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, 187 insertions, 0 deletions

diff --git a/drivers/md/dm-cache-policy.h b/drivers/md/dm-cache-policy.h
new file mode 100644
index 000000000..a1eedcc42
--- /dev/null
+++ b/drivers/md/dm-cache-policy.h
@@ -0,0 +1,187 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2012 Red Hat. All rights reserved.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_CACHE_POLICY_H
+#define DM_CACHE_POLICY_H
+
+#include "dm-cache-block-types.h"
+
+#include <linux/device-mapper.h>
+
+/*----------------------------------------------------------------*/
+
+/*
+ * The cache policy makes the important decisions about which blocks get to
+ * live on the faster cache device.
+ */
+enum policy_operation {
+	POLICY_PROMOTE,
+	POLICY_DEMOTE,
+	POLICY_WRITEBACK
+};
+
+/*
+ * This is the instruction passed back to the core target.
+ */
+struct policy_work {
+	enum policy_operation op;
+	dm_oblock_t oblock;
+	dm_cblock_t cblock;
+};
+
+/*
+ * The cache policy object.  It is envisaged that this structure will be
+ * embedded in a bigger, policy specific structure (ie. use container_of()).
+ */
+struct dm_cache_policy {
+	/*
+	 * Destroys this object.
+	 */
+	void (*destroy)(struct dm_cache_policy *p);
+
+	/*
+	 * Find the location of a block.
+	 *
+	 * Must not block.
+	 *
+	 * Returns 0 if in cache (cblock will be set), -ENOENT if not, < 0 for
+	 * other errors (-EWOULDBLOCK would be typical).  data_dir should be
+	 * READ or WRITE. fast_copy should be set if migrating this block would
+	 * be 'cheap' somehow (eg, discarded data). background_queued will be set
+	 * if a migration has just been queued.
+	 */
+	int (*lookup)(struct dm_cache_policy *p, dm_oblock_t oblock, dm_cblock_t *cblock,
+		      int data_dir, bool fast_copy, bool *background_queued);
+
+	/*
+	 * Sometimes the core target can optimise a migration, eg, the
+	 * block may be discarded, or the bio may cover an entire block.
+	 * In order to optimise it needs the migration immediately though
+	 * so it knows to do something different with the bio.
+	 *
+	 * This method is optional (policy-internal will fallback to using
+	 * lookup).
+	 */
+	int (*lookup_with_work)(struct dm_cache_policy *p,
+				dm_oblock_t oblock, dm_cblock_t *cblock,
+				int data_dir, bool fast_copy,
+				struct policy_work **work);
+
+	/*
+	 * Retrieves background work.  Returns -ENODATA when there's no
+	 * background work.
+	 */
+	int (*get_background_work)(struct dm_cache_policy *p, bool idle,
+				   struct policy_work **result);
+
+	/*
+	 * You must pass in the same work pointer that you were given, not
+	 * a copy.
+	 */
+	void (*complete_background_work)(struct dm_cache_policy *p,
+					 struct policy_work *work,
+					 bool success);
+
+	void (*set_dirty)(struct dm_cache_policy *p, dm_cblock_t cblock);
+	void (*clear_dirty)(struct dm_cache_policy *p, dm_cblock_t cblock);
+
+	/*
+	 * Called when a cache target is first created.  Used to load a
+	 * mapping from the metadata device into the policy.
+	 */
+	int (*load_mapping)(struct dm_cache_policy *p, dm_oblock_t oblock,
+			    dm_cblock_t cblock, bool dirty,
+			    uint32_t hint, bool hint_valid);
+
+	/*
+	 * Drops the mapping, irrespective of whether it's clean or dirty.
+	 * Returns -ENODATA if cblock is not mapped.
+	 */
+	int (*invalidate_mapping)(struct dm_cache_policy *p, dm_cblock_t cblock);
+
+	/*
+	 * Gets the hint for a given cblock.  Called in a single threaded
+	 * context.  So no locking required.
+	 */
+	uint32_t (*get_hint)(struct dm_cache_policy *p, dm_cblock_t cblock);
+
+	/*
+	 * How full is the cache?
+	 */
+	dm_cblock_t (*residency)(struct dm_cache_policy *p);
+
+	/*
+	 * Because of where we sit in the block layer, we can be asked to
+	 * map a lot of little bios that are all in the same block (no
+	 * queue merging has occurred).  To stop the policy being fooled by
+	 * these, the core target sends regular tick() calls to the policy.
+	 * The policy should only count an entry as hit once per tick.
+	 *
+	 * This method is optional.
+	 */
+	void (*tick)(struct dm_cache_policy *p, bool can_block);
+
+	/*
+	 * Configuration.
+	 */
+	int (*emit_config_values)(struct dm_cache_policy *p, char *result,
+				  unsigned int maxlen, ssize_t *sz_ptr);
+	int (*set_config_value)(struct dm_cache_policy *p,
+				const char *key, const char *value);
+
+	void (*allow_migrations)(struct dm_cache_policy *p, bool allow);
+
+	/*
+	 * Book keeping ptr for the policy register, not for general use.
+	 */
+	void *private;
+};
+
+/*----------------------------------------------------------------*/
+
+/*
+ * We maintain a little register of the different policy types.
+ */
+#define CACHE_POLICY_NAME_SIZE 16
+#define CACHE_POLICY_VERSION_SIZE 3
+
+struct dm_cache_policy_type {
+	/* For use by the register code only. */
+	struct list_head list;
+
+	/*
+	 * Policy writers should fill in these fields.  The name field is
+	 * what gets passed on the target line to select your policy.
+	 */
+	char name[CACHE_POLICY_NAME_SIZE];
+	unsigned int version[CACHE_POLICY_VERSION_SIZE];
+
+	/*
+	 * For use by an alias dm_cache_policy_type to point to the
+	 * real dm_cache_policy_type.
+	 */
+	struct dm_cache_policy_type *real;
+
+	/*
+	 * Policies may store a hint for each cache block.
+	 * Currently the size of this hint must be 0 or 4 bytes but we
+	 * expect to relax this in future.
+	 */
+	size_t hint_size;
+
+	struct module *owner;
+	struct dm_cache_policy *(*create)(dm_cblock_t cache_size,
+					  sector_t origin_size,
+					  sector_t block_size);
+};
+
+int dm_cache_policy_register(struct dm_cache_policy_type *type);
+void dm_cache_policy_unregister(struct dm_cache_policy_type *type);
+
+/*----------------------------------------------------------------*/
+
+#endif	/* DM_CACHE_POLICY_H */