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          */
  87 static const u32 twobyte_is_boostable[256 / 32] = {
  88         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  89         /*      ----------------------------------------------          */
  90         W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
  91         W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
  92         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
  93         W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
  94         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
  95         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
  96         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
  97         W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
  98         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
  99         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
 100         W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
 101         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
 102         W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
 103         W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
 104         W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
 105         W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
 106         /*      -----------------------------------------------         */
 107         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 108 };
 109 static const u32 onebyte_has_modrm[256 / 32] = {
 110         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 111         /*      -----------------------------------------------         */
 112         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
 113         W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
 114         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
 115         W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
 116         W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
 117         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
 118         W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
 119         W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
 120         W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
 121         W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
 122         W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
 123         W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
 124         W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
 125         W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
 126         W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
 127         W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1)   /* f0 */
 128         /*      -----------------------------------------------         */
 129         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 130 };
 131 static const u32 twobyte_has_modrm[256 / 32] = {
 132         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 133         /*      -----------------------------------------------         */
 134         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
 135         W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
 136         W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
 137         W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
 138         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
 139         W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
 140         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
 141         W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
 142         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
 143         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
 144         W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
 145         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
 146         W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
 147         W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
 148         W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
 149         W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0)   /* ff */
 150         /*      -----------------------------------------------         */
 151         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 152 };
 153 #undef W
 154
 155 struct kretprobe_blackpoint kretprobe_blacklist[] = {
 156         {"__switch_to", }, /* This function switches only current task, but
 157                               doesn't switch kernel stack.*/
 158         {NULL, NULL}    /* Terminator */
 159 };
 160 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
 161
 162 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
 163 static void __kprobes set_jmp_op(void *from, void *to)
 164 {
 165         struct __arch_jmp_op {
 166                 char op;
 167                 s32 raddr;
 168         } __attribute__((packed)) * jop;
 169         jop = (struct __arch_jmp_op *)from;
 170         jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
 171         jop->op = RELATIVEJUMP_INSTRUCTION;
 172 }
 173
 174 /*
 175  * Check for the REX prefix which can only exist on X86_64
 176  * X86_32 always returns 0
 177  */
 178 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
 179 {
 180 #ifdef CONFIG_X86_64
 181         if ((*insn & 0xf0) == 0x40)
 182                 return 1;
 183 #endif
 184         return 0;
 185 }
 186
 187 /*
 188  * Returns non-zero if opcode is boostable.
 189  * RIP relative instructions are adjusted at copying time in 64 bits mode
 190  */
 191 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
 192 {
 193         kprobe_opcode_t opcode;
 194         kprobe_opcode_t *orig_opcodes = opcodes;
 195
 196         if (search_exception_tables((unsigned long)opcodes))
 197                 return 0;       /* Page fault may occur on this address. */
 198
 199 retry:
 200         if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 201                 return 0;
 202         opcode = *(opcodes++);
 203
 204         /* 2nd-byte opcode */
 205         if (opcode == 0x0f) {
 206                 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 207                         return 0;
 208                 return test_bit(*opcodes,
 209                                 (unsigned long *)twobyte_is_boostable);
 210         }
 211
 212         switch (opcode & 0xf0) {
 213 #ifdef CONFIG_X86_64
 214         case 0x40:
 215                 goto retry; /* REX prefix is boostable */
 216 #endif
 217         case 0x60:
 218                 if (0x63 < opcode && opcode < 0x67)
 219                         goto retry; /* prefixes */
 220                 /* can't boost Address-size override and bound */
 221                 return (opcode != 0x62 && opcode != 0x67);
 222         case 0x70:
 223                 return 0; /* can't boost conditional jump */
 224         case 0xc0:
 225                 /* can't boost software-interruptions */
 226                 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
 227         case 0xd0:
 228                 /* can boost AA* and XLAT */
 229                 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
 230         case 0xe0:
 231                 /* can boost in/out and absolute jmps */
 232                 return ((opcode & 0x04) || opcode == 0xea);
 233         case 0xf0:
 234                 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
 235                         goto retry; /* lock/rep(ne) prefix */
 236                 /* clear and set flags are boostable */
 237                 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
 238         default:
 239                 /* segment override prefixes are boostable */
 240                 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
 241                         goto retry; /* prefixes */
 242                 /* CS override prefix and call are not boostable */
 243                 return (opcode != 0x2e && opcode != 0x9a);
 244         }
 245 }
 246
 247 /*
 248  * Returns non-zero if opcode modifies the interrupt flag.
 249  */
 250 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
 251 {
 252         switch (*insn) {
 253         case 0xfa:              /* cli */
 254         case 0xfb:              /* sti */
 255         case 0xcf:              /* iret/iretd */
 256         case 0x9d:              /* popf/popfd */
 257                 return 1;
 258         }
 259
 260         /*
 261          * on X86_64, 0x40-0x4f are REX prefixes so we need to look
 262          * at the next byte instead.. but of course not recurse infinitely
 263          */
 264         if (is_REX_prefix(insn))
 265                 return is_IF_modifier(++insn);
 266
 267         return 0;
 268 }
 269
 270 /*
 271  * Adjust the displacement if the instruction uses the %rip-relative
 272  * addressing mode.
 273  * If it does, Return the address of the 32-bit displacement word.
 274  * If not, return null.
 275  * Only applicable to 64-bit x86.
 276  */
 277 static void __kprobes fix_riprel(struct kprobe *p)
 278 {
 279 #ifdef CONFIG_X86_64
 280         u8 *insn = p->ainsn.insn;
 281         s64 disp;
 282         int need_modrm;
 283
 284         /* Skip legacy instruction prefixes.  */
 285         while (1) {
 286                 switch (*insn) {
 287                 case 0x66:
 288                 case 0x67:
 289                 case 0x2e:
 290                 case 0x3e:
 291                 case 0x26:
 292                 case 0x64:
 293                 case 0x65:
 294                 case 0x36:
 295                 case 0xf0:
 296                 case 0xf3:
 297                 case 0xf2:
 298                         ++insn;
 299                         continue;
 300                 }
 301                 break;
 302         }
 303
 304         /* Skip REX instruction prefix.  */
 305         if (is_REX_prefix(insn))
 306                 ++insn;
 307
 308         if (*insn == 0x0f) {
 309                 /* Two-byte opcode.  */
 310                 ++insn;
 311                 need_modrm = test_bit(*insn,
 312                                       (unsigned long *)twobyte_has_modrm);
 313         } else
 314                 /* One-byte opcode.  */
 315                 need_modrm = test_bit(*insn,
 316                                       (unsigned long *)onebyte_has_modrm);
 317
 318         if (need_modrm) {
 319                 u8 modrm = *++insn;
 320                 if ((modrm & 0xc7) == 0x05) {
 321                         /* %rip+disp32 addressing mode */
 322                         /* Displacement follows ModRM byte.  */
 323                         ++insn;
 324                         /*
 325                          * The copied instruction uses the %rip-relative
 326                          * addressing mode.  Adjust the displacement for the
 327                          * difference between the original location of this
 328                          * instruction and the location of the copy that will
 329                          * actually be run.  The tricky bit here is making sure
 330                          * that the sign extension happens correctly in this
 331                          * calculation, since we need a signed 32-bit result to
 332                          * be sign-extended to 64 bits when it's added to the
 333                          * %rip value and yield the same 64-bit result that the
 334                          * sign-extension of the original signed 32-bit
 335                          * displacement would have given.
 336                          */
 337                         disp = (u8 *) p->addr + *((s32 *) insn) -
 338                                (u8 *) p->ainsn.insn;
 339                         BUG_ON((s64) (s32) disp != disp); /* Sanity check.  */
 340                         *(s32 *)insn = (s32) disp;
 341                 }
 342         }
 343 #endif
 344 }
 345
 346 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 347 {
 348         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 349
 350         fix_riprel(p);
 351
 352         if (can_boost(p->addr))
 353                 p->ainsn.boostable = 0;
 354         else
 355                 p->ainsn.boostable = -1;
 356
 357         p->opcode = *p->addr;
 358 }
 359
 360 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 361 {
 362         /* insn: must be on special executable page on x86. */
 363         p->ainsn.insn = get_insn_slot();
 364         if (!p->ainsn.insn)
 365                 return -ENOMEM;
 366         arch_copy_kprobe(p);
 367         return 0;
 368 }
 369
 370 void __kprobes arch_arm_kprobe(struct kprobe *p)
 371 {
 372         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 373 }
 374
 375 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 376 {
 377         text_poke(p->addr, &p->opcode, 1);
 378 }
 379
 380 void __kprobes arch_remove_kprobe(struct kprobe *p)
 381 {
 382         if (p->ainsn.insn) {
 383                 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
 384                 p->ainsn.insn = NULL;
 385         }
 386 }
 387
 388 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 389 {
 390         kcb->prev_kprobe.kp = kprobe_running();
 391         kcb->prev_kprobe.status = kcb->kprobe_status;
 392         kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
 393         kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
 394 }
 395
 396 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 397 {
 398         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 399         kcb->kprobe_status = kcb->prev_kprobe.status;
 400         kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
 401         kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
 402 }
 403
 404 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 405                                 struct kprobe_ctlblk *kcb)
 406 {
 407         __get_cpu_var(current_kprobe) = p;
 408         kcb->kprobe_saved_flags = kcb->kprobe_old_flags
 409                 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
 410         if (is_IF_modifier(p->ainsn.insn))
 411                 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
 412 }
 413
 414 static void __kprobes clear_btf(void)
 415 {
 416         if (test_thread_flag(TIF_DEBUGCTLMSR))
 417                 update_debugctlmsr(0);
 418 }
 419
 420 static void __kprobes restore_btf(void)
 421 {
 422         if (test_thread_flag(TIF_DEBUGCTLMSR))
 423                 update_debugctlmsr(current->thread.debugctlmsr);
 424 }
 425
 426 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 427 {
 428         clear_btf();
 429         regs->flags |= X86_EFLAGS_TF;
 430         regs->flags &= ~X86_EFLAGS_IF;
 431         /* single step inline if the instruction is an int3 */
 432         if (p->opcode == BREAKPOINT_INSTRUCTION)
 433                 regs->ip = (unsigned long)p->addr;
 434         else
 435                 regs->ip = (unsigned long)p->ainsn.insn;
 436 }
 437
 438 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 439                                       struct pt_regs *regs)
 440 {
 441         unsigned long *sara = stack_addr(regs);
 442
 443         ri->ret_addr = (kprobe_opcode_t *) *sara;
 444
 445         /* Replace the return addr with trampoline addr */
 446         *sara = (unsigned long) &kretprobe_trampoline;
 447 }
 448
 449 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
 450                                        struct kprobe_ctlblk *kcb)
 451 {
 452 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
 453         if (p->ainsn.boostable == 1 && !p->post_handler) {
 454                 /* Boost up -- we can execute copied instructions directly */
 455                 reset_current_kprobe();
 456                 regs->ip = (unsigned long)p->ainsn.insn;
 457                 preempt_enable_no_resched();
 458                 return;
 459         }
 460 #endif
 461         prepare_singlestep(p, regs);
 462         kcb->kprobe_status = KPROBE_HIT_SS;
 463 }
 464
 465 /*
 466  * We have reentered the kprobe_handler(), since another probe was hit while
 467  * within the handler. We save the original kprobes variables and just single
 468  * step on the instruction of the new probe without calling any user handlers.
 469  */
 470 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
 471                                     struct kprobe_ctlblk *kcb)
 472 {
 473         switch (kcb->kprobe_status) {
 474         case KPROBE_HIT_SSDONE:
 475 #ifdef CONFIG_X86_64
 476                 /* TODO: Provide re-entrancy from post_kprobes_handler() and
 477                  * avoid exception stack corruption while single-stepping on
 478                  * the instruction of the new probe.
 479                  */
 480                 arch_disarm_kprobe(p);
 481                 regs->ip = (unsigned long)p->addr;
 482                 reset_current_kprobe();
 483                 preempt_enable_no_resched();
 484                 break;
 485 #endif
 486         case KPROBE_HIT_ACTIVE:
 487                 save_previous_kprobe(kcb);
 488                 set_current_kprobe(p, regs, kcb);
 489                 kprobes_inc_nmissed_count(p);
 490                 prepare_singlestep(p, regs);
 491                 kcb->kprobe_status = KPROBE_REENTER;
 492                 break;
 493         case KPROBE_HIT_SS:
 494                 if (p == kprobe_running()) {
 495                         regs->flags &= ~X86_EFLAGS_TF;
 496                         regs->flags |= kcb->kprobe_saved_flags;
 497                         return 0;
 498                 } else {
 499                         /* A probe has been hit in the codepath leading up
 500                          * to, or just after, single-stepping of a probed
 501                          * instruction. This entire codepath should strictly
 502                          * reside in .kprobes.text section. Raise a warning
 503                          * to highlight this peculiar case.
 504                          */
 505                 }
 506         default:
 507                 /* impossible cases */
 508                 WARN_ON(1);
 509                 return 0;
 510         }
 511
 512         return 1;
 513 }
 514
 515 /*
 516  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 517  * remain disabled thorough out this function.
 518  */
 519 static int __kprobes kprobe_handler(struct pt_regs *regs)
 520 {
 521         kprobe_opcode_t *addr;
 522         struct kprobe *p;
 523         struct kprobe_ctlblk *kcb;
 524
 525         addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 526         if (*addr != BREAKPOINT_INSTRUCTION) {
 527                 /*
 528                  * The breakpoint instruction was removed right
 529                  * after we hit it.  Another cpu has removed
 530                  * either a probepoint or a debugger breakpoint
 531                  * at this address.  In either case, no further
 532                  * handling of this interrupt is appropriate.
 533                  * Back up over the (now missing) int3 and run
 534                  * the original instruction.
 535                  */
 536                 regs->ip = (unsigned long)addr;
 537                 return 1;
 538         }
 539
 540         /*
 541          * We don't want to be preempted for the entire
 542          * duration of kprobe processing. We conditionally
 543          * re-enable preemption at the end of this function,
 544          * and also in reenter_kprobe() and setup_singlestep().
 545          */
 546         preempt_disable();
 547
 548         kcb = get_kprobe_ctlblk();
 549         p = get_kprobe(addr);
 550
 551         if (p) {
 552                 if (kprobe_running()) {
 553                         if (reenter_kprobe(p, regs, kcb))
 554                                 return 1;
 555                 } else {
 556                         set_current_kprobe(p, regs, kcb);
 557                         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 558
 559                         /*
 560                          * If we have no pre-handler or it returned 0, we
 561                          * continue with normal processing.  If we have a
 562                          * pre-handler and it returned non-zero, it prepped
 563                          * for calling the break_handler below on re-entry
 564                          * for jprobe processing, so get out doing nothing
 565                          * more here.
 566                          */
 567                         if (!p->pre_handler || !p->pre_handler(p, regs))
 568                                 setup_singlestep(p, regs, kcb);
 569                         return 1;
 570                 }
 571         } else if (kprobe_running()) {
 572                 p = __get_cpu_var(current_kprobe);
 573                 if (p->break_handler && p->break_handler(p, regs)) {
 574                         setup_singlestep(p, regs, kcb);
 575                         return 1;
 576                 }
 577         } /* else: not a kprobe fault; let the kernel handle it */
 578
 579         preempt_enable_no_resched();
 580         return 0;
 581 }
 582
 583 /*
 584  * When a retprobed function returns, this code saves registers and
 585  * calls trampoline_handler() runs, which calls the kretprobe's handler.
 586  */
 587 static void __used __kprobes kretprobe_trampoline_holder(void)
 588 {
 589         asm volatile (
 590                         ".global kretprobe_trampoline\n"
 591                         "kretprobe_trampoline: \n"
 592 #ifdef CONFIG_X86_64
 593                         /* We don't bother saving the ss register */
 594                         "       pushq %rsp\n"
 595                         "       pushfq\n"
 596                         /*
 597                          * Skip cs, ip, orig_ax.
 598                          * trampoline_handler() will plug in these values
 599                          */
 600                         "       subq $24, %rsp\n"
 601                         "       pushq %rdi\n"
 602                         "       pushq %rsi\n"
 603                         "       pushq %rdx\n"
 604                         "       pushq %rcx\n"
 605                         "       pushq %rax\n"
 606                         "       pushq %r8\n"
 607                         "       pushq %r9\n"
 608                         "       pushq %r10\n"
 609                         "       pushq %r11\n"
 610                         "       pushq %rbx\n"
 611                         "       pushq %rbp\n"
 612                         "       pushq %r12\n"
 613                         "       pushq %r13\n"
 614                         "       pushq %r14\n"
 615                         "       pushq %r15\n"
 616                         "       movq %rsp, %rdi\n"
 617                         "       call trampoline_handler\n"
 618                         /* Replace saved sp with true return address. */
 619                         "       movq %rax, 152(%rsp)\n"
 620                         "       popq %r15\n"
 621                         "       popq %r14\n"
 622                         "       popq %r13\n"
 623                         "       popq %r12\n"
 624                         "       popq %rbp\n"
 625                         "       popq %rbx\n"
 626                         "       popq %r11\n"
 627                         "       popq %r10\n"
 628                         "       popq %r9\n"
 629                         "       popq %r8\n"
 630                         "       popq %rax\n"
 631                         "       popq %rcx\n"
 632                         "       popq %rdx\n"
 633                         "       popq %rsi\n"
 634                         "       popq %rdi\n"
 635                         /* Skip orig_ax, ip, cs */
 636                         "       addq $24, %rsp\n"
 637                         "       popfq\n"
 638 #else
 639                         "       pushf\n"
 640                         /*
 641                          * Skip cs, ip, orig_ax and gs.
 642                          * trampoline_handler() will plug in these values
 643                          */
 644                         "       subl $16, %esp\n"
 645                         "       pushl %fs\n"
 646                         "       pushl %es\n"
 647                         "       pushl %ds\n"
 648                         "       pushl %eax\n"
 649                         "       pushl %ebp\n"
 650                         "       pushl %edi\n"
 651                         "       pushl %esi\n"
 652                         "       pushl %edx\n"
 653                         "       pushl %ecx\n"
 654                         "       pushl %ebx\n"
 655                         "       movl %esp, %eax\n"
 656                         "       call trampoline_handler\n"
 657                         /* Move flags to cs */
 658                         "       movl 56(%esp), %edx\n"
 659                         "       movl %edx, 52(%esp)\n"
 660                         /* Replace saved flags with true return address. */
 661                         "       movl %eax, 56(%esp)\n"
 662                         "       popl %ebx\n"
 663                         "       popl %ecx\n"
 664                         "       popl %edx\n"
 665                         "       popl %esi\n"
 666                         "       popl %edi\n"
 667                         "       popl %ebp\n"
 668                         "       popl %eax\n"
 669                         /* Skip ds, es, fs, gs, orig_ax and ip */
 670                         "       addl $24, %esp\n"
 671                         "       popf\n"
 672 #endif
 673                         "       ret\n");
 674 }
 675
 676 /*
 677  * Called from kretprobe_trampoline
 678  */
 679 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 680 {
 681         struct kretprobe_instance *ri = NULL;
 682         struct hlist_head *head, empty_rp;
 683         struct hlist_node *node, *tmp;
 684         unsigned long flags, orig_ret_address = 0;
 685         unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
 686
 687         INIT_HLIST_HEAD(&empty_rp);
 688         kretprobe_hash_lock(current, &head, &flags);
 689         /* fixup registers */
 690 #ifdef CONFIG_X86_64
 691         regs->cs = __KERNEL_CS;
 692 #else
 693         regs->cs = __KERNEL_CS | get_kernel_rpl();
 694         regs->gs = 0;
 695 #endif
 696         regs->ip = trampoline_address;
 697         regs->orig_ax = ~0UL;
 698
 699         /*
 700          * It is possible to have multiple instances associated with a given
 701          * task either because multiple functions in the call path have
 702          * return probes installed on them, and/or more than one
 703          * return probe was registered for a target function.
 704          *
 705          * We can handle this because:
 706          *     - instances are always pushed into the head of the list
 707          *     - when multiple return probes are registered for the same
 708          *       function, the (chronologically) first instance's ret_addr
 709          *       will be the real return address, and all the rest will
 710          *       point to kretprobe_trampoline.
 711          */
 712         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 713                 if (ri->task != current)
 714                         /* another task is sharing our hash bucket */
 715                         continue;
 716
 717                 if (ri->rp && ri->rp->handler) {
 718                         __get_cpu_var(current_kprobe) = &ri->rp->kp;
 719                         get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
 720                         ri->rp->handler(ri, regs);
 721                         __get_cpu_var(current_kprobe) = NULL;
 722                 }
 723
 724                 orig_ret_address = (unsigned long)ri->ret_addr;
 725                 recycle_rp_inst(ri, &empty_rp);
 726
 727                 if (orig_ret_address != trampoline_address)
 728                         /*
 729                          * This is the real return address. Any other
 730                          * instances associated with this task are for
 731                          * other calls deeper on the call stack
 732                          */
 733                         break;
 734         }
 735
 736         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 737
 738         kretprobe_hash_unlock(current, &flags);
 739
 740         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 741                 hlist_del(&ri->hlist);
 742                 kfree(ri);
 743         }
 744         return (void *)orig_ret_address;
 745 }
 746
 747 /*
 748  * Called after single-stepping.  p->addr is the address of the
 749  * instruction whose first byte has been replaced by the "int 3"
 750  * instruction.  To avoid the SMP problems that can occur when we
 751  * temporarily put back the original opcode to single-step, we
 752  * single-stepped a copy of the instruction.  The address of this
 753  * copy is p->ainsn.insn.
 754  *
 755  * This function prepares to return from the post-single-step
 756  * interrupt.  We have to fix up the stack as follows:
 757  *
 758  * 0) Except in the case of absolute or indirect jump or call instructions,
 759  * the new ip is relative to the copied instruction.  We need to make
 760  * it relative to the original instruction.
 761  *
 762  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 763  * flags are set in the just-pushed flags, and may need to be cleared.
 764  *
 765  * 2) If the single-stepped instruction was a call, the return address
 766  * that is atop the stack is the address following the copied instruction.
 767  * We need to make it the address following the original instruction.
 768  *
 769  * If this is the first time we've single-stepped the instruction at
 770  * this probepoint, and the instruction is boostable, boost it: add a
 771  * jump instruction after the copied instruction, that jumps to the next
 772  * instruction after the probepoint.
 773  */
 774 static void __kprobes resume_execution(struct kprobe *p,
 775                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 776 {
 777         unsigned long *tos = stack_addr(regs);
 778         unsigned long copy_ip = (unsigned long)p->ainsn.insn;
 779         unsigned long orig_ip = (unsigned long)p->addr;
 780         kprobe_opcode_t *insn = p->ainsn.insn;
 781
 782         /*skip the REX prefix*/
 783         if (is_REX_prefix(insn))
 784                 insn++;
 785
 786         regs->flags &= ~X86_EFLAGS_TF;
 787         switch (*insn) {
 788         case 0x9c:      /* pushfl */
 789                 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
 790                 *tos |= kcb->kprobe_old_flags;
 791                 break;
 792         case 0xc2:      /* iret/ret/lret */
 793         case 0xc3:
 794         case 0xca:
 795         case 0xcb:
 796         case 0xcf:
 797         case 0xea:      /* jmp absolute -- ip is correct */
 798                 /* ip is already adjusted, no more changes required */
 799                 p->ainsn.boostable = 1;
 800                 goto no_change;
 801         case 0xe8:      /* call relative - Fix return addr */
 802                 *tos = orig_ip + (*tos - copy_ip);
 803                 break;
 804 #ifdef CONFIG_X86_32
 805         case 0x9a:      /* call absolute -- same as call absolute, indirect */
 806                 *tos = orig_ip + (*tos - copy_ip);
 807                 goto no_change;
 808 #endif
 809         case 0xff:
 810                 if ((insn[1] & 0x30) == 0x10) {
 811                         /*
 812                          * call absolute, indirect
 813                          * Fix return addr; ip is correct.
 814                          * But this is not boostable
 815                          */
 816                         *tos = orig_ip + (*tos - copy_ip);
 817                         goto no_change;
 818                 } else if (((insn[1] & 0x31) == 0x20) ||
 819                            ((insn[1] & 0x31) == 0x21)) {
 820                         /*
 821                          * jmp near and far, absolute indirect
 822                          * ip is correct. And this is boostable
 823                          */
 824                         p->ainsn.boostable = 1;
 825                         goto no_change;
 826                 }
 827         default:
 828                 break;
 829         }
 830
 831         if (p->ainsn.boostable == 0) {
 832                 if ((regs->ip > copy_ip) &&
 833                     (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
 834                         /*
 835                          * These instructions can be executed directly if it
 836                          * jumps back to correct address.
 837                          */
 838                         set_jmp_op((void *)regs->ip,
 839                                    (void *)orig_ip + (regs->ip - copy_ip));
 840                         p->ainsn.boostable = 1;
 841                 } else {
 842                         p->ainsn.boostable = -1;
 843                 }
 844         }
 845
 846         regs->ip += orig_ip - copy_ip;
 847
 848 no_change:
 849         restore_btf();
 850 }
 851
 852 /*
 853  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 854  * remain disabled thoroughout this function.
 855  */
 856 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
 857 {
 858         struct kprobe *cur = kprobe_running();
 859         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 860
 861         if (!cur)
 862                 return 0;
 863
 864         resume_execution(cur, regs, kcb);
 865         regs->flags |= kcb->kprobe_saved_flags;
 866
 867         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 868                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 869                 cur->post_handler(cur, regs, 0);
 870         }
 871
 872         /* Restore back the original saved kprobes variables and continue. */
 873         if (kcb->kprobe_status == KPROBE_REENTER) {
 874                 restore_previous_kprobe(kcb);
 875                 goto out;
 876         }
 877         reset_current_kprobe();
 878 out:
 879         preempt_enable_no_resched();
 880
 881         /*
 882          * if somebody else is singlestepping across a probe point, flags
 883          * will have TF set, in which case, continue the remaining processing
 884          * of do_debug, as if this is not a probe hit.
 885          */
 886         if (regs->flags & X86_EFLAGS_TF)
 887                 return 0;
 888
 889         return 1;
 890 }
 891
 892 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 893 {
 894         struct kprobe *cur = kprobe_running();
 895         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 896
 897         switch (kcb->kprobe_status) {
 898         case KPROBE_HIT_SS:
 899         case KPROBE_REENTER:
 900                 /*
 901                  * We are here because the instruction being single
 902                  * stepped caused a page fault. We reset the current
 903                  * kprobe and the ip points back to the probe address
 904                  * and allow the page fault handler to continue as a
 905                  * normal page fault.
 906                  */
 907                 regs->ip = (unsigned long)cur->addr;
 908                 regs->flags |= kcb->kprobe_old_flags;
 909                 if (kcb->kprobe_status == KPROBE_REENTER)
 910                         restore_previous_kprobe(kcb);
 911                 else
 912                         reset_current_kprobe();
 913                 preempt_enable_no_resched();
 914                 break;
 915         case KPROBE_HIT_ACTIVE:
 916         case KPROBE_HIT_SSDONE:
 917                 /*
 918                  * We increment the nmissed count for accounting,
 919                  * we can also use npre/npostfault count for accounting
 920                  * these specific fault cases.
 921                  */
 922                 kprobes_inc_nmissed_count(cur);
 923
 924                 /*
 925                  * We come here because instructions in the pre/post
 926                  * handler caused the page_fault, this could happen
 927                  * if handler tries to access user space by
 928                  * copy_from_user(), get_user() etc. Let the
 929                  * user-specified handler try to fix it first.
 930                  */
 931                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 932                         return 1;
 933
 934                 /*
 935                  * In case the user-specified fault handler returned
 936                  * zero, try to fix up.
 937                  */
 938                 if (fixup_exception(regs))
 939                         return 1;
 940
 941                 /*
 942                  * fixup routine could not handle it,
 943                  * Let do_page_fault() fix it.
 944                  */
 945                 break;
 946         default:
 947                 break;
 948         }
 949         return 0;
 950 }
 951
 952 /*
 953  * Wrapper routine for handling exceptions.
 954  */
 955 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 956                                        unsigned long val, void *data)
 957 {
 958         struct die_args *args = data;
 959         int ret = NOTIFY_DONE;
 960
 961         if (args->regs && user_mode_vm(args->regs))
 962                 return ret;
 963
 964         switch (val) {
 965         case DIE_INT3:
 966                 if (kprobe_handler(args->regs))
 967                         ret = NOTIFY_STOP;
 968                 break;
 969         case DIE_DEBUG:
 970                 if (post_kprobe_handler(args->regs))
 971                         ret = NOTIFY_STOP;
 972                 break;
 973         case DIE_GPF:
 974                 /*
 975                  * To be potentially processing a kprobe fault and to
 976                  * trust the result from kprobe_running(), we have
 977                  * be non-preemptible.
 978                  */
 979                 if (!preemptible() && kprobe_running() &&
 980                     kprobe_fault_handler(args->regs, args->trapnr))
 981                         ret = NOTIFY_STOP;
 982                 break;
 983         default:
 984                 break;
 985         }
 986         return ret;
 987 }
 988
 989 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 990 {
 991         struct jprobe *jp = container_of(p, struct jprobe, kp);
 992         unsigned long addr;
 993         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 994
 995         kcb->jprobe_saved_regs = *regs;
 996         kcb->jprobe_saved_sp = stack_addr(regs);
 997         addr = (unsigned long)(kcb->jprobe_saved_sp);
 998
 999         /*
1000          * As Linus pointed out, gcc assumes that the callee
1001          * owns the argument space and could overwrite it, e.g.
1002          * tailcall optimization. So, to be absolutely safe
1003          * we also save and restore enough stack bytes to cover
1004          * the argument area.
1005          */
1006         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
1007                MIN_STACK_SIZE(addr));
1008         regs->flags &= ~X86_EFLAGS_IF;
1009         trace_hardirqs_off();
1010         regs->ip = (unsigned long)(jp->entry);
1011         return 1;
1012 }
1013
1014 void __kprobes jprobe_return(void)
1015 {
1016         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1017
1018         asm volatile (
1019 #ifdef CONFIG_X86_64
1020                         "       xchg   %%rbx,%%rsp      \n"
1021 #else
1022                         "       xchgl   %%ebx,%%esp     \n"
1023 #endif
1024                         "       int3                    \n"
1025                         "       .globl jprobe_return_end\n"
1026                         "       jprobe_return_end:      \n"
1027                         "       nop                     \n"::"b"
1028                         (kcb->jprobe_saved_sp):"memory");
1029 }
1030
1031 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1032 {
1033         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1034         u8 *addr = (u8 *) (regs->ip - 1);
1035         struct jprobe *jp = container_of(p, struct jprobe, kp);
1036
1037         if ((addr > (u8 *) jprobe_return) &&
1038             (addr < (u8 *) jprobe_return_end)) {
1039                 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1040                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1041                         printk(KERN_ERR
1042                                "current sp %p does not match saved sp %p\n",
1043                                stack_addr(regs), kcb->jprobe_saved_sp);
1044                         printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1045                         show_registers(saved_regs);
1046                         printk(KERN_ERR "Current registers\n");
1047                         show_registers(regs);
1048                         BUG();
1049                 }
1050                 *regs = kcb->jprobe_saved_regs;
1051                 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1052                        kcb->jprobes_stack,
1053                        MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1054                 preempt_enable_no_resched();
1055                 return 1;
1056         }
1057         return 0;
1058 }
1059
1060 int __init arch_init_kprobes(void)
1061 {
1062         return 0;
1063 }
1064
1065 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1066 {
1067         return 0;
1068 }