Loading...
Note: File does not exist in v5.4.
1/*
2 * Kernel Probes (KProbes)
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright (C) IBM Corporation, 2002, 2004
19 *
20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 * Probes initial implementation ( includes contributions from
22 * Rusty Russell).
23 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
24 * interface to access function arguments.
25 * 2004-Oct Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
26 * <prasanna@in.ibm.com> adapted for x86_64 from i386.
27 * 2005-Mar Roland McGrath <roland@redhat.com>
28 * Fixed to handle %rip-relative addressing mode correctly.
29 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
30 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
31 * <prasanna@in.ibm.com> added function-return probes.
32 * 2005-May Rusty Lynch <rusty.lynch@intel.com>
33 * Added function return probes functionality
34 * 2006-Feb Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added
35 * kprobe-booster and kretprobe-booster for i386.
36 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
37 * and kretprobe-booster for x86-64
38 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven
39 * <arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com>
40 * unified x86 kprobes code.
41 */
42#include <linux/kprobes.h>
43#include <linux/ptrace.h>
44#include <linux/string.h>
45#include <linux/slab.h>
46#include <linux/hardirq.h>
47#include <linux/preempt.h>
48#include <linux/module.h>
49#include <linux/kdebug.h>
50#include <linux/kallsyms.h>
51#include <linux/ftrace.h>
52
53#include <asm/cacheflush.h>
54#include <asm/desc.h>
55#include <asm/pgtable.h>
56#include <asm/uaccess.h>
57#include <asm/alternative.h>
58#include <asm/insn.h>
59#include <asm/debugreg.h>
60
61#include "kprobes-common.h"
62
63void jprobe_return_end(void);
64
65DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
66DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
67
68#define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
69
70#define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
71 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
72 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
73 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
74 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
75 << (row % 32))
76 /*
77 * Undefined/reserved opcodes, conditional jump, Opcode Extension
78 * Groups, and some special opcodes can not boost.
79 * This is non-const and volatile to keep gcc from statically
80 * optimizing it out, as variable_test_bit makes gcc think only
81 * *(unsigned long*) is used.
82 */
83static volatile u32 twobyte_is_boostable[256 / 32] = {
84 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
85 /* ---------------------------------------------- */
86 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
87 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
88 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
89 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
90 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
91 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
92 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
93 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
94 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
95 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
96 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
97 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
98 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
99 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
100 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
101 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
102 /* ----------------------------------------------- */
103 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
104};
105#undef W
106
107struct kretprobe_blackpoint kretprobe_blacklist[] = {
108 {"__switch_to", }, /* This function switches only current task, but
109 doesn't switch kernel stack.*/
110 {NULL, NULL} /* Terminator */
111};
112
113const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
114
115static void __kprobes __synthesize_relative_insn(void *from, void *to, u8 op)
116{
117 struct __arch_relative_insn {
118 u8 op;
119 s32 raddr;
120 } __attribute__((packed)) *insn;
121
122 insn = (struct __arch_relative_insn *)from;
123 insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
124 insn->op = op;
125}
126
127/* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
128void __kprobes synthesize_reljump(void *from, void *to)
129{
130 __synthesize_relative_insn(from, to, RELATIVEJUMP_OPCODE);
131}
132
133/* Insert a call instruction at address 'from', which calls address 'to'.*/
134void __kprobes synthesize_relcall(void *from, void *to)
135{
136 __synthesize_relative_insn(from, to, RELATIVECALL_OPCODE);
137}
138
139/*
140 * Skip the prefixes of the instruction.
141 */
142static kprobe_opcode_t *__kprobes skip_prefixes(kprobe_opcode_t *insn)
143{
144 insn_attr_t attr;
145
146 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
147 while (inat_is_legacy_prefix(attr)) {
148 insn++;
149 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
150 }
151#ifdef CONFIG_X86_64
152 if (inat_is_rex_prefix(attr))
153 insn++;
154#endif
155 return insn;
156}
157
158/*
159 * Returns non-zero if opcode is boostable.
160 * RIP relative instructions are adjusted at copying time in 64 bits mode
161 */
162int __kprobes can_boost(kprobe_opcode_t *opcodes)
163{
164 kprobe_opcode_t opcode;
165 kprobe_opcode_t *orig_opcodes = opcodes;
166
167 if (search_exception_tables((unsigned long)opcodes))
168 return 0; /* Page fault may occur on this address. */
169
170retry:
171 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
172 return 0;
173 opcode = *(opcodes++);
174
175 /* 2nd-byte opcode */
176 if (opcode == 0x0f) {
177 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
178 return 0;
179 return test_bit(*opcodes,
180 (unsigned long *)twobyte_is_boostable);
181 }
182
183 switch (opcode & 0xf0) {
184#ifdef CONFIG_X86_64
185 case 0x40:
186 goto retry; /* REX prefix is boostable */
187#endif
188 case 0x60:
189 if (0x63 < opcode && opcode < 0x67)
190 goto retry; /* prefixes */
191 /* can't boost Address-size override and bound */
192 return (opcode != 0x62 && opcode != 0x67);
193 case 0x70:
194 return 0; /* can't boost conditional jump */
195 case 0xc0:
196 /* can't boost software-interruptions */
197 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
198 case 0xd0:
199 /* can boost AA* and XLAT */
200 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
201 case 0xe0:
202 /* can boost in/out and absolute jmps */
203 return ((opcode & 0x04) || opcode == 0xea);
204 case 0xf0:
205 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
206 goto retry; /* lock/rep(ne) prefix */
207 /* clear and set flags are boostable */
208 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
209 default:
210 /* segment override prefixes are boostable */
211 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
212 goto retry; /* prefixes */
213 /* CS override prefix and call are not boostable */
214 return (opcode != 0x2e && opcode != 0x9a);
215 }
216}
217
218static unsigned long
219__recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
220{
221 struct kprobe *kp;
222
223 kp = get_kprobe((void *)addr);
224 /* There is no probe, return original address */
225 if (!kp)
226 return addr;
227
228 /*
229 * Basically, kp->ainsn.insn has an original instruction.
230 * However, RIP-relative instruction can not do single-stepping
231 * at different place, __copy_instruction() tweaks the displacement of
232 * that instruction. In that case, we can't recover the instruction
233 * from the kp->ainsn.insn.
234 *
235 * On the other hand, kp->opcode has a copy of the first byte of
236 * the probed instruction, which is overwritten by int3. And
237 * the instruction at kp->addr is not modified by kprobes except
238 * for the first byte, we can recover the original instruction
239 * from it and kp->opcode.
240 */
241 memcpy(buf, kp->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
242 buf[0] = kp->opcode;
243 return (unsigned long)buf;
244}
245
246/*
247 * Recover the probed instruction at addr for further analysis.
248 * Caller must lock kprobes by kprobe_mutex, or disable preemption
249 * for preventing to release referencing kprobes.
250 */
251unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
252{
253 unsigned long __addr;
254
255 __addr = __recover_optprobed_insn(buf, addr);
256 if (__addr != addr)
257 return __addr;
258
259 return __recover_probed_insn(buf, addr);
260}
261
262/* Check if paddr is at an instruction boundary */
263static int __kprobes can_probe(unsigned long paddr)
264{
265 unsigned long addr, __addr, offset = 0;
266 struct insn insn;
267 kprobe_opcode_t buf[MAX_INSN_SIZE];
268
269 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
270 return 0;
271
272 /* Decode instructions */
273 addr = paddr - offset;
274 while (addr < paddr) {
275 /*
276 * Check if the instruction has been modified by another
277 * kprobe, in which case we replace the breakpoint by the
278 * original instruction in our buffer.
279 * Also, jump optimization will change the breakpoint to
280 * relative-jump. Since the relative-jump itself is
281 * normally used, we just go through if there is no kprobe.
282 */
283 __addr = recover_probed_instruction(buf, addr);
284 kernel_insn_init(&insn, (void *)__addr);
285 insn_get_length(&insn);
286
287 /*
288 * Another debugging subsystem might insert this breakpoint.
289 * In that case, we can't recover it.
290 */
291 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
292 return 0;
293 addr += insn.length;
294 }
295
296 return (addr == paddr);
297}
298
299/*
300 * Returns non-zero if opcode modifies the interrupt flag.
301 */
302static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
303{
304 /* Skip prefixes */
305 insn = skip_prefixes(insn);
306
307 switch (*insn) {
308 case 0xfa: /* cli */
309 case 0xfb: /* sti */
310 case 0xcf: /* iret/iretd */
311 case 0x9d: /* popf/popfd */
312 return 1;
313 }
314
315 return 0;
316}
317
318/*
319 * Copy an instruction and adjust the displacement if the instruction
320 * uses the %rip-relative addressing mode.
321 * If it does, Return the address of the 32-bit displacement word.
322 * If not, return null.
323 * Only applicable to 64-bit x86.
324 */
325int __kprobes __copy_instruction(u8 *dest, u8 *src)
326{
327 struct insn insn;
328 kprobe_opcode_t buf[MAX_INSN_SIZE];
329
330 kernel_insn_init(&insn, (void *)recover_probed_instruction(buf, (unsigned long)src));
331 insn_get_length(&insn);
332 /* Another subsystem puts a breakpoint, failed to recover */
333 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
334 return 0;
335 memcpy(dest, insn.kaddr, insn.length);
336
337#ifdef CONFIG_X86_64
338 if (insn_rip_relative(&insn)) {
339 s64 newdisp;
340 u8 *disp;
341 kernel_insn_init(&insn, dest);
342 insn_get_displacement(&insn);
343 /*
344 * The copied instruction uses the %rip-relative addressing
345 * mode. Adjust the displacement for the difference between
346 * the original location of this instruction and the location
347 * of the copy that will actually be run. The tricky bit here
348 * is making sure that the sign extension happens correctly in
349 * this calculation, since we need a signed 32-bit result to
350 * be sign-extended to 64 bits when it's added to the %rip
351 * value and yield the same 64-bit result that the sign-
352 * extension of the original signed 32-bit displacement would
353 * have given.
354 */
355 newdisp = (u8 *) src + (s64) insn.displacement.value - (u8 *) dest;
356 BUG_ON((s64) (s32) newdisp != newdisp); /* Sanity check. */
357 disp = (u8 *) dest + insn_offset_displacement(&insn);
358 *(s32 *) disp = (s32) newdisp;
359 }
360#endif
361 return insn.length;
362}
363
364static void __kprobes arch_copy_kprobe(struct kprobe *p)
365{
366 /* Copy an instruction with recovering if other optprobe modifies it.*/
367 __copy_instruction(p->ainsn.insn, p->addr);
368
369 /*
370 * __copy_instruction can modify the displacement of the instruction,
371 * but it doesn't affect boostable check.
372 */
373 if (can_boost(p->ainsn.insn))
374 p->ainsn.boostable = 0;
375 else
376 p->ainsn.boostable = -1;
377
378 /* Also, displacement change doesn't affect the first byte */
379 p->opcode = p->ainsn.insn[0];
380}
381
382int __kprobes arch_prepare_kprobe(struct kprobe *p)
383{
384 if (alternatives_text_reserved(p->addr, p->addr))
385 return -EINVAL;
386
387 if (!can_probe((unsigned long)p->addr))
388 return -EILSEQ;
389 /* insn: must be on special executable page on x86. */
390 p->ainsn.insn = get_insn_slot();
391 if (!p->ainsn.insn)
392 return -ENOMEM;
393 arch_copy_kprobe(p);
394 return 0;
395}
396
397void __kprobes arch_arm_kprobe(struct kprobe *p)
398{
399 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
400}
401
402void __kprobes arch_disarm_kprobe(struct kprobe *p)
403{
404 text_poke(p->addr, &p->opcode, 1);
405}
406
407void __kprobes arch_remove_kprobe(struct kprobe *p)
408{
409 if (p->ainsn.insn) {
410 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
411 p->ainsn.insn = NULL;
412 }
413}
414
415static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
416{
417 kcb->prev_kprobe.kp = kprobe_running();
418 kcb->prev_kprobe.status = kcb->kprobe_status;
419 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
420 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
421}
422
423static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
424{
425 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
426 kcb->kprobe_status = kcb->prev_kprobe.status;
427 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
428 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
429}
430
431static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
432 struct kprobe_ctlblk *kcb)
433{
434 __this_cpu_write(current_kprobe, p);
435 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
436 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
437 if (is_IF_modifier(p->ainsn.insn))
438 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
439}
440
441static void __kprobes clear_btf(void)
442{
443 if (test_thread_flag(TIF_BLOCKSTEP)) {
444 unsigned long debugctl = get_debugctlmsr();
445
446 debugctl &= ~DEBUGCTLMSR_BTF;
447 update_debugctlmsr(debugctl);
448 }
449}
450
451static void __kprobes restore_btf(void)
452{
453 if (test_thread_flag(TIF_BLOCKSTEP)) {
454 unsigned long debugctl = get_debugctlmsr();
455
456 debugctl |= DEBUGCTLMSR_BTF;
457 update_debugctlmsr(debugctl);
458 }
459}
460
461void __kprobes
462arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
463{
464 unsigned long *sara = stack_addr(regs);
465
466 ri->ret_addr = (kprobe_opcode_t *) *sara;
467
468 /* Replace the return addr with trampoline addr */
469 *sara = (unsigned long) &kretprobe_trampoline;
470}
471
472static void __kprobes
473setup_singlestep(struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb, int reenter)
474{
475 if (setup_detour_execution(p, regs, reenter))
476 return;
477
478#if !defined(CONFIG_PREEMPT)
479 if (p->ainsn.boostable == 1 && !p->post_handler) {
480 /* Boost up -- we can execute copied instructions directly */
481 if (!reenter)
482 reset_current_kprobe();
483 /*
484 * Reentering boosted probe doesn't reset current_kprobe,
485 * nor set current_kprobe, because it doesn't use single
486 * stepping.
487 */
488 regs->ip = (unsigned long)p->ainsn.insn;
489 preempt_enable_no_resched();
490 return;
491 }
492#endif
493 if (reenter) {
494 save_previous_kprobe(kcb);
495 set_current_kprobe(p, regs, kcb);
496 kcb->kprobe_status = KPROBE_REENTER;
497 } else
498 kcb->kprobe_status = KPROBE_HIT_SS;
499 /* Prepare real single stepping */
500 clear_btf();
501 regs->flags |= X86_EFLAGS_TF;
502 regs->flags &= ~X86_EFLAGS_IF;
503 /* single step inline if the instruction is an int3 */
504 if (p->opcode == BREAKPOINT_INSTRUCTION)
505 regs->ip = (unsigned long)p->addr;
506 else
507 regs->ip = (unsigned long)p->ainsn.insn;
508}
509
510/*
511 * We have reentered the kprobe_handler(), since another probe was hit while
512 * within the handler. We save the original kprobes variables and just single
513 * step on the instruction of the new probe without calling any user handlers.
514 */
515static int __kprobes
516reenter_kprobe(struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb)
517{
518 switch (kcb->kprobe_status) {
519 case KPROBE_HIT_SSDONE:
520 case KPROBE_HIT_ACTIVE:
521 kprobes_inc_nmissed_count(p);
522 setup_singlestep(p, regs, kcb, 1);
523 break;
524 case KPROBE_HIT_SS:
525 /* A probe has been hit in the codepath leading up to, or just
526 * after, single-stepping of a probed instruction. This entire
527 * codepath should strictly reside in .kprobes.text section.
528 * Raise a BUG or we'll continue in an endless reentering loop
529 * and eventually a stack overflow.
530 */
531 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
532 p->addr);
533 dump_kprobe(p);
534 BUG();
535 default:
536 /* impossible cases */
537 WARN_ON(1);
538 return 0;
539 }
540
541 return 1;
542}
543
544/*
545 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
546 * remain disabled throughout this function.
547 */
548static int __kprobes kprobe_handler(struct pt_regs *regs)
549{
550 kprobe_opcode_t *addr;
551 struct kprobe *p;
552 struct kprobe_ctlblk *kcb;
553
554 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
555 /*
556 * We don't want to be preempted for the entire
557 * duration of kprobe processing. We conditionally
558 * re-enable preemption at the end of this function,
559 * and also in reenter_kprobe() and setup_singlestep().
560 */
561 preempt_disable();
562
563 kcb = get_kprobe_ctlblk();
564 p = get_kprobe(addr);
565
566 if (p) {
567 if (kprobe_running()) {
568 if (reenter_kprobe(p, regs, kcb))
569 return 1;
570 } else {
571 set_current_kprobe(p, regs, kcb);
572 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
573
574 /*
575 * If we have no pre-handler or it returned 0, we
576 * continue with normal processing. If we have a
577 * pre-handler and it returned non-zero, it prepped
578 * for calling the break_handler below on re-entry
579 * for jprobe processing, so get out doing nothing
580 * more here.
581 */
582 if (!p->pre_handler || !p->pre_handler(p, regs))
583 setup_singlestep(p, regs, kcb, 0);
584 return 1;
585 }
586 } else if (*addr != BREAKPOINT_INSTRUCTION) {
587 /*
588 * The breakpoint instruction was removed right
589 * after we hit it. Another cpu has removed
590 * either a probepoint or a debugger breakpoint
591 * at this address. In either case, no further
592 * handling of this interrupt is appropriate.
593 * Back up over the (now missing) int3 and run
594 * the original instruction.
595 */
596 regs->ip = (unsigned long)addr;
597 preempt_enable_no_resched();
598 return 1;
599 } else if (kprobe_running()) {
600 p = __this_cpu_read(current_kprobe);
601 if (p->break_handler && p->break_handler(p, regs)) {
602 setup_singlestep(p, regs, kcb, 0);
603 return 1;
604 }
605 } /* else: not a kprobe fault; let the kernel handle it */
606
607 preempt_enable_no_resched();
608 return 0;
609}
610
611/*
612 * When a retprobed function returns, this code saves registers and
613 * calls trampoline_handler() runs, which calls the kretprobe's handler.
614 */
615static void __used __kprobes kretprobe_trampoline_holder(void)
616{
617 asm volatile (
618 ".global kretprobe_trampoline\n"
619 "kretprobe_trampoline: \n"
620#ifdef CONFIG_X86_64
621 /* We don't bother saving the ss register */
622 " pushq %rsp\n"
623 " pushfq\n"
624 SAVE_REGS_STRING
625 " movq %rsp, %rdi\n"
626 " call trampoline_handler\n"
627 /* Replace saved sp with true return address. */
628 " movq %rax, 152(%rsp)\n"
629 RESTORE_REGS_STRING
630 " popfq\n"
631#else
632 " pushf\n"
633 SAVE_REGS_STRING
634 " movl %esp, %eax\n"
635 " call trampoline_handler\n"
636 /* Move flags to cs */
637 " movl 56(%esp), %edx\n"
638 " movl %edx, 52(%esp)\n"
639 /* Replace saved flags with true return address. */
640 " movl %eax, 56(%esp)\n"
641 RESTORE_REGS_STRING
642 " popf\n"
643#endif
644 " ret\n");
645}
646
647/*
648 * Called from kretprobe_trampoline
649 */
650static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
651{
652 struct kretprobe_instance *ri = NULL;
653 struct hlist_head *head, empty_rp;
654 struct hlist_node *node, *tmp;
655 unsigned long flags, orig_ret_address = 0;
656 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
657 kprobe_opcode_t *correct_ret_addr = NULL;
658
659 INIT_HLIST_HEAD(&empty_rp);
660 kretprobe_hash_lock(current, &head, &flags);
661 /* fixup registers */
662#ifdef CONFIG_X86_64
663 regs->cs = __KERNEL_CS;
664#else
665 regs->cs = __KERNEL_CS | get_kernel_rpl();
666 regs->gs = 0;
667#endif
668 regs->ip = trampoline_address;
669 regs->orig_ax = ~0UL;
670
671 /*
672 * It is possible to have multiple instances associated with a given
673 * task either because multiple functions in the call path have
674 * return probes installed on them, and/or more than one
675 * return probe was registered for a target function.
676 *
677 * We can handle this because:
678 * - instances are always pushed into the head of the list
679 * - when multiple return probes are registered for the same
680 * function, the (chronologically) first instance's ret_addr
681 * will be the real return address, and all the rest will
682 * point to kretprobe_trampoline.
683 */
684 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
685 if (ri->task != current)
686 /* another task is sharing our hash bucket */
687 continue;
688
689 orig_ret_address = (unsigned long)ri->ret_addr;
690
691 if (orig_ret_address != trampoline_address)
692 /*
693 * This is the real return address. Any other
694 * instances associated with this task are for
695 * other calls deeper on the call stack
696 */
697 break;
698 }
699
700 kretprobe_assert(ri, orig_ret_address, trampoline_address);
701
702 correct_ret_addr = ri->ret_addr;
703 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
704 if (ri->task != current)
705 /* another task is sharing our hash bucket */
706 continue;
707
708 orig_ret_address = (unsigned long)ri->ret_addr;
709 if (ri->rp && ri->rp->handler) {
710 __this_cpu_write(current_kprobe, &ri->rp->kp);
711 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
712 ri->ret_addr = correct_ret_addr;
713 ri->rp->handler(ri, regs);
714 __this_cpu_write(current_kprobe, NULL);
715 }
716
717 recycle_rp_inst(ri, &empty_rp);
718
719 if (orig_ret_address != trampoline_address)
720 /*
721 * This is the real return address. Any other
722 * instances associated with this task are for
723 * other calls deeper on the call stack
724 */
725 break;
726 }
727
728 kretprobe_hash_unlock(current, &flags);
729
730 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
731 hlist_del(&ri->hlist);
732 kfree(ri);
733 }
734 return (void *)orig_ret_address;
735}
736
737/*
738 * Called after single-stepping. p->addr is the address of the
739 * instruction whose first byte has been replaced by the "int 3"
740 * instruction. To avoid the SMP problems that can occur when we
741 * temporarily put back the original opcode to single-step, we
742 * single-stepped a copy of the instruction. The address of this
743 * copy is p->ainsn.insn.
744 *
745 * This function prepares to return from the post-single-step
746 * interrupt. We have to fix up the stack as follows:
747 *
748 * 0) Except in the case of absolute or indirect jump or call instructions,
749 * the new ip is relative to the copied instruction. We need to make
750 * it relative to the original instruction.
751 *
752 * 1) If the single-stepped instruction was pushfl, then the TF and IF
753 * flags are set in the just-pushed flags, and may need to be cleared.
754 *
755 * 2) If the single-stepped instruction was a call, the return address
756 * that is atop the stack is the address following the copied instruction.
757 * We need to make it the address following the original instruction.
758 *
759 * If this is the first time we've single-stepped the instruction at
760 * this probepoint, and the instruction is boostable, boost it: add a
761 * jump instruction after the copied instruction, that jumps to the next
762 * instruction after the probepoint.
763 */
764static void __kprobes
765resume_execution(struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb)
766{
767 unsigned long *tos = stack_addr(regs);
768 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
769 unsigned long orig_ip = (unsigned long)p->addr;
770 kprobe_opcode_t *insn = p->ainsn.insn;
771
772 /* Skip prefixes */
773 insn = skip_prefixes(insn);
774
775 regs->flags &= ~X86_EFLAGS_TF;
776 switch (*insn) {
777 case 0x9c: /* pushfl */
778 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
779 *tos |= kcb->kprobe_old_flags;
780 break;
781 case 0xc2: /* iret/ret/lret */
782 case 0xc3:
783 case 0xca:
784 case 0xcb:
785 case 0xcf:
786 case 0xea: /* jmp absolute -- ip is correct */
787 /* ip is already adjusted, no more changes required */
788 p->ainsn.boostable = 1;
789 goto no_change;
790 case 0xe8: /* call relative - Fix return addr */
791 *tos = orig_ip + (*tos - copy_ip);
792 break;
793#ifdef CONFIG_X86_32
794 case 0x9a: /* call absolute -- same as call absolute, indirect */
795 *tos = orig_ip + (*tos - copy_ip);
796 goto no_change;
797#endif
798 case 0xff:
799 if ((insn[1] & 0x30) == 0x10) {
800 /*
801 * call absolute, indirect
802 * Fix return addr; ip is correct.
803 * But this is not boostable
804 */
805 *tos = orig_ip + (*tos - copy_ip);
806 goto no_change;
807 } else if (((insn[1] & 0x31) == 0x20) ||
808 ((insn[1] & 0x31) == 0x21)) {
809 /*
810 * jmp near and far, absolute indirect
811 * ip is correct. And this is boostable
812 */
813 p->ainsn.boostable = 1;
814 goto no_change;
815 }
816 default:
817 break;
818 }
819
820 if (p->ainsn.boostable == 0) {
821 if ((regs->ip > copy_ip) &&
822 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
823 /*
824 * These instructions can be executed directly if it
825 * jumps back to correct address.
826 */
827 synthesize_reljump((void *)regs->ip,
828 (void *)orig_ip + (regs->ip - copy_ip));
829 p->ainsn.boostable = 1;
830 } else {
831 p->ainsn.boostable = -1;
832 }
833 }
834
835 regs->ip += orig_ip - copy_ip;
836
837no_change:
838 restore_btf();
839}
840
841/*
842 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
843 * remain disabled throughout this function.
844 */
845static int __kprobes post_kprobe_handler(struct pt_regs *regs)
846{
847 struct kprobe *cur = kprobe_running();
848 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
849
850 if (!cur)
851 return 0;
852
853 resume_execution(cur, regs, kcb);
854 regs->flags |= kcb->kprobe_saved_flags;
855
856 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
857 kcb->kprobe_status = KPROBE_HIT_SSDONE;
858 cur->post_handler(cur, regs, 0);
859 }
860
861 /* Restore back the original saved kprobes variables and continue. */
862 if (kcb->kprobe_status == KPROBE_REENTER) {
863 restore_previous_kprobe(kcb);
864 goto out;
865 }
866 reset_current_kprobe();
867out:
868 preempt_enable_no_resched();
869
870 /*
871 * if somebody else is singlestepping across a probe point, flags
872 * will have TF set, in which case, continue the remaining processing
873 * of do_debug, as if this is not a probe hit.
874 */
875 if (regs->flags & X86_EFLAGS_TF)
876 return 0;
877
878 return 1;
879}
880
881int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
882{
883 struct kprobe *cur = kprobe_running();
884 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
885
886 switch (kcb->kprobe_status) {
887 case KPROBE_HIT_SS:
888 case KPROBE_REENTER:
889 /*
890 * We are here because the instruction being single
891 * stepped caused a page fault. We reset the current
892 * kprobe and the ip points back to the probe address
893 * and allow the page fault handler to continue as a
894 * normal page fault.
895 */
896 regs->ip = (unsigned long)cur->addr;
897 regs->flags |= kcb->kprobe_old_flags;
898 if (kcb->kprobe_status == KPROBE_REENTER)
899 restore_previous_kprobe(kcb);
900 else
901 reset_current_kprobe();
902 preempt_enable_no_resched();
903 break;
904 case KPROBE_HIT_ACTIVE:
905 case KPROBE_HIT_SSDONE:
906 /*
907 * We increment the nmissed count for accounting,
908 * we can also use npre/npostfault count for accounting
909 * these specific fault cases.
910 */
911 kprobes_inc_nmissed_count(cur);
912
913 /*
914 * We come here because instructions in the pre/post
915 * handler caused the page_fault, this could happen
916 * if handler tries to access user space by
917 * copy_from_user(), get_user() etc. Let the
918 * user-specified handler try to fix it first.
919 */
920 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
921 return 1;
922
923 /*
924 * In case the user-specified fault handler returned
925 * zero, try to fix up.
926 */
927 if (fixup_exception(regs))
928 return 1;
929
930 /*
931 * fixup routine could not handle it,
932 * Let do_page_fault() fix it.
933 */
934 break;
935 default:
936 break;
937 }
938 return 0;
939}
940
941/*
942 * Wrapper routine for handling exceptions.
943 */
944int __kprobes
945kprobe_exceptions_notify(struct notifier_block *self, unsigned long val, void *data)
946{
947 struct die_args *args = data;
948 int ret = NOTIFY_DONE;
949
950 if (args->regs && user_mode_vm(args->regs))
951 return ret;
952
953 switch (val) {
954 case DIE_INT3:
955 if (kprobe_handler(args->regs))
956 ret = NOTIFY_STOP;
957 break;
958 case DIE_DEBUG:
959 if (post_kprobe_handler(args->regs)) {
960 /*
961 * Reset the BS bit in dr6 (pointed by args->err) to
962 * denote completion of processing
963 */
964 (*(unsigned long *)ERR_PTR(args->err)) &= ~DR_STEP;
965 ret = NOTIFY_STOP;
966 }
967 break;
968 case DIE_GPF:
969 /*
970 * To be potentially processing a kprobe fault and to
971 * trust the result from kprobe_running(), we have
972 * be non-preemptible.
973 */
974 if (!preemptible() && kprobe_running() &&
975 kprobe_fault_handler(args->regs, args->trapnr))
976 ret = NOTIFY_STOP;
977 break;
978 default:
979 break;
980 }
981 return ret;
982}
983
984int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
985{
986 struct jprobe *jp = container_of(p, struct jprobe, kp);
987 unsigned long addr;
988 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
989
990 kcb->jprobe_saved_regs = *regs;
991 kcb->jprobe_saved_sp = stack_addr(regs);
992 addr = (unsigned long)(kcb->jprobe_saved_sp);
993
994 /*
995 * As Linus pointed out, gcc assumes that the callee
996 * owns the argument space and could overwrite it, e.g.
997 * tailcall optimization. So, to be absolutely safe
998 * we also save and restore enough stack bytes to cover
999 * the argument area.
1000 */
1001 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1002 MIN_STACK_SIZE(addr));
1003 regs->flags &= ~X86_EFLAGS_IF;
1004 trace_hardirqs_off();
1005 regs->ip = (unsigned long)(jp->entry);
1006 return 1;
1007}
1008
1009void __kprobes jprobe_return(void)
1010{
1011 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1012
1013 asm volatile (
1014#ifdef CONFIG_X86_64
1015 " xchg %%rbx,%%rsp \n"
1016#else
1017 " xchgl %%ebx,%%esp \n"
1018#endif
1019 " int3 \n"
1020 " .globl jprobe_return_end\n"
1021 " jprobe_return_end: \n"
1022 " nop \n"::"b"
1023 (kcb->jprobe_saved_sp):"memory");
1024}
1025
1026int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1027{
1028 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1029 u8 *addr = (u8 *) (regs->ip - 1);
1030 struct jprobe *jp = container_of(p, struct jprobe, kp);
1031
1032 if ((addr > (u8 *) jprobe_return) &&
1033 (addr < (u8 *) jprobe_return_end)) {
1034 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1035 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1036 printk(KERN_ERR
1037 "current sp %p does not match saved sp %p\n",
1038 stack_addr(regs), kcb->jprobe_saved_sp);
1039 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1040 show_regs(saved_regs);
1041 printk(KERN_ERR "Current registers\n");
1042 show_regs(regs);
1043 BUG();
1044 }
1045 *regs = kcb->jprobe_saved_regs;
1046 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1047 kcb->jprobes_stack,
1048 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1049 preempt_enable_no_resched();
1050 return 1;
1051 }
1052 return 0;
1053}
1054
1055int __init arch_init_kprobes(void)
1056{
1057 return arch_init_optprobes();
1058}
1059
1060int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1061{
1062 return 0;
1063}