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
48 #include <asm/text-patching.h>
49 #include <asm/cacheflush.h>
50 #include <asm/desc.h>
51 #include <linux/uaccess.h>
52 #include <asm/alternative.h>
53 #include <asm/insn.h>
54 #include <asm/debugreg.h>
55 #include <asm/set_memory.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 set_vm_flush_reset_perms(page);
422 /*
423 * First make the page read-only, and only then make it executable to
424 * prevent it from being W+X in between.
425 */
426 set_memory_ro((unsigned long)page, 1);
427
428 /*
429 * TODO: Once additional kernel code protection mechanisms are set, ensure
430 * that the page was not maliciously altered and it is still zeroed.
431 */
432 set_memory_x((unsigned long)page, 1);
433
434 return page;
435 }
436
437 /* Kprobe x86 instruction emulation - only regs->ip or IF flag modifiers */
438
kprobe_emulate_ifmodifiers(struct kprobe * p,struct pt_regs * regs)439 static void kprobe_emulate_ifmodifiers(struct kprobe *p, struct pt_regs *regs)
440 {
441 switch (p->ainsn.opcode) {
442 case 0xfa: /* cli */
443 regs->flags &= ~(X86_EFLAGS_IF);
444 break;
445 case 0xfb: /* sti */
446 regs->flags |= X86_EFLAGS_IF;
447 break;
448 case 0x9c: /* pushf */
449 int3_emulate_push(regs, regs->flags);
450 break;
451 case 0x9d: /* popf */
452 regs->flags = int3_emulate_pop(regs);
453 break;
454 }
455 regs->ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
456 }
457 NOKPROBE_SYMBOL(kprobe_emulate_ifmodifiers);
458
kprobe_emulate_ret(struct kprobe * p,struct pt_regs * regs)459 static void kprobe_emulate_ret(struct kprobe *p, struct pt_regs *regs)
460 {
461 int3_emulate_ret(regs);
462 }
463 NOKPROBE_SYMBOL(kprobe_emulate_ret);
464
kprobe_emulate_call(struct kprobe * p,struct pt_regs * regs)465 static void kprobe_emulate_call(struct kprobe *p, struct pt_regs *regs)
466 {
467 unsigned long func = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
468
469 func += p->ainsn.rel32;
470 int3_emulate_call(regs, func);
471 }
472 NOKPROBE_SYMBOL(kprobe_emulate_call);
473
474 static nokprobe_inline
__kprobe_emulate_jmp(struct kprobe * p,struct pt_regs * regs,bool cond)475 void __kprobe_emulate_jmp(struct kprobe *p, struct pt_regs *regs, bool cond)
476 {
477 unsigned long ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
478
479 if (cond)
480 ip += p->ainsn.rel32;
481 int3_emulate_jmp(regs, ip);
482 }
483
kprobe_emulate_jmp(struct kprobe * p,struct pt_regs * regs)484 static void kprobe_emulate_jmp(struct kprobe *p, struct pt_regs *regs)
485 {
486 __kprobe_emulate_jmp(p, regs, true);
487 }
488 NOKPROBE_SYMBOL(kprobe_emulate_jmp);
489
490 static const unsigned long jcc_mask[6] = {
491 [0] = X86_EFLAGS_OF,
492 [1] = X86_EFLAGS_CF,
493 [2] = X86_EFLAGS_ZF,
494 [3] = X86_EFLAGS_CF | X86_EFLAGS_ZF,
495 [4] = X86_EFLAGS_SF,
496 [5] = X86_EFLAGS_PF,
497 };
498
kprobe_emulate_jcc(struct kprobe * p,struct pt_regs * regs)499 static void kprobe_emulate_jcc(struct kprobe *p, struct pt_regs *regs)
500 {
501 bool invert = p->ainsn.jcc.type & 1;
502 bool match;
503
504 if (p->ainsn.jcc.type < 0xc) {
505 match = regs->flags & jcc_mask[p->ainsn.jcc.type >> 1];
506 } else {
507 match = ((regs->flags & X86_EFLAGS_SF) >> X86_EFLAGS_SF_BIT) ^
508 ((regs->flags & X86_EFLAGS_OF) >> X86_EFLAGS_OF_BIT);
509 if (p->ainsn.jcc.type >= 0xe)
510 match = match || (regs->flags & X86_EFLAGS_ZF);
511 }
512 __kprobe_emulate_jmp(p, regs, (match && !invert) || (!match && invert));
513 }
514 NOKPROBE_SYMBOL(kprobe_emulate_jcc);
515
kprobe_emulate_loop(struct kprobe * p,struct pt_regs * regs)516 static void kprobe_emulate_loop(struct kprobe *p, struct pt_regs *regs)
517 {
518 bool match;
519
520 if (p->ainsn.loop.type != 3) { /* LOOP* */
521 if (p->ainsn.loop.asize == 32)
522 match = ((*(u32 *)®s->cx)--) != 0;
523 #ifdef CONFIG_X86_64
524 else if (p->ainsn.loop.asize == 64)
525 match = ((*(u64 *)®s->cx)--) != 0;
526 #endif
527 else
528 match = ((*(u16 *)®s->cx)--) != 0;
529 } else { /* JCXZ */
530 if (p->ainsn.loop.asize == 32)
531 match = *(u32 *)(®s->cx) == 0;
532 #ifdef CONFIG_X86_64
533 else if (p->ainsn.loop.asize == 64)
534 match = *(u64 *)(®s->cx) == 0;
535 #endif
536 else
537 match = *(u16 *)(®s->cx) == 0;
538 }
539
540 if (p->ainsn.loop.type == 0) /* LOOPNE */
541 match = match && !(regs->flags & X86_EFLAGS_ZF);
542 else if (p->ainsn.loop.type == 1) /* LOOPE */
543 match = match && (regs->flags & X86_EFLAGS_ZF);
544
545 __kprobe_emulate_jmp(p, regs, match);
546 }
547 NOKPROBE_SYMBOL(kprobe_emulate_loop);
548
549 static const int addrmode_regoffs[] = {
550 offsetof(struct pt_regs, ax),
551 offsetof(struct pt_regs, cx),
552 offsetof(struct pt_regs, dx),
553 offsetof(struct pt_regs, bx),
554 offsetof(struct pt_regs, sp),
555 offsetof(struct pt_regs, bp),
556 offsetof(struct pt_regs, si),
557 offsetof(struct pt_regs, di),
558 #ifdef CONFIG_X86_64
559 offsetof(struct pt_regs, r8),
560 offsetof(struct pt_regs, r9),
561 offsetof(struct pt_regs, r10),
562 offsetof(struct pt_regs, r11),
563 offsetof(struct pt_regs, r12),
564 offsetof(struct pt_regs, r13),
565 offsetof(struct pt_regs, r14),
566 offsetof(struct pt_regs, r15),
567 #endif
568 };
569
kprobe_emulate_call_indirect(struct kprobe * p,struct pt_regs * regs)570 static void kprobe_emulate_call_indirect(struct kprobe *p, struct pt_regs *regs)
571 {
572 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
573
574 int3_emulate_call(regs, regs_get_register(regs, offs));
575 }
576 NOKPROBE_SYMBOL(kprobe_emulate_call_indirect);
577
kprobe_emulate_jmp_indirect(struct kprobe * p,struct pt_regs * regs)578 static void kprobe_emulate_jmp_indirect(struct kprobe *p, struct pt_regs *regs)
579 {
580 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
581
582 int3_emulate_jmp(regs, regs_get_register(regs, offs));
583 }
584 NOKPROBE_SYMBOL(kprobe_emulate_jmp_indirect);
585
prepare_emulation(struct kprobe * p,struct insn * insn)586 static int prepare_emulation(struct kprobe *p, struct insn *insn)
587 {
588 insn_byte_t opcode = insn->opcode.bytes[0];
589
590 switch (opcode) {
591 case 0xfa: /* cli */
592 case 0xfb: /* sti */
593 case 0x9c: /* pushfl */
594 case 0x9d: /* popf/popfd */
595 /*
596 * IF modifiers must be emulated since it will enable interrupt while
597 * int3 single stepping.
598 */
599 p->ainsn.emulate_op = kprobe_emulate_ifmodifiers;
600 p->ainsn.opcode = opcode;
601 break;
602 case 0xc2: /* ret/lret */
603 case 0xc3:
604 case 0xca:
605 case 0xcb:
606 p->ainsn.emulate_op = kprobe_emulate_ret;
607 break;
608 case 0x9a: /* far call absolute -- segment is not supported */
609 case 0xea: /* far jmp absolute -- segment is not supported */
610 case 0xcc: /* int3 */
611 case 0xcf: /* iret -- in-kernel IRET is not supported */
612 return -EOPNOTSUPP;
613 break;
614 case 0xe8: /* near call relative */
615 p->ainsn.emulate_op = kprobe_emulate_call;
616 if (insn->immediate.nbytes == 2)
617 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
618 else
619 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
620 break;
621 case 0xeb: /* short jump relative */
622 case 0xe9: /* near jump relative */
623 p->ainsn.emulate_op = kprobe_emulate_jmp;
624 if (insn->immediate.nbytes == 1)
625 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
626 else if (insn->immediate.nbytes == 2)
627 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
628 else
629 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
630 break;
631 case 0x70 ... 0x7f:
632 /* 1 byte conditional jump */
633 p->ainsn.emulate_op = kprobe_emulate_jcc;
634 p->ainsn.jcc.type = opcode & 0xf;
635 p->ainsn.rel32 = *(char *)insn->immediate.bytes;
636 break;
637 case 0x0f:
638 opcode = insn->opcode.bytes[1];
639 if ((opcode & 0xf0) == 0x80) {
640 /* 2 bytes Conditional Jump */
641 p->ainsn.emulate_op = kprobe_emulate_jcc;
642 p->ainsn.jcc.type = opcode & 0xf;
643 if (insn->immediate.nbytes == 2)
644 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
645 else
646 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
647 } else if (opcode == 0x01 &&
648 X86_MODRM_REG(insn->modrm.bytes[0]) == 0 &&
649 X86_MODRM_MOD(insn->modrm.bytes[0]) == 3) {
650 /* VM extensions - not supported */
651 return -EOPNOTSUPP;
652 }
653 break;
654 case 0xe0: /* Loop NZ */
655 case 0xe1: /* Loop */
656 case 0xe2: /* Loop */
657 case 0xe3: /* J*CXZ */
658 p->ainsn.emulate_op = kprobe_emulate_loop;
659 p->ainsn.loop.type = opcode & 0x3;
660 p->ainsn.loop.asize = insn->addr_bytes * 8;
661 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
662 break;
663 case 0xff:
664 /*
665 * Since the 0xff is an extended group opcode, the instruction
666 * is determined by the MOD/RM byte.
667 */
668 opcode = insn->modrm.bytes[0];
669 if ((opcode & 0x30) == 0x10) {
670 if ((opcode & 0x8) == 0x8)
671 return -EOPNOTSUPP; /* far call */
672 /* call absolute, indirect */
673 p->ainsn.emulate_op = kprobe_emulate_call_indirect;
674 } else if ((opcode & 0x30) == 0x20) {
675 if ((opcode & 0x8) == 0x8)
676 return -EOPNOTSUPP; /* far jmp */
677 /* jmp near absolute indirect */
678 p->ainsn.emulate_op = kprobe_emulate_jmp_indirect;
679 } else
680 break;
681
682 if (insn->addr_bytes != sizeof(unsigned long))
683 return -EOPNOTSUPP; /* Don't support different size */
684 if (X86_MODRM_MOD(opcode) != 3)
685 return -EOPNOTSUPP; /* TODO: support memory addressing */
686
687 p->ainsn.indirect.reg = X86_MODRM_RM(opcode);
688 #ifdef CONFIG_X86_64
689 if (X86_REX_B(insn->rex_prefix.value))
690 p->ainsn.indirect.reg += 8;
691 #endif
692 break;
693 default:
694 break;
695 }
696 p->ainsn.size = insn->length;
697
698 return 0;
699 }
700
arch_copy_kprobe(struct kprobe * p)701 static int arch_copy_kprobe(struct kprobe *p)
702 {
703 struct insn insn;
704 kprobe_opcode_t buf[MAX_INSN_SIZE];
705 int ret, len;
706
707 /* Copy an instruction with recovering if other optprobe modifies it.*/
708 len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn);
709 if (!len)
710 return -EINVAL;
711
712 /* Analyze the opcode and setup emulate functions */
713 ret = prepare_emulation(p, &insn);
714 if (ret < 0)
715 return ret;
716
717 /* Add int3 for single-step or booster jmp */
718 len = prepare_singlestep(buf, p, &insn);
719 if (len < 0)
720 return len;
721
722 /* Also, displacement change doesn't affect the first byte */
723 p->opcode = buf[0];
724
725 p->ainsn.tp_len = len;
726 perf_event_text_poke(p->ainsn.insn, NULL, 0, buf, len);
727
728 /* OK, write back the instruction(s) into ROX insn buffer */
729 text_poke(p->ainsn.insn, buf, len);
730
731 return 0;
732 }
733
arch_prepare_kprobe(struct kprobe * p)734 int arch_prepare_kprobe(struct kprobe *p)
735 {
736 int ret;
737
738 if (alternatives_text_reserved(p->addr, p->addr))
739 return -EINVAL;
740
741 if (!can_probe((unsigned long)p->addr))
742 return -EILSEQ;
743
744 memset(&p->ainsn, 0, sizeof(p->ainsn));
745
746 /* insn: must be on special executable page on x86. */
747 p->ainsn.insn = get_insn_slot();
748 if (!p->ainsn.insn)
749 return -ENOMEM;
750
751 ret = arch_copy_kprobe(p);
752 if (ret) {
753 free_insn_slot(p->ainsn.insn, 0);
754 p->ainsn.insn = NULL;
755 }
756
757 return ret;
758 }
759
arch_arm_kprobe(struct kprobe * p)760 void arch_arm_kprobe(struct kprobe *p)
761 {
762 u8 int3 = INT3_INSN_OPCODE;
763
764 text_poke(p->addr, &int3, 1);
765 text_poke_sync();
766 perf_event_text_poke(p->addr, &p->opcode, 1, &int3, 1);
767 }
768
arch_disarm_kprobe(struct kprobe * p)769 void arch_disarm_kprobe(struct kprobe *p)
770 {
771 u8 int3 = INT3_INSN_OPCODE;
772
773 perf_event_text_poke(p->addr, &int3, 1, &p->opcode, 1);
774 text_poke(p->addr, &p->opcode, 1);
775 text_poke_sync();
776 }
777
arch_remove_kprobe(struct kprobe * p)778 void arch_remove_kprobe(struct kprobe *p)
779 {
780 if (p->ainsn.insn) {
781 /* Record the perf event before freeing the slot */
782 perf_event_text_poke(p->ainsn.insn, p->ainsn.insn,
783 p->ainsn.tp_len, NULL, 0);
784 free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
785 p->ainsn.insn = NULL;
786 }
787 }
788
789 static nokprobe_inline void
save_previous_kprobe(struct kprobe_ctlblk * kcb)790 save_previous_kprobe(struct kprobe_ctlblk *kcb)
791 {
792 kcb->prev_kprobe.kp = kprobe_running();
793 kcb->prev_kprobe.status = kcb->kprobe_status;
794 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
795 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
796 }
797
798 static nokprobe_inline void
restore_previous_kprobe(struct kprobe_ctlblk * kcb)799 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
800 {
801 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
802 kcb->kprobe_status = kcb->prev_kprobe.status;
803 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
804 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
805 }
806
807 static nokprobe_inline void
set_current_kprobe(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)808 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
809 struct kprobe_ctlblk *kcb)
810 {
811 __this_cpu_write(current_kprobe, p);
812 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
813 = (regs->flags & X86_EFLAGS_IF);
814 }
815
kprobe_post_process(struct kprobe * cur,struct pt_regs * regs,struct kprobe_ctlblk * kcb)816 static void kprobe_post_process(struct kprobe *cur, struct pt_regs *regs,
817 struct kprobe_ctlblk *kcb)
818 {
819 /* Restore back the original saved kprobes variables and continue. */
820 if (kcb->kprobe_status == KPROBE_REENTER) {
821 /* This will restore both kcb and current_kprobe */
822 restore_previous_kprobe(kcb);
823 } else {
824 /*
825 * Always update the kcb status because
826 * reset_curent_kprobe() doesn't update kcb.
827 */
828 kcb->kprobe_status = KPROBE_HIT_SSDONE;
829 if (cur->post_handler)
830 cur->post_handler(cur, regs, 0);
831 reset_current_kprobe();
832 }
833 }
834 NOKPROBE_SYMBOL(kprobe_post_process);
835
setup_singlestep(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb,int reenter)836 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
837 struct kprobe_ctlblk *kcb, int reenter)
838 {
839 if (setup_detour_execution(p, regs, reenter))
840 return;
841
842 #if !defined(CONFIG_PREEMPTION)
843 if (p->ainsn.boostable) {
844 /* Boost up -- we can execute copied instructions directly */
845 if (!reenter)
846 reset_current_kprobe();
847 /*
848 * Reentering boosted probe doesn't reset current_kprobe,
849 * nor set current_kprobe, because it doesn't use single
850 * stepping.
851 */
852 regs->ip = (unsigned long)p->ainsn.insn;
853 return;
854 }
855 #endif
856 if (reenter) {
857 save_previous_kprobe(kcb);
858 set_current_kprobe(p, regs, kcb);
859 kcb->kprobe_status = KPROBE_REENTER;
860 } else
861 kcb->kprobe_status = KPROBE_HIT_SS;
862
863 if (p->ainsn.emulate_op) {
864 p->ainsn.emulate_op(p, regs);
865 kprobe_post_process(p, regs, kcb);
866 return;
867 }
868
869 /* Disable interrupt, and set ip register on trampoline */
870 regs->flags &= ~X86_EFLAGS_IF;
871 regs->ip = (unsigned long)p->ainsn.insn;
872 }
873 NOKPROBE_SYMBOL(setup_singlestep);
874
875 /*
876 * Called after single-stepping. p->addr is the address of the
877 * instruction whose first byte has been replaced by the "int3"
878 * instruction. To avoid the SMP problems that can occur when we
879 * temporarily put back the original opcode to single-step, we
880 * single-stepped a copy of the instruction. The address of this
881 * copy is p->ainsn.insn. We also doesn't use trap, but "int3" again
882 * right after the copied instruction.
883 * Different from the trap single-step, "int3" single-step can not
884 * handle the instruction which changes the ip register, e.g. jmp,
885 * call, conditional jmp, and the instructions which changes the IF
886 * flags because interrupt must be disabled around the single-stepping.
887 * Such instructions are software emulated, but others are single-stepped
888 * using "int3".
889 *
890 * When the 2nd "int3" handled, the regs->ip and regs->flags needs to
891 * be adjusted, so that we can resume execution on correct code.
892 */
resume_singlestep(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)893 static void resume_singlestep(struct kprobe *p, struct pt_regs *regs,
894 struct kprobe_ctlblk *kcb)
895 {
896 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
897 unsigned long orig_ip = (unsigned long)p->addr;
898
899 /* Restore saved interrupt flag and ip register */
900 regs->flags |= kcb->kprobe_saved_flags;
901 /* Note that regs->ip is executed int3 so must be a step back */
902 regs->ip += (orig_ip - copy_ip) - INT3_INSN_SIZE;
903 }
904 NOKPROBE_SYMBOL(resume_singlestep);
905
906 /*
907 * We have reentered the kprobe_handler(), since another probe was hit while
908 * within the handler. We save the original kprobes variables and just single
909 * step on the instruction of the new probe without calling any user handlers.
910 */
reenter_kprobe(struct kprobe * p,struct pt_regs * regs,struct kprobe_ctlblk * kcb)911 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
912 struct kprobe_ctlblk *kcb)
913 {
914 switch (kcb->kprobe_status) {
915 case KPROBE_HIT_SSDONE:
916 case KPROBE_HIT_ACTIVE:
917 case KPROBE_HIT_SS:
918 kprobes_inc_nmissed_count(p);
919 setup_singlestep(p, regs, kcb, 1);
920 break;
921 case KPROBE_REENTER:
922 /* A probe has been hit in the codepath leading up to, or just
923 * after, single-stepping of a probed instruction. This entire
924 * codepath should strictly reside in .kprobes.text section.
925 * Raise a BUG or we'll continue in an endless reentering loop
926 * and eventually a stack overflow.
927 */
928 pr_err("Unrecoverable kprobe detected.\n");
929 dump_kprobe(p);
930 BUG();
931 default:
932 /* impossible cases */
933 WARN_ON(1);
934 return 0;
935 }
936
937 return 1;
938 }
939 NOKPROBE_SYMBOL(reenter_kprobe);
940
kprobe_is_ss(struct kprobe_ctlblk * kcb)941 static nokprobe_inline int kprobe_is_ss(struct kprobe_ctlblk *kcb)
942 {
943 return (kcb->kprobe_status == KPROBE_HIT_SS ||
944 kcb->kprobe_status == KPROBE_REENTER);
945 }
946
947 /*
948 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
949 * remain disabled throughout this function.
950 */
kprobe_int3_handler(struct pt_regs * regs)951 int kprobe_int3_handler(struct pt_regs *regs)
952 {
953 kprobe_opcode_t *addr;
954 struct kprobe *p;
955 struct kprobe_ctlblk *kcb;
956
957 if (user_mode(regs))
958 return 0;
959
960 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
961 /*
962 * We don't want to be preempted for the entire duration of kprobe
963 * processing. Since int3 and debug trap disables irqs and we clear
964 * IF while singlestepping, it must be no preemptible.
965 */
966
967 kcb = get_kprobe_ctlblk();
968 p = get_kprobe(addr);
969
970 if (p) {
971 if (kprobe_running()) {
972 if (reenter_kprobe(p, regs, kcb))
973 return 1;
974 } else {
975 set_current_kprobe(p, regs, kcb);
976 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
977
978 /*
979 * If we have no pre-handler or it returned 0, we
980 * continue with normal processing. If we have a
981 * pre-handler and it returned non-zero, that means
982 * user handler setup registers to exit to another
983 * instruction, we must skip the single stepping.
984 */
985 if (!p->pre_handler || !p->pre_handler(p, regs))
986 setup_singlestep(p, regs, kcb, 0);
987 else
988 reset_current_kprobe();
989 return 1;
990 }
991 } else if (kprobe_is_ss(kcb)) {
992 p = kprobe_running();
993 if ((unsigned long)p->ainsn.insn < regs->ip &&
994 (unsigned long)p->ainsn.insn + MAX_INSN_SIZE > regs->ip) {
995 /* Most provably this is the second int3 for singlestep */
996 resume_singlestep(p, regs, kcb);
997 kprobe_post_process(p, regs, kcb);
998 return 1;
999 }
1000 }
1001
1002 if (*addr != INT3_INSN_OPCODE) {
1003 /*
1004 * The breakpoint instruction was removed right
1005 * after we hit it. Another cpu has removed
1006 * either a probepoint or a debugger breakpoint
1007 * at this address. In either case, no further
1008 * handling of this interrupt is appropriate.
1009 * Back up over the (now missing) int3 and run
1010 * the original instruction.
1011 */
1012 regs->ip = (unsigned long)addr;
1013 return 1;
1014 } /* else: not a kprobe fault; let the kernel handle it */
1015
1016 return 0;
1017 }
1018 NOKPROBE_SYMBOL(kprobe_int3_handler);
1019
kprobe_fault_handler(struct pt_regs * regs,int trapnr)1020 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
1021 {
1022 struct kprobe *cur = kprobe_running();
1023 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1024
1025 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
1026 /* This must happen on single-stepping */
1027 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
1028 kcb->kprobe_status != KPROBE_REENTER);
1029 /*
1030 * We are here because the instruction being single
1031 * stepped caused a page fault. We reset the current
1032 * kprobe and the ip points back to the probe address
1033 * and allow the page fault handler to continue as a
1034 * normal page fault.
1035 */
1036 regs->ip = (unsigned long)cur->addr;
1037
1038 /*
1039 * If the IF flag was set before the kprobe hit,
1040 * don't touch it:
1041 */
1042 regs->flags |= kcb->kprobe_old_flags;
1043
1044 if (kcb->kprobe_status == KPROBE_REENTER)
1045 restore_previous_kprobe(kcb);
1046 else
1047 reset_current_kprobe();
1048 }
1049
1050 return 0;
1051 }
1052 NOKPROBE_SYMBOL(kprobe_fault_handler);
1053
arch_populate_kprobe_blacklist(void)1054 int __init arch_populate_kprobe_blacklist(void)
1055 {
1056 return kprobe_add_area_blacklist((unsigned long)__entry_text_start,
1057 (unsigned long)__entry_text_end);
1058 }
1059
arch_init_kprobes(void)1060 int __init arch_init_kprobes(void)
1061 {
1062 return 0;
1063 }
1064
arch_trampoline_kprobe(struct kprobe * p)1065 int arch_trampoline_kprobe(struct kprobe *p)
1066 {
1067 return 0;
1068 }
1069