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, 492 insertions, 0 deletions

diff --git a/net/netlink/policy.c b/net/netlink/policy.c
new file mode 100644
index 000000000..08b006c48
--- /dev/null
+++ b/net/netlink/policy.c
@@ -0,0 +1,492 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * NETLINK      Policy advertisement to userspace
+ *
+ * 		Authors:	Johannes Berg <johannes@sipsolutions.net>
+ *
+ * Copyright 2019 Intel Corporation
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+#include <net/netlink.h>
+
+#define INITIAL_POLICIES_ALLOC	10
+
+struct netlink_policy_dump_state {
+	unsigned int policy_idx;
+	unsigned int attr_idx;
+	unsigned int n_alloc;
+	struct {
+		const struct nla_policy *policy;
+		unsigned int maxtype;
+	} policies[] __counted_by(n_alloc);
+};
+
+static int add_policy(struct netlink_policy_dump_state **statep,
+		      const struct nla_policy *policy,
+		      unsigned int maxtype)
+{
+	struct netlink_policy_dump_state *state = *statep;
+	unsigned int old_n_alloc, n_alloc, i;
+
+	if (!policy)
+		return 0;
+
+	for (i = 0; i < state->n_alloc; i++) {
+		if (state->policies[i].policy == policy &&
+		    state->policies[i].maxtype == maxtype)
+			return 0;
+
+		if (!state->policies[i].policy) {
+			state->policies[i].policy = policy;
+			state->policies[i].maxtype = maxtype;
+			return 0;
+		}
+	}
+
+	n_alloc = state->n_alloc + INITIAL_POLICIES_ALLOC;
+	state = krealloc(state, struct_size(state, policies, n_alloc),
+			 GFP_KERNEL);
+	if (!state)
+		return -ENOMEM;
+
+	old_n_alloc = state->n_alloc;
+	state->n_alloc = n_alloc;
+	memset(&state->policies[old_n_alloc], 0,
+	       flex_array_size(state, policies, n_alloc - old_n_alloc));
+
+	state->policies[old_n_alloc].policy = policy;
+	state->policies[old_n_alloc].maxtype = maxtype;
+	*statep = state;
+
+	return 0;
+}
+
+/**
+ * netlink_policy_dump_get_policy_idx - retrieve policy index
+ * @state: the policy dump state
+ * @policy: the policy to find
+ * @maxtype: the policy's maxattr
+ *
+ * Returns: the index of the given policy in the dump state
+ *
+ * Call this to find a policy index when you've added multiple and e.g.
+ * need to tell userspace which command has which policy (by index).
+ *
+ * Note: this will WARN and return 0 if the policy isn't found, which
+ *	 means it wasn't added in the first place, which would be an
+ *	 internal consistency bug.
+ */
+int netlink_policy_dump_get_policy_idx(struct netlink_policy_dump_state *state,
+				       const struct nla_policy *policy,
+				       unsigned int maxtype)
+{
+	unsigned int i;
+
+	if (WARN_ON(!policy))
+                return 0;
+
+	for (i = 0; i < state->n_alloc; i++) {
+		if (state->policies[i].policy == policy &&
+		    state->policies[i].maxtype == maxtype)
+			return i;
+	}
+
+	WARN_ON(1);
+	return 0;
+}
+
+static struct netlink_policy_dump_state *alloc_state(void)
+{
+	struct netlink_policy_dump_state *state;
+
+	state = kzalloc_flex(*state, policies, INITIAL_POLICIES_ALLOC);
+	if (!state)
+		return ERR_PTR(-ENOMEM);
+	state->n_alloc = INITIAL_POLICIES_ALLOC;
+
+	return state;
+}
+
+/**
+ * netlink_policy_dump_add_policy - add a policy to the dump
+ * @pstate: state to add to, may be reallocated, must be %NULL the first time
+ * @policy: the new policy to add to the dump
+ * @maxtype: the new policy's max attr type
+ *
+ * Returns: 0 on success, a negative error code otherwise.
+ *
+ * Call this to allocate a policy dump state, and to add policies to it. This
+ * should be called from the dump start() callback.
+ *
+ * Note: on failures, any previously allocated state is freed.
+ */
+int netlink_policy_dump_add_policy(struct netlink_policy_dump_state **pstate,
+				   const struct nla_policy *policy,
+				   unsigned int maxtype)
+{
+	struct netlink_policy_dump_state *state = *pstate;
+	unsigned int policy_idx;
+	int err;
+
+	if (!state) {
+		state = alloc_state();
+		if (IS_ERR(state))
+			return PTR_ERR(state);
+	}
+
+	/*
+	 * walk the policies and nested ones first, and build
+	 * a linear list of them.
+	 */
+
+	err = add_policy(&state, policy, maxtype);
+	if (err)
+		goto err_try_undo;
+
+	for (policy_idx = 0;
+	     policy_idx < state->n_alloc && state->policies[policy_idx].policy;
+	     policy_idx++) {
+		const struct nla_policy *policy;
+		unsigned int type;
+
+		policy = state->policies[policy_idx].policy;
+
+		for (type = 0;
+		     type <= state->policies[policy_idx].maxtype;
+		     type++) {
+			switch (policy[type].type) {
+			case NLA_NESTED:
+			case NLA_NESTED_ARRAY:
+				err = add_policy(&state,
+						 policy[type].nested_policy,
+						 policy[type].len);
+				if (err)
+					goto err_try_undo;
+				break;
+			default:
+				break;
+			}
+		}
+	}
+
+	*pstate = state;
+	return 0;
+
+err_try_undo:
+	/* Try to preserve reasonable unwind semantics - if we're starting from
+	 * scratch clean up fully, otherwise record what we got and caller will.
+	 */
+	if (!*pstate)
+		netlink_policy_dump_free(state);
+	else
+		*pstate = state;
+	return err;
+}
+
+static bool
+netlink_policy_dump_finished(struct netlink_policy_dump_state *state)
+{
+	return state->policy_idx >= state->n_alloc ||
+	       !state->policies[state->policy_idx].policy;
+}
+
+/**
+ * netlink_policy_dump_loop - dumping loop indicator
+ * @state: the policy dump state
+ *
+ * Returns: %true if the dump continues, %false otherwise
+ *
+ * Note: this frees the dump state when finishing
+ */
+bool netlink_policy_dump_loop(struct netlink_policy_dump_state *state)
+{
+	return !netlink_policy_dump_finished(state);
+}
+
+int netlink_policy_dump_attr_size_estimate(const struct nla_policy *pt)
+{
+	/* nested + type */
+	int common = 2 * nla_attr_size(sizeof(u32));
+
+	switch (pt->type) {
+	case NLA_UNSPEC:
+	case NLA_REJECT:
+		/* these actually don't need any space */
+		return 0;
+	case NLA_NESTED:
+	case NLA_NESTED_ARRAY:
+		/* common, policy idx, policy maxattr */
+		return common + 2 * nla_attr_size(sizeof(u32));
+	case NLA_U8:
+	case NLA_U16:
+	case NLA_U32:
+	case NLA_U64:
+	case NLA_MSECS:
+	case NLA_S8:
+	case NLA_S16:
+	case NLA_S32:
+	case NLA_S64:
+	case NLA_SINT:
+	case NLA_UINT:
+		/* maximum is common, u64 min/max with padding */
+		return common +
+		       2 * (nla_attr_size(0) + nla_attr_size(sizeof(u64)));
+	case NLA_BITFIELD32:
+		return common + nla_attr_size(sizeof(u32));
+	case NLA_STRING:
+	case NLA_NUL_STRING:
+	case NLA_BINARY:
+		/* maximum is common, u32 min-length/max-length */
+		return common + 2 * nla_attr_size(sizeof(u32));
+	case NLA_FLAG:
+		return common;
+	}
+
+	/* this should then cause a warning later */
+	return 0;
+}
+
+static int
+__netlink_policy_dump_write_attr(struct netlink_policy_dump_state *state,
+				 struct sk_buff *skb,
+				 const struct nla_policy *pt,
+				 int nestattr)
+{
+	int estimate = netlink_policy_dump_attr_size_estimate(pt);
+	enum netlink_attribute_type type;
+	struct nlattr *attr;
+
+	attr = nla_nest_start(skb, nestattr);
+	if (!attr)
+		return -ENOBUFS;
+
+	switch (pt->type) {
+	default:
+	case NLA_UNSPEC:
+	case NLA_REJECT:
+		/* skip - use NLA_MIN_LEN to advertise such */
+		nla_nest_cancel(skb, attr);
+		return -ENODATA;
+	case NLA_NESTED:
+		type = NL_ATTR_TYPE_NESTED;
+		fallthrough;
+	case NLA_NESTED_ARRAY:
+		if (pt->type == NLA_NESTED_ARRAY)
+			type = NL_ATTR_TYPE_NESTED_ARRAY;
+		if (state && pt->nested_policy && pt->len &&
+		    (nla_put_u32(skb, NL_POLICY_TYPE_ATTR_POLICY_IDX,
+				 netlink_policy_dump_get_policy_idx(state,
+								    pt->nested_policy,
+								    pt->len)) ||
+		     nla_put_u32(skb, NL_POLICY_TYPE_ATTR_POLICY_MAXTYPE,
+				 pt->len)))
+			goto nla_put_failure;
+		break;
+	case NLA_U8:
+	case NLA_U16:
+	case NLA_U32:
+	case NLA_U64:
+	case NLA_UINT:
+	case NLA_MSECS: {
+		struct netlink_range_validation range;
+
+		if (pt->type == NLA_U8)
+			type = NL_ATTR_TYPE_U8;
+		else if (pt->type == NLA_U16)
+			type = NL_ATTR_TYPE_U16;
+		else if (pt->type == NLA_U32)
+			type = NL_ATTR_TYPE_U32;
+		else if (pt->type == NLA_U64)
+			type = NL_ATTR_TYPE_U64;
+		else
+			type = NL_ATTR_TYPE_UINT;
+
+		if (pt->validation_type == NLA_VALIDATE_MASK) {
+			if (nla_put_u64_64bit(skb, NL_POLICY_TYPE_ATTR_MASK,
+					      pt->mask,
+					      NL_POLICY_TYPE_ATTR_PAD))
+				goto nla_put_failure;
+			break;
+		} else if (pt->validation_type == NLA_VALIDATE_FUNCTION) {
+			break;
+		}
+
+		nla_get_range_unsigned(pt, &range);
+
+		if (nla_put_u64_64bit(skb, NL_POLICY_TYPE_ATTR_MIN_VALUE_U,
+				      range.min, NL_POLICY_TYPE_ATTR_PAD) ||
+		    nla_put_u64_64bit(skb, NL_POLICY_TYPE_ATTR_MAX_VALUE_U,
+				      range.max, NL_POLICY_TYPE_ATTR_PAD))
+			goto nla_put_failure;
+		break;
+	}
+	case NLA_S8:
+	case NLA_S16:
+	case NLA_S32:
+	case NLA_S64:
+	case NLA_SINT: {
+		struct netlink_range_validation_signed range;
+
+		if (pt->type == NLA_S8)
+			type = NL_ATTR_TYPE_S8;
+		else if (pt->type == NLA_S16)
+			type = NL_ATTR_TYPE_S16;
+		else if (pt->type == NLA_S32)
+			type = NL_ATTR_TYPE_S32;
+		else if (pt->type == NLA_S64)
+			type = NL_ATTR_TYPE_S64;
+		else
+			type = NL_ATTR_TYPE_SINT;
+
+		if (pt->validation_type == NLA_VALIDATE_FUNCTION)
+			break;
+
+		nla_get_range_signed(pt, &range);
+
+		if (nla_put_s64(skb, NL_POLICY_TYPE_ATTR_MIN_VALUE_S,
+				range.min, NL_POLICY_TYPE_ATTR_PAD) ||
+		    nla_put_s64(skb, NL_POLICY_TYPE_ATTR_MAX_VALUE_S,
+				range.max, NL_POLICY_TYPE_ATTR_PAD))
+			goto nla_put_failure;
+		break;
+	}
+	case NLA_BITFIELD32:
+		type = NL_ATTR_TYPE_BITFIELD32;
+		if (nla_put_u32(skb, NL_POLICY_TYPE_ATTR_BITFIELD32_MASK,
+				pt->bitfield32_valid))
+			goto nla_put_failure;
+		break;
+	case NLA_STRING:
+	case NLA_NUL_STRING:
+	case NLA_BINARY:
+		if (pt->type == NLA_STRING)
+			type = NL_ATTR_TYPE_STRING;
+		else if (pt->type == NLA_NUL_STRING)
+			type = NL_ATTR_TYPE_NUL_STRING;
+		else
+			type = NL_ATTR_TYPE_BINARY;
+
+		if (pt->validation_type == NLA_VALIDATE_RANGE ||
+		    pt->validation_type == NLA_VALIDATE_RANGE_WARN_TOO_LONG) {
+			struct netlink_range_validation range;
+
+			nla_get_range_unsigned(pt, &range);
+
+			if (range.min &&
+			    nla_put_u32(skb, NL_POLICY_TYPE_ATTR_MIN_LENGTH,
+					range.min))
+				goto nla_put_failure;
+
+			if (range.max < U16_MAX &&
+			    nla_put_u32(skb, NL_POLICY_TYPE_ATTR_MAX_LENGTH,
+					range.max))
+				goto nla_put_failure;
+		} else if (pt->len &&
+			   nla_put_u32(skb, NL_POLICY_TYPE_ATTR_MAX_LENGTH,
+				       pt->len)) {
+			goto nla_put_failure;
+		}
+		break;
+	case NLA_FLAG:
+		type = NL_ATTR_TYPE_FLAG;
+		break;
+	}
+
+	if (nla_put_u32(skb, NL_POLICY_TYPE_ATTR_TYPE, type))
+		goto nla_put_failure;
+
+	nla_nest_end(skb, attr);
+	WARN_ON(attr->nla_len > estimate);
+
+	return 0;
+nla_put_failure:
+	nla_nest_cancel(skb, attr);
+	return -ENOBUFS;
+}
+
+/**
+ * netlink_policy_dump_write_attr - write a given attribute policy
+ * @skb: the message skb to write to
+ * @pt: the attribute's policy
+ * @nestattr: the nested attribute ID to use
+ *
+ * Returns: 0 on success, an error code otherwise; -%ENODATA is
+ *	    special, indicating that there's no policy data and
+ *	    the attribute is generally rejected.
+ */
+int netlink_policy_dump_write_attr(struct sk_buff *skb,
+				   const struct nla_policy *pt,
+				   int nestattr)
+{
+	return __netlink_policy_dump_write_attr(NULL, skb, pt, nestattr);
+}
+
+/**
+ * netlink_policy_dump_write - write current policy dump attributes
+ * @skb: the message skb to write to
+ * @state: the policy dump state
+ *
+ * Returns: 0 on success, an error code otherwise
+ */
+int netlink_policy_dump_write(struct sk_buff *skb,
+			      struct netlink_policy_dump_state *state)
+{
+	const struct nla_policy *pt;
+	struct nlattr *policy;
+	bool again;
+	int err;
+
+send_attribute:
+	again = false;
+
+	pt = &state->policies[state->policy_idx].policy[state->attr_idx];
+
+	policy = nla_nest_start(skb, state->policy_idx);
+	if (!policy)
+		return -ENOBUFS;
+
+	err = __netlink_policy_dump_write_attr(state, skb, pt, state->attr_idx);
+	if (err == -ENODATA) {
+		nla_nest_cancel(skb, policy);
+		again = true;
+		goto next;
+	} else if (err) {
+		goto nla_put_failure;
+	}
+
+	/* finish and move state to next attribute */
+	nla_nest_end(skb, policy);
+
+next:
+	state->attr_idx += 1;
+	if (state->attr_idx > state->policies[state->policy_idx].maxtype) {
+		state->attr_idx = 0;
+		state->policy_idx++;
+	}
+
+	if (again) {
+		if (netlink_policy_dump_finished(state))
+			return -ENODATA;
+		goto send_attribute;
+	}
+
+	return 0;
+
+nla_put_failure:
+	nla_nest_cancel(skb, policy);
+	return -ENOBUFS;
+}
+
+/**
+ * netlink_policy_dump_free - free policy dump state
+ * @state: the policy dump state to free
+ *
+ * Call this from the done() method to ensure dump state is freed.
+ */
+void netlink_policy_dump_free(struct netlink_policy_dump_state *state)
+{
+	kfree(state);
+}