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, 86 insertions, 0 deletions

diff --git a/fs/isofs/util.c b/fs/isofs/util.c
new file mode 100644
index 000000000..42f479da0
--- /dev/null
+++ b/fs/isofs/util.c
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  linux/fs/isofs/util.c
+ */
+
+#include <linux/time.h>
+#include "isofs.h"
+
+/* 
+ * We have to convert from a MM/DD/YY format to the Unix ctime format.
+ * We have to take into account leap years and all of that good stuff.
+ * Unfortunately, the kernel does not have the information on hand to
+ * take into account daylight savings time, but it shouldn't matter.
+ * The time stored should be localtime (with or without DST in effect),
+ * and the timezone offset should hold the offset required to get back
+ * to GMT.  Thus  we should always be correct.
+ */
+
+struct timespec64 iso_date(u8 *p, int flags)
+{
+	int year, month, day, hour, minute, second, tz;
+	struct timespec64 ts;
+
+	if (flags & ISO_DATE_LONG_FORM) {
+		year = (p[0] - '0') * 1000 +
+		       (p[1] - '0') * 100 +
+		       (p[2] - '0') * 10 +
+		       (p[3] - '0') - 1900;
+		month = ((p[4] - '0') * 10 + (p[5] - '0'));
+		day = ((p[6] - '0') * 10 + (p[7] - '0'));
+		hour = ((p[8] - '0') * 10 + (p[9] - '0'));
+		minute = ((p[10] - '0') * 10 + (p[11] - '0'));
+		second = ((p[12] - '0') * 10 + (p[13] - '0'));
+		ts.tv_nsec = ((p[14] - '0') * 10 + (p[15] - '0')) * 10000000;
+		tz = p[16];
+	} else {
+		year = p[0];
+		month = p[1];
+		day = p[2];
+		hour = p[3];
+		minute = p[4];
+		second = p[5];
+		ts.tv_nsec = 0;
+		/* High sierra has no time zone */
+		tz = flags & ISO_DATE_HIGH_SIERRA ? 0 : p[6];
+	}
+
+	if (year < 0) {
+		ts.tv_sec = 0;
+	} else {
+		ts.tv_sec = mktime64(year+1900, month, day, hour, minute, second);
+
+		/* sign extend */
+		if (tz & 0x80)
+			tz |= (-1 << 8);
+
+		/* 
+		 * The timezone offset is unreliable on some disks,
+		 * so we make a sanity check.  In no case is it ever
+		 * more than 13 hours from GMT, which is 52*15min.
+		 * The time is always stored in localtime with the
+		 * timezone offset being what get added to GMT to
+		 * get to localtime.  Thus we need to subtract the offset
+		 * to get to true GMT, which is what we store the time
+		 * as internally.  On the local system, the user may set
+		 * their timezone any way they wish, of course, so GMT
+		 * gets converted back to localtime on the receiving
+		 * system.
+		 *
+		 * NOTE: mkisofs in versions prior to mkisofs-1.10 had
+		 * the sign wrong on the timezone offset.  This has now
+		 * been corrected there too, but if you are getting screwy
+		 * results this may be the explanation.  If enough people
+		 * complain, a user configuration option could be added
+		 * to add the timezone offset in with the wrong sign
+		 * for 'compatibility' with older discs, but I cannot see how
+		 * it will matter that much.
+		 *
+		 * Thanks to kuhlmav@elec.canterbury.ac.nz (Volker Kuhlmann)
+		 * for pointing out the sign error.
+		 */
+		if (-52 <= tz && tz <= 52)
+			ts.tv_sec -= tz * 15 * 60;
+	}
+	return ts;
+}