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, 328 insertions, 0 deletions

diff --git a/net/ethtool/mse.c b/net/ethtool/mse.c
new file mode 100644
index 000000000..e91b74430
--- /dev/null
+++ b/net/ethtool/mse.c
@@ -0,0 +1,328 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/ethtool.h>
+#include <linux/phy.h>
+#include <linux/slab.h>
+
+#include "common.h"
+#include "netlink.h"
+
+/* Channels A-D only; WORST and LINK are exclusive alternatives */
+#define PHY_MSE_CHANNEL_COUNT 4
+
+struct mse_req_info {
+	struct ethnl_req_info base;
+};
+
+struct mse_snapshot_entry {
+	struct phy_mse_snapshot snapshot;
+	int channel;
+};
+
+struct mse_reply_data {
+	struct ethnl_reply_data base;
+	struct phy_mse_capability capability;
+	struct mse_snapshot_entry *snapshots;
+	unsigned int num_snapshots;
+};
+
+static struct mse_reply_data *
+mse_repdata(const struct ethnl_reply_data *reply_base)
+{
+	return container_of(reply_base, struct mse_reply_data, base);
+}
+
+const struct nla_policy ethnl_mse_get_policy[] = {
+	[ETHTOOL_A_MSE_HEADER] = NLA_POLICY_NESTED(ethnl_header_policy_phy),
+};
+
+static int get_snapshot_if_supported(struct phy_device *phydev,
+				     struct mse_reply_data *data,
+				     unsigned int *idx, u32 cap_bit,
+				     enum phy_mse_channel channel)
+{
+	int ret;
+
+	if (data->capability.supported_caps & cap_bit) {
+		ret = phydev->drv->get_mse_snapshot(phydev, channel,
+					&data->snapshots[*idx].snapshot);
+		if (ret)
+			return ret;
+		data->snapshots[*idx].channel = channel;
+		(*idx)++;
+	}
+
+	return 0;
+}
+
+static int mse_get_channels(struct phy_device *phydev,
+			    struct mse_reply_data *data)
+{
+	unsigned int i = 0;
+	int ret;
+
+	if (!data->capability.supported_caps)
+		return 0;
+
+	data->snapshots = kzalloc_objs(*data->snapshots, PHY_MSE_CHANNEL_COUNT);
+	if (!data->snapshots)
+		return -ENOMEM;
+
+	/* Priority 1: Individual channels */
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_A,
+					PHY_MSE_CHANNEL_A);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_B,
+					PHY_MSE_CHANNEL_B);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_C,
+					PHY_MSE_CHANNEL_C);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_D,
+					PHY_MSE_CHANNEL_D);
+	if (ret)
+		return ret;
+
+	/* If any individual channels were found, we are done. */
+	if (i > 0) {
+		data->num_snapshots = i;
+		return 0;
+	}
+
+	/* Priority 2: Worst channel, if no individual channels supported. */
+	ret = get_snapshot_if_supported(phydev, data, &i,
+					PHY_MSE_CAP_WORST_CHANNEL,
+					PHY_MSE_CHANNEL_WORST);
+	if (ret)
+		return ret;
+
+	/* If worst channel was found, we are done. */
+	if (i > 0) {
+		data->num_snapshots = i;
+		return 0;
+	}
+
+	/* Priority 3: Link-wide, if nothing else is supported. */
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_LINK,
+					PHY_MSE_CHANNEL_LINK);
+	if (ret)
+		return ret;
+
+	data->num_snapshots = i;
+	return 0;
+}
+
+static int mse_prepare_data(const struct ethnl_req_info *req_base,
+			    struct ethnl_reply_data *reply_base,
+			    const struct genl_info *info)
+{
+	struct mse_reply_data *data = mse_repdata(reply_base);
+	struct net_device *dev = reply_base->dev;
+	struct phy_device *phydev;
+	int ret;
+
+	phydev = ethnl_req_get_phydev(req_base, info->attrs,
+				      ETHTOOL_A_MSE_HEADER, info->extack);
+	if (IS_ERR(phydev))
+		return PTR_ERR(phydev);
+	if (!phydev)
+		return -EOPNOTSUPP;
+
+	ret = ethnl_ops_begin(dev);
+	if (ret)
+		return ret;
+
+	mutex_lock(&phydev->lock);
+
+	if (!phydev->drv || !phydev->drv->get_mse_capability ||
+	    !phydev->drv->get_mse_snapshot) {
+		ret = -EOPNOTSUPP;
+		goto out_unlock;
+	}
+	if (!phydev->link) {
+		ret = -ENETDOWN;
+		goto out_unlock;
+	}
+
+	ret = phydev->drv->get_mse_capability(phydev, &data->capability);
+	if (ret)
+		goto out_unlock;
+
+	ret = mse_get_channels(phydev, data);
+
+out_unlock:
+	mutex_unlock(&phydev->lock);
+	ethnl_ops_complete(dev);
+	if (ret)
+		kfree(data->snapshots);
+	return ret;
+}
+
+static void mse_cleanup_data(struct ethnl_reply_data *reply_base)
+{
+	struct mse_reply_data *data = mse_repdata(reply_base);
+
+	kfree(data->snapshots);
+}
+
+static int mse_reply_size(const struct ethnl_req_info *req_base,
+			  const struct ethnl_reply_data *reply_base)
+{
+	const struct mse_reply_data *data = mse_repdata(reply_base);
+	size_t len = 0;
+	unsigned int i;
+
+	/* ETHTOOL_A_MSE_CAPABILITIES */
+	len += nla_total_size(0);
+	if (data->capability.supported_caps & PHY_MSE_CAP_AVG)
+		/* ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE */
+		len += nla_total_size(sizeof(u64));
+	if (data->capability.supported_caps & (PHY_MSE_CAP_PEAK |
+					       PHY_MSE_CAP_WORST_PEAK))
+		/* ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE */
+		len += nla_total_size(sizeof(u64));
+	/* ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS */
+	len += nla_total_size(sizeof(u64));
+	/* ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS */
+	len += nla_total_size(sizeof(u64));
+
+	for (i = 0; i < data->num_snapshots; i++) {
+		size_t snapshot_len = 0;
+
+		/* Per-channel nest (e.g., ETHTOOL_A_MSE_CHANNEL_A / _B / _C /
+		 * _D / _WORST_CHANNEL / _LINK)
+		 */
+		snapshot_len += nla_total_size(0);
+
+		if (data->capability.supported_caps & PHY_MSE_CAP_AVG)
+			snapshot_len += nla_total_size(sizeof(u64));
+		if (data->capability.supported_caps & PHY_MSE_CAP_PEAK)
+			snapshot_len += nla_total_size(sizeof(u64));
+		if (data->capability.supported_caps & PHY_MSE_CAP_WORST_PEAK)
+			snapshot_len += nla_total_size(sizeof(u64));
+
+		len += snapshot_len;
+	}
+
+	return len;
+}
+
+static int mse_channel_to_attr(int ch)
+{
+	switch (ch) {
+	case PHY_MSE_CHANNEL_A:
+		return ETHTOOL_A_MSE_CHANNEL_A;
+	case PHY_MSE_CHANNEL_B:
+		return ETHTOOL_A_MSE_CHANNEL_B;
+	case PHY_MSE_CHANNEL_C:
+		return ETHTOOL_A_MSE_CHANNEL_C;
+	case PHY_MSE_CHANNEL_D:
+		return ETHTOOL_A_MSE_CHANNEL_D;
+	case PHY_MSE_CHANNEL_WORST:
+		return ETHTOOL_A_MSE_WORST_CHANNEL;
+	case PHY_MSE_CHANNEL_LINK:
+		return ETHTOOL_A_MSE_LINK;
+	default:
+		return -EINVAL;
+	}
+}
+
+static int mse_fill_reply(struct sk_buff *skb,
+			  const struct ethnl_req_info *req_base,
+			  const struct ethnl_reply_data *reply_base)
+{
+	const struct mse_reply_data *data = mse_repdata(reply_base);
+	struct nlattr *nest;
+	unsigned int i;
+	int ret;
+
+	nest = nla_nest_start(skb, ETHTOOL_A_MSE_CAPABILITIES);
+	if (!nest)
+		return -EMSGSIZE;
+
+	if (data->capability.supported_caps & PHY_MSE_CAP_AVG) {
+		ret = nla_put_uint(skb,
+				   ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE,
+				   data->capability.max_average_mse);
+		if (ret < 0)
+			goto nla_put_nest_failure;
+	}
+
+	if (data->capability.supported_caps & (PHY_MSE_CAP_PEAK |
+					       PHY_MSE_CAP_WORST_PEAK)) {
+		ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE,
+				   data->capability.max_peak_mse);
+		if (ret < 0)
+			goto nla_put_nest_failure;
+	}
+
+	ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS,
+			   data->capability.refresh_rate_ps);
+	if (ret < 0)
+		goto nla_put_nest_failure;
+
+	ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS,
+			   data->capability.num_symbols);
+	if (ret < 0)
+		goto nla_put_nest_failure;
+
+	nla_nest_end(skb, nest);
+
+	for (i = 0; i < data->num_snapshots; i++) {
+		const struct mse_snapshot_entry *s = &data->snapshots[i];
+		int chan_attr;
+
+		chan_attr = mse_channel_to_attr(s->channel);
+		if (chan_attr < 0)
+			return chan_attr;
+
+		nest = nla_nest_start(skb, chan_attr);
+		if (!nest)
+			return -EMSGSIZE;
+
+		if (data->capability.supported_caps & PHY_MSE_CAP_AVG) {
+			ret = nla_put_uint(skb,
+					   ETHTOOL_A_MSE_SNAPSHOT_AVERAGE_MSE,
+					   s->snapshot.average_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+		if (data->capability.supported_caps & PHY_MSE_CAP_PEAK) {
+			ret = nla_put_uint(skb, ETHTOOL_A_MSE_SNAPSHOT_PEAK_MSE,
+					   s->snapshot.peak_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+		if (data->capability.supported_caps & PHY_MSE_CAP_WORST_PEAK) {
+			ret = nla_put_uint(skb,
+					   ETHTOOL_A_MSE_SNAPSHOT_WORST_PEAK_MSE,
+					   s->snapshot.worst_peak_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+
+		nla_nest_end(skb, nest);
+	}
+
+	return 0;
+
+nla_put_nest_failure:
+	nla_nest_cancel(skb, nest);
+	return ret;
+}
+
+const struct ethnl_request_ops ethnl_mse_request_ops = {
+	.request_cmd = ETHTOOL_MSG_MSE_GET,
+	.reply_cmd = ETHTOOL_MSG_MSE_GET_REPLY,
+	.hdr_attr = ETHTOOL_A_MSE_HEADER,
+	.req_info_size = sizeof(struct mse_req_info),
+	.reply_data_size = sizeof(struct mse_reply_data),
+
+	.prepare_data = mse_prepare_data,
+	.cleanup_data = mse_cleanup_data,
+	.reply_size = mse_reply_size,
+	.fill_reply = mse_fill_reply,
+};