projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
x86: kprobes change kprobe_handler flow
[deliverable/linux.git] / arch / x86 / kernel / kprobes.c
1 /*
2 * Kernel Probes (KProbes)
3 *
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.
8 *
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.
13 *
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.
17 *
18 * Copyright (C) IBM Corporation, 2002, 2004
19 *
20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 * Probes initial implementation ( includes contributions from
22 * Rusty Russell).
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.
41 */
42
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>
50
51 #include <asm/cacheflush.h>
52 #include <asm/desc.h>
53 #include <asm/pgtable.h>
54 #include <asm/uaccess.h>
55 #include <asm/alternative.h>
56
57 void jprobe_return_end(void);
58
59 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
60 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
61
62 #ifdef CONFIG_X86_64
63 #define stack_addr(regs) ((unsigned long *)regs->sp)
64 #else
65 /*
66 * "&regs->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 &regs->sp happens to
71 * point to the top of the pre-int3 stack.
72 */
73 #define stack_addr(regs) ((unsigned long *)&regs->sp)
74 #endif
75
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)) \
81 << (row % 32))
82 /*
83 * Undefined/reserved opcodes, conditional jump, Opcode Extension
84 * Groups, and some special opcodes can not boost.
85 */
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 */
107 };
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 */
129 };
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 */
151 };
152 #undef W
153
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 */
158 };
159 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
160
161 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
162 static void __kprobes set_jmp_op(void *from, void *to)
163 {
164 struct __arch_jmp_op {
165 char op;
166 s32 raddr;
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;
171 }
172
173 /*
174 * Check for the REX prefix which can only exist on X86_64
175 * X86_32 always returns 0
176 */
177 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
178 {
179 #ifdef CONFIG_X86_64
180 if ((*insn & 0xf0) == 0x40)
181 return 1;
182 #endif
183 return 0;
184 }
185
186 /*
187 * Returns non-zero if opcode is boostable.
188 * RIP relative instructions are adjusted at copying time in 64 bits mode
189 */
190 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
191 {
192 kprobe_opcode_t opcode;
193 kprobe_opcode_t *orig_opcodes = opcodes;
194
195 retry:
196 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
197 return 0;
198 opcode = *(opcodes++);
199
200 /* 2nd-byte opcode */
201 if (opcode == 0x0f) {
202 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
203 return 0;
204 return test_bit(*opcodes,
205 (unsigned long *)twobyte_is_boostable);
206 }
207
208 switch (opcode & 0xf0) {
209 #ifdef CONFIG_X86_64
210 case 0x40:
211 goto retry; /* REX prefix is boostable */
212 #endif
213 case 0x60:
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);
218 case 0x70:
219 return 0; /* can't boost conditional jump */
220 case 0xc0:
221 /* can't boost software-interruptions */
222 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
223 case 0xd0:
224 /* can boost AA* and XLAT */
225 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
226 case 0xe0:
227 /* can boost in/out and absolute jmps */
228 return ((opcode & 0x04) || opcode == 0xea);
229 case 0xf0:
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));
234 default:
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);
240 }
241 }
242
243 /*
244 * Returns non-zero if opcode modifies the interrupt flag.
245 */
246 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
247 {
248 switch (*insn) {
249 case 0xfa: /* cli */
250 case 0xfb: /* sti */
251 case 0xcf: /* iret/iretd */
252 case 0x9d: /* popf/popfd */
253 return 1;
254 }
255
256 /*
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
259 */
260 if (is_REX_prefix(insn))
261 return is_IF_modifier(++insn);
262
263 return 0;
264 }
265
266 /*
267 * Adjust the displacement if the instruction uses the %rip-relative
268 * addressing mode.
269 * If it does, Return the address of the 32-bit displacement word.
270 * If not, return null.
271 * Only applicable to 64-bit x86.
272 */
273 static void __kprobes fix_riprel(struct kprobe *p)
274 {
275 #ifdef CONFIG_X86_64
276 u8 *insn = p->ainsn.insn;
277 s64 disp;
278 int need_modrm;
279
280 /* Skip legacy instruction prefixes. */
281 while (1) {
282 switch (*insn) {
283 case 0x66:
284 case 0x67:
285 case 0x2e:
286 case 0x3e:
287 case 0x26:
288 case 0x64:
289 case 0x65:
290 case 0x36:
291 case 0xf0:
292 case 0xf3:
293 case 0xf2:
294 ++insn;
295 continue;
296 }
297 break;
298 }
299
300 /* Skip REX instruction prefix. */
301 if (is_REX_prefix(insn))
302 ++insn;
303
304 if (*insn == 0x0f) {
305 /* Two-byte opcode. */
306 ++insn;
307 need_modrm = test_bit(*insn,
308 (unsigned long *)twobyte_has_modrm);
309 } else
310 /* One-byte opcode. */
311 need_modrm = test_bit(*insn,
312 (unsigned long *)onebyte_has_modrm);
313
314 if (need_modrm) {
315 u8 modrm = *++insn;
316 if ((modrm & 0xc7) == 0x05) {
317 /* %rip+disp32 addressing mode */
318 /* Displacement follows ModRM byte. */
319 ++insn;
320 /*
321 * The copied instruction uses the %rip-relative
322 * addressing mode. Adjust the displacement for the
323 * difference between the original location of this
324 * instruction and the location of the copy that will
325 * actually be run. The tricky bit here is making sure
326 * that the sign extension happens correctly in this
327 * calculation, since we need a signed 32-bit result to
328 * be sign-extended to 64 bits when it's added to the
329 * %rip value and yield the same 64-bit result that the
330 * sign-extension of the original signed 32-bit
331 * displacement would have given.
332 */
333 disp = (u8 *) p->addr + *((s32 *) insn) -
334 (u8 *) p->ainsn.insn;
335 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
336 *(s32 *)insn = (s32) disp;
337 }
338 }
339 #endif
340 }
341
342 static void __kprobes arch_copy_kprobe(struct kprobe *p)
343 {
344 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
345
346 fix_riprel(p);
347
348 if (can_boost(p->addr))
349 p->ainsn.boostable = 0;
350 else
351 p->ainsn.boostable = -1;
352
353 p->opcode = *p->addr;
354 }
355
356 int __kprobes arch_prepare_kprobe(struct kprobe *p)
357 {
358 /* insn: must be on special executable page on x86. */
359 p->ainsn.insn = get_insn_slot();
360 if (!p->ainsn.insn)
361 return -ENOMEM;
362 arch_copy_kprobe(p);
363 return 0;
364 }
365
366 void __kprobes arch_arm_kprobe(struct kprobe *p)
367 {
368 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
369 }
370
371 void __kprobes arch_disarm_kprobe(struct kprobe *p)
372 {
373 text_poke(p->addr, &p->opcode, 1);
374 }
375
376 void __kprobes arch_remove_kprobe(struct kprobe *p)
377 {
378 mutex_lock(&kprobe_mutex);
379 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
380 mutex_unlock(&kprobe_mutex);
381 }
382
383 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
384 {
385 kcb->prev_kprobe.kp = kprobe_running();
386 kcb->prev_kprobe.status = kcb->kprobe_status;
387 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
388 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
389 }
390
391 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
392 {
393 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
394 kcb->kprobe_status = kcb->prev_kprobe.status;
395 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
396 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
397 }
398
399 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
400 struct kprobe_ctlblk *kcb)
401 {
402 __get_cpu_var(current_kprobe) = p;
403 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
404 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
405 if (is_IF_modifier(p->ainsn.insn))
406 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
407 }
408
409 static void __kprobes clear_btf(void)
410 {
411 if (test_thread_flag(TIF_DEBUGCTLMSR))
412 wrmsr(MSR_IA32_DEBUGCTLMSR, 0, 0);
413 }
414
415 static void __kprobes restore_btf(void)
416 {
417 if (test_thread_flag(TIF_DEBUGCTLMSR))
418 wrmsr(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr, 0);
419 }
420
421 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
422 {
423 clear_btf();
424 regs->flags |= X86_EFLAGS_TF;
425 regs->flags &= ~X86_EFLAGS_IF;
426 /* single step inline if the instruction is an int3 */
427 if (p->opcode == BREAKPOINT_INSTRUCTION)
428 regs->ip = (unsigned long)p->addr;
429 else
430 regs->ip = (unsigned long)p->ainsn.insn;
431 }
432
433 /* Called with kretprobe_lock held */
434 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
435 struct pt_regs *regs)
436 {
437 unsigned long *sara = stack_addr(regs);
438
439 ri->ret_addr = (kprobe_opcode_t *) *sara;
440
441 /* Replace the return addr with trampoline addr */
442 *sara = (unsigned long) &kretprobe_trampoline;
443 }
444 /*
445 * We have reentered the kprobe_handler(), since another probe was hit while
446 * within the handler. We save the original kprobes variables and just single
447 * step on the instruction of the new probe without calling any user handlers.
448 */
449 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
450 struct kprobe_ctlblk *kcb)
451 {
452 if (kcb->kprobe_status == KPROBE_HIT_SS &&
453 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
454 regs->flags &= ~X86_EFLAGS_TF;
455 regs->flags |= kcb->kprobe_saved_flags;
456 return 0;
457 #ifdef CONFIG_X86_64
458 } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
459 /* TODO: Provide re-entrancy from post_kprobes_handler() and
460 * avoid exception stack corruption while single-stepping on
461 * the instruction of the new probe.
462 */
463 arch_disarm_kprobe(p);
464 regs->ip = (unsigned long)p->addr;
465 reset_current_kprobe();
466 return 1;
467 #endif
468 }
469 save_previous_kprobe(kcb);
470 set_current_kprobe(p, regs, kcb);
471 kprobes_inc_nmissed_count(p);
472 prepare_singlestep(p, regs);
473 kcb->kprobe_status = KPROBE_REENTER;
474 return 1;
475 }
476
477 /*
478 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
479 * remain disabled thorough out this function.
480 */
481 static int __kprobes kprobe_handler(struct pt_regs *regs)
482 {
483 struct kprobe *p;
484 int ret = 0;
485 kprobe_opcode_t *addr;
486 struct kprobe_ctlblk *kcb;
487
488 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
489
490 /*
491 * We don't want to be preempted for the entire
492 * duration of kprobe processing
493 */
494 preempt_disable();
495 kcb = get_kprobe_ctlblk();
496
497 p = get_kprobe(addr);
498 if (p) {
499 /* Check we're not actually recursing */
500 if (kprobe_running()) {
501 ret = reenter_kprobe(p, regs, kcb);
502 if (kcb->kprobe_status == KPROBE_REENTER)
503 {
504 ret = 1;
505 goto out;
506 }
507 goto preempt_out;
508 } else {
509 set_current_kprobe(p, regs, kcb);
510 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
511 if (p->pre_handler && p->pre_handler(p, regs))
512 {
513 /* handler set things up, skip ss setup */
514 ret = 1;
515 goto out;
516 }
517 }
518 } else {
519 if (*addr != BREAKPOINT_INSTRUCTION) {
520 /*
521 * The breakpoint instruction was removed right
522 * after we hit it. Another cpu has removed
523 * either a probepoint or a debugger breakpoint
524 * at this address. In either case, no further
525 * handling of this interrupt is appropriate.
526 * Back up over the (now missing) int3 and run
527 * the original instruction.
528 */
529 regs->ip = (unsigned long)addr;
530 ret = 1;
531 goto preempt_out;
532 }
533 if (kprobe_running()) {
534 p = __get_cpu_var(current_kprobe);
535 if (p->break_handler && p->break_handler(p, regs))
536 goto ss_probe;
537 }
538 /* Not one of ours: let kernel handle it */
539 goto preempt_out;
540 }
541
542 ss_probe:
543 ret = 1;
544 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
545 if (p->ainsn.boostable == 1 && !p->post_handler) {
546 /* Boost up -- we can execute copied instructions directly */
547 reset_current_kprobe();
548 regs->ip = (unsigned long)p->ainsn.insn;
549 goto preempt_out;
550 }
551 #endif
552 prepare_singlestep(p, regs);
553 kcb->kprobe_status = KPROBE_HIT_SS;
554 goto out;
555
556 preempt_out:
557 preempt_enable_no_resched();
558 out:
559 return ret;
560 }
561
562 /*
563 * When a retprobed function returns, this code saves registers and
564 * calls trampoline_handler() runs, which calls the kretprobe's handler.
565 */
566 void __kprobes kretprobe_trampoline_holder(void)
567 {
568 asm volatile (
569 ".global kretprobe_trampoline\n"
570 "kretprobe_trampoline: \n"
571 #ifdef CONFIG_X86_64
572 /* We don't bother saving the ss register */
573 " pushq %rsp\n"
574 " pushfq\n"
575 /*
576 * Skip cs, ip, orig_ax.
577 * trampoline_handler() will plug in these values
578 */
579 " subq $24, %rsp\n"
580 " pushq %rdi\n"
581 " pushq %rsi\n"
582 " pushq %rdx\n"
583 " pushq %rcx\n"
584 " pushq %rax\n"
585 " pushq %r8\n"
586 " pushq %r9\n"
587 " pushq %r10\n"
588 " pushq %r11\n"
589 " pushq %rbx\n"
590 " pushq %rbp\n"
591 " pushq %r12\n"
592 " pushq %r13\n"
593 " pushq %r14\n"
594 " pushq %r15\n"
595 " movq %rsp, %rdi\n"
596 " call trampoline_handler\n"
597 /* Replace saved sp with true return address. */
598 " movq %rax, 152(%rsp)\n"
599 " popq %r15\n"
600 " popq %r14\n"
601 " popq %r13\n"
602 " popq %r12\n"
603 " popq %rbp\n"
604 " popq %rbx\n"
605 " popq %r11\n"
606 " popq %r10\n"
607 " popq %r9\n"
608 " popq %r8\n"
609 " popq %rax\n"
610 " popq %rcx\n"
611 " popq %rdx\n"
612 " popq %rsi\n"
613 " popq %rdi\n"
614 /* Skip orig_ax, ip, cs */
615 " addq $24, %rsp\n"
616 " popfq\n"
617 #else
618 " pushf\n"
619 /*
620 * Skip cs, ip, orig_ax.
621 * trampoline_handler() will plug in these values
622 */
623 " subl $12, %esp\n"
624 " pushl %fs\n"
625 " pushl %ds\n"
626 " pushl %es\n"
627 " pushl %eax\n"
628 " pushl %ebp\n"
629 " pushl %edi\n"
630 " pushl %esi\n"
631 " pushl %edx\n"
632 " pushl %ecx\n"
633 " pushl %ebx\n"
634 " movl %esp, %eax\n"
635 " call trampoline_handler\n"
636 /* Move flags to cs */
637 " movl 52(%esp), %edx\n"
638 " movl %edx, 48(%esp)\n"
639 /* Replace saved flags with true return address. */
640 " movl %eax, 52(%esp)\n"
641 " popl %ebx\n"
642 " popl %ecx\n"
643 " popl %edx\n"
644 " popl %esi\n"
645 " popl %edi\n"
646 " popl %ebp\n"
647 " popl %eax\n"
648 /* Skip ip, orig_ax, es, ds, fs */
649 " addl $20, %esp\n"
650 " popf\n"
651 #endif
652 " ret\n");
653 }
654
655 /*
656 * Called from kretprobe_trampoline
657 */
658 void * __kprobes trampoline_handler(struct pt_regs *regs)
659 {
660 struct kretprobe_instance *ri = NULL;
661 struct hlist_head *head, empty_rp;
662 struct hlist_node *node, *tmp;
663 unsigned long flags, orig_ret_address = 0;
664 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
665
666 INIT_HLIST_HEAD(&empty_rp);
667 spin_lock_irqsave(&kretprobe_lock, flags);
668 head = kretprobe_inst_table_head(current);
669 /* fixup registers */
670 #ifdef CONFIG_X86_64
671 regs->cs = __KERNEL_CS;
672 #else
673 regs->cs = __KERNEL_CS | get_kernel_rpl();
674 #endif
675 regs->ip = trampoline_address;
676 regs->orig_ax = ~0UL;
677
678 /*
679 * It is possible to have multiple instances associated with a given
680 * task either because multiple functions in the call path have
681 * return probes installed on them, and/or more then one
682 * return probe was registered for a target function.
683 *
684 * We can handle this because:
685 * - instances are always pushed into the head of the list
686 * - when multiple return probes are registered for the same
687 * function, the (chronologically) first instance's ret_addr
688 * will be the real return address, and all the rest will
689 * point to kretprobe_trampoline.
690 */
691 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
692 if (ri->task != current)
693 /* another task is sharing our hash bucket */
694 continue;
695
696 if (ri->rp && ri->rp->handler) {
697 __get_cpu_var(current_kprobe) = &ri->rp->kp;
698 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
699 ri->rp->handler(ri, regs);
700 __get_cpu_var(current_kprobe) = NULL;
701 }
702
703 orig_ret_address = (unsigned long)ri->ret_addr;
704 recycle_rp_inst(ri, &empty_rp);
705
706 if (orig_ret_address != trampoline_address)
707 /*
708 * This is the real return address. Any other
709 * instances associated with this task are for
710 * other calls deeper on the call stack
711 */
712 break;
713 }
714
715 kretprobe_assert(ri, orig_ret_address, trampoline_address);
716
717 spin_unlock_irqrestore(&kretprobe_lock, flags);
718
719 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
720 hlist_del(&ri->hlist);
721 kfree(ri);
722 }
723 return (void *)orig_ret_address;
724 }
725
726 /*
727 * Called after single-stepping. p->addr is the address of the
728 * instruction whose first byte has been replaced by the "int 3"
729 * instruction. To avoid the SMP problems that can occur when we
730 * temporarily put back the original opcode to single-step, we
731 * single-stepped a copy of the instruction. The address of this
732 * copy is p->ainsn.insn.
733 *
734 * This function prepares to return from the post-single-step
735 * interrupt. We have to fix up the stack as follows:
736 *
737 * 0) Except in the case of absolute or indirect jump or call instructions,
738 * the new ip is relative to the copied instruction. We need to make
739 * it relative to the original instruction.
740 *
741 * 1) If the single-stepped instruction was pushfl, then the TF and IF
742 * flags are set in the just-pushed flags, and may need to be cleared.
743 *
744 * 2) If the single-stepped instruction was a call, the return address
745 * that is atop the stack is the address following the copied instruction.
746 * We need to make it the address following the original instruction.
747 *
748 * If this is the first time we've single-stepped the instruction at
749 * this probepoint, and the instruction is boostable, boost it: add a
750 * jump instruction after the copied instruction, that jumps to the next
751 * instruction after the probepoint.
752 */
753 static void __kprobes resume_execution(struct kprobe *p,
754 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
755 {
756 unsigned long *tos = stack_addr(regs);
757 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
758 unsigned long orig_ip = (unsigned long)p->addr;
759 kprobe_opcode_t *insn = p->ainsn.insn;
760
761 /*skip the REX prefix*/
762 if (is_REX_prefix(insn))
763 insn++;
764
765 regs->flags &= ~X86_EFLAGS_TF;
766 switch (*insn) {
767 case 0x9c: /* pushfl */
768 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
769 *tos |= kcb->kprobe_old_flags;
770 break;
771 case 0xc2: /* iret/ret/lret */
772 case 0xc3:
773 case 0xca:
774 case 0xcb:
775 case 0xcf:
776 case 0xea: /* jmp absolute -- ip is correct */
777 /* ip is already adjusted, no more changes required */
778 p->ainsn.boostable = 1;
779 goto no_change;
780 case 0xe8: /* call relative - Fix return addr */
781 *tos = orig_ip + (*tos - copy_ip);
782 break;
783 #ifdef CONFIG_X86_32
784 case 0x9a: /* call absolute -- same as call absolute, indirect */
785 *tos = orig_ip + (*tos - copy_ip);
786 goto no_change;
787 #endif
788 case 0xff:
789 if ((insn[1] & 0x30) == 0x10) {
790 /*
791 * call absolute, indirect
792 * Fix return addr; ip is correct.
793 * But this is not boostable
794 */
795 *tos = orig_ip + (*tos - copy_ip);
796 goto no_change;
797 } else if (((insn[1] & 0x31) == 0x20) ||
798 ((insn[1] & 0x31) == 0x21)) {
799 /*
800 * jmp near and far, absolute indirect
801 * ip is correct. And this is boostable
802 */
803 p->ainsn.boostable = 1;
804 goto no_change;
805 }
806 default:
807 break;
808 }
809
810 if (p->ainsn.boostable == 0) {
811 if ((regs->ip > copy_ip) &&
812 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
813 /*
814 * These instructions can be executed directly if it
815 * jumps back to correct address.
816 */
817 set_jmp_op((void *)regs->ip,
818 (void *)orig_ip + (regs->ip - copy_ip));
819 p->ainsn.boostable = 1;
820 } else {
821 p->ainsn.boostable = -1;
822 }
823 }
824
825 regs->ip += orig_ip - copy_ip;
826
827 no_change:
828 restore_btf();
829 }
830
831 /*
832 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
833 * remain disabled thoroughout this function.
834 */
835 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
836 {
837 struct kprobe *cur = kprobe_running();
838 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
839
840 if (!cur)
841 return 0;
842
843 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
844 kcb->kprobe_status = KPROBE_HIT_SSDONE;
845 cur->post_handler(cur, regs, 0);
846 }
847
848 resume_execution(cur, regs, kcb);
849 regs->flags |= kcb->kprobe_saved_flags;
850 trace_hardirqs_fixup_flags(regs->flags);
851
852 /* Restore back the original saved kprobes variables and continue. */
853 if (kcb->kprobe_status == KPROBE_REENTER) {
854 restore_previous_kprobe(kcb);
855 goto out;
856 }
857 reset_current_kprobe();
858 out:
859 preempt_enable_no_resched();
860
861 /*
862 * if somebody else is singlestepping across a probe point, flags
863 * will have TF set, in which case, continue the remaining processing
864 * of do_debug, as if this is not a probe hit.
865 */
866 if (regs->flags & X86_EFLAGS_TF)
867 return 0;
868
869 return 1;
870 }
871
872 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
873 {
874 struct kprobe *cur = kprobe_running();
875 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
876
877 switch (kcb->kprobe_status) {
878 case KPROBE_HIT_SS:
879 case KPROBE_REENTER:
880 /*
881 * We are here because the instruction being single
882 * stepped caused a page fault. We reset the current
883 * kprobe and the ip points back to the probe address
884 * and allow the page fault handler to continue as a
885 * normal page fault.
886 */
887 regs->ip = (unsigned long)cur->addr;
888 regs->flags |= kcb->kprobe_old_flags;
889 if (kcb->kprobe_status == KPROBE_REENTER)
890 restore_previous_kprobe(kcb);
891 else
892 reset_current_kprobe();
893 preempt_enable_no_resched();
894 break;
895 case KPROBE_HIT_ACTIVE:
896 case KPROBE_HIT_SSDONE:
897 /*
898 * We increment the nmissed count for accounting,
899 * we can also use npre/npostfault count for accounting
900 * these specific fault cases.
901 */
902 kprobes_inc_nmissed_count(cur);
903
904 /*
905 * We come here because instructions in the pre/post
906 * handler caused the page_fault, this could happen
907 * if handler tries to access user space by
908 * copy_from_user(), get_user() etc. Let the
909 * user-specified handler try to fix it first.
910 */
911 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
912 return 1;
913
914 /*
915 * In case the user-specified fault handler returned
916 * zero, try to fix up.
917 */
918 if (fixup_exception(regs))
919 return 1;
920
921 /*
922 * fixup routine could not handle it,
923 * Let do_page_fault() fix it.
924 */
925 break;
926 default:
927 break;
928 }
929 return 0;
930 }
931
932 /*
933 * Wrapper routine for handling exceptions.
934 */
935 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
936 unsigned long val, void *data)
937 {
938 struct die_args *args = (struct die_args *)data;
939 int ret = NOTIFY_DONE;
940
941 if (args->regs && user_mode_vm(args->regs))
942 return ret;
943
944 switch (val) {
945 case DIE_INT3:
946 if (kprobe_handler(args->regs))
947 ret = NOTIFY_STOP;
948 break;
949 case DIE_DEBUG:
950 if (post_kprobe_handler(args->regs))
951 ret = NOTIFY_STOP;
952 break;
953 case DIE_GPF:
954 /* kprobe_running() needs smp_processor_id() */
955 preempt_disable();
956 if (kprobe_running() &&
957 kprobe_fault_handler(args->regs, args->trapnr))
958 ret = NOTIFY_STOP;
959 preempt_enable();
960 break;
961 default:
962 break;
963 }
964 return ret;
965 }
966
967 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
968 {
969 struct jprobe *jp = container_of(p, struct jprobe, kp);
970 unsigned long addr;
971 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
972
973 kcb->jprobe_saved_regs = *regs;
974 kcb->jprobe_saved_sp = stack_addr(regs);
975 addr = (unsigned long)(kcb->jprobe_saved_sp);
976
977 /*
978 * As Linus pointed out, gcc assumes that the callee
979 * owns the argument space and could overwrite it, e.g.
980 * tailcall optimization. So, to be absolutely safe
981 * we also save and restore enough stack bytes to cover
982 * the argument area.
983 */
984 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
985 MIN_STACK_SIZE(addr));
986 regs->flags &= ~X86_EFLAGS_IF;
987 trace_hardirqs_off();
988 regs->ip = (unsigned long)(jp->entry);
989 return 1;
990 }
991
992 void __kprobes jprobe_return(void)
993 {
994 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
995
996 asm volatile (
997 #ifdef CONFIG_X86_64
998 " xchg %%rbx,%%rsp \n"
999 #else
1000 " xchgl %%ebx,%%esp \n"
1001 #endif
1002 " int3 \n"
1003 " .globl jprobe_return_end\n"
1004 " jprobe_return_end: \n"
1005 " nop \n"::"b"
1006 (kcb->jprobe_saved_sp):"memory");
1007 }
1008
1009 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1010 {
1011 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1012 u8 *addr = (u8 *) (regs->ip - 1);
1013 struct jprobe *jp = container_of(p, struct jprobe, kp);
1014
1015 if ((addr > (u8 *) jprobe_return) &&
1016 (addr < (u8 *) jprobe_return_end)) {
1017 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1018 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1019 printk(KERN_ERR
1020 "current sp %p does not match saved sp %p\n",
1021 stack_addr(regs), kcb->jprobe_saved_sp);
1022 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1023 show_registers(saved_regs);
1024 printk(KERN_ERR "Current registers\n");
1025 show_registers(regs);
1026 BUG();
1027 }
1028 *regs = kcb->jprobe_saved_regs;
1029 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1030 kcb->jprobes_stack,
1031 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1032 preempt_enable_no_resched();
1033 return 1;
1034 }
1035 return 0;
1036 }
1037
1038 int __init arch_init_kprobes(void)
1039 {
1040 return 0;
1041 }
1042
1043 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1044 {
1045 return 0;
1046 }
Linux kernel - Deliverables
This page took 0.090237 seconds and 6 git commands to generate.