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.
43 #include <linux/kprobes.h>
44 #include <linux/ptrace.h>
45 #include <linux/string.h>
46 #include <linux/slab.h>
47 #include <linux/hardirq.h>
48 #include <linux/preempt.h>
49 #include <linux/module.h>
50 #include <linux/kdebug.h>
51 #include <linux/kallsyms.h>
53 #include <asm/cacheflush.h>
55 #include <asm/pgtable.h>
56 #include <asm/uaccess.h>
57 #include <asm/alternative.h>
60 void jprobe_return_end(void);
62 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
63 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
66 #define stack_addr(regs) ((unsigned long *)regs->sp)
69 * "®s->sp" looks wrong, but it's correct for x86_32. x86_32 CPUs
70 * don't save the ss and esp registers if the CPU is already in kernel
71 * mode when it traps. So for kprobes, regs->sp and regs->ss are not
72 * the [nonexistent] saved stack pointer and ss register, but rather
73 * the top 8 bytes of the pre-int3 stack. So ®s->sp happens to
74 * point to the top of the pre-int3 stack.
76 #define stack_addr(regs) ((unsigned long *)®s->sp)
79 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
80 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
81 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
82 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
83 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
86 * Undefined/reserved opcodes, conditional jump, Opcode Extension
87 * Groups, and some special opcodes can not boost.
89 static const u32 twobyte_is_boostable[256 / 32] = {
90 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
91 /* ---------------------------------------------- */
92 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
93 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
94 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
95 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
96 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
97 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
98 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
99 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
100 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
101 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
102 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
103 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
104 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
105 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
106 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
107 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
108 /* ----------------------------------------------- */
109 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
113 struct kretprobe_blackpoint kretprobe_blacklist[] = {
114 {"__switch_to", }, /* This function switches only current task, but
115 doesn't switch kernel stack.*/
116 {NULL, NULL} /* Terminator */
118 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
120 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
121 static void __kprobes set_jmp_op(void *from, void *to)
123 struct __arch_jmp_op {
126 } __attribute__((packed)) * jop;
127 jop = (struct __arch_jmp_op *)from;
128 jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
129 jop->op = RELATIVEJUMP_INSTRUCTION;
133 * Check for the REX prefix which can only exist on X86_64
134 * X86_32 always returns 0
136 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
139 if ((*insn & 0xf0) == 0x40)
146 * Returns non-zero if opcode is boostable.
147 * RIP relative instructions are adjusted at copying time in 64 bits mode
149 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
151 kprobe_opcode_t opcode;
152 kprobe_opcode_t *orig_opcodes = opcodes;
154 if (search_exception_tables((unsigned long)opcodes))
155 return 0; /* Page fault may occur on this address. */
158 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
160 opcode = *(opcodes++);
162 /* 2nd-byte opcode */
163 if (opcode == 0x0f) {
164 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
166 return test_bit(*opcodes,
167 (unsigned long *)twobyte_is_boostable);
170 switch (opcode & 0xf0) {
173 goto retry; /* REX prefix is boostable */
176 if (0x63 < opcode && opcode < 0x67)
177 goto retry; /* prefixes */
178 /* can't boost Address-size override and bound */
179 return (opcode != 0x62 && opcode != 0x67);
181 return 0; /* can't boost conditional jump */
183 /* can't boost software-interruptions */
184 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
186 /* can boost AA* and XLAT */
187 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
189 /* can boost in/out and absolute jmps */
190 return ((opcode & 0x04) || opcode == 0xea);
192 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
193 goto retry; /* lock/rep(ne) prefix */
194 /* clear and set flags are boostable */
195 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
197 /* segment override prefixes are boostable */
198 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
199 goto retry; /* prefixes */
200 /* CS override prefix and call are not boostable */
201 return (opcode != 0x2e && opcode != 0x9a);
205 /* Recover the probed instruction at addr for further analysis. */
206 static int recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
209 kp = get_kprobe((void *)addr);
214 * Basically, kp->ainsn.insn has an original instruction.
215 * However, RIP-relative instruction can not do single-stepping
216 * at different place, fix_riprel() tweaks the displacement of
217 * that instruction. In that case, we can't recover the instruction
218 * from the kp->ainsn.insn.
220 * On the other hand, kp->opcode has a copy of the first byte of
221 * the probed instruction, which is overwritten by int3. And
222 * the instruction at kp->addr is not modified by kprobes except
223 * for the first byte, we can recover the original instruction
224 * from it and kp->opcode.
226 memcpy(buf, kp->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
231 /* Dummy buffers for kallsyms_lookup */
232 static char __dummy_buf[KSYM_NAME_LEN];
234 /* Check if paddr is at an instruction boundary */
235 static int __kprobes can_probe(unsigned long paddr)
238 unsigned long addr, offset = 0;
240 kprobe_opcode_t buf[MAX_INSN_SIZE];
242 if (!kallsyms_lookup(paddr, NULL, &offset, NULL, __dummy_buf))
245 /* Decode instructions */
246 addr = paddr - offset;
247 while (addr < paddr) {
248 kernel_insn_init(&insn, (void *)addr);
249 insn_get_opcode(&insn);
252 * Check if the instruction has been modified by another
253 * kprobe, in which case we replace the breakpoint by the
254 * original instruction in our buffer.
256 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
257 ret = recover_probed_instruction(buf, addr);
260 * Another debugging subsystem might insert
261 * this breakpoint. In that case, we can't
265 kernel_insn_init(&insn, buf);
267 insn_get_length(&insn);
271 return (addr == paddr);
275 * Returns non-zero if opcode modifies the interrupt flag.
277 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
282 case 0xcf: /* iret/iretd */
283 case 0x9d: /* popf/popfd */
288 * on X86_64, 0x40-0x4f are REX prefixes so we need to look
289 * at the next byte instead.. but of course not recurse infinitely
291 if (is_REX_prefix(insn))
292 return is_IF_modifier(++insn);
298 * Adjust the displacement if the instruction uses the %rip-relative
300 * If it does, Return the address of the 32-bit displacement word.
301 * If not, return null.
302 * Only applicable to 64-bit x86.
304 static void __kprobes fix_riprel(struct kprobe *p)
308 kernel_insn_init(&insn, p->ainsn.insn);
310 if (insn_rip_relative(&insn)) {
313 insn_get_displacement(&insn);
315 * The copied instruction uses the %rip-relative addressing
316 * mode. Adjust the displacement for the difference between
317 * the original location of this instruction and the location
318 * of the copy that will actually be run. The tricky bit here
319 * is making sure that the sign extension happens correctly in
320 * this calculation, since we need a signed 32-bit result to
321 * be sign-extended to 64 bits when it's added to the %rip
322 * value and yield the same 64-bit result that the sign-
323 * extension of the original signed 32-bit displacement would
326 newdisp = (u8 *) p->addr + (s64) insn.displacement.value -
327 (u8 *) p->ainsn.insn;
328 BUG_ON((s64) (s32) newdisp != newdisp); /* Sanity check. */
329 disp = (u8 *) p->ainsn.insn + insn_offset_displacement(&insn);
330 *(s32 *) disp = (s32) newdisp;
335 static void __kprobes arch_copy_kprobe(struct kprobe *p)
337 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
341 if (can_boost(p->addr))
342 p->ainsn.boostable = 0;
344 p->ainsn.boostable = -1;
346 p->opcode = *p->addr;
349 int __kprobes arch_prepare_kprobe(struct kprobe *p)
351 if (!can_probe((unsigned long)p->addr))
353 /* insn: must be on special executable page on x86. */
354 p->ainsn.insn = get_insn_slot();
361 void __kprobes arch_arm_kprobe(struct kprobe *p)
363 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
366 void __kprobes arch_disarm_kprobe(struct kprobe *p)
368 text_poke(p->addr, &p->opcode, 1);
371 void __kprobes arch_remove_kprobe(struct kprobe *p)
374 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
375 p->ainsn.insn = NULL;
379 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
381 kcb->prev_kprobe.kp = kprobe_running();
382 kcb->prev_kprobe.status = kcb->kprobe_status;
383 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
384 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
387 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
389 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
390 kcb->kprobe_status = kcb->prev_kprobe.status;
391 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
392 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
395 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
396 struct kprobe_ctlblk *kcb)
398 __get_cpu_var(current_kprobe) = p;
399 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
400 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
401 if (is_IF_modifier(p->ainsn.insn))
402 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
405 static void __kprobes clear_btf(void)
407 if (test_thread_flag(TIF_DEBUGCTLMSR))
408 update_debugctlmsr(0);
411 static void __kprobes restore_btf(void)
413 if (test_thread_flag(TIF_DEBUGCTLMSR))
414 update_debugctlmsr(current->thread.debugctlmsr);
417 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
420 regs->flags |= X86_EFLAGS_TF;
421 regs->flags &= ~X86_EFLAGS_IF;
422 /* single step inline if the instruction is an int3 */
423 if (p->opcode == BREAKPOINT_INSTRUCTION)
424 regs->ip = (unsigned long)p->addr;
426 regs->ip = (unsigned long)p->ainsn.insn;
429 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
430 struct pt_regs *regs)
432 unsigned long *sara = stack_addr(regs);
434 ri->ret_addr = (kprobe_opcode_t *) *sara;
436 /* Replace the return addr with trampoline addr */
437 *sara = (unsigned long) &kretprobe_trampoline;
440 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
441 struct kprobe_ctlblk *kcb)
443 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
444 if (p->ainsn.boostable == 1 && !p->post_handler) {
445 /* Boost up -- we can execute copied instructions directly */
446 reset_current_kprobe();
447 regs->ip = (unsigned long)p->ainsn.insn;
448 preempt_enable_no_resched();
452 prepare_singlestep(p, regs);
453 kcb->kprobe_status = KPROBE_HIT_SS;
457 * We have reentered the kprobe_handler(), since another probe was hit while
458 * within the handler. We save the original kprobes variables and just single
459 * step on the instruction of the new probe without calling any user handlers.
461 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
462 struct kprobe_ctlblk *kcb)
464 switch (kcb->kprobe_status) {
465 case KPROBE_HIT_SSDONE:
467 /* TODO: Provide re-entrancy from post_kprobes_handler() and
468 * avoid exception stack corruption while single-stepping on
469 * the instruction of the new probe.
471 arch_disarm_kprobe(p);
472 regs->ip = (unsigned long)p->addr;
473 reset_current_kprobe();
474 preempt_enable_no_resched();
477 case KPROBE_HIT_ACTIVE:
478 save_previous_kprobe(kcb);
479 set_current_kprobe(p, regs, kcb);
480 kprobes_inc_nmissed_count(p);
481 prepare_singlestep(p, regs);
482 kcb->kprobe_status = KPROBE_REENTER;
485 /* A probe has been hit in the codepath leading up to, or just
486 * after, single-stepping of a probed instruction. This entire
487 * codepath should strictly reside in .kprobes.text section.
488 * Raise a BUG or we'll continue in an endless reentering loop
489 * and eventually a stack overflow.
491 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
496 /* impossible cases */
505 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
506 * remain disabled thorough out this function.
508 static int __kprobes kprobe_handler(struct pt_regs *regs)
510 kprobe_opcode_t *addr;
512 struct kprobe_ctlblk *kcb;
514 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
515 if (*addr != BREAKPOINT_INSTRUCTION) {
517 * The breakpoint instruction was removed right
518 * after we hit it. Another cpu has removed
519 * either a probepoint or a debugger breakpoint
520 * at this address. In either case, no further
521 * handling of this interrupt is appropriate.
522 * Back up over the (now missing) int3 and run
523 * the original instruction.
525 regs->ip = (unsigned long)addr;
530 * We don't want to be preempted for the entire
531 * duration of kprobe processing. We conditionally
532 * re-enable preemption at the end of this function,
533 * and also in reenter_kprobe() and setup_singlestep().
537 kcb = get_kprobe_ctlblk();
538 p = get_kprobe(addr);
541 if (kprobe_running()) {
542 if (reenter_kprobe(p, regs, kcb))
545 set_current_kprobe(p, regs, kcb);
546 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
549 * If we have no pre-handler or it returned 0, we
550 * continue with normal processing. If we have a
551 * pre-handler and it returned non-zero, it prepped
552 * for calling the break_handler below on re-entry
553 * for jprobe processing, so get out doing nothing
556 if (!p->pre_handler || !p->pre_handler(p, regs))
557 setup_singlestep(p, regs, kcb);
560 } else if (kprobe_running()) {
561 p = __get_cpu_var(current_kprobe);
562 if (p->break_handler && p->break_handler(p, regs)) {
563 setup_singlestep(p, regs, kcb);
566 } /* else: not a kprobe fault; let the kernel handle it */
568 preempt_enable_no_resched();
573 * When a retprobed function returns, this code saves registers and
574 * calls trampoline_handler() runs, which calls the kretprobe's handler.
576 static void __used __kprobes kretprobe_trampoline_holder(void)
579 ".global kretprobe_trampoline\n"
580 "kretprobe_trampoline: \n"
582 /* We don't bother saving the ss register */
586 * Skip cs, ip, orig_ax.
587 * trampoline_handler() will plug in these values
606 " call trampoline_handler\n"
607 /* Replace saved sp with true return address. */
608 " movq %rax, 152(%rsp)\n"
624 /* Skip orig_ax, ip, cs */
630 * Skip cs, ip, orig_ax and gs.
631 * trampoline_handler() will plug in these values
645 " call trampoline_handler\n"
646 /* Move flags to cs */
647 " movl 56(%esp), %edx\n"
648 " movl %edx, 52(%esp)\n"
649 /* Replace saved flags with true return address. */
650 " movl %eax, 56(%esp)\n"
658 /* Skip ds, es, fs, gs, orig_ax and ip */
666 * Called from kretprobe_trampoline
668 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
670 struct kretprobe_instance *ri = NULL;
671 struct hlist_head *head, empty_rp;
672 struct hlist_node *node, *tmp;
673 unsigned long flags, orig_ret_address = 0;
674 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
676 INIT_HLIST_HEAD(&empty_rp);
677 kretprobe_hash_lock(current, &head, &flags);
678 /* fixup registers */
680 regs->cs = __KERNEL_CS;
682 regs->cs = __KERNEL_CS | get_kernel_rpl();
685 regs->ip = trampoline_address;
686 regs->orig_ax = ~0UL;
689 * It is possible to have multiple instances associated with a given
690 * task either because multiple functions in the call path have
691 * return probes installed on them, and/or more than one
692 * return probe was registered for a target function.
694 * We can handle this because:
695 * - instances are always pushed into the head of the list
696 * - when multiple return probes are registered for the same
697 * function, the (chronologically) first instance's ret_addr
698 * will be the real return address, and all the rest will
699 * point to kretprobe_trampoline.
701 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
702 if (ri->task != current)
703 /* another task is sharing our hash bucket */
706 if (ri->rp && ri->rp->handler) {
707 __get_cpu_var(current_kprobe) = &ri->rp->kp;
708 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
709 ri->rp->handler(ri, regs);
710 __get_cpu_var(current_kprobe) = NULL;
713 orig_ret_address = (unsigned long)ri->ret_addr;
714 recycle_rp_inst(ri, &empty_rp);
716 if (orig_ret_address != trampoline_address)
718 * This is the real return address. Any other
719 * instances associated with this task are for
720 * other calls deeper on the call stack
725 kretprobe_assert(ri, orig_ret_address, trampoline_address);
727 kretprobe_hash_unlock(current, &flags);
729 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
730 hlist_del(&ri->hlist);
733 return (void *)orig_ret_address;
737 * Called after single-stepping. p->addr is the address of the
738 * instruction whose first byte has been replaced by the "int 3"
739 * instruction. To avoid the SMP problems that can occur when we
740 * temporarily put back the original opcode to single-step, we
741 * single-stepped a copy of the instruction. The address of this
742 * copy is p->ainsn.insn.
744 * This function prepares to return from the post-single-step
745 * interrupt. We have to fix up the stack as follows:
747 * 0) Except in the case of absolute or indirect jump or call instructions,
748 * the new ip is relative to the copied instruction. We need to make
749 * it relative to the original instruction.
751 * 1) If the single-stepped instruction was pushfl, then the TF and IF
752 * flags are set in the just-pushed flags, and may need to be cleared.
754 * 2) If the single-stepped instruction was a call, the return address
755 * that is atop the stack is the address following the copied instruction.
756 * We need to make it the address following the original instruction.
758 * If this is the first time we've single-stepped the instruction at
759 * this probepoint, and the instruction is boostable, boost it: add a
760 * jump instruction after the copied instruction, that jumps to the next
761 * instruction after the probepoint.
763 static void __kprobes resume_execution(struct kprobe *p,
764 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
766 unsigned long *tos = stack_addr(regs);
767 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
768 unsigned long orig_ip = (unsigned long)p->addr;
769 kprobe_opcode_t *insn = p->ainsn.insn;
771 /*skip the REX prefix*/
772 if (is_REX_prefix(insn))
775 regs->flags &= ~X86_EFLAGS_TF;
777 case 0x9c: /* pushfl */
778 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
779 *tos |= kcb->kprobe_old_flags;
781 case 0xc2: /* iret/ret/lret */
786 case 0xea: /* jmp absolute -- ip is correct */
787 /* ip is already adjusted, no more changes required */
788 p->ainsn.boostable = 1;
790 case 0xe8: /* call relative - Fix return addr */
791 *tos = orig_ip + (*tos - copy_ip);
794 case 0x9a: /* call absolute -- same as call absolute, indirect */
795 *tos = orig_ip + (*tos - copy_ip);
799 if ((insn[1] & 0x30) == 0x10) {
801 * call absolute, indirect
802 * Fix return addr; ip is correct.
803 * But this is not boostable
805 *tos = orig_ip + (*tos - copy_ip);
807 } else if (((insn[1] & 0x31) == 0x20) ||
808 ((insn[1] & 0x31) == 0x21)) {
810 * jmp near and far, absolute indirect
811 * ip is correct. And this is boostable
813 p->ainsn.boostable = 1;
820 if (p->ainsn.boostable == 0) {
821 if ((regs->ip > copy_ip) &&
822 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
824 * These instructions can be executed directly if it
825 * jumps back to correct address.
827 set_jmp_op((void *)regs->ip,
828 (void *)orig_ip + (regs->ip - copy_ip));
829 p->ainsn.boostable = 1;
831 p->ainsn.boostable = -1;
835 regs->ip += orig_ip - copy_ip;
842 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
843 * remain disabled thoroughout this function.
845 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
847 struct kprobe *cur = kprobe_running();
848 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
853 resume_execution(cur, regs, kcb);
854 regs->flags |= kcb->kprobe_saved_flags;
856 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
857 kcb->kprobe_status = KPROBE_HIT_SSDONE;
858 cur->post_handler(cur, regs, 0);
861 /* Restore back the original saved kprobes variables and continue. */
862 if (kcb->kprobe_status == KPROBE_REENTER) {
863 restore_previous_kprobe(kcb);
866 reset_current_kprobe();
868 preempt_enable_no_resched();
871 * if somebody else is singlestepping across a probe point, flags
872 * will have TF set, in which case, continue the remaining processing
873 * of do_debug, as if this is not a probe hit.
875 if (regs->flags & X86_EFLAGS_TF)
881 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
883 struct kprobe *cur = kprobe_running();
884 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
886 switch (kcb->kprobe_status) {
890 * We are here because the instruction being single
891 * stepped caused a page fault. We reset the current
892 * kprobe and the ip points back to the probe address
893 * and allow the page fault handler to continue as a
896 regs->ip = (unsigned long)cur->addr;
897 regs->flags |= kcb->kprobe_old_flags;
898 if (kcb->kprobe_status == KPROBE_REENTER)
899 restore_previous_kprobe(kcb);
901 reset_current_kprobe();
902 preempt_enable_no_resched();
904 case KPROBE_HIT_ACTIVE:
905 case KPROBE_HIT_SSDONE:
907 * We increment the nmissed count for accounting,
908 * we can also use npre/npostfault count for accounting
909 * these specific fault cases.
911 kprobes_inc_nmissed_count(cur);
914 * We come here because instructions in the pre/post
915 * handler caused the page_fault, this could happen
916 * if handler tries to access user space by
917 * copy_from_user(), get_user() etc. Let the
918 * user-specified handler try to fix it first.
920 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
924 * In case the user-specified fault handler returned
925 * zero, try to fix up.
927 if (fixup_exception(regs))
931 * fixup routine could not handle it,
932 * Let do_page_fault() fix it.
942 * Wrapper routine for handling exceptions.
944 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
945 unsigned long val, void *data)
947 struct die_args *args = data;
948 int ret = NOTIFY_DONE;
950 if (args->regs && user_mode_vm(args->regs))
955 if (kprobe_handler(args->regs))
959 if (post_kprobe_handler(args->regs))
964 * To be potentially processing a kprobe fault and to
965 * trust the result from kprobe_running(), we have
966 * be non-preemptible.
968 if (!preemptible() && kprobe_running() &&
969 kprobe_fault_handler(args->regs, args->trapnr))
978 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
980 struct jprobe *jp = container_of(p, struct jprobe, kp);
982 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
984 kcb->jprobe_saved_regs = *regs;
985 kcb->jprobe_saved_sp = stack_addr(regs);
986 addr = (unsigned long)(kcb->jprobe_saved_sp);
989 * As Linus pointed out, gcc assumes that the callee
990 * owns the argument space and could overwrite it, e.g.
991 * tailcall optimization. So, to be absolutely safe
992 * we also save and restore enough stack bytes to cover
995 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
996 MIN_STACK_SIZE(addr));
997 regs->flags &= ~X86_EFLAGS_IF;
998 trace_hardirqs_off();
999 regs->ip = (unsigned long)(jp->entry);
1003 void __kprobes jprobe_return(void)
1005 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1008 #ifdef CONFIG_X86_64
1009 " xchg %%rbx,%%rsp \n"
1011 " xchgl %%ebx,%%esp \n"
1014 " .globl jprobe_return_end\n"
1015 " jprobe_return_end: \n"
1017 (kcb->jprobe_saved_sp):"memory");
1020 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1022 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1023 u8 *addr = (u8 *) (regs->ip - 1);
1024 struct jprobe *jp = container_of(p, struct jprobe, kp);
1026 if ((addr > (u8 *) jprobe_return) &&
1027 (addr < (u8 *) jprobe_return_end)) {
1028 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1029 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1031 "current sp %p does not match saved sp %p\n",
1032 stack_addr(regs), kcb->jprobe_saved_sp);
1033 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1034 show_registers(saved_regs);
1035 printk(KERN_ERR "Current registers\n");
1036 show_registers(regs);
1039 *regs = kcb->jprobe_saved_regs;
1040 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1042 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1043 preempt_enable_no_resched();
1049 int __init arch_init_kprobes(void)
1054 int __kprobes arch_trampoline_kprobe(struct kprobe *p)