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, 266 insertions, 0 deletions

diff --git a/samples/bpf/xdp_sample.bpf.c b/samples/bpf/xdp_sample.bpf.c
new file mode 100644
index 000000000..0eb7e1dca
--- /dev/null
+++ b/samples/bpf/xdp_sample.bpf.c
@@ -0,0 +1,266 @@
+// SPDX-License-Identifier: GPL-2.0
+/*  GPLv2, Copyright(c) 2017 Jesper Dangaard Brouer, Red Hat, Inc. */
+#include "xdp_sample.bpf.h"
+
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_core_read.h>
+#include <bpf/bpf_helpers.h>
+
+array_map rx_cnt SEC(".maps");
+array_map redir_err_cnt SEC(".maps");
+array_map cpumap_enqueue_cnt SEC(".maps");
+array_map cpumap_kthread_cnt SEC(".maps");
+array_map exception_cnt SEC(".maps");
+array_map devmap_xmit_cnt SEC(".maps");
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_HASH);
+	__uint(max_entries, 32 * 32);
+	__type(key, u64);
+	__type(value, struct datarec);
+} devmap_xmit_cnt_multi SEC(".maps");
+
+const volatile int nr_cpus = 0;
+
+/* These can be set before loading so that redundant comparisons can be DCE'd by
+ * the verifier, and only actual matches are tried after loading tp_btf program.
+ * This allows sample to filter tracepoint stats based on net_device.
+ */
+const volatile int from_match[32] = {};
+const volatile int to_match[32] = {};
+
+int cpumap_map_id = 0;
+
+/* Find if b is part of set a, but if a is empty set then evaluate to true */
+#define IN_SET(a, b)                                                 \
+	({                                                           \
+		bool __res = !(a)[0];                                \
+		for (int i = 0; i < ARRAY_SIZE(a) && (a)[i]; i++) { \
+			__res = (a)[i] == (b);                       \
+			if (__res)                                   \
+				break;                               \
+		}                                                    \
+		__res;                                               \
+	})
+
+static __always_inline __u32 xdp_get_err_key(int err)
+{
+	switch (err) {
+	case 0:
+		return 0;
+	case -EINVAL:
+		return 2;
+	case -ENETDOWN:
+		return 3;
+	case -EMSGSIZE:
+		return 4;
+	case -EOPNOTSUPP:
+		return 5;
+	case -ENOSPC:
+		return 6;
+	default:
+		return 1;
+	}
+}
+
+static __always_inline int xdp_redirect_collect_stat(int from, int err)
+{
+	u32 cpu = bpf_get_smp_processor_id();
+	u32 key = XDP_REDIRECT_ERROR;
+	struct datarec *rec;
+	u32 idx;
+
+	if (!IN_SET(from_match, from))
+		return 0;
+
+	key = xdp_get_err_key(err);
+
+	idx = key * nr_cpus + cpu;
+	rec = bpf_map_lookup_elem(&redir_err_cnt, &idx);
+	if (!rec)
+		return 0;
+	if (key)
+		NO_TEAR_INC(rec->dropped);
+	else
+		NO_TEAR_INC(rec->processed);
+	return 0; /* Indicate event was filtered (no further processing)*/
+	/*
+	 * Returning 1 here would allow e.g. a perf-record tracepoint
+	 * to see and record these events, but it doesn't work well
+	 * in-practice as stopping perf-record also unload this
+	 * bpf_prog.  Plus, there is additional overhead of doing so.
+	 */
+}
+
+SEC("tp_btf/xdp_redirect_err")
+int BPF_PROG(tp_xdp_redirect_err, const struct net_device *dev,
+	     const struct bpf_prog *xdp, const void *tgt, int err,
+	     const struct bpf_map *map, u32 index)
+{
+	return xdp_redirect_collect_stat(dev->ifindex, err);
+}
+
+SEC("tp_btf/xdp_redirect_map_err")
+int BPF_PROG(tp_xdp_redirect_map_err, const struct net_device *dev,
+	     const struct bpf_prog *xdp, const void *tgt, int err,
+	     const struct bpf_map *map, u32 index)
+{
+	return xdp_redirect_collect_stat(dev->ifindex, err);
+}
+
+SEC("tp_btf/xdp_redirect")
+int BPF_PROG(tp_xdp_redirect, const struct net_device *dev,
+	     const struct bpf_prog *xdp, const void *tgt, int err,
+	     const struct bpf_map *map, u32 index)
+{
+	return xdp_redirect_collect_stat(dev->ifindex, err);
+}
+
+SEC("tp_btf/xdp_redirect_map")
+int BPF_PROG(tp_xdp_redirect_map, const struct net_device *dev,
+	     const struct bpf_prog *xdp, const void *tgt, int err,
+	     const struct bpf_map *map, u32 index)
+{
+	return xdp_redirect_collect_stat(dev->ifindex, err);
+}
+
+SEC("tp_btf/xdp_cpumap_enqueue")
+int BPF_PROG(tp_xdp_cpumap_enqueue, int map_id, unsigned int processed,
+	     unsigned int drops, int to_cpu)
+{
+	u32 cpu = bpf_get_smp_processor_id();
+	struct datarec *rec;
+	u32 idx;
+
+	if (cpumap_map_id && cpumap_map_id != map_id)
+		return 0;
+
+	idx = to_cpu * nr_cpus + cpu;
+	rec = bpf_map_lookup_elem(&cpumap_enqueue_cnt, &idx);
+	if (!rec)
+		return 0;
+	NO_TEAR_ADD(rec->processed, processed);
+	NO_TEAR_ADD(rec->dropped, drops);
+	/* Record bulk events, then userspace can calc average bulk size */
+	if (processed > 0)
+		NO_TEAR_INC(rec->issue);
+	/* Inception: It's possible to detect overload situations, via
+	 * this tracepoint.  This can be used for creating a feedback
+	 * loop to XDP, which can take appropriate actions to mitigate
+	 * this overload situation.
+	 */
+	return 0;
+}
+
+SEC("tp_btf/xdp_cpumap_kthread")
+int BPF_PROG(tp_xdp_cpumap_kthread, int map_id, unsigned int processed,
+	     unsigned int drops, int sched, struct xdp_cpumap_stats *xdp_stats)
+{
+	struct datarec *rec;
+	u32 cpu;
+
+	if (cpumap_map_id && cpumap_map_id != map_id)
+		return 0;
+
+	cpu = bpf_get_smp_processor_id();
+	rec = bpf_map_lookup_elem(&cpumap_kthread_cnt, &cpu);
+	if (!rec)
+		return 0;
+	NO_TEAR_ADD(rec->processed, processed);
+	NO_TEAR_ADD(rec->dropped, drops);
+	NO_TEAR_ADD(rec->xdp_pass, xdp_stats->pass);
+	NO_TEAR_ADD(rec->xdp_drop, xdp_stats->drop);
+	NO_TEAR_ADD(rec->xdp_redirect, xdp_stats->redirect);
+	/* Count times kthread yielded CPU via schedule call */
+	if (sched)
+		NO_TEAR_INC(rec->issue);
+	return 0;
+}
+
+SEC("tp_btf/xdp_exception")
+int BPF_PROG(tp_xdp_exception, const struct net_device *dev,
+	     const struct bpf_prog *xdp, u32 act)
+{
+	u32 cpu = bpf_get_smp_processor_id();
+	struct datarec *rec;
+	u32 key = act, idx;
+
+	if (!IN_SET(from_match, dev->ifindex))
+		return 0;
+	if (!IN_SET(to_match, dev->ifindex))
+		return 0;
+
+	if (key > XDP_REDIRECT)
+		key = XDP_REDIRECT + 1;
+
+	idx = key * nr_cpus + cpu;
+	rec = bpf_map_lookup_elem(&exception_cnt, &idx);
+	if (!rec)
+		return 0;
+	NO_TEAR_INC(rec->dropped);
+
+	return 0;
+}
+
+SEC("tp_btf/xdp_devmap_xmit")
+int BPF_PROG(tp_xdp_devmap_xmit, const struct net_device *from_dev,
+	     const struct net_device *to_dev, int sent, int drops, int err)
+{
+	struct datarec *rec;
+	int idx_in, idx_out;
+	u32 cpu;
+
+	idx_in = from_dev->ifindex;
+	idx_out = to_dev->ifindex;
+
+	if (!IN_SET(from_match, idx_in))
+		return 0;
+	if (!IN_SET(to_match, idx_out))
+		return 0;
+
+	cpu = bpf_get_smp_processor_id();
+	rec = bpf_map_lookup_elem(&devmap_xmit_cnt, &cpu);
+	if (!rec)
+		return 0;
+	NO_TEAR_ADD(rec->processed, sent);
+	NO_TEAR_ADD(rec->dropped, drops);
+	/* Record bulk events, then userspace can calc average bulk size */
+	NO_TEAR_INC(rec->info);
+	/* Record error cases, where no frame were sent */
+	/* Catch API error of drv ndo_xdp_xmit sent more than count */
+	if (err || drops < 0)
+		NO_TEAR_INC(rec->issue);
+	return 0;
+}
+
+SEC("tp_btf/xdp_devmap_xmit")
+int BPF_PROG(tp_xdp_devmap_xmit_multi, const struct net_device *from_dev,
+	     const struct net_device *to_dev, int sent, int drops, int err)
+{
+	struct datarec empty = {};
+	struct datarec *rec;
+	int idx_in, idx_out;
+	u64 idx;
+
+	idx_in = from_dev->ifindex;
+	idx_out = to_dev->ifindex;
+	idx = idx_in;
+	idx = idx << 32 | idx_out;
+
+	if (!IN_SET(from_match, idx_in))
+		return 0;
+	if (!IN_SET(to_match, idx_out))
+		return 0;
+
+	bpf_map_update_elem(&devmap_xmit_cnt_multi, &idx, &empty, BPF_NOEXIST);
+	rec = bpf_map_lookup_elem(&devmap_xmit_cnt_multi, &idx);
+	if (!rec)
+		return 0;
+
+	NO_TEAR_ADD(rec->processed, sent);
+	NO_TEAR_ADD(rec->dropped, drops);
+	NO_TEAR_INC(rec->info);
+	if (err || drops < 0)
+		NO_TEAR_INC(rec->issue);
+	return 0;
+}