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, 315 insertions, 0 deletions

diff --git a/drivers/firmware/efi/capsule.c b/drivers/firmware/efi/capsule.c
new file mode 100644
index 000000000..dd6252638
--- /dev/null
+++ b/drivers/firmware/efi/capsule.c
@@ -0,0 +1,315 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * EFI capsule support.
+ *
+ * Copyright 2013 Intel Corporation; author Matt Fleming
+ */
+
+#define pr_fmt(fmt) "efi: " fmt
+
+#include <linux/slab.h>
+#include <linux/mutex.h>
+#include <linux/highmem.h>
+#include <linux/efi.h>
+#include <linux/vmalloc.h>
+#include <asm/efi.h>
+#include <asm/io.h>
+
+typedef struct {
+	u64 length;
+	u64 data;
+} efi_capsule_block_desc_t;
+
+static bool capsule_pending;
+static bool stop_capsules;
+static int efi_reset_type = -1;
+
+/*
+ * capsule_mutex serialises access to both capsule_pending and
+ * efi_reset_type and stop_capsules.
+ */
+static DEFINE_MUTEX(capsule_mutex);
+
+/**
+ * efi_capsule_pending - has a capsule been passed to the firmware?
+ * @reset_type: store the type of EFI reset if capsule is pending
+ *
+ * To ensure that the registered capsule is processed correctly by the
+ * firmware we need to perform a specific type of reset. If a capsule is
+ * pending return the reset type in @reset_type.
+ *
+ * This function will race with callers of efi_capsule_update(), for
+ * example, calling this function while somebody else is in
+ * efi_capsule_update() but hasn't reached efi_capsue_update_locked()
+ * will miss the updates to capsule_pending and efi_reset_type after
+ * efi_capsule_update_locked() completes.
+ *
+ * A non-racy use is from platform reboot code because we use
+ * system_state to ensure no capsules can be sent to the firmware once
+ * we're at SYSTEM_RESTART. See efi_capsule_update_locked().
+ */
+bool efi_capsule_pending(int *reset_type)
+{
+	if (!capsule_pending)
+		return false;
+
+	if (reset_type)
+		*reset_type = efi_reset_type;
+
+	return true;
+}
+
+/*
+ * Whitelist of EFI capsule flags that we support.
+ *
+ * We do not handle EFI_CAPSULE_INITIATE_RESET because that would
+ * require us to prepare the kernel for reboot. Refuse to load any
+ * capsules with that flag and any other flags that we do not know how
+ * to handle.
+ */
+#define EFI_CAPSULE_SUPPORTED_FLAG_MASK			\
+	(EFI_CAPSULE_PERSIST_ACROSS_RESET | EFI_CAPSULE_POPULATE_SYSTEM_TABLE)
+
+/**
+ * efi_capsule_supported - does the firmware support the capsule?
+ * @guid: vendor guid of capsule
+ * @flags: capsule flags
+ * @size: size of capsule data
+ * @reset: the reset type required for this capsule
+ *
+ * Check whether a capsule with @flags is supported by the firmware
+ * and that @size doesn't exceed the maximum size for a capsule.
+ *
+ * No attempt is made to check @reset against the reset type required
+ * by any pending capsules because of the races involved.
+ */
+int efi_capsule_supported(efi_guid_t guid, u32 flags, size_t size, int *reset)
+{
+	efi_capsule_header_t capsule;
+	efi_capsule_header_t *cap_list[] = { &capsule };
+	efi_status_t status;
+	u64 max_size;
+
+	if (flags & ~EFI_CAPSULE_SUPPORTED_FLAG_MASK)
+		return -EINVAL;
+
+	capsule.headersize = capsule.imagesize = sizeof(capsule);
+	memcpy(&capsule.guid, &guid, sizeof(efi_guid_t));
+	capsule.flags = flags;
+
+	status = efi.query_capsule_caps(cap_list, 1, &max_size, reset);
+	if (status != EFI_SUCCESS)
+		return efi_status_to_err(status);
+
+	if (size > max_size)
+		return -ENOSPC;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(efi_capsule_supported);
+
+/*
+ * Every scatter gather list (block descriptor) page must end with a
+ * continuation pointer. The last continuation pointer of the last
+ * page must be zero to mark the end of the chain.
+ */
+#define SGLIST_PER_PAGE	((PAGE_SIZE / sizeof(efi_capsule_block_desc_t)) - 1)
+
+/*
+ * How many scatter gather list (block descriptor) pages do we need
+ * to map @count pages?
+ */
+static inline unsigned int sg_pages_num(unsigned int count)
+{
+	return DIV_ROUND_UP(count, SGLIST_PER_PAGE);
+}
+
+/**
+ * efi_capsule_update_locked - pass a single capsule to the firmware
+ * @capsule: capsule to send to the firmware
+ * @sg_pages: array of scatter gather (block descriptor) pages
+ * @reset: the reset type required for @capsule
+ *
+ * Since this function must be called under capsule_mutex check
+ * whether efi_reset_type will conflict with @reset, and atomically
+ * set it and capsule_pending if a capsule was successfully sent to
+ * the firmware.
+ *
+ * We also check to see if the system is about to restart, and if so,
+ * abort. This avoids races between efi_capsule_update() and
+ * efi_capsule_pending().
+ */
+static int
+efi_capsule_update_locked(efi_capsule_header_t *capsule,
+			  struct page **sg_pages, int reset)
+{
+	efi_physical_addr_t sglist_phys;
+	efi_status_t status;
+
+	lockdep_assert_held(&capsule_mutex);
+
+	/*
+	 * If someone has already registered a capsule that requires a
+	 * different reset type, we're out of luck and must abort.
+	 */
+	if (efi_reset_type >= 0 && efi_reset_type != reset) {
+		pr_err("Conflicting capsule reset type %d (%d).\n",
+		       reset, efi_reset_type);
+		return -EINVAL;
+	}
+
+	/*
+	 * If the system is getting ready to restart it may have
+	 * called efi_capsule_pending() to make decisions (such as
+	 * whether to force an EFI reboot), and we're racing against
+	 * that call. Abort in that case.
+	 */
+	if (unlikely(stop_capsules)) {
+		pr_warn("Capsule update raced with reboot, aborting.\n");
+		return -EINVAL;
+	}
+
+	sglist_phys = page_to_phys(sg_pages[0]);
+
+	status = efi.update_capsule(&capsule, 1, sglist_phys);
+	if (status == EFI_SUCCESS) {
+		capsule_pending = true;
+		efi_reset_type = reset;
+	}
+
+	return efi_status_to_err(status);
+}
+
+/**
+ * efi_capsule_update - send a capsule to the firmware
+ * @capsule: capsule to send to firmware
+ * @pages: an array of capsule data pages
+ *
+ * Build a scatter gather list with EFI capsule block descriptors to
+ * map the capsule described by @capsule with its data in @pages and
+ * send it to the firmware via the UpdateCapsule() runtime service.
+ *
+ * @capsule must be a virtual mapping of the complete capsule update in the
+ * kernel address space, as the capsule can be consumed immediately.
+ * A capsule_header_t that describes the entire contents of the capsule
+ * must be at the start of the first data page.
+ *
+ * Even though this function will validate that the firmware supports
+ * the capsule guid, users will likely want to check that
+ * efi_capsule_supported() returns true before calling this function
+ * because it makes it easier to print helpful error messages.
+ *
+ * If the capsule is successfully submitted to the firmware, any
+ * subsequent calls to efi_capsule_pending() will return true. @pages
+ * must not be released or modified if this function returns
+ * successfully.
+ *
+ * Callers must be prepared for this function to fail, which can
+ * happen if we raced with system reboot or if there is already a
+ * pending capsule that has a reset type that conflicts with the one
+ * required by @capsule. Do NOT use efi_capsule_pending() to detect
+ * this conflict since that would be racy. Instead, submit the capsule
+ * to efi_capsule_update() and check the return value.
+ *
+ * Return 0 on success, a converted EFI status code on failure.
+ */
+int efi_capsule_update(efi_capsule_header_t *capsule, phys_addr_t *pages)
+{
+	u32 imagesize = capsule->imagesize;
+	efi_guid_t guid = capsule->guid;
+	unsigned int count, sg_count;
+	u32 flags = capsule->flags;
+	struct page **sg_pages;
+	int rv, reset_type;
+	int i, j;
+
+	rv = efi_capsule_supported(guid, flags, imagesize, &reset_type);
+	if (rv)
+		return rv;
+
+	count = DIV_ROUND_UP(imagesize, PAGE_SIZE);
+	sg_count = sg_pages_num(count);
+
+	sg_pages = kzalloc_objs(*sg_pages, sg_count);
+	if (!sg_pages)
+		return -ENOMEM;
+
+	for (i = 0; i < sg_count; i++) {
+		sg_pages[i] = alloc_page(GFP_KERNEL);
+		if (!sg_pages[i]) {
+			rv = -ENOMEM;
+			goto out;
+		}
+	}
+
+	for (i = 0; i < sg_count; i++) {
+		efi_capsule_block_desc_t *sglist;
+
+		sglist = kmap_atomic(sg_pages[i]);
+
+		for (j = 0; j < SGLIST_PER_PAGE && count > 0; j++) {
+			u64 sz = min_t(u64, imagesize,
+				       PAGE_SIZE - (u64)*pages % PAGE_SIZE);
+
+			sglist[j].length = sz;
+			sglist[j].data = *pages++;
+
+			imagesize -= sz;
+			count--;
+		}
+
+		/* Continuation pointer */
+		sglist[j].length = 0;
+
+		if (i + 1 == sg_count)
+			sglist[j].data = 0;
+		else
+			sglist[j].data = page_to_phys(sg_pages[i + 1]);
+
+#if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
+		/*
+		 * At runtime, the firmware has no way to find out where the
+		 * sglist elements are mapped, if they are mapped in the first
+		 * place. Therefore, on architectures that can only perform
+		 * cache maintenance by virtual address, the firmware is unable
+		 * to perform this maintenance, and so it is up to the OS to do
+		 * it instead.
+		 */
+		efi_capsule_flush_cache_range(sglist, PAGE_SIZE);
+#endif
+		kunmap_atomic(sglist);
+	}
+
+	mutex_lock(&capsule_mutex);
+	rv = efi_capsule_update_locked(capsule, sg_pages, reset_type);
+	mutex_unlock(&capsule_mutex);
+
+out:
+	for (i = 0; rv && i < sg_count; i++) {
+		if (sg_pages[i])
+			__free_page(sg_pages[i]);
+	}
+
+	kfree(sg_pages);
+	return rv;
+}
+EXPORT_SYMBOL_GPL(efi_capsule_update);
+
+static int capsule_reboot_notify(struct notifier_block *nb, unsigned long event, void *cmd)
+{
+	mutex_lock(&capsule_mutex);
+	stop_capsules = true;
+	mutex_unlock(&capsule_mutex);
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block capsule_reboot_nb = {
+	.notifier_call = capsule_reboot_notify,
+};
+
+static int __init capsule_reboot_register(void)
+{
+	return register_reboot_notifier(&capsule_reboot_nb);
+}
+core_initcall(capsule_reboot_register);