arch/x86/kernel/kprobes_64.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 <kenistoj@us.ibm.com> and Prasanna S Panchamukhi
  26  *              <prasanna@in.ibm.com> adapted for x86_64
  27  * 2005-Mar     Roland McGrath <roland@redhat.com>
  28  *              Fixed to handle %rip-relative addressing mode correctly.
  29  * 2005-May     Rusty Lynch <rusty.lynch@intel.com>
  30  *              Added function return probes functionality
  31  * 2007-Dec     Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
  32  *              and kretprobe-booster for x86-64
  33  */
  34
  35 #include <linux/kprobes.h>
  36 #include <linux/ptrace.h>
  37 #include <linux/string.h>
  38 #include <linux/slab.h>
  39 #include <linux/preempt.h>
  40 #include <linux/module.h>
  41 #include <linux/kdebug.h>
  42
  43 #include <asm/cacheflush.h>
  44 #include <asm/desc.h>
  45 #include <asm/pgtable.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/alternative.h>
  48
  49 void jprobe_return_end(void);
  50
  51 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  52 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  53
  54 #define stack_addr(regs) ((unsigned long *)regs->sp)
  55
  56 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
  57         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  58           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  59           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  60           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  61          << (row % 32))
  62         /*
  63          * Undefined/reserved opcodes, conditional jump, Opcode Extension
  64          * Groups, and some special opcodes can not boost.
  65          */
  66 static const u32 twobyte_is_boostable[256 / 32] = {
  67         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  68         /*      ----------------------------------------------          */
  69         W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
  70         W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
  71         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
  72         W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
  73         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
  74         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
  75         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
  76         W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
  77         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
  78         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
  79         W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
  80         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
  81         W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
  82         W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
  83         W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
  84         W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
  85         /*      -----------------------------------------------         */
  86         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  87 };
  88 static const u32 onebyte_has_modrm[256 / 32] = {
  89         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  90         /*      -----------------------------------------------         */
  91         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
  92         W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
  93         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
  94         W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
  95         W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
  96         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
  97         W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
  98         W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
  99         W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
 100         W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
 101         W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
 102         W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
 103         W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
 104         W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
 105         W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
 106         W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1)   /* f0 */
 107         /*      -----------------------------------------------         */
 108         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 109 };
 110 static const u32 twobyte_has_modrm[256 / 32] = {
 111         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 112         /*      -----------------------------------------------         */
 113         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
 114         W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
 115         W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
 116         W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
 117         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
 118         W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
 119         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
 120         W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
 121         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
 122         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
 123         W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
 124         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
 125         W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
 126         W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
 127         W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
 128         W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0)   /* ff */
 129         /*      -----------------------------------------------         */
 130         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 131 };
 132 #undef W
 133
 134 struct kretprobe_blackpoint kretprobe_blacklist[] = {
 135         {"__switch_to", }, /* This function switches only current task, but
 136                               doesn't switch kernel stack.*/
 137         {NULL, NULL}    /* Terminator */
 138 };
 139 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
 140
 141 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
 142 static __always_inline void set_jmp_op(void *from, void *to)
 143 {
 144         struct __arch_jmp_op {
 145                 char op;
 146                 s32 raddr;
 147         } __attribute__((packed)) * jop;
 148         jop = (struct __arch_jmp_op *)from;
 149         jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
 150         jop->op = RELATIVEJUMP_INSTRUCTION;
 151 }
 152
 153 /*
 154  * returns non-zero if opcode is boostable.
 155  * RIP relative instructions are adjusted at copying time
 156  */
 157 static __always_inline int can_boost(kprobe_opcode_t *opcodes)
 158 {
 159         kprobe_opcode_t opcode;
 160         kprobe_opcode_t *orig_opcodes = opcodes;
 161
 162 retry:
 163         if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 164                 return 0;
 165         opcode = *(opcodes++);
 166
 167         /* 2nd-byte opcode */
 168         if (opcode == 0x0f) {
 169                 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 170                         return 0;
 171                 return test_bit(*opcodes,
 172                                 (unsigned long *)twobyte_is_boostable);
 173         }
 174
 175         switch (opcode & 0xf0) {
 176         case 0x40:
 177                 goto retry; /* REX prefix is boostable */
 178         case 0x60:
 179                 if (0x63 < opcode && opcode < 0x67)
 180                         goto retry; /* prefixes */
 181                 /* can't boost Address-size override and bound */
 182                 return (opcode != 0x62 && opcode != 0x67);
 183         case 0x70:
 184                 return 0; /* can't boost conditional jump */
 185         case 0xc0:
 186                 /* can't boost software-interruptions */
 187                 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
 188         case 0xd0:
 189                 /* can boost AA* and XLAT */
 190                 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
 191         case 0xe0:
 192                 /* can boost in/out and absolute jmps */
 193                 return ((opcode & 0x04) || opcode == 0xea);
 194         case 0xf0:
 195                 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
 196                         goto retry; /* lock/rep(ne) prefix */
 197                 /* clear and set flags are boostable */
 198                 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
 199         default:
 200                 /* segment override prefixes are boostable */
 201                 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
 202                         goto retry; /* prefixes */
 203                 /* CS override prefix and call are not boostable */
 204                 return (opcode != 0x2e && opcode != 0x9a);
 205         }
 206 }
 207
 208 /*
 209  * returns non-zero if opcode modifies the interrupt flag.
 210  */
 211 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
 212 {
 213         switch (*insn) {
 214         case 0xfa:              /* cli */
 215         case 0xfb:              /* sti */
 216         case 0xcf:              /* iret/iretd */
 217         case 0x9d:              /* popf/popfd */
 218                 return 1;
 219         }
 220
 221         /*
 222          * on 64 bit x86, 0x40-0x4f are prefixes so we need to look
 223          * at the next byte instead.. but of course not recurse infinitely
 224          */
 225         if (*insn  >= 0x40 && *insn <= 0x4f)
 226                 return is_IF_modifier(++insn);
 227         return 0;
 228 }
 229
 230 /*
 231  * Adjust the displacement if the instruction uses the %rip-relative
 232  * addressing mode.
 233  * If it does, Return the address of the 32-bit displacement word.
 234  * If not, return null.
 235  */
 236 static void __kprobes fix_riprel(struct kprobe *p)
 237 {
 238         u8 *insn = p->ainsn.insn;
 239         s64 disp;
 240         int need_modrm;
 241
 242         /* Skip legacy instruction prefixes.  */
 243         while (1) {
 244                 switch (*insn) {
 245                 case 0x66:
 246                 case 0x67:
 247                 case 0x2e:
 248                 case 0x3e:
 249                 case 0x26:
 250                 case 0x64:
 251                 case 0x65:
 252                 case 0x36:
 253                 case 0xf0:
 254                 case 0xf3:
 255                 case 0xf2:
 256                         ++insn;
 257                         continue;
 258                 }
 259                 break;
 260         }
 261
 262         /* Skip REX instruction prefix.  */
 263         if ((*insn & 0xf0) == 0x40)
 264                 ++insn;
 265
 266         if (*insn == 0x0f) {    /* Two-byte opcode.  */
 267                 ++insn;
 268                 need_modrm = test_bit(*insn,
 269                                       (unsigned long *)twobyte_has_modrm);
 270         } else                  /* One-byte opcode.  */
 271                 need_modrm = test_bit(*insn,
 272                                       (unsigned long *)onebyte_has_modrm);
 273
 274         if (need_modrm) {
 275                 u8 modrm = *++insn;
 276                 if ((modrm & 0xc7) == 0x05) { /* %rip+disp32 addressing mode */
 277                         /* Displacement follows ModRM byte.  */
 278                         ++insn;
 279                         /*
 280                          * The copied instruction uses the %rip-relative
 281                          * addressing mode.  Adjust the displacement for the
 282                          * difference between the original location of this
 283                          * instruction and the location of the copy that will
 284                          * actually be run.  The tricky bit here is making sure
 285                          * that the sign extension happens correctly in this
 286                          * calculation, since we need a signed 32-bit result to
 287                          * be sign-extended to 64 bits when it's added to the
 288                          * %rip value and yield the same 64-bit result that the
 289                          * sign-extension of the original signed 32-bit
 290                          * displacement would have given.
 291                          */
 292                         disp = (u8 *) p->addr + *((s32 *) insn) -
 293                                (u8 *) p->ainsn.insn;
 294                         BUG_ON((s64) (s32) disp != disp); /* Sanity check.  */
 295                         *(s32 *)insn = (s32) disp;
 296                 }
 297         }
 298 }
 299
 300 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 301 {
 302         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 303         fix_riprel(p);
 304         if (can_boost(p->addr))
 305                 p->ainsn.boostable = 0;
 306         else
 307                 p->ainsn.boostable = -1;
 308
 309         p->opcode = *p->addr;
 310 }
 311
 312 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 313 {
 314         /* insn: must be on special executable page on x86. */
 315         p->ainsn.insn = get_insn_slot();
 316         if (!p->ainsn.insn)
 317                 return -ENOMEM;
 318         arch_copy_kprobe(p);
 319         return 0;
 320 }
 321
 322 void __kprobes arch_arm_kprobe(struct kprobe *p)
 323 {
 324         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 325 }
 326
 327 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 328 {
 329         text_poke(p->addr, &p->opcode, 1);
 330 }
 331
 332 void __kprobes arch_remove_kprobe(struct kprobe *p)
 333 {
 334         mutex_lock(&kprobe_mutex);
 335         free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
 336         mutex_unlock(&kprobe_mutex);
 337 }
 338
 339 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 340 {
 341         kcb->prev_kprobe.kp = kprobe_running();
 342         kcb->prev_kprobe.status = kcb->kprobe_status;
 343         kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
 344         kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
 345 }
 346
 347 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 348 {
 349         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 350         kcb->kprobe_status = kcb->prev_kprobe.status;
 351         kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
 352         kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
 353 }
 354
 355 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 356                                 struct kprobe_ctlblk *kcb)
 357 {
 358         __get_cpu_var(current_kprobe) = p;
 359         kcb->kprobe_saved_flags = kcb->kprobe_old_flags
 360                 = (regs->flags & (TF_MASK | IF_MASK));
 361         if (is_IF_modifier(p->ainsn.insn))
 362                 kcb->kprobe_saved_flags &= ~IF_MASK;
 363 }
 364
 365 static __always_inline void clear_btf(void)
 366 {
 367         if (test_thread_flag(TIF_DEBUGCTLMSR))
 368                 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
 369 }
 370
 371 static __always_inline void restore_btf(void)
 372 {
 373         if (test_thread_flag(TIF_DEBUGCTLMSR))
 374                 wrmsrl(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr);
 375 }
 376
 377 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 378 {
 379         clear_btf();
 380         regs->flags |= TF_MASK;
 381         regs->flags &= ~IF_MASK;
 382         /*single step inline if the instruction is an int3*/
 383         if (p->opcode == BREAKPOINT_INSTRUCTION)
 384                 regs->ip = (unsigned long)p->addr;
 385         else
 386                 regs->ip = (unsigned long)p->ainsn.insn;
 387 }
 388
 389 /* Called with kretprobe_lock held */
 390 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 391                                       struct pt_regs *regs)
 392 {
 393         unsigned long *sara = stack_addr(regs);
 394
 395         ri->ret_addr = (kprobe_opcode_t *) *sara;
 396
 397         /* Replace the return addr with trampoline addr */
 398         *sara = (unsigned long) &kretprobe_trampoline;
 399 }
 400
 401 /*
 402  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 403  * remain disabled thorough out this function.
 404  */
 405 static int __kprobes kprobe_handler(struct pt_regs *regs)
 406 {
 407         struct kprobe *p;
 408         int ret = 0;
 409         kprobe_opcode_t *addr;
 410         struct kprobe_ctlblk *kcb;
 411
 412         addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 413
 414         /*
 415          * We don't want to be preempted for the entire
 416          * duration of kprobe processing
 417          */
 418         preempt_disable();
 419         kcb = get_kprobe_ctlblk();
 420
 421         /* Check we're not actually recursing */
 422         if (kprobe_running()) {
 423                 p = get_kprobe(addr);
 424                 if (p) {
 425                         if (kcb->kprobe_status == KPROBE_HIT_SS &&
 426                                 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
 427                                 regs->flags &= ~TF_MASK;
 428                                 regs->flags |= kcb->kprobe_saved_flags;
 429                                 goto no_kprobe;
 430                         } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
 431                                 /* TODO: Provide re-entrancy from
 432                                  * post_kprobes_handler() and avoid exception
 433                                  * stack corruption while single-stepping on
 434                                  * the instruction of the new probe.
 435                                  */
 436                                 arch_disarm_kprobe(p);
 437                                 regs->ip = (unsigned long)p->addr;
 438                                 reset_current_kprobe();
 439                                 return 1;
 440                         }
 441                         /* We have reentered the kprobe_handler(), since
 442                          * another probe was hit while within the handler.
 443                          * We here save the original kprobes variables and
 444                          * just single step on the instruction of the new probe
 445                          * without calling any user handlers.
 446                          */
 447                         save_previous_kprobe(kcb);
 448                         set_current_kprobe(p, regs, kcb);
 449                         kprobes_inc_nmissed_count(p);
 450                         prepare_singlestep(p, regs);
 451                         kcb->kprobe_status = KPROBE_REENTER;
 452                         return 1;
 453                 } else {
 454                         if (*addr != BREAKPOINT_INSTRUCTION) {
 455                         /* The breakpoint instruction was removed by
 456                          * another cpu right after we hit, no further
 457                          * handling of this interrupt is appropriate
 458                          */
 459                                 regs->ip = (unsigned long)addr;
 460                                 ret = 1;
 461                                 goto no_kprobe;
 462                         }
 463                         p = __get_cpu_var(current_kprobe);
 464                         if (p->break_handler && p->break_handler(p, regs)) {
 465                                 goto ss_probe;
 466                         }
 467                 }
 468                 goto no_kprobe;
 469         }
 470
 471         p = get_kprobe(addr);
 472         if (!p) {
 473                 if (*addr != BREAKPOINT_INSTRUCTION) {
 474                         /*
 475                          * The breakpoint instruction was removed right
 476                          * after we hit it.  Another cpu has removed
 477                          * either a probepoint or a debugger breakpoint
 478                          * at this address.  In either case, no further
 479                          * handling of this interrupt is appropriate.
 480                          * Back up over the (now missing) int3 and run
 481                          * the original instruction.
 482                          */
 483                         regs->ip = (unsigned long)addr;
 484                         ret = 1;
 485                 }
 486                 /* Not one of ours: let kernel handle it */
 487                 goto no_kprobe;
 488         }
 489
 490         set_current_kprobe(p, regs, kcb);
 491         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 492
 493         if (p->pre_handler && p->pre_handler(p, regs))
 494                 /* handler has already set things up, so skip ss setup */
 495                 return 1;
 496
 497 ss_probe:
 498 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
 499         if (p->ainsn.boostable == 1 && !p->post_handler) {
 500                 /* Boost up -- we can execute copied instructions directly */
 501                 reset_current_kprobe();
 502                 regs->ip = (unsigned long)p->ainsn.insn;
 503                 preempt_enable_no_resched();
 504                 return 1;
 505         }
 506 #endif
 507         prepare_singlestep(p, regs);
 508         kcb->kprobe_status = KPROBE_HIT_SS;
 509         return 1;
 510
 511 no_kprobe:
 512         preempt_enable_no_resched();
 513         return ret;
 514 }
 515
 516 /*
 517  * When a retprobed function returns, this code saves registers and
 518  * calls trampoline_handler() runs, which calls the kretprobe's handler.
 519  */
 520  void __kprobes kretprobe_trampoline_holder(void)
 521  {
 522         asm volatile (  ".global kretprobe_trampoline\n"
 523                         "kretprobe_trampoline: \n"
 524                         /* We don't bother saving the ss register */
 525                         "       pushq %rsp\n"
 526                         "       pushfq\n"
 527                         /*
 528                          * Skip cs, ip, orig_ax.
 529                          * trampoline_handler() will plug in these values
 530                          */
 531                         "       subq $24, %rsp\n"
 532                         "       pushq %rdi\n"
 533                         "       pushq %rsi\n"
 534                         "       pushq %rdx\n"
 535                         "       pushq %rcx\n"
 536                         "       pushq %rax\n"
 537                         "       pushq %r8\n"
 538                         "       pushq %r9\n"
 539                         "       pushq %r10\n"
 540                         "       pushq %r11\n"
 541                         "       pushq %rbx\n"
 542                         "       pushq %rbp\n"
 543                         "       pushq %r12\n"
 544                         "       pushq %r13\n"
 545                         "       pushq %r14\n"
 546                         "       pushq %r15\n"
 547                         "       movq %rsp, %rdi\n"
 548                         "       call trampoline_handler\n"
 549                         /* Replace saved sp with true return address. */
 550                         "       movq %rax, 152(%rsp)\n"
 551                         "       popq %r15\n"
 552                         "       popq %r14\n"
 553                         "       popq %r13\n"
 554                         "       popq %r12\n"
 555                         "       popq %rbp\n"
 556                         "       popq %rbx\n"
 557                         "       popq %r11\n"
 558                         "       popq %r10\n"
 559                         "       popq %r9\n"
 560                         "       popq %r8\n"
 561                         "       popq %rax\n"
 562                         "       popq %rcx\n"
 563                         "       popq %rdx\n"
 564                         "       popq %rsi\n"
 565                         "       popq %rdi\n"
 566                         /* Skip orig_ax, ip, cs */
 567                         "       addq $24, %rsp\n"
 568                         "       popfq\n"
 569                         "       ret\n");
 570  }
 571
 572 /*
 573  * Called from kretprobe_trampoline
 574  */
 575 fastcall void * __kprobes trampoline_handler(struct pt_regs *regs)
 576 {
 577         struct kretprobe_instance *ri = NULL;
 578         struct hlist_head *head, empty_rp;
 579         struct hlist_node *node, *tmp;
 580         unsigned long flags, orig_ret_address = 0;
 581         unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
 582
 583         INIT_HLIST_HEAD(&empty_rp);
 584         spin_lock_irqsave(&kretprobe_lock, flags);
 585         head = kretprobe_inst_table_head(current);
 586         /* fixup registers */
 587         regs->cs = __KERNEL_CS;
 588         regs->ip = trampoline_address;
 589         regs->orig_ax = ~0UL;
 590
 591         /*
 592          * It is possible to have multiple instances associated with a given
 593          * task either because multiple functions in the call path have
 594          * return probes installed on them, and/or more then one
 595          * return probe was registered for a target function.
 596          *
 597          * We can handle this because:
 598          *     - instances are always pushed into the head of the list
 599          *     - when multiple return probes are registered for the same
 600          *       function, the (chronologically) first instance's ret_addr
 601          *       will be the real return address, and all the rest will
 602          *       point to kretprobe_trampoline.
 603          */
 604         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 605                 if (ri->task != current)
 606                         /* another task is sharing our hash bucket */
 607                         continue;
 608
 609                 if (ri->rp && ri->rp->handler) {
 610                         __get_cpu_var(current_kprobe) = &ri->rp->kp;
 611                         get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
 612                         ri->rp->handler(ri, regs);
 613                         __get_cpu_var(current_kprobe) = NULL;
 614                 }
 615
 616                 orig_ret_address = (unsigned long)ri->ret_addr;
 617                 recycle_rp_inst(ri, &empty_rp);
 618
 619                 if (orig_ret_address != trampoline_address)
 620                         /*
 621                          * This is the real return address. Any other
 622                          * instances associated with this task are for
 623                          * other calls deeper on the call stack
 624                          */
 625                         break;
 626         }
 627
 628         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 629
 630         spin_unlock_irqrestore(&kretprobe_lock, flags);
 631
 632         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 633                 hlist_del(&ri->hlist);
 634                 kfree(ri);
 635         }
 636         return (void *)orig_ret_address;
 637 }
 638
 639 /*
 640  * Called after single-stepping.  p->addr is the address of the
 641  * instruction whose first byte has been replaced by the "int 3"
 642  * instruction.  To avoid the SMP problems that can occur when we
 643  * temporarily put back the original opcode to single-step, we
 644  * single-stepped a copy of the instruction.  The address of this
 645  * copy is p->ainsn.insn.
 646  *
 647  * This function prepares to return from the post-single-step
 648  * interrupt.  We have to fix up the stack as follows:
 649  *
 650  * 0) Except in the case of absolute or indirect jump or call instructions,
 651  * the new ip is relative to the copied instruction.  We need to make
 652  * it relative to the original instruction.
 653  *
 654  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 655  * flags are set in the just-pushed flags, and may need to be cleared.
 656  *
 657  * 2) If the single-stepped instruction was a call, the return address
 658  * that is atop the stack is the address following the copied instruction.
 659  * We need to make it the address following the original instruction.
 660  *
 661  * If this is the first time we've single-stepped the instruction at
 662  * this probepoint, and the instruction is boostable, boost it: add a
 663  * jump instruction after the copied instruction, that jumps to the next
 664  * instruction after the probepoint.
 665  */
 666 static void __kprobes resume_execution(struct kprobe *p,
 667                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 668 {
 669         unsigned long *tos = stack_addr(regs);
 670         unsigned long copy_ip = (unsigned long)p->ainsn.insn;
 671         unsigned long orig_ip = (unsigned long)p->addr;
 672         kprobe_opcode_t *insn = p->ainsn.insn;
 673
 674         /*skip the REX prefix*/
 675         if (*insn >= 0x40 && *insn <= 0x4f)
 676                 insn++;
 677
 678         regs->flags &= ~TF_MASK;
 679         switch (*insn) {
 680         case 0x9c:      /* pushfl */
 681                 *tos &= ~(TF_MASK | IF_MASK);
 682                 *tos |= kcb->kprobe_old_flags;
 683                 break;
 684         case 0xc2:      /* iret/ret/lret */
 685         case 0xc3:
 686         case 0xca:
 687         case 0xcb:
 688         case 0xcf:
 689         case 0xea:      /* jmp absolute -- ip is correct */
 690                 /* ip is already adjusted, no more changes required */
 691                 p->ainsn.boostable = 1;
 692                 goto no_change;
 693         case 0xe8:      /* call relative - Fix return addr */
 694                 *tos = orig_ip + (*tos - copy_ip);
 695                 break;
 696         case 0xff:
 697                 if ((insn[1] & 0x30) == 0x10) {
 698                         /*
 699                          * call absolute, indirect
 700                          * Fix return addr; ip is correct.
 701                          * But this is not boostable
 702                          */
 703                         *tos = orig_ip + (*tos - copy_ip);
 704                         goto no_change;
 705                 } else if (((insn[1] & 0x31) == 0x20) ||
 706                            ((insn[1] & 0x31) == 0x21)) {
 707                         /*
 708                          * jmp near and far, absolute indirect
 709                          * ip is correct. And this is boostable
 710                          */
 711                         p->ainsn.boostable = 1;
 712                         goto no_change;
 713                 }
 714         default:
 715                 break;
 716         }
 717
 718         if (p->ainsn.boostable == 0) {
 719                 if ((regs->ip > copy_ip) &&
 720                     (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
 721                         /*
 722                          * These instructions can be executed directly if it
 723                          * jumps back to correct address.
 724                          */
 725                         set_jmp_op((void *)regs->ip,
 726                                    (void *)orig_ip + (regs->ip - copy_ip));
 727                         p->ainsn.boostable = 1;
 728                 } else {
 729                         p->ainsn.boostable = -1;
 730                 }
 731         }
 732
 733         regs->ip += orig_ip - copy_ip;
 734
 735 no_change:
 736         restore_btf();
 737
 738         return;
 739 }
 740
 741 /*
 742  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 743  * remain disabled thoroughout this function.
 744  */
 745 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
 746 {
 747         struct kprobe *cur = kprobe_running();
 748         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 749
 750         if (!cur)
 751                 return 0;
 752
 753         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 754                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 755                 cur->post_handler(cur, regs, 0);
 756         }
 757
 758         resume_execution(cur, regs, kcb);
 759         regs->flags |= kcb->kprobe_saved_flags;
 760         trace_hardirqs_fixup_flags(regs->flags);
 761
 762         /* Restore back the original saved kprobes variables and continue. */
 763         if (kcb->kprobe_status == KPROBE_REENTER) {
 764                 restore_previous_kprobe(kcb);
 765                 goto out;
 766         }
 767         reset_current_kprobe();
 768 out:
 769         preempt_enable_no_resched();
 770
 771         /*
 772          * if somebody else is singlestepping across a probe point, flags
 773          * will have TF set, in which case, continue the remaining processing
 774          * of do_debug, as if this is not a probe hit.
 775          */
 776         if (regs->flags & TF_MASK)
 777                 return 0;
 778
 779         return 1;
 780 }
 781
 782 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 783 {
 784         struct kprobe *cur = kprobe_running();
 785         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 786         const struct exception_table_entry *fixup;
 787
 788         switch(kcb->kprobe_status) {
 789         case KPROBE_HIT_SS:
 790         case KPROBE_REENTER:
 791                 /*
 792                  * We are here because the instruction being single
 793                  * stepped caused a page fault. We reset the current
 794                  * kprobe and the ip points back to the probe address
 795                  * and allow the page fault handler to continue as a
 796                  * normal page fault.
 797                  */
 798                 regs->ip = (unsigned long)cur->addr;
 799                 regs->flags |= kcb->kprobe_old_flags;
 800                 if (kcb->kprobe_status == KPROBE_REENTER)
 801                         restore_previous_kprobe(kcb);
 802                 else
 803                         reset_current_kprobe();
 804                 preempt_enable_no_resched();
 805                 break;
 806         case KPROBE_HIT_ACTIVE:
 807         case KPROBE_HIT_SSDONE:
 808                 /*
 809                  * We increment the nmissed count for accounting,
 810                  * we can also use npre/npostfault count for accounting
 811                  * these specific fault cases.
 812                  */
 813                 kprobes_inc_nmissed_count(cur);
 814
 815                 /*
 816                  * We come here because instructions in the pre/post
 817                  * handler caused the page_fault, this could happen
 818                  * if handler tries to access user space by
 819                  * copy_from_user(), get_user() etc. Let the
 820                  * user-specified handler try to fix it first.
 821                  */
 822                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 823                         return 1;
 824
 825                 /*
 826                  * In case the user-specified fault handler returned
 827                  * zero, try to fix up.
 828                  */
 829                 fixup = search_exception_tables(regs->ip);
 830                 if (fixup) {
 831                         regs->ip = fixup->fixup;
 832                         return 1;
 833                 }
 834
 835                 /*
 836                  * fixup routine could not handle it,
 837                  * Let do_page_fault() fix it.
 838                  */
 839                 break;
 840         default:
 841                 break;
 842         }
 843         return 0;
 844 }
 845
 846 /*
 847  * Wrapper routine for handling exceptions.
 848  */
 849 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 850                                        unsigned long val, void *data)
 851 {
 852         struct die_args *args = (struct die_args *)data;
 853         int ret = NOTIFY_DONE;
 854
 855         if (args->regs && user_mode_vm(args->regs))
 856                 return ret;
 857
 858         switch (val) {
 859         case DIE_INT3:
 860                 if (kprobe_handler(args->regs))
 861                         ret = NOTIFY_STOP;
 862                 break;
 863         case DIE_DEBUG:
 864                 if (post_kprobe_handler(args->regs))
 865                         ret = NOTIFY_STOP;
 866                 break;
 867         case DIE_GPF:
 868                 /* kprobe_running() needs smp_processor_id() */
 869                 preempt_disable();
 870                 if (kprobe_running() &&
 871                     kprobe_fault_handler(args->regs, args->trapnr))
 872                         ret = NOTIFY_STOP;
 873                 preempt_enable();
 874                 break;
 875         default:
 876                 break;
 877         }
 878         return ret;
 879 }
 880
 881 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 882 {
 883         struct jprobe *jp = container_of(p, struct jprobe, kp);
 884         unsigned long addr;
 885         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 886
 887         kcb->jprobe_saved_regs = *regs;
 888         kcb->jprobe_saved_sp = stack_addr(regs);
 889         addr = (unsigned long)(kcb->jprobe_saved_sp);
 890
 891         /*
 892          * As Linus pointed out, gcc assumes that the callee
 893          * owns the argument space and could overwrite it, e.g.
 894          * tailcall optimization. So, to be absolutely safe
 895          * we also save and restore enough stack bytes to cover
 896          * the argument area.
 897          */
 898         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
 899                         MIN_STACK_SIZE(addr));
 900         regs->flags &= ~IF_MASK;
 901         trace_hardirqs_off();
 902         regs->ip = (unsigned long)(jp->entry);
 903         return 1;
 904 }
 905
 906 void __kprobes jprobe_return(void)
 907 {
 908         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 909
 910         asm volatile ("       xchg   %%rbx,%%rsp     \n"
 911                       "       int3                      \n"
 912                       "       .globl jprobe_return_end  \n"
 913                       "       jprobe_return_end:        \n"
 914                       "       nop                       \n"::"b"
 915                       (kcb->jprobe_saved_sp):"memory");
 916 }
 917
 918 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 919 {
 920         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 921         u8 *addr = (u8 *) (regs->ip - 1);
 922         struct jprobe *jp = container_of(p, struct jprobe, kp);
 923
 924         if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
 925                 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
 926                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
 927                         printk("current sp %p does not match saved sp %p\n",
 928                                stack_addr(regs), kcb->jprobe_saved_sp);
 929                         printk("Saved registers for jprobe %p\n", jp);
 930                         show_registers(saved_regs);
 931                         printk("Current registers\n");
 932                         show_registers(regs);
 933                         BUG();
 934                 }
 935                 *regs = kcb->jprobe_saved_regs;
 936                 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
 937                        kcb->jprobes_stack,
 938                        MIN_STACK_SIZE(kcb->jprobe_saved_sp));
 939                 preempt_enable_no_resched();
 940                 return 1;
 941         }
 942         return 0;
 943 }
 944
 945 int __init arch_init_kprobes(void)
 946 {
 947         return 0;
 948 }
 949
 950 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
 951 {
 952         return 0;
 953 }