[PATCH v11 07/12] x86: Add emulation code for UMIP instructions

[Ricardo Neri]

The feature User-Mode Instruction Prevention present in recent Intel
processor prevents a group of instructions (sgdt, sidt, sldt, smsw, and
str) from being executed with CPL > 0. Otherwise, a general protection
fault is issued.

Rather than relaying to the user space the general protection fault caused
by the UMIP-protected instructions (in the form of a SIGSEGV signal), it
can be trapped and the instruction emulated to provide a dummy result.
This allows to both conserve the current kernel behavior and not reveal the
system resources that UMIP intends to protect (i.e., the locations of the
global descriptor and interrupt descriptor tables, the segment selectors of
the local descriptor table, the value of the task state register and the
contents of the CR0 register).

This emulation is needed because certain applications (e.g., WineHQ and
DOSEMU2) rely on this subset of instructions to function. Given that sldt
and str are not commonly used in programs that run on WineHQ or DOSEMU2,
they are not emulated. Also, emulation is provided only for 32-bit
processes; 64-bit processes that attempt to use the instructions that UMIP
protects will receive the SIGSEGV signal issued as a consequence of the
general protection fault.

The instructions protected by UMIP can be split in two groups. Those which
return a kernel memory address (sgdt and sidt) and those which return a
value (smsw, sldt and str; the last two not emulated).

For the instructions that return a kernel memory address, applications such
as WineHQ rely on the result being located in the kernel memory space, not
the actual location of the table. The result is emulated as a hard-coded
value that lies close to the top of the kernel memory. The limit for the
GDT and the IDT are set to zero.

The instruction smsw is emulated to return the value that the register CR0
has at boot time as set in the head_32.

Care is taken to appropriately emulate the results when segmentation is
used. That is, rather than relying on USER_DS and USER_CS, the function
insn_get_addr_ref() inspects the segment descriptor pointed by the
registers in pt_regs. This ensures that we correctly obtain the segment
base address and the address and operand sizes even if the user space
application uses a local descriptor table.

Cc: Andy Lutomirski <luto@xxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: H. Peter Anvin <hpa@xxxxxxxxx>
Cc: Borislav Petkov <bp@xxxxxxx>
Cc: Brian Gerst <brgerst@xxxxxxxxx>
Cc: Chen Yucong <slaoub@xxxxxxxxx>
Cc: Chris Metcalf <cmetcalf@xxxxxxxxxxxx>
Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
Cc: Fenghua Yu <fenghua.yu@xxxxxxxxx>
Cc: Huang Rui <ray.huang@xxxxxxx>
Cc: Jiri Slaby <jslaby@xxxxxxx>
Cc: Jonathan Corbet <corbet@xxxxxxx>
Cc: Michael S. Tsirkin <mst@xxxxxxxxxx>
Cc: Paul Gortmaker <paul.gortmaker@xxxxxxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Ravi V. Shankar <ravi.v.shankar@xxxxxxxxx>
Cc: Shuah Khan <shuah@xxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Cc: Tony Luck <tony.luck@xxxxxxxxx>
Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
Cc: x86@xxxxxxxxxx
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
---
 arch/x86/include/asm/umip.h |  12 ++
 arch/x86/kernel/Makefile    |   1 +
 arch/x86/kernel/umip.c      | 321 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 334 insertions(+)
 create mode 100644 arch/x86/include/asm/umip.h
 create mode 100644 arch/x86/kernel/umip.c

diff --git a/arch/x86/include/asm/umip.h b/arch/x86/include/asm/umip.h
new file mode 100644
index 0000000..db43f2a
--- /dev/null
+++ b/arch/x86/include/asm/umip.h
@@ -0,0 +1,12 @@
+#ifndef _ASM_X86_UMIP_H
+#define _ASM_X86_UMIP_H
+
+#include <linux/types.h>
+#include <asm/ptrace.h>
+
+#ifdef CONFIG_X86_INTEL_UMIP
+bool fixup_umip_exception(struct pt_regs *regs);
+#else
+static inline bool fixup_umip_exception(struct pt_regs *regs) { return false; }
+#endif  /* CONFIG_X86_INTEL_UMIP */
+#endif  /* _ASM_X86_UMIP_H */
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 295abaa..81bb565 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -127,6 +127,7 @@ obj-$(CONFIG_EFI)			+= sysfb_efi.o
 obj-$(CONFIG_PERF_EVENTS)		+= perf_regs.o
 obj-$(CONFIG_TRACING)			+= tracepoint.o
 obj-$(CONFIG_SCHED_MC_PRIO)		+= itmt.o
+obj-$(CONFIG_X86_INTEL_UMIP)		+= umip.o
 
 obj-$(CONFIG_UNWINDER_ORC)		+= unwind_orc.o
 obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
diff --git a/arch/x86/kernel/umip.c b/arch/x86/kernel/umip.c
new file mode 100644
index 0000000..d80b816
--- /dev/null
+++ b/arch/x86/kernel/umip.c
@@ -0,0 +1,321 @@
+/*
+ * umip.c Emulation for instruction protected by the Intel User-Mode
+ * Instruction Prevention feature
+ *
+ * Copyright (c) 2017, Intel Corporation.
+ * Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
+ */
+
+#include <linux/uaccess.h>
+#include <asm/umip.h>
+#include <asm/traps.h>
+#include <asm/insn.h>
+#include <asm/insn-eval.h>
+
+/** DOC: Emulation for User-Mode Instruction Prevention (UMIP)
+ *
+ * The feature User-Mode Instruction Prevention present in recent Intel
+ * processor prevents a group of instructions (sgdt, sidt, sldt, smsw, and str)
+ * from being executed with CPL > 0. Otherwise, a general protection fault is
+ * issued.
+ *
+ * Rather than relaying to the user space the general protection fault caused by
+ * the UMIP-protected instructions (in the form of a SIGSEGV signal), it can be
+ * trapped and emulate the result of such instructions to provide dummy values.
+ * This allows to both conserve the current kernel behavior and not reveal the
+ * system resources that UMIP intends to protect (i.e., the locations of the
+ * global descriptor and interrupt descriptor tables, the segment selectors of
+ * the local descriptor table, the value of the task state register and the
+ * contents of the CR0 register).
+ *
+ * This emulation is needed because certain applications (e.g., WineHQ and
+ * DOSEMU2) rely on this subset of instructions to function.
+ *
+ * The instructions protected by UMIP can be split in two groups. Those which
+ * return a kernel memory address (sgdt and sidt) and those which return a
+ * value (sldt, str and smsw).
+ *
+ * For the instructions that return a kernel memory address, applications
+ * such as WineHQ rely on the result being located in the kernel memory space,
+ * not the actual location of the table. The result is emulated as a hard-coded
+ * value that, lies close to the top of the kernel memory. The limit for the GDT
+ * and the IDT are set to zero.
+ *
+ * Given that sldt and str are not commonly used in programs that run on WineHQ
+ * or DOSEMU2, they are not emulated.
+ *
+ * The instruction smsw is emulated to return the value that the register CR0
+ * has at boot time as set in the head_32.
+ *
+ * Also, emulation is provided only for 32-bit processes; 64-bit processes
+ * that attempt to use the instructions that UMIP protects will receive the
+ * SIGSEGV signal issued as a consequence of the general protection fault.
+ *
+ * Care is taken to appropriately emulate the results when segmentation is
+ * used. That is, rather than relying on USER_DS and USER_CS, the function
+ * insn_get_addr_ref() inspects the segment descriptor pointed by the
+ * registers in pt_regs. This ensures that we correctly obtain the segment
+ * base address and the address and operand sizes even if the user space
+ * application uses a local descriptor table.
+ */
+
+#define UMIP_DUMMY_GDT_BASE 0xfffe0000
+#define UMIP_DUMMY_IDT_BASE 0xffff0000
+
+/*
+ * The SGDT and SIDT instructions store the contents of the global descriptor
+ * table and interrupt table registers, respectively. The destination is a
+ * memory operand of X+2 bytes. X bytes are used to store the base address of
+ * the table and 2 bytes are used to store the limit. In 32-bit processes, the
+ * only processes for which emulation is provided, X has a value of 4.
+ */
+#define UMIP_GDT_IDT_BASE_SIZE 4
+#define UMIP_GDT_IDT_LIMIT_SIZE 2
+
+#define	UMIP_INST_SGDT	0	/* 0F 01 /0 */
+#define	UMIP_INST_SIDT	1	/* 0F 01 /1 */
+#define	UMIP_INST_SMSW	3	/* 0F 01 /4 */
+
+/**
+ * identify_insn() - Identify a UMIP-protected instruction
+ * @insn:	Instruction structure with opcode and ModRM byte.
+ *
+ * From the opcode and ModRM.reg in @insn identify, if any, a UMIP-protected
+ * instruction that can be emulated.
+ *
+ * Returns:
+ *
+ * On success, a constant identifying a specific UMIP-protected instruction that
+ * can be emulated.
+ *
+ * -EINVAL on error or when not an UMIP-protected instruction that can be
+ * emulated.
+ */
+static int identify_insn(struct insn *insn)
+{
+	/* By getting modrm we also get the opcode. */
+	insn_get_modrm(insn);
+
+	if (!insn->modrm.nbytes)
+		return -EINVAL;
+
+	/* All the instructions of interest start with 0x0f. */
+	if (insn->opcode.bytes[0] != 0xf)
+		return -EINVAL;
+
+	if (insn->opcode.bytes[1] == 0x1) {
+		switch (X86_MODRM_REG(insn->modrm.value)) {
+		case 0:
+			return UMIP_INST_SGDT;
+		case 1:
+			return UMIP_INST_SIDT;
+		case 4:
+			return UMIP_INST_SMSW;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	/* SLDT AND STR are not emulated */
+	return -EINVAL;
+}
+
+/**
+ * emulate_umip_insn() - Emulate UMIP instructions and return dummy values
+ * @insn:	Instruction structure with operands
+ * @umip_inst:	A constant indicating the instruction to emulate
+ * @data:	Buffer into which the dummy result is stored
+ * @data_size:	Size of the emulated result
+ *
+ * Emulate an instruction protected by UMIP and provide a dummy result. The
+ * result of the emulation is saved in @data. The size of the results depends
+ * on both the instruction and type of operand (register vs memory address).
+ * The size of the result is updated in @data_size. Caller is responsible
+ * of providing a @data buffer of at least UMIP_GDT_IDT_BASE_SIZE +
+ * UMIP_GDT_IDT_LIMIT_SIZE bytes.
+ *
+ * Returns:
+ *
+ * 0 on success, -EINVAL on error while emulating.
+ */
+static int emulate_umip_insn(struct insn *insn, int umip_inst,
+			     unsigned char *data, int *data_size)
+{
+	unsigned long dummy_base_addr, dummy_value;
+	unsigned short dummy_limit = 0;
+
+	if (!data || !data_size || !insn)
+		return -EINVAL;
+	/*
+	 * These two instructions return the base address and limit of the
+	 * global and interrupt descriptor table, respectively. According to the
+	 * Intel Software Development manual, the base address can be 24-bit,
+	 * 32-bit or 64-bit. Limit is always 16-bit. If the operand size is
+	 * 16-bit, the returned value of the base address is supposed to be a
+	 * zero-extended 24-byte number. However, it seems that a 32-byte number
+	 * is always returned irrespective of the operand size.
+	 */
+	if (umip_inst == UMIP_INST_SGDT || umip_inst == UMIP_INST_SIDT) {
+		/* SGDT and SIDT do not use registers operands. */
+		if (X86_MODRM_MOD(insn->modrm.value) == 3)
+			return -EINVAL;
+
+		if (umip_inst == UMIP_INST_SGDT)
+			dummy_base_addr = UMIP_DUMMY_GDT_BASE;
+		else
+			dummy_base_addr = UMIP_DUMMY_IDT_BASE;
+
+		*data_size = UMIP_GDT_IDT_LIMIT_SIZE + UMIP_GDT_IDT_BASE_SIZE;
+
+		memcpy(data + 2, &dummy_base_addr, UMIP_GDT_IDT_BASE_SIZE);
+		memcpy(data, &dummy_limit, UMIP_GDT_IDT_LIMIT_SIZE);
+
+	} else if (umip_inst == UMIP_INST_SMSW) {
+		dummy_value = CR0_STATE;
+
+		/*
+		 * Even though the CR0 register has 4 bytes, the number
+		 * of bytes to be copied in the result buffer is determined
+		 * by whether the operand is a register or a memory location.
+		 * If operand is a register, return as many bytes as the operand
+		 * size. If operand is memory, return only the two least
+		 * siginificant bytes of CR0.
+		 */
+		if (X86_MODRM_MOD(insn->modrm.value) == 3)
+			*data_size = insn->opnd_bytes;
+		else
+			*data_size = 2;
+
+		memcpy(data, &dummy_value, *data_size);
+	/* STR and SLDT  are not emulated */
+	} else {
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+/**
+ * fixup_umip_exception() - Fixup a general protection fault caused by UMIP
+ * @regs:	Registers as saved when entering the #GP handler
+ *
+ * The instructions sgdt, sidt, str, smsw, sldt cause a general protection
+ * fault if executed with CPL > 0 (i.e., from user space). If the offending
+ * user-space process is not in long mode, this function fixes the exception
+ * up and provides dummy results for sgdt, sidt and smsw; str and sldt are not
+ * fixed up. Also long mode user-space processes are not fixed up.
+ *
+ * If operands are memory addresses, results are copied to user-space memory as
+ * indicated by the instruction pointed by eIP using the registers indicated in
+ * the instruction operands. If operands are registers, results are copied into
+ * the context that was saved when entering kernel mode.
+ *
+ * Returns:
+ *
+ * True if emulation was successful; false if not.
+ */
+bool fixup_umip_exception(struct pt_regs *regs)
+{
+	int not_copied, nr_copied, reg_offset, dummy_data_size, umip_inst;
+	unsigned long seg_base = 0, *reg_addr;
+	/* 10 bytes is the maximum size of the result of UMIP instructions */
+	unsigned char dummy_data[10] = { 0 };
+	unsigned char buf[MAX_INSN_SIZE];
+	void __user *uaddr;
+	struct insn insn;
+	char seg_defs;
+
+	if (!regs)
+		return false;
+
+	/* Do not emulate 64-bit processes. */
+	if (user_64bit_mode(regs))
+		return false;
+
+	/*
+	 * If not in user-space long mode, a custom code segment could be in
+	 * use. This is true in protected mode (if the process defined a local
+	 * descriptor table), or virtual-8086 mode. In most of the cases
+	 * seg_base will be zero as in USER_CS.
+	 */
+	if (!user_64bit_mode(regs))
+		seg_base = insn_get_seg_base(regs, INAT_SEG_REG_CS);
+
+	if (seg_base == -1L)
+		return false;
+
+	not_copied = copy_from_user(buf, (void __user *)(seg_base + regs->ip),
+				    sizeof(buf));
+	nr_copied = sizeof(buf) - not_copied;
+
+	/*
+	 * The copy_from_user above could have failed if user code is protected
+	 * by a memory protection key. Give up on emulation in such a case.
+	 * Should we issue a page fault?
+	 */
+	if (!nr_copied)
+		return false;
+
+	insn_init(&insn, buf, nr_copied, user_64bit_mode(regs));
+
+	/*
+	 * Override the default operand and address sizes with what is specified
+	 * in the code segment descriptor. The instruction decoder only sets
+	 * the address size it to either 4 or 8 address bytes and does nothing
+	 * for the operand bytes. This OK for most of the cases, but we could
+	 * have special cases where, for instance, a 16-bit code segment
+	 * descriptor is used.
+	 * If there is an address override prefix, the instruction decoder
+	 * correctly updates these values, even for 16-bit defaults.
+	 */
+	seg_defs = insn_get_code_seg_params(regs);
+	if (seg_defs == -EINVAL)
+		return false;
+
+	insn.addr_bytes = INSN_CODE_SEG_ADDR_SZ(seg_defs);
+	insn.opnd_bytes = INSN_CODE_SEG_OPND_SZ(seg_defs);
+
+	insn_get_length(&insn);
+	if (nr_copied < insn.length)
+		return false;
+
+	umip_inst = identify_insn(&insn);
+	if (umip_inst < 0)
+		return false;
+
+	if (emulate_umip_insn(&insn, umip_inst, dummy_data, &dummy_data_size))
+		return false;
+
+	/*
+	 * If operand is a register, write result to the copy of the register
+	 * value that was pushed to the stack when entering into kernel mode.
+	 * Upon exit, the value we write will be restored to the actual hardware
+	 * register.
+	 */
+	if (X86_MODRM_MOD(insn.modrm.value) == 3) {
+		reg_offset = insn_get_modrm_rm_off(&insn, regs);
+
+		/*
+		 * Negative values are usually errors. In memory addressing,
+		 * the exception is -EDOM. Since we expect a register operand,
+		 * all negative values are errors.
+		 */
+		if (reg_offset < 0)
+			return false;
+
+		reg_addr = (unsigned long *)((unsigned long)regs + reg_offset);
+		memcpy(reg_addr, dummy_data, dummy_data_size);
+	} else {
+		uaddr = insn_get_addr_ref(&insn, regs);
+		if ((unsigned long)uaddr == -1L)
+			return false;
+
+		nr_copied = copy_to_user(uaddr, dummy_data, dummy_data_size);
+		if (nr_copied  > 0)
+			return false;
+	}
+
+	/* increase IP to let the program keep going */
+	regs->ip += insn.length;
+	return true;
+}
-- 
2.7.4