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  */
  32
  33 #include <linux/kprobes.h>
  34 #include <linux/ptrace.h>
  35 #include <linux/string.h>
  36 #include <linux/slab.h>
  37 #include <linux/preempt.h>
  38 #include <linux/module.h>
  39 #include <linux/kdebug.h>
  40
  41 #include <asm/pgtable.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/alternative.h>
  44
  45 void jprobe_return_end(void);
  46 static void __kprobes arch_copy_kprobe(struct kprobe *p);
  47
  48 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  49 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  50
  51 struct kretprobe_blackpoint kretprobe_blacklist[] = {
  52         {"__switch_to", }, /* This function switches only current task, but
  53                               doesn't switch kernel stack.*/
  54         {NULL, NULL}    /* Terminator */
  55 };
  56 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
  57
  58 /*
  59  * returns non-zero if opcode modifies the interrupt flag.
  60  */
  61 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
  62 {
  63         switch (*insn) {
  64         case 0xfa:              /* cli */
  65         case 0xfb:              /* sti */
  66         case 0xcf:              /* iret/iretd */
  67         case 0x9d:              /* popf/popfd */
  68                 return 1;
  69         }
  70
  71         if (*insn  >= 0x40 && *insn <= 0x4f && *++insn == 0xcf)
  72                 return 1;
  73         return 0;
  74 }
  75
  76 int __kprobes arch_prepare_kprobe(struct kprobe *p)
  77 {
  78         /* insn: must be on special executable page on x86_64. */
  79         p->ainsn.insn = get_insn_slot();
  80         if (!p->ainsn.insn) {
  81                 return -ENOMEM;
  82         }
  83         arch_copy_kprobe(p);
  84         return 0;
  85 }
  86
  87 /*
  88  * Determine if the instruction uses the %rip-relative addressing mode.
  89  * If it does, return the address of the 32-bit displacement word.
  90  * If not, return null.
  91  */
  92 static s32 __kprobes *is_riprel(u8 *insn)
  93 {
  94 #define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf)                \
  95         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  96           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  97           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  98           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  99          << (row % 64))
 100         static const u64 onebyte_has_modrm[256 / 64] = {
 101                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 102                 /*      -------------------------------         */
 103                 W(0x00, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 00 */
 104                 W(0x10, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 10 */
 105                 W(0x20, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 20 */
 106                 W(0x30, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0), /* 30 */
 107                 W(0x40, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 40 */
 108                 W(0x50, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 50 */
 109                 W(0x60, 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0)| /* 60 */
 110                 W(0x70, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 70 */
 111                 W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
 112                 W(0x90, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 90 */
 113                 W(0xa0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* a0 */
 114                 W(0xb0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* b0 */
 115                 W(0xc0, 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0)| /* c0 */
 116                 W(0xd0, 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1)| /* d0 */
 117                 W(0xe0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* e0 */
 118                 W(0xf0, 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1)  /* f0 */
 119                 /*      -------------------------------         */
 120                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 121         };
 122         static const u64 twobyte_has_modrm[256 / 64] = {
 123                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 124                 /*      -------------------------------         */
 125                 W(0x00, 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1)| /* 0f */
 126                 W(0x10, 1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0)| /* 1f */
 127                 W(0x20, 1,1,1,1,1,0,1,0,1,1,1,1,1,1,1,1)| /* 2f */
 128                 W(0x30, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 3f */
 129                 W(0x40, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 4f */
 130                 W(0x50, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 5f */
 131                 W(0x60, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 6f */
 132                 W(0x70, 1,1,1,1,1,1,1,0,0,0,0,0,1,1,1,1), /* 7f */
 133                 W(0x80, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 8f */
 134                 W(0x90, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 9f */
 135                 W(0xa0, 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1)| /* af */
 136                 W(0xb0, 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1), /* bf */
 137                 W(0xc0, 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0)| /* cf */
 138                 W(0xd0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* df */
 139                 W(0xe0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* ef */
 140                 W(0xf0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0)  /* ff */
 141                 /*      -------------------------------         */
 142                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 143         };
 144 #undef  W
 145         int need_modrm;
 146
 147         /* Skip legacy instruction prefixes.  */
 148         while (1) {
 149                 switch (*insn) {
 150                 case 0x66:
 151                 case 0x67:
 152                 case 0x2e:
 153                 case 0x3e:
 154                 case 0x26:
 155                 case 0x64:
 156                 case 0x65:
 157                 case 0x36:
 158                 case 0xf0:
 159                 case 0xf3:
 160                 case 0xf2:
 161                         ++insn;
 162                         continue;
 163                 }
 164                 break;
 165         }
 166
 167         /* Skip REX instruction prefix.  */
 168         if ((*insn & 0xf0) == 0x40)
 169                 ++insn;
 170
 171         if (*insn == 0x0f) {    /* Two-byte opcode.  */
 172                 ++insn;
 173                 need_modrm = test_bit(*insn, twobyte_has_modrm);
 174         } else {                /* One-byte opcode.  */
 175                 need_modrm = test_bit(*insn, onebyte_has_modrm);
 176         }
 177
 178         if (need_modrm) {
 179                 u8 modrm = *++insn;
 180                 if ((modrm & 0xc7) == 0x05) { /* %rip+disp32 addressing mode */
 181                         /* Displacement follows ModRM byte.  */
 182                         return (s32 *) ++insn;
 183                 }
 184         }
 185
 186         /* No %rip-relative addressing mode here.  */
 187         return NULL;
 188 }
 189
 190 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 191 {
 192         s32 *ripdisp;
 193         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE);
 194         ripdisp = is_riprel(p->ainsn.insn);
 195         if (ripdisp) {
 196                 /*
 197                  * The copied instruction uses the %rip-relative
 198                  * addressing mode.  Adjust the displacement for the
 199                  * difference between the original location of this
 200                  * instruction and the location of the copy that will
 201                  * actually be run.  The tricky bit here is making sure
 202                  * that the sign extension happens correctly in this
 203                  * calculation, since we need a signed 32-bit result to
 204                  * be sign-extended to 64 bits when it's added to the
 205                  * %rip value and yield the same 64-bit result that the
 206                  * sign-extension of the original signed 32-bit
 207                  * displacement would have given.
 208                  */
 209                 s64 disp = (u8 *) p->addr + *ripdisp - (u8 *) p->ainsn.insn;
 210                 BUG_ON((s64) (s32) disp != disp); /* Sanity check.  */
 211                 *ripdisp = disp;
 212         }
 213         p->opcode = *p->addr;
 214 }
 215
 216 void __kprobes arch_arm_kprobe(struct kprobe *p)
 217 {
 218         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 219 }
 220
 221 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 222 {
 223         text_poke(p->addr, &p->opcode, 1);
 224 }
 225
 226 void __kprobes arch_remove_kprobe(struct kprobe *p)
 227 {
 228         mutex_lock(&kprobe_mutex);
 229         free_insn_slot(p->ainsn.insn, 0);
 230         mutex_unlock(&kprobe_mutex);
 231 }
 232
 233 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 234 {
 235         kcb->prev_kprobe.kp = kprobe_running();
 236         kcb->prev_kprobe.status = kcb->kprobe_status;
 237         kcb->prev_kprobe.old_rflags = kcb->kprobe_old_rflags;
 238         kcb->prev_kprobe.saved_rflags = kcb->kprobe_saved_rflags;
 239 }
 240
 241 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 242 {
 243         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 244         kcb->kprobe_status = kcb->prev_kprobe.status;
 245         kcb->kprobe_old_rflags = kcb->prev_kprobe.old_rflags;
 246         kcb->kprobe_saved_rflags = kcb->prev_kprobe.saved_rflags;
 247 }
 248
 249 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 250                                 struct kprobe_ctlblk *kcb)
 251 {
 252         __get_cpu_var(current_kprobe) = p;
 253         kcb->kprobe_saved_rflags = kcb->kprobe_old_rflags
 254                 = (regs->flags & (TF_MASK | IF_MASK));
 255         if (is_IF_modifier(p->ainsn.insn))
 256                 kcb->kprobe_saved_rflags &= ~IF_MASK;
 257 }
 258
 259 static __always_inline void clear_btf(void)
 260 {
 261         if (test_thread_flag(TIF_DEBUGCTLMSR))
 262                 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
 263 }
 264
 265 static __always_inline void restore_btf(void)
 266 {
 267         if (test_thread_flag(TIF_DEBUGCTLMSR))
 268                 wrmsrl(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr);
 269 }
 270
 271 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 272 {
 273         clear_btf();
 274         regs->flags |= TF_MASK;
 275         regs->flags &= ~IF_MASK;
 276         /*single step inline if the instruction is an int3*/
 277         if (p->opcode == BREAKPOINT_INSTRUCTION)
 278                 regs->ip = (unsigned long)p->addr;
 279         else
 280                 regs->ip = (unsigned long)p->ainsn.insn;
 281 }
 282
 283 /* Called with kretprobe_lock held */
 284 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 285                                       struct pt_regs *regs)
 286 {
 287         unsigned long *sara = (unsigned long *)regs->sp;
 288
 289         ri->ret_addr = (kprobe_opcode_t *) *sara;
 290         /* Replace the return addr with trampoline addr */
 291         *sara = (unsigned long) &kretprobe_trampoline;
 292 }
 293
 294 int __kprobes kprobe_handler(struct pt_regs *regs)
 295 {
 296         struct kprobe *p;
 297         int ret = 0;
 298         kprobe_opcode_t *addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 299         struct kprobe_ctlblk *kcb;
 300
 301         /*
 302          * We don't want to be preempted for the entire
 303          * duration of kprobe processing
 304          */
 305         preempt_disable();
 306         kcb = get_kprobe_ctlblk();
 307
 308         /* Check we're not actually recursing */
 309         if (kprobe_running()) {
 310                 p = get_kprobe(addr);
 311                 if (p) {
 312                         if (kcb->kprobe_status == KPROBE_HIT_SS &&
 313                                 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
 314                                 regs->flags &= ~TF_MASK;
 315                                 regs->flags |= kcb->kprobe_saved_rflags;
 316                                 goto no_kprobe;
 317                         } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
 318                                 /* TODO: Provide re-entrancy from
 319                                  * post_kprobes_handler() and avoid exception
 320                                  * stack corruption while single-stepping on
 321                                  * the instruction of the new probe.
 322                                  */
 323                                 arch_disarm_kprobe(p);
 324                                 regs->ip = (unsigned long)p->addr;
 325                                 reset_current_kprobe();
 326                                 ret = 1;
 327                         } else {
 328                                 /* We have reentered the kprobe_handler(), since
 329                                  * another probe was hit while within the
 330                                  * handler. We here save the original kprobe
 331                                  * variables and just single step on instruction
 332                                  * of the new probe without calling any user
 333                                  * handlers.
 334                                  */
 335                                 save_previous_kprobe(kcb);
 336                                 set_current_kprobe(p, regs, kcb);
 337                                 kprobes_inc_nmissed_count(p);
 338                                 prepare_singlestep(p, regs);
 339                                 kcb->kprobe_status = KPROBE_REENTER;
 340                                 return 1;
 341                         }
 342                 } else {
 343                         if (*addr != BREAKPOINT_INSTRUCTION) {
 344                         /* The breakpoint instruction was removed by
 345                          * another cpu right after we hit, no further
 346                          * handling of this interrupt is appropriate
 347                          */
 348                                 regs->ip = (unsigned long)addr;
 349                                 ret = 1;
 350                                 goto no_kprobe;
 351                         }
 352                         p = __get_cpu_var(current_kprobe);
 353                         if (p->break_handler && p->break_handler(p, regs)) {
 354                                 goto ss_probe;
 355                         }
 356                 }
 357                 goto no_kprobe;
 358         }
 359
 360         p = get_kprobe(addr);
 361         if (!p) {
 362                 if (*addr != BREAKPOINT_INSTRUCTION) {
 363                         /*
 364                          * The breakpoint instruction was removed right
 365                          * after we hit it.  Another cpu has removed
 366                          * either a probepoint or a debugger breakpoint
 367                          * at this address.  In either case, no further
 368                          * handling of this interrupt is appropriate.
 369                          * Back up over the (now missing) int3 and run
 370                          * the original instruction.
 371                          */
 372                         regs->ip = (unsigned long)addr;
 373                         ret = 1;
 374                 }
 375                 /* Not one of ours: let kernel handle it */
 376                 goto no_kprobe;
 377         }
 378
 379         set_current_kprobe(p, regs, kcb);
 380         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 381
 382         if (p->pre_handler && p->pre_handler(p, regs))
 383                 /* handler has already set things up, so skip ss setup */
 384                 return 1;
 385
 386 ss_probe:
 387         prepare_singlestep(p, regs);
 388         kcb->kprobe_status = KPROBE_HIT_SS;
 389         return 1;
 390
 391 no_kprobe:
 392         preempt_enable_no_resched();
 393         return ret;
 394 }
 395
 396 /*
 397  * For function-return probes, init_kprobes() establishes a probepoint
 398  * here. When a retprobed function returns, this probe is hit and
 399  * trampoline_probe_handler() runs, calling the kretprobe's handler.
 400  */
 401  void kretprobe_trampoline_holder(void)
 402  {
 403         asm volatile (  ".global kretprobe_trampoline\n"
 404                         "kretprobe_trampoline: \n"
 405                         "nop\n");
 406  }
 407
 408 /*
 409  * Called when we hit the probe point at kretprobe_trampoline
 410  */
 411 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
 412 {
 413         struct kretprobe_instance *ri = NULL;
 414         struct hlist_head *head, empty_rp;
 415         struct hlist_node *node, *tmp;
 416         unsigned long flags, orig_ret_address = 0;
 417         unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
 418
 419         INIT_HLIST_HEAD(&empty_rp);
 420         spin_lock_irqsave(&kretprobe_lock, flags);
 421         head = kretprobe_inst_table_head(current);
 422
 423         /*
 424          * It is possible to have multiple instances associated with a given
 425          * task either because an multiple functions in the call path
 426          * have a return probe installed on them, and/or more then one return
 427          * return probe was registered for a target function.
 428          *
 429          * We can handle this because:
 430          *     - instances are always inserted at the head of the list
 431          *     - when multiple return probes are registered for the same
 432          *       function, the first instance's ret_addr will point to the
 433          *       real return address, and all the rest will point to
 434          *       kretprobe_trampoline
 435          */
 436         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 437                 if (ri->task != current)
 438                         /* another task is sharing our hash bucket */
 439                         continue;
 440
 441                 if (ri->rp && ri->rp->handler)
 442                         ri->rp->handler(ri, regs);
 443
 444                 orig_ret_address = (unsigned long)ri->ret_addr;
 445                 recycle_rp_inst(ri, &empty_rp);
 446
 447                 if (orig_ret_address != trampoline_address)
 448                         /*
 449                          * This is the real return address. Any other
 450                          * instances associated with this task are for
 451                          * other calls deeper on the call stack
 452                          */
 453                         break;
 454         }
 455
 456         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 457         regs->ip = orig_ret_address;
 458
 459         reset_current_kprobe();
 460         spin_unlock_irqrestore(&kretprobe_lock, flags);
 461         preempt_enable_no_resched();
 462
 463         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 464                 hlist_del(&ri->hlist);
 465                 kfree(ri);
 466         }
 467         /*
 468          * By returning a non-zero value, we are telling
 469          * kprobe_handler() that we don't want the post_handler
 470          * to run (and have re-enabled preemption)
 471          */
 472         return 1;
 473 }
 474
 475 /*
 476  * Called after single-stepping.  p->addr is the address of the
 477  * instruction whose first byte has been replaced by the "int 3"
 478  * instruction.  To avoid the SMP problems that can occur when we
 479  * temporarily put back the original opcode to single-step, we
 480  * single-stepped a copy of the instruction.  The address of this
 481  * copy is p->ainsn.insn.
 482  *
 483  * This function prepares to return from the post-single-step
 484  * interrupt.  We have to fix up the stack as follows:
 485  *
 486  * 0) Except in the case of absolute or indirect jump or call instructions,
 487  * the new ip is relative to the copied instruction.  We need to make
 488  * it relative to the original instruction.
 489  *
 490  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 491  * flags are set in the just-pushed flags, and may need to be cleared.
 492  *
 493  * 2) If the single-stepped instruction was a call, the return address
 494  * that is atop the stack is the address following the copied instruction.
 495  * We need to make it the address following the original instruction.
 496  */
 497 static void __kprobes resume_execution(struct kprobe *p,
 498                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 499 {
 500         unsigned long *tos = (unsigned long *)regs->sp;
 501         unsigned long copy_rip = (unsigned long)p->ainsn.insn;
 502         unsigned long orig_rip = (unsigned long)p->addr;
 503         kprobe_opcode_t *insn = p->ainsn.insn;
 504
 505         /*skip the REX prefix*/
 506         if (*insn >= 0x40 && *insn <= 0x4f)
 507                 insn++;
 508
 509         regs->flags &= ~TF_MASK;
 510         switch (*insn) {
 511         case 0x9c:      /* pushfl */
 512                 *tos &= ~(TF_MASK | IF_MASK);
 513                 *tos |= kcb->kprobe_old_rflags;
 514                 break;
 515         case 0xc2:      /* iret/ret/lret */
 516         case 0xc3:
 517         case 0xca:
 518         case 0xcb:
 519         case 0xcf:
 520         case 0xea:      /* jmp absolute -- ip is correct */
 521                 /* ip is already adjusted, no more changes required */
 522                 goto no_change;
 523         case 0xe8:      /* call relative - Fix return addr */
 524                 *tos = orig_rip + (*tos - copy_rip);
 525                 break;
 526         case 0xff:
 527                 if ((insn[1] & 0x30) == 0x10) {
 528                         /* call absolute, indirect */
 529                         /* Fix return addr; ip is correct. */
 530                         *tos = orig_rip + (*tos - copy_rip);
 531                         goto no_change;
 532                 } else if (((insn[1] & 0x31) == 0x20) ||        /* jmp near, absolute indirect */
 533                            ((insn[1] & 0x31) == 0x21)) {        /* jmp far, absolute indirect */
 534                         /* ip is correct. */
 535                         goto no_change;
 536                 }
 537         default:
 538                 break;
 539         }
 540
 541         regs->ip = orig_rip + (regs->ip - copy_rip);
 542
 543 no_change:
 544         restore_btf();
 545
 546         return;
 547 }
 548
 549 int __kprobes post_kprobe_handler(struct pt_regs *regs)
 550 {
 551         struct kprobe *cur = kprobe_running();
 552         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 553
 554         if (!cur)
 555                 return 0;
 556
 557         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 558                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 559                 cur->post_handler(cur, regs, 0);
 560         }
 561
 562         resume_execution(cur, regs, kcb);
 563         regs->flags |= kcb->kprobe_saved_rflags;
 564         trace_hardirqs_fixup_flags(regs->flags);
 565
 566         /* Restore the original saved kprobes variables and continue. */
 567         if (kcb->kprobe_status == KPROBE_REENTER) {
 568                 restore_previous_kprobe(kcb);
 569                 goto out;
 570         }
 571         reset_current_kprobe();
 572 out:
 573         preempt_enable_no_resched();
 574
 575         /*
 576          * if somebody else is singlestepping across a probe point, flags
 577          * will have TF set, in which case, continue the remaining processing
 578          * of do_debug, as if this is not a probe hit.
 579          */
 580         if (regs->flags & TF_MASK)
 581                 return 0;
 582
 583         return 1;
 584 }
 585
 586 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 587 {
 588         struct kprobe *cur = kprobe_running();
 589         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 590         const struct exception_table_entry *fixup;
 591
 592         switch(kcb->kprobe_status) {
 593         case KPROBE_HIT_SS:
 594         case KPROBE_REENTER:
 595                 /*
 596                  * We are here because the instruction being single
 597                  * stepped caused a page fault. We reset the current
 598                  * kprobe and the ip points back to the probe address
 599                  * and allow the page fault handler to continue as a
 600                  * normal page fault.
 601                  */
 602                 regs->ip = (unsigned long)cur->addr;
 603                 regs->flags |= kcb->kprobe_old_rflags;
 604                 if (kcb->kprobe_status == KPROBE_REENTER)
 605                         restore_previous_kprobe(kcb);
 606                 else
 607                         reset_current_kprobe();
 608                 preempt_enable_no_resched();
 609                 break;
 610         case KPROBE_HIT_ACTIVE:
 611         case KPROBE_HIT_SSDONE:
 612                 /*
 613                  * We increment the nmissed count for accounting,
 614                  * we can also use npre/npostfault count for accouting
 615                  * these specific fault cases.
 616                  */
 617                 kprobes_inc_nmissed_count(cur);
 618
 619                 /*
 620                  * We come here because instructions in the pre/post
 621                  * handler caused the page_fault, this could happen
 622                  * if handler tries to access user space by
 623                  * copy_from_user(), get_user() etc. Let the
 624                  * user-specified handler try to fix it first.
 625                  */
 626                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 627                         return 1;
 628
 629                 /*
 630                  * In case the user-specified fault handler returned
 631                  * zero, try to fix up.
 632                  */
 633                 fixup = search_exception_tables(regs->ip);
 634                 if (fixup) {
 635                         regs->ip = fixup->fixup;
 636                         return 1;
 637                 }
 638
 639                 /*
 640                  * fixup() could not handle it,
 641                  * Let do_page_fault() fix it.
 642                  */
 643                 break;
 644         default:
 645                 break;
 646         }
 647         return 0;
 648 }
 649
 650 /*
 651  * Wrapper routine for handling exceptions.
 652  */
 653 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 654                                        unsigned long val, void *data)
 655 {
 656         struct die_args *args = (struct die_args *)data;
 657         int ret = NOTIFY_DONE;
 658
 659         if (args->regs && user_mode(args->regs))
 660                 return ret;
 661
 662         switch (val) {
 663         case DIE_INT3:
 664                 if (kprobe_handler(args->regs))
 665                         ret = NOTIFY_STOP;
 666                 break;
 667         case DIE_DEBUG:
 668                 if (post_kprobe_handler(args->regs))
 669                         ret = NOTIFY_STOP;
 670                 break;
 671         case DIE_GPF:
 672                 /* kprobe_running() needs smp_processor_id() */
 673                 preempt_disable();
 674                 if (kprobe_running() &&
 675                     kprobe_fault_handler(args->regs, args->trapnr))
 676                         ret = NOTIFY_STOP;
 677                 preempt_enable();
 678                 break;
 679         default:
 680                 break;
 681         }
 682         return ret;
 683 }
 684
 685 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 686 {
 687         struct jprobe *jp = container_of(p, struct jprobe, kp);
 688         unsigned long addr;
 689         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 690
 691         kcb->jprobe_saved_regs = *regs;
 692         kcb->jprobe_saved_rsp = (long *) regs->sp;
 693         addr = (unsigned long)(kcb->jprobe_saved_rsp);
 694         /*
 695          * As Linus pointed out, gcc assumes that the callee
 696          * owns the argument space and could overwrite it, e.g.
 697          * tailcall optimization. So, to be absolutely safe
 698          * we also save and restore enough stack bytes to cover
 699          * the argument area.
 700          */
 701         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
 702                         MIN_STACK_SIZE(addr));
 703         regs->flags &= ~IF_MASK;
 704         trace_hardirqs_off();
 705         regs->ip = (unsigned long)(jp->entry);
 706         return 1;
 707 }
 708
 709 void __kprobes jprobe_return(void)
 710 {
 711         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 712
 713         asm volatile ("       xchg   %%rbx,%%rsp     \n"
 714                       "       int3                      \n"
 715                       "       .globl jprobe_return_end  \n"
 716                       "       jprobe_return_end:        \n"
 717                       "       nop                       \n"::"b"
 718                       (kcb->jprobe_saved_rsp):"memory");
 719 }
 720
 721 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 722 {
 723         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 724         u8 *addr = (u8 *) (regs->ip - 1);
 725         unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_rsp);
 726         struct jprobe *jp = container_of(p, struct jprobe, kp);
 727
 728         if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
 729                 if ((unsigned long *)regs->sp != kcb->jprobe_saved_rsp) {
 730                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
 731                         printk("current sp %p does not match saved sp %p\n",
 732                                (long *)regs->sp, kcb->jprobe_saved_rsp);
 733                         printk("Saved registers for jprobe %p\n", jp);
 734                         show_registers(saved_regs);
 735                         printk("Current registers\n");
 736                         show_registers(regs);
 737                         BUG();
 738                 }
 739                 *regs = kcb->jprobe_saved_regs;
 740                 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
 741                        MIN_STACK_SIZE(stack_addr));
 742                 preempt_enable_no_resched();
 743                 return 1;
 744         }
 745         return 0;
 746 }
 747
 748 static struct kprobe trampoline_p = {
 749         .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
 750         .pre_handler = trampoline_probe_handler
 751 };
 752
 753 int __init arch_init_kprobes(void)
 754 {
 755         return register_kprobe(&trampoline_p);
 756 }
 757
 758 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
 759 {
 760         if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
 761                 return 1;
 762
 763         return 0;
 764 }