xref: /openbmc/linux/arch/x86/kernel/umip.c (revision dc6a81c3)
1 /*
2  * umip.c Emulation for instruction protected by the User-Mode Instruction
3  * Prevention feature
4  *
5  * Copyright (c) 2017, Intel Corporation.
6  * Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
7  */
8 
9 #include <linux/uaccess.h>
10 #include <asm/umip.h>
11 #include <asm/traps.h>
12 #include <asm/insn.h>
13 #include <asm/insn-eval.h>
14 #include <linux/ratelimit.h>
15 
16 #undef pr_fmt
17 #define pr_fmt(fmt) "umip: " fmt
18 
19 /** DOC: Emulation for User-Mode Instruction Prevention (UMIP)
20  *
21  * User-Mode Instruction Prevention is a security feature present in recent
22  * x86 processors that, when enabled, prevents a group of instructions (SGDT,
23  * SIDT, SLDT, SMSW and STR) from being run in user mode by issuing a general
24  * protection fault if the instruction is executed with CPL > 0.
25  *
26  * Rather than relaying to the user space the general protection fault caused by
27  * the UMIP-protected instructions (in the form of a SIGSEGV signal), it can be
28  * trapped and emulate the result of such instructions to provide dummy values.
29  * This allows to both conserve the current kernel behavior and not reveal the
30  * system resources that UMIP intends to protect (i.e., the locations of the
31  * global descriptor and interrupt descriptor tables, the segment selectors of
32  * the local descriptor table, the value of the task state register and the
33  * contents of the CR0 register).
34  *
35  * This emulation is needed because certain applications (e.g., WineHQ and
36  * DOSEMU2) rely on this subset of instructions to function.
37  *
38  * The instructions protected by UMIP can be split in two groups. Those which
39  * return a kernel memory address (SGDT and SIDT) and those which return a
40  * value (SLDT, STR and SMSW).
41  *
42  * For the instructions that return a kernel memory address, applications
43  * such as WineHQ rely on the result being located in the kernel memory space,
44  * not the actual location of the table. The result is emulated as a hard-coded
45  * value that, lies close to the top of the kernel memory. The limit for the GDT
46  * and the IDT are set to zero.
47  *
48  * Given that SLDT and STR are not commonly used in programs that run on WineHQ
49  * or DOSEMU2, they are not emulated.
50  *
51  * The instruction smsw is emulated to return the value that the register CR0
52  * has at boot time as set in the head_32.
53  *
54  * Emulation is provided for both 32-bit and 64-bit processes.
55  *
56  * Care is taken to appropriately emulate the results when segmentation is
57  * used. That is, rather than relying on USER_DS and USER_CS, the function
58  * insn_get_addr_ref() inspects the segment descriptor pointed by the
59  * registers in pt_regs. This ensures that we correctly obtain the segment
60  * base address and the address and operand sizes even if the user space
61  * application uses a local descriptor table.
62  */
63 
64 #define UMIP_DUMMY_GDT_BASE 0xfffffffffffe0000ULL
65 #define UMIP_DUMMY_IDT_BASE 0xffffffffffff0000ULL
66 
67 /*
68  * The SGDT and SIDT instructions store the contents of the global descriptor
69  * table and interrupt table registers, respectively. The destination is a
70  * memory operand of X+2 bytes. X bytes are used to store the base address of
71  * the table and 2 bytes are used to store the limit. In 32-bit processes X
72  * has a value of 4, in 64-bit processes X has a value of 8.
73  */
74 #define UMIP_GDT_IDT_BASE_SIZE_64BIT 8
75 #define UMIP_GDT_IDT_BASE_SIZE_32BIT 4
76 #define UMIP_GDT_IDT_LIMIT_SIZE 2
77 
78 #define	UMIP_INST_SGDT	0	/* 0F 01 /0 */
79 #define	UMIP_INST_SIDT	1	/* 0F 01 /1 */
80 #define	UMIP_INST_SMSW	2	/* 0F 01 /4 */
81 #define	UMIP_INST_SLDT  3       /* 0F 00 /0 */
82 #define	UMIP_INST_STR   4       /* 0F 00 /1 */
83 
84 const char * const umip_insns[5] = {
85 	[UMIP_INST_SGDT] = "SGDT",
86 	[UMIP_INST_SIDT] = "SIDT",
87 	[UMIP_INST_SMSW] = "SMSW",
88 	[UMIP_INST_SLDT] = "SLDT",
89 	[UMIP_INST_STR] = "STR",
90 };
91 
92 #define umip_pr_err(regs, fmt, ...) \
93 	umip_printk(regs, KERN_ERR, fmt, ##__VA_ARGS__)
94 #define umip_pr_warn(regs, fmt, ...) \
95 	umip_printk(regs, KERN_WARNING, fmt,  ##__VA_ARGS__)
96 
97 /**
98  * umip_printk() - Print a rate-limited message
99  * @regs:	Register set with the context in which the warning is printed
100  * @log_level:	Kernel log level to print the message
101  * @fmt:	The text string to print
102  *
103  * Print the text contained in @fmt. The print rate is limited to bursts of 5
104  * messages every two minutes. The purpose of this customized version of
105  * printk() is to print messages when user space processes use any of the
106  * UMIP-protected instructions. Thus, the printed text is prepended with the
107  * task name and process ID number of the current task as well as the
108  * instruction and stack pointers in @regs as seen when entering kernel mode.
109  *
110  * Returns:
111  *
112  * None.
113  */
114 static __printf(3, 4)
115 void umip_printk(const struct pt_regs *regs, const char *log_level,
116 		 const char *fmt, ...)
117 {
118 	/* Bursts of 5 messages every two minutes */
119 	static DEFINE_RATELIMIT_STATE(ratelimit, 2 * 60 * HZ, 5);
120 	struct task_struct *tsk = current;
121 	struct va_format vaf;
122 	va_list args;
123 
124 	if (!__ratelimit(&ratelimit))
125 		return;
126 
127 	va_start(args, fmt);
128 	vaf.fmt = fmt;
129 	vaf.va = &args;
130 	printk("%s" pr_fmt("%s[%d] ip:%lx sp:%lx: %pV"), log_level, tsk->comm,
131 	       task_pid_nr(tsk), regs->ip, regs->sp, &vaf);
132 	va_end(args);
133 }
134 
135 /**
136  * identify_insn() - Identify a UMIP-protected instruction
137  * @insn:	Instruction structure with opcode and ModRM byte.
138  *
139  * From the opcode and ModRM.reg in @insn identify, if any, a UMIP-protected
140  * instruction that can be emulated.
141  *
142  * Returns:
143  *
144  * On success, a constant identifying a specific UMIP-protected instruction that
145  * can be emulated.
146  *
147  * -EINVAL on error or when not an UMIP-protected instruction that can be
148  * emulated.
149  */
150 static int identify_insn(struct insn *insn)
151 {
152 	/* By getting modrm we also get the opcode. */
153 	insn_get_modrm(insn);
154 
155 	if (!insn->modrm.nbytes)
156 		return -EINVAL;
157 
158 	/* All the instructions of interest start with 0x0f. */
159 	if (insn->opcode.bytes[0] != 0xf)
160 		return -EINVAL;
161 
162 	if (insn->opcode.bytes[1] == 0x1) {
163 		switch (X86_MODRM_REG(insn->modrm.value)) {
164 		case 0:
165 			return UMIP_INST_SGDT;
166 		case 1:
167 			return UMIP_INST_SIDT;
168 		case 4:
169 			return UMIP_INST_SMSW;
170 		default:
171 			return -EINVAL;
172 		}
173 	} else if (insn->opcode.bytes[1] == 0x0) {
174 		if (X86_MODRM_REG(insn->modrm.value) == 0)
175 			return UMIP_INST_SLDT;
176 		else if (X86_MODRM_REG(insn->modrm.value) == 1)
177 			return UMIP_INST_STR;
178 		else
179 			return -EINVAL;
180 	} else {
181 		return -EINVAL;
182 	}
183 }
184 
185 /**
186  * emulate_umip_insn() - Emulate UMIP instructions and return dummy values
187  * @insn:	Instruction structure with operands
188  * @umip_inst:	A constant indicating the instruction to emulate
189  * @data:	Buffer into which the dummy result is stored
190  * @data_size:	Size of the emulated result
191  * @x86_64:	true if process is 64-bit, false otherwise
192  *
193  * Emulate an instruction protected by UMIP and provide a dummy result. The
194  * result of the emulation is saved in @data. The size of the results depends
195  * on both the instruction and type of operand (register vs memory address).
196  * The size of the result is updated in @data_size. Caller is responsible
197  * of providing a @data buffer of at least UMIP_GDT_IDT_BASE_SIZE +
198  * UMIP_GDT_IDT_LIMIT_SIZE bytes.
199  *
200  * Returns:
201  *
202  * 0 on success, -EINVAL on error while emulating.
203  */
204 static int emulate_umip_insn(struct insn *insn, int umip_inst,
205 			     unsigned char *data, int *data_size, bool x86_64)
206 {
207 	if (!data || !data_size || !insn)
208 		return -EINVAL;
209 	/*
210 	 * These two instructions return the base address and limit of the
211 	 * global and interrupt descriptor table, respectively. According to the
212 	 * Intel Software Development manual, the base address can be 24-bit,
213 	 * 32-bit or 64-bit. Limit is always 16-bit. If the operand size is
214 	 * 16-bit, the returned value of the base address is supposed to be a
215 	 * zero-extended 24-byte number. However, it seems that a 32-byte number
216 	 * is always returned irrespective of the operand size.
217 	 */
218 	if (umip_inst == UMIP_INST_SGDT || umip_inst == UMIP_INST_SIDT) {
219 		u64 dummy_base_addr;
220 		u16 dummy_limit = 0;
221 
222 		/* SGDT and SIDT do not use registers operands. */
223 		if (X86_MODRM_MOD(insn->modrm.value) == 3)
224 			return -EINVAL;
225 
226 		if (umip_inst == UMIP_INST_SGDT)
227 			dummy_base_addr = UMIP_DUMMY_GDT_BASE;
228 		else
229 			dummy_base_addr = UMIP_DUMMY_IDT_BASE;
230 
231 		/*
232 		 * 64-bit processes use the entire dummy base address.
233 		 * 32-bit processes use the lower 32 bits of the base address.
234 		 * dummy_base_addr is always 64 bits, but we memcpy the correct
235 		 * number of bytes from it to the destination.
236 		 */
237 		if (x86_64)
238 			*data_size = UMIP_GDT_IDT_BASE_SIZE_64BIT;
239 		else
240 			*data_size = UMIP_GDT_IDT_BASE_SIZE_32BIT;
241 
242 		memcpy(data + 2, &dummy_base_addr, *data_size);
243 
244 		*data_size += UMIP_GDT_IDT_LIMIT_SIZE;
245 		memcpy(data, &dummy_limit, UMIP_GDT_IDT_LIMIT_SIZE);
246 
247 	} else if (umip_inst == UMIP_INST_SMSW) {
248 		unsigned long dummy_value = CR0_STATE;
249 
250 		/*
251 		 * Even though the CR0 register has 4 bytes, the number
252 		 * of bytes to be copied in the result buffer is determined
253 		 * by whether the operand is a register or a memory location.
254 		 * If operand is a register, return as many bytes as the operand
255 		 * size. If operand is memory, return only the two least
256 		 * siginificant bytes of CR0.
257 		 */
258 		if (X86_MODRM_MOD(insn->modrm.value) == 3)
259 			*data_size = insn->opnd_bytes;
260 		else
261 			*data_size = 2;
262 
263 		memcpy(data, &dummy_value, *data_size);
264 	/* STR and SLDT  are not emulated */
265 	} else {
266 		return -EINVAL;
267 	}
268 
269 	return 0;
270 }
271 
272 /**
273  * force_sig_info_umip_fault() - Force a SIGSEGV with SEGV_MAPERR
274  * @addr:	Address that caused the signal
275  * @regs:	Register set containing the instruction pointer
276  *
277  * Force a SIGSEGV signal with SEGV_MAPERR as the error code. This function is
278  * intended to be used to provide a segmentation fault when the result of the
279  * UMIP emulation could not be copied to the user space memory.
280  *
281  * Returns: none
282  */
283 static void force_sig_info_umip_fault(void __user *addr, struct pt_regs *regs)
284 {
285 	struct task_struct *tsk = current;
286 
287 	tsk->thread.cr2		= (unsigned long)addr;
288 	tsk->thread.error_code	= X86_PF_USER | X86_PF_WRITE;
289 	tsk->thread.trap_nr	= X86_TRAP_PF;
290 
291 	force_sig_fault(SIGSEGV, SEGV_MAPERR, addr);
292 
293 	if (!(show_unhandled_signals && unhandled_signal(tsk, SIGSEGV)))
294 		return;
295 
296 	umip_pr_err(regs, "segfault in emulation. error%x\n",
297 		    X86_PF_USER | X86_PF_WRITE);
298 }
299 
300 /**
301  * fixup_umip_exception() - Fixup a general protection fault caused by UMIP
302  * @regs:	Registers as saved when entering the #GP handler
303  *
304  * The instructions SGDT, SIDT, STR, SMSW and SLDT cause a general protection
305  * fault if executed with CPL > 0 (i.e., from user space). This function fixes
306  * the exception up and provides dummy results for SGDT, SIDT and SMSW; STR
307  * and SLDT are not fixed up.
308  *
309  * If operands are memory addresses, results are copied to user-space memory as
310  * indicated by the instruction pointed by eIP using the registers indicated in
311  * the instruction operands. If operands are registers, results are copied into
312  * the context that was saved when entering kernel mode.
313  *
314  * Returns:
315  *
316  * True if emulation was successful; false if not.
317  */
318 bool fixup_umip_exception(struct pt_regs *regs)
319 {
320 	int not_copied, nr_copied, reg_offset, dummy_data_size, umip_inst;
321 	unsigned long seg_base = 0, *reg_addr;
322 	/* 10 bytes is the maximum size of the result of UMIP instructions */
323 	unsigned char dummy_data[10] = { 0 };
324 	unsigned char buf[MAX_INSN_SIZE];
325 	void __user *uaddr;
326 	struct insn insn;
327 	int seg_defs;
328 
329 	if (!regs)
330 		return false;
331 
332 	/*
333 	 * If not in user-space long mode, a custom code segment could be in
334 	 * use. This is true in protected mode (if the process defined a local
335 	 * descriptor table), or virtual-8086 mode. In most of the cases
336 	 * seg_base will be zero as in USER_CS.
337 	 */
338 	if (!user_64bit_mode(regs))
339 		seg_base = insn_get_seg_base(regs, INAT_SEG_REG_CS);
340 
341 	if (seg_base == -1L)
342 		return false;
343 
344 	not_copied = copy_from_user(buf, (void __user *)(seg_base + regs->ip),
345 				    sizeof(buf));
346 	nr_copied = sizeof(buf) - not_copied;
347 
348 	/*
349 	 * The copy_from_user above could have failed if user code is protected
350 	 * by a memory protection key. Give up on emulation in such a case.
351 	 * Should we issue a page fault?
352 	 */
353 	if (!nr_copied)
354 		return false;
355 
356 	insn_init(&insn, buf, nr_copied, user_64bit_mode(regs));
357 
358 	/*
359 	 * Override the default operand and address sizes with what is specified
360 	 * in the code segment descriptor. The instruction decoder only sets
361 	 * the address size it to either 4 or 8 address bytes and does nothing
362 	 * for the operand bytes. This OK for most of the cases, but we could
363 	 * have special cases where, for instance, a 16-bit code segment
364 	 * descriptor is used.
365 	 * If there is an address override prefix, the instruction decoder
366 	 * correctly updates these values, even for 16-bit defaults.
367 	 */
368 	seg_defs = insn_get_code_seg_params(regs);
369 	if (seg_defs == -EINVAL)
370 		return false;
371 
372 	insn.addr_bytes = INSN_CODE_SEG_ADDR_SZ(seg_defs);
373 	insn.opnd_bytes = INSN_CODE_SEG_OPND_SZ(seg_defs);
374 
375 	insn_get_length(&insn);
376 	if (nr_copied < insn.length)
377 		return false;
378 
379 	umip_inst = identify_insn(&insn);
380 	if (umip_inst < 0)
381 		return false;
382 
383 	umip_pr_warn(regs, "%s instruction cannot be used by applications.\n",
384 			umip_insns[umip_inst]);
385 
386 	/* Do not emulate (spoof) SLDT or STR. */
387 	if (umip_inst == UMIP_INST_STR || umip_inst == UMIP_INST_SLDT)
388 		return false;
389 
390 	umip_pr_warn(regs, "For now, expensive software emulation returns the result.\n");
391 
392 	if (emulate_umip_insn(&insn, umip_inst, dummy_data, &dummy_data_size,
393 			      user_64bit_mode(regs)))
394 		return false;
395 
396 	/*
397 	 * If operand is a register, write result to the copy of the register
398 	 * value that was pushed to the stack when entering into kernel mode.
399 	 * Upon exit, the value we write will be restored to the actual hardware
400 	 * register.
401 	 */
402 	if (X86_MODRM_MOD(insn.modrm.value) == 3) {
403 		reg_offset = insn_get_modrm_rm_off(&insn, regs);
404 
405 		/*
406 		 * Negative values are usually errors. In memory addressing,
407 		 * the exception is -EDOM. Since we expect a register operand,
408 		 * all negative values are errors.
409 		 */
410 		if (reg_offset < 0)
411 			return false;
412 
413 		reg_addr = (unsigned long *)((unsigned long)regs + reg_offset);
414 		memcpy(reg_addr, dummy_data, dummy_data_size);
415 	} else {
416 		uaddr = insn_get_addr_ref(&insn, regs);
417 		if ((unsigned long)uaddr == -1L)
418 			return false;
419 
420 		nr_copied = copy_to_user(uaddr, dummy_data, dummy_data_size);
421 		if (nr_copied  > 0) {
422 			/*
423 			 * If copy fails, send a signal and tell caller that
424 			 * fault was fixed up.
425 			 */
426 			force_sig_info_umip_fault(uaddr, regs);
427 			return true;
428 		}
429 	}
430 
431 	/* increase IP to let the program keep going */
432 	regs->ip += insn.length;
433 	return true;
434 }
435