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 INSN is boostable.
168 * RIP relative instructions are adjusted at copying time in 64 bits mode
170 int can_boost(struct insn *insn, void *addr)
172 kprobe_opcode_t opcode;
174 if (search_exception_tables((unsigned long)addr))
175 return 0; /* Page fault may occur on this address. */
177 /* 2nd-byte opcode */
178 if (insn->opcode.nbytes == 2)
179 return test_bit(insn->opcode.bytes[1],
180 (unsigned long *)twobyte_is_boostable);
182 if (insn->opcode.nbytes != 1)
185 /* Can't boost Address-size override prefix */
186 if (unlikely(inat_is_address_size_prefix(insn->attr)))
189 opcode = insn->opcode.bytes[0];
191 switch (opcode & 0xf0) {
193 /* can't boost "bound" */
194 return (opcode != 0x62);
196 return 0; /* can't boost conditional jump */
198 return opcode != 0x9a; /* can't boost call far */
200 /* can't boost software-interruptions */
201 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
203 /* can boost AA* and XLAT */
204 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
206 /* can boost in/out and absolute jmps */
207 return ((opcode & 0x04) || opcode == 0xea);
209 /* clear and set flags are boostable */
210 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
212 /* CS override prefix and call are not boostable */
213 return (opcode != 0x2e && opcode != 0x9a);
218 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr)
223 kp = get_kprobe((void *)addr);
224 faddr = ftrace_location(addr);
226 * Addresses inside the ftrace location are refused by
227 * arch_check_ftrace_location(). Something went terribly wrong
228 * if such an address is checked here.
230 if (WARN_ON(faddr && faddr != addr))
233 * Use the current code if it is not modified by Kprobe
234 * and it cannot be modified by ftrace.
240 * Basically, kp->ainsn.insn has an original instruction.
241 * However, RIP-relative instruction can not do single-stepping
242 * at different place, __copy_instruction() tweaks the displacement of
243 * that instruction. In that case, we can't recover the instruction
244 * from the kp->ainsn.insn.
246 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
247 * of the first byte of the probed instruction, which is overwritten
248 * by int3. And the instruction at kp->addr is not modified by kprobes
249 * except for the first byte, we can recover the original instruction
250 * from it and kp->opcode.
252 * In case of Kprobes using ftrace, we do not have a copy of
253 * the original instruction. In fact, the ftrace location might
254 * be modified at anytime and even could be in an inconsistent state.
255 * Fortunately, we know that the original code is the ideal 5-byte
258 if (probe_kernel_read(buf, (void *)addr,
259 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
263 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5);
266 return (unsigned long)buf;
270 * Recover the probed instruction at addr for further analysis.
271 * Caller must lock kprobes by kprobe_mutex, or disable preemption
272 * for preventing to release referencing kprobes.
273 * Returns zero if the instruction can not get recovered (or access failed).
275 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
277 unsigned long __addr;
279 __addr = __recover_optprobed_insn(buf, addr);
283 return __recover_probed_insn(buf, addr);
286 /* Check if paddr is at an instruction boundary */
287 static int can_probe(unsigned long paddr)
289 unsigned long addr, __addr, offset = 0;
291 kprobe_opcode_t buf[MAX_INSN_SIZE];
293 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
296 /* Decode instructions */
297 addr = paddr - offset;
298 while (addr < paddr) {
300 * Check if the instruction has been modified by another
301 * kprobe, in which case we replace the breakpoint by the
302 * original instruction in our buffer.
303 * Also, jump optimization will change the breakpoint to
304 * relative-jump. Since the relative-jump itself is
305 * normally used, we just go through if there is no kprobe.
307 __addr = recover_probed_instruction(buf, addr);
310 kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE);
311 insn_get_length(&insn);
314 * Another debugging subsystem might insert this breakpoint.
315 * In that case, we can't recover it.
317 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
322 return (addr == paddr);
326 * Returns non-zero if opcode modifies the interrupt flag.
328 static int is_IF_modifier(kprobe_opcode_t *insn)
331 insn = skip_prefixes(insn);
336 case 0xcf: /* iret/iretd */
337 case 0x9d: /* popf/popfd */
345 * Copy an instruction with recovering modified instruction by kprobes
346 * and adjust the displacement if the instruction uses the %rip-relative
348 * This returns the length of copied instruction, or 0 if it has an error.
350 int __copy_instruction(u8 *dest, u8 *src, struct insn *insn)
352 kprobe_opcode_t buf[MAX_INSN_SIZE];
353 unsigned long recovered_insn =
354 recover_probed_instruction(buf, (unsigned long)src);
356 if (!recovered_insn || !insn)
359 /* This can access kernel text if given address is not recovered */
360 if (probe_kernel_read(dest, (void *)recovered_insn, MAX_INSN_SIZE))
363 kernel_insn_init(insn, dest, MAX_INSN_SIZE);
364 insn_get_length(insn);
366 /* Another subsystem puts a breakpoint, failed to recover */
367 if (insn->opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
371 /* Only x86_64 has RIP relative instructions */
372 if (insn_rip_relative(insn)) {
376 * The copied instruction uses the %rip-relative addressing
377 * mode. Adjust the displacement for the difference between
378 * the original location of this instruction and the location
379 * of the copy that will actually be run. The tricky bit here
380 * is making sure that the sign extension happens correctly in
381 * this calculation, since we need a signed 32-bit result to
382 * be sign-extended to 64 bits when it's added to the %rip
383 * value and yield the same 64-bit result that the sign-
384 * extension of the original signed 32-bit displacement would
387 newdisp = (u8 *) src + (s64) insn->displacement.value
389 if ((s64) (s32) newdisp != newdisp) {
390 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp);
391 pr_err("\tSrc: %p, Dest: %p, old disp: %x\n",
392 src, dest, insn->displacement.value);
395 disp = (u8 *) dest + insn_offset_displacement(insn);
396 *(s32 *) disp = (s32) newdisp;
402 /* Prepare reljump right after instruction to boost */
403 static void prepare_boost(struct kprobe *p, struct insn *insn)
405 if (can_boost(insn, p->addr) &&
406 MAX_INSN_SIZE - insn->length >= RELATIVEJUMP_SIZE) {
408 * These instructions can be executed directly if it
409 * jumps back to correct address.
411 synthesize_reljump(p->ainsn.insn + insn->length,
412 p->addr + insn->length);
413 p->ainsn.boostable = true;
415 p->ainsn.boostable = false;
419 static int arch_copy_kprobe(struct kprobe *p)
424 set_memory_rw((unsigned long)p->ainsn.insn & PAGE_MASK, 1);
426 /* Copy an instruction with recovering if other optprobe modifies it.*/
427 len = __copy_instruction(p->ainsn.insn, p->addr, &insn);
432 * __copy_instruction can modify the displacement of the instruction,
433 * but it doesn't affect boostable check.
435 prepare_boost(p, &insn);
437 set_memory_ro((unsigned long)p->ainsn.insn & PAGE_MASK, 1);
439 /* Check whether the instruction modifies Interrupt Flag or not */
440 p->ainsn.if_modifier = is_IF_modifier(p->ainsn.insn);
442 /* Also, displacement change doesn't affect the first byte */
443 p->opcode = p->ainsn.insn[0];
448 int arch_prepare_kprobe(struct kprobe *p)
450 if (alternatives_text_reserved(p->addr, p->addr))
453 if (!can_probe((unsigned long)p->addr))
455 /* insn: must be on special executable page on x86. */
456 p->ainsn.insn = get_insn_slot();
460 return arch_copy_kprobe(p);
463 void arch_arm_kprobe(struct kprobe *p)
465 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
468 void arch_disarm_kprobe(struct kprobe *p)
470 text_poke(p->addr, &p->opcode, 1);
473 void arch_remove_kprobe(struct kprobe *p)
476 free_insn_slot(p->ainsn.insn, p->ainsn.boostable);
477 p->ainsn.insn = NULL;
481 static nokprobe_inline void
482 save_previous_kprobe(struct kprobe_ctlblk *kcb)
484 kcb->prev_kprobe.kp = kprobe_running();
485 kcb->prev_kprobe.status = kcb->kprobe_status;
486 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
487 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
490 static nokprobe_inline void
491 restore_previous_kprobe(struct kprobe_ctlblk *kcb)
493 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
494 kcb->kprobe_status = kcb->prev_kprobe.status;
495 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
496 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
499 static nokprobe_inline void
500 set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
501 struct kprobe_ctlblk *kcb)
503 __this_cpu_write(current_kprobe, p);
504 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
505 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
506 if (p->ainsn.if_modifier)
507 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
510 static nokprobe_inline void clear_btf(void)
512 if (test_thread_flag(TIF_BLOCKSTEP)) {
513 unsigned long debugctl = get_debugctlmsr();
515 debugctl &= ~DEBUGCTLMSR_BTF;
516 update_debugctlmsr(debugctl);
520 static nokprobe_inline void restore_btf(void)
522 if (test_thread_flag(TIF_BLOCKSTEP)) {
523 unsigned long debugctl = get_debugctlmsr();
525 debugctl |= DEBUGCTLMSR_BTF;
526 update_debugctlmsr(debugctl);
530 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
532 unsigned long *sara = stack_addr(regs);
534 ri->ret_addr = (kprobe_opcode_t *) *sara;
536 /* Replace the return addr with trampoline addr */
537 *sara = (unsigned long) &kretprobe_trampoline;
539 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
541 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs,
542 struct kprobe_ctlblk *kcb, int reenter)
544 if (setup_detour_execution(p, regs, reenter))
547 #if !defined(CONFIG_PREEMPT)
548 if (p->ainsn.boostable && !p->post_handler) {
549 /* Boost up -- we can execute copied instructions directly */
551 reset_current_kprobe();
553 * Reentering boosted probe doesn't reset current_kprobe,
554 * nor set current_kprobe, because it doesn't use single
557 regs->ip = (unsigned long)p->ainsn.insn;
558 preempt_enable_no_resched();
563 save_previous_kprobe(kcb);
564 set_current_kprobe(p, regs, kcb);
565 kcb->kprobe_status = KPROBE_REENTER;
567 kcb->kprobe_status = KPROBE_HIT_SS;
568 /* Prepare real single stepping */
570 regs->flags |= X86_EFLAGS_TF;
571 regs->flags &= ~X86_EFLAGS_IF;
572 /* single step inline if the instruction is an int3 */
573 if (p->opcode == BREAKPOINT_INSTRUCTION)
574 regs->ip = (unsigned long)p->addr;
576 regs->ip = (unsigned long)p->ainsn.insn;
578 NOKPROBE_SYMBOL(setup_singlestep);
581 * We have reentered the kprobe_handler(), since another probe was hit while
582 * within the handler. We save the original kprobes variables and just single
583 * step on the instruction of the new probe without calling any user handlers.
585 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
586 struct kprobe_ctlblk *kcb)
588 switch (kcb->kprobe_status) {
589 case KPROBE_HIT_SSDONE:
590 case KPROBE_HIT_ACTIVE:
592 kprobes_inc_nmissed_count(p);
593 setup_singlestep(p, regs, kcb, 1);
596 /* A probe has been hit in the codepath leading up to, or just
597 * after, single-stepping of a probed instruction. This entire
598 * codepath should strictly reside in .kprobes.text section.
599 * Raise a BUG or we'll continue in an endless reentering loop
600 * and eventually a stack overflow.
602 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
607 /* impossible cases */
614 NOKPROBE_SYMBOL(reenter_kprobe);
617 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
618 * remain disabled throughout this function.
620 int kprobe_int3_handler(struct pt_regs *regs)
622 kprobe_opcode_t *addr;
624 struct kprobe_ctlblk *kcb;
629 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
631 * We don't want to be preempted for the entire
632 * duration of kprobe processing. We conditionally
633 * re-enable preemption at the end of this function,
634 * and also in reenter_kprobe() and setup_singlestep().
638 kcb = get_kprobe_ctlblk();
639 p = get_kprobe(addr);
642 if (kprobe_running()) {
643 if (reenter_kprobe(p, regs, kcb))
646 set_current_kprobe(p, regs, kcb);
647 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
650 * If we have no pre-handler or it returned 0, we
651 * continue with normal processing. If we have a
652 * pre-handler and it returned non-zero, it prepped
653 * for calling the break_handler below on re-entry
654 * for jprobe processing, so get out doing nothing
657 if (!p->pre_handler || !p->pre_handler(p, regs))
658 setup_singlestep(p, regs, kcb, 0);
661 } else if (*addr != BREAKPOINT_INSTRUCTION) {
663 * The breakpoint instruction was removed right
664 * after we hit it. Another cpu has removed
665 * either a probepoint or a debugger breakpoint
666 * at this address. In either case, no further
667 * handling of this interrupt is appropriate.
668 * Back up over the (now missing) int3 and run
669 * the original instruction.
671 regs->ip = (unsigned long)addr;
672 preempt_enable_no_resched();
674 } else if (kprobe_running()) {
675 p = __this_cpu_read(current_kprobe);
676 if (p->break_handler && p->break_handler(p, regs)) {
677 if (!skip_singlestep(p, regs, kcb))
678 setup_singlestep(p, regs, kcb, 0);
681 } /* else: not a kprobe fault; let the kernel handle it */
683 preempt_enable_no_resched();
686 NOKPROBE_SYMBOL(kprobe_int3_handler);
689 * When a retprobed function returns, this code saves registers and
690 * calls trampoline_handler() runs, which calls the kretprobe's handler.
693 ".global kretprobe_trampoline\n"
694 ".type kretprobe_trampoline, @function\n"
695 "kretprobe_trampoline:\n"
697 /* We don't bother saving the ss register */
702 " call trampoline_handler\n"
703 /* Replace saved sp with true return address. */
704 " movq %rax, 152(%rsp)\n"
711 " call trampoline_handler\n"
712 /* Move flags to cs */
713 " movl 56(%esp), %edx\n"
714 " movl %edx, 52(%esp)\n"
715 /* Replace saved flags with true return address. */
716 " movl %eax, 56(%esp)\n"
721 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
723 NOKPROBE_SYMBOL(kretprobe_trampoline);
724 STACK_FRAME_NON_STANDARD(kretprobe_trampoline);
727 * Called from kretprobe_trampoline
729 __visible __used void *trampoline_handler(struct pt_regs *regs)
731 struct kretprobe_instance *ri = NULL;
732 struct hlist_head *head, empty_rp;
733 struct hlist_node *tmp;
734 unsigned long flags, orig_ret_address = 0;
735 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
736 kprobe_opcode_t *correct_ret_addr = NULL;
738 INIT_HLIST_HEAD(&empty_rp);
739 kretprobe_hash_lock(current, &head, &flags);
740 /* fixup registers */
742 regs->cs = __KERNEL_CS;
744 regs->cs = __KERNEL_CS | get_kernel_rpl();
747 regs->ip = trampoline_address;
748 regs->orig_ax = ~0UL;
751 * It is possible to have multiple instances associated with a given
752 * task either because multiple functions in the call path have
753 * return probes installed on them, and/or more than one
754 * return probe was registered for a target function.
756 * We can handle this because:
757 * - instances are always pushed into the head of the list
758 * - when multiple return probes are registered for the same
759 * function, the (chronologically) first instance's ret_addr
760 * will be the real return address, and all the rest will
761 * point to kretprobe_trampoline.
763 hlist_for_each_entry(ri, head, hlist) {
764 if (ri->task != current)
765 /* another task is sharing our hash bucket */
768 orig_ret_address = (unsigned long)ri->ret_addr;
770 if (orig_ret_address != trampoline_address)
772 * This is the real return address. Any other
773 * instances associated with this task are for
774 * other calls deeper on the call stack
779 kretprobe_assert(ri, orig_ret_address, trampoline_address);
781 correct_ret_addr = ri->ret_addr;
782 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
783 if (ri->task != current)
784 /* another task is sharing our hash bucket */
787 orig_ret_address = (unsigned long)ri->ret_addr;
788 if (ri->rp && ri->rp->handler) {
789 __this_cpu_write(current_kprobe, &ri->rp->kp);
790 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
791 ri->ret_addr = correct_ret_addr;
792 ri->rp->handler(ri, regs);
793 __this_cpu_write(current_kprobe, NULL);
796 recycle_rp_inst(ri, &empty_rp);
798 if (orig_ret_address != trampoline_address)
800 * This is the real return address. Any other
801 * instances associated with this task are for
802 * other calls deeper on the call stack
807 kretprobe_hash_unlock(current, &flags);
809 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
810 hlist_del(&ri->hlist);
813 return (void *)orig_ret_address;
815 NOKPROBE_SYMBOL(trampoline_handler);
818 * Called after single-stepping. p->addr is the address of the
819 * instruction whose first byte has been replaced by the "int 3"
820 * instruction. To avoid the SMP problems that can occur when we
821 * temporarily put back the original opcode to single-step, we
822 * single-stepped a copy of the instruction. The address of this
823 * copy is p->ainsn.insn.
825 * This function prepares to return from the post-single-step
826 * interrupt. We have to fix up the stack as follows:
828 * 0) Except in the case of absolute or indirect jump or call instructions,
829 * the new ip is relative to the copied instruction. We need to make
830 * it relative to the original instruction.
832 * 1) If the single-stepped instruction was pushfl, then the TF and IF
833 * flags are set in the just-pushed flags, and may need to be cleared.
835 * 2) If the single-stepped instruction was a call, the return address
836 * that is atop the stack is the address following the copied instruction.
837 * We need to make it the address following the original instruction.
839 * If this is the first time we've single-stepped the instruction at
840 * this probepoint, and the instruction is boostable, boost it: add a
841 * jump instruction after the copied instruction, that jumps to the next
842 * instruction after the probepoint.
844 static void resume_execution(struct kprobe *p, struct pt_regs *regs,
845 struct kprobe_ctlblk *kcb)
847 unsigned long *tos = stack_addr(regs);
848 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
849 unsigned long orig_ip = (unsigned long)p->addr;
850 kprobe_opcode_t *insn = p->ainsn.insn;
853 insn = skip_prefixes(insn);
855 regs->flags &= ~X86_EFLAGS_TF;
857 case 0x9c: /* pushfl */
858 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
859 *tos |= kcb->kprobe_old_flags;
861 case 0xc2: /* iret/ret/lret */
866 case 0xea: /* jmp absolute -- ip is correct */
867 /* ip is already adjusted, no more changes required */
868 p->ainsn.boostable = true;
870 case 0xe8: /* call relative - Fix return addr */
871 *tos = orig_ip + (*tos - copy_ip);
874 case 0x9a: /* call absolute -- same as call absolute, indirect */
875 *tos = orig_ip + (*tos - copy_ip);
879 if ((insn[1] & 0x30) == 0x10) {
881 * call absolute, indirect
882 * Fix return addr; ip is correct.
883 * But this is not boostable
885 *tos = orig_ip + (*tos - copy_ip);
887 } else if (((insn[1] & 0x31) == 0x20) ||
888 ((insn[1] & 0x31) == 0x21)) {
890 * jmp near and far, absolute indirect
891 * ip is correct. And this is boostable
893 p->ainsn.boostable = true;
900 regs->ip += orig_ip - copy_ip;
905 NOKPROBE_SYMBOL(resume_execution);
908 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
909 * remain disabled throughout this function.
911 int kprobe_debug_handler(struct pt_regs *regs)
913 struct kprobe *cur = kprobe_running();
914 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
919 resume_execution(cur, regs, kcb);
920 regs->flags |= kcb->kprobe_saved_flags;
922 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
923 kcb->kprobe_status = KPROBE_HIT_SSDONE;
924 cur->post_handler(cur, regs, 0);
927 /* Restore back the original saved kprobes variables and continue. */
928 if (kcb->kprobe_status == KPROBE_REENTER) {
929 restore_previous_kprobe(kcb);
932 reset_current_kprobe();
934 preempt_enable_no_resched();
937 * if somebody else is singlestepping across a probe point, flags
938 * will have TF set, in which case, continue the remaining processing
939 * of do_debug, as if this is not a probe hit.
941 if (regs->flags & X86_EFLAGS_TF)
946 NOKPROBE_SYMBOL(kprobe_debug_handler);
948 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
950 struct kprobe *cur = kprobe_running();
951 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
953 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) {
954 /* This must happen on single-stepping */
955 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS &&
956 kcb->kprobe_status != KPROBE_REENTER);
958 * We are here because the instruction being single
959 * stepped caused a page fault. We reset the current
960 * kprobe and the ip points back to the probe address
961 * and allow the page fault handler to continue as a
964 regs->ip = (unsigned long)cur->addr;
966 * Trap flag (TF) has been set here because this fault
967 * happened where the single stepping will be done.
968 * So clear it by resetting the current kprobe:
970 regs->flags &= ~X86_EFLAGS_TF;
973 * If the TF flag was set before the kprobe hit,
976 regs->flags |= kcb->kprobe_old_flags;
978 if (kcb->kprobe_status == KPROBE_REENTER)
979 restore_previous_kprobe(kcb);
981 reset_current_kprobe();
982 preempt_enable_no_resched();
983 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE ||
984 kcb->kprobe_status == KPROBE_HIT_SSDONE) {
986 * We increment the nmissed count for accounting,
987 * we can also use npre/npostfault count for accounting
988 * these specific fault cases.
990 kprobes_inc_nmissed_count(cur);
993 * We come here because instructions in the pre/post
994 * handler caused the page_fault, this could happen
995 * if handler tries to access user space by
996 * copy_from_user(), get_user() etc. Let the
997 * user-specified handler try to fix it first.
999 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
1003 * In case the user-specified fault handler returned
1004 * zero, try to fix up.
1006 if (fixup_exception(regs, trapnr))
1010 * fixup routine could not handle it,
1011 * Let do_page_fault() fix it.
1017 NOKPROBE_SYMBOL(kprobe_fault_handler);
1020 * Wrapper routine for handling exceptions.
1022 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
1025 struct die_args *args = data;
1026 int ret = NOTIFY_DONE;
1028 if (args->regs && user_mode(args->regs))
1031 if (val == DIE_GPF) {
1033 * To be potentially processing a kprobe fault and to
1034 * trust the result from kprobe_running(), we have
1035 * be non-preemptible.
1037 if (!preemptible() && kprobe_running() &&
1038 kprobe_fault_handler(args->regs, args->trapnr))
1043 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1045 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
1047 struct jprobe *jp = container_of(p, struct jprobe, kp);
1049 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1051 kcb->jprobe_saved_regs = *regs;
1052 kcb->jprobe_saved_sp = stack_addr(regs);
1053 addr = (unsigned long)(kcb->jprobe_saved_sp);
1056 * As Linus pointed out, gcc assumes that the callee
1057 * owns the argument space and could overwrite it, e.g.
1058 * tailcall optimization. So, to be absolutely safe
1059 * we also save and restore enough stack bytes to cover
1060 * the argument area.
1061 * Use __memcpy() to avoid KASAN stack out-of-bounds reports as we copy
1062 * raw stack chunk with redzones:
1064 __memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr, MIN_STACK_SIZE(addr));
1065 regs->flags &= ~X86_EFLAGS_IF;
1066 trace_hardirqs_off();
1067 regs->ip = (unsigned long)(jp->entry);
1070 * jprobes use jprobe_return() which skips the normal return
1071 * path of the function, and this messes up the accounting of the
1072 * function graph tracer to get messed up.
1074 * Pause function graph tracing while performing the jprobe function.
1076 pause_graph_tracing();
1079 NOKPROBE_SYMBOL(setjmp_pre_handler);
1081 void jprobe_return(void)
1083 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1085 /* Unpoison stack redzones in the frames we are going to jump over. */
1086 kasan_unpoison_stack_above_sp_to(kcb->jprobe_saved_sp);
1089 #ifdef CONFIG_X86_64
1090 " xchg %%rbx,%%rsp \n"
1092 " xchgl %%ebx,%%esp \n"
1095 " .globl jprobe_return_end\n"
1096 " jprobe_return_end: \n"
1098 (kcb->jprobe_saved_sp):"memory");
1100 NOKPROBE_SYMBOL(jprobe_return);
1101 NOKPROBE_SYMBOL(jprobe_return_end);
1103 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1105 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1106 u8 *addr = (u8 *) (regs->ip - 1);
1107 struct jprobe *jp = container_of(p, struct jprobe, kp);
1108 void *saved_sp = kcb->jprobe_saved_sp;
1110 if ((addr > (u8 *) jprobe_return) &&
1111 (addr < (u8 *) jprobe_return_end)) {
1112 if (stack_addr(regs) != saved_sp) {
1113 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1115 "current sp %p does not match saved sp %p\n",
1116 stack_addr(regs), saved_sp);
1117 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1118 show_regs(saved_regs);
1119 printk(KERN_ERR "Current registers\n");
1123 /* It's OK to start function graph tracing again */
1124 unpause_graph_tracing();
1125 *regs = kcb->jprobe_saved_regs;
1126 __memcpy(saved_sp, kcb->jprobes_stack, MIN_STACK_SIZE(saved_sp));
1127 preempt_enable_no_resched();
1132 NOKPROBE_SYMBOL(longjmp_break_handler);
1134 bool arch_within_kprobe_blacklist(unsigned long addr)
1136 return (addr >= (unsigned long)__kprobes_text_start &&
1137 addr < (unsigned long)__kprobes_text_end) ||
1138 (addr >= (unsigned long)__entry_text_start &&
1139 addr < (unsigned long)__entry_text_end);
1142 int __init arch_init_kprobes(void)
1147 int arch_trampoline_kprobe(struct kprobe *p)