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 */
111 static const u32 onebyte_has_modrm[256 / 32] = {
112 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
113 /* ----------------------------------------------- */
114 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
115 W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
116 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
117 W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
118 W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
119 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
120 W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
121 W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
122 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
123 W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
124 W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
125 W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
126 W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
127 W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
128 W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
129 W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) /* f0 */
130 /* ----------------------------------------------- */
131 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
133 static const u32 twobyte_has_modrm[256 / 32] = {
134 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
135 /* ----------------------------------------------- */
136 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
137 W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
138 W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
139 W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
140 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
141 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
142 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
143 W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
144 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
145 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
146 W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
147 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
148 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
149 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
150 W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
151 W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0) /* ff */
152 /* ----------------------------------------------- */
153 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
157 struct kretprobe_blackpoint kretprobe_blacklist[] = {
158 {"__switch_to", }, /* This function switches only current task, but
159 doesn't switch kernel stack.*/
160 {NULL, NULL} /* Terminator */
162 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
164 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
165 static void __kprobes set_jmp_op(void *from, void *to)
167 struct __arch_jmp_op {
170 } __attribute__((packed)) * jop;
171 jop = (struct __arch_jmp_op *)from;
172 jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
173 jop->op = RELATIVEJUMP_INSTRUCTION;
177 * Check for the REX prefix which can only exist on X86_64
178 * X86_32 always returns 0
180 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
183 if ((*insn & 0xf0) == 0x40)
190 * Returns non-zero if opcode is boostable.
191 * RIP relative instructions are adjusted at copying time in 64 bits mode
193 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
195 kprobe_opcode_t opcode;
196 kprobe_opcode_t *orig_opcodes = opcodes;
198 if (search_exception_tables((unsigned long)opcodes))
199 return 0; /* Page fault may occur on this address. */
202 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
204 opcode = *(opcodes++);
206 /* 2nd-byte opcode */
207 if (opcode == 0x0f) {
208 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
210 return test_bit(*opcodes,
211 (unsigned long *)twobyte_is_boostable);
214 switch (opcode & 0xf0) {
217 goto retry; /* REX prefix is boostable */
220 if (0x63 < opcode && opcode < 0x67)
221 goto retry; /* prefixes */
222 /* can't boost Address-size override and bound */
223 return (opcode != 0x62 && opcode != 0x67);
225 return 0; /* can't boost conditional jump */
227 /* can't boost software-interruptions */
228 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
230 /* can boost AA* and XLAT */
231 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
233 /* can boost in/out and absolute jmps */
234 return ((opcode & 0x04) || opcode == 0xea);
236 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
237 goto retry; /* lock/rep(ne) prefix */
238 /* clear and set flags are boostable */
239 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
241 /* segment override prefixes are boostable */
242 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
243 goto retry; /* prefixes */
244 /* CS override prefix and call are not boostable */
245 return (opcode != 0x2e && opcode != 0x9a);
249 /* Recover the probed instruction at addr for further analysis. */
250 static int recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
253 kp = get_kprobe((void *)addr);
258 * Basically, kp->ainsn.insn has an original instruction.
259 * However, RIP-relative instruction can not do single-stepping
260 * at different place, fix_riprel() tweaks the displacement of
261 * that instruction. In that case, we can't recover the instruction
262 * from the kp->ainsn.insn.
264 * On the other hand, kp->opcode has a copy of the first byte of
265 * the probed instruction, which is overwritten by int3. And
266 * the instruction at kp->addr is not modified by kprobes except
267 * for the first byte, we can recover the original instruction
268 * from it and kp->opcode.
270 memcpy(buf, kp->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
275 /* Dummy buffers for kallsyms_lookup */
276 static char __dummy_buf[KSYM_NAME_LEN];
278 /* Check if paddr is at an instruction boundary */
279 static int __kprobes can_probe(unsigned long paddr)
282 unsigned long addr, offset = 0;
284 kprobe_opcode_t buf[MAX_INSN_SIZE];
286 if (!kallsyms_lookup(paddr, NULL, &offset, NULL, __dummy_buf))
289 /* Decode instructions */
290 addr = paddr - offset;
291 while (addr < paddr) {
292 kernel_insn_init(&insn, (void *)addr);
293 insn_get_opcode(&insn);
296 * Check if the instruction has been modified by another
297 * kprobe, in which case we replace the breakpoint by the
298 * original instruction in our buffer.
300 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
301 ret = recover_probed_instruction(buf, addr);
304 * Another debugging subsystem might insert
305 * this breakpoint. In that case, we can't
309 kernel_insn_init(&insn, buf);
311 insn_get_length(&insn);
315 return (addr == paddr);
319 * Returns non-zero if opcode modifies the interrupt flag.
321 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
326 case 0xcf: /* iret/iretd */
327 case 0x9d: /* popf/popfd */
332 * on X86_64, 0x40-0x4f are REX prefixes so we need to look
333 * at the next byte instead.. but of course not recurse infinitely
335 if (is_REX_prefix(insn))
336 return is_IF_modifier(++insn);
342 * Adjust the displacement if the instruction uses the %rip-relative
344 * If it does, Return the address of the 32-bit displacement word.
345 * If not, return null.
346 * Only applicable to 64-bit x86.
348 static void __kprobes fix_riprel(struct kprobe *p)
351 u8 *insn = p->ainsn.insn;
355 /* Skip legacy instruction prefixes. */
375 /* Skip REX instruction prefix. */
376 if (is_REX_prefix(insn))
380 /* Two-byte opcode. */
382 need_modrm = test_bit(*insn,
383 (unsigned long *)twobyte_has_modrm);
385 /* One-byte opcode. */
386 need_modrm = test_bit(*insn,
387 (unsigned long *)onebyte_has_modrm);
391 if ((modrm & 0xc7) == 0x05) {
392 /* %rip+disp32 addressing mode */
393 /* Displacement follows ModRM byte. */
396 * The copied instruction uses the %rip-relative
397 * addressing mode. Adjust the displacement for the
398 * difference between the original location of this
399 * instruction and the location of the copy that will
400 * actually be run. The tricky bit here is making sure
401 * that the sign extension happens correctly in this
402 * calculation, since we need a signed 32-bit result to
403 * be sign-extended to 64 bits when it's added to the
404 * %rip value and yield the same 64-bit result that the
405 * sign-extension of the original signed 32-bit
406 * displacement would have given.
408 disp = (u8 *) p->addr + *((s32 *) insn) -
409 (u8 *) p->ainsn.insn;
410 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
411 *(s32 *)insn = (s32) disp;
417 static void __kprobes arch_copy_kprobe(struct kprobe *p)
419 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
423 if (can_boost(p->addr))
424 p->ainsn.boostable = 0;
426 p->ainsn.boostable = -1;
428 p->opcode = *p->addr;
431 int __kprobes arch_prepare_kprobe(struct kprobe *p)
433 if (!can_probe((unsigned long)p->addr))
435 /* insn: must be on special executable page on x86. */
436 p->ainsn.insn = get_insn_slot();
443 void __kprobes arch_arm_kprobe(struct kprobe *p)
445 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
448 void __kprobes arch_disarm_kprobe(struct kprobe *p)
450 text_poke(p->addr, &p->opcode, 1);
453 void __kprobes arch_remove_kprobe(struct kprobe *p)
456 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
457 p->ainsn.insn = NULL;
461 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
463 kcb->prev_kprobe.kp = kprobe_running();
464 kcb->prev_kprobe.status = kcb->kprobe_status;
465 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
466 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
469 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
471 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
472 kcb->kprobe_status = kcb->prev_kprobe.status;
473 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
474 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
477 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
478 struct kprobe_ctlblk *kcb)
480 __get_cpu_var(current_kprobe) = p;
481 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
482 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
483 if (is_IF_modifier(p->ainsn.insn))
484 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
487 static void __kprobes clear_btf(void)
489 if (test_thread_flag(TIF_DEBUGCTLMSR))
490 update_debugctlmsr(0);
493 static void __kprobes restore_btf(void)
495 if (test_thread_flag(TIF_DEBUGCTLMSR))
496 update_debugctlmsr(current->thread.debugctlmsr);
499 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
502 regs->flags |= X86_EFLAGS_TF;
503 regs->flags &= ~X86_EFLAGS_IF;
504 /* single step inline if the instruction is an int3 */
505 if (p->opcode == BREAKPOINT_INSTRUCTION)
506 regs->ip = (unsigned long)p->addr;
508 regs->ip = (unsigned long)p->ainsn.insn;
511 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
512 struct pt_regs *regs)
514 unsigned long *sara = stack_addr(regs);
516 ri->ret_addr = (kprobe_opcode_t *) *sara;
518 /* Replace the return addr with trampoline addr */
519 *sara = (unsigned long) &kretprobe_trampoline;
522 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
523 struct kprobe_ctlblk *kcb)
525 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
526 if (p->ainsn.boostable == 1 && !p->post_handler) {
527 /* Boost up -- we can execute copied instructions directly */
528 reset_current_kprobe();
529 regs->ip = (unsigned long)p->ainsn.insn;
530 preempt_enable_no_resched();
534 prepare_singlestep(p, regs);
535 kcb->kprobe_status = KPROBE_HIT_SS;
539 * We have reentered the kprobe_handler(), since another probe was hit while
540 * within the handler. We save the original kprobes variables and just single
541 * step on the instruction of the new probe without calling any user handlers.
543 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
544 struct kprobe_ctlblk *kcb)
546 switch (kcb->kprobe_status) {
547 case KPROBE_HIT_SSDONE:
549 /* TODO: Provide re-entrancy from post_kprobes_handler() and
550 * avoid exception stack corruption while single-stepping on
551 * the instruction of the new probe.
553 arch_disarm_kprobe(p);
554 regs->ip = (unsigned long)p->addr;
555 reset_current_kprobe();
556 preempt_enable_no_resched();
559 case KPROBE_HIT_ACTIVE:
560 save_previous_kprobe(kcb);
561 set_current_kprobe(p, regs, kcb);
562 kprobes_inc_nmissed_count(p);
563 prepare_singlestep(p, regs);
564 kcb->kprobe_status = KPROBE_REENTER;
567 if (p == kprobe_running()) {
568 regs->flags &= ~X86_EFLAGS_TF;
569 regs->flags |= kcb->kprobe_saved_flags;
572 /* A probe has been hit in the codepath leading up
573 * to, or just after, single-stepping of a probed
574 * instruction. This entire codepath should strictly
575 * reside in .kprobes.text section. Raise a warning
576 * to highlight this peculiar case.
580 /* impossible cases */
589 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
590 * remain disabled thorough out this function.
592 static int __kprobes kprobe_handler(struct pt_regs *regs)
594 kprobe_opcode_t *addr;
596 struct kprobe_ctlblk *kcb;
598 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
599 if (*addr != BREAKPOINT_INSTRUCTION) {
601 * The breakpoint instruction was removed right
602 * after we hit it. Another cpu has removed
603 * either a probepoint or a debugger breakpoint
604 * at this address. In either case, no further
605 * handling of this interrupt is appropriate.
606 * Back up over the (now missing) int3 and run
607 * the original instruction.
609 regs->ip = (unsigned long)addr;
614 * We don't want to be preempted for the entire
615 * duration of kprobe processing. We conditionally
616 * re-enable preemption at the end of this function,
617 * and also in reenter_kprobe() and setup_singlestep().
621 kcb = get_kprobe_ctlblk();
622 p = get_kprobe(addr);
625 if (kprobe_running()) {
626 if (reenter_kprobe(p, regs, kcb))
629 set_current_kprobe(p, regs, kcb);
630 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
633 * If we have no pre-handler or it returned 0, we
634 * continue with normal processing. If we have a
635 * pre-handler and it returned non-zero, it prepped
636 * for calling the break_handler below on re-entry
637 * for jprobe processing, so get out doing nothing
640 if (!p->pre_handler || !p->pre_handler(p, regs))
641 setup_singlestep(p, regs, kcb);
644 } else if (kprobe_running()) {
645 p = __get_cpu_var(current_kprobe);
646 if (p->break_handler && p->break_handler(p, regs)) {
647 setup_singlestep(p, regs, kcb);
650 } /* else: not a kprobe fault; let the kernel handle it */
652 preempt_enable_no_resched();
657 * When a retprobed function returns, this code saves registers and
658 * calls trampoline_handler() runs, which calls the kretprobe's handler.
660 static void __used __kprobes kretprobe_trampoline_holder(void)
663 ".global kretprobe_trampoline\n"
664 "kretprobe_trampoline: \n"
666 /* We don't bother saving the ss register */
670 * Skip cs, ip, orig_ax.
671 * trampoline_handler() will plug in these values
690 " call trampoline_handler\n"
691 /* Replace saved sp with true return address. */
692 " movq %rax, 152(%rsp)\n"
708 /* Skip orig_ax, ip, cs */
714 * Skip cs, ip, orig_ax and gs.
715 * trampoline_handler() will plug in these values
729 " call trampoline_handler\n"
730 /* Move flags to cs */
731 " movl 56(%esp), %edx\n"
732 " movl %edx, 52(%esp)\n"
733 /* Replace saved flags with true return address. */
734 " movl %eax, 56(%esp)\n"
742 /* Skip ds, es, fs, gs, orig_ax and ip */
750 * Called from kretprobe_trampoline
752 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
754 struct kretprobe_instance *ri = NULL;
755 struct hlist_head *head, empty_rp;
756 struct hlist_node *node, *tmp;
757 unsigned long flags, orig_ret_address = 0;
758 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
760 INIT_HLIST_HEAD(&empty_rp);
761 kretprobe_hash_lock(current, &head, &flags);
762 /* fixup registers */
764 regs->cs = __KERNEL_CS;
766 regs->cs = __KERNEL_CS | get_kernel_rpl();
769 regs->ip = trampoline_address;
770 regs->orig_ax = ~0UL;
773 * It is possible to have multiple instances associated with a given
774 * task either because multiple functions in the call path have
775 * return probes installed on them, and/or more than one
776 * return probe was registered for a target function.
778 * We can handle this because:
779 * - instances are always pushed into the head of the list
780 * - when multiple return probes are registered for the same
781 * function, the (chronologically) first instance's ret_addr
782 * will be the real return address, and all the rest will
783 * point to kretprobe_trampoline.
785 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
786 if (ri->task != current)
787 /* another task is sharing our hash bucket */
790 if (ri->rp && ri->rp->handler) {
791 __get_cpu_var(current_kprobe) = &ri->rp->kp;
792 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
793 ri->rp->handler(ri, regs);
794 __get_cpu_var(current_kprobe) = NULL;
797 orig_ret_address = (unsigned long)ri->ret_addr;
798 recycle_rp_inst(ri, &empty_rp);
800 if (orig_ret_address != trampoline_address)
802 * This is the real return address. Any other
803 * instances associated with this task are for
804 * other calls deeper on the call stack
809 kretprobe_assert(ri, orig_ret_address, trampoline_address);
811 kretprobe_hash_unlock(current, &flags);
813 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
814 hlist_del(&ri->hlist);
817 return (void *)orig_ret_address;
821 * Called after single-stepping. p->addr is the address of the
822 * instruction whose first byte has been replaced by the "int 3"
823 * instruction. To avoid the SMP problems that can occur when we
824 * temporarily put back the original opcode to single-step, we
825 * single-stepped a copy of the instruction. The address of this
826 * copy is p->ainsn.insn.
828 * This function prepares to return from the post-single-step
829 * interrupt. We have to fix up the stack as follows:
831 * 0) Except in the case of absolute or indirect jump or call instructions,
832 * the new ip is relative to the copied instruction. We need to make
833 * it relative to the original instruction.
835 * 1) If the single-stepped instruction was pushfl, then the TF and IF
836 * flags are set in the just-pushed flags, and may need to be cleared.
838 * 2) If the single-stepped instruction was a call, the return address
839 * that is atop the stack is the address following the copied instruction.
840 * We need to make it the address following the original instruction.
842 * If this is the first time we've single-stepped the instruction at
843 * this probepoint, and the instruction is boostable, boost it: add a
844 * jump instruction after the copied instruction, that jumps to the next
845 * instruction after the probepoint.
847 static void __kprobes resume_execution(struct kprobe *p,
848 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
850 unsigned long *tos = stack_addr(regs);
851 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
852 unsigned long orig_ip = (unsigned long)p->addr;
853 kprobe_opcode_t *insn = p->ainsn.insn;
855 /*skip the REX prefix*/
856 if (is_REX_prefix(insn))
859 regs->flags &= ~X86_EFLAGS_TF;
861 case 0x9c: /* pushfl */
862 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
863 *tos |= kcb->kprobe_old_flags;
865 case 0xc2: /* iret/ret/lret */
870 case 0xea: /* jmp absolute -- ip is correct */
871 /* ip is already adjusted, no more changes required */
872 p->ainsn.boostable = 1;
874 case 0xe8: /* call relative - Fix return addr */
875 *tos = orig_ip + (*tos - copy_ip);
878 case 0x9a: /* call absolute -- same as call absolute, indirect */
879 *tos = orig_ip + (*tos - copy_ip);
883 if ((insn[1] & 0x30) == 0x10) {
885 * call absolute, indirect
886 * Fix return addr; ip is correct.
887 * But this is not boostable
889 *tos = orig_ip + (*tos - copy_ip);
891 } else if (((insn[1] & 0x31) == 0x20) ||
892 ((insn[1] & 0x31) == 0x21)) {
894 * jmp near and far, absolute indirect
895 * ip is correct. And this is boostable
897 p->ainsn.boostable = 1;
904 if (p->ainsn.boostable == 0) {
905 if ((regs->ip > copy_ip) &&
906 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
908 * These instructions can be executed directly if it
909 * jumps back to correct address.
911 set_jmp_op((void *)regs->ip,
912 (void *)orig_ip + (regs->ip - copy_ip));
913 p->ainsn.boostable = 1;
915 p->ainsn.boostable = -1;
919 regs->ip += orig_ip - copy_ip;
926 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
927 * remain disabled thoroughout this function.
929 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
931 struct kprobe *cur = kprobe_running();
932 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
937 resume_execution(cur, regs, kcb);
938 regs->flags |= kcb->kprobe_saved_flags;
940 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
941 kcb->kprobe_status = KPROBE_HIT_SSDONE;
942 cur->post_handler(cur, regs, 0);
945 /* Restore back the original saved kprobes variables and continue. */
946 if (kcb->kprobe_status == KPROBE_REENTER) {
947 restore_previous_kprobe(kcb);
950 reset_current_kprobe();
952 preempt_enable_no_resched();
955 * if somebody else is singlestepping across a probe point, flags
956 * will have TF set, in which case, continue the remaining processing
957 * of do_debug, as if this is not a probe hit.
959 if (regs->flags & X86_EFLAGS_TF)
965 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
967 struct kprobe *cur = kprobe_running();
968 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
970 switch (kcb->kprobe_status) {
974 * We are here because the instruction being single
975 * stepped caused a page fault. We reset the current
976 * kprobe and the ip points back to the probe address
977 * and allow the page fault handler to continue as a
980 regs->ip = (unsigned long)cur->addr;
981 regs->flags |= kcb->kprobe_old_flags;
982 if (kcb->kprobe_status == KPROBE_REENTER)
983 restore_previous_kprobe(kcb);
985 reset_current_kprobe();
986 preempt_enable_no_resched();
988 case KPROBE_HIT_ACTIVE:
989 case KPROBE_HIT_SSDONE:
991 * We increment the nmissed count for accounting,
992 * we can also use npre/npostfault count for accounting
993 * these specific fault cases.
995 kprobes_inc_nmissed_count(cur);
998 * We come here because instructions in the pre/post
999 * handler caused the page_fault, this could happen
1000 * if handler tries to access user space by
1001 * copy_from_user(), get_user() etc. Let the
1002 * user-specified handler try to fix it first.
1004 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
1008 * In case the user-specified fault handler returned
1009 * zero, try to fix up.
1011 if (fixup_exception(regs))
1015 * fixup routine could not handle it,
1016 * Let do_page_fault() fix it.
1026 * Wrapper routine for handling exceptions.
1028 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
1029 unsigned long val, void *data)
1031 struct die_args *args = data;
1032 int ret = NOTIFY_DONE;
1034 if (args->regs && user_mode_vm(args->regs))
1039 if (kprobe_handler(args->regs))
1043 if (post_kprobe_handler(args->regs))
1048 * To be potentially processing a kprobe fault and to
1049 * trust the result from kprobe_running(), we have
1050 * be non-preemptible.
1052 if (!preemptible() && kprobe_running() &&
1053 kprobe_fault_handler(args->regs, args->trapnr))
1062 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
1064 struct jprobe *jp = container_of(p, struct jprobe, kp);
1066 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1068 kcb->jprobe_saved_regs = *regs;
1069 kcb->jprobe_saved_sp = stack_addr(regs);
1070 addr = (unsigned long)(kcb->jprobe_saved_sp);
1073 * As Linus pointed out, gcc assumes that the callee
1074 * owns the argument space and could overwrite it, e.g.
1075 * tailcall optimization. So, to be absolutely safe
1076 * we also save and restore enough stack bytes to cover
1077 * the argument area.
1079 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1080 MIN_STACK_SIZE(addr));
1081 regs->flags &= ~X86_EFLAGS_IF;
1082 trace_hardirqs_off();
1083 regs->ip = (unsigned long)(jp->entry);
1087 void __kprobes jprobe_return(void)
1089 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1092 #ifdef CONFIG_X86_64
1093 " xchg %%rbx,%%rsp \n"
1095 " xchgl %%ebx,%%esp \n"
1098 " .globl jprobe_return_end\n"
1099 " jprobe_return_end: \n"
1101 (kcb->jprobe_saved_sp):"memory");
1104 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1106 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1107 u8 *addr = (u8 *) (regs->ip - 1);
1108 struct jprobe *jp = container_of(p, struct jprobe, kp);
1110 if ((addr > (u8 *) jprobe_return) &&
1111 (addr < (u8 *) jprobe_return_end)) {
1112 if (stack_addr(regs) != kcb->jprobe_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), kcb->jprobe_saved_sp);
1117 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1118 show_registers(saved_regs);
1119 printk(KERN_ERR "Current registers\n");
1120 show_registers(regs);
1123 *regs = kcb->jprobe_saved_regs;
1124 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1126 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1127 preempt_enable_no_resched();
1133 int __init arch_init_kprobes(void)
1138 int __kprobes arch_trampoline_kprobe(struct kprobe *p)