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/preempt.h>
48 #include <linux/module.h>
49 #include <linux/kdebug.h>
51 #include <asm/cacheflush.h>
53 #include <asm/pgtable.h>
54 #include <asm/uaccess.h>
55 #include <asm/alternative.h>
57 void jprobe_return_end(void);
59 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
60 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
63 #define stack_addr(regs) ((unsigned long *)regs->sp)
66 * "®s->sp" looks wrong, but it's correct for x86_32. x86_32 CPUs
67 * don't save the ss and esp registers if the CPU is already in kernel
68 * mode when it traps. So for kprobes, regs->sp and regs->ss are not
69 * the [nonexistent] saved stack pointer and ss register, but rather
70 * the top 8 bytes of the pre-int3 stack. So ®s->sp happens to
71 * point to the top of the pre-int3 stack.
73 #define stack_addr(regs) ((unsigned long *)®s->sp)
76 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
77 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
78 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
79 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
80 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
83 * Undefined/reserved opcodes, conditional jump, Opcode Extension
84 * Groups, and some special opcodes can not boost.
86 static const u32 twobyte_is_boostable[256 / 32] = {
87 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
88 /* ---------------------------------------------- */
89 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
90 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
91 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
92 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
93 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
94 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
95 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
96 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
97 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
98 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
99 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
100 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
101 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
102 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
103 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
104 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
105 /* ----------------------------------------------- */
106 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
108 static const u32 onebyte_has_modrm[256 / 32] = {
109 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
110 /* ----------------------------------------------- */
111 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
112 W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
113 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
114 W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
115 W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
116 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
117 W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
118 W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
119 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
120 W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
121 W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
122 W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
123 W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
124 W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
125 W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
126 W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) /* f0 */
127 /* ----------------------------------------------- */
128 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
130 static const u32 twobyte_has_modrm[256 / 32] = {
131 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
132 /* ----------------------------------------------- */
133 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
134 W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
135 W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
136 W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
137 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
138 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
139 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
140 W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
141 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
142 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
143 W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
144 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
145 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
146 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
147 W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
148 W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0) /* ff */
149 /* ----------------------------------------------- */
150 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
154 struct kretprobe_blackpoint kretprobe_blacklist[] = {
155 {"__switch_to", }, /* This function switches only current task, but
156 doesn't switch kernel stack.*/
157 {NULL, NULL} /* Terminator */
159 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
161 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
162 static void __kprobes set_jmp_op(void *from, void *to)
164 struct __arch_jmp_op {
167 } __attribute__((packed)) * jop;
168 jop = (struct __arch_jmp_op *)from;
169 jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
170 jop->op = RELATIVEJUMP_INSTRUCTION;
174 * Check for the REX prefix which can only exist on X86_64
175 * X86_32 always returns 0
177 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
180 if ((*insn & 0xf0) == 0x40)
187 * Returns non-zero if opcode is boostable.
188 * RIP relative instructions are adjusted at copying time in 64 bits mode
190 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
192 kprobe_opcode_t opcode;
193 kprobe_opcode_t *orig_opcodes = opcodes;
196 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
198 opcode = *(opcodes++);
200 /* 2nd-byte opcode */
201 if (opcode == 0x0f) {
202 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
204 return test_bit(*opcodes,
205 (unsigned long *)twobyte_is_boostable);
208 switch (opcode & 0xf0) {
211 goto retry; /* REX prefix is boostable */
214 if (0x63 < opcode && opcode < 0x67)
215 goto retry; /* prefixes */
216 /* can't boost Address-size override and bound */
217 return (opcode != 0x62 && opcode != 0x67);
219 return 0; /* can't boost conditional jump */
221 /* can't boost software-interruptions */
222 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
224 /* can boost AA* and XLAT */
225 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
227 /* can boost in/out and absolute jmps */
228 return ((opcode & 0x04) || opcode == 0xea);
230 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
231 goto retry; /* lock/rep(ne) prefix */
232 /* clear and set flags are boostable */
233 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
235 /* segment override prefixes are boostable */
236 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
237 goto retry; /* prefixes */
238 /* CS override prefix and call are not boostable */
239 return (opcode != 0x2e && opcode != 0x9a);
244 * Returns non-zero if opcode modifies the interrupt flag.
246 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
251 case 0xcf: /* iret/iretd */
252 case 0x9d: /* popf/popfd */
257 * on X86_64, 0x40-0x4f are REX prefixes so we need to look
258 * at the next byte instead.. but of course not recurse infinitely
260 if (is_REX_prefix(insn))
261 return is_IF_modifier(++insn);
268 * Adjust the displacement if the instruction uses the %rip-relative
270 * If it does, Return the address of the 32-bit displacement word.
271 * If not, return null.
273 static void __kprobes fix_riprel(struct kprobe *p)
275 u8 *insn = p->ainsn.insn;
279 /* Skip legacy instruction prefixes. */
299 /* Skip REX instruction prefix. */
300 if (is_REX_prefix(insn))
304 /* Two-byte opcode. */
306 need_modrm = test_bit(*insn,
307 (unsigned long *)twobyte_has_modrm);
309 /* One-byte opcode. */
310 need_modrm = test_bit(*insn,
311 (unsigned long *)onebyte_has_modrm);
315 if ((modrm & 0xc7) == 0x05) {
316 /* %rip+disp32 addressing mode */
317 /* Displacement follows ModRM byte. */
320 * The copied instruction uses the %rip-relative
321 * addressing mode. Adjust the displacement for the
322 * difference between the original location of this
323 * instruction and the location of the copy that will
324 * actually be run. The tricky bit here is making sure
325 * that the sign extension happens correctly in this
326 * calculation, since we need a signed 32-bit result to
327 * be sign-extended to 64 bits when it's added to the
328 * %rip value and yield the same 64-bit result that the
329 * sign-extension of the original signed 32-bit
330 * displacement would have given.
332 disp = (u8 *) p->addr + *((s32 *) insn) -
333 (u8 *) p->ainsn.insn;
334 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
335 *(s32 *)insn = (s32) disp;
341 static void __kprobes arch_copy_kprobe(struct kprobe *p)
343 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
347 if (can_boost(p->addr))
348 p->ainsn.boostable = 0;
350 p->ainsn.boostable = -1;
352 p->opcode = *p->addr;
355 int __kprobes arch_prepare_kprobe(struct kprobe *p)
357 /* insn: must be on special executable page on x86. */
358 p->ainsn.insn = get_insn_slot();
365 void __kprobes arch_arm_kprobe(struct kprobe *p)
367 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
370 void __kprobes arch_disarm_kprobe(struct kprobe *p)
372 text_poke(p->addr, &p->opcode, 1);
375 void __kprobes arch_remove_kprobe(struct kprobe *p)
377 mutex_lock(&kprobe_mutex);
378 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
379 mutex_unlock(&kprobe_mutex);
382 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
384 kcb->prev_kprobe.kp = kprobe_running();
385 kcb->prev_kprobe.status = kcb->kprobe_status;
386 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
387 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
390 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
392 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
393 kcb->kprobe_status = kcb->prev_kprobe.status;
394 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
395 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
398 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
399 struct kprobe_ctlblk *kcb)
401 __get_cpu_var(current_kprobe) = p;
402 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
403 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
404 if (is_IF_modifier(p->ainsn.insn))
405 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
408 static void __kprobes clear_btf(void)
410 if (test_thread_flag(TIF_DEBUGCTLMSR))
411 wrmsr(MSR_IA32_DEBUGCTLMSR, 0, 0);
414 static void __kprobes restore_btf(void)
416 if (test_thread_flag(TIF_DEBUGCTLMSR))
417 wrmsr(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr, 0);
420 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
423 regs->flags |= X86_EFLAGS_TF;
424 regs->flags &= ~X86_EFLAGS_IF;
425 /* single step inline if the instruction is an int3 */
426 if (p->opcode == BREAKPOINT_INSTRUCTION)
427 regs->ip = (unsigned long)p->addr;
429 regs->ip = (unsigned long)p->ainsn.insn;
432 /* Called with kretprobe_lock held */
433 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
434 struct pt_regs *regs)
436 unsigned long *sara = stack_addr(regs);
438 ri->ret_addr = (kprobe_opcode_t *) *sara;
440 /* Replace the return addr with trampoline addr */
441 *sara = (unsigned long) &kretprobe_trampoline;
444 * We have reentered the kprobe_handler(), since another probe was hit while
445 * within the handler. We save the original kprobes variables and just single
446 * step on the instruction of the new probe without calling any user handlers.
448 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
449 struct kprobe_ctlblk *kcb)
451 if (kcb->kprobe_status == KPROBE_HIT_SS &&
452 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
453 regs->flags &= ~X86_EFLAGS_TF;
454 regs->flags |= kcb->kprobe_saved_flags;
457 } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
458 /* TODO: Provide re-entrancy from post_kprobes_handler() and
459 * avoid exception stack corruption while single-stepping on
460 * the instruction of the new probe.
462 arch_disarm_kprobe(p);
463 regs->ip = (unsigned long)p->addr;
464 reset_current_kprobe();
468 save_previous_kprobe(kcb);
469 set_current_kprobe(p, regs, kcb);
470 kprobes_inc_nmissed_count(p);
471 prepare_singlestep(p, regs);
472 kcb->kprobe_status = KPROBE_REENTER;
477 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
478 * remain disabled thorough out this function.
480 static int __kprobes kprobe_handler(struct pt_regs *regs)
484 kprobe_opcode_t *addr;
485 struct kprobe_ctlblk *kcb;
487 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
490 * We don't want to be preempted for the entire
491 * duration of kprobe processing
494 kcb = get_kprobe_ctlblk();
496 /* Check we're not actually recursing */
497 if (kprobe_running()) {
498 p = get_kprobe(addr);
500 ret = reenter_kprobe(p, regs, kcb);
501 if (kcb->kprobe_status == KPROBE_REENTER)
504 if (*addr != BREAKPOINT_INSTRUCTION) {
505 /* The breakpoint instruction was removed by
506 * another cpu right after we hit, no further
507 * handling of this interrupt is appropriate
509 regs->ip = (unsigned long)addr;
513 p = __get_cpu_var(current_kprobe);
514 if (p->break_handler && p->break_handler(p, regs))
520 p = get_kprobe(addr);
522 if (*addr != BREAKPOINT_INSTRUCTION) {
524 * The breakpoint instruction was removed right
525 * after we hit it. Another cpu has removed
526 * either a probepoint or a debugger breakpoint
527 * at this address. In either case, no further
528 * handling of this interrupt is appropriate.
529 * Back up over the (now missing) int3 and run
530 * the original instruction.
532 regs->ip = (unsigned long)addr;
535 /* Not one of ours: let kernel handle it */
539 set_current_kprobe(p, regs, kcb);
540 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
542 if (p->pre_handler && p->pre_handler(p, regs))
543 /* handler has already set things up, so skip ss setup */
547 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
548 if (p->ainsn.boostable == 1 && !p->post_handler) {
549 /* Boost up -- we can execute copied instructions directly */
550 reset_current_kprobe();
551 regs->ip = (unsigned long)p->ainsn.insn;
552 preempt_enable_no_resched();
556 prepare_singlestep(p, regs);
557 kcb->kprobe_status = KPROBE_HIT_SS;
561 preempt_enable_no_resched();
566 * When a retprobed function returns, this code saves registers and
567 * calls trampoline_handler() runs, which calls the kretprobe's handler.
569 void __kprobes kretprobe_trampoline_holder(void)
572 ".global kretprobe_trampoline\n"
573 "kretprobe_trampoline: \n"
575 /* We don't bother saving the ss register */
579 * Skip cs, ip, orig_ax.
580 * trampoline_handler() will plug in these values
599 " call trampoline_handler\n"
600 /* Replace saved sp with true return address. */
601 " movq %rax, 152(%rsp)\n"
617 /* Skip orig_ax, ip, cs */
623 * Skip cs, ip, orig_ax.
624 * trampoline_handler() will plug in these values
638 " call trampoline_handler\n"
639 /* Move flags to cs */
640 " movl 52(%esp), %edx\n"
641 " movl %edx, 48(%esp)\n"
642 /* Replace saved flags with true return address. */
643 " movl %eax, 52(%esp)\n"
651 /* Skip ip, orig_ax, es, ds, fs */
659 * Called from kretprobe_trampoline
661 void * __kprobes trampoline_handler(struct pt_regs *regs)
663 struct kretprobe_instance *ri = NULL;
664 struct hlist_head *head, empty_rp;
665 struct hlist_node *node, *tmp;
666 unsigned long flags, orig_ret_address = 0;
667 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
669 INIT_HLIST_HEAD(&empty_rp);
670 spin_lock_irqsave(&kretprobe_lock, flags);
671 head = kretprobe_inst_table_head(current);
672 /* fixup registers */
674 regs->cs = __KERNEL_CS;
676 regs->cs = __KERNEL_CS | get_kernel_rpl();
678 regs->ip = trampoline_address;
679 regs->orig_ax = ~0UL;
682 * It is possible to have multiple instances associated with a given
683 * task either because multiple functions in the call path have
684 * return probes installed on them, and/or more then one
685 * return probe was registered for a target function.
687 * We can handle this because:
688 * - instances are always pushed into the head of the list
689 * - when multiple return probes are registered for the same
690 * function, the (chronologically) first instance's ret_addr
691 * will be the real return address, and all the rest will
692 * point to kretprobe_trampoline.
694 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
695 if (ri->task != current)
696 /* another task is sharing our hash bucket */
699 if (ri->rp && ri->rp->handler) {
700 __get_cpu_var(current_kprobe) = &ri->rp->kp;
701 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
702 ri->rp->handler(ri, regs);
703 __get_cpu_var(current_kprobe) = NULL;
706 orig_ret_address = (unsigned long)ri->ret_addr;
707 recycle_rp_inst(ri, &empty_rp);
709 if (orig_ret_address != trampoline_address)
711 * This is the real return address. Any other
712 * instances associated with this task are for
713 * other calls deeper on the call stack
718 kretprobe_assert(ri, orig_ret_address, trampoline_address);
720 spin_unlock_irqrestore(&kretprobe_lock, flags);
722 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
723 hlist_del(&ri->hlist);
726 return (void *)orig_ret_address;
730 * Called after single-stepping. p->addr is the address of the
731 * instruction whose first byte has been replaced by the "int 3"
732 * instruction. To avoid the SMP problems that can occur when we
733 * temporarily put back the original opcode to single-step, we
734 * single-stepped a copy of the instruction. The address of this
735 * copy is p->ainsn.insn.
737 * This function prepares to return from the post-single-step
738 * interrupt. We have to fix up the stack as follows:
740 * 0) Except in the case of absolute or indirect jump or call instructions,
741 * the new ip is relative to the copied instruction. We need to make
742 * it relative to the original instruction.
744 * 1) If the single-stepped instruction was pushfl, then the TF and IF
745 * flags are set in the just-pushed flags, and may need to be cleared.
747 * 2) If the single-stepped instruction was a call, the return address
748 * that is atop the stack is the address following the copied instruction.
749 * We need to make it the address following the original instruction.
751 * If this is the first time we've single-stepped the instruction at
752 * this probepoint, and the instruction is boostable, boost it: add a
753 * jump instruction after the copied instruction, that jumps to the next
754 * instruction after the probepoint.
756 static void __kprobes resume_execution(struct kprobe *p,
757 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
759 unsigned long *tos = stack_addr(regs);
760 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
761 unsigned long orig_ip = (unsigned long)p->addr;
762 kprobe_opcode_t *insn = p->ainsn.insn;
764 /*skip the REX prefix*/
765 if (is_REX_prefix(insn))
768 regs->flags &= ~X86_EFLAGS_TF;
770 case 0x9c: /* pushfl */
771 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
772 *tos |= kcb->kprobe_old_flags;
774 case 0xc2: /* iret/ret/lret */
779 case 0xea: /* jmp absolute -- ip is correct */
780 /* ip is already adjusted, no more changes required */
781 p->ainsn.boostable = 1;
783 case 0xe8: /* call relative - Fix return addr */
784 *tos = orig_ip + (*tos - copy_ip);
787 case 0x9a: /* call absolute -- same as call absolute, indirect */
788 *tos = orig_ip + (*tos - copy_ip);
792 if ((insn[1] & 0x30) == 0x10) {
794 * call absolute, indirect
795 * Fix return addr; ip is correct.
796 * But this is not boostable
798 *tos = orig_ip + (*tos - copy_ip);
800 } else if (((insn[1] & 0x31) == 0x20) ||
801 ((insn[1] & 0x31) == 0x21)) {
803 * jmp near and far, absolute indirect
804 * ip is correct. And this is boostable
806 p->ainsn.boostable = 1;
813 if (p->ainsn.boostable == 0) {
814 if ((regs->ip > copy_ip) &&
815 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
817 * These instructions can be executed directly if it
818 * jumps back to correct address.
820 set_jmp_op((void *)regs->ip,
821 (void *)orig_ip + (regs->ip - copy_ip));
822 p->ainsn.boostable = 1;
824 p->ainsn.boostable = -1;
828 regs->ip += orig_ip - copy_ip;
835 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
836 * remain disabled thoroughout this function.
838 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
840 struct kprobe *cur = kprobe_running();
841 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
846 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
847 kcb->kprobe_status = KPROBE_HIT_SSDONE;
848 cur->post_handler(cur, regs, 0);
851 resume_execution(cur, regs, kcb);
852 regs->flags |= kcb->kprobe_saved_flags;
853 trace_hardirqs_fixup_flags(regs->flags);
855 /* Restore back the original saved kprobes variables and continue. */
856 if (kcb->kprobe_status == KPROBE_REENTER) {
857 restore_previous_kprobe(kcb);
860 reset_current_kprobe();
862 preempt_enable_no_resched();
865 * if somebody else is singlestepping across a probe point, flags
866 * will have TF set, in which case, continue the remaining processing
867 * of do_debug, as if this is not a probe hit.
869 if (regs->flags & X86_EFLAGS_TF)
875 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
877 struct kprobe *cur = kprobe_running();
878 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
880 switch (kcb->kprobe_status) {
884 * We are here because the instruction being single
885 * stepped caused a page fault. We reset the current
886 * kprobe and the ip points back to the probe address
887 * and allow the page fault handler to continue as a
890 regs->ip = (unsigned long)cur->addr;
891 regs->flags |= kcb->kprobe_old_flags;
892 if (kcb->kprobe_status == KPROBE_REENTER)
893 restore_previous_kprobe(kcb);
895 reset_current_kprobe();
896 preempt_enable_no_resched();
898 case KPROBE_HIT_ACTIVE:
899 case KPROBE_HIT_SSDONE:
901 * We increment the nmissed count for accounting,
902 * we can also use npre/npostfault count for accounting
903 * these specific fault cases.
905 kprobes_inc_nmissed_count(cur);
908 * We come here because instructions in the pre/post
909 * handler caused the page_fault, this could happen
910 * if handler tries to access user space by
911 * copy_from_user(), get_user() etc. Let the
912 * user-specified handler try to fix it first.
914 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
918 * In case the user-specified fault handler returned
919 * zero, try to fix up.
921 if (fixup_exception(regs))
925 * fixup routine could not handle it,
926 * Let do_page_fault() fix it.
936 * Wrapper routine for handling exceptions.
938 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
939 unsigned long val, void *data)
941 struct die_args *args = (struct die_args *)data;
942 int ret = NOTIFY_DONE;
944 if (args->regs && user_mode_vm(args->regs))
949 if (kprobe_handler(args->regs))
953 if (post_kprobe_handler(args->regs))
957 /* kprobe_running() needs smp_processor_id() */
959 if (kprobe_running() &&
960 kprobe_fault_handler(args->regs, args->trapnr))
970 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
972 struct jprobe *jp = container_of(p, struct jprobe, kp);
974 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
976 kcb->jprobe_saved_regs = *regs;
977 kcb->jprobe_saved_sp = stack_addr(regs);
978 addr = (unsigned long)(kcb->jprobe_saved_sp);
981 * As Linus pointed out, gcc assumes that the callee
982 * owns the argument space and could overwrite it, e.g.
983 * tailcall optimization. So, to be absolutely safe
984 * we also save and restore enough stack bytes to cover
987 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
988 MIN_STACK_SIZE(addr));
989 regs->flags &= ~X86_EFLAGS_IF;
990 trace_hardirqs_off();
991 regs->ip = (unsigned long)(jp->entry);
995 void __kprobes jprobe_return(void)
997 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1000 #ifdef CONFIG_X86_64
1001 " xchg %%rbx,%%rsp \n"
1003 " xchgl %%ebx,%%esp \n"
1006 " .globl jprobe_return_end\n"
1007 " jprobe_return_end: \n"
1009 (kcb->jprobe_saved_sp):"memory");
1012 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1014 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1015 u8 *addr = (u8 *) (regs->ip - 1);
1016 struct jprobe *jp = container_of(p, struct jprobe, kp);
1018 if ((addr > (u8 *) jprobe_return) &&
1019 (addr < (u8 *) jprobe_return_end)) {
1020 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1021 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1023 "current sp %p does not match saved sp %p\n",
1024 stack_addr(regs), kcb->jprobe_saved_sp);
1025 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1026 show_registers(saved_regs);
1027 printk(KERN_ERR "Current registers\n");
1028 show_registers(regs);
1031 *regs = kcb->jprobe_saved_regs;
1032 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1034 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1035 preempt_enable_no_resched();
1041 int __init arch_init_kprobes(void)
1046 int __kprobes arch_trampoline_kprobe(struct kprobe *p)