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/module.h>
49 #include <linux/kdebug.h>
50 #include <linux/kallsyms.h>
51 #include <linux/ftrace.h>
52 #include <linux/frame.h>
54 #include <asm/cacheflush.h>
56 #include <asm/pgtable.h>
57 #include <asm/uaccess.h>
58 #include <asm/alternative.h>
60 #include <asm/debugreg.h>
64 void jprobe_return_end(void);
66 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
67 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
69 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
71 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
72 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
73 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
74 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
75 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
78 * Undefined/reserved opcodes, conditional jump, Opcode Extension
79 * Groups, and some special opcodes can not boost.
80 * This is non-const and volatile to keep gcc from statically
81 * optimizing it out, as variable_test_bit makes gcc think only
82 * *(unsigned long*) is used.
84 static volatile u32 twobyte_is_boostable[256 / 32] = {
85 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
86 /* ---------------------------------------------- */
87 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
88 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
89 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
90 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
91 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
92 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
93 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
94 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
95 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
96 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
97 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
98 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
99 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
100 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
101 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
102 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
103 /* ----------------------------------------------- */
104 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
108 struct kretprobe_blackpoint kretprobe_blacklist[] = {
109 {"__switch_to", }, /* This function switches only current task, but
110 doesn't switch kernel stack.*/
111 {NULL, NULL} /* Terminator */
114 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
116 static nokprobe_inline void
117 __synthesize_relative_insn(void *from, void *to, u8 op)
119 struct __arch_relative_insn {
124 insn = (struct __arch_relative_insn *)from;
125 insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
129 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
130 void synthesize_reljump(void *from, void *to)
132 __synthesize_relative_insn(from, to, RELATIVEJUMP_OPCODE);
134 NOKPROBE_SYMBOL(synthesize_reljump);
136 /* Insert a call instruction at address 'from', which calls address 'to'.*/
137 void synthesize_relcall(void *from, void *to)
139 __synthesize_relative_insn(from, to, RELATIVECALL_OPCODE);
141 NOKPROBE_SYMBOL(synthesize_relcall);
144 * Skip the prefixes of the instruction.
146 static kprobe_opcode_t *skip_prefixes(kprobe_opcode_t *insn)
150 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
151 while (inat_is_legacy_prefix(attr)) {
153 attr = inat_get_opcode_attribute((insn_byte_t)*insn);
156 if (inat_is_rex_prefix(attr))
161 NOKPROBE_SYMBOL(skip_prefixes);
164 * Returns non-zero if opcode is boostable.
165 * RIP relative instructions are adjusted at copying time in 64 bits mode
167 int can_boost(kprobe_opcode_t *opcodes)
169 kprobe_opcode_t opcode;
170 kprobe_opcode_t *orig_opcodes = opcodes;
172 if (search_exception_tables((unsigned long)opcodes))
173 return 0; /* Page fault may occur on this address. */
176 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
178 opcode = *(opcodes++);
180 /* 2nd-byte opcode */
181 if (opcode == 0x0f) {
182 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
184 return test_bit(*opcodes,
185 (unsigned long *)twobyte_is_boostable);
188 switch (opcode & 0xf0) {
191 goto retry; /* REX prefix is boostable */
194 if (0x63 < opcode && opcode < 0x67)
195 goto retry; /* prefixes */
196 /* can't boost Address-size override and bound */
197 return (opcode != 0x62 && opcode != 0x67);
199 return 0; /* can't boost conditional jump */
201 /* can't boost software-interruptions */
202 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
204 /* can boost AA* and XLAT */
205 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
207 /* can boost in/out and absolute jmps */
208 return ((opcode & 0x04) || opcode == 0xea);
210 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
211 goto retry; /* lock/rep(ne) prefix */
212 /* clear and set flags are boostable */
213 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
215 /* segment override prefixes are boostable */
216 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
217 goto retry; /* prefixes */
218 /* CS override prefix and call are not boostable */
219 return (opcode != 0x2e && opcode != 0x9a);
224 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
229 kp = get_kprobe((void *)addr);
230 faddr = ftrace_location(addr);
232 * Addresses inside the ftrace location are refused by
233 * arch_check_ftrace_location(). Something went terribly wrong
234 * if such an address is checked here.
236 if (WARN_ON(faddr && faddr != addr))
239 * Use the current code if it is not modified by Kprobe
240 * and it cannot be modified by ftrace.
246 * Basically, kp->ainsn.insn has an original instruction.
247 * However, RIP-relative instruction can not do single-stepping
248 * at different place, __copy_instruction() tweaks the displacement of
249 * that instruction. In that case, we can't recover the instruction
250 * from the kp->ainsn.insn.
252 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
253 * of the first byte of the probed instruction, which is overwritten
254 * by int3. And the instruction at kp->addr is not modified by kprobes
255 * except for the first byte, we can recover the original instruction
256 * from it and kp->opcode.
258 * In case of Kprobes using ftrace, we do not have a copy of
259 * the original instruction. In fact, the ftrace location might
260 * be modified at anytime and even could be in an inconsistent state.
261 * Fortunately, we know that the original code is the ideal 5-byte
264 memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
266 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
269 return (unsigned long)buf;
273 * Recover the probed instruction at addr for further analysis.
274 * Caller must lock kprobes by kprobe_mutex, or disable preemption
275 * for preventing to release referencing kprobes.
276 * Returns zero if the instruction can not get recovered.
278 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
280 unsigned long __addr;
282 __addr = __recover_optprobed_insn(buf, addr);
286 return __recover_probed_insn(buf, addr);
289 /* Check if paddr is at an instruction boundary */
290 static int can_probe(unsigned long paddr)
292 unsigned long addr, __addr, offset = 0;
294 kprobe_opcode_t buf[MAX_INSN_SIZE];
296 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
299 /* Decode instructions */
300 addr = paddr - offset;
301 while (addr < paddr) {
303 * Check if the instruction has been modified by another
304 * kprobe, in which case we replace the breakpoint by the
305 * original instruction in our buffer.
306 * Also, jump optimization will change the breakpoint to
307 * relative-jump. Since the relative-jump itself is
308 * normally used, we just go through if there is no kprobe.
310 __addr = recover_probed_instruction(buf, addr);
313 kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
314 insn_get_length(&insn);
317 * Another debugging subsystem might insert this breakpoint.
318 * In that case, we can't recover it.
320 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
325 return (addr == paddr);
329 * Returns non-zero if opcode modifies the interrupt flag.
331 static int is_IF_modifier(kprobe_opcode_t *insn)
334 insn = skip_prefixes(insn);
339 case 0xcf: /* iret/iretd */
340 case 0x9d: /* popf/popfd */
348 * Copy an instruction and adjust the displacement if the instruction
349 * uses the %rip-relative addressing mode.
350 * If it does, Return the address of the 32-bit displacement word.
351 * If not, return null.
352 * Only applicable to 64-bit x86.
354 int __copy_instruction(u8 *dest, u8 *src)
357 kprobe_opcode_t buf[MAX_INSN_SIZE];
359 unsigned long recovered_insn =
360 recover_probed_instruction(buf, (unsigned long)src);
364 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
365 insn_get_length(&insn);
366 length = insn.length;
368 /* Another subsystem puts a breakpoint, failed to recover */
369 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
371 memcpy(dest, insn.kaddr, length);
374 if (insn_rip_relative(&insn)) {
377 kernel_insn_init(&insn, dest, length);
378 insn_get_displacement(&insn);
380 * The copied instruction uses the %rip-relative addressing
381 * mode. Adjust the displacement for the difference between
382 * the original location of this instruction and the location
383 * of the copy that will actually be run. The tricky bit here
384 * is making sure that the sign extension happens correctly in
385 * this calculation, since we need a signed 32-bit result to
386 * be sign-extended to 64 bits when it's added to the %rip
387 * value and yield the same 64-bit result that the sign-
388 * extension of the original signed 32-bit displacement would
391 newdisp = (u8 *) src + (s64) insn.displacement.value - (u8 *) dest;
392 if ((s64) (s32) newdisp != newdisp) {
393 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
394 pr_err("\tSrc: %p, Dest: %p, old disp: %x\n", src, dest, insn.displacement.value);
397 disp = (u8 *) dest + insn_offset_displacement(&insn);
398 *(s32 *) disp = (s32) newdisp;
404 static int arch_copy_kprobe(struct kprobe *p)
408 /* Copy an instruction with recovering if other optprobe modifies it.*/
409 ret = __copy_instruction(p->ainsn.insn, p->addr);
414 * __copy_instruction can modify the displacement of the instruction,
415 * but it doesn't affect boostable check.
417 if (can_boost(p->ainsn.insn))
418 p->ainsn.boostable = 0;
420 p->ainsn.boostable = -1;
422 /* Check whether the instruction modifies Interrupt Flag or not */
423 p->ainsn.if_modifier = is_IF_modifier(p->ainsn.insn);
425 /* Also, displacement change doesn't affect the first byte */
426 p->opcode = p->ainsn.insn[0];
431 int arch_prepare_kprobe(struct kprobe *p)
433 if (alternatives_text_reserved(p->addr, p->addr))
436 if (!can_probe((unsigned long)p->addr))
438 /* insn: must be on special executable page on x86. */
439 p->ainsn.insn = get_insn_slot();
443 return arch_copy_kprobe(p);
446 void arch_arm_kprobe(struct kprobe *p)
448 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
451 void arch_disarm_kprobe(struct kprobe *p)
453 text_poke(p->addr, &p->opcode, 1);
456 void arch_remove_kprobe(struct kprobe *p)
459 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
460 p->ainsn.insn = NULL;
464 static nokprobe_inline void
465 save_previous_kprobe(struct kprobe_ctlblk *kcb)
467 kcb->prev_kprobe.kp = kprobe_running();
468 kcb->prev_kprobe.status = kcb->kprobe_status;
469 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
470 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
473 static nokprobe_inline void
474 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
476 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
477 kcb->kprobe_status = kcb->prev_kprobe.status;
478 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
479 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
482 static nokprobe_inline void
483 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
484 struct kprobe_ctlblk *kcb)
486 __this_cpu_write(current_kprobe, p);
487 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
488 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
489 if (p->ainsn.if_modifier)
490 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
493 static nokprobe_inline void clear_btf(void)
495 if (test_thread_flag(TIF_BLOCKSTEP)) {
496 unsigned long debugctl = get_debugctlmsr();
498 debugctl &= ~DEBUGCTLMSR_BTF;
499 update_debugctlmsr(debugctl);
503 static nokprobe_inline void restore_btf(void)
505 if (test_thread_flag(TIF_BLOCKSTEP)) {
506 unsigned long debugctl = get_debugctlmsr();
508 debugctl |= DEBUGCTLMSR_BTF;
509 update_debugctlmsr(debugctl);
513 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
515 unsigned long *sara = stack_addr(regs);
517 ri->ret_addr = (kprobe_opcode_t *) *sara;
519 /* Replace the return addr with trampoline addr */
520 *sara = (unsigned long) &kretprobe_trampoline;
522 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
524 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
525 struct kprobe_ctlblk *kcb, int reenter)
527 if (setup_detour_execution(p, regs, reenter))
530 #if !defined(CONFIG_PREEMPT)
531 if (p->ainsn.boostable == 1 && !p->post_handler) {
532 /* Boost up -- we can execute copied instructions directly */
534 reset_current_kprobe();
536 * Reentering boosted probe doesn't reset current_kprobe,
537 * nor set current_kprobe, because it doesn't use single
540 regs->ip = (unsigned long)p->ainsn.insn;
541 preempt_enable_no_resched();
546 save_previous_kprobe(kcb);
547 set_current_kprobe(p, regs, kcb);
548 kcb->kprobe_status = KPROBE_REENTER;
550 kcb->kprobe_status = KPROBE_HIT_SS;
551 /* Prepare real single stepping */
553 regs->flags |= X86_EFLAGS_TF;
554 regs->flags &= ~X86_EFLAGS_IF;
555 /* single step inline if the instruction is an int3 */
556 if (p->opcode == BREAKPOINT_INSTRUCTION)
557 regs->ip = (unsigned long)p->addr;
559 regs->ip = (unsigned long)p->ainsn.insn;
561 NOKPROBE_SYMBOL(setup_singlestep);
564 * We have reentered the kprobe_handler(), since another probe was hit while
565 * within the handler. We save the original kprobes variables and just single
566 * step on the instruction of the new probe without calling any user handlers.
568 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
569 struct kprobe_ctlblk *kcb)
571 switch (kcb->kprobe_status) {
572 case KPROBE_HIT_SSDONE:
573 case KPROBE_HIT_ACTIVE:
575 kprobes_inc_nmissed_count(p);
576 setup_singlestep(p, regs, kcb, 1);
579 /* A probe has been hit in the codepath leading up to, or just
580 * after, single-stepping of a probed instruction. This entire
581 * codepath should strictly reside in .kprobes.text section.
582 * Raise a BUG or we'll continue in an endless reentering loop
583 * and eventually a stack overflow.
585 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
590 /* impossible cases */
597 NOKPROBE_SYMBOL(reenter_kprobe);
600 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
601 * remain disabled throughout this function.
603 int kprobe_int3_handler(struct pt_regs *regs)
605 kprobe_opcode_t *addr;
607 struct kprobe_ctlblk *kcb;
612 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
614 * We don't want to be preempted for the entire
615 * duration of kprobe processing. We conditionally
616 * re-enable preemption at the end of this function,
617 * and also in reenter_kprobe() and setup_singlestep().
621 kcb = get_kprobe_ctlblk();
622 p = get_kprobe(addr);
625 if (kprobe_running()) {
626 if (reenter_kprobe(p, regs, kcb))
629 set_current_kprobe(p, regs, kcb);
630 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
633 * If we have no pre-handler or it returned 0, we
634 * continue with normal processing. If we have a
635 * pre-handler and it returned non-zero, it prepped
636 * for calling the break_handler below on re-entry
637 * for jprobe processing, so get out doing nothing
640 if (!p->pre_handler || !p->pre_handler(p, regs))
641 setup_singlestep(p, regs, kcb, 0);
644 } else if (*addr != BREAKPOINT_INSTRUCTION) {
646 * The breakpoint instruction was removed right
647 * after we hit it. Another cpu has removed
648 * either a probepoint or a debugger breakpoint
649 * at this address. In either case, no further
650 * handling of this interrupt is appropriate.
651 * Back up over the (now missing) int3 and run
652 * the original instruction.
654 regs->ip = (unsigned long)addr;
655 preempt_enable_no_resched();
657 } else if (kprobe_running()) {
658 p = __this_cpu_read(current_kprobe);
659 if (p->break_handler && p->break_handler(p, regs)) {
660 if (!skip_singlestep(p, regs, kcb))
661 setup_singlestep(p, regs, kcb, 0);
664 } /* else: not a kprobe fault; let the kernel handle it */
666 preempt_enable_no_resched();
669 NOKPROBE_SYMBOL(kprobe_int3_handler);
672 * When a retprobed function returns, this code saves registers and
673 * calls trampoline_handler() runs, which calls the kretprobe's handler.
676 ".global kretprobe_trampoline\n"
677 ".type kretprobe_trampoline, @function\n"
678 "kretprobe_trampoline:\n"
680 /* We don't bother saving the ss register */
685 " call trampoline_handler\n"
686 /* Replace saved sp with true return address. */
687 " movq %rax, 152(%rsp)\n"
694 " call trampoline_handler\n"
695 /* Move flags to cs */
696 " movl 56(%esp), %edx\n"
697 " movl %edx, 52(%esp)\n"
698 /* Replace saved flags with true return address. */
699 " movl %eax, 56(%esp)\n"
704 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
706 NOKPROBE_SYMBOL(kretprobe_trampoline);
707 STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
710 * Called from kretprobe_trampoline
712 __visible __used void *trampoline_handler(struct pt_regs *regs)
714 struct kretprobe_instance *ri = NULL;
715 struct hlist_head *head, empty_rp;
716 struct hlist_node *tmp;
717 unsigned long flags, orig_ret_address = 0;
718 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
719 kprobe_opcode_t *correct_ret_addr = NULL;
721 INIT_HLIST_HEAD(&empty_rp);
722 kretprobe_hash_lock(current, &head, &flags);
723 /* fixup registers */
725 regs->cs = __KERNEL_CS;
727 regs->cs = __KERNEL_CS | get_kernel_rpl();
730 regs->ip = trampoline_address;
731 regs->orig_ax = ~0UL;
734 * It is possible to have multiple instances associated with a given
735 * task either because multiple functions in the call path have
736 * return probes installed on them, and/or more than one
737 * return probe was registered for a target function.
739 * We can handle this because:
740 * - instances are always pushed into the head of the list
741 * - when multiple return probes are registered for the same
742 * function, the (chronologically) first instance's ret_addr
743 * will be the real return address, and all the rest will
744 * point to kretprobe_trampoline.
746 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
747 if (ri->task != current)
748 /* another task is sharing our hash bucket */
751 orig_ret_address = (unsigned long)ri->ret_addr;
753 if (orig_ret_address != trampoline_address)
755 * This is the real return address. Any other
756 * instances associated with this task are for
757 * other calls deeper on the call stack
762 kretprobe_assert(ri, orig_ret_address, trampoline_address);
764 correct_ret_addr = ri->ret_addr;
765 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
766 if (ri->task != current)
767 /* another task is sharing our hash bucket */
770 orig_ret_address = (unsigned long)ri->ret_addr;
771 if (ri->rp && ri->rp->handler) {
772 __this_cpu_write(current_kprobe, &ri->rp->kp);
773 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
774 ri->ret_addr = correct_ret_addr;
775 ri->rp->handler(ri, regs);
776 __this_cpu_write(current_kprobe, NULL);
779 recycle_rp_inst(ri, &empty_rp);
781 if (orig_ret_address != trampoline_address)
783 * This is the real return address. Any other
784 * instances associated with this task are for
785 * other calls deeper on the call stack
790 kretprobe_hash_unlock(current, &flags);
792 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
793 hlist_del(&ri->hlist);
796 return (void *)orig_ret_address;
798 NOKPROBE_SYMBOL(trampoline_handler);
801 * Called after single-stepping. p->addr is the address of the
802 * instruction whose first byte has been replaced by the "int 3"
803 * instruction. To avoid the SMP problems that can occur when we
804 * temporarily put back the original opcode to single-step, we
805 * single-stepped a copy of the instruction. The address of this
806 * copy is p->ainsn.insn.
808 * This function prepares to return from the post-single-step
809 * interrupt. We have to fix up the stack as follows:
811 * 0) Except in the case of absolute or indirect jump or call instructions,
812 * the new ip is relative to the copied instruction. We need to make
813 * it relative to the original instruction.
815 * 1) If the single-stepped instruction was pushfl, then the TF and IF
816 * flags are set in the just-pushed flags, and may need to be cleared.
818 * 2) If the single-stepped instruction was a call, the return address
819 * that is atop the stack is the address following the copied instruction.
820 * We need to make it the address following the original instruction.
822 * If this is the first time we've single-stepped the instruction at
823 * this probepoint, and the instruction is boostable, boost it: add a
824 * jump instruction after the copied instruction, that jumps to the next
825 * instruction after the probepoint.
827 static void resume_execution(struct kprobe *p, struct pt_regs *regs,
828 struct kprobe_ctlblk *kcb)
830 unsigned long *tos = stack_addr(regs);
831 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
832 unsigned long orig_ip = (unsigned long)p->addr;
833 kprobe_opcode_t *insn = p->ainsn.insn;
836 insn = skip_prefixes(insn);
838 regs->flags &= ~X86_EFLAGS_TF;
840 case 0x9c: /* pushfl */
841 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
842 *tos |= kcb->kprobe_old_flags;
844 case 0xc2: /* iret/ret/lret */
849 case 0xea: /* jmp absolute -- ip is correct */
850 /* ip is already adjusted, no more changes required */
851 p->ainsn.boostable = 1;
853 case 0xe8: /* call relative - Fix return addr */
854 *tos = orig_ip + (*tos - copy_ip);
857 case 0x9a: /* call absolute -- same as call absolute, indirect */
858 *tos = orig_ip + (*tos - copy_ip);
862 if ((insn[1] & 0x30) == 0x10) {
864 * call absolute, indirect
865 * Fix return addr; ip is correct.
866 * But this is not boostable
868 *tos = orig_ip + (*tos - copy_ip);
870 } else if (((insn[1] & 0x31) == 0x20) ||
871 ((insn[1] & 0x31) == 0x21)) {
873 * jmp near and far, absolute indirect
874 * ip is correct. And this is boostable
876 p->ainsn.boostable = 1;
883 if (p->ainsn.boostable == 0) {
884 if ((regs->ip > copy_ip) &&
885 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
887 * These instructions can be executed directly if it
888 * jumps back to correct address.
890 synthesize_reljump((void *)regs->ip,
891 (void *)orig_ip + (regs->ip - copy_ip));
892 p->ainsn.boostable = 1;
894 p->ainsn.boostable = -1;
898 regs->ip += orig_ip - copy_ip;
903 NOKPROBE_SYMBOL(resume_execution);
906 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
907 * remain disabled throughout this function.
909 int kprobe_debug_handler(struct pt_regs *regs)
911 struct kprobe *cur = kprobe_running();
912 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
917 resume_execution(cur, regs, kcb);
918 regs->flags |= kcb->kprobe_saved_flags;
920 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
921 kcb->kprobe_status = KPROBE_HIT_SSDONE;
922 cur->post_handler(cur, regs, 0);
925 /* Restore back the original saved kprobes variables and continue. */
926 if (kcb->kprobe_status == KPROBE_REENTER) {
927 restore_previous_kprobe(kcb);
930 reset_current_kprobe();
932 preempt_enable_no_resched();
935 * if somebody else is singlestepping across a probe point, flags
936 * will have TF set, in which case, continue the remaining processing
937 * of do_debug, as if this is not a probe hit.
939 if (regs->flags & X86_EFLAGS_TF)
944 NOKPROBE_SYMBOL(kprobe_debug_handler);
946 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
948 struct kprobe *cur = kprobe_running();
949 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
951 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
952 /* This must happen on single-stepping */
953 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
954 kcb->kprobe_status != KPROBE_REENTER);
956 * We are here because the instruction being single
957 * stepped caused a page fault. We reset the current
958 * kprobe and the ip points back to the probe address
959 * and allow the page fault handler to continue as a
962 regs->ip = (unsigned long)cur->addr;
963 regs->flags |= kcb->kprobe_old_flags;
964 if (kcb->kprobe_status == KPROBE_REENTER)
965 restore_previous_kprobe(kcb);
967 reset_current_kprobe();
968 preempt_enable_no_resched();
969 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
970 kcb->kprobe_status == KPROBE_HIT_SSDONE) {
972 * We increment the nmissed count for accounting,
973 * we can also use npre/npostfault count for accounting
974 * these specific fault cases.
976 kprobes_inc_nmissed_count(cur);
979 * We come here because instructions in the pre/post
980 * handler caused the page_fault, this could happen
981 * if handler tries to access user space by
982 * copy_from_user(), get_user() etc. Let the
983 * user-specified handler try to fix it first.
985 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
989 * In case the user-specified fault handler returned
990 * zero, try to fix up.
992 if (fixup_exception(regs, trapnr))
996 * fixup routine could not handle it,
997 * Let do_page_fault() fix it.
1003 NOKPROBE_SYMBOL(kprobe_fault_handler);
1006 * Wrapper routine for handling exceptions.
1008 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
1011 struct die_args *args = data;
1012 int ret = NOTIFY_DONE;
1014 if (args->regs && user_mode(args->regs))
1017 if (val == DIE_GPF) {
1019 * To be potentially processing a kprobe fault and to
1020 * trust the result from kprobe_running(), we have
1021 * be non-preemptible.
1023 if (!preemptible() && kprobe_running() &&
1024 kprobe_fault_handler(args->regs, args->trapnr))
1029 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1031 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
1033 struct jprobe *jp = container_of(p, struct jprobe, kp);
1035 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1037 kcb->jprobe_saved_regs = *regs;
1038 kcb->jprobe_saved_sp = stack_addr(regs);
1039 addr = (unsigned long)(kcb->jprobe_saved_sp);
1042 * As Linus pointed out, gcc assumes that the callee
1043 * owns the argument space and could overwrite it, e.g.
1044 * tailcall optimization. So, to be absolutely safe
1045 * we also save and restore enough stack bytes to cover
1046 * the argument area.
1048 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1049 MIN_STACK_SIZE(addr));
1050 regs->flags &= ~X86_EFLAGS_IF;
1051 trace_hardirqs_off();
1052 regs->ip = (unsigned long)(jp->entry);
1055 * jprobes use jprobe_return() which skips the normal return
1056 * path of the function, and this messes up the accounting of the
1057 * function graph tracer to get messed up.
1059 * Pause function graph tracing while performing the jprobe function.
1061 pause_graph_tracing();
1064 NOKPROBE_SYMBOL(setjmp_pre_handler);
1066 void jprobe_return(void)
1068 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1071 #ifdef CONFIG_X86_64
1072 " xchg %%rbx,%%rsp \n"
1074 " xchgl %%ebx,%%esp \n"
1077 " .globl jprobe_return_end\n"
1078 " jprobe_return_end: \n"
1080 (kcb->jprobe_saved_sp):"memory");
1082 NOKPROBE_SYMBOL(jprobe_return);
1083 NOKPROBE_SYMBOL(jprobe_return_end);
1085 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1087 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1088 u8 *addr = (u8 *) (regs->ip - 1);
1089 struct jprobe *jp = container_of(p, struct jprobe, kp);
1090 void *saved_sp = kcb->jprobe_saved_sp;
1092 if ((addr > (u8 *) jprobe_return) &&
1093 (addr < (u8 *) jprobe_return_end)) {
1094 if (stack_addr(regs) != saved_sp) {
1095 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1097 "current sp %p does not match saved sp %p\n",
1098 stack_addr(regs), saved_sp);
1099 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1100 show_regs(saved_regs);
1101 printk(KERN_ERR "Current registers\n");
1105 /* It's OK to start function graph tracing again */
1106 unpause_graph_tracing();
1107 *regs = kcb->jprobe_saved_regs;
1108 memcpy(saved_sp, kcb->jprobes_stack, MIN_STACK_SIZE(saved_sp));
1109 preempt_enable_no_resched();
1114 NOKPROBE_SYMBOL(longjmp_break_handler);
1116 bool arch_within_kprobe_blacklist(unsigned long addr)
1118 return (addr >= (unsigned long)__kprobes_text_start &&
1119 addr < (unsigned long)__kprobes_text_end) ||
1120 (addr >= (unsigned long)__entry_text_start &&
1121 addr < (unsigned long)__entry_text_end);
1124 int __init arch_init_kprobes(void)
1129 int arch_trampoline_kprobe(struct kprobe *p)