Loading...
Note: File does not exist in v3.5.6.
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/ftrace.h>
41#include <linux/frame.h>
42#include <linux/kasan.h>
43#include <linux/moduleloader.h>
44#include <linux/vmalloc.h>
45#include <linux/pgtable.h>
46
47#include <asm/text-patching.h>
48#include <asm/cacheflush.h>
49#include <asm/desc.h>
50#include <linux/uaccess.h>
51#include <asm/alternative.h>
52#include <asm/insn.h>
53#include <asm/debugreg.h>
54#include <asm/set_memory.h>
55
56#include "common.h"
57
58DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
59DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
60
61#define stack_addr(regs) ((unsigned long *)regs->sp)
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 */
76static 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
100struct 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
106const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
107
108static nokprobe_inline void
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'.*/
122void synthesize_reljump(void *dest, void *from, void *to)
123{
124 __synthesize_relative_insn(dest, from, to, JMP32_INSN_OPCODE);
125}
126NOKPROBE_SYMBOL(synthesize_reljump);
127
128/* Insert a call instruction at address 'from', which calls address 'to'.*/
129void synthesize_relcall(void *dest, void *from, void *to)
130{
131 __synthesize_relative_insn(dest, from, to, CALL_INSN_OPCODE);
132}
133NOKPROBE_SYMBOL(synthesize_relcall);
134
135/*
136 * Skip the prefixes of the instruction.
137 */
138static kprobe_opcode_t *skip_prefixes(kprobe_opcode_t *insn)
139{
140 insn_attr_t attr;
141
142 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
143 while (inat_is_legacy_prefix(attr)) {
144 insn++;
145 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
146 }
147#ifdef CONFIG_X86_64
148 if (inat_is_rex_prefix(attr))
149 insn++;
150#endif
151 return insn;
152}
153NOKPROBE_SYMBOL(skip_prefixes);
154
155/*
156 * Returns non-zero if INSN is boostable.
157 * RIP relative instructions are adjusted at copying time in 64 bits mode
158 */
159int can_boost(struct insn *insn, void *addr)
160{
161 kprobe_opcode_t opcode;
162
163 if (search_exception_tables((unsigned long)addr))
164 return 0; /* Page fault may occur on this address. */
165
166 /* 2nd-byte opcode */
167 if (insn->opcode.nbytes == 2)
168 return test_bit(insn->opcode.bytes[1],
169 (unsigned long *)twobyte_is_boostable);
170
171 if (insn->opcode.nbytes != 1)
172 return 0;
173
174 /* Can't boost Address-size override prefix */
175 if (unlikely(inat_is_address_size_prefix(insn->attr)))
176 return 0;
177
178 opcode = insn->opcode.bytes[0];
179
180 switch (opcode & 0xf0) {
181 case 0x60:
182 /* can't boost "bound" */
183 return (opcode != 0x62);
184 case 0x70:
185 return 0; /* can't boost conditional jump */
186 case 0x90:
187 return opcode != 0x9a; /* can't boost call far */
188 case 0xc0:
189 /* can't boost software-interruptions */
190 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
191 case 0xd0:
192 /* can boost AA* and XLAT */
193 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
194 case 0xe0:
195 /* can boost in/out and absolute jmps */
196 return ((opcode & 0x04) || opcode == 0xea);
197 case 0xf0:
198 /* clear and set flags are boostable */
199 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
200 default:
201 /* CS override prefix and call are not boostable */
202 return (opcode != 0x2e && opcode != 0x9a);
203 }
204}
205
206static unsigned long
207__recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
208{
209 struct kprobe *kp;
210 unsigned long faddr;
211
212 kp = get_kprobe((void *)addr);
213 faddr = ftrace_location(addr);
214 /*
215 * Addresses inside the ftrace location are refused by
216 * arch_check_ftrace_location(). Something went terribly wrong
217 * if such an address is checked here.
218 */
219 if (WARN_ON(faddr && faddr != addr))
220 return 0UL;
221 /*
222 * Use the current code if it is not modified by Kprobe
223 * and it cannot be modified by ftrace.
224 */
225 if (!kp && !faddr)
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, in case on normal Kprobe, kp->opcode has a copy
236 * of the first byte of the probed instruction, which is overwritten
237 * by int3. And the instruction at kp->addr is not modified by kprobes
238 * except for the first byte, we can recover the original instruction
239 * from it and kp->opcode.
240 *
241 * In case of Kprobes using ftrace, we do not have a copy of
242 * the original instruction. In fact, the ftrace location might
243 * be modified at anytime and even could be in an inconsistent state.
244 * Fortunately, we know that the original code is the ideal 5-byte
245 * long NOP.
246 */
247 if (copy_from_kernel_nofault(buf, (void *)addr,
248 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
249 return 0UL;
250
251 if (faddr)
252 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
253 else
254 buf[0] = kp->opcode;
255 return (unsigned long)buf;
256}
257
258/*
259 * Recover the probed instruction at addr for further analysis.
260 * Caller must lock kprobes by kprobe_mutex, or disable preemption
261 * for preventing to release referencing kprobes.
262 * Returns zero if the instruction can not get recovered (or access failed).
263 */
264unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
265{
266 unsigned long __addr;
267
268 __addr = __recover_optprobed_insn(buf, addr);
269 if (__addr != addr)
270 return __addr;
271
272 return __recover_probed_insn(buf, addr);
273}
274
275/* Check if paddr is at an instruction boundary */
276static int can_probe(unsigned long paddr)
277{
278 unsigned long addr, __addr, offset = 0;
279 struct insn insn;
280 kprobe_opcode_t buf[MAX_INSN_SIZE];
281
282 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
283 return 0;
284
285 /* Decode instructions */
286 addr = paddr - offset;
287 while (addr < paddr) {
288 /*
289 * Check if the instruction has been modified by another
290 * kprobe, in which case we replace the breakpoint by the
291 * original instruction in our buffer.
292 * Also, jump optimization will change the breakpoint to
293 * relative-jump. Since the relative-jump itself is
294 * normally used, we just go through if there is no kprobe.
295 */
296 __addr = recover_probed_instruction(buf, addr);
297 if (!__addr)
298 return 0;
299 kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
300 insn_get_length(&insn);
301
302 /*
303 * Another debugging subsystem might insert this breakpoint.
304 * In that case, we can't recover it.
305 */
306 if (insn.opcode.bytes[0] == INT3_INSN_OPCODE)
307 return 0;
308 addr += insn.length;
309 }
310
311 return (addr == paddr);
312}
313
314/*
315 * Returns non-zero if opcode modifies the interrupt flag.
316 */
317static int is_IF_modifier(kprobe_opcode_t *insn)
318{
319 /* Skip prefixes */
320 insn = skip_prefixes(insn);
321
322 switch (*insn) {
323 case 0xfa: /* cli */
324 case 0xfb: /* sti */
325 case 0xcf: /* iret/iretd */
326 case 0x9d: /* popf/popfd */
327 return 1;
328 }
329
330 return 0;
331}
332
333/*
334 * Copy an instruction with recovering modified instruction by kprobes
335 * and adjust the displacement if the instruction uses the %rip-relative
336 * addressing mode. Note that since @real will be the final place of copied
337 * instruction, displacement must be adjust by @real, not @dest.
338 * This returns the length of copied instruction, or 0 if it has an error.
339 */
340int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn)
341{
342 kprobe_opcode_t buf[MAX_INSN_SIZE];
343 unsigned long recovered_insn =
344 recover_probed_instruction(buf, (unsigned long)src);
345
346 if (!recovered_insn || !insn)
347 return 0;
348
349 /* This can access kernel text if given address is not recovered */
350 if (copy_from_kernel_nofault(dest, (void *)recovered_insn,
351 MAX_INSN_SIZE))
352 return 0;
353
354 kernel_insn_init(insn, dest, MAX_INSN_SIZE);
355 insn_get_length(insn);
356
357 /* We can not probe force emulate prefixed instruction */
358 if (insn_has_emulate_prefix(insn))
359 return 0;
360
361 /* Another subsystem puts a breakpoint, failed to recover */
362 if (insn->opcode.bytes[0] == INT3_INSN_OPCODE)
363 return 0;
364
365 /* We should not singlestep on the exception masking instructions */
366 if (insn_masking_exception(insn))
367 return 0;
368
369#ifdef CONFIG_X86_64
370 /* Only x86_64 has RIP relative instructions */
371 if (insn_rip_relative(insn)) {
372 s64 newdisp;
373 u8 *disp;
374 /*
375 * The copied instruction uses the %rip-relative addressing
376 * mode. Adjust the displacement for the difference between
377 * the original location of this instruction and the location
378 * of the copy that will actually be run. The tricky bit here
379 * is making sure that the sign extension happens correctly in
380 * this calculation, since we need a signed 32-bit result to
381 * be sign-extended to 64 bits when it's added to the %rip
382 * value and yield the same 64-bit result that the sign-
383 * extension of the original signed 32-bit displacement would
384 * have given.
385 */
386 newdisp = (u8 *) src + (s64) insn->displacement.value
387 - (u8 *) real;
388 if ((s64) (s32) newdisp != newdisp) {
389 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
390 return 0;
391 }
392 disp = (u8 *) dest + insn_offset_displacement(insn);
393 *(s32 *) disp = (s32) newdisp;
394 }
395#endif
396 return insn->length;
397}
398
399/* Prepare reljump right after instruction to boost */
400static int prepare_boost(kprobe_opcode_t *buf, struct kprobe *p,
401 struct insn *insn)
402{
403 int len = insn->length;
404
405 if (can_boost(insn, p->addr) &&
406 MAX_INSN_SIZE - len >= JMP32_INSN_SIZE) {
407 /*
408 * These instructions can be executed directly if it
409 * jumps back to correct address.
410 */
411 synthesize_reljump(buf + len, p->ainsn.insn + len,
412 p->addr + insn->length);
413 len += JMP32_INSN_SIZE;
414 p->ainsn.boostable = true;
415 } else {
416 p->ainsn.boostable = false;
417 }
418
419 return len;
420}
421
422/* Make page to RO mode when allocate it */
423void *alloc_insn_page(void)
424{
425 void *page;
426
427 page = module_alloc(PAGE_SIZE);
428 if (!page)
429 return NULL;
430
431 set_vm_flush_reset_perms(page);
432 /*
433 * First make the page read-only, and only then make it executable to
434 * prevent it from being W+X in between.
435 */
436 set_memory_ro((unsigned long)page, 1);
437
438 /*
439 * TODO: Once additional kernel code protection mechanisms are set, ensure
440 * that the page was not maliciously altered and it is still zeroed.
441 */
442 set_memory_x((unsigned long)page, 1);
443
444 return page;
445}
446
447/* Recover page to RW mode before releasing it */
448void free_insn_page(void *page)
449{
450 module_memfree(page);
451}
452
453static int arch_copy_kprobe(struct kprobe *p)
454{
455 struct insn insn;
456 kprobe_opcode_t buf[MAX_INSN_SIZE];
457 int len;
458
459 /* Copy an instruction with recovering if other optprobe modifies it.*/
460 len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn);
461 if (!len)
462 return -EINVAL;
463
464 /*
465 * __copy_instruction can modify the displacement of the instruction,
466 * but it doesn't affect boostable check.
467 */
468 len = prepare_boost(buf, p, &insn);
469
470 /* Check whether the instruction modifies Interrupt Flag or not */
471 p->ainsn.if_modifier = is_IF_modifier(buf);
472
473 /* Also, displacement change doesn't affect the first byte */
474 p->opcode = buf[0];
475
476 p->ainsn.tp_len = len;
477 perf_event_text_poke(p->ainsn.insn, NULL, 0, buf, len);
478
479 /* OK, write back the instruction(s) into ROX insn buffer */
480 text_poke(p->ainsn.insn, buf, len);
481
482 return 0;
483}
484
485int arch_prepare_kprobe(struct kprobe *p)
486{
487 int ret;
488
489 if (alternatives_text_reserved(p->addr, p->addr))
490 return -EINVAL;
491
492 if (!can_probe((unsigned long)p->addr))
493 return -EILSEQ;
494 /* insn: must be on special executable page on x86. */
495 p->ainsn.insn = get_insn_slot();
496 if (!p->ainsn.insn)
497 return -ENOMEM;
498
499 ret = arch_copy_kprobe(p);
500 if (ret) {
501 free_insn_slot(p->ainsn.insn, 0);
502 p->ainsn.insn = NULL;
503 }
504
505 return ret;
506}
507
508void arch_arm_kprobe(struct kprobe *p)
509{
510 u8 int3 = INT3_INSN_OPCODE;
511
512 text_poke(p->addr, &int3, 1);
513 text_poke_sync();
514 perf_event_text_poke(p->addr, &p->opcode, 1, &int3, 1);
515}
516
517void arch_disarm_kprobe(struct kprobe *p)
518{
519 u8 int3 = INT3_INSN_OPCODE;
520
521 perf_event_text_poke(p->addr, &int3, 1, &p->opcode, 1);
522 text_poke(p->addr, &p->opcode, 1);
523 text_poke_sync();
524}
525
526void arch_remove_kprobe(struct kprobe *p)
527{
528 if (p->ainsn.insn) {
529 /* Record the perf event before freeing the slot */
530 perf_event_text_poke(p->ainsn.insn, p->ainsn.insn,
531 p->ainsn.tp_len, NULL, 0);
532 free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
533 p->ainsn.insn = NULL;
534 }
535}
536
537static nokprobe_inline void
538save_previous_kprobe(struct kprobe_ctlblk *kcb)
539{
540 kcb->prev_kprobe.kp = kprobe_running();
541 kcb->prev_kprobe.status = kcb->kprobe_status;
542 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
543 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
544}
545
546static nokprobe_inline void
547restore_previous_kprobe(struct kprobe_ctlblk *kcb)
548{
549 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
550 kcb->kprobe_status = kcb->prev_kprobe.status;
551 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
552 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
553}
554
555static nokprobe_inline void
556set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
557 struct kprobe_ctlblk *kcb)
558{
559 __this_cpu_write(current_kprobe, p);
560 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
561 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
562 if (p->ainsn.if_modifier)
563 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
564}
565
566static nokprobe_inline void clear_btf(void)
567{
568 if (test_thread_flag(TIF_BLOCKSTEP)) {
569 unsigned long debugctl = get_debugctlmsr();
570
571 debugctl &= ~DEBUGCTLMSR_BTF;
572 update_debugctlmsr(debugctl);
573 }
574}
575
576static nokprobe_inline void restore_btf(void)
577{
578 if (test_thread_flag(TIF_BLOCKSTEP)) {
579 unsigned long debugctl = get_debugctlmsr();
580
581 debugctl |= DEBUGCTLMSR_BTF;
582 update_debugctlmsr(debugctl);
583 }
584}
585
586void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
587{
588 unsigned long *sara = stack_addr(regs);
589
590 ri->ret_addr = (kprobe_opcode_t *) *sara;
591 ri->fp = sara;
592
593 /* Replace the return addr with trampoline addr */
594 *sara = (unsigned long) &kretprobe_trampoline;
595}
596NOKPROBE_SYMBOL(arch_prepare_kretprobe);
597
598static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
599 struct kprobe_ctlblk *kcb, int reenter)
600{
601 if (setup_detour_execution(p, regs, reenter))
602 return;
603
604#if !defined(CONFIG_PREEMPTION)
605 if (p->ainsn.boostable && !p->post_handler) {
606 /* Boost up -- we can execute copied instructions directly */
607 if (!reenter)
608 reset_current_kprobe();
609 /*
610 * Reentering boosted probe doesn't reset current_kprobe,
611 * nor set current_kprobe, because it doesn't use single
612 * stepping.
613 */
614 regs->ip = (unsigned long)p->ainsn.insn;
615 return;
616 }
617#endif
618 if (reenter) {
619 save_previous_kprobe(kcb);
620 set_current_kprobe(p, regs, kcb);
621 kcb->kprobe_status = KPROBE_REENTER;
622 } else
623 kcb->kprobe_status = KPROBE_HIT_SS;
624 /* Prepare real single stepping */
625 clear_btf();
626 regs->flags |= X86_EFLAGS_TF;
627 regs->flags &= ~X86_EFLAGS_IF;
628 /* single step inline if the instruction is an int3 */
629 if (p->opcode == INT3_INSN_OPCODE)
630 regs->ip = (unsigned long)p->addr;
631 else
632 regs->ip = (unsigned long)p->ainsn.insn;
633}
634NOKPROBE_SYMBOL(setup_singlestep);
635
636/*
637 * We have reentered the kprobe_handler(), since another probe was hit while
638 * within the handler. We save the original kprobes variables and just single
639 * step on the instruction of the new probe without calling any user handlers.
640 */
641static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
642 struct kprobe_ctlblk *kcb)
643{
644 switch (kcb->kprobe_status) {
645 case KPROBE_HIT_SSDONE:
646 case KPROBE_HIT_ACTIVE:
647 case KPROBE_HIT_SS:
648 kprobes_inc_nmissed_count(p);
649 setup_singlestep(p, regs, kcb, 1);
650 break;
651 case KPROBE_REENTER:
652 /* A probe has been hit in the codepath leading up to, or just
653 * after, single-stepping of a probed instruction. This entire
654 * codepath should strictly reside in .kprobes.text section.
655 * Raise a BUG or we'll continue in an endless reentering loop
656 * and eventually a stack overflow.
657 */
658 pr_err("Unrecoverable kprobe detected.\n");
659 dump_kprobe(p);
660 BUG();
661 default:
662 /* impossible cases */
663 WARN_ON(1);
664 return 0;
665 }
666
667 return 1;
668}
669NOKPROBE_SYMBOL(reenter_kprobe);
670
671/*
672 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
673 * remain disabled throughout this function.
674 */
675int kprobe_int3_handler(struct pt_regs *regs)
676{
677 kprobe_opcode_t *addr;
678 struct kprobe *p;
679 struct kprobe_ctlblk *kcb;
680
681 if (user_mode(regs))
682 return 0;
683
684 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
685 /*
686 * We don't want to be preempted for the entire duration of kprobe
687 * processing. Since int3 and debug trap disables irqs and we clear
688 * IF while singlestepping, it must be no preemptible.
689 */
690
691 kcb = get_kprobe_ctlblk();
692 p = get_kprobe(addr);
693
694 if (p) {
695 if (kprobe_running()) {
696 if (reenter_kprobe(p, regs, kcb))
697 return 1;
698 } else {
699 set_current_kprobe(p, regs, kcb);
700 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
701
702 /*
703 * If we have no pre-handler or it returned 0, we
704 * continue with normal processing. If we have a
705 * pre-handler and it returned non-zero, that means
706 * user handler setup registers to exit to another
707 * instruction, we must skip the single stepping.
708 */
709 if (!p->pre_handler || !p->pre_handler(p, regs))
710 setup_singlestep(p, regs, kcb, 0);
711 else
712 reset_current_kprobe();
713 return 1;
714 }
715 } else if (*addr != INT3_INSN_OPCODE) {
716 /*
717 * The breakpoint instruction was removed right
718 * after we hit it. Another cpu has removed
719 * either a probepoint or a debugger breakpoint
720 * at this address. In either case, no further
721 * handling of this interrupt is appropriate.
722 * Back up over the (now missing) int3 and run
723 * the original instruction.
724 */
725 regs->ip = (unsigned long)addr;
726 return 1;
727 } /* else: not a kprobe fault; let the kernel handle it */
728
729 return 0;
730}
731NOKPROBE_SYMBOL(kprobe_int3_handler);
732
733/*
734 * When a retprobed function returns, this code saves registers and
735 * calls trampoline_handler() runs, which calls the kretprobe's handler.
736 */
737asm(
738 ".text\n"
739 ".global kretprobe_trampoline\n"
740 ".type kretprobe_trampoline, @function\n"
741 "kretprobe_trampoline:\n"
742 /* We don't bother saving the ss register */
743#ifdef CONFIG_X86_64
744 " pushq %rsp\n"
745 " pushfq\n"
746 SAVE_REGS_STRING
747 " movq %rsp, %rdi\n"
748 " call trampoline_handler\n"
749 /* Replace saved sp with true return address. */
750 " movq %rax, 19*8(%rsp)\n"
751 RESTORE_REGS_STRING
752 " popfq\n"
753#else
754 " pushl %esp\n"
755 " pushfl\n"
756 SAVE_REGS_STRING
757 " movl %esp, %eax\n"
758 " call trampoline_handler\n"
759 /* Replace saved sp with true return address. */
760 " movl %eax, 15*4(%esp)\n"
761 RESTORE_REGS_STRING
762 " popfl\n"
763#endif
764 " ret\n"
765 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
766);
767NOKPROBE_SYMBOL(kretprobe_trampoline);
768STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
769
770/*
771 * Called from kretprobe_trampoline
772 */
773__used __visible void *trampoline_handler(struct pt_regs *regs)
774{
775 struct kretprobe_instance *ri = NULL;
776 struct hlist_head *head, empty_rp;
777 struct hlist_node *tmp;
778 unsigned long flags, orig_ret_address = 0;
779 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
780 kprobe_opcode_t *correct_ret_addr = NULL;
781 void *frame_pointer;
782 bool skipped = false;
783
784 /*
785 * Set a dummy kprobe for avoiding kretprobe recursion.
786 * Since kretprobe never run in kprobe handler, kprobe must not
787 * be running at this point.
788 */
789 kprobe_busy_begin();
790
791 INIT_HLIST_HEAD(&empty_rp);
792 kretprobe_hash_lock(current, &head, &flags);
793 /* fixup registers */
794 regs->cs = __KERNEL_CS;
795#ifdef CONFIG_X86_32
796 regs->cs |= get_kernel_rpl();
797 regs->gs = 0;
798#endif
799 /* We use pt_regs->sp for return address holder. */
800 frame_pointer = ®s->sp;
801 regs->ip = trampoline_address;
802 regs->orig_ax = ~0UL;
803
804 /*
805 * It is possible to have multiple instances associated with a given
806 * task either because multiple functions in the call path have
807 * return probes installed on them, and/or more than one
808 * return probe was registered for a target function.
809 *
810 * We can handle this because:
811 * - instances are always pushed into the head of the list
812 * - when multiple return probes are registered for the same
813 * function, the (chronologically) first instance's ret_addr
814 * will be the real return address, and all the rest will
815 * point to kretprobe_trampoline.
816 */
817 hlist_for_each_entry(ri, head, hlist) {
818 if (ri->task != current)
819 /* another task is sharing our hash bucket */
820 continue;
821 /*
822 * Return probes must be pushed on this hash list correct
823 * order (same as return order) so that it can be popped
824 * correctly. However, if we find it is pushed it incorrect
825 * order, this means we find a function which should not be
826 * probed, because the wrong order entry is pushed on the
827 * path of processing other kretprobe itself.
828 */
829 if (ri->fp != frame_pointer) {
830 if (!skipped)
831 pr_warn("kretprobe is stacked incorrectly. Trying to fixup.\n");
832 skipped = true;
833 continue;
834 }
835
836 orig_ret_address = (unsigned long)ri->ret_addr;
837 if (skipped)
838 pr_warn("%ps must be blacklisted because of incorrect kretprobe order\n",
839 ri->rp->kp.addr);
840
841 if (orig_ret_address != trampoline_address)
842 /*
843 * This is the real return address. Any other
844 * instances associated with this task are for
845 * other calls deeper on the call stack
846 */
847 break;
848 }
849
850 kretprobe_assert(ri, orig_ret_address, trampoline_address);
851
852 correct_ret_addr = ri->ret_addr;
853 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
854 if (ri->task != current)
855 /* another task is sharing our hash bucket */
856 continue;
857 if (ri->fp != frame_pointer)
858 continue;
859
860 orig_ret_address = (unsigned long)ri->ret_addr;
861 if (ri->rp && ri->rp->handler) {
862 __this_cpu_write(current_kprobe, &ri->rp->kp);
863 ri->ret_addr = correct_ret_addr;
864 ri->rp->handler(ri, regs);
865 __this_cpu_write(current_kprobe, &kprobe_busy);
866 }
867
868 recycle_rp_inst(ri, &empty_rp);
869
870 if (orig_ret_address != trampoline_address)
871 /*
872 * This is the real return address. Any other
873 * instances associated with this task are for
874 * other calls deeper on the call stack
875 */
876 break;
877 }
878
879 kretprobe_hash_unlock(current, &flags);
880
881 kprobe_busy_end();
882
883 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
884 hlist_del(&ri->hlist);
885 kfree(ri);
886 }
887 return (void *)orig_ret_address;
888}
889NOKPROBE_SYMBOL(trampoline_handler);
890
891/*
892 * Called after single-stepping. p->addr is the address of the
893 * instruction whose first byte has been replaced by the "int 3"
894 * instruction. To avoid the SMP problems that can occur when we
895 * temporarily put back the original opcode to single-step, we
896 * single-stepped a copy of the instruction. The address of this
897 * copy is p->ainsn.insn.
898 *
899 * This function prepares to return from the post-single-step
900 * interrupt. We have to fix up the stack as follows:
901 *
902 * 0) Except in the case of absolute or indirect jump or call instructions,
903 * the new ip is relative to the copied instruction. We need to make
904 * it relative to the original instruction.
905 *
906 * 1) If the single-stepped instruction was pushfl, then the TF and IF
907 * flags are set in the just-pushed flags, and may need to be cleared.
908 *
909 * 2) If the single-stepped instruction was a call, the return address
910 * that is atop the stack is the address following the copied instruction.
911 * We need to make it the address following the original instruction.
912 *
913 * If this is the first time we've single-stepped the instruction at
914 * this probepoint, and the instruction is boostable, boost it: add a
915 * jump instruction after the copied instruction, that jumps to the next
916 * instruction after the probepoint.
917 */
918static void resume_execution(struct kprobe *p, struct pt_regs *regs,
919 struct kprobe_ctlblk *kcb)
920{
921 unsigned long *tos = stack_addr(regs);
922 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
923 unsigned long orig_ip = (unsigned long)p->addr;
924 kprobe_opcode_t *insn = p->ainsn.insn;
925
926 /* Skip prefixes */
927 insn = skip_prefixes(insn);
928
929 regs->flags &= ~X86_EFLAGS_TF;
930 switch (*insn) {
931 case 0x9c: /* pushfl */
932 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
933 *tos |= kcb->kprobe_old_flags;
934 break;
935 case 0xc2: /* iret/ret/lret */
936 case 0xc3:
937 case 0xca:
938 case 0xcb:
939 case 0xcf:
940 case 0xea: /* jmp absolute -- ip is correct */
941 /* ip is already adjusted, no more changes required */
942 p->ainsn.boostable = true;
943 goto no_change;
944 case 0xe8: /* call relative - Fix return addr */
945 *tos = orig_ip + (*tos - copy_ip);
946 break;
947#ifdef CONFIG_X86_32
948 case 0x9a: /* call absolute -- same as call absolute, indirect */
949 *tos = orig_ip + (*tos - copy_ip);
950 goto no_change;
951#endif
952 case 0xff:
953 if ((insn[1] & 0x30) == 0x10) {
954 /*
955 * call absolute, indirect
956 * Fix return addr; ip is correct.
957 * But this is not boostable
958 */
959 *tos = orig_ip + (*tos - copy_ip);
960 goto no_change;
961 } else if (((insn[1] & 0x31) == 0x20) ||
962 ((insn[1] & 0x31) == 0x21)) {
963 /*
964 * jmp near and far, absolute indirect
965 * ip is correct. And this is boostable
966 */
967 p->ainsn.boostable = true;
968 goto no_change;
969 }
970 default:
971 break;
972 }
973
974 regs->ip += orig_ip - copy_ip;
975
976no_change:
977 restore_btf();
978}
979NOKPROBE_SYMBOL(resume_execution);
980
981/*
982 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
983 * remain disabled throughout this function.
984 */
985int kprobe_debug_handler(struct pt_regs *regs)
986{
987 struct kprobe *cur = kprobe_running();
988 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
989
990 if (!cur)
991 return 0;
992
993 resume_execution(cur, regs, kcb);
994 regs->flags |= kcb->kprobe_saved_flags;
995
996 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
997 kcb->kprobe_status = KPROBE_HIT_SSDONE;
998 cur->post_handler(cur, regs, 0);
999 }
1000
1001 /* Restore back the original saved kprobes variables and continue. */
1002 if (kcb->kprobe_status == KPROBE_REENTER) {
1003 restore_previous_kprobe(kcb);
1004 goto out;
1005 }
1006 reset_current_kprobe();
1007out:
1008 /*
1009 * if somebody else is singlestepping across a probe point, flags
1010 * will have TF set, in which case, continue the remaining processing
1011 * of do_debug, as if this is not a probe hit.
1012 */
1013 if (regs->flags & X86_EFLAGS_TF)
1014 return 0;
1015
1016 return 1;
1017}
1018NOKPROBE_SYMBOL(kprobe_debug_handler);
1019
1020int 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 * Trap flag (TF) has been set here because this fault
1039 * happened where the single stepping will be done.
1040 * So clear it by resetting the current kprobe:
1041 */
1042 regs->flags &= ~X86_EFLAGS_TF;
1043
1044 /*
1045 * If the TF flag was set before the kprobe hit,
1046 * don't touch it:
1047 */
1048 regs->flags |= kcb->kprobe_old_flags;
1049
1050 if (kcb->kprobe_status == KPROBE_REENTER)
1051 restore_previous_kprobe(kcb);
1052 else
1053 reset_current_kprobe();
1054 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
1055 kcb->kprobe_status == KPROBE_HIT_SSDONE) {
1056 /*
1057 * We increment the nmissed count for accounting,
1058 * we can also use npre/npostfault count for accounting
1059 * these specific fault cases.
1060 */
1061 kprobes_inc_nmissed_count(cur);
1062
1063 /*
1064 * We come here because instructions in the pre/post
1065 * handler caused the page_fault, this could happen
1066 * if handler tries to access user space by
1067 * copy_from_user(), get_user() etc. Let the
1068 * user-specified handler try to fix it first.
1069 */
1070 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
1071 return 1;
1072 }
1073
1074 return 0;
1075}
1076NOKPROBE_SYMBOL(kprobe_fault_handler);
1077
1078int __init arch_populate_kprobe_blacklist(void)
1079{
1080 return kprobe_add_area_blacklist((unsigned long)__entry_text_start,
1081 (unsigned long)__entry_text_end);
1082}
1083
1084int __init arch_init_kprobes(void)
1085{
1086 return 0;
1087}
1088
1089int arch_trampoline_kprobe(struct kprobe *p)
1090{
1091 return 0;
1092}