1/*
2 *  Kernel Probes (KProbes)
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright (C) IBM Corporation, 2002, 2004
19 *
20 * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 *		Probes initial implementation ( includes contributions from
22 *		Rusty Russell).
23 * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
24 *		interface to access function arguments.
25 * 2004-Oct	Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
26 *		<prasanna@in.ibm.com> adapted for x86_64 from i386.
27 * 2005-Mar	Roland McGrath <roland@redhat.com>
28 *		Fixed to handle %rip-relative addressing mode correctly.
29 * 2005-May	Hien Nguyen <hien@us.ibm.com>, Jim Keniston
30 *		<jkenisto@us.ibm.com> and Prasanna S Panchamukhi
31 *		<prasanna@in.ibm.com> added function-return probes.
32 * 2005-May	Rusty Lynch <rusty.lynch@intel.com>
33 *		Added function return probes functionality
34 * 2006-Feb	Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added
35 *		kprobe-booster and kretprobe-booster for i386.
36 * 2007-Dec	Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
37 *		and kretprobe-booster for x86-64
38 * 2007-Dec	Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven
39 *		<arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com>
40 *		unified x86 kprobes code.
41 */
42#include <linux/kprobes.h>
43#include <linux/ptrace.h>
44#include <linux/string.h>
45#include <linux/slab.h>
46#include <linux/hardirq.h>
47#include <linux/preempt.h>
48#include <linux/module.h>
49#include <linux/kdebug.h>
50#include <linux/kallsyms.h>
51#include <linux/ftrace.h>
52
53#include <asm/cacheflush.h>
54#include <asm/desc.h>
55#include <asm/pgtable.h>
56#include <asm/uaccess.h>
57#include <asm/alternative.h>
58#include <asm/insn.h>
59#include <asm/debugreg.h>
60
61#include "common.h"
62
63void jprobe_return_end(void);
64
65DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
66DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
67
68#define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
69
70#define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
71	(((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
72	  (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
73	  (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
74	  (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
75	 << (row % 32))
76	/*
77	 * Undefined/reserved opcodes, conditional jump, Opcode Extension
78	 * Groups, and some special opcodes can not boost.
79	 * This is non-const and volatile to keep gcc from statically
80	 * optimizing it out, as variable_test_bit makes gcc think only
81	 * *(unsigned long*) is used.
82	 */
83static volatile u32 twobyte_is_boostable[256 / 32] = {
84	/*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
85	/*      ----------------------------------------------          */
86	W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
87	W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
88	W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
89	W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
90	W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
91	W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
92	W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
93	W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
94	W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
95	W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
96	W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
97	W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
98	W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
99	W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
100	W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
101	W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
102	/*      -----------------------------------------------         */
103	/*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
104};
105#undef W
106
107struct kretprobe_blackpoint kretprobe_blacklist[] = {
108	{"__switch_to", }, /* This function switches only current task, but
109			      doesn't switch kernel stack.*/
110	{NULL, NULL}	/* Terminator */
111};
112
113const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
114
115static nokprobe_inline void
116__synthesize_relative_insn(void *from, void *to, u8 op)
117{
118	struct __arch_relative_insn {
119		u8 op;
120		s32 raddr;
121	} __packed *insn;
122
123	insn = (struct __arch_relative_insn *)from;
124	insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
125	insn->op = op;
126}
127
128/* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
129void synthesize_reljump(void *from, void *to)
130{
131	__synthesize_relative_insn(from, to, RELATIVEJUMP_OPCODE);
132}
133NOKPROBE_SYMBOL(synthesize_reljump);
134
135/* Insert a call instruction at address 'from', which calls address 'to'.*/
136void synthesize_relcall(void *from, void *to)
137{
138	__synthesize_relative_insn(from, to, RELATIVECALL_OPCODE);
139}
140NOKPROBE_SYMBOL(synthesize_relcall);
141
142/*
143 * Skip the prefixes of the instruction.
144 */
145static kprobe_opcode_t *skip_prefixes(kprobe_opcode_t *insn)
146{
147	insn_attr_t attr;
148
149	attr = inat_get_opcode_attribute((insn_byte_t)*insn);
150	while (inat_is_legacy_prefix(attr)) {
151		insn++;
152		attr = inat_get_opcode_attribute((insn_byte_t)*insn);
153	}
154#ifdef CONFIG_X86_64
155	if (inat_is_rex_prefix(attr))
156		insn++;
157#endif
158	return insn;
159}
160NOKPROBE_SYMBOL(skip_prefixes);
161
162/*
163 * Returns non-zero if opcode is boostable.
164 * RIP relative instructions are adjusted at copying time in 64 bits mode
165 */
166int can_boost(kprobe_opcode_t *opcodes)
167{
168	kprobe_opcode_t opcode;
169	kprobe_opcode_t *orig_opcodes = opcodes;
170
171	if (search_exception_tables((unsigned long)opcodes))
172		return 0;	/* Page fault may occur on this address. */
173
174retry:
175	if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
176		return 0;
177	opcode = *(opcodes++);
178
179	/* 2nd-byte opcode */
180	if (opcode == 0x0f) {
181		if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
182			return 0;
183		return test_bit(*opcodes,
184				(unsigned long *)twobyte_is_boostable);
185	}
186
187	switch (opcode & 0xf0) {
188#ifdef CONFIG_X86_64
189	case 0x40:
190		goto retry; /* REX prefix is boostable */
191#endif
192	case 0x60:
193		if (0x63 < opcode && opcode < 0x67)
194			goto retry; /* prefixes */
195		/* can't boost Address-size override and bound */
196		return (opcode != 0x62 && opcode != 0x67);
197	case 0x70:
198		return 0; /* can't boost conditional jump */
199	case 0xc0:
200		/* can't boost software-interruptions */
201		return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
202	case 0xd0:
203		/* can boost AA* and XLAT */
204		return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
205	case 0xe0:
206		/* can boost in/out and absolute jmps */
207		return ((opcode & 0x04) || opcode == 0xea);
208	case 0xf0:
209		if ((opcode & 0x0c) == 0 && opcode != 0xf1)
210			goto retry; /* lock/rep(ne) prefix */
211		/* clear and set flags are boostable */
212		return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
213	default:
214		/* segment override prefixes are boostable */
215		if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
216			goto retry; /* prefixes */
217		/* CS override prefix and call are not boostable */
218		return (opcode != 0x2e && opcode != 0x9a);
219	}
220}
221
222static unsigned long
223__recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
224{
225	struct kprobe *kp;
226	unsigned long faddr;
227
228	kp = get_kprobe((void *)addr);
229	faddr = ftrace_location(addr);
230	/*
231	 * Addresses inside the ftrace location are refused by
232	 * arch_check_ftrace_location(). Something went terribly wrong
233	 * if such an address is checked here.
234	 */
235	if (WARN_ON(faddr && faddr != addr))
236		return 0UL;
237	/*
238	 * Use the current code if it is not modified by Kprobe
239	 * and it cannot be modified by ftrace.
240	 */
241	if (!kp && !faddr)
242		return addr;
243
244	/*
245	 * Basically, kp->ainsn.insn has an original instruction.
246	 * However, RIP-relative instruction can not do single-stepping
247	 * at different place, __copy_instruction() tweaks the displacement of
248	 * that instruction. In that case, we can't recover the instruction
249	 * from the kp->ainsn.insn.
250	 *
251	 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
252	 * of the first byte of the probed instruction, which is overwritten
253	 * by int3. And the instruction at kp->addr is not modified by kprobes
254	 * except for the first byte, we can recover the original instruction
255	 * from it and kp->opcode.
256	 *
257	 * In case of Kprobes using ftrace, we do not have a copy of
258	 * the original instruction. In fact, the ftrace location might
259	 * be modified at anytime and even could be in an inconsistent state.
260	 * Fortunately, we know that the original code is the ideal 5-byte
261	 * long NOP.
262	 */
263	memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
264	if (faddr)
265		memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
266	else
267		buf[0] = kp->opcode;
268	return (unsigned long)buf;
269}
270
271/*
272 * Recover the probed instruction at addr for further analysis.
273 * Caller must lock kprobes by kprobe_mutex, or disable preemption
274 * for preventing to release referencing kprobes.
275 * Returns zero if the instruction can not get recovered.
276 */
277unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
278{
279	unsigned long __addr;
280
281	__addr = __recover_optprobed_insn(buf, addr);
282	if (__addr != addr)
283		return __addr;
284
285	return __recover_probed_insn(buf, addr);
286}
287
288/* Check if paddr is at an instruction boundary */
289static int can_probe(unsigned long paddr)
290{
291	unsigned long addr, __addr, offset = 0;
292	struct insn insn;
293	kprobe_opcode_t buf[MAX_INSN_SIZE];
294
295	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
296		return 0;
297
298	/* Decode instructions */
299	addr = paddr - offset;
300	while (addr < paddr) {
301		/*
302		 * Check if the instruction has been modified by another
303		 * kprobe, in which case we replace the breakpoint by the
304		 * original instruction in our buffer.
305		 * Also, jump optimization will change the breakpoint to
306		 * relative-jump. Since the relative-jump itself is
307		 * normally used, we just go through if there is no kprobe.
308		 */
309		__addr = recover_probed_instruction(buf, addr);
310		if (!__addr)
311			return 0;
312		kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
313		insn_get_length(&insn);
314
315		/*
316		 * Another debugging subsystem might insert this breakpoint.
317		 * In that case, we can't recover it.
318		 */
319		if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
320			return 0;
321		addr += insn.length;
322	}
323
324	return (addr == paddr);
325}
326
327/*
328 * Returns non-zero if opcode modifies the interrupt flag.
329 */
330static int is_IF_modifier(kprobe_opcode_t *insn)
331{
332	/* Skip prefixes */
333	insn = skip_prefixes(insn);
334
335	switch (*insn) {
336	case 0xfa:		/* cli */
337	case 0xfb:		/* sti */
338	case 0xcf:		/* iret/iretd */
339	case 0x9d:		/* popf/popfd */
340		return 1;
341	}
342
343	return 0;
344}
345
346/*
347 * Copy an instruction and adjust the displacement if the instruction
348 * uses the %rip-relative addressing mode.
349 * If it does, Return the address of the 32-bit displacement word.
350 * If not, return null.
351 * Only applicable to 64-bit x86.
352 */
353int __copy_instruction(u8 *dest, u8 *src)
354{
355	struct insn insn;
356	kprobe_opcode_t buf[MAX_INSN_SIZE];
357	int length;
358	unsigned long recovered_insn =
359		recover_probed_instruction(buf, (unsigned long)src);
360
361	if (!recovered_insn)
362		return 0;
363	kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
364	insn_get_length(&insn);
365	length = insn.length;
366
367	/* Another subsystem puts a breakpoint, failed to recover */
368	if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
369		return 0;
370	memcpy(dest, insn.kaddr, length);
371
372#ifdef CONFIG_X86_64
373	if (insn_rip_relative(&insn)) {
374		s64 newdisp;
375		u8 *disp;
376		kernel_insn_init(&insn, dest, length);
377		insn_get_displacement(&insn);
378		/*
379		 * The copied instruction uses the %rip-relative addressing
380		 * mode.  Adjust the displacement for the difference between
381		 * the original location of this instruction and the location
382		 * of the copy that will actually be run.  The tricky bit here
383		 * is making sure that the sign extension happens correctly in
384		 * this calculation, since we need a signed 32-bit result to
385		 * be sign-extended to 64 bits when it's added to the %rip
386		 * value and yield the same 64-bit result that the sign-
387		 * extension of the original signed 32-bit displacement would
388		 * have given.
389		 */
390		newdisp = (u8 *) src + (s64) insn.displacement.value - (u8 *) dest;
391		if ((s64) (s32) newdisp != newdisp) {
392			pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
393			pr_err("\tSrc: %p, Dest: %p, old disp: %x\n", src, dest, insn.displacement.value);
394			return 0;
395		}
396		disp = (u8 *) dest + insn_offset_displacement(&insn);
397		*(s32 *) disp = (s32) newdisp;
398	}
399#endif
400	return length;
401}
402
403static int arch_copy_kprobe(struct kprobe *p)
404{
405	int ret;
406
407	/* Copy an instruction with recovering if other optprobe modifies it.*/
408	ret = __copy_instruction(p->ainsn.insn, p->addr);
409	if (!ret)
410		return -EINVAL;
411
412	/*
413	 * __copy_instruction can modify the displacement of the instruction,
414	 * but it doesn't affect boostable check.
415	 */
416	if (can_boost(p->ainsn.insn))
417		p->ainsn.boostable = 0;
418	else
419		p->ainsn.boostable = -1;
420
421	/* Check whether the instruction modifies Interrupt Flag or not */
422	p->ainsn.if_modifier = is_IF_modifier(p->ainsn.insn);
423
424	/* Also, displacement change doesn't affect the first byte */
425	p->opcode = p->ainsn.insn[0];
426
427	return 0;
428}
429
430int arch_prepare_kprobe(struct kprobe *p)
431{
432	if (alternatives_text_reserved(p->addr, p->addr))
433		return -EINVAL;
434
435	if (!can_probe((unsigned long)p->addr))
436		return -EILSEQ;
437	/* insn: must be on special executable page on x86. */
438	p->ainsn.insn = get_insn_slot();
439	if (!p->ainsn.insn)
440		return -ENOMEM;
441
442	return arch_copy_kprobe(p);
443}
444
445void arch_arm_kprobe(struct kprobe *p)
446{
447	text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
448}
449
450void arch_disarm_kprobe(struct kprobe *p)
451{
452	text_poke(p->addr, &p->opcode, 1);
453}
454
455void arch_remove_kprobe(struct kprobe *p)
456{
457	if (p->ainsn.insn) {
458		free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
459		p->ainsn.insn = NULL;
460	}
461}
462
463static nokprobe_inline void
464save_previous_kprobe(struct kprobe_ctlblk *kcb)
465{
466	kcb->prev_kprobe.kp = kprobe_running();
467	kcb->prev_kprobe.status = kcb->kprobe_status;
468	kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
469	kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
470}
471
472static nokprobe_inline void
473restore_previous_kprobe(struct kprobe_ctlblk *kcb)
474{
475	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
476	kcb->kprobe_status = kcb->prev_kprobe.status;
477	kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
478	kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
479}
480
481static nokprobe_inline void
482set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
483		   struct kprobe_ctlblk *kcb)
484{
485	__this_cpu_write(current_kprobe, p);
486	kcb->kprobe_saved_flags = kcb->kprobe_old_flags
487		= (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
488	if (p->ainsn.if_modifier)
489		kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
490}
491
492static nokprobe_inline void clear_btf(void)
493{
494	if (test_thread_flag(TIF_BLOCKSTEP)) {
495		unsigned long debugctl = get_debugctlmsr();
496
497		debugctl &= ~DEBUGCTLMSR_BTF;
498		update_debugctlmsr(debugctl);
499	}
500}
501
502static nokprobe_inline void restore_btf(void)
503{
504	if (test_thread_flag(TIF_BLOCKSTEP)) {
505		unsigned long debugctl = get_debugctlmsr();
506
507		debugctl |= DEBUGCTLMSR_BTF;
508		update_debugctlmsr(debugctl);
509	}
510}
511
512void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
513{
514	unsigned long *sara = stack_addr(regs);
515
516	ri->ret_addr = (kprobe_opcode_t *) *sara;
517
518	/* Replace the return addr with trampoline addr */
519	*sara = (unsigned long) &kretprobe_trampoline;
520}
521NOKPROBE_SYMBOL(arch_prepare_kretprobe);
522
523static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
524			     struct kprobe_ctlblk *kcb, int reenter)
525{
526	if (setup_detour_execution(p, regs, reenter))
527		return;
528
529#if !defined(CONFIG_PREEMPT)
530	if (p->ainsn.boostable == 1 && !p->post_handler) {
531		/* Boost up -- we can execute copied instructions directly */
532		if (!reenter)
533			reset_current_kprobe();
534		/*
535		 * Reentering boosted probe doesn't reset current_kprobe,
536		 * nor set current_kprobe, because it doesn't use single
537		 * stepping.
538		 */
539		regs->ip = (unsigned long)p->ainsn.insn;
540		preempt_enable_no_resched();
541		return;
542	}
543#endif
544	if (reenter) {
545		save_previous_kprobe(kcb);
546		set_current_kprobe(p, regs, kcb);
547		kcb->kprobe_status = KPROBE_REENTER;
548	} else
549		kcb->kprobe_status = KPROBE_HIT_SS;
550	/* Prepare real single stepping */
551	clear_btf();
552	regs->flags |= X86_EFLAGS_TF;
553	regs->flags &= ~X86_EFLAGS_IF;
554	/* single step inline if the instruction is an int3 */
555	if (p->opcode == BREAKPOINT_INSTRUCTION)
556		regs->ip = (unsigned long)p->addr;
557	else
558		regs->ip = (unsigned long)p->ainsn.insn;
559}
560NOKPROBE_SYMBOL(setup_singlestep);
561
562/*
563 * We have reentered the kprobe_handler(), since another probe was hit while
564 * within the handler. We save the original kprobes variables and just single
565 * step on the instruction of the new probe without calling any user handlers.
566 */
567static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
568			  struct kprobe_ctlblk *kcb)
569{
570	switch (kcb->kprobe_status) {
571	case KPROBE_HIT_SSDONE:
572	case KPROBE_HIT_ACTIVE:
573	case KPROBE_HIT_SS:
574		kprobes_inc_nmissed_count(p);
575		setup_singlestep(p, regs, kcb, 1);
576		break;
577	case KPROBE_REENTER:
578		/* A probe has been hit in the codepath leading up to, or just
579		 * after, single-stepping of a probed instruction. This entire
580		 * codepath should strictly reside in .kprobes.text section.
581		 * Raise a BUG or we'll continue in an endless reentering loop
582		 * and eventually a stack overflow.
583		 */
584		printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
585		       p->addr);
586		dump_kprobe(p);
587		BUG();
588	default:
589		/* impossible cases */
590		WARN_ON(1);
591		return 0;
592	}
593
594	return 1;
595}
596NOKPROBE_SYMBOL(reenter_kprobe);
597
598/*
599 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
600 * remain disabled throughout this function.
601 */
602int kprobe_int3_handler(struct pt_regs *regs)
603{
604	kprobe_opcode_t *addr;
605	struct kprobe *p;
606	struct kprobe_ctlblk *kcb;
607
608	if (user_mode(regs))
609		return 0;
610
611	addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
612	/*
613	 * We don't want to be preempted for the entire
614	 * duration of kprobe processing. We conditionally
615	 * re-enable preemption at the end of this function,
616	 * and also in reenter_kprobe() and setup_singlestep().
617	 */
618	preempt_disable();
619
620	kcb = get_kprobe_ctlblk();
621	p = get_kprobe(addr);
622
623	if (p) {
624		if (kprobe_running()) {
625			if (reenter_kprobe(p, regs, kcb))
626				return 1;
627		} else {
628			set_current_kprobe(p, regs, kcb);
629			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
630
631			/*
632			 * If we have no pre-handler or it returned 0, we
633			 * continue with normal processing.  If we have a
634			 * pre-handler and it returned non-zero, it prepped
635			 * for calling the break_handler below on re-entry
636			 * for jprobe processing, so get out doing nothing
637			 * more here.
638			 */
639			if (!p->pre_handler || !p->pre_handler(p, regs))
640				setup_singlestep(p, regs, kcb, 0);
641			return 1;
642		}
643	} else if (*addr != BREAKPOINT_INSTRUCTION) {
644		/*
645		 * The breakpoint instruction was removed right
646		 * after we hit it.  Another cpu has removed
647		 * either a probepoint or a debugger breakpoint
648		 * at this address.  In either case, no further
649		 * handling of this interrupt is appropriate.
650		 * Back up over the (now missing) int3 and run
651		 * the original instruction.
652		 */
653		regs->ip = (unsigned long)addr;
654		preempt_enable_no_resched();
655		return 1;
656	} else if (kprobe_running()) {
657		p = __this_cpu_read(current_kprobe);
658		if (p->break_handler && p->break_handler(p, regs)) {
659			if (!skip_singlestep(p, regs, kcb))
660				setup_singlestep(p, regs, kcb, 0);
661			return 1;
662		}
663	} /* else: not a kprobe fault; let the kernel handle it */
664
665	preempt_enable_no_resched();
666	return 0;
667}
668NOKPROBE_SYMBOL(kprobe_int3_handler);
669
670/*
671 * When a retprobed function returns, this code saves registers and
672 * calls trampoline_handler() runs, which calls the kretprobe's handler.
673 */
674static void __used kretprobe_trampoline_holder(void)
675{
676	asm volatile (
677			".global kretprobe_trampoline\n"
678			"kretprobe_trampoline: \n"
679#ifdef CONFIG_X86_64
680			/* We don't bother saving the ss register */
681			"	pushq %rsp\n"
682			"	pushfq\n"
683			SAVE_REGS_STRING
684			"	movq %rsp, %rdi\n"
685			"	call trampoline_handler\n"
686			/* Replace saved sp with true return address. */
687			"	movq %rax, 152(%rsp)\n"
688			RESTORE_REGS_STRING
689			"	popfq\n"
690#else
691			"	pushf\n"
692			SAVE_REGS_STRING
693			"	movl %esp, %eax\n"
694			"	call trampoline_handler\n"
695			/* Move flags to cs */
696			"	movl 56(%esp), %edx\n"
697			"	movl %edx, 52(%esp)\n"
698			/* Replace saved flags with true return address. */
699			"	movl %eax, 56(%esp)\n"
700			RESTORE_REGS_STRING
701			"	popf\n"
702#endif
703			"	ret\n");
704}
705NOKPROBE_SYMBOL(kretprobe_trampoline_holder);
706NOKPROBE_SYMBOL(kretprobe_trampoline);
707
708/*
709 * Called from kretprobe_trampoline
710 */
711__visible __used void *trampoline_handler(struct pt_regs *regs)
712{
713	struct kretprobe_instance *ri = NULL;
714	struct hlist_head *head, empty_rp;
715	struct hlist_node *tmp;
716	unsigned long flags, orig_ret_address = 0;
717	unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
718	kprobe_opcode_t *correct_ret_addr = NULL;
719
720	INIT_HLIST_HEAD(&empty_rp);
721	kretprobe_hash_lock(current, &head, &flags);
722	/* fixup registers */
723#ifdef CONFIG_X86_64
724	regs->cs = __KERNEL_CS;
725#else
726	regs->cs = __KERNEL_CS | get_kernel_rpl();
727	regs->gs = 0;
728#endif
729	regs->ip = trampoline_address;
730	regs->orig_ax = ~0UL;
731
732	/*
733	 * It is possible to have multiple instances associated with a given
734	 * task either because multiple functions in the call path have
735	 * return probes installed on them, and/or more than one
736	 * return probe was registered for a target function.
737	 *
738	 * We can handle this because:
739	 *     - instances are always pushed into the head of the list
740	 *     - when multiple return probes are registered for the same
741	 *	 function, the (chronologically) first instance's ret_addr
742	 *	 will be the real return address, and all the rest will
743	 *	 point to kretprobe_trampoline.
744	 */
745	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
746		if (ri->task != current)
747			/* another task is sharing our hash bucket */
748			continue;
749
750		orig_ret_address = (unsigned long)ri->ret_addr;
751
752		if (orig_ret_address != trampoline_address)
753			/*
754			 * This is the real return address. Any other
755			 * instances associated with this task are for
756			 * other calls deeper on the call stack
757			 */
758			break;
759	}
760
761	kretprobe_assert(ri, orig_ret_address, trampoline_address);
762
763	correct_ret_addr = ri->ret_addr;
764	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
765		if (ri->task != current)
766			/* another task is sharing our hash bucket */
767			continue;
768
769		orig_ret_address = (unsigned long)ri->ret_addr;
770		if (ri->rp && ri->rp->handler) {
771			__this_cpu_write(current_kprobe, &ri->rp->kp);
772			get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
773			ri->ret_addr = correct_ret_addr;
774			ri->rp->handler(ri, regs);
775			__this_cpu_write(current_kprobe, NULL);
776		}
777
778		recycle_rp_inst(ri, &empty_rp);
779
780		if (orig_ret_address != trampoline_address)
781			/*
782			 * This is the real return address. Any other
783			 * instances associated with this task are for
784			 * other calls deeper on the call stack
785			 */
786			break;
787	}
788
789	kretprobe_hash_unlock(current, &flags);
790
791	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
792		hlist_del(&ri->hlist);
793		kfree(ri);
794	}
795	return (void *)orig_ret_address;
796}
797NOKPROBE_SYMBOL(trampoline_handler);
798
799/*
800 * Called after single-stepping.  p->addr is the address of the
801 * instruction whose first byte has been replaced by the "int 3"
802 * instruction.  To avoid the SMP problems that can occur when we
803 * temporarily put back the original opcode to single-step, we
804 * single-stepped a copy of the instruction.  The address of this
805 * copy is p->ainsn.insn.
806 *
807 * This function prepares to return from the post-single-step
808 * interrupt.  We have to fix up the stack as follows:
809 *
810 * 0) Except in the case of absolute or indirect jump or call instructions,
811 * the new ip is relative to the copied instruction.  We need to make
812 * it relative to the original instruction.
813 *
814 * 1) If the single-stepped instruction was pushfl, then the TF and IF
815 * flags are set in the just-pushed flags, and may need to be cleared.
816 *
817 * 2) If the single-stepped instruction was a call, the return address
818 * that is atop the stack is the address following the copied instruction.
819 * We need to make it the address following the original instruction.
820 *
821 * If this is the first time we've single-stepped the instruction at
822 * this probepoint, and the instruction is boostable, boost it: add a
823 * jump instruction after the copied instruction, that jumps to the next
824 * instruction after the probepoint.
825 */
826static void resume_execution(struct kprobe *p, struct pt_regs *regs,
827			     struct kprobe_ctlblk *kcb)
828{
829	unsigned long *tos = stack_addr(regs);
830	unsigned long copy_ip = (unsigned long)p->ainsn.insn;
831	unsigned long orig_ip = (unsigned long)p->addr;
832	kprobe_opcode_t *insn = p->ainsn.insn;
833
834	/* Skip prefixes */
835	insn = skip_prefixes(insn);
836
837	regs->flags &= ~X86_EFLAGS_TF;
838	switch (*insn) {
839	case 0x9c:	/* pushfl */
840		*tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
841		*tos |= kcb->kprobe_old_flags;
842		break;
843	case 0xc2:	/* iret/ret/lret */
844	case 0xc3:
845	case 0xca:
846	case 0xcb:
847	case 0xcf:
848	case 0xea:	/* jmp absolute -- ip is correct */
849		/* ip is already adjusted, no more changes required */
850		p->ainsn.boostable = 1;
851		goto no_change;
852	case 0xe8:	/* call relative - Fix return addr */
853		*tos = orig_ip + (*tos - copy_ip);
854		break;
855#ifdef CONFIG_X86_32
856	case 0x9a:	/* call absolute -- same as call absolute, indirect */
857		*tos = orig_ip + (*tos - copy_ip);
858		goto no_change;
859#endif
860	case 0xff:
861		if ((insn[1] & 0x30) == 0x10) {
862			/*
863			 * call absolute, indirect
864			 * Fix return addr; ip is correct.
865			 * But this is not boostable
866			 */
867			*tos = orig_ip + (*tos - copy_ip);
868			goto no_change;
869		} else if (((insn[1] & 0x31) == 0x20) ||
870			   ((insn[1] & 0x31) == 0x21)) {
871			/*
872			 * jmp near and far, absolute indirect
873			 * ip is correct. And this is boostable
874			 */
875			p->ainsn.boostable = 1;
876			goto no_change;
877		}
878	default:
879		break;
880	}
881
882	if (p->ainsn.boostable == 0) {
883		if ((regs->ip > copy_ip) &&
884		    (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
885			/*
886			 * These instructions can be executed directly if it
887			 * jumps back to correct address.
888			 */
889			synthesize_reljump((void *)regs->ip,
890				(void *)orig_ip + (regs->ip - copy_ip));
891			p->ainsn.boostable = 1;
892		} else {
893			p->ainsn.boostable = -1;
894		}
895	}
896
897	regs->ip += orig_ip - copy_ip;
898
899no_change:
900	restore_btf();
901}
902NOKPROBE_SYMBOL(resume_execution);
903
904/*
905 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
906 * remain disabled throughout this function.
907 */
908int kprobe_debug_handler(struct pt_regs *regs)
909{
910	struct kprobe *cur = kprobe_running();
911	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
912
913	if (!cur)
914		return 0;
915
916	resume_execution(cur, regs, kcb);
917	regs->flags |= kcb->kprobe_saved_flags;
918
919	if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
920		kcb->kprobe_status = KPROBE_HIT_SSDONE;
921		cur->post_handler(cur, regs, 0);
922	}
923
924	/* Restore back the original saved kprobes variables and continue. */
925	if (kcb->kprobe_status == KPROBE_REENTER) {
926		restore_previous_kprobe(kcb);
927		goto out;
928	}
929	reset_current_kprobe();
930out:
931	preempt_enable_no_resched();
932
933	/*
934	 * if somebody else is singlestepping across a probe point, flags
935	 * will have TF set, in which case, continue the remaining processing
936	 * of do_debug, as if this is not a probe hit.
937	 */
938	if (regs->flags & X86_EFLAGS_TF)
939		return 0;
940
941	return 1;
942}
943NOKPROBE_SYMBOL(kprobe_debug_handler);
944
945int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
946{
947	struct kprobe *cur = kprobe_running();
948	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
949
950	if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
951		/* This must happen on single-stepping */
952		WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
953			kcb->kprobe_status != KPROBE_REENTER);
954		/*
955		 * We are here because the instruction being single
956		 * stepped caused a page fault. We reset the current
957		 * kprobe and the ip points back to the probe address
958		 * and allow the page fault handler to continue as a
959		 * normal page fault.
960		 */
961		regs->ip = (unsigned long)cur->addr;
962		regs->flags |= kcb->kprobe_old_flags;
963		if (kcb->kprobe_status == KPROBE_REENTER)
964			restore_previous_kprobe(kcb);
965		else
966			reset_current_kprobe();
967		preempt_enable_no_resched();
968	} else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
969		   kcb->kprobe_status == KPROBE_HIT_SSDONE) {
970		/*
971		 * We increment the nmissed count for accounting,
972		 * we can also use npre/npostfault count for accounting
973		 * these specific fault cases.
974		 */
975		kprobes_inc_nmissed_count(cur);
976
977		/*
978		 * We come here because instructions in the pre/post
979		 * handler caused the page_fault, this could happen
980		 * if handler tries to access user space by
981		 * copy_from_user(), get_user() etc. Let the
982		 * user-specified handler try to fix it first.
983		 */
984		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
985			return 1;
986
987		/*
988		 * In case the user-specified fault handler returned
989		 * zero, try to fix up.
990		 */
991		if (fixup_exception(regs))
992			return 1;
993
994		/*
995		 * fixup routine could not handle it,
996		 * Let do_page_fault() fix it.
997		 */
998	}
999
1000	return 0;
1001}
1002NOKPROBE_SYMBOL(kprobe_fault_handler);
1003
1004/*
1005 * Wrapper routine for handling exceptions.
1006 */
1007int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
1008			     void *data)
1009{
1010	struct die_args *args = data;
1011	int ret = NOTIFY_DONE;
1012
1013	if (args->regs && user_mode(args->regs))
1014		return ret;
1015
1016	if (val == DIE_GPF) {
1017		/*
1018		 * To be potentially processing a kprobe fault and to
1019		 * trust the result from kprobe_running(), we have
1020		 * be non-preemptible.
1021		 */
1022		if (!preemptible() && kprobe_running() &&
1023		    kprobe_fault_handler(args->regs, args->trapnr))
1024			ret = NOTIFY_STOP;
1025	}
1026	return ret;
1027}
1028NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1029
1030int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
1031{
1032	struct jprobe *jp = container_of(p, struct jprobe, kp);
1033	unsigned long addr;
1034	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1035
1036	kcb->jprobe_saved_regs = *regs;
1037	kcb->jprobe_saved_sp = stack_addr(regs);
1038	addr = (unsigned long)(kcb->jprobe_saved_sp);
1039
1040	/*
1041	 * As Linus pointed out, gcc assumes that the callee
1042	 * owns the argument space and could overwrite it, e.g.
1043	 * tailcall optimization. So, to be absolutely safe
1044	 * we also save and restore enough stack bytes to cover
1045	 * the argument area.
1046	 */
1047	memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1048	       MIN_STACK_SIZE(addr));
1049	regs->flags &= ~X86_EFLAGS_IF;
1050	trace_hardirqs_off();
1051	regs->ip = (unsigned long)(jp->entry);
1052
1053	/*
1054	 * jprobes use jprobe_return() which skips the normal return
1055	 * path of the function, and this messes up the accounting of the
1056	 * function graph tracer to get messed up.
1057	 *
1058	 * Pause function graph tracing while performing the jprobe function.
1059	 */
1060	pause_graph_tracing();
1061	return 1;
1062}
1063NOKPROBE_SYMBOL(setjmp_pre_handler);
1064
1065void jprobe_return(void)
1066{
1067	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1068
1069	asm volatile (
1070#ifdef CONFIG_X86_64
1071			"       xchg   %%rbx,%%rsp	\n"
1072#else
1073			"       xchgl   %%ebx,%%esp	\n"
1074#endif
1075			"       int3			\n"
1076			"       .globl jprobe_return_end\n"
1077			"       jprobe_return_end:	\n"
1078			"       nop			\n"::"b"
1079			(kcb->jprobe_saved_sp):"memory");
1080}
1081NOKPROBE_SYMBOL(jprobe_return);
1082NOKPROBE_SYMBOL(jprobe_return_end);
1083
1084int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1085{
1086	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1087	u8 *addr = (u8 *) (regs->ip - 1);
1088	struct jprobe *jp = container_of(p, struct jprobe, kp);
1089	void *saved_sp = kcb->jprobe_saved_sp;
1090
1091	if ((addr > (u8 *) jprobe_return) &&
1092	    (addr < (u8 *) jprobe_return_end)) {
1093		if (stack_addr(regs) != saved_sp) {
1094			struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1095			printk(KERN_ERR
1096			       "current sp %p does not match saved sp %p\n",
1097			       stack_addr(regs), saved_sp);
1098			printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1099			show_regs(saved_regs);
1100			printk(KERN_ERR "Current registers\n");
1101			show_regs(regs);
1102			BUG();
1103		}
1104		/* It's OK to start function graph tracing again */
1105		unpause_graph_tracing();
1106		*regs = kcb->jprobe_saved_regs;
1107		memcpy(saved_sp, kcb->jprobes_stack, MIN_STACK_SIZE(saved_sp));
1108		preempt_enable_no_resched();
1109		return 1;
1110	}
1111	return 0;
1112}
1113NOKPROBE_SYMBOL(longjmp_break_handler);
1114
1115bool arch_within_kprobe_blacklist(unsigned long addr)
1116{
1117	return  (addr >= (unsigned long)__kprobes_text_start &&
1118		 addr < (unsigned long)__kprobes_text_end) ||
1119		(addr >= (unsigned long)__entry_text_start &&
1120		 addr < (unsigned long)__entry_text_end);
1121}
1122
1123int __init arch_init_kprobes(void)
1124{
1125	return 0;
1126}
1127
1128int arch_trampoline_kprobe(struct kprobe *p)
1129{
1130	return 0;
1131}
1132