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/kgdb.h>
41 #include <linux/ftrace.h>
42 #include <linux/kasan.h>
43 #include <linux/moduleloader.h>
44 #include <linux/objtool.h>
45 #include <linux/vmalloc.h>
46 #include <linux/pgtable.h>
47 #include <linux/set_memory.h>
48
49 #include <asm/text-patching.h>
50 #include <asm/cacheflush.h>
51 #include <asm/desc.h>
52 #include <linux/uaccess.h>
53 #include <asm/alternative.h>
54 #include <asm/insn.h>
55 #include <asm/debugreg.h>
56 #include <asm/ibt.h>
57
58 #include "common.h"
59
60 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
61 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
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
__synthesize_relative_insn(void * dest,void * from,void * to,u8 op)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'.*/
synthesize_reljump(void * dest,void * from,void * 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'.*/
synthesize_relcall(void * dest,void * from,void * 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 */
can_boost(struct insn * insn,void * addr)139 int can_boost(struct insn *insn, void *addr)
140 {
141 kprobe_opcode_t opcode;
142 insn_byte_t prefix;
143 int i;
144
145 if (search_exception_tables((unsigned long)addr))
146 return 0; /* Page fault may occur on this address. */
147
148 /* 2nd-byte opcode */
149 if (insn->opcode.nbytes == 2)
150 return test_bit(insn->opcode.bytes[1],
151 (unsigned long *)twobyte_is_boostable);
152
153 if (insn->opcode.nbytes != 1)
154 return 0;
155
156 for_each_insn_prefix(insn, i, prefix) {
157 insn_attr_t attr;
158
159 attr = inat_get_opcode_attribute(prefix);
160 /* Can't boost Address-size override prefix and CS override prefix */
161 if (prefix == 0x2e || inat_is_address_size_prefix(attr))
162 return 0;
163 }
164
165 opcode = insn->opcode.bytes[0];
166
167 switch (opcode) {
168 case 0x62: /* bound */
169 case 0x70 ... 0x7f: /* Conditional jumps */
170 case 0x9a: /* Call far */
171 case 0xc0 ... 0xc1: /* Grp2 */
172 case 0xcc ... 0xce: /* software exceptions */
173 case 0xd0 ... 0xd3: /* Grp2 */
174 case 0xd6: /* (UD) */
175 case 0xd8 ... 0xdf: /* ESC */
176 case 0xe0 ... 0xe3: /* LOOP*, JCXZ */
177 case 0xe8 ... 0xe9: /* near Call, JMP */
178 case 0xeb: /* Short JMP */
179 case 0xf0 ... 0xf4: /* LOCK/REP, HLT */
180 case 0xf6 ... 0xf7: /* Grp3 */
181 case 0xfe: /* Grp4 */
182 /* ... are not boostable */
183 return 0;
184 case 0xff: /* Grp5 */
185 /* Only indirect jmp is boostable */
186 return X86_MODRM_REG(insn->modrm.bytes[0]) == 4;
187 default:
188 return 1;
189 }
190 }
191
192 static unsigned long
__recover_probed_insn(kprobe_opcode_t * buf,unsigned long addr)193 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
194 {
195 struct kprobe *kp;
196 bool faddr;
197
198 kp = get_kprobe((void *)addr);
199 faddr = ftrace_location(addr) == addr;
200 /*
201 * Use the current code if it is not modified by Kprobe
202 * and it cannot be modified by ftrace.
203 */
204 if (!kp && !faddr)
205 return addr;
206
207 /*
208 * Basically, kp->ainsn.insn has an original instruction.
209 * However, RIP-relative instruction can not do single-stepping
210 * at different place, __copy_instruction() tweaks the displacement of
211 * that instruction. In that case, we can't recover the instruction
212 * from the kp->ainsn.insn.
213 *
214 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
215 * of the first byte of the probed instruction, which is overwritten
216 * by int3. And the instruction at kp->addr is not modified by kprobes
217 * except for the first byte, we can recover the original instruction
218 * from it and kp->opcode.
219 *
220 * In case of Kprobes using ftrace, we do not have a copy of
221 * the original instruction. In fact, the ftrace location might
222 * be modified at anytime and even could be in an inconsistent state.
223 * Fortunately, we know that the original code is the ideal 5-byte
224 * long NOP.
225 */
226 if (copy_from_kernel_nofault(buf, (void *)addr,
227 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
228 return 0UL;
229
230 if (faddr)
231 memcpy(buf, x86_nops[5], 5);
232 else
233 buf[0] = kp->opcode;
234 return (unsigned long)buf;
235 }
236
237 /*
238 * Recover the probed instruction at addr for further analysis.
239 * Caller must lock kprobes by kprobe_mutex, or disable preemption
240 * for preventing to release referencing kprobes.
241 * Returns zero if the instruction can not get recovered (or access failed).
242 */
recover_probed_instruction(kprobe_opcode_t * buf,unsigned long addr)243 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
244 {
245 unsigned long __addr;
246
247 __addr = __recover_optprobed_insn(buf, addr);
248 if (__addr != addr)
249 return __addr;
250
251 return __recover_probed_insn(buf, addr);
252 }
253
254 /* Check if paddr is at an instruction boundary */
can_probe(unsigned long paddr)255 static int can_probe(unsigned long paddr)
256 {
257 unsigned long addr, __addr, offset = 0;
258 struct insn insn;
259 kprobe_opcode_t buf[MAX_INSN_SIZE];
260
261 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
262 return 0;
263
264 /* Decode instructions */
265 addr = paddr - offset;
266 while (addr < paddr) {
267 int ret;
268
269 /*
270 * Check if the instruction has been modified by another
271 * kprobe, in which case we replace the breakpoint by the
272 * original instruction in our buffer.
273 * Also, jump optimization will change the breakpoint to
274 * relative-jump. Since the relative-jump itself is
275 * normally used, we just go through if there is no kprobe.
276 */
277 __addr = recover_probed_instruction(buf, addr);
278 if (!__addr)
279 return 0;
280
281 ret = insn_decode_kernel(&insn, (void *)__addr);
282 if (ret < 0)
283 return 0;
284
285 #ifdef CONFIG_KGDB
286 /*
287 * If there is a dynamically installed kgdb sw breakpoint,
288 * this function should not be probed.
289 */
290 if (insn.opcode.bytes[0] == INT3_INSN_OPCODE &&
291 kgdb_has_hit_break(addr))
292 return 0;
293 #endif
294 addr += insn.length;
295 }
296
297 return (addr == paddr);
298 }
299
300 /* If x86 supports IBT (ENDBR) it must be skipped. */
arch_adjust_kprobe_addr(unsigned long addr,unsigned long offset,bool * on_func_entry)301 kprobe_opcode_t *arch_adjust_kprobe_addr(unsigned long addr, unsigned long offset,
302 bool *on_func_entry)
303 {
304 if (is_endbr(*(u32 *)addr)) {
305 *on_func_entry = !offset || offset == 4;
306 if (*on_func_entry)
307 offset = 4;
308
309 } else {
310 *on_func_entry = !offset;
311 }
312
313 return (kprobe_opcode_t *)(addr + offset);
314 }
315
316 /*
317 * Copy an instruction with recovering modified instruction by kprobes
318 * and adjust the displacement if the instruction uses the %rip-relative
319 * addressing mode. Note that since @real will be the final place of copied
320 * instruction, displacement must be adjust by @real, not @dest.
321 * This returns the length of copied instruction, or 0 if it has an error.
322 */
__copy_instruction(u8 * dest,u8 * src,u8 * real,struct insn * insn)323 int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn)
324 {
325 kprobe_opcode_t buf[MAX_INSN_SIZE];
326 unsigned long recovered_insn = recover_probed_instruction(buf, (unsigned long)src);
327 int ret;
328
329 if (!recovered_insn || !insn)
330 return 0;
331
332 /* This can access kernel text if given address is not recovered */
333 if (copy_from_kernel_nofault(dest, (void *)recovered_insn,
334 MAX_INSN_SIZE))
335 return 0;
336
337 ret = insn_decode_kernel(insn, dest);
338 if (ret < 0)
339 return 0;
340
341 /* We can not probe force emulate prefixed instruction */
342 if (insn_has_emulate_prefix(insn))
343 return 0;
344
345 /* Another subsystem puts a breakpoint, failed to recover */
346 if (insn->opcode.bytes[0] == INT3_INSN_OPCODE)
347 return 0;
348
349 /* We should not singlestep on the exception masking instructions */
350 if (insn_masking_exception(insn))
351 return 0;
352
353 #ifdef CONFIG_X86_64
354 /* Only x86_64 has RIP relative instructions */
355 if (insn_rip_relative(insn)) {
356 s64 newdisp;
357 u8 *disp;
358 /*
359 * The copied instruction uses the %rip-relative addressing
360 * mode. Adjust the displacement for the difference between
361 * the original location of this instruction and the location
362 * of the copy that will actually be run. The tricky bit here
363 * is making sure that the sign extension happens correctly in
364 * this calculation, since we need a signed 32-bit result to
365 * be sign-extended to 64 bits when it's added to the %rip
366 * value and yield the same 64-bit result that the sign-
367 * extension of the original signed 32-bit displacement would
368 * have given.
369 */
370 newdisp = (u8 *) src + (s64) insn->displacement.value
371 - (u8 *) real;
372 if ((s64) (s32) newdisp != newdisp) {
373 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
374 return 0;
375 }
376 disp = (u8 *) dest + insn_offset_displacement(insn);
377 *(s32 *) disp = (s32) newdisp;
378 }
379 #endif
380 return insn->length;
381 }
382
383 /* Prepare reljump or int3 right after instruction */
prepare_singlestep(kprobe_opcode_t * buf,struct kprobe * p,struct insn * insn)384 static int prepare_singlestep(kprobe_opcode_t *buf, struct kprobe *p,
385 struct insn *insn)
386 {
387 int len = insn->length;
388
389 if (!IS_ENABLED(CONFIG_PREEMPTION) &&
390 !p->post_handler && can_boost(insn, p->addr) &&
391 MAX_INSN_SIZE - len >= JMP32_INSN_SIZE) {
392 /*
393 * These instructions can be executed directly if it
394 * jumps back to correct address.
395 */
396 synthesize_reljump(buf + len, p->ainsn.insn + len,
397 p->addr + insn->length);
398 len += JMP32_INSN_SIZE;
399 p->ainsn.boostable = 1;
400 } else {
401 /* Otherwise, put an int3 for trapping singlestep */
402 if (MAX_INSN_SIZE - len < INT3_INSN_SIZE)
403 return -ENOSPC;
404
405 buf[len] = INT3_INSN_OPCODE;
406 len += INT3_INSN_SIZE;
407 }
408
409 return len;
410 }
411
412 /* Make page to RO mode when allocate it */
alloc_insn_page(void)413 void *alloc_insn_page(void)
414 {
415 void *page;
416
417 page = module_alloc(PAGE_SIZE);
418 if (!page)
419 return NULL;
420
421 /*
422 * TODO: Once additional kernel code protection mechanisms are set, ensure
423 * that the page was not maliciously altered and it is still zeroed.
424 */
425 set_memory_rox((unsigned long)page, 1);
426
427 return page;
428 }
429
430 /* Kprobe x86 instruction emulation - only regs->ip or IF flag modifiers */
431
kprobe_emulate_ifmodifiers(struct kprobe * p,struct pt_regs * regs)432 static void kprobe_emulate_ifmodifiers(struct kprobe *p, struct pt_regs *regs)
433 {
434 switch (p->ainsn.opcode) {
435 case 0xfa: /* cli */
436 regs->flags &= ~(X86_EFLAGS_IF);
437 break;
438 case 0xfb: /* sti */
439 regs->flags |= X86_EFLAGS_IF;
440 break;
441 case 0x9c: /* pushf */
442 int3_emulate_push(regs, regs->flags);
443 break;
444 case 0x9d: /* popf */
445 regs->flags = int3_emulate_pop(regs);
446 break;
447 }
448 regs->ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
449 }
450 NOKPROBE_SYMBOL(kprobe_emulate_ifmodifiers);
451
kprobe_emulate_ret(struct kprobe * p,struct pt_regs * regs)452 static void kprobe_emulate_ret(struct kprobe *p, struct pt_regs *regs)
453 {
454 int3_emulate_ret(regs);
455 }
456 NOKPROBE_SYMBOL(kprobe_emulate_ret);
457
kprobe_emulate_call(struct kprobe * p,struct pt_regs * regs)458 static void kprobe_emulate_call(struct kprobe *p, struct pt_regs *regs)
459 {
460 unsigned long func = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
461
462 func += p->ainsn.rel32;
463 int3_emulate_call(regs, func);
464 }
465 NOKPROBE_SYMBOL(kprobe_emulate_call);
466
kprobe_emulate_jmp(struct kprobe * p,struct pt_regs * regs)467 static void kprobe_emulate_jmp(struct kprobe *p, struct pt_regs *regs)
468 {
469 unsigned long ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
470
471 ip += p->ainsn.rel32;
472 int3_emulate_jmp(regs, ip);
473 }
474 NOKPROBE_SYMBOL(kprobe_emulate_jmp);
475
kprobe_emulate_jcc(struct kprobe * p,struct pt_regs * regs)476 static void kprobe_emulate_jcc(struct kprobe *p, struct pt_regs *regs)
477 {
478 unsigned long ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
479
480 int3_emulate_jcc(regs, p->ainsn.jcc.type, ip, p->ainsn.rel32);
481 }
482 NOKPROBE_SYMBOL(kprobe_emulate_jcc);
483
kprobe_emulate_loop(struct kprobe * p,struct pt_regs * regs)484 static void kprobe_emulate_loop(struct kprobe *p, struct pt_regs *regs)
485 {
486 unsigned long ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
487 bool match;
488
489 if (p->ainsn.loop.type != 3) { /* LOOP* */
490 if (p->ainsn.loop.asize == 32)
491 match = ((*(u32 *)®s->cx)--) != 0;
492 #ifdef CONFIG_X86_64
493 else if (p->ainsn.loop.asize == 64)
494 match = ((*(u64 *)®s->cx)--) != 0;
495 #endif
496 else
497 match = ((*(u16 *)®s->cx)--) != 0;
498 } else { /* JCXZ */
499 if (p->ainsn.loop.asize == 32)
500 match = *(u32 *)(®s->cx) == 0;
501 #ifdef CONFIG_X86_64
502 else if (p->ainsn.loop.asize == 64)
503 match = *(u64 *)(®s->cx) == 0;
504 #endif
505 else
506 match = *(u16 *)(®s->cx) == 0;
507 }
508
509 if (p->ainsn.loop.type == 0) /* LOOPNE */
510 match = match && !(regs->flags & X86_EFLAGS_ZF);
511 else if (p->ainsn.loop.type == 1) /* LOOPE */
512 match = match && (regs->flags & X86_EFLAGS_ZF);
513
514 if (match)
515 ip += p->ainsn.rel32;
516 int3_emulate_jmp(regs, ip);
517 }
518 NOKPROBE_SYMBOL(kprobe_emulate_loop);
519
520 static const int addrmode_regoffs[] = {
521 offsetof(struct pt_regs, ax),
522 offsetof(struct pt_regs, cx),
523 offsetof(struct pt_regs, dx),
524 offsetof(struct pt_regs, bx),
525 offsetof(struct pt_regs, sp),
526 offsetof(struct pt_regs, bp),
527 offsetof(struct pt_regs, si),
528 offsetof(struct pt_regs, di),
529 #ifdef CONFIG_X86_64
530 offsetof(struct pt_regs, r8),
531 offsetof(struct pt_regs, r9),
532 offsetof(struct pt_regs, r10),
533 offsetof(struct pt_regs, r11),
534 offsetof(struct pt_regs, r12),
535 offsetof(struct pt_regs, r13),
536 offsetof(struct pt_regs, r14),
537 offsetof(struct pt_regs, r15),
538 #endif
539 };
540
kprobe_emulate_call_indirect(struct kprobe * p,struct pt_regs * regs)541 static void kprobe_emulate_call_indirect(struct kprobe *p, struct pt_regs *regs)
542 {
543 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
544
545 int3_emulate_call(regs, regs_get_register(regs, offs));
546 }
547 NOKPROBE_SYMBOL(kprobe_emulate_call_indirect);
548
kprobe_emulate_jmp_indirect(struct kprobe * p,struct pt_regs * regs)549 static void kprobe_emulate_jmp_indirect(struct kprobe *p, struct pt_regs *regs)
550 {
551 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
552
553 int3_emulate_jmp(regs, regs_get_register(regs, offs));
554 }
555 NOKPROBE_SYMBOL(kprobe_emulate_jmp_indirect);
556
prepare_emulation(struct kprobe * p,struct insn * insn)557 static int prepare_emulation(struct kprobe *p, struct insn *insn)
558 {
559 insn_byte_t opcode = insn->opcode.bytes[0];
560
561 switch (opcode) {
562 case 0xfa: /* cli */
563 case 0xfb: /* sti */
564 case 0x9c: /* pushfl */
565 case 0x9d: /* popf/popfd */
566 /*
567 * IF modifiers must be emulated since it will enable interrupt while
568 * int3 single stepping.
569 */
570 p->ainsn.emulate_op = kprobe_emulate_ifmodifiers;
571 p->ainsn.opcode = opcode;
572 break;
573 case 0xc2: /* ret/lret */
574 case 0xc3:
575 case 0xca:
576 case 0xcb:
577 p->ainsn.emulate_op = kprobe_emulate_ret;
578 break;
579 case 0x9a: /* far call absolute -- segment is not supported */
580 case 0xea: /* far jmp absolute -- segment is not supported */
581 case 0xcc: /* int3 */
582 case 0xcf: /* iret -- in-kernel IRET is not supported */
583 return -EOPNOTSUPP;
584 break;
585 case 0xe8: /* near call relative */
586 p->ainsn.emulate_op = kprobe_emulate_call;
587 if (insn->immediate.nbytes == 2)
588 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
589 else
590 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
591 break;
592 case 0xeb: /* short jump relative */
593 case 0xe9: /* near jump relative */
594 p->ainsn.emulate_op = kprobe_emulate_jmp;
595 if (insn->immediate.nbytes == 1)
596 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
597 else if (insn->immediate.nbytes == 2)
598 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
599 else
600 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
601 break;
602 case 0x70 ... 0x7f:
603 /* 1 byte conditional jump */
604 p->ainsn.emulate_op = kprobe_emulate_jcc;
605 p->ainsn.jcc.type = opcode & 0xf;
606 p->ainsn.rel32 = insn->immediate.value;
607 break;
608 case 0x0f:
609 opcode = insn->opcode.bytes[1];
610 if ((opcode & 0xf0) == 0x80) {
611 /* 2 bytes Conditional Jump */
612 p->ainsn.emulate_op = kprobe_emulate_jcc;
613 p->ainsn.jcc.type = opcode & 0xf;
614 if (insn->immediate.nbytes == 2)
615 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
616 else
617 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
618 } else if (opcode == 0x01 &&
619 X86_MODRM_REG(insn->modrm.bytes[0]) == 0 &&
620 X86_MODRM_MOD(insn->modrm.bytes[0]) == 3) {
621 /* VM extensions - not supported */
622 return -EOPNOTSUPP;
623 }
624 break;
625 case 0xe0: /* Loop NZ */
626 case 0xe1: /* Loop */
627 case 0xe2: /* Loop */
628 case 0xe3: /* J*CXZ */
629 p->ainsn.emulate_op = kprobe_emulate_loop;
630 p->ainsn.loop.type = opcode & 0x3;
631 p->ainsn.loop.asize = insn->addr_bytes * 8;
632 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
633 break;
634 case 0xff:
635 /*
636 * Since the 0xff is an extended group opcode, the instruction
637 * is determined by the MOD/RM byte.
638 */
639 opcode = insn->modrm.bytes[0];
640 switch (X86_MODRM_REG(opcode)) {
641 case 0b010: /* FF /2, call near, absolute indirect */
642 p->ainsn.emulate_op = kprobe_emulate_call_indirect;
643 break;
644 case 0b100: /* FF /4, jmp near, absolute indirect */
645 p->ainsn.emulate_op = kprobe_emulate_jmp_indirect;
646 break;
647 case 0b011: /* FF /3, call far, absolute indirect */
648 case 0b101: /* FF /5, jmp far, absolute indirect */
649 return -EOPNOTSUPP;
650 }
651
652 if (!p->ainsn.emulate_op)
653 break;
654
655 if (insn->addr_bytes != sizeof(unsigned long))
656 return -EOPNOTSUPP; /* Don't support different size */
657 if (X86_MODRM_MOD(opcode) != 3)
658 return -EOPNOTSUPP; /* TODO: support memory addressing */
659
660 p->ainsn.indirect.reg = X86_MODRM_RM(opcode);
661 #ifdef CONFIG_X86_64
662 if (X86_REX_B(insn->rex_prefix.value))
663 p->ainsn.indirect.reg += 8;
664 #endif
665 break;
666 default:
667 break;
668 }
669 p->ainsn.size = insn->length;
670
671 return 0;
672 }
673
arch_copy_kprobe(struct kprobe * p)674 static int arch_copy_kprobe(struct kprobe *p)
675 {
676 struct insn insn;
677 kprobe_opcode_t buf[MAX_INSN_SIZE];
678 int ret, len;
679
680 /* Copy an instruction with recovering if other optprobe modifies it.*/
681 len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn);
682 if (!len)
683 return -EINVAL;
684
685 /* Analyze the opcode and setup emulate functions */
686 ret = prepare_emulation(p, &insn);
687 if (ret < 0)
688 return ret;
689
690 /* Add int3 for single-step or booster jmp */
691 len = prepare_singlestep(buf, p, &insn);
692 if (len < 0)
693 return len;
694
695 /* Also, displacement change doesn't affect the first byte */
696 p->opcode = buf[0];
697
698 p->ainsn.tp_len = len;
699 perf_event_text_poke(p->ainsn.insn, NULL, 0, buf, len);
700
701 /* OK, write back the instruction(s) into ROX insn buffer */
702 text_poke(p->ainsn.insn, buf, len);
703
704 return 0;
705 }
706
arch_prepare_kprobe(struct kprobe * p)707 int arch_prepare_kprobe(struct kprobe *p)
708 {
709 int ret;
710
711 if (alternatives_text_reserved(p->addr, p->addr))
712 return -EINVAL;
713
714 if (!can_probe((unsigned long)p->addr))
715 return -EILSEQ;
716
717 memset(&p->ainsn, 0, sizeof(p->ainsn));
718
719 /* insn: must be on special executable page on x86. */
720 p->ainsn.insn = get_insn_slot();
721 if (!p->ainsn.insn)
722 return -ENOMEM;
723
724 ret = arch_copy_kprobe(p);
725 if (ret) {
726 free_insn_slot(p->ainsn.insn, 0);
727 p->ainsn.insn = NULL;
728 }
729
730 return ret;
731 }
732
arch_arm_kprobe(struct kprobe * p)733 void arch_arm_kprobe(struct kprobe *p)
734 {
735 u8 int3 = INT3_INSN_OPCODE;
736
737 text_poke(p->addr, &int3, 1);
738 text_poke_sync();
739 perf_event_text_poke(p->addr, &p->opcode, 1, &int3, 1);
740 }
741
arch_disarm_kprobe(struct kprobe * p)742 void arch_disarm_kprobe(struct kprobe *p)
743 {
744 u8 int3 = INT3_INSN_OPCODE;
745
746 perf_event_text_poke(p->addr, &int3, 1, &p->opcode, 1);
747 text_poke(p->addr, &p->opcode, 1);
748 text_poke_sync();
749 }
750
arch_remove_kprobe(struct kprobe * p)751 void arch_remove_kprobe(struct kprobe *p)
752 {
753 if (p->ainsn.insn) {
754 /* Record the perf event before freeing the slot */
755 perf_event_text_poke(p->ainsn.insn, p->ainsn.insn,
756 p->ainsn.tp_len, NULL, 0);
757 free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
758 p->ainsn.insn = NULL;
759 }
760 }
761
762 static nokprobe_inline void
save_previous_kprobe(struct kprobe_ctlblk * kcb)763 save_previous_kprobe(struct kprobe_ctlblk *kcb)
764 {
765 kcb->prev_kprobe.kp = kprobe_running();
766 kcb->prev_kprobe.status = kcb->kprobe_status;
767 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
768 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
769 }
770
771 static nokprobe_inline void
restore_previous_kprobe(struct kprobe_ctlblk * kcb)772 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
773 {
774 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
775 kcb->kprobe_status = kcb->prev_kprobe.status;
776 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
777 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
778 }
779
780 static nokprobe_inline void
set_current_kprobe(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)781 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
782 struct kprobe_ctlblk *kcb)
783 {
784 __this_cpu_write(current_kprobe, p);
785 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
786 = (regs->flags & X86_EFLAGS_IF);
787 }
788
kprobe_post_process(struct kprobe * cur,struct pt_regs * regs,struct kprobe_ctlblk * kcb)789 static void kprobe_post_process(struct kprobe *cur, struct pt_regs *regs,
790 struct kprobe_ctlblk *kcb)
791 {
792 /* Restore back the original saved kprobes variables and continue. */
793 if (kcb->kprobe_status == KPROBE_REENTER) {
794 /* This will restore both kcb and current_kprobe */
795 restore_previous_kprobe(kcb);
796 } else {
797 /*
798 * Always update the kcb status because
799 * reset_curent_kprobe() doesn't update kcb.
800 */
801 kcb->kprobe_status = KPROBE_HIT_SSDONE;
802 if (cur->post_handler)
803 cur->post_handler(cur, regs, 0);
804 reset_current_kprobe();
805 }
806 }
807 NOKPROBE_SYMBOL(kprobe_post_process);
808
setup_singlestep(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb,int reenter)809 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
810 struct kprobe_ctlblk *kcb, int reenter)
811 {
812 if (setup_detour_execution(p, regs, reenter))
813 return;
814
815 #if !defined(CONFIG_PREEMPTION)
816 if (p->ainsn.boostable) {
817 /* Boost up -- we can execute copied instructions directly */
818 if (!reenter)
819 reset_current_kprobe();
820 /*
821 * Reentering boosted probe doesn't reset current_kprobe,
822 * nor set current_kprobe, because it doesn't use single
823 * stepping.
824 */
825 regs->ip = (unsigned long)p->ainsn.insn;
826 return;
827 }
828 #endif
829 if (reenter) {
830 save_previous_kprobe(kcb);
831 set_current_kprobe(p, regs, kcb);
832 kcb->kprobe_status = KPROBE_REENTER;
833 } else
834 kcb->kprobe_status = KPROBE_HIT_SS;
835
836 if (p->ainsn.emulate_op) {
837 p->ainsn.emulate_op(p, regs);
838 kprobe_post_process(p, regs, kcb);
839 return;
840 }
841
842 /* Disable interrupt, and set ip register on trampoline */
843 regs->flags &= ~X86_EFLAGS_IF;
844 regs->ip = (unsigned long)p->ainsn.insn;
845 }
846 NOKPROBE_SYMBOL(setup_singlestep);
847
848 /*
849 * Called after single-stepping. p->addr is the address of the
850 * instruction whose first byte has been replaced by the "int3"
851 * instruction. To avoid the SMP problems that can occur when we
852 * temporarily put back the original opcode to single-step, we
853 * single-stepped a copy of the instruction. The address of this
854 * copy is p->ainsn.insn. We also doesn't use trap, but "int3" again
855 * right after the copied instruction.
856 * Different from the trap single-step, "int3" single-step can not
857 * handle the instruction which changes the ip register, e.g. jmp,
858 * call, conditional jmp, and the instructions which changes the IF
859 * flags because interrupt must be disabled around the single-stepping.
860 * Such instructions are software emulated, but others are single-stepped
861 * using "int3".
862 *
863 * When the 2nd "int3" handled, the regs->ip and regs->flags needs to
864 * be adjusted, so that we can resume execution on correct code.
865 */
resume_singlestep(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)866 static void resume_singlestep(struct kprobe *p, struct pt_regs *regs,
867 struct kprobe_ctlblk *kcb)
868 {
869 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
870 unsigned long orig_ip = (unsigned long)p->addr;
871
872 /* Restore saved interrupt flag and ip register */
873 regs->flags |= kcb->kprobe_saved_flags;
874 /* Note that regs->ip is executed int3 so must be a step back */
875 regs->ip += (orig_ip - copy_ip) - INT3_INSN_SIZE;
876 }
877 NOKPROBE_SYMBOL(resume_singlestep);
878
879 /*
880 * We have reentered the kprobe_handler(), since another probe was hit while
881 * within the handler. We save the original kprobes variables and just single
882 * step on the instruction of the new probe without calling any user handlers.
883 */
reenter_kprobe(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)884 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
885 struct kprobe_ctlblk *kcb)
886 {
887 switch (kcb->kprobe_status) {
888 case KPROBE_HIT_SSDONE:
889 case KPROBE_HIT_ACTIVE:
890 case KPROBE_HIT_SS:
891 kprobes_inc_nmissed_count(p);
892 setup_singlestep(p, regs, kcb, 1);
893 break;
894 case KPROBE_REENTER:
895 /* A probe has been hit in the codepath leading up to, or just
896 * after, single-stepping of a probed instruction. This entire
897 * codepath should strictly reside in .kprobes.text section.
898 * Raise a BUG or we'll continue in an endless reentering loop
899 * and eventually a stack overflow.
900 */
901 pr_err("Unrecoverable kprobe detected.\n");
902 dump_kprobe(p);
903 BUG();
904 default:
905 /* impossible cases */
906 WARN_ON(1);
907 return 0;
908 }
909
910 return 1;
911 }
912 NOKPROBE_SYMBOL(reenter_kprobe);
913
kprobe_is_ss(struct kprobe_ctlblk * kcb)914 static nokprobe_inline int kprobe_is_ss(struct kprobe_ctlblk *kcb)
915 {
916 return (kcb->kprobe_status == KPROBE_HIT_SS ||
917 kcb->kprobe_status == KPROBE_REENTER);
918 }
919
920 /*
921 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
922 * remain disabled throughout this function.
923 */
kprobe_int3_handler(struct pt_regs * regs)924 int kprobe_int3_handler(struct pt_regs *regs)
925 {
926 kprobe_opcode_t *addr;
927 struct kprobe *p;
928 struct kprobe_ctlblk *kcb;
929
930 if (user_mode(regs))
931 return 0;
932
933 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
934 /*
935 * We don't want to be preempted for the entire duration of kprobe
936 * processing. Since int3 and debug trap disables irqs and we clear
937 * IF while singlestepping, it must be no preemptible.
938 */
939
940 kcb = get_kprobe_ctlblk();
941 p = get_kprobe(addr);
942
943 if (p) {
944 if (kprobe_running()) {
945 if (reenter_kprobe(p, regs, kcb))
946 return 1;
947 } else {
948 set_current_kprobe(p, regs, kcb);
949 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
950
951 /*
952 * If we have no pre-handler or it returned 0, we
953 * continue with normal processing. If we have a
954 * pre-handler and it returned non-zero, that means
955 * user handler setup registers to exit to another
956 * instruction, we must skip the single stepping.
957 */
958 if (!p->pre_handler || !p->pre_handler(p, regs))
959 setup_singlestep(p, regs, kcb, 0);
960 else
961 reset_current_kprobe();
962 return 1;
963 }
964 } else if (kprobe_is_ss(kcb)) {
965 p = kprobe_running();
966 if ((unsigned long)p->ainsn.insn < regs->ip &&
967 (unsigned long)p->ainsn.insn + MAX_INSN_SIZE > regs->ip) {
968 /* Most provably this is the second int3 for singlestep */
969 resume_singlestep(p, regs, kcb);
970 kprobe_post_process(p, regs, kcb);
971 return 1;
972 }
973 } /* else: not a kprobe fault; let the kernel handle it */
974
975 return 0;
976 }
977 NOKPROBE_SYMBOL(kprobe_int3_handler);
978
kprobe_fault_handler(struct pt_regs * regs,int trapnr)979 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
980 {
981 struct kprobe *cur = kprobe_running();
982 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
983
984 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
985 /* This must happen on single-stepping */
986 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
987 kcb->kprobe_status != KPROBE_REENTER);
988 /*
989 * We are here because the instruction being single
990 * stepped caused a page fault. We reset the current
991 * kprobe and the ip points back to the probe address
992 * and allow the page fault handler to continue as a
993 * normal page fault.
994 */
995 regs->ip = (unsigned long)cur->addr;
996
997 /*
998 * If the IF flag was set before the kprobe hit,
999 * don't touch it:
1000 */
1001 regs->flags |= kcb->kprobe_old_flags;
1002
1003 if (kcb->kprobe_status == KPROBE_REENTER)
1004 restore_previous_kprobe(kcb);
1005 else
1006 reset_current_kprobe();
1007 }
1008
1009 return 0;
1010 }
1011 NOKPROBE_SYMBOL(kprobe_fault_handler);
1012
arch_populate_kprobe_blacklist(void)1013 int __init arch_populate_kprobe_blacklist(void)
1014 {
1015 return kprobe_add_area_blacklist((unsigned long)__entry_text_start,
1016 (unsigned long)__entry_text_end);
1017 }
1018
arch_init_kprobes(void)1019 int __init arch_init_kprobes(void)
1020 {
1021 return 0;
1022 }
1023
arch_trampoline_kprobe(struct kprobe * p)1024 int arch_trampoline_kprobe(struct kprobe *p)
1025 {
1026 return 0;
1027 }
1028