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, 517 insertions, 0 deletions

diff --git a/lib/crypto/mpi/mpih-div.c b/lib/crypto/mpi/mpih-div.c
new file mode 100644
index 000000000..be70ee2e4
--- /dev/null
+++ b/lib/crypto/mpi/mpih-div.c
@@ -0,0 +1,517 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* mpihelp-div.c  -  MPI helper functions
+ *	Copyright (C) 1994, 1996 Free Software Foundation, Inc.
+ *	Copyright (C) 1998, 1999 Free Software Foundation, Inc.
+ *
+ * This file is part of GnuPG.
+ *
+ * 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.
+ *	 The GNU MP Library itself is published under the LGPL;
+ *	 however I decided to publish this code under the plain GPL.
+ */
+
+#include "mpi-internal.h"
+#include "longlong.h"
+
+#ifndef UMUL_TIME
+#define UMUL_TIME 1
+#endif
+#ifndef UDIV_TIME
+#define UDIV_TIME UMUL_TIME
+#endif
+
+
+mpi_limb_t
+mpihelp_mod_1(mpi_ptr_t dividend_ptr, mpi_size_t dividend_size,
+			mpi_limb_t divisor_limb)
+{
+	mpi_size_t i;
+	mpi_limb_t n1, n0, r;
+	mpi_limb_t dummy __maybe_unused;
+
+	/* Botch: Should this be handled at all?  Rely on callers?	*/
+	if (!dividend_size)
+		return 0;
+
+	/* If multiplication is much faster than division, and the
+	 * dividend is large, pre-invert the divisor, and use
+	 * only multiplications in the inner loop.
+	 *
+	 * This test should be read:
+	 *	 Does it ever help to use udiv_qrnnd_preinv?
+	 *	   && Does what we save compensate for the inversion overhead?
+	 */
+	if (UDIV_TIME > (2 * UMUL_TIME + 6)
+			&& (UDIV_TIME - (2 * UMUL_TIME + 6)) * dividend_size > UDIV_TIME) {
+		int normalization_steps;
+
+		normalization_steps = count_leading_zeros(divisor_limb);
+		if (normalization_steps) {
+			mpi_limb_t divisor_limb_inverted;
+
+			divisor_limb <<= normalization_steps;
+
+			/* Compute (2**2N - 2**N * DIVISOR_LIMB) / DIVISOR_LIMB.  The
+			 * result is a (N+1)-bit approximation to 1/DIVISOR_LIMB, with the
+			 * most significant bit (with weight 2**N) implicit.
+			 *
+			 * Special case for DIVISOR_LIMB == 100...000.
+			 */
+			if (!(divisor_limb << 1))
+				divisor_limb_inverted = ~(mpi_limb_t)0;
+			else
+				udiv_qrnnd(divisor_limb_inverted, dummy,
+						-divisor_limb, 0, divisor_limb);
+
+			n1 = dividend_ptr[dividend_size - 1];
+			r = n1 >> (BITS_PER_MPI_LIMB - normalization_steps);
+
+			/* Possible optimization:
+			 * if (r == 0
+			 * && divisor_limb > ((n1 << normalization_steps)
+			 *		       | (dividend_ptr[dividend_size - 2] >> ...)))
+			 * ...one division less...
+			 */
+			for (i = dividend_size - 2; i >= 0; i--) {
+				n0 = dividend_ptr[i];
+				UDIV_QRNND_PREINV(dummy, r, r,
+						((n1 << normalization_steps)
+						 | (n0 >> (BITS_PER_MPI_LIMB - normalization_steps))),
+						divisor_limb, divisor_limb_inverted);
+				n1 = n0;
+			}
+			UDIV_QRNND_PREINV(dummy, r, r,
+					n1 << normalization_steps,
+					divisor_limb, divisor_limb_inverted);
+			return r >> normalization_steps;
+		} else {
+			mpi_limb_t divisor_limb_inverted;
+
+			/* Compute (2**2N - 2**N * DIVISOR_LIMB) / DIVISOR_LIMB.  The
+			 * result is a (N+1)-bit approximation to 1/DIVISOR_LIMB, with the
+			 * most significant bit (with weight 2**N) implicit.
+			 *
+			 * Special case for DIVISOR_LIMB == 100...000.
+			 */
+			if (!(divisor_limb << 1))
+				divisor_limb_inverted = ~(mpi_limb_t)0;
+			else
+				udiv_qrnnd(divisor_limb_inverted, dummy,
+						-divisor_limb, 0, divisor_limb);
+
+			i = dividend_size - 1;
+			r = dividend_ptr[i];
+
+			if (r >= divisor_limb)
+				r = 0;
+			else
+				i--;
+
+			for ( ; i >= 0; i--) {
+				n0 = dividend_ptr[i];
+				UDIV_QRNND_PREINV(dummy, r, r,
+						n0, divisor_limb, divisor_limb_inverted);
+			}
+			return r;
+		}
+	} else {
+		if (UDIV_NEEDS_NORMALIZATION) {
+			int normalization_steps;
+
+			normalization_steps = count_leading_zeros(divisor_limb);
+			if (normalization_steps) {
+				divisor_limb <<= normalization_steps;
+
+				n1 = dividend_ptr[dividend_size - 1];
+				r = n1 >> (BITS_PER_MPI_LIMB - normalization_steps);
+
+				/* Possible optimization:
+				 * if (r == 0
+				 * && divisor_limb > ((n1 << normalization_steps)
+				 *		   | (dividend_ptr[dividend_size - 2] >> ...)))
+				 * ...one division less...
+				 */
+				for (i = dividend_size - 2; i >= 0; i--) {
+					n0 = dividend_ptr[i];
+					udiv_qrnnd(dummy, r, r,
+						((n1 << normalization_steps)
+						 | (n0 >> (BITS_PER_MPI_LIMB - normalization_steps))),
+						divisor_limb);
+					n1 = n0;
+				}
+				udiv_qrnnd(dummy, r, r,
+						n1 << normalization_steps,
+						divisor_limb);
+				return r >> normalization_steps;
+			}
+		}
+		/* No normalization needed, either because udiv_qrnnd doesn't require
+		 * it, or because DIVISOR_LIMB is already normalized.
+		 */
+		i = dividend_size - 1;
+		r = dividend_ptr[i];
+
+		if (r >= divisor_limb)
+			r = 0;
+		else
+			i--;
+
+		for (; i >= 0; i--) {
+			n0 = dividend_ptr[i];
+			udiv_qrnnd(dummy, r, r, n0, divisor_limb);
+		}
+		return r;
+	}
+}
+
+/* Divide num (NP/NSIZE) by den (DP/DSIZE) and write
+ * the NSIZE-DSIZE least significant quotient limbs at QP
+ * and the DSIZE long remainder at NP.	If QEXTRA_LIMBS is
+ * non-zero, generate that many fraction bits and append them after the
+ * other quotient limbs.
+ * Return the most significant limb of the quotient, this is always 0 or 1.
+ *
+ * Preconditions:
+ * 0. NSIZE >= DSIZE.
+ * 1. The most significant bit of the divisor must be set.
+ * 2. QP must either not overlap with the input operands at all, or
+ *    QP + DSIZE >= NP must hold true.	(This means that it's
+ *    possible to put the quotient in the high part of NUM, right after the
+ *    remainder in NUM.
+ * 3. NSIZE >= DSIZE, even if QEXTRA_LIMBS is non-zero.
+ */
+
+mpi_limb_t
+mpihelp_divrem(mpi_ptr_t qp, mpi_size_t qextra_limbs,
+	       mpi_ptr_t np, mpi_size_t nsize, mpi_ptr_t dp, mpi_size_t dsize)
+{
+	mpi_limb_t most_significant_q_limb = 0;
+
+	switch (dsize) {
+	case 0:
+		/* We are asked to divide by zero, so go ahead and do it!  (To make
+		   the compiler not remove this statement, return the value.)  */
+		/*
+		 * existing clients of this function have been modified
+		 * not to call it with dsize == 0, so this should not happen
+		 */
+		return 1 / dsize;
+
+	case 1:
+		{
+			mpi_size_t i;
+			mpi_limb_t n1;
+			mpi_limb_t d;
+
+			d = dp[0];
+			n1 = np[nsize - 1];
+
+			if (n1 >= d) {
+				n1 -= d;
+				most_significant_q_limb = 1;
+			}
+
+			qp += qextra_limbs;
+			for (i = nsize - 2; i >= 0; i--)
+				udiv_qrnnd(qp[i], n1, n1, np[i], d);
+			qp -= qextra_limbs;
+
+			for (i = qextra_limbs - 1; i >= 0; i--)
+				udiv_qrnnd(qp[i], n1, n1, 0, d);
+
+			np[0] = n1;
+		}
+		break;
+
+	case 2:
+		{
+			mpi_size_t i;
+			mpi_limb_t n1, n0, n2;
+			mpi_limb_t d1, d0;
+
+			np += nsize - 2;
+			d1 = dp[1];
+			d0 = dp[0];
+			n1 = np[1];
+			n0 = np[0];
+
+			if (n1 >= d1 && (n1 > d1 || n0 >= d0)) {
+				sub_ddmmss(n1, n0, n1, n0, d1, d0);
+				most_significant_q_limb = 1;
+			}
+
+			for (i = qextra_limbs + nsize - 2 - 1; i >= 0; i--) {
+				mpi_limb_t q;
+				mpi_limb_t r;
+
+				if (i >= qextra_limbs)
+					np--;
+				else
+					np[0] = 0;
+
+				if (n1 == d1) {
+					/* Q should be either 111..111 or 111..110.  Need special
+					 * treatment of this rare case as normal division would
+					 * give overflow.  */
+					q = ~(mpi_limb_t) 0;
+
+					r = n0 + d1;
+					if (r < d1) {	/* Carry in the addition? */
+						add_ssaaaa(n1, n0, r - d0,
+							   np[0], 0, d0);
+						qp[i] = q;
+						continue;
+					}
+					n1 = d0 - (d0 != 0 ? 1 : 0);
+					n0 = -d0;
+				} else {
+					udiv_qrnnd(q, r, n1, n0, d1);
+					umul_ppmm(n1, n0, d0, q);
+				}
+
+				n2 = np[0];
+q_test:
+				if (n1 > r || (n1 == r && n0 > n2)) {
+					/* The estimated Q was too large.  */
+					q--;
+					sub_ddmmss(n1, n0, n1, n0, 0, d0);
+					r += d1;
+					if (r >= d1)	/* If not carry, test Q again.  */
+						goto q_test;
+				}
+
+				qp[i] = q;
+				sub_ddmmss(n1, n0, r, n2, n1, n0);
+			}
+			np[1] = n1;
+			np[0] = n0;
+		}
+		break;
+
+	default:
+		{
+			mpi_size_t i;
+			mpi_limb_t dX, d1, n0;
+
+			np += nsize - dsize;
+			dX = dp[dsize - 1];
+			d1 = dp[dsize - 2];
+			n0 = np[dsize - 1];
+
+			if (n0 >= dX) {
+				if (n0 > dX
+				    || mpihelp_cmp(np, dp, dsize - 1) >= 0) {
+					mpihelp_sub_n(np, np, dp, dsize);
+					n0 = np[dsize - 1];
+					most_significant_q_limb = 1;
+				}
+			}
+
+			for (i = qextra_limbs + nsize - dsize - 1; i >= 0; i--) {
+				mpi_limb_t q;
+				mpi_limb_t n1, n2;
+				mpi_limb_t cy_limb;
+
+				if (i >= qextra_limbs) {
+					np--;
+					n2 = np[dsize];
+				} else {
+					n2 = np[dsize - 1];
+					MPN_COPY_DECR(np + 1, np, dsize - 1);
+					np[0] = 0;
+				}
+
+				if (n0 == dX) {
+					/* This might over-estimate q, but it's probably not worth
+					 * the extra code here to find out.  */
+					q = ~(mpi_limb_t) 0;
+				} else {
+					mpi_limb_t r;
+
+					udiv_qrnnd(q, r, n0, np[dsize - 1], dX);
+					umul_ppmm(n1, n0, d1, q);
+
+					while (n1 > r
+					       || (n1 == r
+						   && n0 > np[dsize - 2])) {
+						q--;
+						r += dX;
+						if (r < dX)	/* I.e. "carry in previous addition?" */
+							break;
+						n1 -= n0 < d1;
+						n0 -= d1;
+					}
+				}
+
+				/* Possible optimization: We already have (q * n0) and (1 * n1)
+				 * after the calculation of q.  Taking advantage of that, we
+				 * could make this loop make two iterations less.  */
+				cy_limb = mpihelp_submul_1(np, dp, dsize, q);
+
+				if (n2 != cy_limb) {
+					mpihelp_add_n(np, np, dp, dsize);
+					q--;
+				}
+
+				qp[i] = q;
+				n0 = np[dsize - 1];
+			}
+		}
+	}
+
+	return most_significant_q_limb;
+}
+
+/****************
+ * Divide (DIVIDEND_PTR,,DIVIDEND_SIZE) by DIVISOR_LIMB.
+ * Write DIVIDEND_SIZE limbs of quotient at QUOT_PTR.
+ * Return the single-limb remainder.
+ * There are no constraints on the value of the divisor.
+ *
+ * QUOT_PTR and DIVIDEND_PTR might point to the same limb.
+ */
+
+mpi_limb_t
+mpihelp_divmod_1(mpi_ptr_t quot_ptr,
+		mpi_ptr_t dividend_ptr, mpi_size_t dividend_size,
+		mpi_limb_t divisor_limb)
+{
+	mpi_size_t i;
+	mpi_limb_t n1, n0, r;
+	mpi_limb_t dummy __maybe_unused;
+
+	if (!dividend_size)
+		return 0;
+
+	/* If multiplication is much faster than division, and the
+	 * dividend is large, pre-invert the divisor, and use
+	 * only multiplications in the inner loop.
+	 *
+	 * This test should be read:
+	 * Does it ever help to use udiv_qrnnd_preinv?
+	 * && Does what we save compensate for the inversion overhead?
+	 */
+	if (UDIV_TIME > (2 * UMUL_TIME + 6)
+			&& (UDIV_TIME - (2 * UMUL_TIME + 6)) * dividend_size > UDIV_TIME) {
+		int normalization_steps;
+
+		normalization_steps = count_leading_zeros(divisor_limb);
+		if (normalization_steps) {
+			mpi_limb_t divisor_limb_inverted;
+
+			divisor_limb <<= normalization_steps;
+
+			/* Compute (2**2N - 2**N * DIVISOR_LIMB) / DIVISOR_LIMB.  The
+			 * result is a (N+1)-bit approximation to 1/DIVISOR_LIMB, with the
+			 * most significant bit (with weight 2**N) implicit.
+			 */
+			/* Special case for DIVISOR_LIMB == 100...000.  */
+			if (!(divisor_limb << 1))
+				divisor_limb_inverted = ~(mpi_limb_t)0;
+			else
+				udiv_qrnnd(divisor_limb_inverted, dummy,
+						-divisor_limb, 0, divisor_limb);
+
+			n1 = dividend_ptr[dividend_size - 1];
+			r = n1 >> (BITS_PER_MPI_LIMB - normalization_steps);
+
+			/* Possible optimization:
+			 * if (r == 0
+			 * && divisor_limb > ((n1 << normalization_steps)
+			 *		       | (dividend_ptr[dividend_size - 2] >> ...)))
+			 * ...one division less...
+			 */
+			for (i = dividend_size - 2; i >= 0; i--) {
+				n0 = dividend_ptr[i];
+				UDIV_QRNND_PREINV(quot_ptr[i + 1], r, r,
+						((n1 << normalization_steps)
+						 | (n0 >> (BITS_PER_MPI_LIMB - normalization_steps))),
+						divisor_limb, divisor_limb_inverted);
+				n1 = n0;
+			}
+			UDIV_QRNND_PREINV(quot_ptr[0], r, r,
+					n1 << normalization_steps,
+					divisor_limb, divisor_limb_inverted);
+			return r >> normalization_steps;
+		} else {
+			mpi_limb_t divisor_limb_inverted;
+
+			/* Compute (2**2N - 2**N * DIVISOR_LIMB) / DIVISOR_LIMB.  The
+			 * result is a (N+1)-bit approximation to 1/DIVISOR_LIMB, with the
+			 * most significant bit (with weight 2**N) implicit.
+			 */
+			/* Special case for DIVISOR_LIMB == 100...000.  */
+			if (!(divisor_limb << 1))
+				divisor_limb_inverted = ~(mpi_limb_t) 0;
+			else
+				udiv_qrnnd(divisor_limb_inverted, dummy,
+						-divisor_limb, 0, divisor_limb);
+
+			i = dividend_size - 1;
+			r = dividend_ptr[i];
+
+			if (r >= divisor_limb)
+				r = 0;
+			else
+				quot_ptr[i--] = 0;
+
+			for ( ; i >= 0; i--) {
+				n0 = dividend_ptr[i];
+				UDIV_QRNND_PREINV(quot_ptr[i], r, r,
+						n0, divisor_limb, divisor_limb_inverted);
+			}
+			return r;
+		}
+	} else {
+		if (UDIV_NEEDS_NORMALIZATION) {
+			int normalization_steps;
+
+			normalization_steps = count_leading_zeros(divisor_limb);
+			if (normalization_steps) {
+				divisor_limb <<= normalization_steps;
+
+				n1 = dividend_ptr[dividend_size - 1];
+				r = n1 >> (BITS_PER_MPI_LIMB - normalization_steps);
+
+				/* Possible optimization:
+				 * if (r == 0
+				 * && divisor_limb > ((n1 << normalization_steps)
+				 *		   | (dividend_ptr[dividend_size - 2] >> ...)))
+				 * ...one division less...
+				 */
+				for (i = dividend_size - 2; i >= 0; i--) {
+					n0 = dividend_ptr[i];
+					udiv_qrnnd(quot_ptr[i + 1], r, r,
+						((n1 << normalization_steps)
+						 | (n0 >> (BITS_PER_MPI_LIMB - normalization_steps))),
+						divisor_limb);
+					n1 = n0;
+				}
+				udiv_qrnnd(quot_ptr[0], r, r,
+						n1 << normalization_steps,
+						divisor_limb);
+				return r >> normalization_steps;
+			}
+		}
+		/* No normalization needed, either because udiv_qrnnd doesn't require
+		 * it, or because DIVISOR_LIMB is already normalized.
+		 */
+		i = dividend_size - 1;
+		r = dividend_ptr[i];
+
+		if (r >= divisor_limb)
+			r = 0;
+		else
+			quot_ptr[i--] = 0;
+
+		for (; i >= 0; i--) {
+			n0 = dividend_ptr[i];
+			udiv_qrnnd(quot_ptr[i], r, r, n0, divisor_limb);
+		}
+		return r;
+	}
+}