL;
+
+	if (!rcu_access_pointer(*trie)) {
+		node = kmem_cache_zalloc(node_cache, GFP_KERNEL);
+		if (unlikely(!node))
+			return -ENOMEM;
+		RCU_INIT_POINTER(node->peer, peer);
+		list_add_tail(&node->peer_list, &peer->allowedips_list);
+		copy_and_assign_cidr(node, key, cidr, bits);
+		connect_node(trie, 2, node);
+		return 0;
+	}
+	if (node_placement(*trie, key, cidr, bits, &node, lock)) {
+		rcu_assign_pointer(node->peer, peer);
+		list_move_tail(&node->peer_list, &peer->allowedips_list);
+		return 0;
+	}
+
+	newnode = kmem_cache_zalloc(node_cache, GFP_KERNEL);
+	if (unlikely(!newnode))
+		return -ENOMEM;
+	RCU_INIT_POINTER(newnode->peer, peer);
+	list_add_tail(&newnode->peer_list, &peer->allowedips_list);
+	copy_and_assign_cidr(newnode, key, cidr, bits);
+
+	if (!node) {
+		down = rcu_dereference_protected(*trie, lockdep_is_held(lock));
+	} else {
+		const u8 bit = choose(node, key);
+		down = rcu_dereference_protected(node->bit[bit], lockdep_is_held(lock));
+		if (!down) {
+			connect_node(&node->bit[bit], bit, newnode);
+			return 0;
+		}
+	}
+	cidr = min(cidr, common_bits(down, key, bits));
+	parent = node;
+
+	if (newnode->cidr == cidr) {
+		choose_and_connect_node(newnode, down);
+		if (!parent)
+			connect_node(trie, 2, newnode);
+		else
+			choose_and_connect_node(parent, newnode);
+		return 0;
+	}
+
+	node = kmem_cache_zalloc(node_cache, GFP_KERNEL);
+	if (unlikely(!node)) {
+		list_del(&newnode->peer_list);
+		kmem_cache_free(node_cache, newnode);
+		return -ENOMEM;
+	}
+	INIT_LIST_HEAD(&node->peer_list);
+	copy_and_assign_cidr(node, newnode->bits, cidr, bits);
+
+	choose_and_connect_node(node, down);
+	choose_and_connect_node(node, newnode);
+	if (!parent)
+		connect_node(trie, 2, node);
+	else
+		choose_and_connect_node(parent, node);
+	return 0;
+}
+
+static void remove_node(struct allowedips_node *node, struct mutex *lock)
+{
+	struct allowedips_node *child, **parent_bit, *parent;
+	bool free_parent;
+
+	list_del_init(&node->peer_list);
+	RCU_INIT_POINTER(node->peer, NULL);
+	if (node->bit[0] && node->bit[1])
+		return;
+	child = rcu_dereference_protected(node->bit[!rcu_access_pointer(node->bit[0])],
+					  lockdep_is_held(lock));
+	if (child)
+		child->parent_bit_packed = node->parent_bit_packed;
+	parent_bit = (struct allowedips_node **)(node->parent_bit_packed & ~3UL);
+	*parent_bit = child;
+	parent = (void *)parent_bit -
+			offsetof(struct allowedips_node, bit[node->parent_bit_packed & 1]);
+	free_parent = !rcu_access_pointer(node->bit[0]) && !rcu_access_pointer(node->bit[1]) &&
+			(node->parent_bit_packed & 3) <= 1 && !rcu_access_pointer(parent->peer);
+	if (free_parent)
+		child = rcu_dereference_protected(parent->bit[!(node->parent_bit_packed & 1)],
+						  lockdep_is_held(lock));
+	kfree_rcu(node, rcu);
+	if (!free_parent)
+		return;
+	if (child)
+		child->parent_bit_packed = parent->parent_bit_packed;
+	*(struct allowedips_node **)(parent->parent_bit_packed & ~3UL) = child;
+	kfree_rcu(parent, rcu);
+}
+
+static int remove(struct allowedips_node __rcu **trie, u8 bits, const u8 *key,
+		  u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	struct allowedips_node *node;
+
+	if (unlikely(cidr > bits))
+		return -EINVAL;
+	if (!rcu_access_pointer(*trie) || !node_placement(*trie, key, cidr, bits, &node, lock) ||
+	    peer != rcu_access_pointer(node->peer))
+		return 0;
+
+	remove_node(node, lock);
+	return 0;
+}
+
+void wg_allowedips_init(struct allowedips *table)
+{
+	table->root4 = table->root6 = NULL;
+	table->seq = 1;
+}
+
+void wg_allowedips_free(struct allowedips *table, struct mutex *lock)
+{
+	struct allowedips_node __rcu *old4 = table->root4, *old6 = table->root6;
+
+	++table->seq;
+	RCU_INIT_POINTER(table->root4, NULL);
+	RCU_INIT_POINTER(table->root6, NULL);
+	if (rcu_access_pointer(old4)) {
+		struct allowedips_node *node = rcu_dereference_protected(old4,
+							lockdep_is_held(lock));
+
+		root_remove_peer_lists(node);
+		call_rcu(&node->rcu, root_free_rcu);
+	}
+	if (rcu_access_pointer(old6)) {
+		struct allowedips_node *node = rcu_dereference_protected(old6,
+							lockdep_is_held(lock));
+
+		root_remove_peer_lists(node);
+		call_rcu(&node->rcu, root_free_rcu);
+	}
+}
+
+int wg_allowedips_insert_v4(struct allowedips *table, const struct in_addr *ip,
+			    u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	/* Aligned so it can be passed to fls */
+	u8 key[4] __aligned(__alignof(u32));
+
+	++table->seq;
+	swap_endian(key, (const u8 *)ip, 32);
+	return add(&table->root4, 32, key, cidr, peer, lock);
+}
+
+int wg_allowedips_insert_v6(struct allowedips *table, const struct in6_addr *ip,
+			    u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	/* Aligned so it can be passed to fls64 */
+	u8 key[16] __aligned(__alignof(u64));
+
+	++table->seq;
+	swap_endian(key, (const u8 *)ip, 128);
+	return add(&table->root6, 128, key, cidr, peer, lock);
+}
+
+int wg_allowedips_remove_v4(struct allowedips *table, const struct in_addr *ip,
+			    u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	/* Aligned so it can be passed to fls */
+	u8 key[4] __aligned(__alignof(u32));
+
+	++table->seq;
+	swap_endian(key, (const u8 *)ip, 32);
+	return remove(&table->root4, 32, key, cidr, peer, lock);
+}
+
+int wg_allowedips_remove_v6(struct allowedips *table, const struct in6_addr *ip,
+			    u8 cidr, struct wg_peer *peer, struct mutex *lock)
+{
+	/* Aligned so it can be passed to fls64 */
+	u8 key[16] __aligned(__alignof(u64));
+
+	++table->seq;
+	swap_endian(key, (const u8 *)ip, 128);
+	return remove(&table->root6, 128, key, cidr, peer, lock);
+}
+
+void wg_allowedips_remove_by_peer(struct allowedips *table,
+				  struct wg_peer *peer, struct mutex *lock)
+{
+	struct allowedips_node *node, *tmp;
+
+	if (list_empty(&peer->allowedips_list))
+		return;
+	++table->seq;
+	list_for_each_entry_safe(node, tmp, &peer->allowedips_list, peer_list)
+		remove_node(node, lock);
+}
+
+int wg_allowedips_read_node(struct allowedips_node *node, u8 ip[16], u8 *cidr)
+{
+	const unsigned int cidr_bytes = DIV_ROUND_UP(node->cidr, 8U);
+	swap_endian(ip, node->bits, node->bitlen);
+	memset(ip + cidr_bytes, 0, node->bitlen / 8U - cidr_bytes);
+	if (node->cidr)
+		ip[cidr_bytes - 1U] &= ~0U << (-node->cidr % 8U);
+
+	*cidr = node->cidr;
+	return node->bitlen == 32 ? AF_INET : AF_INET6;
+}
+
+/* Returns a strong reference to a peer */
+struct wg_peer *wg_allowedips_lookup_dst(struct allowedips *table,
+					 struct sk_buff *skb)
+{
+	if (skb->protocol == htons(ETH_P_IP))
+		return lookup(table->root4, 32, &ip_hdr(skb)->daddr);
+	else if (skb->protocol == htons(ETH_P_IPV6))
+		return lookup(table->root6, 128, &ipv6_hdr(skb)->daddr);
+	return NULL;
+}
+
+/* Returns a strong reference to a peer */
+struct wg_peer *wg_allowedips_lookup_src(struct allowedips *table,
+					 struct sk_buff *skb)
+{
+	if (skb->protocol == htons(ETH_P_IP))
+		return lookup(table->root4, 32, &ip_hdr(skb)->saddr);
+	else if (skb->protocol == htons(ETH_P_IPV6))
+		return lookup(table->root6, 128, &ipv6_hdr(skb)->saddr);
+	return NULL;
+}
+
+int __init wg_allowedips_slab_init(void)
+{
+	node_cache = KMEM_CACHE(allowedips_node, 0);
+	return node_cache ? 0 : -ENOMEM;
+}
+
+void wg_allowedips_slab_uninit(void)
+{
+	rcu_barrier();
+	kmem_cache_destroy(node_cache);
+}
+
+#include "selftest/allowedips.c"