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>
52 #include <asm/cacheflush.h>
54 #include <asm/pgtable.h>
55 #include <asm/uaccess.h>
56 #include <asm/alternative.h>
58 void jprobe_return_end(void);
60 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
61 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
64 #define stack_addr(regs) ((unsigned long *)regs->sp)
67 * "®s->sp" looks wrong, but it's correct for x86_32. x86_32 CPUs
68 * don't save the ss and esp registers if the CPU is already in kernel
69 * mode when it traps. So for kprobes, regs->sp and regs->ss are not
70 * the [nonexistent] saved stack pointer and ss register, but rather
71 * the top 8 bytes of the pre-int3 stack. So ®s->sp happens to
72 * point to the top of the pre-int3 stack.
74 #define stack_addr(regs) ((unsigned long *)®s->sp)
77 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
78 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
79 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
80 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
81 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
84 * Undefined/reserved opcodes, conditional jump, Opcode Extension
85 * Groups, and some special opcodes can not boost.
87 static const 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, 0) , /* 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 */
109 static const u32 onebyte_has_modrm[256 / 32] = {
110 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
111 /* ----------------------------------------------- */
112 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
113 W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
114 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
115 W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
116 W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
117 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
118 W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
119 W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
120 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
121 W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
122 W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
123 W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
124 W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
125 W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
126 W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
127 W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) /* f0 */
128 /* ----------------------------------------------- */
129 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
131 static const u32 twobyte_has_modrm[256 / 32] = {
132 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
133 /* ----------------------------------------------- */
134 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
135 W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
136 W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
137 W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
138 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
139 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
140 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
141 W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
142 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
143 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
144 W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
145 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
146 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
147 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
148 W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
149 W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0) /* ff */
150 /* ----------------------------------------------- */
151 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
155 struct kretprobe_blackpoint kretprobe_blacklist[] = {
156 {"__switch_to", }, /* This function switches only current task, but
157 doesn't switch kernel stack.*/
158 {NULL, NULL} /* Terminator */
160 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
162 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
163 static void __kprobes set_jmp_op(void *from, void *to)
165 struct __arch_jmp_op {
168 } __attribute__((packed)) * jop;
169 jop = (struct __arch_jmp_op *)from;
170 jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
171 jop->op = RELATIVEJUMP_INSTRUCTION;
175 * Check for the REX prefix which can only exist on X86_64
176 * X86_32 always returns 0
178 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
181 if ((*insn & 0xf0) == 0x40)
188 * Returns non-zero if opcode is boostable.
189 * RIP relative instructions are adjusted at copying time in 64 bits mode
191 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
193 kprobe_opcode_t opcode;
194 kprobe_opcode_t *orig_opcodes = opcodes;
197 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
199 opcode = *(opcodes++);
201 /* 2nd-byte opcode */
202 if (opcode == 0x0f) {
203 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
205 return test_bit(*opcodes,
206 (unsigned long *)twobyte_is_boostable);
209 switch (opcode & 0xf0) {
212 goto retry; /* REX prefix is boostable */
215 if (0x63 < opcode && opcode < 0x67)
216 goto retry; /* prefixes */
217 /* can't boost Address-size override and bound */
218 return (opcode != 0x62 && opcode != 0x67);
220 return 0; /* can't boost conditional jump */
222 /* can't boost software-interruptions */
223 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
225 /* can boost AA* and XLAT */
226 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
228 /* can boost in/out and absolute jmps */
229 return ((opcode & 0x04) || opcode == 0xea);
231 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
232 goto retry; /* lock/rep(ne) prefix */
233 /* clear and set flags are boostable */
234 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
236 /* segment override prefixes are boostable */
237 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
238 goto retry; /* prefixes */
239 /* CS override prefix and call are not boostable */
240 return (opcode != 0x2e && opcode != 0x9a);
245 * Returns non-zero if opcode modifies the interrupt flag.
247 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
252 case 0xcf: /* iret/iretd */
253 case 0x9d: /* popf/popfd */
258 * on X86_64, 0x40-0x4f are REX prefixes so we need to look
259 * at the next byte instead.. but of course not recurse infinitely
261 if (is_REX_prefix(insn))
262 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.
272 * Only applicable to 64-bit x86.
274 static void __kprobes fix_riprel(struct kprobe *p)
277 u8 *insn = p->ainsn.insn;
281 /* Skip legacy instruction prefixes. */
301 /* Skip REX instruction prefix. */
302 if (is_REX_prefix(insn))
306 /* Two-byte opcode. */
308 need_modrm = test_bit(*insn,
309 (unsigned long *)twobyte_has_modrm);
311 /* One-byte opcode. */
312 need_modrm = test_bit(*insn,
313 (unsigned long *)onebyte_has_modrm);
317 if ((modrm & 0xc7) == 0x05) {
318 /* %rip+disp32 addressing mode */
319 /* Displacement follows ModRM byte. */
322 * The copied instruction uses the %rip-relative
323 * addressing mode. Adjust the displacement for the
324 * difference between the original location of this
325 * instruction and the location of the copy that will
326 * actually be run. The tricky bit here is making sure
327 * that the sign extension happens correctly in this
328 * calculation, since we need a signed 32-bit result to
329 * be sign-extended to 64 bits when it's added to the
330 * %rip value and yield the same 64-bit result that the
331 * sign-extension of the original signed 32-bit
332 * displacement would have given.
334 disp = (u8 *) p->addr + *((s32 *) insn) -
335 (u8 *) p->ainsn.insn;
336 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
337 *(s32 *)insn = (s32) disp;
343 static void __kprobes arch_copy_kprobe(struct kprobe *p)
345 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
349 if (can_boost(p->addr))
350 p->ainsn.boostable = 0;
352 p->ainsn.boostable = -1;
354 p->opcode = *p->addr;
357 int __kprobes arch_prepare_kprobe(struct kprobe *p)
359 /* insn: must be on special executable page on x86. */
360 p->ainsn.insn = get_insn_slot();
367 void __kprobes arch_arm_kprobe(struct kprobe *p)
369 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
372 void __kprobes arch_disarm_kprobe(struct kprobe *p)
374 text_poke(p->addr, &p->opcode, 1);
377 void __kprobes arch_remove_kprobe(struct kprobe *p)
379 mutex_lock(&kprobe_mutex);
380 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
381 mutex_unlock(&kprobe_mutex);
384 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
386 kcb->prev_kprobe.kp = kprobe_running();
387 kcb->prev_kprobe.status = kcb->kprobe_status;
388 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
389 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
392 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
394 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
395 kcb->kprobe_status = kcb->prev_kprobe.status;
396 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
397 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
400 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
401 struct kprobe_ctlblk *kcb)
403 __get_cpu_var(current_kprobe) = p;
404 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
405 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
406 if (is_IF_modifier(p->ainsn.insn))
407 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
410 static void __kprobes clear_btf(void)
412 if (test_thread_flag(TIF_DEBUGCTLMSR))
413 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
416 static void __kprobes restore_btf(void)
418 if (test_thread_flag(TIF_DEBUGCTLMSR))
419 wrmsrl(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr);
422 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
425 regs->flags |= X86_EFLAGS_TF;
426 regs->flags &= ~X86_EFLAGS_IF;
427 /* single step inline if the instruction is an int3 */
428 if (p->opcode == BREAKPOINT_INSTRUCTION)
429 regs->ip = (unsigned long)p->addr;
431 regs->ip = (unsigned long)p->ainsn.insn;
434 /* Called with kretprobe_lock held */
435 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
436 struct pt_regs *regs)
438 unsigned long *sara = stack_addr(regs);
440 ri->ret_addr = (kprobe_opcode_t *) *sara;
442 /* Replace the return addr with trampoline addr */
443 *sara = (unsigned long) &kretprobe_trampoline;
446 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
447 struct kprobe_ctlblk *kcb)
449 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
450 if (p->ainsn.boostable == 1 && !p->post_handler) {
451 /* Boost up -- we can execute copied instructions directly */
452 reset_current_kprobe();
453 regs->ip = (unsigned long)p->ainsn.insn;
454 preempt_enable_no_resched();
458 prepare_singlestep(p, regs);
459 kcb->kprobe_status = KPROBE_HIT_SS;
463 * We have reentered the kprobe_handler(), since another probe was hit while
464 * within the handler. We save the original kprobes variables and just single
465 * step on the instruction of the new probe without calling any user handlers.
467 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
468 struct kprobe_ctlblk *kcb)
470 switch (kcb->kprobe_status) {
471 case KPROBE_HIT_SSDONE:
473 /* TODO: Provide re-entrancy from post_kprobes_handler() and
474 * avoid exception stack corruption while single-stepping on
475 * the instruction of the new probe.
477 arch_disarm_kprobe(p);
478 regs->ip = (unsigned long)p->addr;
479 reset_current_kprobe();
480 preempt_enable_no_resched();
483 case KPROBE_HIT_ACTIVE:
484 save_previous_kprobe(kcb);
485 set_current_kprobe(p, regs, kcb);
486 kprobes_inc_nmissed_count(p);
487 prepare_singlestep(p, regs);
488 kcb->kprobe_status = KPROBE_REENTER;
491 if (p == kprobe_running()) {
492 regs->flags &= ~TF_MASK;
493 regs->flags |= kcb->kprobe_saved_flags;
496 /* A probe has been hit in the codepath leading up
497 * to, or just after, single-stepping of a probed
498 * instruction. This entire codepath should strictly
499 * reside in .kprobes.text section. Raise a warning
500 * to highlight this peculiar case.
504 /* impossible cases */
513 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
514 * remain disabled thorough out this function.
516 static int __kprobes kprobe_handler(struct pt_regs *regs)
518 kprobe_opcode_t *addr;
520 struct kprobe_ctlblk *kcb;
522 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
523 if (*addr != BREAKPOINT_INSTRUCTION) {
525 * The breakpoint instruction was removed right
526 * after we hit it. Another cpu has removed
527 * either a probepoint or a debugger breakpoint
528 * at this address. In either case, no further
529 * handling of this interrupt is appropriate.
530 * Back up over the (now missing) int3 and run
531 * the original instruction.
533 regs->ip = (unsigned long)addr;
538 * We don't want to be preempted for the entire
539 * duration of kprobe processing. We conditionally
540 * re-enable preemption at the end of this function,
541 * and also in reenter_kprobe() and setup_singlestep().
545 kcb = get_kprobe_ctlblk();
546 p = get_kprobe(addr);
549 if (kprobe_running()) {
550 if (reenter_kprobe(p, regs, kcb))
553 set_current_kprobe(p, regs, kcb);
554 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
557 * If we have no pre-handler or it returned 0, we
558 * continue with normal processing. If we have a
559 * pre-handler and it returned non-zero, it prepped
560 * for calling the break_handler below on re-entry
561 * for jprobe processing, so get out doing nothing
564 if (!p->pre_handler || !p->pre_handler(p, regs))
565 setup_singlestep(p, regs, kcb);
568 } else if (kprobe_running()) {
569 p = __get_cpu_var(current_kprobe);
570 if (p->break_handler && p->break_handler(p, regs)) {
571 setup_singlestep(p, regs, kcb);
574 } /* else: not a kprobe fault; let the kernel handle it */
576 preempt_enable_no_resched();
581 * When a retprobed function returns, this code saves registers and
582 * calls trampoline_handler() runs, which calls the kretprobe's handler.
584 static void __used __kprobes kretprobe_trampoline_holder(void)
587 ".global kretprobe_trampoline\n"
588 "kretprobe_trampoline: \n"
590 /* We don't bother saving the ss register */
594 * Skip cs, ip, orig_ax.
595 * trampoline_handler() will plug in these values
614 " call trampoline_handler\n"
615 /* Replace saved sp with true return address. */
616 " movq %rax, 152(%rsp)\n"
632 /* Skip orig_ax, ip, cs */
638 * Skip cs, ip, orig_ax.
639 * trampoline_handler() will plug in these values
653 " call trampoline_handler\n"
654 /* Move flags to cs */
655 " movl 52(%esp), %edx\n"
656 " movl %edx, 48(%esp)\n"
657 /* Replace saved flags with true return address. */
658 " movl %eax, 52(%esp)\n"
666 /* Skip ip, orig_ax, es, ds, fs */
674 * Called from kretprobe_trampoline
676 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
678 struct kretprobe_instance *ri = NULL;
679 struct hlist_head *head, empty_rp;
680 struct hlist_node *node, *tmp;
681 unsigned long flags, orig_ret_address = 0;
682 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
684 INIT_HLIST_HEAD(&empty_rp);
685 spin_lock_irqsave(&kretprobe_lock, flags);
686 head = kretprobe_inst_table_head(current);
687 /* fixup registers */
689 regs->cs = __KERNEL_CS;
691 regs->cs = __KERNEL_CS | get_kernel_rpl();
693 regs->ip = trampoline_address;
694 regs->orig_ax = ~0UL;
697 * It is possible to have multiple instances associated with a given
698 * task either because multiple functions in the call path have
699 * return probes installed on them, and/or more then one
700 * return probe was registered for a target function.
702 * We can handle this because:
703 * - instances are always pushed into the head of the list
704 * - when multiple return probes are registered for the same
705 * function, the (chronologically) first instance's ret_addr
706 * will be the real return address, and all the rest will
707 * point to kretprobe_trampoline.
709 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
710 if (ri->task != current)
711 /* another task is sharing our hash bucket */
714 if (ri->rp && ri->rp->handler) {
715 __get_cpu_var(current_kprobe) = &ri->rp->kp;
716 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
717 ri->rp->handler(ri, regs);
718 __get_cpu_var(current_kprobe) = NULL;
721 orig_ret_address = (unsigned long)ri->ret_addr;
722 recycle_rp_inst(ri, &empty_rp);
724 if (orig_ret_address != trampoline_address)
726 * This is the real return address. Any other
727 * instances associated with this task are for
728 * other calls deeper on the call stack
733 kretprobe_assert(ri, orig_ret_address, trampoline_address);
735 spin_unlock_irqrestore(&kretprobe_lock, flags);
737 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
738 hlist_del(&ri->hlist);
741 return (void *)orig_ret_address;
745 * Called after single-stepping. p->addr is the address of the
746 * instruction whose first byte has been replaced by the "int 3"
747 * instruction. To avoid the SMP problems that can occur when we
748 * temporarily put back the original opcode to single-step, we
749 * single-stepped a copy of the instruction. The address of this
750 * copy is p->ainsn.insn.
752 * This function prepares to return from the post-single-step
753 * interrupt. We have to fix up the stack as follows:
755 * 0) Except in the case of absolute or indirect jump or call instructions,
756 * the new ip is relative to the copied instruction. We need to make
757 * it relative to the original instruction.
759 * 1) If the single-stepped instruction was pushfl, then the TF and IF
760 * flags are set in the just-pushed flags, and may need to be cleared.
762 * 2) If the single-stepped instruction was a call, the return address
763 * that is atop the stack is the address following the copied instruction.
764 * We need to make it the address following the original instruction.
766 * If this is the first time we've single-stepped the instruction at
767 * this probepoint, and the instruction is boostable, boost it: add a
768 * jump instruction after the copied instruction, that jumps to the next
769 * instruction after the probepoint.
771 static void __kprobes resume_execution(struct kprobe *p,
772 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
774 unsigned long *tos = stack_addr(regs);
775 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
776 unsigned long orig_ip = (unsigned long)p->addr;
777 kprobe_opcode_t *insn = p->ainsn.insn;
779 /*skip the REX prefix*/
780 if (is_REX_prefix(insn))
783 regs->flags &= ~X86_EFLAGS_TF;
785 case 0x9c: /* pushfl */
786 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
787 *tos |= kcb->kprobe_old_flags;
789 case 0xc2: /* iret/ret/lret */
794 case 0xea: /* jmp absolute -- ip is correct */
795 /* ip is already adjusted, no more changes required */
796 p->ainsn.boostable = 1;
798 case 0xe8: /* call relative - Fix return addr */
799 *tos = orig_ip + (*tos - copy_ip);
802 case 0x9a: /* call absolute -- same as call absolute, indirect */
803 *tos = orig_ip + (*tos - copy_ip);
807 if ((insn[1] & 0x30) == 0x10) {
809 * call absolute, indirect
810 * Fix return addr; ip is correct.
811 * But this is not boostable
813 *tos = orig_ip + (*tos - copy_ip);
815 } else if (((insn[1] & 0x31) == 0x20) ||
816 ((insn[1] & 0x31) == 0x21)) {
818 * jmp near and far, absolute indirect
819 * ip is correct. And this is boostable
821 p->ainsn.boostable = 1;
828 if (p->ainsn.boostable == 0) {
829 if ((regs->ip > copy_ip) &&
830 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
832 * These instructions can be executed directly if it
833 * jumps back to correct address.
835 set_jmp_op((void *)regs->ip,
836 (void *)orig_ip + (regs->ip - copy_ip));
837 p->ainsn.boostable = 1;
839 p->ainsn.boostable = -1;
843 regs->ip += orig_ip - copy_ip;
850 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
851 * remain disabled thoroughout this function.
853 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
855 struct kprobe *cur = kprobe_running();
856 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
861 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
862 kcb->kprobe_status = KPROBE_HIT_SSDONE;
863 cur->post_handler(cur, regs, 0);
866 resume_execution(cur, regs, kcb);
867 regs->flags |= kcb->kprobe_saved_flags;
868 trace_hardirqs_fixup_flags(regs->flags);
870 /* Restore back the original saved kprobes variables and continue. */
871 if (kcb->kprobe_status == KPROBE_REENTER) {
872 restore_previous_kprobe(kcb);
875 reset_current_kprobe();
877 preempt_enable_no_resched();
880 * if somebody else is singlestepping across a probe point, flags
881 * will have TF set, in which case, continue the remaining processing
882 * of do_debug, as if this is not a probe hit.
884 if (regs->flags & X86_EFLAGS_TF)
890 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
892 struct kprobe *cur = kprobe_running();
893 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
895 switch (kcb->kprobe_status) {
899 * We are here because the instruction being single
900 * stepped caused a page fault. We reset the current
901 * kprobe and the ip points back to the probe address
902 * and allow the page fault handler to continue as a
905 regs->ip = (unsigned long)cur->addr;
906 regs->flags |= kcb->kprobe_old_flags;
907 if (kcb->kprobe_status == KPROBE_REENTER)
908 restore_previous_kprobe(kcb);
910 reset_current_kprobe();
911 preempt_enable_no_resched();
913 case KPROBE_HIT_ACTIVE:
914 case KPROBE_HIT_SSDONE:
916 * We increment the nmissed count for accounting,
917 * we can also use npre/npostfault count for accounting
918 * these specific fault cases.
920 kprobes_inc_nmissed_count(cur);
923 * We come here because instructions in the pre/post
924 * handler caused the page_fault, this could happen
925 * if handler tries to access user space by
926 * copy_from_user(), get_user() etc. Let the
927 * user-specified handler try to fix it first.
929 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
933 * In case the user-specified fault handler returned
934 * zero, try to fix up.
936 if (fixup_exception(regs))
940 * fixup routine could not handle it,
941 * Let do_page_fault() fix it.
951 * Wrapper routine for handling exceptions.
953 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
954 unsigned long val, void *data)
956 struct die_args *args = data;
957 int ret = NOTIFY_DONE;
959 if (args->regs && user_mode_vm(args->regs))
964 if (kprobe_handler(args->regs))
968 if (post_kprobe_handler(args->regs))
973 * To be potentially processing a kprobe fault and to
974 * trust the result from kprobe_running(), we have
975 * be non-preemptible.
977 if (!preemptible() && kprobe_running() &&
978 kprobe_fault_handler(args->regs, args->trapnr))
987 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
989 struct jprobe *jp = container_of(p, struct jprobe, kp);
991 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
993 kcb->jprobe_saved_regs = *regs;
994 kcb->jprobe_saved_sp = stack_addr(regs);
995 addr = (unsigned long)(kcb->jprobe_saved_sp);
998 * As Linus pointed out, gcc assumes that the callee
999 * owns the argument space and could overwrite it, e.g.
1000 * tailcall optimization. So, to be absolutely safe
1001 * we also save and restore enough stack bytes to cover
1002 * the argument area.
1004 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1005 MIN_STACK_SIZE(addr));
1006 regs->flags &= ~X86_EFLAGS_IF;
1007 trace_hardirqs_off();
1008 regs->ip = (unsigned long)(jp->entry);
1012 void __kprobes jprobe_return(void)
1014 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1017 #ifdef CONFIG_X86_64
1018 " xchg %%rbx,%%rsp \n"
1020 " xchgl %%ebx,%%esp \n"
1023 " .globl jprobe_return_end\n"
1024 " jprobe_return_end: \n"
1026 (kcb->jprobe_saved_sp):"memory");
1029 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1031 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1032 u8 *addr = (u8 *) (regs->ip - 1);
1033 struct jprobe *jp = container_of(p, struct jprobe, kp);
1035 if ((addr > (u8 *) jprobe_return) &&
1036 (addr < (u8 *) jprobe_return_end)) {
1037 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1038 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1040 "current sp %p does not match saved sp %p\n",
1041 stack_addr(regs), kcb->jprobe_saved_sp);
1042 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1043 show_registers(saved_regs);
1044 printk(KERN_ERR "Current registers\n");
1045 show_registers(regs);
1048 *regs = kcb->jprobe_saved_regs;
1049 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1051 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1052 preempt_enable_no_resched();
1058 int __init arch_init_kprobes(void)
1063 int __kprobes arch_trampoline_kprobe(struct kprobe *p)