xref: /linux/arch/x86/kernel/kprobes/core.c (revision b77e0ce62d63a761ffb7f7245a215a49f5921c2f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Kernel Probes (KProbes)
4  *
5  * Copyright (C) IBM Corporation, 2002, 2004
6  *
7  * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
8  *		Probes initial implementation ( includes contributions from
9  *		Rusty Russell).
10  * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
11  *		interface to access function arguments.
12  * 2004-Oct	Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
13  *		<prasanna@in.ibm.com> adapted for x86_64 from i386.
14  * 2005-Mar	Roland McGrath <roland@redhat.com>
15  *		Fixed to handle %rip-relative addressing mode correctly.
16  * 2005-May	Hien Nguyen <hien@us.ibm.com>, Jim Keniston
17  *		<jkenisto@us.ibm.com> and Prasanna S Panchamukhi
18  *		<prasanna@in.ibm.com> added function-return probes.
19  * 2005-May	Rusty Lynch <rusty.lynch@intel.com>
20  *		Added function return probes functionality
21  * 2006-Feb	Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added
22  *		kprobe-booster and kretprobe-booster for i386.
23  * 2007-Dec	Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
24  *		and kretprobe-booster for x86-64
25  * 2007-Dec	Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven
26  *		<arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com>
27  *		unified x86 kprobes code.
28  */
29 #include <linux/kprobes.h>
30 #include <linux/ptrace.h>
31 #include <linux/string.h>
32 #include <linux/slab.h>
33 #include <linux/hardirq.h>
34 #include <linux/preempt.h>
35 #include <linux/sched/debug.h>
36 #include <linux/perf_event.h>
37 #include <linux/extable.h>
38 #include <linux/kdebug.h>
39 #include <linux/kallsyms.h>
40 #include <linux/ftrace.h>
41 #include <linux/kasan.h>
42 #include <linux/moduleloader.h>
43 #include <linux/objtool.h>
44 #include <linux/vmalloc.h>
45 #include <linux/pgtable.h>
46 
47 #include <asm/text-patching.h>
48 #include <asm/cacheflush.h>
49 #include <asm/desc.h>
50 #include <linux/uaccess.h>
51 #include <asm/alternative.h>
52 #include <asm/insn.h>
53 #include <asm/debugreg.h>
54 #include <asm/set_memory.h>
55 
56 #include "common.h"
57 
58 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
59 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
60 
61 #define stack_addr(regs) ((unsigned long *)regs->sp)
62 
63 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
64 	(((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
65 	  (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
66 	  (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
67 	  (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
68 	 << (row % 32))
69 	/*
70 	 * Undefined/reserved opcodes, conditional jump, Opcode Extension
71 	 * Groups, and some special opcodes can not boost.
72 	 * This is non-const and volatile to keep gcc from statically
73 	 * optimizing it out, as variable_test_bit makes gcc think only
74 	 * *(unsigned long*) is used.
75 	 */
76 static volatile u32 twobyte_is_boostable[256 / 32] = {
77 	/*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
78 	/*      ----------------------------------------------          */
79 	W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
80 	W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
81 	W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
82 	W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
83 	W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
84 	W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
85 	W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
86 	W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
87 	W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
88 	W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
89 	W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
90 	W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
91 	W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
92 	W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
93 	W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
94 	W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
95 	/*      -----------------------------------------------         */
96 	/*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
97 };
98 #undef W
99 
100 struct kretprobe_blackpoint kretprobe_blacklist[] = {
101 	{"__switch_to", }, /* This function switches only current task, but
102 			      doesn't switch kernel stack.*/
103 	{NULL, NULL}	/* Terminator */
104 };
105 
106 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
107 
108 static nokprobe_inline void
109 __synthesize_relative_insn(void *dest, void *from, void *to, u8 op)
110 {
111 	struct __arch_relative_insn {
112 		u8 op;
113 		s32 raddr;
114 	} __packed *insn;
115 
116 	insn = (struct __arch_relative_insn *)dest;
117 	insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
118 	insn->op = op;
119 }
120 
121 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
122 void synthesize_reljump(void *dest, void *from, void *to)
123 {
124 	__synthesize_relative_insn(dest, from, to, JMP32_INSN_OPCODE);
125 }
126 NOKPROBE_SYMBOL(synthesize_reljump);
127 
128 /* Insert a call instruction at address 'from', which calls address 'to'.*/
129 void synthesize_relcall(void *dest, void *from, void *to)
130 {
131 	__synthesize_relative_insn(dest, from, to, CALL_INSN_OPCODE);
132 }
133 NOKPROBE_SYMBOL(synthesize_relcall);
134 
135 /*
136  * Returns non-zero if INSN is boostable.
137  * RIP relative instructions are adjusted at copying time in 64 bits mode
138  */
139 int can_boost(struct insn *insn, void *addr)
140 {
141 	kprobe_opcode_t opcode;
142 
143 	if (search_exception_tables((unsigned long)addr))
144 		return 0;	/* Page fault may occur on this address. */
145 
146 	/* 2nd-byte opcode */
147 	if (insn->opcode.nbytes == 2)
148 		return test_bit(insn->opcode.bytes[1],
149 				(unsigned long *)twobyte_is_boostable);
150 
151 	if (insn->opcode.nbytes != 1)
152 		return 0;
153 
154 	/* Can't boost Address-size override prefix */
155 	if (unlikely(inat_is_address_size_prefix(insn->attr)))
156 		return 0;
157 
158 	opcode = insn->opcode.bytes[0];
159 
160 	switch (opcode & 0xf0) {
161 	case 0x60:
162 		/* can't boost "bound" */
163 		return (opcode != 0x62);
164 	case 0x70:
165 		return 0; /* can't boost conditional jump */
166 	case 0x90:
167 		return opcode != 0x9a;	/* can't boost call far */
168 	case 0xc0:
169 		/* can't boost software-interruptions */
170 		return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
171 	case 0xd0:
172 		/* can boost AA* and XLAT */
173 		return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
174 	case 0xe0:
175 		/* can boost in/out and absolute jmps */
176 		return ((opcode & 0x04) || opcode == 0xea);
177 	case 0xf0:
178 		/* clear and set flags are boostable */
179 		return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
180 	default:
181 		/* CS override prefix and call are not boostable */
182 		return (opcode != 0x2e && opcode != 0x9a);
183 	}
184 }
185 
186 static unsigned long
187 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
188 {
189 	struct kprobe *kp;
190 	unsigned long faddr;
191 
192 	kp = get_kprobe((void *)addr);
193 	faddr = ftrace_location(addr);
194 	/*
195 	 * Addresses inside the ftrace location are refused by
196 	 * arch_check_ftrace_location(). Something went terribly wrong
197 	 * if such an address is checked here.
198 	 */
199 	if (WARN_ON(faddr && faddr != addr))
200 		return 0UL;
201 	/*
202 	 * Use the current code if it is not modified by Kprobe
203 	 * and it cannot be modified by ftrace.
204 	 */
205 	if (!kp && !faddr)
206 		return addr;
207 
208 	/*
209 	 * Basically, kp->ainsn.insn has an original instruction.
210 	 * However, RIP-relative instruction can not do single-stepping
211 	 * at different place, __copy_instruction() tweaks the displacement of
212 	 * that instruction. In that case, we can't recover the instruction
213 	 * from the kp->ainsn.insn.
214 	 *
215 	 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
216 	 * of the first byte of the probed instruction, which is overwritten
217 	 * by int3. And the instruction at kp->addr is not modified by kprobes
218 	 * except for the first byte, we can recover the original instruction
219 	 * from it and kp->opcode.
220 	 *
221 	 * In case of Kprobes using ftrace, we do not have a copy of
222 	 * the original instruction. In fact, the ftrace location might
223 	 * be modified at anytime and even could be in an inconsistent state.
224 	 * Fortunately, we know that the original code is the ideal 5-byte
225 	 * long NOP.
226 	 */
227 	if (copy_from_kernel_nofault(buf, (void *)addr,
228 		MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
229 		return 0UL;
230 
231 	if (faddr)
232 		memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
233 	else
234 		buf[0] = kp->opcode;
235 	return (unsigned long)buf;
236 }
237 
238 /*
239  * Recover the probed instruction at addr for further analysis.
240  * Caller must lock kprobes by kprobe_mutex, or disable preemption
241  * for preventing to release referencing kprobes.
242  * Returns zero if the instruction can not get recovered (or access failed).
243  */
244 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
245 {
246 	unsigned long __addr;
247 
248 	__addr = __recover_optprobed_insn(buf, addr);
249 	if (__addr != addr)
250 		return __addr;
251 
252 	return __recover_probed_insn(buf, addr);
253 }
254 
255 /* Check if paddr is at an instruction boundary */
256 static int can_probe(unsigned long paddr)
257 {
258 	unsigned long addr, __addr, offset = 0;
259 	struct insn insn;
260 	kprobe_opcode_t buf[MAX_INSN_SIZE];
261 
262 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
263 		return 0;
264 
265 	/* Decode instructions */
266 	addr = paddr - offset;
267 	while (addr < paddr) {
268 		/*
269 		 * Check if the instruction has been modified by another
270 		 * kprobe, in which case we replace the breakpoint by the
271 		 * original instruction in our buffer.
272 		 * Also, jump optimization will change the breakpoint to
273 		 * relative-jump. Since the relative-jump itself is
274 		 * normally used, we just go through if there is no kprobe.
275 		 */
276 		__addr = recover_probed_instruction(buf, addr);
277 		if (!__addr)
278 			return 0;
279 		kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
280 		insn_get_length(&insn);
281 
282 		/*
283 		 * Another debugging subsystem might insert this breakpoint.
284 		 * In that case, we can't recover it.
285 		 */
286 		if (insn.opcode.bytes[0] == INT3_INSN_OPCODE)
287 			return 0;
288 		addr += insn.length;
289 	}
290 
291 	return (addr == paddr);
292 }
293 
294 /*
295  * Copy an instruction with recovering modified instruction by kprobes
296  * and adjust the displacement if the instruction uses the %rip-relative
297  * addressing mode. Note that since @real will be the final place of copied
298  * instruction, displacement must be adjust by @real, not @dest.
299  * This returns the length of copied instruction, or 0 if it has an error.
300  */
301 int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn)
302 {
303 	kprobe_opcode_t buf[MAX_INSN_SIZE];
304 	unsigned long recovered_insn =
305 		recover_probed_instruction(buf, (unsigned long)src);
306 
307 	if (!recovered_insn || !insn)
308 		return 0;
309 
310 	/* This can access kernel text if given address is not recovered */
311 	if (copy_from_kernel_nofault(dest, (void *)recovered_insn,
312 			MAX_INSN_SIZE))
313 		return 0;
314 
315 	kernel_insn_init(insn, dest, MAX_INSN_SIZE);
316 	insn_get_length(insn);
317 
318 	/* We can not probe force emulate prefixed instruction */
319 	if (insn_has_emulate_prefix(insn))
320 		return 0;
321 
322 	/* Another subsystem puts a breakpoint, failed to recover */
323 	if (insn->opcode.bytes[0] == INT3_INSN_OPCODE)
324 		return 0;
325 
326 	/* We should not singlestep on the exception masking instructions */
327 	if (insn_masking_exception(insn))
328 		return 0;
329 
330 #ifdef CONFIG_X86_64
331 	/* Only x86_64 has RIP relative instructions */
332 	if (insn_rip_relative(insn)) {
333 		s64 newdisp;
334 		u8 *disp;
335 		/*
336 		 * The copied instruction uses the %rip-relative addressing
337 		 * mode.  Adjust the displacement for the difference between
338 		 * the original location of this instruction and the location
339 		 * of the copy that will actually be run.  The tricky bit here
340 		 * is making sure that the sign extension happens correctly in
341 		 * this calculation, since we need a signed 32-bit result to
342 		 * be sign-extended to 64 bits when it's added to the %rip
343 		 * value and yield the same 64-bit result that the sign-
344 		 * extension of the original signed 32-bit displacement would
345 		 * have given.
346 		 */
347 		newdisp = (u8 *) src + (s64) insn->displacement.value
348 			  - (u8 *) real;
349 		if ((s64) (s32) newdisp != newdisp) {
350 			pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
351 			return 0;
352 		}
353 		disp = (u8 *) dest + insn_offset_displacement(insn);
354 		*(s32 *) disp = (s32) newdisp;
355 	}
356 #endif
357 	return insn->length;
358 }
359 
360 /* Prepare reljump right after instruction to boost */
361 static int prepare_boost(kprobe_opcode_t *buf, struct kprobe *p,
362 			  struct insn *insn)
363 {
364 	int len = insn->length;
365 
366 	if (can_boost(insn, p->addr) &&
367 	    MAX_INSN_SIZE - len >= JMP32_INSN_SIZE) {
368 		/*
369 		 * These instructions can be executed directly if it
370 		 * jumps back to correct address.
371 		 */
372 		synthesize_reljump(buf + len, p->ainsn.insn + len,
373 				   p->addr + insn->length);
374 		len += JMP32_INSN_SIZE;
375 		p->ainsn.boostable = 1;
376 	} else {
377 		p->ainsn.boostable = 0;
378 	}
379 
380 	return len;
381 }
382 
383 /* Make page to RO mode when allocate it */
384 void *alloc_insn_page(void)
385 {
386 	void *page;
387 
388 	page = module_alloc(PAGE_SIZE);
389 	if (!page)
390 		return NULL;
391 
392 	set_vm_flush_reset_perms(page);
393 	/*
394 	 * First make the page read-only, and only then make it executable to
395 	 * prevent it from being W+X in between.
396 	 */
397 	set_memory_ro((unsigned long)page, 1);
398 
399 	/*
400 	 * TODO: Once additional kernel code protection mechanisms are set, ensure
401 	 * that the page was not maliciously altered and it is still zeroed.
402 	 */
403 	set_memory_x((unsigned long)page, 1);
404 
405 	return page;
406 }
407 
408 /* Recover page to RW mode before releasing it */
409 void free_insn_page(void *page)
410 {
411 	module_memfree(page);
412 }
413 
414 static void set_resume_flags(struct kprobe *p, struct insn *insn)
415 {
416 	insn_byte_t opcode = insn->opcode.bytes[0];
417 
418 	switch (opcode) {
419 	case 0xfa:		/* cli */
420 	case 0xfb:		/* sti */
421 	case 0x9d:		/* popf/popfd */
422 		/* Check whether the instruction modifies Interrupt Flag or not */
423 		p->ainsn.if_modifier = 1;
424 		break;
425 	case 0x9c:	/* pushfl */
426 		p->ainsn.is_pushf = 1;
427 		break;
428 	case 0xcf:	/* iret */
429 		p->ainsn.if_modifier = 1;
430 		fallthrough;
431 	case 0xc2:	/* ret/lret */
432 	case 0xc3:
433 	case 0xca:
434 	case 0xcb:
435 	case 0xea:	/* jmp absolute -- ip is correct */
436 		/* ip is already adjusted, no more changes required */
437 		p->ainsn.is_abs_ip = 1;
438 		/* Without resume jump, this is boostable */
439 		p->ainsn.boostable = 1;
440 		break;
441 	case 0xe8:	/* call relative - Fix return addr */
442 		p->ainsn.is_call = 1;
443 		break;
444 #ifdef CONFIG_X86_32
445 	case 0x9a:	/* call absolute -- same as call absolute, indirect */
446 		p->ainsn.is_call = 1;
447 		p->ainsn.is_abs_ip = 1;
448 		break;
449 #endif
450 	case 0xff:
451 		opcode = insn->opcode.bytes[1];
452 		if ((opcode & 0x30) == 0x10) {
453 			/*
454 			 * call absolute, indirect
455 			 * Fix return addr; ip is correct.
456 			 * But this is not boostable
457 			 */
458 			p->ainsn.is_call = 1;
459 			p->ainsn.is_abs_ip = 1;
460 			break;
461 		} else if (((opcode & 0x31) == 0x20) ||
462 			   ((opcode & 0x31) == 0x21)) {
463 			/*
464 			 * jmp near and far, absolute indirect
465 			 * ip is correct.
466 			 */
467 			p->ainsn.is_abs_ip = 1;
468 			/* Without resume jump, this is boostable */
469 			p->ainsn.boostable = 1;
470 		}
471 		break;
472 	}
473 }
474 
475 static int arch_copy_kprobe(struct kprobe *p)
476 {
477 	struct insn insn;
478 	kprobe_opcode_t buf[MAX_INSN_SIZE];
479 	int len;
480 
481 	/* Copy an instruction with recovering if other optprobe modifies it.*/
482 	len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn);
483 	if (!len)
484 		return -EINVAL;
485 
486 	/*
487 	 * __copy_instruction can modify the displacement of the instruction,
488 	 * but it doesn't affect boostable check.
489 	 */
490 	len = prepare_boost(buf, p, &insn);
491 
492 	/* Analyze the opcode and set resume flags */
493 	set_resume_flags(p, &insn);
494 
495 	/* Also, displacement change doesn't affect the first byte */
496 	p->opcode = buf[0];
497 
498 	p->ainsn.tp_len = len;
499 	perf_event_text_poke(p->ainsn.insn, NULL, 0, buf, len);
500 
501 	/* OK, write back the instruction(s) into ROX insn buffer */
502 	text_poke(p->ainsn.insn, buf, len);
503 
504 	return 0;
505 }
506 
507 int arch_prepare_kprobe(struct kprobe *p)
508 {
509 	int ret;
510 
511 	if (alternatives_text_reserved(p->addr, p->addr))
512 		return -EINVAL;
513 
514 	if (!can_probe((unsigned long)p->addr))
515 		return -EILSEQ;
516 
517 	memset(&p->ainsn, 0, sizeof(p->ainsn));
518 
519 	/* insn: must be on special executable page on x86. */
520 	p->ainsn.insn = get_insn_slot();
521 	if (!p->ainsn.insn)
522 		return -ENOMEM;
523 
524 	ret = arch_copy_kprobe(p);
525 	if (ret) {
526 		free_insn_slot(p->ainsn.insn, 0);
527 		p->ainsn.insn = NULL;
528 	}
529 
530 	return ret;
531 }
532 
533 void arch_arm_kprobe(struct kprobe *p)
534 {
535 	u8 int3 = INT3_INSN_OPCODE;
536 
537 	text_poke(p->addr, &int3, 1);
538 	text_poke_sync();
539 	perf_event_text_poke(p->addr, &p->opcode, 1, &int3, 1);
540 }
541 
542 void arch_disarm_kprobe(struct kprobe *p)
543 {
544 	u8 int3 = INT3_INSN_OPCODE;
545 
546 	perf_event_text_poke(p->addr, &int3, 1, &p->opcode, 1);
547 	text_poke(p->addr, &p->opcode, 1);
548 	text_poke_sync();
549 }
550 
551 void arch_remove_kprobe(struct kprobe *p)
552 {
553 	if (p->ainsn.insn) {
554 		/* Record the perf event before freeing the slot */
555 		perf_event_text_poke(p->ainsn.insn, p->ainsn.insn,
556 				     p->ainsn.tp_len, NULL, 0);
557 		free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
558 		p->ainsn.insn = NULL;
559 	}
560 }
561 
562 static nokprobe_inline void
563 save_previous_kprobe(struct kprobe_ctlblk *kcb)
564 {
565 	kcb->prev_kprobe.kp = kprobe_running();
566 	kcb->prev_kprobe.status = kcb->kprobe_status;
567 	kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
568 	kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
569 }
570 
571 static nokprobe_inline void
572 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
573 {
574 	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
575 	kcb->kprobe_status = kcb->prev_kprobe.status;
576 	kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
577 	kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
578 }
579 
580 static nokprobe_inline void
581 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
582 		   struct kprobe_ctlblk *kcb)
583 {
584 	__this_cpu_write(current_kprobe, p);
585 	kcb->kprobe_saved_flags = kcb->kprobe_old_flags
586 		= (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
587 	if (p->ainsn.if_modifier)
588 		kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
589 }
590 
591 static nokprobe_inline void clear_btf(void)
592 {
593 	if (test_thread_flag(TIF_BLOCKSTEP)) {
594 		unsigned long debugctl = get_debugctlmsr();
595 
596 		debugctl &= ~DEBUGCTLMSR_BTF;
597 		update_debugctlmsr(debugctl);
598 	}
599 }
600 
601 static nokprobe_inline void restore_btf(void)
602 {
603 	if (test_thread_flag(TIF_BLOCKSTEP)) {
604 		unsigned long debugctl = get_debugctlmsr();
605 
606 		debugctl |= DEBUGCTLMSR_BTF;
607 		update_debugctlmsr(debugctl);
608 	}
609 }
610 
611 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
612 {
613 	unsigned long *sara = stack_addr(regs);
614 
615 	ri->ret_addr = (kprobe_opcode_t *) *sara;
616 	ri->fp = sara;
617 
618 	/* Replace the return addr with trampoline addr */
619 	*sara = (unsigned long) &kretprobe_trampoline;
620 }
621 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
622 
623 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
624 			     struct kprobe_ctlblk *kcb, int reenter)
625 {
626 	if (setup_detour_execution(p, regs, reenter))
627 		return;
628 
629 #if !defined(CONFIG_PREEMPTION)
630 	if (p->ainsn.boostable && !p->post_handler) {
631 		/* Boost up -- we can execute copied instructions directly */
632 		if (!reenter)
633 			reset_current_kprobe();
634 		/*
635 		 * Reentering boosted probe doesn't reset current_kprobe,
636 		 * nor set current_kprobe, because it doesn't use single
637 		 * stepping.
638 		 */
639 		regs->ip = (unsigned long)p->ainsn.insn;
640 		return;
641 	}
642 #endif
643 	if (reenter) {
644 		save_previous_kprobe(kcb);
645 		set_current_kprobe(p, regs, kcb);
646 		kcb->kprobe_status = KPROBE_REENTER;
647 	} else
648 		kcb->kprobe_status = KPROBE_HIT_SS;
649 	/* Prepare real single stepping */
650 	clear_btf();
651 	regs->flags |= X86_EFLAGS_TF;
652 	regs->flags &= ~X86_EFLAGS_IF;
653 	/* single step inline if the instruction is an int3 */
654 	if (p->opcode == INT3_INSN_OPCODE)
655 		regs->ip = (unsigned long)p->addr;
656 	else
657 		regs->ip = (unsigned long)p->ainsn.insn;
658 }
659 NOKPROBE_SYMBOL(setup_singlestep);
660 
661 /*
662  * We have reentered the kprobe_handler(), since another probe was hit while
663  * within the handler. We save the original kprobes variables and just single
664  * step on the instruction of the new probe without calling any user handlers.
665  */
666 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
667 			  struct kprobe_ctlblk *kcb)
668 {
669 	switch (kcb->kprobe_status) {
670 	case KPROBE_HIT_SSDONE:
671 	case KPROBE_HIT_ACTIVE:
672 	case KPROBE_HIT_SS:
673 		kprobes_inc_nmissed_count(p);
674 		setup_singlestep(p, regs, kcb, 1);
675 		break;
676 	case KPROBE_REENTER:
677 		/* A probe has been hit in the codepath leading up to, or just
678 		 * after, single-stepping of a probed instruction. This entire
679 		 * codepath should strictly reside in .kprobes.text section.
680 		 * Raise a BUG or we'll continue in an endless reentering loop
681 		 * and eventually a stack overflow.
682 		 */
683 		pr_err("Unrecoverable kprobe detected.\n");
684 		dump_kprobe(p);
685 		BUG();
686 	default:
687 		/* impossible cases */
688 		WARN_ON(1);
689 		return 0;
690 	}
691 
692 	return 1;
693 }
694 NOKPROBE_SYMBOL(reenter_kprobe);
695 
696 /*
697  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
698  * remain disabled throughout this function.
699  */
700 int kprobe_int3_handler(struct pt_regs *regs)
701 {
702 	kprobe_opcode_t *addr;
703 	struct kprobe *p;
704 	struct kprobe_ctlblk *kcb;
705 
706 	if (user_mode(regs))
707 		return 0;
708 
709 	addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
710 	/*
711 	 * We don't want to be preempted for the entire duration of kprobe
712 	 * processing. Since int3 and debug trap disables irqs and we clear
713 	 * IF while singlestepping, it must be no preemptible.
714 	 */
715 
716 	kcb = get_kprobe_ctlblk();
717 	p = get_kprobe(addr);
718 
719 	if (p) {
720 		if (kprobe_running()) {
721 			if (reenter_kprobe(p, regs, kcb))
722 				return 1;
723 		} else {
724 			set_current_kprobe(p, regs, kcb);
725 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
726 
727 			/*
728 			 * If we have no pre-handler or it returned 0, we
729 			 * continue with normal processing.  If we have a
730 			 * pre-handler and it returned non-zero, that means
731 			 * user handler setup registers to exit to another
732 			 * instruction, we must skip the single stepping.
733 			 */
734 			if (!p->pre_handler || !p->pre_handler(p, regs))
735 				setup_singlestep(p, regs, kcb, 0);
736 			else
737 				reset_current_kprobe();
738 			return 1;
739 		}
740 	} else if (*addr != INT3_INSN_OPCODE) {
741 		/*
742 		 * The breakpoint instruction was removed right
743 		 * after we hit it.  Another cpu has removed
744 		 * either a probepoint or a debugger breakpoint
745 		 * at this address.  In either case, no further
746 		 * handling of this interrupt is appropriate.
747 		 * Back up over the (now missing) int3 and run
748 		 * the original instruction.
749 		 */
750 		regs->ip = (unsigned long)addr;
751 		return 1;
752 	} /* else: not a kprobe fault; let the kernel handle it */
753 
754 	return 0;
755 }
756 NOKPROBE_SYMBOL(kprobe_int3_handler);
757 
758 /*
759  * When a retprobed function returns, this code saves registers and
760  * calls trampoline_handler() runs, which calls the kretprobe's handler.
761  */
762 asm(
763 	".text\n"
764 	".global kretprobe_trampoline\n"
765 	".type kretprobe_trampoline, @function\n"
766 	"kretprobe_trampoline:\n"
767 	/* We don't bother saving the ss register */
768 #ifdef CONFIG_X86_64
769 	"	pushq %rsp\n"
770 	"	pushfq\n"
771 	SAVE_REGS_STRING
772 	"	movq %rsp, %rdi\n"
773 	"	call trampoline_handler\n"
774 	/* Replace saved sp with true return address. */
775 	"	movq %rax, 19*8(%rsp)\n"
776 	RESTORE_REGS_STRING
777 	"	popfq\n"
778 #else
779 	"	pushl %esp\n"
780 	"	pushfl\n"
781 	SAVE_REGS_STRING
782 	"	movl %esp, %eax\n"
783 	"	call trampoline_handler\n"
784 	/* Replace saved sp with true return address. */
785 	"	movl %eax, 15*4(%esp)\n"
786 	RESTORE_REGS_STRING
787 	"	popfl\n"
788 #endif
789 	"	ret\n"
790 	".size kretprobe_trampoline, .-kretprobe_trampoline\n"
791 );
792 NOKPROBE_SYMBOL(kretprobe_trampoline);
793 STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
794 
795 
796 /*
797  * Called from kretprobe_trampoline
798  */
799 __used __visible void *trampoline_handler(struct pt_regs *regs)
800 {
801 	/* fixup registers */
802 	regs->cs = __KERNEL_CS;
803 #ifdef CONFIG_X86_32
804 	regs->gs = 0;
805 #endif
806 	regs->ip = (unsigned long)&kretprobe_trampoline;
807 	regs->orig_ax = ~0UL;
808 
809 	return (void *)kretprobe_trampoline_handler(regs, &kretprobe_trampoline, &regs->sp);
810 }
811 NOKPROBE_SYMBOL(trampoline_handler);
812 
813 /*
814  * Called after single-stepping.  p->addr is the address of the
815  * instruction whose first byte has been replaced by the "int 3"
816  * instruction.  To avoid the SMP problems that can occur when we
817  * temporarily put back the original opcode to single-step, we
818  * single-stepped a copy of the instruction.  The address of this
819  * copy is p->ainsn.insn.
820  *
821  * This function prepares to return from the post-single-step
822  * interrupt.  We have to fix up the stack as follows:
823  *
824  * 0) Except in the case of absolute or indirect jump or call instructions,
825  * the new ip is relative to the copied instruction.  We need to make
826  * it relative to the original instruction.
827  *
828  * 1) If the single-stepped instruction was pushfl, then the TF and IF
829  * flags are set in the just-pushed flags, and may need to be cleared.
830  *
831  * 2) If the single-stepped instruction was a call, the return address
832  * that is atop the stack is the address following the copied instruction.
833  * We need to make it the address following the original instruction.
834  */
835 static void resume_execution(struct kprobe *p, struct pt_regs *regs,
836 			     struct kprobe_ctlblk *kcb)
837 {
838 	unsigned long *tos = stack_addr(regs);
839 	unsigned long copy_ip = (unsigned long)p->ainsn.insn;
840 	unsigned long orig_ip = (unsigned long)p->addr;
841 
842 	regs->flags &= ~X86_EFLAGS_TF;
843 
844 	/* Fixup the contents of top of stack */
845 	if (p->ainsn.is_pushf) {
846 		*tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
847 		*tos |= kcb->kprobe_old_flags;
848 	} else if (p->ainsn.is_call) {
849 		*tos = orig_ip + (*tos - copy_ip);
850 	}
851 
852 	if (!p->ainsn.is_abs_ip)
853 		regs->ip += orig_ip - copy_ip;
854 
855 	restore_btf();
856 }
857 NOKPROBE_SYMBOL(resume_execution);
858 
859 /*
860  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
861  * remain disabled throughout this function.
862  */
863 int kprobe_debug_handler(struct pt_regs *regs)
864 {
865 	struct kprobe *cur = kprobe_running();
866 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
867 
868 	if (!cur)
869 		return 0;
870 
871 	resume_execution(cur, regs, kcb);
872 	regs->flags |= kcb->kprobe_saved_flags;
873 
874 	if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
875 		kcb->kprobe_status = KPROBE_HIT_SSDONE;
876 		cur->post_handler(cur, regs, 0);
877 	}
878 
879 	/* Restore back the original saved kprobes variables and continue. */
880 	if (kcb->kprobe_status == KPROBE_REENTER) {
881 		restore_previous_kprobe(kcb);
882 		goto out;
883 	}
884 	reset_current_kprobe();
885 out:
886 	/*
887 	 * if somebody else is singlestepping across a probe point, flags
888 	 * will have TF set, in which case, continue the remaining processing
889 	 * of do_debug, as if this is not a probe hit.
890 	 */
891 	if (regs->flags & X86_EFLAGS_TF)
892 		return 0;
893 
894 	return 1;
895 }
896 NOKPROBE_SYMBOL(kprobe_debug_handler);
897 
898 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
899 {
900 	struct kprobe *cur = kprobe_running();
901 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
902 
903 	if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
904 		/* This must happen on single-stepping */
905 		WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
906 			kcb->kprobe_status != KPROBE_REENTER);
907 		/*
908 		 * We are here because the instruction being single
909 		 * stepped caused a page fault. We reset the current
910 		 * kprobe and the ip points back to the probe address
911 		 * and allow the page fault handler to continue as a
912 		 * normal page fault.
913 		 */
914 		regs->ip = (unsigned long)cur->addr;
915 		/*
916 		 * Trap flag (TF) has been set here because this fault
917 		 * happened where the single stepping will be done.
918 		 * So clear it by resetting the current kprobe:
919 		 */
920 		regs->flags &= ~X86_EFLAGS_TF;
921 		/*
922 		 * Since the single step (trap) has been cancelled,
923 		 * we need to restore BTF here.
924 		 */
925 		restore_btf();
926 
927 		/*
928 		 * If the TF flag was set before the kprobe hit,
929 		 * don't touch it:
930 		 */
931 		regs->flags |= kcb->kprobe_old_flags;
932 
933 		if (kcb->kprobe_status == KPROBE_REENTER)
934 			restore_previous_kprobe(kcb);
935 		else
936 			reset_current_kprobe();
937 	} else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
938 		   kcb->kprobe_status == KPROBE_HIT_SSDONE) {
939 		/*
940 		 * We increment the nmissed count for accounting,
941 		 * we can also use npre/npostfault count for accounting
942 		 * these specific fault cases.
943 		 */
944 		kprobes_inc_nmissed_count(cur);
945 
946 		/*
947 		 * We come here because instructions in the pre/post
948 		 * handler caused the page_fault, this could happen
949 		 * if handler tries to access user space by
950 		 * copy_from_user(), get_user() etc. Let the
951 		 * user-specified handler try to fix it first.
952 		 */
953 		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
954 			return 1;
955 	}
956 
957 	return 0;
958 }
959 NOKPROBE_SYMBOL(kprobe_fault_handler);
960 
961 int __init arch_populate_kprobe_blacklist(void)
962 {
963 	return kprobe_add_area_blacklist((unsigned long)__entry_text_start,
964 					 (unsigned long)__entry_text_end);
965 }
966 
967 int __init arch_init_kprobes(void)
968 {
969 	return 0;
970 }
971 
972 int arch_trampoline_kprobe(struct kprobe *p)
973 {
974 	return 0;
975 }
976