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, 230 insertions, 0 deletions

diff --git a/lib/crypto/mpi/mpi-div.c b/lib/crypto/mpi/mpi-div.c
new file mode 100644
index 000000000..6e5044e72
--- /dev/null
+++ b/lib/crypto/mpi/mpi-div.c
@@ -0,0 +1,230 @@
+/* mpi-div.c  -  MPI functions
+ * Copyright (C) 1994, 1996, 1998, 2001, 2002,
+ *               2003 Free Software Foundation, Inc.
+ *
+ * This file is part of Libgcrypt.
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ *	 Actually it's the same code with only minor changes in the
+ *	 way the data is stored; this is to support the abstraction
+ *	 of an optional secure memory allocation which may be used
+ *	 to avoid revealing of sensitive data due to paging etc.
+ */
+
+#include "mpi-internal.h"
+#include "longlong.h"
+
+int mpi_tdiv_qr(MPI quot, MPI rem, MPI num, MPI den);
+
+int mpi_fdiv_r(MPI rem, MPI dividend, MPI divisor)
+{
+	int divisor_sign = divisor->sign;
+	MPI temp_divisor = NULL;
+	int err;
+
+	/* We need the original value of the divisor after the remainder has been
+	 * preliminary calculated.	We have to copy it to temporary space if it's
+	 * the same variable as REM.
+	 */
+	if (rem == divisor) {
+		temp_divisor = mpi_copy(divisor);
+		if (!temp_divisor)
+			return -ENOMEM;
+		divisor = temp_divisor;
+	}
+
+	err = mpi_tdiv_r(rem, dividend, divisor);
+	if (err)
+		goto free_temp_divisor;
+
+	if (((divisor_sign?1:0) ^ (dividend->sign?1:0)) && rem->nlimbs)
+		err = mpi_add(rem, rem, divisor);
+
+free_temp_divisor:
+	mpi_free(temp_divisor);
+
+	return err;
+}
+
+/* If den == quot, den needs temporary storage.
+ * If den == rem, den needs temporary storage.
+ * If num == quot, num needs temporary storage.
+ * If den has temporary storage, it can be normalized while being copied,
+ *   i.e no extra storage should be allocated.
+ */
+
+int mpi_tdiv_r(MPI rem, MPI num, MPI den)
+{
+	return mpi_tdiv_qr(NULL, rem, num, den);
+}
+
+int mpi_tdiv_qr(MPI quot, MPI rem, MPI num, MPI den)
+{
+	mpi_ptr_t np, dp;
+	mpi_ptr_t qp, rp;
+	mpi_size_t nsize = num->nlimbs;
+	mpi_size_t dsize = den->nlimbs;
+	mpi_size_t qsize, rsize;
+	mpi_size_t sign_remainder = num->sign;
+	mpi_size_t sign_quotient = num->sign ^ den->sign;
+	unsigned int normalization_steps;
+	mpi_limb_t q_limb;
+	mpi_ptr_t marker[5];
+	int markidx = 0;
+	int err;
+
+	/* Ensure space is enough for quotient and remainder.
+	 * We need space for an extra limb in the remainder, because it's
+	 * up-shifted (normalized) below.
+	 */
+	rsize = nsize + 1;
+	err = mpi_resize(rem, rsize);
+	if (err)
+		return err;
+
+	qsize = rsize - dsize;	  /* qsize cannot be bigger than this.	*/
+	if (qsize <= 0) {
+		if (num != rem) {
+			rem->nlimbs = num->nlimbs;
+			rem->sign = num->sign;
+			MPN_COPY(rem->d, num->d, nsize);
+		}
+		if (quot) {
+			/* This needs to follow the assignment to rem, in case the
+			 * numerator and quotient are the same.
+			 */
+			quot->nlimbs = 0;
+			quot->sign = 0;
+		}
+		return 0;
+	}
+
+	if (quot) {
+		err = mpi_resize(quot, qsize);
+		if (err)
+			return err;
+	}
+
+	/* Read pointers here, when reallocation is finished.  */
+	np = num->d;
+	dp = den->d;
+	rp = rem->d;
+
+	/* Optimize division by a single-limb divisor.  */
+	if (dsize == 1) {
+		mpi_limb_t rlimb;
+		if (quot) {
+			qp = quot->d;
+			rlimb = mpihelp_divmod_1(qp, np, nsize, dp[0]);
+			qsize -= qp[qsize - 1] == 0;
+			quot->nlimbs = qsize;
+			quot->sign = sign_quotient;
+		} else
+			rlimb = mpihelp_mod_1(np, nsize, dp[0]);
+		rp[0] = rlimb;
+		rsize = rlimb != 0?1:0;
+		rem->nlimbs = rsize;
+		rem->sign = sign_remainder;
+		return 0;
+	}
+
+	err = -ENOMEM;
+	if (quot) {
+		qp = quot->d;
+		/* Make sure QP and NP point to different objects.  Otherwise the
+		 * numerator would be gradually overwritten by the quotient limbs.
+		 */
+		if (qp == np) { /* Copy NP object to temporary space.  */
+			np = marker[markidx++] = mpi_alloc_limb_space(nsize);
+			if (!np)
+				goto out_free_marker;
+			MPN_COPY(np, qp, nsize);
+		}
+	} else /* Put quotient at top of remainder. */
+		qp = rp + dsize;
+
+	normalization_steps = count_leading_zeros(dp[dsize - 1]);
+
+	/* Normalize the denominator, i.e. make its most significant bit set by
+	 * shifting it NORMALIZATION_STEPS bits to the left.  Also shift the
+	 * numerator the same number of steps (to keep the quotient the same!).
+	 */
+	if (normalization_steps) {
+		mpi_ptr_t tp;
+		mpi_limb_t nlimb;
+
+		/* Shift up the denominator setting the most significant bit of
+		 * the most significant word.  Use temporary storage not to clobber
+		 * the original contents of the denominator.
+		 */
+		tp = marker[markidx++] = mpi_alloc_limb_space(dsize);
+		if (!tp)
+			goto out_free_marker;
+		mpihelp_lshift(tp, dp, dsize, normalization_steps);
+		dp = tp;
+
+		/* Shift up the numerator, possibly introducing a new most
+		 * significant word.  Move the shifted numerator in the remainder
+		 * meanwhile.
+		 */
+		nlimb = mpihelp_lshift(rp, np, nsize, normalization_steps);
+		if (nlimb) {
+			rp[nsize] = nlimb;
+			rsize = nsize + 1;
+		} else
+			rsize = nsize;
+	} else {
+		/* The denominator is already normalized, as required.	Copy it to
+		 * temporary space if it overlaps with the quotient or remainder.
+		 */
+		if (dp == rp || (quot && (dp == qp))) {
+			mpi_ptr_t tp;
+
+			tp = marker[markidx++] = mpi_alloc_limb_space(dsize);
+			if (!tp)
+				goto out_free_marker;
+			MPN_COPY(tp, dp, dsize);
+			dp = tp;
+		}
+
+		/* Move the numerator to the remainder.  */
+		if (rp != np)
+			MPN_COPY(rp, np, nsize);
+
+		rsize = nsize;
+	}
+
+	q_limb = mpihelp_divrem(qp, 0, rp, rsize, dp, dsize);
+
+	if (quot) {
+		qsize = rsize - dsize;
+		if (q_limb) {
+			qp[qsize] = q_limb;
+			qsize += 1;
+		}
+
+		quot->nlimbs = qsize;
+		quot->sign = sign_quotient;
+	}
+
+	rsize = dsize;
+	MPN_NORMALIZE(rp, rsize);
+
+	if (normalization_steps && rsize) {
+		mpihelp_rshift(rp, rp, rsize, normalization_steps);
+		rsize -= rp[rsize - 1] == 0?1:0;
+	}
+
+	rem->nlimbs = rsize;
+	rem->sign	= sign_remainder;
+
+	err = 0;
+
+out_free_marker:
+	while (markidx) {
+		markidx--;
+		mpi_free_limb_space(marker[markidx]);
+	}
+
+	return err;
+}