2 * Kernel Probes (KProbes)
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.
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.
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.
18 * Copyright (C) IBM Corporation, 2002, 2004
20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 * Probes initial implementation ( includes contributions from
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.
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/sched/debug.h>
49 #include <linux/extable.h>
50 #include <linux/kdebug.h>
51 #include <linux/kallsyms.h>
52 #include <linux/ftrace.h>
53 #include <linux/frame.h>
54 #include <linux/kasan.h>
56 #include <asm/text-patching.h>
57 #include <asm/cacheflush.h>
59 #include <asm/pgtable.h>
60 #include <linux/uaccess.h>
61 #include <asm/alternative.h>
63 #include <asm/debugreg.h>
67 void jprobe_return_end(void);
69 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
70 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
72 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
74 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
75 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
76 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
77 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
78 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
81 * Undefined/reserved opcodes, conditional jump, Opcode Extension
82 * Groups, and some special opcodes can not boost.
83 * This is non-const and volatile to keep gcc from statically
84 * optimizing it out, as variable_test_bit makes gcc think only
85 * *(unsigned long*) is used.
87 static volatile u32 twobyte_is_boostable[256 / 32] = {
88 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
89 /* ---------------------------------------------- */
90 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
91 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
92 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
93 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
94 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
95 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
96 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
97 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
98 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
99 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
100 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
101 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
102 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
103 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
104 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
105 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
106 /* ----------------------------------------------- */
107 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
111 struct kretprobe_blackpoint kretprobe_blacklist[] = {
112 {"__switch_to", }, /* This function switches only current task, but
113 doesn't switch kernel stack.*/
114 {NULL, NULL} /* Terminator */
117 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
119 static nokprobe_inline void
120 __synthesize_relative_insn(void *from, void *to, u8 op)
122 struct __arch_relative_insn {
127 insn = (struct __arch_relative_insn *)from;
128 insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
132 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
133 void synthesize_reljump(void *from, void *to)
135 __synthesize_relative_insn(from, to, RELATIVEJUMP_OPCODE);
137 NOKPROBE_SYMBOL(synthesize_reljump);
139 /* Insert a call instruction at address 'from', which calls address 'to'.*/
140 void synthesize_relcall(void *from, void *to)
142 __synthesize_relative_insn(from, to, RELATIVECALL_OPCODE);
144 NOKPROBE_SYMBOL(synthesize_relcall);
147 * Skip the prefixes of the instruction.
149 static kprobe_opcode_t *skip_prefixes(kprobe_opcode_t *insn)
153 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
154 while (inat_is_legacy_prefix(attr)) {
156 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
159 if (inat_is_rex_prefix(attr))
164 NOKPROBE_SYMBOL(skip_prefixes);
167 * Returns non-zero if opcode is boostable.
168 * RIP relative instructions are adjusted at copying time in 64 bits mode
170 int can_boost(kprobe_opcode_t *opcodes, void *addr)
172 kprobe_opcode_t opcode;
173 kprobe_opcode_t *orig_opcodes = opcodes;
175 if (search_exception_tables((unsigned long)addr))
176 return 0; /* Page fault may occur on this address. */
179 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
181 opcode = *(opcodes++);
183 /* 2nd-byte opcode */
184 if (opcode == 0x0f) {
185 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
187 return test_bit(*opcodes,
188 (unsigned long *)twobyte_is_boostable);
191 switch (opcode & 0xf0) {
194 goto retry; /* REX prefix is boostable */
197 if (0x63 < opcode && opcode < 0x67)
198 goto retry; /* prefixes */
199 /* can't boost Address-size override and bound */
200 return (opcode != 0x62 && opcode != 0x67);
202 return 0; /* can't boost conditional jump */
204 /* can't boost software-interruptions */
205 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
207 /* can boost AA* and XLAT */
208 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
210 /* can boost in/out and absolute jmps */
211 return ((opcode & 0x04) || opcode == 0xea);
213 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
214 goto retry; /* lock/rep(ne) prefix */
215 /* clear and set flags are boostable */
216 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
218 /* segment override prefixes are boostable */
219 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
220 goto retry; /* prefixes */
221 /* CS override prefix and call are not boostable */
222 return (opcode != 0x2e && opcode != 0x9a);
227 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
232 kp = get_kprobe((void *)addr);
233 faddr = ftrace_location(addr);
235 * Addresses inside the ftrace location are refused by
236 * arch_check_ftrace_location(). Something went terribly wrong
237 * if such an address is checked here.
239 if (WARN_ON(faddr && faddr != addr))
242 * Use the current code if it is not modified by Kprobe
243 * and it cannot be modified by ftrace.
249 * Basically, kp->ainsn.insn has an original instruction.
250 * However, RIP-relative instruction can not do single-stepping
251 * at different place, __copy_instruction() tweaks the displacement of
252 * that instruction. In that case, we can't recover the instruction
253 * from the kp->ainsn.insn.
255 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
256 * of the first byte of the probed instruction, which is overwritten
257 * by int3. And the instruction at kp->addr is not modified by kprobes
258 * except for the first byte, we can recover the original instruction
259 * from it and kp->opcode.
261 * In case of Kprobes using ftrace, we do not have a copy of
262 * the original instruction. In fact, the ftrace location might
263 * be modified at anytime and even could be in an inconsistent state.
264 * Fortunately, we know that the original code is the ideal 5-byte
267 memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
269 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
272 return (unsigned long)buf;
276 * Recover the probed instruction at addr for further analysis.
277 * Caller must lock kprobes by kprobe_mutex, or disable preemption
278 * for preventing to release referencing kprobes.
279 * Returns zero if the instruction can not get recovered.
281 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
283 unsigned long __addr;
285 __addr = __recover_optprobed_insn(buf, addr);
289 return __recover_probed_insn(buf, addr);
292 /* Check if paddr is at an instruction boundary */
293 static int can_probe(unsigned long paddr)
295 unsigned long addr, __addr, offset = 0;
297 kprobe_opcode_t buf[MAX_INSN_SIZE];
299 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
302 /* Decode instructions */
303 addr = paddr - offset;
304 while (addr < paddr) {
306 * Check if the instruction has been modified by another
307 * kprobe, in which case we replace the breakpoint by the
308 * original instruction in our buffer.
309 * Also, jump optimization will change the breakpoint to
310 * relative-jump. Since the relative-jump itself is
311 * normally used, we just go through if there is no kprobe.
313 __addr = recover_probed_instruction(buf, addr);
316 kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
317 insn_get_length(&insn);
320 * Another debugging subsystem might insert this breakpoint.
321 * In that case, we can't recover it.
323 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
328 return (addr == paddr);
332 * Returns non-zero if opcode modifies the interrupt flag.
334 static int is_IF_modifier(kprobe_opcode_t *insn)
337 insn = skip_prefixes(insn);
342 case 0xcf: /* iret/iretd */
343 case 0x9d: /* popf/popfd */
351 * Copy an instruction and adjust the displacement if the instruction
352 * uses the %rip-relative addressing mode.
353 * If it does, Return the address of the 32-bit displacement word.
354 * If not, return null.
355 * Only applicable to 64-bit x86.
357 int __copy_instruction(u8 *dest, u8 *src)
360 kprobe_opcode_t buf[MAX_INSN_SIZE];
362 unsigned long recovered_insn =
363 recover_probed_instruction(buf, (unsigned long)src);
367 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
368 insn_get_length(&insn);
369 length = insn.length;
371 /* Another subsystem puts a breakpoint, failed to recover */
372 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
374 memcpy(dest, insn.kaddr, length);
377 if (insn_rip_relative(&insn)) {
380 kernel_insn_init(&insn, dest, length);
381 insn_get_displacement(&insn);
383 * The copied instruction uses the %rip-relative addressing
384 * mode. Adjust the displacement for the difference between
385 * the original location of this instruction and the location
386 * of the copy that will actually be run. The tricky bit here
387 * is making sure that the sign extension happens correctly in
388 * this calculation, since we need a signed 32-bit result to
389 * be sign-extended to 64 bits when it's added to the %rip
390 * value and yield the same 64-bit result that the sign-
391 * extension of the original signed 32-bit displacement would
394 newdisp = (u8 *) src + (s64) insn.displacement.value - (u8 *) dest;
395 if ((s64) (s32) newdisp != newdisp) {
396 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
397 pr_err("\tSrc: %p, Dest: %p, old disp: %x\n", src, dest, insn.displacement.value);
400 disp = (u8 *) dest + insn_offset_displacement(&insn);
401 *(s32 *) disp = (s32) newdisp;
407 static int arch_copy_kprobe(struct kprobe *p)
411 /* Copy an instruction with recovering if other optprobe modifies it.*/
412 ret = __copy_instruction(p->ainsn.insn, p->addr);
417 * __copy_instruction can modify the displacement of the instruction,
418 * but it doesn't affect boostable check.
420 if (can_boost(p->ainsn.insn, p->addr))
421 p->ainsn.boostable = 0;
423 p->ainsn.boostable = -1;
425 /* Check whether the instruction modifies Interrupt Flag or not */
426 p->ainsn.if_modifier = is_IF_modifier(p->ainsn.insn);
428 /* Also, displacement change doesn't affect the first byte */
429 p->opcode = p->ainsn.insn[0];
434 int arch_prepare_kprobe(struct kprobe *p)
436 if (alternatives_text_reserved(p->addr, p->addr))
439 if (!can_probe((unsigned long)p->addr))
441 /* insn: must be on special executable page on x86. */
442 p->ainsn.insn = get_insn_slot();
446 return arch_copy_kprobe(p);
449 void arch_arm_kprobe(struct kprobe *p)
451 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
454 void arch_disarm_kprobe(struct kprobe *p)
456 text_poke(p->addr, &p->opcode, 1);
459 void arch_remove_kprobe(struct kprobe *p)
462 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
463 p->ainsn.insn = NULL;
467 static nokprobe_inline void
468 save_previous_kprobe(struct kprobe_ctlblk *kcb)
470 kcb->prev_kprobe.kp = kprobe_running();
471 kcb->prev_kprobe.status = kcb->kprobe_status;
472 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
473 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
476 static nokprobe_inline void
477 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
479 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
480 kcb->kprobe_status = kcb->prev_kprobe.status;
481 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
482 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
485 static nokprobe_inline void
486 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
487 struct kprobe_ctlblk *kcb)
489 __this_cpu_write(current_kprobe, p);
490 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
491 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
492 if (p->ainsn.if_modifier)
493 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
496 static nokprobe_inline void clear_btf(void)
498 if (test_thread_flag(TIF_BLOCKSTEP)) {
499 unsigned long debugctl = get_debugctlmsr();
501 debugctl &= ~DEBUGCTLMSR_BTF;
502 update_debugctlmsr(debugctl);
506 static nokprobe_inline void restore_btf(void)
508 if (test_thread_flag(TIF_BLOCKSTEP)) {
509 unsigned long debugctl = get_debugctlmsr();
511 debugctl |= DEBUGCTLMSR_BTF;
512 update_debugctlmsr(debugctl);
516 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
518 unsigned long *sara = stack_addr(regs);
520 ri->ret_addr = (kprobe_opcode_t *) *sara;
522 /* Replace the return addr with trampoline addr */
523 *sara = (unsigned long) &kretprobe_trampoline;
525 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
527 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
528 struct kprobe_ctlblk *kcb, int reenter)
530 if (setup_detour_execution(p, regs, reenter))
533 #if !defined(CONFIG_PREEMPT)
534 if (p->ainsn.boostable == 1 && !p->post_handler) {
535 /* Boost up -- we can execute copied instructions directly */
537 reset_current_kprobe();
539 * Reentering boosted probe doesn't reset current_kprobe,
540 * nor set current_kprobe, because it doesn't use single
543 regs->ip = (unsigned long)p->ainsn.insn;
544 preempt_enable_no_resched();
549 save_previous_kprobe(kcb);
550 set_current_kprobe(p, regs, kcb);
551 kcb->kprobe_status = KPROBE_REENTER;
553 kcb->kprobe_status = KPROBE_HIT_SS;
554 /* Prepare real single stepping */
556 regs->flags |= X86_EFLAGS_TF;
557 regs->flags &= ~X86_EFLAGS_IF;
558 /* single step inline if the instruction is an int3 */
559 if (p->opcode == BREAKPOINT_INSTRUCTION)
560 regs->ip = (unsigned long)p->addr;
562 regs->ip = (unsigned long)p->ainsn.insn;
564 NOKPROBE_SYMBOL(setup_singlestep);
567 * We have reentered the kprobe_handler(), since another probe was hit while
568 * within the handler. We save the original kprobes variables and just single
569 * step on the instruction of the new probe without calling any user handlers.
571 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
572 struct kprobe_ctlblk *kcb)
574 switch (kcb->kprobe_status) {
575 case KPROBE_HIT_SSDONE:
576 case KPROBE_HIT_ACTIVE:
578 kprobes_inc_nmissed_count(p);
579 setup_singlestep(p, regs, kcb, 1);
582 /* A probe has been hit in the codepath leading up to, or just
583 * after, single-stepping of a probed instruction. This entire
584 * codepath should strictly reside in .kprobes.text section.
585 * Raise a BUG or we'll continue in an endless reentering loop
586 * and eventually a stack overflow.
588 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
593 /* impossible cases */
600 NOKPROBE_SYMBOL(reenter_kprobe);
603 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
604 * remain disabled throughout this function.
606 int kprobe_int3_handler(struct pt_regs *regs)
608 kprobe_opcode_t *addr;
610 struct kprobe_ctlblk *kcb;
615 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
617 * We don't want to be preempted for the entire
618 * duration of kprobe processing. We conditionally
619 * re-enable preemption at the end of this function,
620 * and also in reenter_kprobe() and setup_singlestep().
624 kcb = get_kprobe_ctlblk();
625 p = get_kprobe(addr);
628 if (kprobe_running()) {
629 if (reenter_kprobe(p, regs, kcb))
632 set_current_kprobe(p, regs, kcb);
633 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
636 * If we have no pre-handler or it returned 0, we
637 * continue with normal processing. If we have a
638 * pre-handler and it returned non-zero, it prepped
639 * for calling the break_handler below on re-entry
640 * for jprobe processing, so get out doing nothing
643 if (!p->pre_handler || !p->pre_handler(p, regs))
644 setup_singlestep(p, regs, kcb, 0);
647 } else if (*addr != BREAKPOINT_INSTRUCTION) {
649 * The breakpoint instruction was removed right
650 * after we hit it. Another cpu has removed
651 * either a probepoint or a debugger breakpoint
652 * at this address. In either case, no further
653 * handling of this interrupt is appropriate.
654 * Back up over the (now missing) int3 and run
655 * the original instruction.
657 regs->ip = (unsigned long)addr;
658 preempt_enable_no_resched();
660 } else if (kprobe_running()) {
661 p = __this_cpu_read(current_kprobe);
662 if (p->break_handler && p->break_handler(p, regs)) {
663 if (!skip_singlestep(p, regs, kcb))
664 setup_singlestep(p, regs, kcb, 0);
667 } /* else: not a kprobe fault; let the kernel handle it */
669 preempt_enable_no_resched();
672 NOKPROBE_SYMBOL(kprobe_int3_handler);
675 * When a retprobed function returns, this code saves registers and
676 * calls trampoline_handler() runs, which calls the kretprobe's handler.
679 ".global kretprobe_trampoline\n"
680 ".type kretprobe_trampoline, @function\n"
681 "kretprobe_trampoline:\n"
683 /* We don't bother saving the ss register */
688 " call trampoline_handler\n"
689 /* Replace saved sp with true return address. */
690 " movq %rax, 152(%rsp)\n"
697 " call trampoline_handler\n"
698 /* Move flags to cs */
699 " movl 56(%esp), %edx\n"
700 " movl %edx, 52(%esp)\n"
701 /* Replace saved flags with true return address. */
702 " movl %eax, 56(%esp)\n"
707 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
709 NOKPROBE_SYMBOL(kretprobe_trampoline);
710 STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
713 * Called from kretprobe_trampoline
715 __visible __used void *trampoline_handler(struct pt_regs *regs)
717 struct kretprobe_instance *ri = NULL;
718 struct hlist_head *head, empty_rp;
719 struct hlist_node *tmp;
720 unsigned long flags, orig_ret_address = 0;
721 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
722 kprobe_opcode_t *correct_ret_addr = NULL;
724 INIT_HLIST_HEAD(&empty_rp);
725 kretprobe_hash_lock(current, &head, &flags);
726 /* fixup registers */
728 regs->cs = __KERNEL_CS;
730 regs->cs = __KERNEL_CS | get_kernel_rpl();
733 regs->ip = trampoline_address;
734 regs->orig_ax = ~0UL;
737 * It is possible to have multiple instances associated with a given
738 * task either because multiple functions in the call path have
739 * return probes installed on them, and/or more than one
740 * return probe was registered for a target function.
742 * We can handle this because:
743 * - instances are always pushed into the head of the list
744 * - when multiple return probes are registered for the same
745 * function, the (chronologically) first instance's ret_addr
746 * will be the real return address, and all the rest will
747 * point to kretprobe_trampoline.
749 hlist_for_each_entry(ri, head, hlist) {
750 if (ri->task != current)
751 /* another task is sharing our hash bucket */
754 orig_ret_address = (unsigned long)ri->ret_addr;
756 if (orig_ret_address != trampoline_address)
758 * This is the real return address. Any other
759 * instances associated with this task are for
760 * other calls deeper on the call stack
765 kretprobe_assert(ri, orig_ret_address, trampoline_address);
767 correct_ret_addr = ri->ret_addr;
768 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
769 if (ri->task != current)
770 /* another task is sharing our hash bucket */
773 orig_ret_address = (unsigned long)ri->ret_addr;
774 if (ri->rp && ri->rp->handler) {
775 __this_cpu_write(current_kprobe, &ri->rp->kp);
776 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
777 ri->ret_addr = correct_ret_addr;
778 ri->rp->handler(ri, regs);
779 __this_cpu_write(current_kprobe, NULL);
782 recycle_rp_inst(ri, &empty_rp);
784 if (orig_ret_address != trampoline_address)
786 * This is the real return address. Any other
787 * instances associated with this task are for
788 * other calls deeper on the call stack
793 kretprobe_hash_unlock(current, &flags);
795 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
796 hlist_del(&ri->hlist);
799 return (void *)orig_ret_address;
801 NOKPROBE_SYMBOL(trampoline_handler);
804 * Called after single-stepping. p->addr is the address of the
805 * instruction whose first byte has been replaced by the "int 3"
806 * instruction. To avoid the SMP problems that can occur when we
807 * temporarily put back the original opcode to single-step, we
808 * single-stepped a copy of the instruction. The address of this
809 * copy is p->ainsn.insn.
811 * This function prepares to return from the post-single-step
812 * interrupt. We have to fix up the stack as follows:
814 * 0) Except in the case of absolute or indirect jump or call instructions,
815 * the new ip is relative to the copied instruction. We need to make
816 * it relative to the original instruction.
818 * 1) If the single-stepped instruction was pushfl, then the TF and IF
819 * flags are set in the just-pushed flags, and may need to be cleared.
821 * 2) If the single-stepped instruction was a call, the return address
822 * that is atop the stack is the address following the copied instruction.
823 * We need to make it the address following the original instruction.
825 * If this is the first time we've single-stepped the instruction at
826 * this probepoint, and the instruction is boostable, boost it: add a
827 * jump instruction after the copied instruction, that jumps to the next
828 * instruction after the probepoint.
830 static void resume_execution(struct kprobe *p, struct pt_regs *regs,
831 struct kprobe_ctlblk *kcb)
833 unsigned long *tos = stack_addr(regs);
834 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
835 unsigned long orig_ip = (unsigned long)p->addr;
836 kprobe_opcode_t *insn = p->ainsn.insn;
839 insn = skip_prefixes(insn);
841 regs->flags &= ~X86_EFLAGS_TF;
843 case 0x9c: /* pushfl */
844 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
845 *tos |= kcb->kprobe_old_flags;
847 case 0xc2: /* iret/ret/lret */
852 case 0xea: /* jmp absolute -- ip is correct */
853 /* ip is already adjusted, no more changes required */
854 p->ainsn.boostable = 1;
856 case 0xe8: /* call relative - Fix return addr */
857 *tos = orig_ip + (*tos - copy_ip);
860 case 0x9a: /* call absolute -- same as call absolute, indirect */
861 *tos = orig_ip + (*tos - copy_ip);
865 if ((insn[1] & 0x30) == 0x10) {
867 * call absolute, indirect
868 * Fix return addr; ip is correct.
869 * But this is not boostable
871 *tos = orig_ip + (*tos - copy_ip);
873 } else if (((insn[1] & 0x31) == 0x20) ||
874 ((insn[1] & 0x31) == 0x21)) {
876 * jmp near and far, absolute indirect
877 * ip is correct. And this is boostable
879 p->ainsn.boostable = 1;
886 if (p->ainsn.boostable == 0) {
887 if ((regs->ip > copy_ip) &&
888 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
890 * These instructions can be executed directly if it
891 * jumps back to correct address.
893 synthesize_reljump((void *)regs->ip,
894 (void *)orig_ip + (regs->ip - copy_ip));
895 p->ainsn.boostable = 1;
897 p->ainsn.boostable = -1;
901 regs->ip += orig_ip - copy_ip;
906 NOKPROBE_SYMBOL(resume_execution);
909 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
910 * remain disabled throughout this function.
912 int kprobe_debug_handler(struct pt_regs *regs)
914 struct kprobe *cur = kprobe_running();
915 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
920 resume_execution(cur, regs, kcb);
921 regs->flags |= kcb->kprobe_saved_flags;
923 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
924 kcb->kprobe_status = KPROBE_HIT_SSDONE;
925 cur->post_handler(cur, regs, 0);
928 /* Restore back the original saved kprobes variables and continue. */
929 if (kcb->kprobe_status == KPROBE_REENTER) {
930 restore_previous_kprobe(kcb);
933 reset_current_kprobe();
935 preempt_enable_no_resched();
938 * if somebody else is singlestepping across a probe point, flags
939 * will have TF set, in which case, continue the remaining processing
940 * of do_debug, as if this is not a probe hit.
942 if (regs->flags & X86_EFLAGS_TF)
947 NOKPROBE_SYMBOL(kprobe_debug_handler);
949 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
951 struct kprobe *cur = kprobe_running();
952 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
954 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
955 /* This must happen on single-stepping */
956 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
957 kcb->kprobe_status != KPROBE_REENTER);
959 * We are here because the instruction being single
960 * stepped caused a page fault. We reset the current
961 * kprobe and the ip points back to the probe address
962 * and allow the page fault handler to continue as a
965 regs->ip = (unsigned long)cur->addr;
967 * Trap flag (TF) has been set here because this fault
968 * happened where the single stepping will be done.
969 * So clear it by resetting the current kprobe:
971 regs->flags &= ~X86_EFLAGS_TF;
974 * If the TF flag was set before the kprobe hit,
977 regs->flags |= kcb->kprobe_old_flags;
979 if (kcb->kprobe_status == KPROBE_REENTER)
980 restore_previous_kprobe(kcb);
982 reset_current_kprobe();
983 preempt_enable_no_resched();
984 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
985 kcb->kprobe_status == KPROBE_HIT_SSDONE) {
987 * We increment the nmissed count for accounting,
988 * we can also use npre/npostfault count for accounting
989 * these specific fault cases.
991 kprobes_inc_nmissed_count(cur);
994 * We come here because instructions in the pre/post
995 * handler caused the page_fault, this could happen
996 * if handler tries to access user space by
997 * copy_from_user(), get_user() etc. Let the
998 * user-specified handler try to fix it first.
1000 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
1004 * In case the user-specified fault handler returned
1005 * zero, try to fix up.
1007 if (fixup_exception(regs, trapnr))
1011 * fixup routine could not handle it,
1012 * Let do_page_fault() fix it.
1018 NOKPROBE_SYMBOL(kprobe_fault_handler);
1021 * Wrapper routine for handling exceptions.
1023 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
1026 struct die_args *args = data;
1027 int ret = NOTIFY_DONE;
1029 if (args->regs && user_mode(args->regs))
1032 if (val == DIE_GPF) {
1034 * To be potentially processing a kprobe fault and to
1035 * trust the result from kprobe_running(), we have
1036 * be non-preemptible.
1038 if (!preemptible() && kprobe_running() &&
1039 kprobe_fault_handler(args->regs, args->trapnr))
1044 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1046 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
1048 struct jprobe *jp = container_of(p, struct jprobe, kp);
1050 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1052 kcb->jprobe_saved_regs = *regs;
1053 kcb->jprobe_saved_sp = stack_addr(regs);
1054 addr = (unsigned long)(kcb->jprobe_saved_sp);
1057 * As Linus pointed out, gcc assumes that the callee
1058 * owns the argument space and could overwrite it, e.g.
1059 * tailcall optimization. So, to be absolutely safe
1060 * we also save and restore enough stack bytes to cover
1061 * the argument area.
1062 * Use __memcpy() to avoid KASAN stack out-of-bounds reports as we copy
1063 * raw stack chunk with redzones:
1065 __memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr, MIN_STACK_SIZE(addr));
1066 regs->flags &= ~X86_EFLAGS_IF;
1067 trace_hardirqs_off();
1068 regs->ip = (unsigned long)(jp->entry);
1071 * jprobes use jprobe_return() which skips the normal return
1072 * path of the function, and this messes up the accounting of the
1073 * function graph tracer to get messed up.
1075 * Pause function graph tracing while performing the jprobe function.
1077 pause_graph_tracing();
1080 NOKPROBE_SYMBOL(setjmp_pre_handler);
1082 void jprobe_return(void)
1084 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1086 /* Unpoison stack redzones in the frames we are going to jump over. */
1087 kasan_unpoison_stack_above_sp_to(kcb->jprobe_saved_sp);
1090 #ifdef CONFIG_X86_64
1091 " xchg %%rbx,%%rsp \n"
1093 " xchgl %%ebx,%%esp \n"
1096 " .globl jprobe_return_end\n"
1097 " jprobe_return_end: \n"
1099 (kcb->jprobe_saved_sp):"memory");
1101 NOKPROBE_SYMBOL(jprobe_return);
1102 NOKPROBE_SYMBOL(jprobe_return_end);
1104 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1106 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1107 u8 *addr = (u8 *) (regs->ip - 1);
1108 struct jprobe *jp = container_of(p, struct jprobe, kp);
1109 void *saved_sp = kcb->jprobe_saved_sp;
1111 if ((addr > (u8 *) jprobe_return) &&
1112 (addr < (u8 *) jprobe_return_end)) {
1113 if (stack_addr(regs) != saved_sp) {
1114 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1116 "current sp %p does not match saved sp %p\n",
1117 stack_addr(regs), saved_sp);
1118 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1119 show_regs(saved_regs);
1120 printk(KERN_ERR "Current registers\n");
1124 /* It's OK to start function graph tracing again */
1125 unpause_graph_tracing();
1126 *regs = kcb->jprobe_saved_regs;
1127 __memcpy(saved_sp, kcb->jprobes_stack, MIN_STACK_SIZE(saved_sp));
1128 preempt_enable_no_resched();
1133 NOKPROBE_SYMBOL(longjmp_break_handler);
1135 bool arch_within_kprobe_blacklist(unsigned long addr)
1137 return (addr >= (unsigned long)__kprobes_text_start &&
1138 addr < (unsigned long)__kprobes_text_end) ||
1139 (addr >= (unsigned long)__entry_text_start &&
1140 addr < (unsigned long)__entry_text_end);
1143 int __init arch_init_kprobes(void)
1148 int arch_trampoline_kprobe(struct kprobe *p)