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/hardirq.h>
  48 #include <linux/preempt.h>
  49 #include <linux/module.h>
  50 #include <linux/kdebug.h>
  51
  52 #include <asm/cacheflush.h>
  53 #include <asm/desc.h>
  54 #include <asm/pgtable.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/alternative.h>
  57
  58 void jprobe_return_end(void);
  59
  60 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  61 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  62
  63 #ifdef CONFIG_X86_64
  64 #define stack_addr(regs) ((unsigned long *)regs->sp)
  65 #else
  66 /*
  67  * "&regs->sp" looks wrong, but it's correct for x86_32.  x86_32 CPUs
  68  * don't save the ss and esp registers if the CPU is already in kernel
  69  * mode when it traps.  So for kprobes, regs->sp and regs->ss are not
  70  * the [nonexistent] saved stack pointer and ss register, but rather
  71  * the top 8 bytes of the pre-int3 stack.  So &regs->sp happens to
  72  * point to the top of the pre-int3 stack.
  73  */
  74 #define stack_addr(regs) ((unsigned long *)&regs->sp)
  75 #endif
  76
  77 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
  78         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  79           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  80           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  81           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  82          << (row % 32))
  83         /*
  84          * Undefined/reserved opcodes, conditional jump, Opcode Extension
  85          * Groups, and some special opcodes can not boost.
  86          * This is non-const to keep gcc from statically optimizing it out, as
  87          * variable_test_bit makes gcc think only *(unsigned long*) is used.
  88          */
  89 static 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          */
 110 };
 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          */
 132 };
 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          */
 154 };
 155 #undef W
 156
 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 */
 161 };
 162 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
 163
 164 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
 165 static void __kprobes set_jmp_op(void *from, void *to)
 166 {
 167         struct __arch_jmp_op {
 168                 char op;
 169                 s32 raddr;
 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;
 174 }
 175
 176 /*
 177  * Check for the REX prefix which can only exist on X86_64
 178  * X86_32 always returns 0
 179  */
 180 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
 181 {
 182 #ifdef CONFIG_X86_64
 183         if ((*insn & 0xf0) == 0x40)
 184                 return 1;
 185 #endif
 186         return 0;
 187 }
 188
 189 /*
 190  * Returns non-zero if opcode is boostable.
 191  * RIP relative instructions are adjusted at copying time in 64 bits mode
 192  */
 193 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
 194 {
 195         kprobe_opcode_t opcode;
 196         kprobe_opcode_t *orig_opcodes = opcodes;
 197
 198         if (search_exception_tables((unsigned long)opcodes))
 199                 return 0;       /* Page fault may occur on this address. */
 200
 201 retry:
 202         if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 203                 return 0;
 204         opcode = *(opcodes++);
 205
 206         /* 2nd-byte opcode */
 207         if (opcode == 0x0f) {
 208                 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 209                         return 0;
 210                 return test_bit(*opcodes,
 211                                 (unsigned long *)twobyte_is_boostable);
 212         }
 213
 214         switch (opcode & 0xf0) {
 215 #ifdef CONFIG_X86_64
 216         case 0x40:
 217                 goto retry; /* REX prefix is boostable */
 218 #endif
 219         case 0x60:
 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);
 224         case 0x70:
 225                 return 0; /* can't boost conditional jump */
 226         case 0xc0:
 227                 /* can't boost software-interruptions */
 228                 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
 229         case 0xd0:
 230                 /* can boost AA* and XLAT */
 231                 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
 232         case 0xe0:
 233                 /* can boost in/out and absolute jmps */
 234                 return ((opcode & 0x04) || opcode == 0xea);
 235         case 0xf0:
 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));
 240         default:
 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);
 246         }
 247 }
 248
 249 /*
 250  * Returns non-zero if opcode modifies the interrupt flag.
 251  */
 252 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
 253 {
 254         switch (*insn) {
 255         case 0xfa:              /* cli */
 256         case 0xfb:              /* sti */
 257         case 0xcf:              /* iret/iretd */
 258         case 0x9d:              /* popf/popfd */
 259                 return 1;
 260         }
 261
 262         /*
 263          * on X86_64, 0x40-0x4f are REX prefixes so we need to look
 264          * at the next byte instead.. but of course not recurse infinitely
 265          */
 266         if (is_REX_prefix(insn))
 267                 return is_IF_modifier(++insn);
 268
 269         return 0;
 270 }
 271
 272 /*
 273  * Adjust the displacement if the instruction uses the %rip-relative
 274  * addressing mode.
 275  * If it does, Return the address of the 32-bit displacement word.
 276  * If not, return null.
 277  * Only applicable to 64-bit x86.
 278  */
 279 static void __kprobes fix_riprel(struct kprobe *p)
 280 {
 281 #ifdef CONFIG_X86_64
 282         u8 *insn = p->ainsn.insn;
 283         s64 disp;
 284         int need_modrm;
 285
 286         /* Skip legacy instruction prefixes.  */
 287         while (1) {
 288                 switch (*insn) {
 289                 case 0x66:
 290                 case 0x67:
 291                 case 0x2e:
 292                 case 0x3e:
 293                 case 0x26:
 294                 case 0x64:
 295                 case 0x65:
 296                 case 0x36:
 297                 case 0xf0:
 298                 case 0xf3:
 299                 case 0xf2:
 300                         ++insn;
 301                         continue;
 302                 }
 303                 break;
 304         }
 305
 306         /* Skip REX instruction prefix.  */
 307         if (is_REX_prefix(insn))
 308                 ++insn;
 309
 310         if (*insn == 0x0f) {
 311                 /* Two-byte opcode.  */
 312                 ++insn;
 313                 need_modrm = test_bit(*insn,
 314                                       (unsigned long *)twobyte_has_modrm);
 315         } else
 316                 /* One-byte opcode.  */
 317                 need_modrm = test_bit(*insn,
 318                                       (unsigned long *)onebyte_has_modrm);
 319
 320         if (need_modrm) {
 321                 u8 modrm = *++insn;
 322                 if ((modrm & 0xc7) == 0x05) {
 323                         /* %rip+disp32 addressing mode */
 324                         /* Displacement follows ModRM byte.  */
 325                         ++insn;
 326                         /*
 327                          * The copied instruction uses the %rip-relative
 328                          * addressing mode.  Adjust the displacement for the
 329                          * difference between the original location of this
 330                          * instruction and the location of the copy that will
 331                          * actually be run.  The tricky bit here is making sure
 332                          * that the sign extension happens correctly in this
 333                          * calculation, since we need a signed 32-bit result to
 334                          * be sign-extended to 64 bits when it's added to the
 335                          * %rip value and yield the same 64-bit result that the
 336                          * sign-extension of the original signed 32-bit
 337                          * displacement would have given.
 338                          */
 339                         disp = (u8 *) p->addr + *((s32 *) insn) -
 340                                (u8 *) p->ainsn.insn;
 341                         BUG_ON((s64) (s32) disp != disp); /* Sanity check.  */
 342                         *(s32 *)insn = (s32) disp;
 343                 }
 344         }
 345 #endif
 346 }
 347
 348 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 349 {
 350         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 351
 352         fix_riprel(p);
 353
 354         if (can_boost(p->addr))
 355                 p->ainsn.boostable = 0;
 356         else
 357                 p->ainsn.boostable = -1;
 358
 359         p->opcode = *p->addr;
 360 }
 361
 362 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 363 {
 364         /* insn: must be on special executable page on x86. */
 365         p->ainsn.insn = get_insn_slot();
 366         if (!p->ainsn.insn)
 367                 return -ENOMEM;
 368         arch_copy_kprobe(p);
 369         return 0;
 370 }
 371
 372 void __kprobes arch_arm_kprobe(struct kprobe *p)
 373 {
 374         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 375 }
 376
 377 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 378 {
 379         text_poke(p->addr, &p->opcode, 1);
 380 }
 381
 382 void __kprobes arch_remove_kprobe(struct kprobe *p)
 383 {
 384         if (p->ainsn.insn) {
 385                 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
 386                 p->ainsn.insn = NULL;
 387         }
 388 }
 389
 390 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 391 {
 392         kcb->prev_kprobe.kp = kprobe_running();
 393         kcb->prev_kprobe.status = kcb->kprobe_status;
 394         kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
 395         kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
 396 }
 397
 398 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 399 {
 400         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 401         kcb->kprobe_status = kcb->prev_kprobe.status;
 402         kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
 403         kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
 404 }
 405
 406 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 407                                 struct kprobe_ctlblk *kcb)
 408 {
 409         __get_cpu_var(current_kprobe) = p;
 410         kcb->kprobe_saved_flags = kcb->kprobe_old_flags
 411                 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
 412         if (is_IF_modifier(p->ainsn.insn))
 413                 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
 414 }
 415
 416 static void __kprobes clear_btf(void)
 417 {
 418         if (test_thread_flag(TIF_DEBUGCTLMSR))
 419                 update_debugctlmsr(0);
 420 }
 421
 422 static void __kprobes restore_btf(void)
 423 {
 424         if (test_thread_flag(TIF_DEBUGCTLMSR))
 425                 update_debugctlmsr(current->thread.debugctlmsr);
 426 }
 427
 428 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 429 {
 430         clear_btf();
 431         regs->flags |= X86_EFLAGS_TF;
 432         regs->flags &= ~X86_EFLAGS_IF;
 433         /* single step inline if the instruction is an int3 */
 434         if (p->opcode == BREAKPOINT_INSTRUCTION)
 435                 regs->ip = (unsigned long)p->addr;
 436         else
 437                 regs->ip = (unsigned long)p->ainsn.insn;
 438 }
 439
 440 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 441                                       struct pt_regs *regs)
 442 {
 443         unsigned long *sara = stack_addr(regs);
 444
 445         ri->ret_addr = (kprobe_opcode_t *) *sara;
 446
 447         /* Replace the return addr with trampoline addr */
 448         *sara = (unsigned long) &kretprobe_trampoline;
 449 }
 450
 451 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
 452                                        struct kprobe_ctlblk *kcb)
 453 {
 454 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
 455         if (p->ainsn.boostable == 1 && !p->post_handler) {
 456                 /* Boost up -- we can execute copied instructions directly */
 457                 reset_current_kprobe();
 458                 regs->ip = (unsigned long)p->ainsn.insn;
 459                 preempt_enable_no_resched();
 460                 return;
 461         }
 462 #endif
 463         prepare_singlestep(p, regs);
 464         kcb->kprobe_status = KPROBE_HIT_SS;
 465 }
 466
 467 /*
 468  * We have reentered the kprobe_handler(), since another probe was hit while
 469  * within the handler. We save the original kprobes variables and just single
 470  * step on the instruction of the new probe without calling any user handlers.
 471  */
 472 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
 473                                     struct kprobe_ctlblk *kcb)
 474 {
 475         switch (kcb->kprobe_status) {
 476         case KPROBE_HIT_SSDONE:
 477 #ifdef CONFIG_X86_64
 478                 /* TODO: Provide re-entrancy from post_kprobes_handler() and
 479                  * avoid exception stack corruption while single-stepping on
 480                  * the instruction of the new probe.
 481                  */
 482                 arch_disarm_kprobe(p);
 483                 regs->ip = (unsigned long)p->addr;
 484                 reset_current_kprobe();
 485                 preempt_enable_no_resched();
 486                 break;
 487 #endif
 488         case KPROBE_HIT_ACTIVE:
 489                 save_previous_kprobe(kcb);
 490                 set_current_kprobe(p, regs, kcb);
 491                 kprobes_inc_nmissed_count(p);
 492                 prepare_singlestep(p, regs);
 493                 kcb->kprobe_status = KPROBE_REENTER;
 494                 break;
 495         case KPROBE_HIT_SS:
 496                 if (p == kprobe_running()) {
 497                         regs->flags &= ~X86_EFLAGS_TF;
 498                         regs->flags |= kcb->kprobe_saved_flags;
 499                         return 0;
 500                 } else {
 501                         /* A probe has been hit in the codepath leading up
 502                          * to, or just after, single-stepping of a probed
 503                          * instruction. This entire codepath should strictly
 504                          * reside in .kprobes.text section. Raise a warning
 505                          * to highlight this peculiar case.
 506                          */
 507                 }
 508         default:
 509                 /* impossible cases */
 510                 WARN_ON(1);
 511                 return 0;
 512         }
 513
 514         return 1;
 515 }
 516
 517 /*
 518  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 519  * remain disabled thorough out this function.
 520  */
 521 static int __kprobes kprobe_handler(struct pt_regs *regs)
 522 {
 523         kprobe_opcode_t *addr;
 524         struct kprobe *p;
 525         struct kprobe_ctlblk *kcb;
 526
 527         addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 528         if (*addr != BREAKPOINT_INSTRUCTION) {
 529                 /*
 530                  * The breakpoint instruction was removed right
 531                  * after we hit it.  Another cpu has removed
 532                  * either a probepoint or a debugger breakpoint
 533                  * at this address.  In either case, no further
 534                  * handling of this interrupt is appropriate.
 535                  * Back up over the (now missing) int3 and run
 536                  * the original instruction.
 537                  */
 538                 regs->ip = (unsigned long)addr;
 539                 return 1;
 540         }
 541
 542         /*
 543          * We don't want to be preempted for the entire
 544          * duration of kprobe processing. We conditionally
 545          * re-enable preemption at the end of this function,
 546          * and also in reenter_kprobe() and setup_singlestep().
 547          */
 548         preempt_disable();
 549
 550         kcb = get_kprobe_ctlblk();
 551         p = get_kprobe(addr);
 552
 553         if (p) {
 554                 if (kprobe_running()) {
 555                         if (reenter_kprobe(p, regs, kcb))
 556                                 return 1;
 557                 } else {
 558                         set_current_kprobe(p, regs, kcb);
 559                         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 560
 561                         /*
 562                          * If we have no pre-handler or it returned 0, we
 563                          * continue with normal processing.  If we have a
 564                          * pre-handler and it returned non-zero, it prepped
 565                          * for calling the break_handler below on re-entry
 566                          * for jprobe processing, so get out doing nothing
 567                          * more here.
 568                          */
 569                         if (!p->pre_handler || !p->pre_handler(p, regs))
 570                                 setup_singlestep(p, regs, kcb);
 571                         return 1;
 572                 }
 573         } else if (kprobe_running()) {
 574                 p = __get_cpu_var(current_kprobe);
 575                 if (p->break_handler && p->break_handler(p, regs)) {
 576                         setup_singlestep(p, regs, kcb);
 577                         return 1;
 578                 }
 579         } /* else: not a kprobe fault; let the kernel handle it */
 580
 581         preempt_enable_no_resched();
 582         return 0;
 583 }
 584
 585 /*
 586  * When a retprobed function returns, this code saves registers and
 587  * calls trampoline_handler() runs, which calls the kretprobe's handler.
 588  */
 589 static void __used __kprobes kretprobe_trampoline_holder(void)
 590 {
 591         asm volatile (
 592                         ".global kretprobe_trampoline\n"
 593                         "kretprobe_trampoline: \n"
 594 #ifdef CONFIG_X86_64
 595                         /* We don't bother saving the ss register */
 596                         "       pushq %rsp\n"
 597                         "       pushfq\n"
 598                         /*
 599                          * Skip cs, ip, orig_ax.
 600                          * trampoline_handler() will plug in these values
 601                          */
 602                         "       subq $24, %rsp\n"
 603                         "       pushq %rdi\n"
 604                         "       pushq %rsi\n"
 605                         "       pushq %rdx\n"
 606                         "       pushq %rcx\n"
 607                         "       pushq %rax\n"
 608                         "       pushq %r8\n"
 609                         "       pushq %r9\n"
 610                         "       pushq %r10\n"
 611                         "       pushq %r11\n"
 612                         "       pushq %rbx\n"
 613                         "       pushq %rbp\n"
 614                         "       pushq %r12\n"
 615                         "       pushq %r13\n"
 616                         "       pushq %r14\n"
 617                         "       pushq %r15\n"
 618                         "       movq %rsp, %rdi\n"
 619                         "       call trampoline_handler\n"
 620                         /* Replace saved sp with true return address. */
 621                         "       movq %rax, 152(%rsp)\n"
 622                         "       popq %r15\n"
 623                         "       popq %r14\n"
 624                         "       popq %r13\n"
 625                         "       popq %r12\n"
 626                         "       popq %rbp\n"
 627                         "       popq %rbx\n"
 628                         "       popq %r11\n"
 629                         "       popq %r10\n"
 630                         "       popq %r9\n"
 631                         "       popq %r8\n"
 632                         "       popq %rax\n"
 633                         "       popq %rcx\n"
 634                         "       popq %rdx\n"
 635                         "       popq %rsi\n"
 636                         "       popq %rdi\n"
 637                         /* Skip orig_ax, ip, cs */
 638                         "       addq $24, %rsp\n"
 639                         "       popfq\n"
 640 #else
 641                         "       pushf\n"
 642                         /*
 643                          * Skip cs, ip, orig_ax and gs.
 644                          * trampoline_handler() will plug in these values
 645                          */
 646                         "       subl $16, %esp\n"
 647                         "       pushl %fs\n"
 648                         "       pushl %es\n"
 649                         "       pushl %ds\n"
 650                         "       pushl %eax\n"
 651                         "       pushl %ebp\n"
 652                         "       pushl %edi\n"
 653                         "       pushl %esi\n"
 654                         "       pushl %edx\n"
 655                         "       pushl %ecx\n"
 656                         "       pushl %ebx\n"
 657                         "       movl %esp, %eax\n"
 658                         "       call trampoline_handler\n"
 659                         /* Move flags to cs */
 660                         "       movl 56(%esp), %edx\n"
 661                         "       movl %edx, 52(%esp)\n"
 662                         /* Replace saved flags with true return address. */
 663                         "       movl %eax, 56(%esp)\n"
 664                         "       popl %ebx\n"
 665                         "       popl %ecx\n"
 666                         "       popl %edx\n"
 667                         "       popl %esi\n"
 668                         "       popl %edi\n"
 669                         "       popl %ebp\n"
 670                         "       popl %eax\n"
 671                         /* Skip ds, es, fs, gs, orig_ax and ip */
 672                         "       addl $24, %esp\n"
 673                         "       popf\n"
 674 #endif
 675                         "       ret\n");
 676 }
 677
 678 /*
 679  * Called from kretprobe_trampoline
 680  */
 681 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 682 {
 683         struct kretprobe_instance *ri = NULL;
 684         struct hlist_head *head, empty_rp;
 685         struct hlist_node *node, *tmp;
 686         unsigned long flags, orig_ret_address = 0;
 687         unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
 688
 689         INIT_HLIST_HEAD(&empty_rp);
 690         kretprobe_hash_lock(current, &head, &flags);
 691         /* fixup registers */
 692 #ifdef CONFIG_X86_64
 693         regs->cs = __KERNEL_CS;
 694 #else
 695         regs->cs = __KERNEL_CS | get_kernel_rpl();
 696         regs->gs = 0;
 697 #endif
 698         regs->ip = trampoline_address;
 699         regs->orig_ax = ~0UL;
 700
 701         /*
 702          * It is possible to have multiple instances associated with a given
 703          * task either because multiple functions in the call path have
 704          * return probes installed on them, and/or more than one
 705          * return probe was registered for a target function.
 706          *
 707          * We can handle this because:
 708          *     - instances are always pushed into the head of the list
 709          *     - when multiple return probes are registered for the same
 710          *       function, the (chronologically) first instance's ret_addr
 711          *       will be the real return address, and all the rest will
 712          *       point to kretprobe_trampoline.
 713          */
 714         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 715                 if (ri->task != current)
 716                         /* another task is sharing our hash bucket */
 717                         continue;
 718
 719                 if (ri->rp && ri->rp->handler) {
 720                         __get_cpu_var(current_kprobe) = &ri->rp->kp;
 721                         get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
 722                         ri->rp->handler(ri, regs);
 723                         __get_cpu_var(current_kprobe) = NULL;
 724                 }
 725
 726                 orig_ret_address = (unsigned long)ri->ret_addr;
 727                 recycle_rp_inst(ri, &empty_rp);
 728
 729                 if (orig_ret_address != trampoline_address)
 730                         /*
 731                          * This is the real return address. Any other
 732                          * instances associated with this task are for
 733                          * other calls deeper on the call stack
 734                          */
 735                         break;
 736         }
 737
 738         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 739
 740         kretprobe_hash_unlock(current, &flags);
 741
 742         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 743                 hlist_del(&ri->hlist);
 744                 kfree(ri);
 745         }
 746         return (void *)orig_ret_address;
 747 }
 748
 749 /*
 750  * Called after single-stepping.  p->addr is the address of the
 751  * instruction whose first byte has been replaced by the "int 3"
 752  * instruction.  To avoid the SMP problems that can occur when we
 753  * temporarily put back the original opcode to single-step, we
 754  * single-stepped a copy of the instruction.  The address of this
 755  * copy is p->ainsn.insn.
 756  *
 757  * This function prepares to return from the post-single-step
 758  * interrupt.  We have to fix up the stack as follows:
 759  *
 760  * 0) Except in the case of absolute or indirect jump or call instructions,
 761  * the new ip is relative to the copied instruction.  We need to make
 762  * it relative to the original instruction.
 763  *
 764  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 765  * flags are set in the just-pushed flags, and may need to be cleared.
 766  *
 767  * 2) If the single-stepped instruction was a call, the return address
 768  * that is atop the stack is the address following the copied instruction.
 769  * We need to make it the address following the original instruction.
 770  *
 771  * If this is the first time we've single-stepped the instruction at
 772  * this probepoint, and the instruction is boostable, boost it: add a
 773  * jump instruction after the copied instruction, that jumps to the next
 774  * instruction after the probepoint.
 775  */
 776 static void __kprobes resume_execution(struct kprobe *p,
 777                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 778 {
 779         unsigned long *tos = stack_addr(regs);
 780         unsigned long copy_ip = (unsigned long)p->ainsn.insn;
 781         unsigned long orig_ip = (unsigned long)p->addr;
 782         kprobe_opcode_t *insn = p->ainsn.insn;
 783
 784         /*skip the REX prefix*/
 785         if (is_REX_prefix(insn))
 786                 insn++;
 787
 788         regs->flags &= ~X86_EFLAGS_TF;
 789         switch (*insn) {
 790         case 0x9c:      /* pushfl */
 791                 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
 792                 *tos |= kcb->kprobe_old_flags;
 793                 break;
 794         case 0xc2:      /* iret/ret/lret */
 795         case 0xc3:
 796         case 0xca:
 797         case 0xcb:
 798         case 0xcf:
 799         case 0xea:      /* jmp absolute -- ip is correct */
 800                 /* ip is already adjusted, no more changes required */
 801                 p->ainsn.boostable = 1;
 802                 goto no_change;
 803         case 0xe8:      /* call relative - Fix return addr */
 804                 *tos = orig_ip + (*tos - copy_ip);
 805                 break;
 806 #ifdef CONFIG_X86_32
 807         case 0x9a:      /* call absolute -- same as call absolute, indirect */
 808                 *tos = orig_ip + (*tos - copy_ip);
 809                 goto no_change;
 810 #endif
 811         case 0xff:
 812                 if ((insn[1] & 0x30) == 0x10) {
 813                         /*
 814                          * call absolute, indirect
 815                          * Fix return addr; ip is correct.
 816                          * But this is not boostable
 817                          */
 818                         *tos = orig_ip + (*tos - copy_ip);
 819                         goto no_change;
 820                 } else if (((insn[1] & 0x31) == 0x20) ||
 821                            ((insn[1] & 0x31) == 0x21)) {
 822                         /*
 823                          * jmp near and far, absolute indirect
 824                          * ip is correct. And this is boostable
 825                          */
 826                         p->ainsn.boostable = 1;
 827                         goto no_change;
 828                 }
 829         default:
 830                 break;
 831         }
 832
 833         if (p->ainsn.boostable == 0) {
 834                 if ((regs->ip > copy_ip) &&
 835                     (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
 836                         /*
 837                          * These instructions can be executed directly if it
 838                          * jumps back to correct address.
 839                          */
 840                         set_jmp_op((void *)regs->ip,
 841                                    (void *)orig_ip + (regs->ip - copy_ip));
 842                         p->ainsn.boostable = 1;
 843                 } else {
 844                         p->ainsn.boostable = -1;
 845                 }
 846         }
 847
 848         regs->ip += orig_ip - copy_ip;
 849
 850 no_change:
 851         restore_btf();
 852 }
 853
 854 /*
 855  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 856  * remain disabled thoroughout this function.
 857  */
 858 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
 859 {
 860         struct kprobe *cur = kprobe_running();
 861         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 862
 863         if (!cur)
 864                 return 0;
 865
 866         resume_execution(cur, regs, kcb);
 867         regs->flags |= kcb->kprobe_saved_flags;
 868
 869         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 870                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 871                 cur->post_handler(cur, regs, 0);
 872         }
 873
 874         /* Restore back the original saved kprobes variables and continue. */
 875         if (kcb->kprobe_status == KPROBE_REENTER) {
 876                 restore_previous_kprobe(kcb);
 877                 goto out;
 878         }
 879         reset_current_kprobe();
 880 out:
 881         preempt_enable_no_resched();
 882
 883         /*
 884          * if somebody else is singlestepping across a probe point, flags
 885          * will have TF set, in which case, continue the remaining processing
 886          * of do_debug, as if this is not a probe hit.
 887          */
 888         if (regs->flags & X86_EFLAGS_TF)
 889                 return 0;
 890
 891         return 1;
 892 }
 893
 894 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 895 {
 896         struct kprobe *cur = kprobe_running();
 897         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 898
 899         switch (kcb->kprobe_status) {
 900         case KPROBE_HIT_SS:
 901         case KPROBE_REENTER:
 902                 /*
 903                  * We are here because the instruction being single
 904                  * stepped caused a page fault. We reset the current
 905                  * kprobe and the ip points back to the probe address
 906                  * and allow the page fault handler to continue as a
 907                  * normal page fault.
 908                  */
 909                 regs->ip = (unsigned long)cur->addr;
 910                 regs->flags |= kcb->kprobe_old_flags;
 911                 if (kcb->kprobe_status == KPROBE_REENTER)
 912                         restore_previous_kprobe(kcb);
 913                 else
 914                         reset_current_kprobe();
 915                 preempt_enable_no_resched();
 916                 break;
 917         case KPROBE_HIT_ACTIVE:
 918         case KPROBE_HIT_SSDONE:
 919                 /*
 920                  * We increment the nmissed count for accounting,
 921                  * we can also use npre/npostfault count for accounting
 922                  * these specific fault cases.
 923                  */
 924                 kprobes_inc_nmissed_count(cur);
 925
 926                 /*
 927                  * We come here because instructions in the pre/post
 928                  * handler caused the page_fault, this could happen
 929                  * if handler tries to access user space by
 930                  * copy_from_user(), get_user() etc. Let the
 931                  * user-specified handler try to fix it first.
 932                  */
 933                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 934                         return 1;
 935
 936                 /*
 937                  * In case the user-specified fault handler returned
 938                  * zero, try to fix up.
 939                  */
 940                 if (fixup_exception(regs))
 941                         return 1;
 942
 943                 /*
 944                  * fixup routine could not handle it,
 945                  * Let do_page_fault() fix it.
 946                  */
 947                 break;
 948         default:
 949                 break;
 950         }
 951         return 0;
 952 }
 953
 954 /*
 955  * Wrapper routine for handling exceptions.
 956  */
 957 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 958                                        unsigned long val, void *data)
 959 {
 960         struct die_args *args = data;
 961         int ret = NOTIFY_DONE;
 962
 963         if (args->regs && user_mode_vm(args->regs))
 964                 return ret;
 965
 966         switch (val) {
 967         case DIE_INT3:
 968                 if (kprobe_handler(args->regs))
 969                         ret = NOTIFY_STOP;
 970                 break;
 971         case DIE_DEBUG:
 972                 if (post_kprobe_handler(args->regs))
 973                         ret = NOTIFY_STOP;
 974                 break;
 975         case DIE_GPF:
 976                 /*
 977                  * To be potentially processing a kprobe fault and to
 978                  * trust the result from kprobe_running(), we have
 979                  * be non-preemptible.
 980                  */
 981                 if (!preemptible() && kprobe_running() &&
 982                     kprobe_fault_handler(args->regs, args->trapnr))
 983                         ret = NOTIFY_STOP;
 984                 break;
 985         default:
 986                 break;
 987         }
 988         return ret;
 989 }
 990
 991 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 992 {
 993         struct jprobe *jp = container_of(p, struct jprobe, kp);
 994         unsigned long addr;
 995         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 996
 997         kcb->jprobe_saved_regs = *regs;
 998         kcb->jprobe_saved_sp = stack_addr(regs);
 999         addr = (unsigned long)(kcb->jprobe_saved_sp);
1000
1001         /*
1002          * As Linus pointed out, gcc assumes that the callee
1003          * owns the argument space and could overwrite it, e.g.
1004          * tailcall optimization. So, to be absolutely safe
1005          * we also save and restore enough stack bytes to cover
1006          * the argument area.
1007          */
1008         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1009                MIN_STACK_SIZE(addr));
1010         regs->flags &= ~X86_EFLAGS_IF;
1011         trace_hardirqs_off();
1012         regs->ip = (unsigned long)(jp->entry);
1013         return 1;
1014 }
1015
1016 void __kprobes jprobe_return(void)
1017 {
1018         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1019
1020         asm volatile (
1021 #ifdef CONFIG_X86_64
1022                         "       xchg   %%rbx,%%rsp      \n"
1023 #else
1024                         "       xchgl   %%ebx,%%esp     \n"
1025 #endif
1026                         "       int3                    \n"
1027                         "       .globl jprobe_return_end\n"
1028                         "       jprobe_return_end:      \n"
1029                         "       nop                     \n"::"b"
1030                         (kcb->jprobe_saved_sp):"memory");
1031 }
1032
1033 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1034 {
1035         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1036         u8 *addr = (u8 *) (regs->ip - 1);
1037         struct jprobe *jp = container_of(p, struct jprobe, kp);
1038
1039         if ((addr > (u8 *) jprobe_return) &&
1040             (addr < (u8 *) jprobe_return_end)) {
1041                 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1042                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1043                         printk(KERN_ERR
1044                                "current sp %p does not match saved sp %p\n",
1045                                stack_addr(regs), kcb->jprobe_saved_sp);
1046                         printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1047                         show_registers(saved_regs);
1048                         printk(KERN_ERR "Current registers\n");
1049                         show_registers(regs);
1050                         BUG();
1051                 }
1052                 *regs = kcb->jprobe_saved_regs;
1053                 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1054                        kcb->jprobes_stack,
1055                        MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1056                 preempt_enable_no_resched();
1057                 return 1;
1058         }
1059         return 0;
1060 }
1061
1062 int __init arch_init_kprobes(void)
1063 {
1064         return 0;
1065 }
1066
1067 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1068 {
1069         return 0;
1070 }