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, 155 insertions, 0 deletions

diff --git a/lib/glob.c b/lib/glob.c
new file mode 100644
index 000000000..c80d9dd73
--- /dev/null
+++ b/lib/glob.c
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: (GPL-2.0 OR MIT)
+#include <linux/module.h>
+#include <linux/glob.h>
+#include <linux/export.h>
+
+/*
+ * The only reason this code can be compiled as a module is because the
+ * ATA code that depends on it can be as well.  In practice, they're
+ * both usually compiled in and the module overhead goes away.
+ */
+MODULE_DESCRIPTION("glob(7) matching");
+MODULE_LICENSE("Dual MIT/GPL");
+
+static bool __pure glob_match_str(char const *pat, char const *str,
+				  char const *str_end);
+
+/**
+ * glob_match - Shell-style pattern matching, like !fnmatch(pat, str, 0)
+ * @pat: Shell-style pattern to match, e.g. "*.[ch]".
+ * @str: String to match.  The pattern must match the entire string.
+ *
+ * Perform shell-style glob matching, returning true (1) if the match
+ * succeeds, or false (0) if it fails.  Equivalent to !fnmatch(@pat, @str, 0).
+ *
+ * Pattern metacharacters are ?, *, [ and \.
+ * (And, inside character classes, !, - and ].)
+ *
+ * This is a small and simple implementation intended for device denylists
+ * where a string is matched against a number of patterns.  Thus, it
+ * does not preprocess the patterns.  It is non-recursive, and run-time
+ * is at most quadratic: strlen(@str)*strlen(@pat).
+ *
+ * An example of the worst case is glob_match("*aaaaa", "aaaaaaaaaa");
+ * it takes 6 passes over the pattern before matching the string.
+ *
+ * Like !fnmatch(@pat, @str, 0) and unlike the shell, this does NOT
+ * treat / or leading . specially; it isn't actually used for pathnames.
+ *
+ * Note that according to glob(7) (and unlike bash), character classes
+ * are complemented by a leading !; this does not support the regex-style
+ * [^a-z] syntax.
+ *
+ * An opening bracket without a matching close is matched literally.
+ */
+bool __pure glob_match(char const *pat, char const *str)
+{
+	return glob_match_str(pat, str, NULL);
+}
+EXPORT_SYMBOL(glob_match);
+
+/**
+ * glob_match_len - glob match against a length-bounded string
+ * @pat: Shell-style pattern to match.
+ * @str: String to match.  Need not be NUL-terminated.
+ * @len: Number of bytes of @str that may be read.
+ *
+ * Like glob_match(), but @str is only read up to @len bytes, so it can be
+ * used on buffers that are not NUL-terminated (e.g. trace event fields).
+ * A NUL byte within @len still terminates the string.
+ */
+bool __pure glob_match_len(char const *pat, char const *str, size_t len)
+{
+	return glob_match_str(pat, str, str + len);
+}
+EXPORT_SYMBOL(glob_match_len);
+
+static bool __pure glob_match_str(char const *pat, char const *str,
+				  char const *str_end)
+{
+	/*
+	 * Backtrack to previous * on mismatch and retry starting one
+	 * character later in the string.  Because * matches all characters
+	 * (no exception for /), it can be easily proved that there's
+	 * never a need to backtrack multiple levels.
+	 */
+	char const *back_pat = NULL, *back_str = NULL;
+
+	/*
+	 * Loop over each token (character or class) in pat, matching
+	 * it against the remaining unmatched tail of str.  Return false
+	 * on mismatch, or true after matching the trailing nul bytes.
+	 */
+	for (;;) {
+		unsigned char c = (str_end && str >= str_end) ? '\0' : *str;
+		unsigned char d = *pat++;
+
+		str++;
+
+		switch (d) {
+		case '?':	/* Wildcard: anything but nul */
+			if (c == '\0')
+				return false;
+			break;
+		case '*':	/* Any-length wildcard */
+			if (*pat == '\0')	/* Optimize trailing * case */
+				return true;
+			back_pat = pat;
+			back_str = --str;	/* Allow zero-length match */
+			break;
+		case '[': {	/* Character class */
+			if (c == '\0')	/* No possible match */
+				return false;
+			bool match = false, inverted = (*pat == '!');
+			char const *class = inverted ? pat + 1 : pat;
+			unsigned char a = *class++;
+
+			/*
+			 * Iterate over each span in the character class.
+			 * A span is either a single character a, or a
+			 * range a-b.  The first span may begin with ']'.
+			 */
+			do {
+				unsigned char b = a;
+
+				if (a == '\0')	/* Malformed */
+					goto literal;
+
+				if (class[0] == '-' && class[1] != ']') {
+					b = class[1];
+
+					if (b == '\0')
+						goto literal;
+
+					class += 2;
+					/* Any special action if a > b? */
+				}
+				if (a <= c && c <= b)
+					match = true;
+			} while ((a = *class++) != ']');
+
+			if (match == inverted)
+				goto backtrack;
+			pat = class;
+			}
+			break;
+		case '\\':
+			d = *pat++;
+			fallthrough;
+		default:	/* Literal character */
+literal:
+			if (c == d) {
+				if (d == '\0')
+					return true;
+				break;
+			}
+backtrack:
+			if (c == '\0' || !back_pat)
+				return false;	/* No point continuing */
+			/* Try again from last *, one character later in str. */
+			pat = back_pat;
+			str = ++back_str;
+			break;
+		}
+	}
+}