1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
5 * Copyright (C) IBM Corporation, 2002, 2004
7 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
8 * Probes initial implementation ( includes contributions from
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.
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>
48 #include <asm/text-patching.h>
49 #include <asm/cacheflush.h>
51 #include <linux/uaccess.h>
52 #include <asm/alternative.h>
54 #include <asm/debugreg.h>
55 #include <asm/set_memory.h>
59 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
60 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
62 #define stack_addr(regs) ((unsigned long *)regs->sp)
64 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
65 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
66 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
67 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
68 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
71 * Undefined/reserved opcodes, conditional jump, Opcode Extension
72 * Groups, and some special opcodes can not boost.
73 * This is non-const and volatile to keep gcc from statically
74 * optimizing it out, as variable_test_bit makes gcc think only
75 * *(unsigned long*) is used.
77 static volatile u32 twobyte_is_boostable[256 / 32] = {
78 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
79 /* ---------------------------------------------- */
80 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
81 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
82 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
83 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
84 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
85 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
86 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
87 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
88 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
89 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
90 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
91 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
92 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
93 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
94 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
95 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
96 /* ----------------------------------------------- */
97 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
101 struct kretprobe_blackpoint kretprobe_blacklist[] = {
102 {"__switch_to", }, /* This function switches only current task, but
103 doesn't switch kernel stack.*/
104 {NULL, NULL} /* Terminator */
107 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
109 static nokprobe_inline void
110 __synthesize_relative_insn(void *dest, void *from, void *to, u8 op)
112 struct __arch_relative_insn {
117 insn = (struct __arch_relative_insn *)dest;
118 insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
122 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
123 void synthesize_reljump(void *dest, void *from, void *to)
125 __synthesize_relative_insn(dest, from, to, JMP32_INSN_OPCODE);
127 NOKPROBE_SYMBOL(synthesize_reljump);
129 /* Insert a call instruction at address 'from', which calls address 'to'.*/
130 void synthesize_relcall(void *dest, void *from, void *to)
132 __synthesize_relative_insn(dest, from, to, CALL_INSN_OPCODE);
134 NOKPROBE_SYMBOL(synthesize_relcall);
137 * Returns non-zero if INSN is boostable.
138 * RIP relative instructions are adjusted at copying time in 64 bits mode
140 int can_boost(struct insn *insn, void *addr)
142 kprobe_opcode_t opcode;
146 if (search_exception_tables((unsigned long)addr))
147 return 0; /* Page fault may occur on this address. */
149 /* 2nd-byte opcode */
150 if (insn->opcode.nbytes == 2)
151 return test_bit(insn->opcode.bytes[1],
152 (unsigned long *)twobyte_is_boostable);
154 if (insn->opcode.nbytes != 1)
157 for_each_insn_prefix(insn, i, prefix) {
160 attr = inat_get_opcode_attribute(prefix);
161 /* Can't boost Address-size override prefix and CS override prefix */
162 if (prefix == 0x2e || inat_is_address_size_prefix(attr))
166 opcode = insn->opcode.bytes[0];
169 case 0x62: /* bound */
170 case 0x70 ... 0x7f: /* Conditional jumps */
171 case 0x9a: /* Call far */
172 case 0xc0 ... 0xc1: /* Grp2 */
173 case 0xcc ... 0xce: /* software exceptions */
174 case 0xd0 ... 0xd3: /* Grp2 */
175 case 0xd6: /* (UD) */
176 case 0xd8 ... 0xdf: /* ESC */
177 case 0xe0 ... 0xe3: /* LOOP*, JCXZ */
178 case 0xe8 ... 0xe9: /* near Call, JMP */
179 case 0xeb: /* Short JMP */
180 case 0xf0 ... 0xf4: /* LOCK/REP, HLT */
181 case 0xf6 ... 0xf7: /* Grp3 */
182 case 0xfe: /* Grp4 */
183 /* ... are not boostable */
185 case 0xff: /* Grp5 */
186 /* Only indirect jmp is boostable */
187 return X86_MODRM_REG(insn->modrm.bytes[0]) == 4;
194 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
199 kp = get_kprobe((void *)addr);
200 faddr = ftrace_location(addr);
202 * Addresses inside the ftrace location are refused by
203 * arch_check_ftrace_location(). Something went terribly wrong
204 * if such an address is checked here.
206 if (WARN_ON(faddr && faddr != addr))
209 * Use the current code if it is not modified by Kprobe
210 * and it cannot be modified by ftrace.
216 * Basically, kp->ainsn.insn has an original instruction.
217 * However, RIP-relative instruction can not do single-stepping
218 * at different place, __copy_instruction() tweaks the displacement of
219 * that instruction. In that case, we can't recover the instruction
220 * from the kp->ainsn.insn.
222 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
223 * of the first byte of the probed instruction, which is overwritten
224 * by int3. And the instruction at kp->addr is not modified by kprobes
225 * except for the first byte, we can recover the original instruction
226 * from it and kp->opcode.
228 * In case of Kprobes using ftrace, we do not have a copy of
229 * the original instruction. In fact, the ftrace location might
230 * be modified at anytime and even could be in an inconsistent state.
231 * Fortunately, we know that the original code is the ideal 5-byte
234 if (copy_from_kernel_nofault(buf, (void *)addr,
235 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
239 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
242 return (unsigned long)buf;
246 * Recover the probed instruction at addr for further analysis.
247 * Caller must lock kprobes by kprobe_mutex, or disable preemption
248 * for preventing to release referencing kprobes.
249 * Returns zero if the instruction can not get recovered (or access failed).
251 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
253 unsigned long __addr;
255 __addr = __recover_optprobed_insn(buf, addr);
259 return __recover_probed_insn(buf, addr);
262 /* Check if paddr is at an instruction boundary */
263 static int can_probe(unsigned long paddr)
265 unsigned long addr, __addr, offset = 0;
267 kprobe_opcode_t buf[MAX_INSN_SIZE];
269 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
272 /* Decode instructions */
273 addr = paddr - offset;
274 while (addr < paddr) {
278 * Check if the instruction has been modified by another
279 * kprobe, in which case we replace the breakpoint by the
280 * original instruction in our buffer.
281 * Also, jump optimization will change the breakpoint to
282 * relative-jump. Since the relative-jump itself is
283 * normally used, we just go through if there is no kprobe.
285 __addr = recover_probed_instruction(buf, addr);
289 ret = insn_decode(&insn, (void *)__addr, MAX_INSN_SIZE, INSN_MODE_KERN);
295 * If there is a dynamically installed kgdb sw breakpoint,
296 * this function should not be probed.
298 if (insn.opcode.bytes[0] == INT3_INSN_OPCODE &&
299 kgdb_has_hit_break(addr))
305 return (addr == paddr);
309 * Copy an instruction with recovering modified instruction by kprobes
310 * and adjust the displacement if the instruction uses the %rip-relative
311 * addressing mode. Note that since @real will be the final place of copied
312 * instruction, displacement must be adjust by @real, not @dest.
313 * This returns the length of copied instruction, or 0 if it has an error.
315 int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn)
317 kprobe_opcode_t buf[MAX_INSN_SIZE];
318 unsigned long recovered_insn = recover_probed_instruction(buf, (unsigned long)src);
321 if (!recovered_insn || !insn)
324 /* This can access kernel text if given address is not recovered */
325 if (copy_from_kernel_nofault(dest, (void *)recovered_insn,
329 ret = insn_decode(insn, dest, MAX_INSN_SIZE, INSN_MODE_KERN);
333 /* We can not probe force emulate prefixed instruction */
334 if (insn_has_emulate_prefix(insn))
337 /* Another subsystem puts a breakpoint, failed to recover */
338 if (insn->opcode.bytes[0] == INT3_INSN_OPCODE)
341 /* We should not singlestep on the exception masking instructions */
342 if (insn_masking_exception(insn))
346 /* Only x86_64 has RIP relative instructions */
347 if (insn_rip_relative(insn)) {
351 * The copied instruction uses the %rip-relative addressing
352 * mode. Adjust the displacement for the difference between
353 * the original location of this instruction and the location
354 * of the copy that will actually be run. The tricky bit here
355 * is making sure that the sign extension happens correctly in
356 * this calculation, since we need a signed 32-bit result to
357 * be sign-extended to 64 bits when it's added to the %rip
358 * value and yield the same 64-bit result that the sign-
359 * extension of the original signed 32-bit displacement would
362 newdisp = (u8 *) src + (s64) insn->displacement.value
364 if ((s64) (s32) newdisp != newdisp) {
365 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
368 disp = (u8 *) dest + insn_offset_displacement(insn);
369 *(s32 *) disp = (s32) newdisp;
375 /* Prepare reljump or int3 right after instruction */
376 static int prepare_singlestep(kprobe_opcode_t *buf, struct kprobe *p,
379 int len = insn->length;
381 if (!IS_ENABLED(CONFIG_PREEMPTION) &&
382 !p->post_handler && can_boost(insn, p->addr) &&
383 MAX_INSN_SIZE - len >= JMP32_INSN_SIZE) {
385 * These instructions can be executed directly if it
386 * jumps back to correct address.
388 synthesize_reljump(buf + len, p->ainsn.insn + len,
389 p->addr + insn->length);
390 len += JMP32_INSN_SIZE;
391 p->ainsn.boostable = 1;
393 /* Otherwise, put an int3 for trapping singlestep */
394 if (MAX_INSN_SIZE - len < INT3_INSN_SIZE)
397 buf[len] = INT3_INSN_OPCODE;
398 len += INT3_INSN_SIZE;
404 /* Make page to RO mode when allocate it */
405 void *alloc_insn_page(void)
409 page = module_alloc(PAGE_SIZE);
413 set_vm_flush_reset_perms(page);
415 * First make the page read-only, and only then make it executable to
416 * prevent it from being W+X in between.
418 set_memory_ro((unsigned long)page, 1);
421 * TODO: Once additional kernel code protection mechanisms are set, ensure
422 * that the page was not maliciously altered and it is still zeroed.
424 set_memory_x((unsigned long)page, 1);
429 /* Recover page to RW mode before releasing it */
430 void free_insn_page(void *page)
432 module_memfree(page);
435 /* Kprobe x86 instruction emulation - only regs->ip or IF flag modifiers */
437 static void kprobe_emulate_ifmodifiers(struct kprobe *p, struct pt_regs *regs)
439 switch (p->ainsn.opcode) {
441 regs->flags &= ~(X86_EFLAGS_IF);
444 regs->flags |= X86_EFLAGS_IF;
446 case 0x9c: /* pushf */
447 int3_emulate_push(regs, regs->flags);
449 case 0x9d: /* popf */
450 regs->flags = int3_emulate_pop(regs);
453 regs->ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
455 NOKPROBE_SYMBOL(kprobe_emulate_ifmodifiers);
457 static void kprobe_emulate_ret(struct kprobe *p, struct pt_regs *regs)
459 int3_emulate_ret(regs);
461 NOKPROBE_SYMBOL(kprobe_emulate_ret);
463 static void kprobe_emulate_call(struct kprobe *p, struct pt_regs *regs)
465 unsigned long func = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
467 func += p->ainsn.rel32;
468 int3_emulate_call(regs, func);
470 NOKPROBE_SYMBOL(kprobe_emulate_call);
472 static nokprobe_inline
473 void __kprobe_emulate_jmp(struct kprobe *p, struct pt_regs *regs, bool cond)
475 unsigned long ip = regs->ip - INT3_INSN_SIZE + p->ainsn.size;
478 ip += p->ainsn.rel32;
479 int3_emulate_jmp(regs, ip);
482 static void kprobe_emulate_jmp(struct kprobe *p, struct pt_regs *regs)
484 __kprobe_emulate_jmp(p, regs, true);
486 NOKPROBE_SYMBOL(kprobe_emulate_jmp);
488 static const unsigned long jcc_mask[6] = {
492 [3] = X86_EFLAGS_CF | X86_EFLAGS_ZF,
497 static void kprobe_emulate_jcc(struct kprobe *p, struct pt_regs *regs)
499 bool invert = p->ainsn.jcc.type & 1;
502 if (p->ainsn.jcc.type < 0xc) {
503 match = regs->flags & jcc_mask[p->ainsn.jcc.type >> 1];
505 match = ((regs->flags & X86_EFLAGS_SF) >> X86_EFLAGS_SF_BIT) ^
506 ((regs->flags & X86_EFLAGS_OF) >> X86_EFLAGS_OF_BIT);
507 if (p->ainsn.jcc.type >= 0xe)
508 match = match || (regs->flags & X86_EFLAGS_ZF);
510 __kprobe_emulate_jmp(p, regs, (match && !invert) || (!match && invert));
512 NOKPROBE_SYMBOL(kprobe_emulate_jcc);
514 static void kprobe_emulate_loop(struct kprobe *p, struct pt_regs *regs)
518 if (p->ainsn.loop.type != 3) { /* LOOP* */
519 if (p->ainsn.loop.asize == 32)
520 match = ((*(u32 *)®s->cx)--) != 0;
522 else if (p->ainsn.loop.asize == 64)
523 match = ((*(u64 *)®s->cx)--) != 0;
526 match = ((*(u16 *)®s->cx)--) != 0;
528 if (p->ainsn.loop.asize == 32)
529 match = *(u32 *)(®s->cx) == 0;
531 else if (p->ainsn.loop.asize == 64)
532 match = *(u64 *)(®s->cx) == 0;
535 match = *(u16 *)(®s->cx) == 0;
538 if (p->ainsn.loop.type == 0) /* LOOPNE */
539 match = match && !(regs->flags & X86_EFLAGS_ZF);
540 else if (p->ainsn.loop.type == 1) /* LOOPE */
541 match = match && (regs->flags & X86_EFLAGS_ZF);
543 __kprobe_emulate_jmp(p, regs, match);
545 NOKPROBE_SYMBOL(kprobe_emulate_loop);
547 static const int addrmode_regoffs[] = {
548 offsetof(struct pt_regs, ax),
549 offsetof(struct pt_regs, cx),
550 offsetof(struct pt_regs, dx),
551 offsetof(struct pt_regs, bx),
552 offsetof(struct pt_regs, sp),
553 offsetof(struct pt_regs, bp),
554 offsetof(struct pt_regs, si),
555 offsetof(struct pt_regs, di),
557 offsetof(struct pt_regs, r8),
558 offsetof(struct pt_regs, r9),
559 offsetof(struct pt_regs, r10),
560 offsetof(struct pt_regs, r11),
561 offsetof(struct pt_regs, r12),
562 offsetof(struct pt_regs, r13),
563 offsetof(struct pt_regs, r14),
564 offsetof(struct pt_regs, r15),
568 static void kprobe_emulate_call_indirect(struct kprobe *p, struct pt_regs *regs)
570 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
572 int3_emulate_push(regs, regs->ip - INT3_INSN_SIZE + p->ainsn.size);
573 int3_emulate_jmp(regs, regs_get_register(regs, offs));
575 NOKPROBE_SYMBOL(kprobe_emulate_call_indirect);
577 static void kprobe_emulate_jmp_indirect(struct kprobe *p, struct pt_regs *regs)
579 unsigned long offs = addrmode_regoffs[p->ainsn.indirect.reg];
581 int3_emulate_jmp(regs, regs_get_register(regs, offs));
583 NOKPROBE_SYMBOL(kprobe_emulate_jmp_indirect);
585 static int prepare_emulation(struct kprobe *p, struct insn *insn)
587 insn_byte_t opcode = insn->opcode.bytes[0];
592 case 0x9c: /* pushfl */
593 case 0x9d: /* popf/popfd */
595 * IF modifiers must be emulated since it will enable interrupt while
596 * int3 single stepping.
598 p->ainsn.emulate_op = kprobe_emulate_ifmodifiers;
599 p->ainsn.opcode = opcode;
601 case 0xc2: /* ret/lret */
605 p->ainsn.emulate_op = kprobe_emulate_ret;
607 case 0x9a: /* far call absolute -- segment is not supported */
608 case 0xea: /* far jmp absolute -- segment is not supported */
609 case 0xcc: /* int3 */
610 case 0xcf: /* iret -- in-kernel IRET is not supported */
613 case 0xe8: /* near call relative */
614 p->ainsn.emulate_op = kprobe_emulate_call;
615 if (insn->immediate.nbytes == 2)
616 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
618 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
620 case 0xeb: /* short jump relative */
621 case 0xe9: /* near jump relative */
622 p->ainsn.emulate_op = kprobe_emulate_jmp;
623 if (insn->immediate.nbytes == 1)
624 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
625 else if (insn->immediate.nbytes == 2)
626 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
628 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
631 /* 1 byte conditional jump */
632 p->ainsn.emulate_op = kprobe_emulate_jcc;
633 p->ainsn.jcc.type = opcode & 0xf;
634 p->ainsn.rel32 = *(char *)insn->immediate.bytes;
637 opcode = insn->opcode.bytes[1];
638 if ((opcode & 0xf0) == 0x80) {
639 /* 2 bytes Conditional Jump */
640 p->ainsn.emulate_op = kprobe_emulate_jcc;
641 p->ainsn.jcc.type = opcode & 0xf;
642 if (insn->immediate.nbytes == 2)
643 p->ainsn.rel32 = *(s16 *)&insn->immediate.value;
645 p->ainsn.rel32 = *(s32 *)&insn->immediate.value;
646 } else if (opcode == 0x01 &&
647 X86_MODRM_REG(insn->modrm.bytes[0]) == 0 &&
648 X86_MODRM_MOD(insn->modrm.bytes[0]) == 3) {
649 /* VM extensions - not supported */
653 case 0xe0: /* Loop NZ */
654 case 0xe1: /* Loop */
655 case 0xe2: /* Loop */
656 case 0xe3: /* J*CXZ */
657 p->ainsn.emulate_op = kprobe_emulate_loop;
658 p->ainsn.loop.type = opcode & 0x3;
659 p->ainsn.loop.asize = insn->addr_bytes * 8;
660 p->ainsn.rel32 = *(s8 *)&insn->immediate.value;
664 * Since the 0xff is an extended group opcode, the instruction
665 * is determined by the MOD/RM byte.
667 opcode = insn->modrm.bytes[0];
668 if ((opcode & 0x30) == 0x10) {
669 if ((opcode & 0x8) == 0x8)
670 return -EOPNOTSUPP; /* far call */
671 /* call absolute, indirect */
672 p->ainsn.emulate_op = kprobe_emulate_call_indirect;
673 } else if ((opcode & 0x30) == 0x20) {
674 if ((opcode & 0x8) == 0x8)
675 return -EOPNOTSUPP; /* far jmp */
676 /* jmp near absolute indirect */
677 p->ainsn.emulate_op = kprobe_emulate_jmp_indirect;
681 if (insn->addr_bytes != sizeof(unsigned long))
682 return -EOPNOTSUPP; /* Don't support differnt size */
683 if (X86_MODRM_MOD(opcode) != 3)
684 return -EOPNOTSUPP; /* TODO: support memory addressing */
686 p->ainsn.indirect.reg = X86_MODRM_RM(opcode);
688 if (X86_REX_B(insn->rex_prefix.value))
689 p->ainsn.indirect.reg += 8;
695 p->ainsn.size = insn->length;
700 static int arch_copy_kprobe(struct kprobe *p)
703 kprobe_opcode_t buf[MAX_INSN_SIZE];
706 /* Copy an instruction with recovering if other optprobe modifies it.*/
707 len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn);
711 /* Analyze the opcode and setup emulate functions */
712 ret = prepare_emulation(p, &insn);
716 /* Add int3 for single-step or booster jmp */
717 len = prepare_singlestep(buf, p, &insn);
721 /* Also, displacement change doesn't affect the first byte */
724 p->ainsn.tp_len = len;
725 perf_event_text_poke(p->ainsn.insn, NULL, 0, buf, len);
727 /* OK, write back the instruction(s) into ROX insn buffer */
728 text_poke(p->ainsn.insn, buf, len);
733 int arch_prepare_kprobe(struct kprobe *p)
737 if (alternatives_text_reserved(p->addr, p->addr))
740 if (!can_probe((unsigned long)p->addr))
743 memset(&p->ainsn, 0, sizeof(p->ainsn));
745 /* insn: must be on special executable page on x86. */
746 p->ainsn.insn = get_insn_slot();
750 ret = arch_copy_kprobe(p);
752 free_insn_slot(p->ainsn.insn, 0);
753 p->ainsn.insn = NULL;
759 void arch_arm_kprobe(struct kprobe *p)
761 u8 int3 = INT3_INSN_OPCODE;
763 text_poke(p->addr, &int3, 1);
765 perf_event_text_poke(p->addr, &p->opcode, 1, &int3, 1);
768 void arch_disarm_kprobe(struct kprobe *p)
770 u8 int3 = INT3_INSN_OPCODE;
772 perf_event_text_poke(p->addr, &int3, 1, &p->opcode, 1);
773 text_poke(p->addr, &p->opcode, 1);
777 void arch_remove_kprobe(struct kprobe *p)
780 /* Record the perf event before freeing the slot */
781 perf_event_text_poke(p->ainsn.insn, p->ainsn.insn,
782 p->ainsn.tp_len, NULL, 0);
783 free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
784 p->ainsn.insn = NULL;
788 static nokprobe_inline void
789 save_previous_kprobe(struct kprobe_ctlblk *kcb)
791 kcb->prev_kprobe.kp = kprobe_running();
792 kcb->prev_kprobe.status = kcb->kprobe_status;
793 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
794 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
797 static nokprobe_inline void
798 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
800 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
801 kcb->kprobe_status = kcb->prev_kprobe.status;
802 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
803 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
806 static nokprobe_inline void
807 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
808 struct kprobe_ctlblk *kcb)
810 __this_cpu_write(current_kprobe, p);
811 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
812 = (regs->flags & X86_EFLAGS_IF);
815 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
817 unsigned long *sara = stack_addr(regs);
819 ri->ret_addr = (kprobe_opcode_t *) *sara;
822 /* Replace the return addr with trampoline addr */
823 *sara = (unsigned long) &kretprobe_trampoline;
825 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
827 static void kprobe_post_process(struct kprobe *cur, struct pt_regs *regs,
828 struct kprobe_ctlblk *kcb)
830 /* Restore back the original saved kprobes variables and continue. */
831 if (kcb->kprobe_status == KPROBE_REENTER) {
832 /* This will restore both kcb and current_kprobe */
833 restore_previous_kprobe(kcb);
836 * Always update the kcb status because
837 * reset_curent_kprobe() doesn't update kcb.
839 kcb->kprobe_status = KPROBE_HIT_SSDONE;
840 if (cur->post_handler)
841 cur->post_handler(cur, regs, 0);
842 reset_current_kprobe();
845 NOKPROBE_SYMBOL(kprobe_post_process);
847 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
848 struct kprobe_ctlblk *kcb, int reenter)
850 if (setup_detour_execution(p, regs, reenter))
853 #if !defined(CONFIG_PREEMPTION)
854 if (p->ainsn.boostable) {
855 /* Boost up -- we can execute copied instructions directly */
857 reset_current_kprobe();
859 * Reentering boosted probe doesn't reset current_kprobe,
860 * nor set current_kprobe, because it doesn't use single
863 regs->ip = (unsigned long)p->ainsn.insn;
868 save_previous_kprobe(kcb);
869 set_current_kprobe(p, regs, kcb);
870 kcb->kprobe_status = KPROBE_REENTER;
872 kcb->kprobe_status = KPROBE_HIT_SS;
874 if (p->ainsn.emulate_op) {
875 p->ainsn.emulate_op(p, regs);
876 kprobe_post_process(p, regs, kcb);
880 /* Disable interrupt, and set ip register on trampoline */
881 regs->flags &= ~X86_EFLAGS_IF;
882 regs->ip = (unsigned long)p->ainsn.insn;
884 NOKPROBE_SYMBOL(setup_singlestep);
887 * Called after single-stepping. p->addr is the address of the
888 * instruction whose first byte has been replaced by the "int3"
889 * instruction. To avoid the SMP problems that can occur when we
890 * temporarily put back the original opcode to single-step, we
891 * single-stepped a copy of the instruction. The address of this
892 * copy is p->ainsn.insn. We also doesn't use trap, but "int3" again
893 * right after the copied instruction.
894 * Different from the trap single-step, "int3" single-step can not
895 * handle the instruction which changes the ip register, e.g. jmp,
896 * call, conditional jmp, and the instructions which changes the IF
897 * flags because interrupt must be disabled around the single-stepping.
898 * Such instructions are software emulated, but others are single-stepped
901 * When the 2nd "int3" handled, the regs->ip and regs->flags needs to
902 * be adjusted, so that we can resume execution on correct code.
904 static void resume_singlestep(struct kprobe *p, struct pt_regs *regs,
905 struct kprobe_ctlblk *kcb)
907 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
908 unsigned long orig_ip = (unsigned long)p->addr;
910 /* Restore saved interrupt flag and ip register */
911 regs->flags |= kcb->kprobe_saved_flags;
912 /* Note that regs->ip is executed int3 so must be a step back */
913 regs->ip += (orig_ip - copy_ip) - INT3_INSN_SIZE;
915 NOKPROBE_SYMBOL(resume_singlestep);
918 * We have reentered the kprobe_handler(), since another probe was hit while
919 * within the handler. We save the original kprobes variables and just single
920 * step on the instruction of the new probe without calling any user handlers.
922 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
923 struct kprobe_ctlblk *kcb)
925 switch (kcb->kprobe_status) {
926 case KPROBE_HIT_SSDONE:
927 case KPROBE_HIT_ACTIVE:
929 kprobes_inc_nmissed_count(p);
930 setup_singlestep(p, regs, kcb, 1);
933 /* A probe has been hit in the codepath leading up to, or just
934 * after, single-stepping of a probed instruction. This entire
935 * codepath should strictly reside in .kprobes.text section.
936 * Raise a BUG or we'll continue in an endless reentering loop
937 * and eventually a stack overflow.
939 pr_err("Unrecoverable kprobe detected.\n");
943 /* impossible cases */
950 NOKPROBE_SYMBOL(reenter_kprobe);
952 static nokprobe_inline int kprobe_is_ss(struct kprobe_ctlblk *kcb)
954 return (kcb->kprobe_status == KPROBE_HIT_SS ||
955 kcb->kprobe_status == KPROBE_REENTER);
959 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
960 * remain disabled throughout this function.
962 int kprobe_int3_handler(struct pt_regs *regs)
964 kprobe_opcode_t *addr;
966 struct kprobe_ctlblk *kcb;
971 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
973 * We don't want to be preempted for the entire duration of kprobe
974 * processing. Since int3 and debug trap disables irqs and we clear
975 * IF while singlestepping, it must be no preemptible.
978 kcb = get_kprobe_ctlblk();
979 p = get_kprobe(addr);
982 if (kprobe_running()) {
983 if (reenter_kprobe(p, regs, kcb))
986 set_current_kprobe(p, regs, kcb);
987 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
990 * If we have no pre-handler or it returned 0, we
991 * continue with normal processing. If we have a
992 * pre-handler and it returned non-zero, that means
993 * user handler setup registers to exit to another
994 * instruction, we must skip the single stepping.
996 if (!p->pre_handler || !p->pre_handler(p, regs))
997 setup_singlestep(p, regs, kcb, 0);
999 reset_current_kprobe();
1002 } else if (kprobe_is_ss(kcb)) {
1003 p = kprobe_running();
1004 if ((unsigned long)p->ainsn.insn < regs->ip &&
1005 (unsigned long)p->ainsn.insn + MAX_INSN_SIZE > regs->ip) {
1006 /* Most provably this is the second int3 for singlestep */
1007 resume_singlestep(p, regs, kcb);
1008 kprobe_post_process(p, regs, kcb);
1013 if (*addr != INT3_INSN_OPCODE) {
1015 * The breakpoint instruction was removed right
1016 * after we hit it. Another cpu has removed
1017 * either a probepoint or a debugger breakpoint
1018 * at this address. In either case, no further
1019 * handling of this interrupt is appropriate.
1020 * Back up over the (now missing) int3 and run
1021 * the original instruction.
1023 regs->ip = (unsigned long)addr;
1025 } /* else: not a kprobe fault; let the kernel handle it */
1029 NOKPROBE_SYMBOL(kprobe_int3_handler);
1032 * When a retprobed function returns, this code saves registers and
1033 * calls trampoline_handler() runs, which calls the kretprobe's handler.
1037 ".global kretprobe_trampoline\n"
1038 ".type kretprobe_trampoline, @function\n"
1039 "kretprobe_trampoline:\n"
1040 /* We don't bother saving the ss register */
1041 #ifdef CONFIG_X86_64
1045 " movq %rsp, %rdi\n"
1046 " call trampoline_handler\n"
1047 /* Replace saved sp with true return address. */
1048 " movq %rax, 19*8(%rsp)\n"
1055 " movl %esp, %eax\n"
1056 " call trampoline_handler\n"
1057 /* Replace saved sp with true return address. */
1058 " movl %eax, 15*4(%esp)\n"
1063 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
1065 NOKPROBE_SYMBOL(kretprobe_trampoline);
1066 STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
1070 * Called from kretprobe_trampoline
1072 __used __visible void *trampoline_handler(struct pt_regs *regs)
1074 /* fixup registers */
1075 regs->cs = __KERNEL_CS;
1076 #ifdef CONFIG_X86_32
1079 regs->ip = (unsigned long)&kretprobe_trampoline;
1080 regs->orig_ax = ~0UL;
1082 return (void *)kretprobe_trampoline_handler(regs, &kretprobe_trampoline, ®s->sp);
1084 NOKPROBE_SYMBOL(trampoline_handler);
1086 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
1088 struct kprobe *cur = kprobe_running();
1089 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1091 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
1092 /* This must happen on single-stepping */
1093 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
1094 kcb->kprobe_status != KPROBE_REENTER);
1096 * We are here because the instruction being single
1097 * stepped caused a page fault. We reset the current
1098 * kprobe and the ip points back to the probe address
1099 * and allow the page fault handler to continue as a
1100 * normal page fault.
1102 regs->ip = (unsigned long)cur->addr;
1105 * If the IF flag was set before the kprobe hit,
1108 regs->flags |= kcb->kprobe_old_flags;
1110 if (kcb->kprobe_status == KPROBE_REENTER)
1111 restore_previous_kprobe(kcb);
1113 reset_current_kprobe();
1114 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
1115 kcb->kprobe_status == KPROBE_HIT_SSDONE) {
1117 * We increment the nmissed count for accounting,
1118 * we can also use npre/npostfault count for accounting
1119 * these specific fault cases.
1121 kprobes_inc_nmissed_count(cur);
1124 * We come here because instructions in the pre/post
1125 * handler caused the page_fault, this could happen
1126 * if handler tries to access user space by
1127 * copy_from_user(), get_user() etc. Let the
1128 * user-specified handler try to fix it first.
1130 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
1136 NOKPROBE_SYMBOL(kprobe_fault_handler);
1138 int __init arch_populate_kprobe_blacklist(void)
1140 return kprobe_add_area_blacklist((unsigned long)__entry_text_start,
1141 (unsigned long)__entry_text_end);
1144 int __init arch_init_kprobes(void)
1149 int arch_trampoline_kprobe(struct kprobe *p)