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, 273 insertions, 0 deletions

diff --git a/drivers/virt/acrn/ioeventfd.c b/drivers/virt/acrn/ioeventfd.c
new file mode 100644
index 000000000..66fe9f3ce
--- /dev/null
+++ b/drivers/virt/acrn/ioeventfd.c
@@ -0,0 +1,273 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * ACRN HSM eventfd - use eventfd objects to signal expected I/O requests
+ *
+ * Copyright (C) 2020 Intel Corporation. All rights reserved.
+ *
+ * Authors:
+ *	Shuo Liu <shuo.a.liu@intel.com>
+ *	Yakui Zhao <yakui.zhao@intel.com>
+ */
+
+#include <linux/eventfd.h>
+#include <linux/slab.h>
+
+#include "acrn_drv.h"
+
+/**
+ * struct hsm_ioeventfd - Properties of HSM ioeventfd
+ * @list:	Entry within &acrn_vm.ioeventfds of ioeventfds of a VM
+ * @eventfd:	Eventfd of the HSM ioeventfd
+ * @addr:	Address of I/O range
+ * @data:	Data for matching
+ * @length:	Length of I/O range
+ * @type:	Type of I/O range (ACRN_IOREQ_TYPE_MMIO/ACRN_IOREQ_TYPE_PORTIO)
+ * @wildcard:	Data matching or not
+ */
+struct hsm_ioeventfd {
+	struct list_head	list;
+	struct eventfd_ctx	*eventfd;
+	u64			addr;
+	u64			data;
+	int			length;
+	int			type;
+	bool			wildcard;
+};
+
+static inline int ioreq_type_from_flags(int flags)
+{
+	return flags & ACRN_IOEVENTFD_FLAG_PIO ?
+		       ACRN_IOREQ_TYPE_PORTIO : ACRN_IOREQ_TYPE_MMIO;
+}
+
+static void acrn_ioeventfd_shutdown(struct acrn_vm *vm, struct hsm_ioeventfd *p)
+{
+	lockdep_assert_held(&vm->ioeventfds_lock);
+
+	eventfd_ctx_put(p->eventfd);
+	list_del(&p->list);
+	kfree(p);
+}
+
+static bool hsm_ioeventfd_is_conflict(struct acrn_vm *vm,
+				      struct hsm_ioeventfd *ioeventfd)
+{
+	struct hsm_ioeventfd *p;
+
+	lockdep_assert_held(&vm->ioeventfds_lock);
+
+	/* Either one is wildcard, the data matching will be skipped. */
+	list_for_each_entry(p, &vm->ioeventfds, list)
+		if (p->eventfd == ioeventfd->eventfd &&
+		    p->addr == ioeventfd->addr &&
+		    p->type == ioeventfd->type &&
+		    (p->wildcard || ioeventfd->wildcard ||
+			p->data == ioeventfd->data))
+			return true;
+
+	return false;
+}
+
+/*
+ * Assign an eventfd to a VM and create a HSM ioeventfd associated with the
+ * eventfd. The properties of the HSM ioeventfd are built from a &struct
+ * acrn_ioeventfd.
+ */
+static int acrn_ioeventfd_assign(struct acrn_vm *vm,
+				 struct acrn_ioeventfd *args)
+{
+	struct eventfd_ctx *eventfd;
+	struct hsm_ioeventfd *p;
+	int ret;
+
+	/* Check for range overflow */
+	if (args->addr + args->len < args->addr)
+		return -EINVAL;
+
+	/*
+	 * Currently, acrn_ioeventfd is used to support vhost. 1,2,4,8 width
+	 * accesses can cover vhost's requirements.
+	 */
+	if (!(args->len == 1 || args->len == 2 ||
+	      args->len == 4 || args->len == 8))
+		return -EINVAL;
+
+	eventfd = eventfd_ctx_fdget(args->fd);
+	if (IS_ERR(eventfd))
+		return PTR_ERR(eventfd);
+
+	p = kzalloc_obj(*p);
+	if (!p) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	INIT_LIST_HEAD(&p->list);
+	p->addr = args->addr;
+	p->length = args->len;
+	p->eventfd = eventfd;
+	p->type = ioreq_type_from_flags(args->flags);
+
+	/*
+	 * ACRN_IOEVENTFD_FLAG_DATAMATCH flag is set in virtio 1.0 support, the
+	 * writing of notification register of each virtqueue may trigger the
+	 * notification. There is no data matching requirement.
+	 */
+	if (args->flags & ACRN_IOEVENTFD_FLAG_DATAMATCH)
+		p->data = args->data;
+	else
+		p->wildcard = true;
+
+	mutex_lock(&vm->ioeventfds_lock);
+
+	if (hsm_ioeventfd_is_conflict(vm, p)) {
+		ret = -EEXIST;
+		goto unlock_fail;
+	}
+
+	/* register the I/O range into ioreq client */
+	ret = acrn_ioreq_range_add(vm->ioeventfd_client, p->type,
+				   p->addr, p->addr + p->length - 1);
+	if (ret < 0)
+		goto unlock_fail;
+
+	list_add_tail(&p->list, &vm->ioeventfds);
+	mutex_unlock(&vm->ioeventfds_lock);
+
+	return 0;
+
+unlock_fail:
+	mutex_unlock(&vm->ioeventfds_lock);
+	kfree(p);
+fail:
+	eventfd_ctx_put(eventfd);
+	return ret;
+}
+
+static int acrn_ioeventfd_deassign(struct acrn_vm *vm,
+				   struct acrn_ioeventfd *args)
+{
+	struct hsm_ioeventfd *p;
+	struct eventfd_ctx *eventfd;
+
+	eventfd = eventfd_ctx_fdget(args->fd);
+	if (IS_ERR(eventfd))
+		return PTR_ERR(eventfd);
+
+	mutex_lock(&vm->ioeventfds_lock);
+	list_for_each_entry(p, &vm->ioeventfds, list) {
+		if (p->eventfd != eventfd)
+			continue;
+
+		acrn_ioreq_range_del(vm->ioeventfd_client, p->type,
+				     p->addr, p->addr + p->length - 1);
+		acrn_ioeventfd_shutdown(vm, p);
+		break;
+	}
+	mutex_unlock(&vm->ioeventfds_lock);
+
+	eventfd_ctx_put(eventfd);
+	return 0;
+}
+
+static struct hsm_ioeventfd *hsm_ioeventfd_match(struct acrn_vm *vm, u64 addr,
+						 u64 data, int len, int type)
+{
+	struct hsm_ioeventfd *p = NULL;
+
+	lockdep_assert_held(&vm->ioeventfds_lock);
+
+	list_for_each_entry(p, &vm->ioeventfds, list) {
+		if (p->type == type && p->addr == addr && p->length >= len &&
+		    (p->wildcard || p->data == data))
+			return p;
+	}
+
+	return NULL;
+}
+
+static int acrn_ioeventfd_handler(struct acrn_ioreq_client *client,
+				  struct acrn_io_request *req)
+{
+	struct hsm_ioeventfd *p;
+	u64 addr, val;
+	int size;
+
+	if (req->type == ACRN_IOREQ_TYPE_MMIO) {
+		/*
+		 * I/O requests are dispatched by range check only, so a
+		 * acrn_ioreq_client need process both READ and WRITE accesses
+		 * of same range. READ accesses are safe to be ignored here
+		 * because virtio PCI devices write the notify registers for
+		 * notification.
+		 */
+		if (req->reqs.mmio_request.direction == ACRN_IOREQ_DIR_READ) {
+			/* reading does nothing and return 0 */
+			req->reqs.mmio_request.value = 0;
+			return 0;
+		}
+		addr = req->reqs.mmio_request.address;
+		size = req->reqs.mmio_request.size;
+		val = req->reqs.mmio_request.value;
+	} else {
+		if (req->reqs.pio_request.direction == ACRN_IOREQ_DIR_READ) {
+			/* reading does nothing and return 0 */
+			req->reqs.pio_request.value = 0;
+			return 0;
+		}
+		addr = req->reqs.pio_request.address;
+		size = req->reqs.pio_request.size;
+		val = req->reqs.pio_request.value;
+	}
+
+	mutex_lock(&client->vm->ioeventfds_lock);
+	p = hsm_ioeventfd_match(client->vm, addr, val, size, req->type);
+	if (p)
+		eventfd_signal(p->eventfd);
+	mutex_unlock(&client->vm->ioeventfds_lock);
+
+	return 0;
+}
+
+int acrn_ioeventfd_config(struct acrn_vm *vm, struct acrn_ioeventfd *args)
+{
+	int ret;
+
+	if (args->flags & ACRN_IOEVENTFD_FLAG_DEASSIGN)
+		ret = acrn_ioeventfd_deassign(vm, args);
+	else
+		ret = acrn_ioeventfd_assign(vm, args);
+
+	return ret;
+}
+
+int acrn_ioeventfd_init(struct acrn_vm *vm)
+{
+	char name[ACRN_NAME_LEN];
+
+	mutex_init(&vm->ioeventfds_lock);
+	INIT_LIST_HEAD(&vm->ioeventfds);
+	snprintf(name, sizeof(name), "ioeventfd-%u", vm->vmid);
+	vm->ioeventfd_client = acrn_ioreq_client_create(vm,
+							acrn_ioeventfd_handler,
+							NULL, false, name);
+	if (!vm->ioeventfd_client) {
+		dev_err(acrn_dev.this_device, "Failed to create ioeventfd ioreq client!\n");
+		return -EINVAL;
+	}
+
+	dev_dbg(acrn_dev.this_device, "VM %u ioeventfd init.\n", vm->vmid);
+	return 0;
+}
+
+void acrn_ioeventfd_deinit(struct acrn_vm *vm)
+{
+	struct hsm_ioeventfd *p, *next;
+
+	dev_dbg(acrn_dev.this_device, "VM %u ioeventfd deinit.\n", vm->vmid);
+	acrn_ioreq_client_destroy(vm->ioeventfd_client);
+	mutex_lock(&vm->ioeventfds_lock);
+	list_for_each_entry_safe(p, next, &vm->ioeventfds, list)
+		acrn_ioeventfd_shutdown(vm, p);
+	mutex_unlock(&vm->ioeventfds_lock);
+}