arch/x86_64/kernel/kprobes.c

   1 /*
   2  *  Kernel Probes (KProbes)
   3  *  arch/x86_64/kernel/kprobes.c
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18  *
  19  * Copyright (C) IBM Corporation, 2002, 2004
  20  *
  21  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  22  *              Probes initial implementation ( includes contributions from
  23  *              Rusty Russell).
  24  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  25  *              interface to access function arguments.
  26  * 2004-Oct     Jim Keniston <kenistoj@us.ibm.com> and Prasanna S Panchamukhi
  27  *              <prasanna@in.ibm.com> adapted for x86_64
  28  * 2005-Mar     Roland McGrath <roland@redhat.com>
  29  *              Fixed to handle %rip-relative addressing mode correctly.
  30  * 2005-May     Rusty Lynch <rusty.lynch@intel.com>
  31  *              Added function return probes functionality
  32  */
  33
  34 #include <linux/kprobes.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include <linux/preempt.h>
  39 #include <linux/module.h>
  40
  41 #include <asm/cacheflush.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/kdebug.h>
  44 #include <asm/uaccess.h>
  45
  46 void jprobe_return_end(void);
  47 static void __kprobes arch_copy_kprobe(struct kprobe *p);
  48
  49 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  50 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  51
  52 /*
  53  * returns non-zero if opcode modifies the interrupt flag.
  54  */
  55 static __always_inline int is_IF_modifier(kprobe_opcode_t *insn)
  56 {
  57         switch (*insn) {
  58         case 0xfa:              /* cli */
  59         case 0xfb:              /* sti */
  60         case 0xcf:              /* iret/iretd */
  61         case 0x9d:              /* popf/popfd */
  62                 return 1;
  63         }
  64
  65         if (*insn  >= 0x40 && *insn <= 0x4f && *++insn == 0xcf)
  66                 return 1;
  67         return 0;
  68 }
  69
  70 int __kprobes arch_prepare_kprobe(struct kprobe *p)
  71 {
  72         /* insn: must be on special executable page on x86_64. */
  73         p->ainsn.insn = get_insn_slot();
  74         if (!p->ainsn.insn) {
  75                 return -ENOMEM;
  76         }
  77         arch_copy_kprobe(p);
  78         return 0;
  79 }
  80
  81 /*
  82  * Determine if the instruction uses the %rip-relative addressing mode.
  83  * If it does, return the address of the 32-bit displacement word.
  84  * If not, return null.
  85  */
  86 static s32 __kprobes *is_riprel(u8 *insn)
  87 {
  88 #define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf)                \
  89         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  90           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  91           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  92           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  93          << (row % 64))
  94         static const u64 onebyte_has_modrm[256 / 64] = {
  95                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
  96                 /*      -------------------------------         */
  97                 W(0x00, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 00 */
  98                 W(0x10, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 10 */
  99                 W(0x20, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 20 */
 100                 W(0x30, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0), /* 30 */
 101                 W(0x40, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 40 */
 102                 W(0x50, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 50 */
 103                 W(0x60, 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0)| /* 60 */
 104                 W(0x70, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 70 */
 105                 W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
 106                 W(0x90, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 90 */
 107                 W(0xa0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* a0 */
 108                 W(0xb0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* b0 */
 109                 W(0xc0, 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0)| /* c0 */
 110                 W(0xd0, 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1)| /* d0 */
 111                 W(0xe0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* e0 */
 112                 W(0xf0, 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1)  /* f0 */
 113                 /*      -------------------------------         */
 114                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 115         };
 116         static const u64 twobyte_has_modrm[256 / 64] = {
 117                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 118                 /*      -------------------------------         */
 119                 W(0x00, 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1)| /* 0f */
 120                 W(0x10, 1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0)| /* 1f */
 121                 W(0x20, 1,1,1,1,1,0,1,0,1,1,1,1,1,1,1,1)| /* 2f */
 122                 W(0x30, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 3f */
 123                 W(0x40, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 4f */
 124                 W(0x50, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 5f */
 125                 W(0x60, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 6f */
 126                 W(0x70, 1,1,1,1,1,1,1,0,0,0,0,0,1,1,1,1), /* 7f */
 127                 W(0x80, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 8f */
 128                 W(0x90, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 9f */
 129                 W(0xa0, 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1)| /* af */
 130                 W(0xb0, 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1), /* bf */
 131                 W(0xc0, 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0)| /* cf */
 132                 W(0xd0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* df */
 133                 W(0xe0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* ef */
 134                 W(0xf0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0)  /* ff */
 135                 /*      -------------------------------         */
 136                 /*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
 137         };
 138 #undef  W
 139         int need_modrm;
 140
 141         /* Skip legacy instruction prefixes.  */
 142         while (1) {
 143                 switch (*insn) {
 144                 case 0x66:
 145                 case 0x67:
 146                 case 0x2e:
 147                 case 0x3e:
 148                 case 0x26:
 149                 case 0x64:
 150                 case 0x65:
 151                 case 0x36:
 152                 case 0xf0:
 153                 case 0xf3:
 154                 case 0xf2:
 155                         ++insn;
 156                         continue;
 157                 }
 158                 break;
 159         }
 160
 161         /* Skip REX instruction prefix.  */
 162         if ((*insn & 0xf0) == 0x40)
 163                 ++insn;
 164
 165         if (*insn == 0x0f) {    /* Two-byte opcode.  */
 166                 ++insn;
 167                 need_modrm = test_bit(*insn, twobyte_has_modrm);
 168         } else {                /* One-byte opcode.  */
 169                 need_modrm = test_bit(*insn, onebyte_has_modrm);
 170         }
 171
 172         if (need_modrm) {
 173                 u8 modrm = *++insn;
 174                 if ((modrm & 0xc7) == 0x05) { /* %rip+disp32 addressing mode */
 175                         /* Displacement follows ModRM byte.  */
 176                         return (s32 *) ++insn;
 177                 }
 178         }
 179
 180         /* No %rip-relative addressing mode here.  */
 181         return NULL;
 182 }
 183
 184 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 185 {
 186         s32 *ripdisp;
 187         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE);
 188         ripdisp = is_riprel(p->ainsn.insn);
 189         if (ripdisp) {
 190                 /*
 191                  * The copied instruction uses the %rip-relative
 192                  * addressing mode.  Adjust the displacement for the
 193                  * difference between the original location of this
 194                  * instruction and the location of the copy that will
 195                  * actually be run.  The tricky bit here is making sure
 196                  * that the sign extension happens correctly in this
 197                  * calculation, since we need a signed 32-bit result to
 198                  * be sign-extended to 64 bits when it's added to the
 199                  * %rip value and yield the same 64-bit result that the
 200                  * sign-extension of the original signed 32-bit
 201                  * displacement would have given.
 202                  */
 203                 s64 disp = (u8 *) p->addr + *ripdisp - (u8 *) p->ainsn.insn;
 204                 BUG_ON((s64) (s32) disp != disp); /* Sanity check.  */
 205                 *ripdisp = disp;
 206         }
 207         p->opcode = *p->addr;
 208 }
 209
 210 void __kprobes arch_arm_kprobe(struct kprobe *p)
 211 {
 212         *p->addr = BREAKPOINT_INSTRUCTION;
 213         flush_icache_range((unsigned long) p->addr,
 214                            (unsigned long) p->addr + sizeof(kprobe_opcode_t));
 215 }
 216
 217 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 218 {
 219         *p->addr = p->opcode;
 220         flush_icache_range((unsigned long) p->addr,
 221                            (unsigned long) p->addr + sizeof(kprobe_opcode_t));
 222 }
 223
 224 void __kprobes arch_remove_kprobe(struct kprobe *p)
 225 {
 226         mutex_lock(&kprobe_mutex);
 227         free_insn_slot(p->ainsn.insn);
 228         mutex_unlock(&kprobe_mutex);
 229 }
 230
 231 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 232 {
 233         kcb->prev_kprobe.kp = kprobe_running();
 234         kcb->prev_kprobe.status = kcb->kprobe_status;
 235         kcb->prev_kprobe.old_rflags = kcb->kprobe_old_rflags;
 236         kcb->prev_kprobe.saved_rflags = kcb->kprobe_saved_rflags;
 237 }
 238
 239 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 240 {
 241         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 242         kcb->kprobe_status = kcb->prev_kprobe.status;
 243         kcb->kprobe_old_rflags = kcb->prev_kprobe.old_rflags;
 244         kcb->kprobe_saved_rflags = kcb->prev_kprobe.saved_rflags;
 245 }
 246
 247 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 248                                 struct kprobe_ctlblk *kcb)
 249 {
 250         __get_cpu_var(current_kprobe) = p;
 251         kcb->kprobe_saved_rflags = kcb->kprobe_old_rflags
 252                 = (regs->eflags & (TF_MASK | IF_MASK));
 253         if (is_IF_modifier(p->ainsn.insn))
 254                 kcb->kprobe_saved_rflags &= ~IF_MASK;
 255 }
 256
 257 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 258 {
 259         regs->eflags |= TF_MASK;
 260         regs->eflags &= ~IF_MASK;
 261         /*single step inline if the instruction is an int3*/
 262         if (p->opcode == BREAKPOINT_INSTRUCTION)
 263                 regs->rip = (unsigned long)p->addr;
 264         else
 265                 regs->rip = (unsigned long)p->ainsn.insn;
 266 }
 267
 268 /* Called with kretprobe_lock held */
 269 void __kprobes arch_prepare_kretprobe(struct kretprobe *rp,
 270                                       struct pt_regs *regs)
 271 {
 272         unsigned long *sara = (unsigned long *)regs->rsp;
 273         struct kretprobe_instance *ri;
 274
 275         if ((ri = get_free_rp_inst(rp)) != NULL) {
 276                 ri->rp = rp;
 277                 ri->task = current;
 278                 ri->ret_addr = (kprobe_opcode_t *) *sara;
 279
 280                 /* Replace the return addr with trampoline addr */
 281                 *sara = (unsigned long) &kretprobe_trampoline;
 282
 283                 add_rp_inst(ri);
 284         } else {
 285                 rp->nmissed++;
 286         }
 287 }
 288
 289 int __kprobes kprobe_handler(struct pt_regs *regs)
 290 {
 291         struct kprobe *p;
 292         int ret = 0;
 293         kprobe_opcode_t *addr = (kprobe_opcode_t *)(regs->rip - sizeof(kprobe_opcode_t));
 294         struct kprobe_ctlblk *kcb;
 295
 296         /*
 297          * We don't want to be preempted for the entire
 298          * duration of kprobe processing
 299          */
 300         preempt_disable();
 301         kcb = get_kprobe_ctlblk();
 302
 303         /* Check we're not actually recursing */
 304         if (kprobe_running()) {
 305                 p = get_kprobe(addr);
 306                 if (p) {
 307                         if (kcb->kprobe_status == KPROBE_HIT_SS &&
 308                                 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
 309                                 regs->eflags &= ~TF_MASK;
 310                                 regs->eflags |= kcb->kprobe_saved_rflags;
 311                                 goto no_kprobe;
 312                         } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
 313                                 /* TODO: Provide re-entrancy from
 314                                  * post_kprobes_handler() and avoid exception
 315                                  * stack corruption while single-stepping on
 316                                  * the instruction of the new probe.
 317                                  */
 318                                 arch_disarm_kprobe(p);
 319                                 regs->rip = (unsigned long)p->addr;
 320                                 reset_current_kprobe();
 321                                 ret = 1;
 322                         } else {
 323                                 /* We have reentered the kprobe_handler(), since
 324                                  * another probe was hit while within the
 325                                  * handler. We here save the original kprobe
 326                                  * variables and just single step on instruction
 327                                  * of the new probe without calling any user
 328                                  * handlers.
 329                                  */
 330                                 save_previous_kprobe(kcb);
 331                                 set_current_kprobe(p, regs, kcb);
 332                                 kprobes_inc_nmissed_count(p);
 333                                 prepare_singlestep(p, regs);
 334                                 kcb->kprobe_status = KPROBE_REENTER;
 335                                 return 1;
 336                         }
 337                 } else {
 338                         if (*addr != BREAKPOINT_INSTRUCTION) {
 339                         /* The breakpoint instruction was removed by
 340                          * another cpu right after we hit, no further
 341                          * handling of this interrupt is appropriate
 342                          */
 343                                 regs->rip = (unsigned long)addr;
 344                                 ret = 1;
 345                                 goto no_kprobe;
 346                         }
 347                         p = __get_cpu_var(current_kprobe);
 348                         if (p->break_handler && p->break_handler(p, regs)) {
 349                                 goto ss_probe;
 350                         }
 351                 }
 352                 goto no_kprobe;
 353         }
 354
 355         p = get_kprobe(addr);
 356         if (!p) {
 357                 if (*addr != BREAKPOINT_INSTRUCTION) {
 358                         /*
 359                          * The breakpoint instruction was removed right
 360                          * after we hit it.  Another cpu has removed
 361                          * either a probepoint or a debugger breakpoint
 362                          * at this address.  In either case, no further
 363                          * handling of this interrupt is appropriate.
 364                          * Back up over the (now missing) int3 and run
 365                          * the original instruction.
 366                          */
 367                         regs->rip = (unsigned long)addr;
 368                         ret = 1;
 369                 }
 370                 /* Not one of ours: let kernel handle it */
 371                 goto no_kprobe;
 372         }
 373
 374         set_current_kprobe(p, regs, kcb);
 375         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 376
 377         if (p->pre_handler && p->pre_handler(p, regs))
 378                 /* handler has already set things up, so skip ss setup */
 379                 return 1;
 380
 381 ss_probe:
 382         prepare_singlestep(p, regs);
 383         kcb->kprobe_status = KPROBE_HIT_SS;
 384         return 1;
 385
 386 no_kprobe:
 387         preempt_enable_no_resched();
 388         return ret;
 389 }
 390
 391 /*
 392  * For function-return probes, init_kprobes() establishes a probepoint
 393  * here. When a retprobed function returns, this probe is hit and
 394  * trampoline_probe_handler() runs, calling the kretprobe's handler.
 395  */
 396  void kretprobe_trampoline_holder(void)
 397  {
 398         asm volatile (  ".global kretprobe_trampoline\n"
 399                         "kretprobe_trampoline: \n"
 400                         "nop\n");
 401  }
 402
 403 /*
 404  * Called when we hit the probe point at kretprobe_trampoline
 405  */
 406 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
 407 {
 408         struct kretprobe_instance *ri = NULL;
 409         struct hlist_head *head;
 410         struct hlist_node *node, *tmp;
 411         unsigned long flags, orig_ret_address = 0;
 412         unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
 413
 414         spin_lock_irqsave(&kretprobe_lock, flags);
 415         head = kretprobe_inst_table_head(current);
 416
 417         /*
 418          * It is possible to have multiple instances associated with a given
 419          * task either because an multiple functions in the call path
 420          * have a return probe installed on them, and/or more then one return
 421          * return probe was registered for a target function.
 422          *
 423          * We can handle this because:
 424          *     - instances are always inserted at the head of the list
 425          *     - when multiple return probes are registered for the same
 426          *       function, the first instance's ret_addr will point to the
 427          *       real return address, and all the rest will point to
 428          *       kretprobe_trampoline
 429          */
 430         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 431                 if (ri->task != current)
 432                         /* another task is sharing our hash bucket */
 433                         continue;
 434
 435                 if (ri->rp && ri->rp->handler)
 436                         ri->rp->handler(ri, regs);
 437
 438                 orig_ret_address = (unsigned long)ri->ret_addr;
 439                 recycle_rp_inst(ri);
 440
 441                 if (orig_ret_address != trampoline_address)
 442                         /*
 443                          * This is the real return address. Any other
 444                          * instances associated with this task are for
 445                          * other calls deeper on the call stack
 446                          */
 447                         break;
 448         }
 449
 450         BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
 451         regs->rip = orig_ret_address;
 452
 453         reset_current_kprobe();
 454         spin_unlock_irqrestore(&kretprobe_lock, flags);
 455         preempt_enable_no_resched();
 456
 457         /*
 458          * By returning a non-zero value, we are telling
 459          * kprobe_handler() that we don't want the post_handler
 460          * to run (and have re-enabled preemption)
 461          */
 462         return 1;
 463 }
 464
 465 /*
 466  * Called after single-stepping.  p->addr is the address of the
 467  * instruction whose first byte has been replaced by the "int 3"
 468  * instruction.  To avoid the SMP problems that can occur when we
 469  * temporarily put back the original opcode to single-step, we
 470  * single-stepped a copy of the instruction.  The address of this
 471  * copy is p->ainsn.insn.
 472  *
 473  * This function prepares to return from the post-single-step
 474  * interrupt.  We have to fix up the stack as follows:
 475  *
 476  * 0) Except in the case of absolute or indirect jump or call instructions,
 477  * the new rip is relative to the copied instruction.  We need to make
 478  * it relative to the original instruction.
 479  *
 480  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 481  * flags are set in the just-pushed eflags, and may need to be cleared.
 482  *
 483  * 2) If the single-stepped instruction was a call, the return address
 484  * that is atop the stack is the address following the copied instruction.
 485  * We need to make it the address following the original instruction.
 486  */
 487 static void __kprobes resume_execution(struct kprobe *p,
 488                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 489 {
 490         unsigned long *tos = (unsigned long *)regs->rsp;
 491         unsigned long next_rip = 0;
 492         unsigned long copy_rip = (unsigned long)p->ainsn.insn;
 493         unsigned long orig_rip = (unsigned long)p->addr;
 494         kprobe_opcode_t *insn = p->ainsn.insn;
 495
 496         /*skip the REX prefix*/
 497         if (*insn >= 0x40 && *insn <= 0x4f)
 498                 insn++;
 499
 500         switch (*insn) {
 501         case 0x9c:              /* pushfl */
 502                 *tos &= ~(TF_MASK | IF_MASK);
 503                 *tos |= kcb->kprobe_old_rflags;
 504                 break;
 505         case 0xc3:              /* ret/lret */
 506         case 0xcb:
 507         case 0xc2:
 508         case 0xca:
 509                 regs->eflags &= ~TF_MASK;
 510                 /* rip is already adjusted, no more changes required*/
 511                 return;
 512         case 0xe8:              /* call relative - Fix return addr */
 513                 *tos = orig_rip + (*tos - copy_rip);
 514                 break;
 515         case 0xff:
 516                 if ((insn[1] & 0x30) == 0x10) {
 517                         /* call absolute, indirect */
 518                         /* Fix return addr; rip is correct. */
 519                         next_rip = regs->rip;
 520                         *tos = orig_rip + (*tos - copy_rip);
 521                 } else if (((insn[1] & 0x31) == 0x20) ||        /* jmp near, absolute indirect */
 522                            ((insn[1] & 0x31) == 0x21)) {        /* jmp far, absolute indirect */
 523                         /* rip is correct. */
 524                         next_rip = regs->rip;
 525                 }
 526                 break;
 527         case 0xea:              /* jmp absolute -- rip is correct */
 528                 next_rip = regs->rip;
 529                 break;
 530         default:
 531                 break;
 532         }
 533
 534         regs->eflags &= ~TF_MASK;
 535         if (next_rip) {
 536                 regs->rip = next_rip;
 537         } else {
 538                 regs->rip = orig_rip + (regs->rip - copy_rip);
 539         }
 540 }
 541
 542 int __kprobes post_kprobe_handler(struct pt_regs *regs)
 543 {
 544         struct kprobe *cur = kprobe_running();
 545         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 546
 547         if (!cur)
 548                 return 0;
 549
 550         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 551                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 552                 cur->post_handler(cur, regs, 0);
 553         }
 554
 555         resume_execution(cur, regs, kcb);
 556         regs->eflags |= kcb->kprobe_saved_rflags;
 557
 558         /* Restore the original saved kprobes variables and continue. */
 559         if (kcb->kprobe_status == KPROBE_REENTER) {
 560                 restore_previous_kprobe(kcb);
 561                 goto out;
 562         }
 563         reset_current_kprobe();
 564 out:
 565         preempt_enable_no_resched();
 566
 567         /*
 568          * if somebody else is singlestepping across a probe point, eflags
 569          * will have TF set, in which case, continue the remaining processing
 570          * of do_debug, as if this is not a probe hit.
 571          */
 572         if (regs->eflags & TF_MASK)
 573                 return 0;
 574
 575         return 1;
 576 }
 577
 578 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 579 {
 580         struct kprobe *cur = kprobe_running();
 581         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 582         const struct exception_table_entry *fixup;
 583
 584         switch(kcb->kprobe_status) {
 585         case KPROBE_HIT_SS:
 586         case KPROBE_REENTER:
 587                 /*
 588                  * We are here because the instruction being single
 589                  * stepped caused a page fault. We reset the current
 590                  * kprobe and the rip points back to the probe address
 591                  * and allow the page fault handler to continue as a
 592                  * normal page fault.
 593                  */
 594                 regs->rip = (unsigned long)cur->addr;
 595                 regs->eflags |= kcb->kprobe_old_rflags;
 596                 if (kcb->kprobe_status == KPROBE_REENTER)
 597                         restore_previous_kprobe(kcb);
 598                 else
 599                         reset_current_kprobe();
 600                 preempt_enable_no_resched();
 601                 break;
 602         case KPROBE_HIT_ACTIVE:
 603         case KPROBE_HIT_SSDONE:
 604                 /*
 605                  * We increment the nmissed count for accounting,
 606                  * we can also use npre/npostfault count for accouting
 607                  * these specific fault cases.
 608                  */
 609                 kprobes_inc_nmissed_count(cur);
 610
 611                 /*
 612                  * We come here because instructions in the pre/post
 613                  * handler caused the page_fault, this could happen
 614                  * if handler tries to access user space by
 615                  * copy_from_user(), get_user() etc. Let the
 616                  * user-specified handler try to fix it first.
 617                  */
 618                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 619                         return 1;
 620
 621                 /*
 622                  * In case the user-specified fault handler returned
 623                  * zero, try to fix up.
 624                  */
 625                 fixup = search_exception_tables(regs->rip);
 626                 if (fixup) {
 627                         regs->rip = fixup->fixup;
 628                         return 1;
 629                 }
 630
 631                 /*
 632                  * fixup() could not handle it,
 633                  * Let do_page_fault() fix it.
 634                  */
 635                 break;
 636         default:
 637                 break;
 638         }
 639         return 0;
 640 }
 641
 642 /*
 643  * Wrapper routine for handling exceptions.
 644  */
 645 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 646                                        unsigned long val, void *data)
 647 {
 648         struct die_args *args = (struct die_args *)data;
 649         int ret = NOTIFY_DONE;
 650
 651         if (args->regs && user_mode(args->regs))
 652                 return ret;
 653
 654         switch (val) {
 655         case DIE_INT3:
 656                 if (kprobe_handler(args->regs))
 657                         ret = NOTIFY_STOP;
 658                 break;
 659         case DIE_DEBUG:
 660                 if (post_kprobe_handler(args->regs))
 661                         ret = NOTIFY_STOP;
 662                 break;
 663         case DIE_GPF:
 664         case DIE_PAGE_FAULT:
 665                 /* kprobe_running() needs smp_processor_id() */
 666                 preempt_disable();
 667                 if (kprobe_running() &&
 668                     kprobe_fault_handler(args->regs, args->trapnr))
 669                         ret = NOTIFY_STOP;
 670                 preempt_enable();
 671                 break;
 672         default:
 673                 break;
 674         }
 675         return ret;
 676 }
 677
 678 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 679 {
 680         struct jprobe *jp = container_of(p, struct jprobe, kp);
 681         unsigned long addr;
 682         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 683
 684         kcb->jprobe_saved_regs = *regs;
 685         kcb->jprobe_saved_rsp = (long *) regs->rsp;
 686         addr = (unsigned long)(kcb->jprobe_saved_rsp);
 687         /*
 688          * As Linus pointed out, gcc assumes that the callee
 689          * owns the argument space and could overwrite it, e.g.
 690          * tailcall optimization. So, to be absolutely safe
 691          * we also save and restore enough stack bytes to cover
 692          * the argument area.
 693          */
 694         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
 695                         MIN_STACK_SIZE(addr));
 696         regs->eflags &= ~IF_MASK;
 697         regs->rip = (unsigned long)(jp->entry);
 698         return 1;
 699 }
 700
 701 void __kprobes jprobe_return(void)
 702 {
 703         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 704
 705         asm volatile ("       xchg   %%rbx,%%rsp     \n"
 706                       "       int3                      \n"
 707                       "       .globl jprobe_return_end  \n"
 708                       "       jprobe_return_end:        \n"
 709                       "       nop                       \n"::"b"
 710                       (kcb->jprobe_saved_rsp):"memory");
 711 }
 712
 713 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 714 {
 715         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 716         u8 *addr = (u8 *) (regs->rip - 1);
 717         unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_rsp);
 718         struct jprobe *jp = container_of(p, struct jprobe, kp);
 719
 720         if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
 721                 if ((long *)regs->rsp != kcb->jprobe_saved_rsp) {
 722                         struct pt_regs *saved_regs =
 723                             container_of(kcb->jprobe_saved_rsp,
 724                                             struct pt_regs, rsp);
 725                         printk("current rsp %p does not match saved rsp %p\n",
 726                                (long *)regs->rsp, kcb->jprobe_saved_rsp);
 727                         printk("Saved registers for jprobe %p\n", jp);
 728                         show_registers(saved_regs);
 729                         printk("Current registers\n");
 730                         show_registers(regs);
 731                         BUG();
 732                 }
 733                 *regs = kcb->jprobe_saved_regs;
 734                 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
 735                        MIN_STACK_SIZE(stack_addr));
 736                 preempt_enable_no_resched();
 737                 return 1;
 738         }
 739         return 0;
 740 }
 741
 742 static struct kprobe trampoline_p = {
 743         .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
 744         .pre_handler = trampoline_probe_handler
 745 };
 746
 747 int __init arch_init_kprobes(void)
 748 {
 749         return register_kprobe(&trampoline_p);
 750 }