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, 270 insertions, 0 deletions

diff --git a/drivers/md/dm-ps-io-affinity.c b/drivers/md/dm-ps-io-affinity.c
new file mode 100644
index 000000000..526d94d13
--- /dev/null
+++ b/drivers/md/dm-ps-io-affinity.c
@@ -0,0 +1,270 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020 Oracle Corporation
+ *
+ * Module Author: Mike Christie
+ */
+#include "dm-path-selector.h"
+
+#include <linux/device-mapper.h>
+#include <linux/module.h>
+
+#define DM_MSG_PREFIX "multipath io-affinity"
+
+struct path_info {
+	struct dm_path *path;
+	cpumask_var_t cpumask;
+	refcount_t refcount;
+	bool failed;
+};
+
+struct selector {
+	struct path_info **path_map;
+	cpumask_var_t path_mask;
+	atomic_t map_misses;
+};
+
+static void ioa_free_path(struct selector *s, unsigned int cpu)
+{
+	struct path_info *pi = s->path_map[cpu];
+
+	if (!pi)
+		return;
+
+	if (refcount_dec_and_test(&pi->refcount)) {
+		cpumask_clear_cpu(cpu, s->path_mask);
+		free_cpumask_var(pi->cpumask);
+		kfree(pi);
+
+		s->path_map[cpu] = NULL;
+	}
+}
+
+static int ioa_add_path(struct path_selector *ps, struct dm_path *path,
+			int argc, char **argv, char **error)
+{
+	struct selector *s = ps->context;
+	struct path_info *pi = NULL;
+	unsigned int cpu;
+	int ret;
+
+	if (argc != 1) {
+		*error = "io-affinity ps: invalid number of arguments";
+		return -EINVAL;
+	}
+
+	pi = kzalloc_obj(*pi);
+	if (!pi) {
+		*error = "io-affinity ps: Error allocating path context";
+		return -ENOMEM;
+	}
+
+	pi->path = path;
+	path->pscontext = pi;
+	refcount_set(&pi->refcount, 1);
+
+	if (!zalloc_cpumask_var(&pi->cpumask, GFP_KERNEL)) {
+		*error = "io-affinity ps: Error allocating cpumask context";
+		ret = -ENOMEM;
+		goto free_pi;
+	}
+
+	ret = cpumask_parse(argv[0], pi->cpumask);
+	if (ret) {
+		*error = "io-affinity ps: invalid cpumask";
+		ret = -EINVAL;
+		goto free_mask;
+	}
+
+	for_each_cpu(cpu, pi->cpumask) {
+		if (cpu >= nr_cpu_ids) {
+			DMWARN_LIMIT("Ignoring mapping for CPU %u. Max CPU is %u",
+				     cpu, nr_cpu_ids);
+			break;
+		}
+
+		if (s->path_map[cpu]) {
+			DMWARN("CPU mapping for %u exists. Ignoring.", cpu);
+			continue;
+		}
+
+		cpumask_set_cpu(cpu, s->path_mask);
+		s->path_map[cpu] = pi;
+		refcount_inc(&pi->refcount);
+	}
+
+	if (refcount_dec_and_test(&pi->refcount)) {
+		*error = "io-affinity ps: No new/valid CPU mapping found";
+		ret = -EINVAL;
+		goto free_mask;
+	}
+
+	return 0;
+
+free_mask:
+	free_cpumask_var(pi->cpumask);
+free_pi:
+	kfree(pi);
+	return ret;
+}
+
+static int ioa_create(struct path_selector *ps, unsigned int argc, char **argv)
+{
+	struct selector *s;
+
+	s = kmalloc_obj(*s);
+	if (!s)
+		return -ENOMEM;
+
+	s->path_map = kzalloc_objs(struct path_info *, nr_cpu_ids);
+	if (!s->path_map)
+		goto free_selector;
+
+	if (!zalloc_cpumask_var(&s->path_mask, GFP_KERNEL))
+		goto free_map;
+
+	atomic_set(&s->map_misses, 0);
+	ps->context = s;
+	return 0;
+
+free_map:
+	kfree(s->path_map);
+free_selector:
+	kfree(s);
+	return -ENOMEM;
+}
+
+static void ioa_destroy(struct path_selector *ps)
+{
+	struct selector *s = ps->context;
+	unsigned int cpu;
+
+	for_each_cpu(cpu, s->path_mask)
+		ioa_free_path(s, cpu);
+
+	free_cpumask_var(s->path_mask);
+	kfree(s->path_map);
+	kfree(s);
+
+	ps->context = NULL;
+}
+
+static int ioa_status(struct path_selector *ps, struct dm_path *path,
+		      status_type_t type, char *result, unsigned int maxlen)
+{
+	struct selector *s = ps->context;
+	struct path_info *pi;
+	int sz = 0;
+
+	if (!path) {
+		DMEMIT("0 ");
+		return sz;
+	}
+
+	switch (type) {
+	case STATUSTYPE_INFO:
+		DMEMIT("%d ", atomic_read(&s->map_misses));
+		break;
+	case STATUSTYPE_TABLE:
+		pi = path->pscontext;
+		DMEMIT("%*pb ", cpumask_pr_args(pi->cpumask));
+		break;
+	case STATUSTYPE_IMA:
+		*result = '\0';
+		break;
+	}
+
+	return sz;
+}
+
+static void ioa_fail_path(struct path_selector *ps, struct dm_path *p)
+{
+	struct path_info *pi = p->pscontext;
+
+	pi->failed = true;
+}
+
+static int ioa_reinstate_path(struct path_selector *ps, struct dm_path *p)
+{
+	struct path_info *pi = p->pscontext;
+
+	pi->failed = false;
+	return 0;
+}
+
+static struct dm_path *ioa_select_path(struct path_selector *ps,
+				       size_t nr_bytes)
+{
+	unsigned int cpu, node;
+	struct selector *s = ps->context;
+	const struct cpumask *cpumask;
+	struct path_info *pi;
+	int i;
+
+	cpu = get_cpu();
+
+	pi = s->path_map[cpu];
+	if (pi && !pi->failed)
+		goto done;
+
+	/*
+	 * Perf is not optimal, but we at least try the local node then just
+	 * try not to fail.
+	 */
+	if (!pi)
+		atomic_inc(&s->map_misses);
+
+	node = cpu_to_node(cpu);
+	cpumask = cpumask_of_node(node);
+	for_each_cpu(i, cpumask) {
+		pi = s->path_map[i];
+		if (pi && !pi->failed)
+			goto done;
+	}
+
+	for_each_cpu(i, s->path_mask) {
+		pi = s->path_map[i];
+		if (pi && !pi->failed)
+			goto done;
+	}
+	pi = NULL;
+
+done:
+	put_cpu();
+	return pi ? pi->path : NULL;
+}
+
+static struct path_selector_type ioa_ps = {
+	.name		= "io-affinity",
+	.module		= THIS_MODULE,
+	.table_args	= 1,
+	.info_args	= 1,
+	.create		= ioa_create,
+	.destroy	= ioa_destroy,
+	.status		= ioa_status,
+	.add_path	= ioa_add_path,
+	.fail_path	= ioa_fail_path,
+	.reinstate_path	= ioa_reinstate_path,
+	.select_path	= ioa_select_path,
+};
+
+static int __init dm_ioa_init(void)
+{
+	int ret = dm_register_path_selector(&ioa_ps);
+
+	if (ret < 0)
+		DMERR("register failed %d", ret);
+	return ret;
+}
+
+static void __exit dm_ioa_exit(void)
+{
+	dm_unregister_path_selector(&ioa_ps);
+}
+
+module_init(dm_ioa_init);
+module_exit(dm_ioa_exit);
+
+MODULE_DESCRIPTION(DM_NAME " multipath path selector that selects paths based on the CPU IO is being executed on");
+MODULE_AUTHOR("Mike Christie <michael.christie@oracle.com>");
+MODULE_LICENSE("GPL");