aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/kprobes.c
diff options
context:
space:
mode:
authorMasami Hiramatsu <mhiramat@redhat.com>2008-01-30 07:31:21 -0500
committerIngo Molnar <mingo@elte.hu>2008-01-30 07:31:21 -0500
commitd6be29b871e285d33be0e3025929e2d6bcabb0c0 (patch)
tree5dbcd8ef953d19e3978062d337c76e4fa103150b /arch/x86/kernel/kprobes.c
parent8533bbe9f87b01f49ff951f665ea1988252fa3c2 (diff)
x86: kprobes code for x86 unification
This patch unifies kprobes code. - Unify kprobes_*.h to kprobes.h - Unify kprobes_*.c to kprobes.c (Differences are separated by ifdefs) - Most differences are related to REX prefix and rip relatives. - Two inline assembly code are different. - One difference in kprobe_handlre() - One fixup exception code is different, but it will be unified if mm/extable_*.c are unified. - Merge history logs into arch/x86/kernel/kprobes.c. Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Signed-off-by: Jim Keniston <jkenisto@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Diffstat (limited to 'arch/x86/kernel/kprobes.c')
-rw-r--r--arch/x86/kernel/kprobes.c1045
1 files changed, 1045 insertions, 0 deletions
diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
new file mode 100644
index 00000000000..9aadd4d4a22
--- /dev/null
+++ b/arch/x86/kernel/kprobes.c
@@ -0,0 +1,1045 @@
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
57void jprobe_return_end(void);
58
59DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
60DEFINE_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 */
86static 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};
108static 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};
130static 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
154struct 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};
159const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
160
161/* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
162static __always_inline void 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 * Returns non-zero if opcode is boostable.
175 * RIP relative instructions are adjusted at copying time in 64 bits mode
176 */
177static __always_inline int can_boost(kprobe_opcode_t *opcodes)
178{
179 kprobe_opcode_t opcode;
180 kprobe_opcode_t *orig_opcodes = opcodes;
181
182retry:
183 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
184 return 0;
185 opcode = *(opcodes++);
186
187 /* 2nd-byte opcode */
188 if (opcode == 0x0f) {
189 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
190 return 0;
191 return test_bit(*opcodes,
192 (unsigned long *)twobyte_is_boostable);
193 }
194
195 switch (opcode & 0xf0) {
196#ifdef CONFIG_X86_64
197 case 0x40:
198 goto retry; /* REX prefix is boostable */
199#endif
200 case 0x60:
201 if (0x63 < opcode && opcode < 0x67)
202 goto retry; /* prefixes */
203 /* can't boost Address-size override and bound */
204 return (opcode != 0x62 && opcode != 0x67);
205 case 0x70:
206 return 0; /* can't boost conditional jump */
207 case 0xc0:
208 /* can't boost software-interruptions */
209 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
210 case 0xd0:
211 /* can boost AA* and XLAT */
212 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
213 case 0xe0:
214 /* can boost in/out and absolute jmps */
215 return ((opcode & 0x04) || opcode == 0xea);
216 case 0xf0:
217 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
218 goto retry; /* lock/rep(ne) prefix */
219 /* clear and set flags are boostable */
220 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
221 default:
222 /* segment override prefixes are boostable */
223 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
224 goto retry; /* prefixes */
225 /* CS override prefix and call are not boostable */
226 return (opcode != 0x2e && opcode != 0x9a);
227 }
228}
229
230/*
231 * Returns non-zero if opcode modifies the interrupt flag.
232 */
233static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
234{
235 switch (*insn) {
236 case 0xfa: /* cli */
237 case 0xfb: /* sti */
238 case 0xcf: /* iret/iretd */
239 case 0x9d: /* popf/popfd */
240 return 1;
241 }
242#ifdef CONFIG_X86_64
243 /*
244 * on 64 bit x86, 0x40-0x4f are prefixes so we need to look
245 * at the next byte instead.. but of course not recurse infinitely
246 */
247 if (*insn >= 0x40 && *insn <= 0x4f)
248 return is_IF_modifier(++insn);
249#endif
250 return 0;
251}
252
253#ifdef CONFIG_X86_64
254/*
255 * Adjust the displacement if the instruction uses the %rip-relative
256 * addressing mode.
257 * If it does, Return the address of the 32-bit displacement word.
258 * If not, return null.
259 */
260static void __kprobes fix_riprel(struct kprobe *p)
261{
262 u8 *insn = p->ainsn.insn;
263 s64 disp;
264 int need_modrm;
265
266 /* Skip legacy instruction prefixes. */
267 while (1) {
268 switch (*insn) {
269 case 0x66:
270 case 0x67:
271 case 0x2e:
272 case 0x3e:
273 case 0x26:
274 case 0x64:
275 case 0x65:
276 case 0x36:
277 case 0xf0:
278 case 0xf3:
279 case 0xf2:
280 ++insn;
281 continue;
282 }
283 break;
284 }
285
286 /* Skip REX instruction prefix. */
287 if ((*insn & 0xf0) == 0x40)
288 ++insn;
289
290 if (*insn == 0x0f) {
291 /* Two-byte opcode. */
292 ++insn;
293 need_modrm = test_bit(*insn,
294 (unsigned long *)twobyte_has_modrm);
295 } else
296 /* One-byte opcode. */
297 need_modrm = test_bit(*insn,
298 (unsigned long *)onebyte_has_modrm);
299
300 if (need_modrm) {
301 u8 modrm = *++insn;
302 if ((modrm & 0xc7) == 0x05) {
303 /* %rip+disp32 addressing mode */
304 /* Displacement follows ModRM byte. */
305 ++insn;
306 /*
307 * The copied instruction uses the %rip-relative
308 * addressing mode. Adjust the displacement for the
309 * difference between the original location of this
310 * instruction and the location of the copy that will
311 * actually be run. The tricky bit here is making sure
312 * that the sign extension happens correctly in this
313 * calculation, since we need a signed 32-bit result to
314 * be sign-extended to 64 bits when it's added to the
315 * %rip value and yield the same 64-bit result that the
316 * sign-extension of the original signed 32-bit
317 * displacement would have given.
318 */
319 disp = (u8 *) p->addr + *((s32 *) insn) -
320 (u8 *) p->ainsn.insn;
321 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
322 *(s32 *)insn = (s32) disp;
323 }
324 }
325}
326#endif
327
328static void __kprobes arch_copy_kprobe(struct kprobe *p)
329{
330 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
331#ifdef CONFIG_X86_64
332 fix_riprel(p);
333#endif
334 if (can_boost(p->addr))
335 p->ainsn.boostable = 0;
336 else
337 p->ainsn.boostable = -1;
338
339 p->opcode = *p->addr;
340}
341
342int __kprobes arch_prepare_kprobe(struct kprobe *p)
343{
344 /* insn: must be on special executable page on x86. */
345 p->ainsn.insn = get_insn_slot();
346 if (!p->ainsn.insn)
347 return -ENOMEM;
348 arch_copy_kprobe(p);
349 return 0;
350}
351
352void __kprobes arch_arm_kprobe(struct kprobe *p)
353{
354 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
355}
356
357void __kprobes arch_disarm_kprobe(struct kprobe *p)
358{
359 text_poke(p->addr, &p->opcode, 1);
360}
361
362void __kprobes arch_remove_kprobe(struct kprobe *p)
363{
364 mutex_lock(&kprobe_mutex);
365 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
366 mutex_unlock(&kprobe_mutex);
367}
368
369static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
370{
371 kcb->prev_kprobe.kp = kprobe_running();
372 kcb->prev_kprobe.status = kcb->kprobe_status;
373 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
374 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
375}
376
377static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
378{
379 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
380 kcb->kprobe_status = kcb->prev_kprobe.status;
381 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
382 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
383}
384
385static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
386 struct kprobe_ctlblk *kcb)
387{
388 __get_cpu_var(current_kprobe) = p;
389 kcb->kprobe_saved_flags = kcb->kprobe_old_flags
390 = (regs->flags & (TF_MASK | IF_MASK));
391 if (is_IF_modifier(p->ainsn.insn))
392 kcb->kprobe_saved_flags &= ~IF_MASK;
393}
394
395static __always_inline void clear_btf(void)
396{
397 if (test_thread_flag(TIF_DEBUGCTLMSR))
398 wrmsr(MSR_IA32_DEBUGCTLMSR, 0, 0);
399}
400
401static __always_inline void restore_btf(void)
402{
403 if (test_thread_flag(TIF_DEBUGCTLMSR))
404 wrmsr(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr, 0);
405}
406
407static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
408{
409 clear_btf();
410 regs->flags |= TF_MASK;
411 regs->flags &= ~IF_MASK;
412 /*single step inline if the instruction is an int3*/
413 if (p->opcode == BREAKPOINT_INSTRUCTION)
414 regs->ip = (unsigned long)p->addr;
415 else
416 regs->ip = (unsigned long)p->ainsn.insn;
417}
418
419/* Called with kretprobe_lock held */
420void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
421 struct pt_regs *regs)
422{
423 unsigned long *sara = stack_addr(regs);
424
425 ri->ret_addr = (kprobe_opcode_t *) *sara;
426
427 /* Replace the return addr with trampoline addr */
428 *sara = (unsigned long) &kretprobe_trampoline;
429}
430
431/*
432 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
433 * remain disabled thorough out this function.
434 */
435static int __kprobes kprobe_handler(struct pt_regs *regs)
436{
437 struct kprobe *p;
438 int ret = 0;
439 kprobe_opcode_t *addr;
440 struct kprobe_ctlblk *kcb;
441
442 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
443
444 /*
445 * We don't want to be preempted for the entire
446 * duration of kprobe processing
447 */
448 preempt_disable();
449 kcb = get_kprobe_ctlblk();
450
451 /* Check we're not actually recursing */
452 if (kprobe_running()) {
453 p = get_kprobe(addr);
454 if (p) {
455 if (kcb->kprobe_status == KPROBE_HIT_SS &&
456 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
457 regs->flags &= ~TF_MASK;
458 regs->flags |= kcb->kprobe_saved_flags;
459 goto no_kprobe;
460#ifdef CONFIG_X86_64
461 } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
462 /* TODO: Provide re-entrancy from
463 * post_kprobes_handler() and avoid exception
464 * stack corruption while single-stepping on
465 * the instruction of the new probe.
466 */
467 arch_disarm_kprobe(p);
468 regs->ip = (unsigned long)p->addr;
469 reset_current_kprobe();
470 return 1;
471#endif
472 }
473 /* We have reentered the kprobe_handler(), since
474 * another probe was hit while within the handler.
475 * We here save the original kprobes variables and
476 * just single step on the instruction of the new probe
477 * without calling any user handlers.
478 */
479 save_previous_kprobe(kcb);
480 set_current_kprobe(p, regs, kcb);
481 kprobes_inc_nmissed_count(p);
482 prepare_singlestep(p, regs);
483 kcb->kprobe_status = KPROBE_REENTER;
484 return 1;
485 } else {
486 if (*addr != BREAKPOINT_INSTRUCTION) {
487 /* The breakpoint instruction was removed by
488 * another cpu right after we hit, no further
489 * handling of this interrupt is appropriate
490 */
491 regs->ip = (unsigned long)addr;
492 ret = 1;
493 goto no_kprobe;
494 }
495 p = __get_cpu_var(current_kprobe);
496 if (p->break_handler && p->break_handler(p, regs))
497 goto ss_probe;
498 }
499 goto no_kprobe;
500 }
501
502 p = get_kprobe(addr);
503 if (!p) {
504 if (*addr != BREAKPOINT_INSTRUCTION) {
505 /*
506 * The breakpoint instruction was removed right
507 * after we hit it. Another cpu has removed
508 * either a probepoint or a debugger breakpoint
509 * at this address. In either case, no further
510 * handling of this interrupt is appropriate.
511 * Back up over the (now missing) int3 and run
512 * the original instruction.
513 */
514 regs->ip = (unsigned long)addr;
515 ret = 1;
516 }
517 /* Not one of ours: let kernel handle it */
518 goto no_kprobe;
519 }
520
521 set_current_kprobe(p, regs, kcb);
522 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
523
524 if (p->pre_handler && p->pre_handler(p, regs))
525 /* handler has already set things up, so skip ss setup */
526 return 1;
527
528ss_probe:
529#if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
530 if (p->ainsn.boostable == 1 && !p->post_handler) {
531 /* Boost up -- we can execute copied instructions directly */
532 reset_current_kprobe();
533 regs->ip = (unsigned long)p->ainsn.insn;
534 preempt_enable_no_resched();
535 return 1;
536 }
537#endif
538 prepare_singlestep(p, regs);
539 kcb->kprobe_status = KPROBE_HIT_SS;
540 return 1;
541
542no_kprobe:
543 preempt_enable_no_resched();
544 return ret;
545}
546
547/*
548 * When a retprobed function returns, this code saves registers and
549 * calls trampoline_handler() runs, which calls the kretprobe's handler.
550 */
551 void __kprobes kretprobe_trampoline_holder(void)
552 {
553 asm volatile (
554 ".global kretprobe_trampoline\n"
555 "kretprobe_trampoline: \n"
556#ifdef CONFIG_X86_64
557 /* We don't bother saving the ss register */
558 " pushq %rsp\n"
559 " pushfq\n"
560 /*
561 * Skip cs, ip, orig_ax.
562 * trampoline_handler() will plug in these values
563 */
564 " subq $24, %rsp\n"
565 " pushq %rdi\n"
566 " pushq %rsi\n"
567 " pushq %rdx\n"
568 " pushq %rcx\n"
569 " pushq %rax\n"
570 " pushq %r8\n"
571 " pushq %r9\n"
572 " pushq %r10\n"
573 " pushq %r11\n"
574 " pushq %rbx\n"
575 " pushq %rbp\n"
576 " pushq %r12\n"
577 " pushq %r13\n"
578 " pushq %r14\n"
579 " pushq %r15\n"
580 " movq %rsp, %rdi\n"
581 " call trampoline_handler\n"
582 /* Replace saved sp with true return address. */
583 " movq %rax, 152(%rsp)\n"
584 " popq %r15\n"
585 " popq %r14\n"
586 " popq %r13\n"
587 " popq %r12\n"
588 " popq %rbp\n"
589 " popq %rbx\n"
590 " popq %r11\n"
591 " popq %r10\n"
592 " popq %r9\n"
593 " popq %r8\n"
594 " popq %rax\n"
595 " popq %rcx\n"
596 " popq %rdx\n"
597 " popq %rsi\n"
598 " popq %rdi\n"
599 /* Skip orig_ax, ip, cs */
600 " addq $24, %rsp\n"
601 " popfq\n"
602#else
603 " pushf\n"
604 /*
605 * Skip cs, ip, orig_ax.
606 * trampoline_handler() will plug in these values
607 */
608 " subl $12, %esp\n"
609 " pushl %fs\n"
610 " pushl %ds\n"
611 " pushl %es\n"
612 " pushl %eax\n"
613 " pushl %ebp\n"
614 " pushl %edi\n"
615 " pushl %esi\n"
616 " pushl %edx\n"
617 " pushl %ecx\n"
618 " pushl %ebx\n"
619 " movl %esp, %eax\n"
620 " call trampoline_handler\n"
621 /* Move flags to cs */
622 " movl 52(%esp), %edx\n"
623 " movl %edx, 48(%esp)\n"
624 /* Replace saved flags with true return address. */
625 " movl %eax, 52(%esp)\n"
626 " popl %ebx\n"
627 " popl %ecx\n"
628 " popl %edx\n"
629 " popl %esi\n"
630 " popl %edi\n"
631 " popl %ebp\n"
632 " popl %eax\n"
633 /* Skip ip, orig_ax, es, ds, fs */
634 " addl $20, %esp\n"
635 " popf\n"
636#endif
637 " ret\n");
638 }
639
640/*
641 * Called from kretprobe_trampoline
642 */
643void * __kprobes trampoline_handler(struct pt_regs *regs)
644{
645 struct kretprobe_instance *ri = NULL;
646 struct hlist_head *head, empty_rp;
647 struct hlist_node *node, *tmp;
648 unsigned long flags, orig_ret_address = 0;
649 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
650
651 INIT_HLIST_HEAD(&empty_rp);
652 spin_lock_irqsave(&kretprobe_lock, flags);
653 head = kretprobe_inst_table_head(current);
654 /* fixup registers */
655#ifdef CONFIG_X86_64
656 regs->cs = __KERNEL_CS;
657#else
658 regs->cs = __KERNEL_CS | get_kernel_rpl();
659#endif
660 regs->ip = trampoline_address;
661 regs->orig_ax = ~0UL;
662
663 /*
664 * It is possible to have multiple instances associated with a given
665 * task either because multiple functions in the call path have
666 * return probes installed on them, and/or more then one
667 * return probe was registered for a target function.
668 *
669 * We can handle this because:
670 * - instances are always pushed into the head of the list
671 * - when multiple return probes are registered for the same
672 * function, the (chronologically) first instance's ret_addr
673 * will be the real return address, and all the rest will
674 * point to kretprobe_trampoline.
675 */
676 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
677 if (ri->task != current)
678 /* another task is sharing our hash bucket */
679 continue;
680
681 if (ri->rp && ri->rp->handler) {
682 __get_cpu_var(current_kprobe) = &ri->rp->kp;
683 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
684 ri->rp->handler(ri, regs);
685 __get_cpu_var(current_kprobe) = NULL;
686 }
687
688 orig_ret_address = (unsigned long)ri->ret_addr;
689 recycle_rp_inst(ri, &empty_rp);
690
691 if (orig_ret_address != trampoline_address)
692 /*
693 * This is the real return address. Any other
694 * instances associated with this task are for
695 * other calls deeper on the call stack
696 */
697 break;
698 }
699
700 kretprobe_assert(ri, orig_ret_address, trampoline_address);
701
702 spin_unlock_irqrestore(&kretprobe_lock, flags);
703
704 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
705 hlist_del(&ri->hlist);
706 kfree(ri);
707 }
708 return (void *)orig_ret_address;
709}
710
711/*
712 * Called after single-stepping. p->addr is the address of the
713 * instruction whose first byte has been replaced by the "int 3"
714 * instruction. To avoid the SMP problems that can occur when we
715 * temporarily put back the original opcode to single-step, we
716 * single-stepped a copy of the instruction. The address of this
717 * copy is p->ainsn.insn.
718 *
719 * This function prepares to return from the post-single-step
720 * interrupt. We have to fix up the stack as follows:
721 *
722 * 0) Except in the case of absolute or indirect jump or call instructions,
723 * the new ip is relative to the copied instruction. We need to make
724 * it relative to the original instruction.
725 *
726 * 1) If the single-stepped instruction was pushfl, then the TF and IF
727 * flags are set in the just-pushed flags, and may need to be cleared.
728 *
729 * 2) If the single-stepped instruction was a call, the return address
730 * that is atop the stack is the address following the copied instruction.
731 * We need to make it the address following the original instruction.
732 *
733 * If this is the first time we've single-stepped the instruction at
734 * this probepoint, and the instruction is boostable, boost it: add a
735 * jump instruction after the copied instruction, that jumps to the next
736 * instruction after the probepoint.
737 */
738static void __kprobes resume_execution(struct kprobe *p,
739 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
740{
741 unsigned long *tos = stack_addr(regs);
742 unsigned long copy_ip = (unsigned long)p->ainsn.insn;
743 unsigned long orig_ip = (unsigned long)p->addr;
744 kprobe_opcode_t *insn = p->ainsn.insn;
745
746#ifdef CONFIG_X86_64
747 /*skip the REX prefix*/
748 if (*insn >= 0x40 && *insn <= 0x4f)
749 insn++;
750#endif
751
752 regs->flags &= ~TF_MASK;
753 switch (*insn) {
754 case 0x9c: /* pushfl */
755 *tos &= ~(TF_MASK | IF_MASK);
756 *tos |= kcb->kprobe_old_flags;
757 break;
758 case 0xc2: /* iret/ret/lret */
759 case 0xc3:
760 case 0xca:
761 case 0xcb:
762 case 0xcf:
763 case 0xea: /* jmp absolute -- ip is correct */
764 /* ip is already adjusted, no more changes required */
765 p->ainsn.boostable = 1;
766 goto no_change;
767 case 0xe8: /* call relative - Fix return addr */
768 *tos = orig_ip + (*tos - copy_ip);
769 break;
770#ifndef CONFIG_X86_64
771 case 0x9a: /* call absolute -- same as call absolute, indirect */
772 *tos = orig_ip + (*tos - copy_ip);
773 goto no_change;
774#endif
775 case 0xff:
776 if ((insn[1] & 0x30) == 0x10) {
777 /*
778 * call absolute, indirect
779 * Fix return addr; ip is correct.
780 * But this is not boostable
781 */
782 *tos = orig_ip + (*tos - copy_ip);
783 goto no_change;
784 } else if (((insn[1] & 0x31) == 0x20) ||
785 ((insn[1] & 0x31) == 0x21)) {
786 /*
787 * jmp near and far, absolute indirect
788 * ip is correct. And this is boostable
789 */
790 p->ainsn.boostable = 1;
791 goto no_change;
792 }
793 default:
794 break;
795 }
796
797 if (p->ainsn.boostable == 0) {
798 if ((regs->ip > copy_ip) &&
799 (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
800 /*
801 * These instructions can be executed directly if it
802 * jumps back to correct address.
803 */
804 set_jmp_op((void *)regs->ip,
805 (void *)orig_ip + (regs->ip - copy_ip));
806 p->ainsn.boostable = 1;
807 } else {
808 p->ainsn.boostable = -1;
809 }
810 }
811
812 regs->ip += orig_ip - copy_ip;
813
814no_change:
815 restore_btf();
816
817 return;
818}
819
820/*
821 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
822 * remain disabled thoroughout this function.
823 */
824static int __kprobes post_kprobe_handler(struct pt_regs *regs)
825{
826 struct kprobe *cur = kprobe_running();
827 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
828
829 if (!cur)
830 return 0;
831
832 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
833 kcb->kprobe_status = KPROBE_HIT_SSDONE;
834 cur->post_handler(cur, regs, 0);
835 }
836
837 resume_execution(cur, regs, kcb);
838 regs->flags |= kcb->kprobe_saved_flags;
839 trace_hardirqs_fixup_flags(regs->flags);
840
841 /* Restore back the original saved kprobes variables and continue. */
842 if (kcb->kprobe_status == KPROBE_REENTER) {
843 restore_previous_kprobe(kcb);
844 goto out;
845 }
846 reset_current_kprobe();
847out:
848 preempt_enable_no_resched();
849
850 /*
851 * if somebody else is singlestepping across a probe point, flags
852 * will have TF set, in which case, continue the remaining processing
853 * of do_debug, as if this is not a probe hit.
854 */
855 if (regs->flags & TF_MASK)
856 return 0;
857
858 return 1;
859}
860
861int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
862{
863 struct kprobe *cur = kprobe_running();
864 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
865
866 switch (kcb->kprobe_status) {
867 case KPROBE_HIT_SS:
868 case KPROBE_REENTER:
869 /*
870 * We are here because the instruction being single
871 * stepped caused a page fault. We reset the current
872 * kprobe and the ip points back to the probe address
873 * and allow the page fault handler to continue as a
874 * normal page fault.
875 */
876 regs->ip = (unsigned long)cur->addr;
877 regs->flags |= kcb->kprobe_old_flags;
878 if (kcb->kprobe_status == KPROBE_REENTER)
879 restore_previous_kprobe(kcb);
880 else
881 reset_current_kprobe();
882 preempt_enable_no_resched();
883 break;
884 case KPROBE_HIT_ACTIVE:
885 case KPROBE_HIT_SSDONE:
886 /*
887 * We increment the nmissed count for accounting,
888 * we can also use npre/npostfault count for accounting
889 * these specific fault cases.
890 */
891 kprobes_inc_nmissed_count(cur);
892
893 /*
894 * We come here because instructions in the pre/post
895 * handler caused the page_fault, this could happen
896 * if handler tries to access user space by
897 * copy_from_user(), get_user() etc. Let the
898 * user-specified handler try to fix it first.
899 */
900 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
901 return 1;
902
903 /*
904 * In case the user-specified fault handler returned
905 * zero, try to fix up.
906 */
907#ifdef CONFIG_X86_64
908 {
909 const struct exception_table_entry *fixup;
910 fixup = search_exception_tables(regs->ip);
911 if (fixup) {
912 regs->ip = fixup->fixup;
913 return 1;
914 }
915 }
916#else
917 if (fixup_exception(regs))
918 return 1;
919#endif
920 /*
921 * fixup routine could not handle it,
922 * Let do_page_fault() fix it.
923 */
924 break;
925 default:
926 break;
927 }
928 return 0;
929}
930
931/*
932 * Wrapper routine for handling exceptions.
933 */
934int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
935 unsigned long val, void *data)
936{
937 struct die_args *args = (struct die_args *)data;
938 int ret = NOTIFY_DONE;
939
940 if (args->regs && user_mode_vm(args->regs))
941 return ret;
942
943 switch (val) {
944 case DIE_INT3:
945 if (kprobe_handler(args->regs))
946 ret = NOTIFY_STOP;
947 break;
948 case DIE_DEBUG:
949 if (post_kprobe_handler(args->regs))
950 ret = NOTIFY_STOP;
951 break;
952 case DIE_GPF:
953 /* kprobe_running() needs smp_processor_id() */
954 preempt_disable();
955 if (kprobe_running() &&
956 kprobe_fault_handler(args->regs, args->trapnr))
957 ret = NOTIFY_STOP;
958 preempt_enable();
959 break;
960 default:
961 break;
962 }
963 return ret;
964}
965
966int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
967{
968 struct jprobe *jp = container_of(p, struct jprobe, kp);
969 unsigned long addr;
970 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
971
972 kcb->jprobe_saved_regs = *regs;
973 kcb->jprobe_saved_sp = stack_addr(regs);
974 addr = (unsigned long)(kcb->jprobe_saved_sp);
975
976 /*
977 * As Linus pointed out, gcc assumes that the callee
978 * owns the argument space and could overwrite it, e.g.
979 * tailcall optimization. So, to be absolutely safe
980 * we also save and restore enough stack bytes to cover
981 * the argument area.
982 */
983 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
984 MIN_STACK_SIZE(addr));
985 regs->flags &= ~IF_MASK;
986 trace_hardirqs_off();
987 regs->ip = (unsigned long)(jp->entry);
988 return 1;
989}
990
991void __kprobes jprobe_return(void)
992{
993 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
994
995 asm volatile (
996#ifdef CONFIG_X86_64
997 " xchg %%rbx,%%rsp \n"
998#else
999 " xchgl %%ebx,%%esp \n"
1000#endif
1001 " int3 \n"
1002 " .globl jprobe_return_end\n"
1003 " jprobe_return_end: \n"
1004 " nop \n"::"b"
1005 (kcb->jprobe_saved_sp):"memory");
1006}
1007
1008int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1009{
1010 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1011 u8 *addr = (u8 *) (regs->ip - 1);
1012 struct jprobe *jp = container_of(p, struct jprobe, kp);
1013
1014 if ((addr > (u8 *) jprobe_return) &&
1015 (addr < (u8 *) jprobe_return_end)) {
1016 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1017 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1018 printk(KERN_ERR
1019 "current sp %p does not match saved sp %p\n",
1020 stack_addr(regs), kcb->jprobe_saved_sp);
1021 printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1022 show_registers(saved_regs);
1023 printk(KERN_ERR "Current registers\n");
1024 show_registers(regs);
1025 BUG();
1026 }
1027 *regs = kcb->jprobe_saved_regs;
1028 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1029 kcb->jprobes_stack,
1030 MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1031 preempt_enable_no_resched();
1032 return 1;
1033 }
1034 return 0;
1035}
1036
1037int __init arch_init_kprobes(void)
1038{
1039 return 0;
1040}
1041
1042int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1043{
1044 return 0;
1045}
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-11 07:40:20 -0500