arch/x86/kernel/kprobes.c

   1 /*
   2  *  Kernel Probes (KProbes)
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright (C) IBM Corporation, 2002, 2004
  19  *
  20  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  21  *              Probes initial implementation ( includes contributions from
  22  *              Rusty Russell).
  23  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  24  *              interface to access function arguments.
  25  * 2004-Oct     Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  26  *              <prasanna@in.ibm.com> adapted for x86_64 from i386.
  27  * 2005-Mar     Roland McGrath <roland@redhat.com>
  28  *              Fixed to handle %rip-relative addressing mode correctly.
  29  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  30  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  31  *              <prasanna@in.ibm.com> added function-return probes.
  32  * 2005-May     Rusty Lynch <rusty.lynch@intel.com>
  33  *              Added function return probes functionality
  34  * 2006-Feb     Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added
  35  *              kprobe-booster and kretprobe-booster for i386.
  36  * 2007-Dec     Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
  37  *              and kretprobe-booster for x86-64
  38  * 2007-Dec     Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven
  39  *              <arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com>
  40  *              unified x86 kprobes code.
  41  */
  42
  43 #include <linux/kprobes.h>
  44 #include <linux/ptrace.h>
  45 #include <linux/string.h>
  46 #include <linux/slab.h>
  47 #include <linux/hardirq.h>
  48 #include <linux/preempt.h>
  49 #include <linux/module.h>
  50 #include <linux/kdebug.h>
  51 #include <linux/kallsyms.h>
  52
  53 #include <asm/cacheflush.h>
  54 #include <asm/desc.h>
  55 #include <asm/pgtable.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/alternative.h>
  58 #include <asm/insn.h>
  59 #include <asm/debugreg.h>
  60
  61 void jprobe_return_end(void);
  62
  63 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  64 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  65
  66 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
  67
  68 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
  69         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  70           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  71           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  72           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  73          << (row % 32))
  74         /*
  75          * Undefined/reserved opcodes, conditional jump, Opcode Extension
  76          * Groups, and some special opcodes can not boost.
  77          */
  78 static const u32 twobyte_is_boostable[256 / 32] = {
  79         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  80         /*      ----------------------------------------------          */
  81         W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
  82         W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
  83         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
  84         W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
  85         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
  86         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
  87         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
  88         W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
  89         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
  90         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
  91         W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
  92         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
  93         W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
  94         W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
  95         W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
  96         W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
  97         /*      -----------------------------------------------         */
  98         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  99 };
 100 #undef W
 101
 102 struct kretprobe_blackpoint kretprobe_blacklist[] = {
 103         {"__switch_to", }, /* This function switches only current task, but
 104                               doesn't switch kernel stack.*/
 105         {NULL, NULL}    /* Terminator */
 106 };
 107 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
 108
 109 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
 110 static void __kprobes set_jmp_op(void *from, void *to)
 111 {
 112         struct __arch_jmp_op {
 113                 char op;
 114                 s32 raddr;
 115         } __attribute__((packed)) * jop;
 116         jop = (struct __arch_jmp_op *)from;
 117         jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
 118         jop->op = RELATIVEJUMP_INSTRUCTION;
 119 }
 120
 121 /*
 122  * Check for the REX prefix which can only exist on X86_64
 123  * X86_32 always returns 0
 124  */
 125 static int __kprobes is_REX_prefix(kprobe_opcode_t *insn)
 126 {
 127 #ifdef CONFIG_X86_64
 128         if ((*insn & 0xf0) == 0x40)
 129                 return 1;
 130 #endif
 131         return 0;
 132 }
 133
 134 /*
 135  * Returns non-zero if opcode is boostable.
 136  * RIP relative instructions are adjusted at copying time in 64 bits mode
 137  */
 138 static int __kprobes can_boost(kprobe_opcode_t *opcodes)
 139 {
 140         kprobe_opcode_t opcode;
 141         kprobe_opcode_t *orig_opcodes = opcodes;
 142
 143         if (search_exception_tables((unsigned long)opcodes))
 144                 return 0;       /* Page fault may occur on this address. */
 145
 146 retry:
 147         if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 148                 return 0;
 149         opcode = *(opcodes++);
 150
 151         /* 2nd-byte opcode */
 152         if (opcode == 0x0f) {
 153                 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 154                         return 0;
 155                 return test_bit(*opcodes,
 156                                 (unsigned long *)twobyte_is_boostable);
 157         }
 158
 159         switch (opcode & 0xf0) {
 160 #ifdef CONFIG_X86_64
 161         case 0x40:
 162                 goto retry; /* REX prefix is boostable */
 163 #endif
 164         case 0x60:
 165                 if (0x63 < opcode && opcode < 0x67)
 166                         goto retry; /* prefixes */
 167                 /* can't boost Address-size override and bound */
 168                 return (opcode != 0x62 && opcode != 0x67);
 169         case 0x70:
 170                 return 0; /* can't boost conditional jump */
 171         case 0xc0:
 172                 /* can't boost software-interruptions */
 173                 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
 174         case 0xd0:
 175                 /* can boost AA* and XLAT */
 176                 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
 177         case 0xe0:
 178                 /* can boost in/out and absolute jmps */
 179                 return ((opcode & 0x04) || opcode == 0xea);
 180         case 0xf0:
 181                 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
 182                         goto retry; /* lock/rep(ne) prefix */
 183                 /* clear and set flags are boostable */
 184                 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
 185         default:
 186                 /* segment override prefixes are boostable */
 187                 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
 188                         goto retry; /* prefixes */
 189                 /* CS override prefix and call are not boostable */
 190                 return (opcode != 0x2e && opcode != 0x9a);
 191         }
 192 }
 193
 194 /* Recover the probed instruction at addr for further analysis. */
 195 static int recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
 196 {
 197         struct kprobe *kp;
 198         kp = get_kprobe((void *)addr);
 199         if (!kp)
 200                 return -EINVAL;
 201
 202         /*
 203          *  Basically, kp->ainsn.insn has an original instruction.
 204          *  However, RIP-relative instruction can not do single-stepping
 205          *  at different place, fix_riprel() tweaks the displacement of
 206          *  that instruction. In that case, we can't recover the instruction
 207          *  from the kp->ainsn.insn.
 208          *
 209          *  On the other hand, kp->opcode has a copy of the first byte of
 210          *  the probed instruction, which is overwritten by int3. And
 211          *  the instruction at kp->addr is not modified by kprobes except
 212          *  for the first byte, we can recover the original instruction
 213          *  from it and kp->opcode.
 214          */
 215         memcpy(buf, kp->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 216         buf[0] = kp->opcode;
 217         return 0;
 218 }
 219
 220 /* Dummy buffers for kallsyms_lookup */
 221 static char __dummy_buf[KSYM_NAME_LEN];
 222
 223 /* Check if paddr is at an instruction boundary */
 224 static int __kprobes can_probe(unsigned long paddr)
 225 {
 226         int ret;
 227         unsigned long addr, offset = 0;
 228         struct insn insn;
 229         kprobe_opcode_t buf[MAX_INSN_SIZE];
 230
 231         if (!kallsyms_lookup(paddr, NULL, &offset, NULL, __dummy_buf))
 232                 return 0;
 233
 234         /* Decode instructions */
 235         addr = paddr - offset;
 236         while (addr < paddr) {
 237                 kernel_insn_init(&insn, (void *)addr);
 238                 insn_get_opcode(&insn);
 239
 240                 /*
 241                  * Check if the instruction has been modified by another
 242                  * kprobe, in which case we replace the breakpoint by the
 243                  * original instruction in our buffer.
 244                  */
 245                 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
 246                         ret = recover_probed_instruction(buf, addr);
 247                         if (ret)
 248                                 /*
 249                                  * Another debugging subsystem might insert
 250                                  * this breakpoint. In that case, we can't
 251                                  * recover it.
 252                                  */
 253                                 return 0;
 254                         kernel_insn_init(&insn, buf);
 255                 }
 256                 insn_get_length(&insn);
 257                 addr += insn.length;
 258         }
 259
 260         return (addr == paddr);
 261 }
 262
 263 /*
 264  * Returns non-zero if opcode modifies the interrupt flag.
 265  */
 266 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
 267 {
 268         switch (*insn) {
 269         case 0xfa:              /* cli */
 270         case 0xfb:              /* sti */
 271         case 0xcf:              /* iret/iretd */
 272         case 0x9d:              /* popf/popfd */
 273                 return 1;
 274         }
 275
 276         /*
 277          * on X86_64, 0x40-0x4f are REX prefixes so we need to look
 278          * at the next byte instead.. but of course not recurse infinitely
 279          */
 280         if (is_REX_prefix(insn))
 281                 return is_IF_modifier(++insn);
 282
 283         return 0;
 284 }
 285
 286 /*
 287  * Adjust the displacement if the instruction uses the %rip-relative
 288  * addressing mode.
 289  * If it does, Return the address of the 32-bit displacement word.
 290  * If not, return null.
 291  * Only applicable to 64-bit x86.
 292  */
 293 static void __kprobes fix_riprel(struct kprobe *p)
 294 {
 295 #ifdef CONFIG_X86_64
 296         struct insn insn;
 297         kernel_insn_init(&insn, p->ainsn.insn);
 298
 299         if (insn_rip_relative(&insn)) {
 300                 s64 newdisp;
 301                 u8 *disp;
 302                 insn_get_displacement(&insn);
 303                 /*
 304                  * The copied instruction uses the %rip-relative addressing
 305                  * mode.  Adjust the displacement for the difference between
 306                  * the original location of this instruction and the location
 307                  * of the copy that will actually be run.  The tricky bit here
 308                  * is making sure that the sign extension happens correctly in
 309                  * this calculation, since we need a signed 32-bit result to
 310                  * be sign-extended to 64 bits when it's added to the %rip
 311                  * value and yield the same 64-bit result that the sign-
 312                  * extension of the original signed 32-bit displacement would
 313                  * have given.
 314                  */
 315                 newdisp = (u8 *) p->addr + (s64) insn.displacement.value -
 316                           (u8 *) p->ainsn.insn;
 317                 BUG_ON((s64) (s32) newdisp != newdisp); /* Sanity check.  */
 318                 disp = (u8 *) p->ainsn.insn + insn_offset_displacement(&insn);
 319                 *(s32 *) disp = (s32) newdisp;
 320         }
 321 #endif
 322 }
 323
 324 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 325 {
 326         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 327
 328         fix_riprel(p);
 329
 330         if (can_boost(p->addr))
 331                 p->ainsn.boostable = 0;
 332         else
 333                 p->ainsn.boostable = -1;
 334
 335         p->opcode = *p->addr;
 336 }
 337
 338 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 339 {
 340         if (!can_probe((unsigned long)p->addr))
 341                 return -EILSEQ;
 342         /* insn: must be on special executable page on x86. */
 343         p->ainsn.insn = get_insn_slot();
 344         if (!p->ainsn.insn)
 345                 return -ENOMEM;
 346         arch_copy_kprobe(p);
 347         return 0;
 348 }
 349
 350 void __kprobes arch_arm_kprobe(struct kprobe *p)
 351 {
 352         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 353 }
 354
 355 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 356 {
 357         text_poke(p->addr, &p->opcode, 1);
 358 }
 359
 360 void __kprobes arch_remove_kprobe(struct kprobe *p)
 361 {
 362         if (p->ainsn.insn) {
 363                 free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
 364                 p->ainsn.insn = NULL;
 365         }
 366 }
 367
 368 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 369 {
 370         kcb->prev_kprobe.kp = kprobe_running();
 371         kcb->prev_kprobe.status = kcb->kprobe_status;
 372         kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
 373         kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
 374 }
 375
 376 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 377 {
 378         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 379         kcb->kprobe_status = kcb->prev_kprobe.status;
 380         kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
 381         kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
 382 }
 383
 384 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 385                                 struct kprobe_ctlblk *kcb)
 386 {
 387         __get_cpu_var(current_kprobe) = p;
 388         kcb->kprobe_saved_flags = kcb->kprobe_old_flags
 389                 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF));
 390         if (is_IF_modifier(p->ainsn.insn))
 391                 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF;
 392 }
 393
 394 static void __kprobes clear_btf(void)
 395 {
 396         if (test_thread_flag(TIF_DEBUGCTLMSR))
 397                 update_debugctlmsr(0);
 398 }
 399
 400 static void __kprobes restore_btf(void)
 401 {
 402         if (test_thread_flag(TIF_DEBUGCTLMSR))
 403                 update_debugctlmsr(current->thread.debugctlmsr);
 404 }
 405
 406 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 407 {
 408         clear_btf();
 409         regs->flags |= X86_EFLAGS_TF;
 410         regs->flags &= ~X86_EFLAGS_IF;
 411         /* single step inline if the instruction is an int3 */
 412         if (p->opcode == BREAKPOINT_INSTRUCTION)
 413                 regs->ip = (unsigned long)p->addr;
 414         else
 415                 regs->ip = (unsigned long)p->ainsn.insn;
 416 }
 417
 418 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 419                                       struct pt_regs *regs)
 420 {
 421         unsigned long *sara = stack_addr(regs);
 422
 423         ri->ret_addr = (kprobe_opcode_t *) *sara;
 424
 425         /* Replace the return addr with trampoline addr */
 426         *sara = (unsigned long) &kretprobe_trampoline;
 427 }
 428
 429 static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
 430                                        struct kprobe_ctlblk *kcb)
 431 {
 432 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
 433         if (p->ainsn.boostable == 1 && !p->post_handler) {
 434                 /* Boost up -- we can execute copied instructions directly */
 435                 reset_current_kprobe();
 436                 regs->ip = (unsigned long)p->ainsn.insn;
 437                 preempt_enable_no_resched();
 438                 return;
 439         }
 440 #endif
 441         prepare_singlestep(p, regs);
 442         kcb->kprobe_status = KPROBE_HIT_SS;
 443 }
 444
 445 /*
 446  * We have reentered the kprobe_handler(), since another probe was hit while
 447  * within the handler. We save the original kprobes variables and just single
 448  * step on the instruction of the new probe without calling any user handlers.
 449  */
 450 static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
 451                                     struct kprobe_ctlblk *kcb)
 452 {
 453         switch (kcb->kprobe_status) {
 454         case KPROBE_HIT_SSDONE:
 455         case KPROBE_HIT_ACTIVE:
 456                 save_previous_kprobe(kcb);
 457                 set_current_kprobe(p, regs, kcb);
 458                 kprobes_inc_nmissed_count(p);
 459                 prepare_singlestep(p, regs);
 460                 kcb->kprobe_status = KPROBE_REENTER;
 461                 break;
 462         case KPROBE_HIT_SS:
 463                 /* A probe has been hit in the codepath leading up to, or just
 464                  * after, single-stepping of a probed instruction. This entire
 465                  * codepath should strictly reside in .kprobes.text section.
 466                  * Raise a BUG or we'll continue in an endless reentering loop
 467                  * and eventually a stack overflow.
 468                  */
 469                 printk(KERN_WARNING "Unrecoverable kprobe detected at %p.\n",
 470                        p->addr);
 471                 dump_kprobe(p);
 472                 BUG();
 473         default:
 474                 /* impossible cases */
 475                 WARN_ON(1);
 476                 return 0;
 477         }
 478
 479         return 1;
 480 }
 481
 482 /*
 483  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 484  * remain disabled thorough out this function.
 485  */
 486 static int __kprobes kprobe_handler(struct pt_regs *regs)
 487 {
 488         kprobe_opcode_t *addr;
 489         struct kprobe *p;
 490         struct kprobe_ctlblk *kcb;
 491
 492         addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 493         if (*addr != BREAKPOINT_INSTRUCTION) {
 494                 /*
 495                  * The breakpoint instruction was removed right
 496                  * after we hit it.  Another cpu has removed
 497                  * either a probepoint or a debugger breakpoint
 498                  * at this address.  In either case, no further
 499                  * handling of this interrupt is appropriate.
 500                  * Back up over the (now missing) int3 and run
 501                  * the original instruction.
 502                  */
 503                 regs->ip = (unsigned long)addr;
 504                 return 1;
 505         }
 506
 507         /*
 508          * We don't want to be preempted for the entire
 509          * duration of kprobe processing. We conditionally
 510          * re-enable preemption at the end of this function,
 511          * and also in reenter_kprobe() and setup_singlestep().
 512          */
 513         preempt_disable();
 514
 515         kcb = get_kprobe_ctlblk();
 516         p = get_kprobe(addr);
 517
 518         if (p) {
 519                 if (kprobe_running()) {
 520                         if (reenter_kprobe(p, regs, kcb))
 521                                 return 1;
 522                 } else {
 523                         set_current_kprobe(p, regs, kcb);
 524                         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 525
 526                         /*
 527                          * If we have no pre-handler or it returned 0, we
 528                          * continue with normal processing.  If we have a
 529                          * pre-handler and it returned non-zero, it prepped
 530                          * for calling the break_handler below on re-entry
 531                          * for jprobe processing, so get out doing nothing
 532                          * more here.
 533                          */
 534                         if (!p->pre_handler || !p->pre_handler(p, regs))
 535                                 setup_singlestep(p, regs, kcb);
 536                         return 1;
 537                 }
 538         } else if (kprobe_running()) {
 539                 p = __get_cpu_var(current_kprobe);
 540                 if (p->break_handler && p->break_handler(p, regs)) {
 541                         setup_singlestep(p, regs, kcb);
 542                         return 1;
 543                 }
 544         } /* else: not a kprobe fault; let the kernel handle it */
 545
 546         preempt_enable_no_resched();
 547         return 0;
 548 }
 549
 550 /*
 551  * When a retprobed function returns, this code saves registers and
 552  * calls trampoline_handler() runs, which calls the kretprobe's handler.
 553  */
 554 static void __used __kprobes kretprobe_trampoline_holder(void)
 555 {
 556         asm volatile (
 557                         ".global kretprobe_trampoline\n"
 558                         "kretprobe_trampoline: \n"
 559 #ifdef CONFIG_X86_64
 560                         /* We don't bother saving the ss register */
 561                         "       pushq %rsp\n"
 562                         "       pushfq\n"
 563                         /*
 564                          * Skip cs, ip, orig_ax.
 565                          * trampoline_handler() will plug in these values
 566                          */
 567                         "       subq $24, %rsp\n"
 568                         "       pushq %rdi\n"
 569                         "       pushq %rsi\n"
 570                         "       pushq %rdx\n"
 571                         "       pushq %rcx\n"
 572                         "       pushq %rax\n"
 573                         "       pushq %r8\n"
 574                         "       pushq %r9\n"
 575                         "       pushq %r10\n"
 576                         "       pushq %r11\n"
 577                         "       pushq %rbx\n"
 578                         "       pushq %rbp\n"
 579                         "       pushq %r12\n"
 580                         "       pushq %r13\n"
 581                         "       pushq %r14\n"
 582                         "       pushq %r15\n"
 583                         "       movq %rsp, %rdi\n"
 584                         "       call trampoline_handler\n"
 585                         /* Replace saved sp with true return address. */
 586                         "       movq %rax, 152(%rsp)\n"
 587                         "       popq %r15\n"
 588                         "       popq %r14\n"
 589                         "       popq %r13\n"
 590                         "       popq %r12\n"
 591                         "       popq %rbp\n"
 592                         "       popq %rbx\n"
 593                         "       popq %r11\n"
 594                         "       popq %r10\n"
 595                         "       popq %r9\n"
 596                         "       popq %r8\n"
 597                         "       popq %rax\n"
 598                         "       popq %rcx\n"
 599                         "       popq %rdx\n"
 600                         "       popq %rsi\n"
 601                         "       popq %rdi\n"
 602                         /* Skip orig_ax, ip, cs */
 603                         "       addq $24, %rsp\n"
 604                         "       popfq\n"
 605 #else
 606                         "       pushf\n"
 607                         /*
 608                          * Skip cs, ip, orig_ax and gs.
 609                          * trampoline_handler() will plug in these values
 610                          */
 611                         "       subl $16, %esp\n"
 612                         "       pushl %fs\n"
 613                         "       pushl %es\n"
 614                         "       pushl %ds\n"
 615                         "       pushl %eax\n"
 616                         "       pushl %ebp\n"
 617                         "       pushl %edi\n"
 618                         "       pushl %esi\n"
 619                         "       pushl %edx\n"
 620                         "       pushl %ecx\n"
 621                         "       pushl %ebx\n"
 622                         "       movl %esp, %eax\n"
 623                         "       call trampoline_handler\n"
 624                         /* Move flags to cs */
 625                         "       movl 56(%esp), %edx\n"
 626                         "       movl %edx, 52(%esp)\n"
 627                         /* Replace saved flags with true return address. */
 628                         "       movl %eax, 56(%esp)\n"
 629                         "       popl %ebx\n"
 630                         "       popl %ecx\n"
 631                         "       popl %edx\n"
 632                         "       popl %esi\n"
 633                         "       popl %edi\n"
 634                         "       popl %ebp\n"
 635                         "       popl %eax\n"
 636                         /* Skip ds, es, fs, gs, orig_ax and ip */
 637                         "       addl $24, %esp\n"
 638                         "       popf\n"
 639 #endif
 640                         "       ret\n");
 641 }
 642
 643 /*
 644  * Called from kretprobe_trampoline
 645  */
 646 static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 647 {
 648         struct kretprobe_instance *ri = NULL;
 649         struct hlist_head *head, empty_rp;
 650         struct hlist_node *node, *tmp;
 651         unsigned long flags, orig_ret_address = 0;
 652         unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
 653
 654         INIT_HLIST_HEAD(&empty_rp);
 655         kretprobe_hash_lock(current, &head, &flags);
 656         /* fixup registers */
 657 #ifdef CONFIG_X86_64
 658         regs->cs = __KERNEL_CS;
 659 #else
 660         regs->cs = __KERNEL_CS | get_kernel_rpl();
 661         regs->gs = 0;
 662 #endif
 663         regs->ip = trampoline_address;
 664         regs->orig_ax = ~0UL;
 665
 666         /*
 667          * It is possible to have multiple instances associated with a given
 668          * task either because multiple functions in the call path have
 669          * return probes installed on them, and/or more than one
 670          * return probe was registered for a target function.
 671          *
 672          * We can handle this because:
 673          *     - instances are always pushed into the head of the list
 674          *     - when multiple return probes are registered for the same
 675          *       function, the (chronologically) first instance's ret_addr
 676          *       will be the real return address, and all the rest will
 677          *       point to kretprobe_trampoline.
 678          */
 679         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 680                 if (ri->task != current)
 681                         /* another task is sharing our hash bucket */
 682                         continue;
 683
 684                 if (ri->rp && ri->rp->handler) {
 685                         __get_cpu_var(current_kprobe) = &ri->rp->kp;
 686                         get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
 687                         ri->rp->handler(ri, regs);
 688                         __get_cpu_var(current_kprobe) = NULL;
 689                 }
 690
 691                 orig_ret_address = (unsigned long)ri->ret_addr;
 692                 recycle_rp_inst(ri, &empty_rp);
 693
 694                 if (orig_ret_address != trampoline_address)
 695                         /*
 696                          * This is the real return address. Any other
 697                          * instances associated with this task are for
 698                          * other calls deeper on the call stack
 699                          */
 700                         break;
 701         }
 702
 703         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 704
 705         kretprobe_hash_unlock(current, &flags);
 706
 707         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 708                 hlist_del(&ri->hlist);
 709                 kfree(ri);
 710         }
 711         return (void *)orig_ret_address;
 712 }
 713
 714 /*
 715  * Called after single-stepping.  p->addr is the address of the
 716  * instruction whose first byte has been replaced by the "int 3"
 717  * instruction.  To avoid the SMP problems that can occur when we
 718  * temporarily put back the original opcode to single-step, we
 719  * single-stepped a copy of the instruction.  The address of this
 720  * copy is p->ainsn.insn.
 721  *
 722  * This function prepares to return from the post-single-step
 723  * interrupt.  We have to fix up the stack as follows:
 724  *
 725  * 0) Except in the case of absolute or indirect jump or call instructions,
 726  * the new ip is relative to the copied instruction.  We need to make
 727  * it relative to the original instruction.
 728  *
 729  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 730  * flags are set in the just-pushed flags, and may need to be cleared.
 731  *
 732  * 2) If the single-stepped instruction was a call, the return address
 733  * that is atop the stack is the address following the copied instruction.
 734  * We need to make it the address following the original instruction.
 735  *
 736  * If this is the first time we've single-stepped the instruction at
 737  * this probepoint, and the instruction is boostable, boost it: add a
 738  * jump instruction after the copied instruction, that jumps to the next
 739  * instruction after the probepoint.
 740  */
 741 static void __kprobes resume_execution(struct kprobe *p,
 742                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 743 {
 744         unsigned long *tos = stack_addr(regs);
 745         unsigned long copy_ip = (unsigned long)p->ainsn.insn;
 746         unsigned long orig_ip = (unsigned long)p->addr;
 747         kprobe_opcode_t *insn = p->ainsn.insn;
 748
 749         /*skip the REX prefix*/
 750         if (is_REX_prefix(insn))
 751                 insn++;
 752
 753         regs->flags &= ~X86_EFLAGS_TF;
 754         switch (*insn) {
 755         case 0x9c:      /* pushfl */
 756                 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF);
 757                 *tos |= kcb->kprobe_old_flags;
 758                 break;
 759         case 0xc2:      /* iret/ret/lret */
 760         case 0xc3:
 761         case 0xca:
 762         case 0xcb:
 763         case 0xcf:
 764         case 0xea:      /* jmp absolute -- ip is correct */
 765                 /* ip is already adjusted, no more changes required */
 766                 p->ainsn.boostable = 1;
 767                 goto no_change;
 768         case 0xe8:      /* call relative - Fix return addr */
 769                 *tos = orig_ip + (*tos - copy_ip);
 770                 break;
 771 #ifdef CONFIG_X86_32
 772         case 0x9a:      /* call absolute -- same as call absolute, indirect */
 773                 *tos = orig_ip + (*tos - copy_ip);
 774                 goto no_change;
 775 #endif
 776         case 0xff:
 777                 if ((insn[1] & 0x30) == 0x10) {
 778                         /*
 779                          * call absolute, indirect
 780                          * Fix return addr; ip is correct.
 781                          * But this is not boostable
 782                          */
 783                         *tos = orig_ip + (*tos - copy_ip);
 784                         goto no_change;
 785                 } else if (((insn[1] & 0x31) == 0x20) ||
 786                            ((insn[1] & 0x31) == 0x21)) {
 787                         /*
 788                          * jmp near and far, absolute indirect
 789                          * ip is correct. And this is boostable
 790                          */
 791                         p->ainsn.boostable = 1;
 792                         goto no_change;
 793                 }
 794         default:
 795                 break;
 796         }
 797
 798         if (p->ainsn.boostable == 0) {
 799                 if ((regs->ip > copy_ip) &&
 800                     (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
 801                         /*
 802                          * These instructions can be executed directly if it
 803                          * jumps back to correct address.
 804                          */
 805                         set_jmp_op((void *)regs->ip,
 806                                    (void *)orig_ip + (regs->ip - copy_ip));
 807                         p->ainsn.boostable = 1;
 808                 } else {
 809                         p->ainsn.boostable = -1;
 810                 }
 811         }
 812
 813         regs->ip += orig_ip - copy_ip;
 814
 815 no_change:
 816         restore_btf();
 817 }
 818
 819 /*
 820  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 821  * remain disabled thoroughout this function.
 822  */
 823 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
 824 {
 825         struct kprobe *cur = kprobe_running();
 826         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 827
 828         if (!cur)
 829                 return 0;
 830
 831         resume_execution(cur, regs, kcb);
 832         regs->flags |= kcb->kprobe_saved_flags;
 833
 834         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 835                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 836                 cur->post_handler(cur, regs, 0);
 837         }
 838
 839         /* Restore back the original saved kprobes variables and continue. */
 840         if (kcb->kprobe_status == KPROBE_REENTER) {
 841                 restore_previous_kprobe(kcb);
 842                 goto out;
 843         }
 844         reset_current_kprobe();
 845 out:
 846         preempt_enable_no_resched();
 847
 848         /*
 849          * if somebody else is singlestepping across a probe point, flags
 850          * will have TF set, in which case, continue the remaining processing
 851          * of do_debug, as if this is not a probe hit.
 852          */
 853         if (regs->flags & X86_EFLAGS_TF)
 854                 return 0;
 855
 856         return 1;
 857 }
 858
 859 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 860 {
 861         struct kprobe *cur = kprobe_running();
 862         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 863
 864         switch (kcb->kprobe_status) {
 865         case KPROBE_HIT_SS:
 866         case KPROBE_REENTER:
 867                 /*
 868                  * We are here because the instruction being single
 869                  * stepped caused a page fault. We reset the current
 870                  * kprobe and the ip points back to the probe address
 871                  * and allow the page fault handler to continue as a
 872                  * normal page fault.
 873                  */
 874                 regs->ip = (unsigned long)cur->addr;
 875                 regs->flags |= kcb->kprobe_old_flags;
 876                 if (kcb->kprobe_status == KPROBE_REENTER)
 877                         restore_previous_kprobe(kcb);
 878                 else
 879                         reset_current_kprobe();
 880                 preempt_enable_no_resched();
 881                 break;
 882         case KPROBE_HIT_ACTIVE:
 883         case KPROBE_HIT_SSDONE:
 884                 /*
 885                  * We increment the nmissed count for accounting,
 886                  * we can also use npre/npostfault count for accounting
 887                  * these specific fault cases.
 888                  */
 889                 kprobes_inc_nmissed_count(cur);
 890
 891                 /*
 892                  * We come here because instructions in the pre/post
 893                  * handler caused the page_fault, this could happen
 894                  * if handler tries to access user space by
 895                  * copy_from_user(), get_user() etc. Let the
 896                  * user-specified handler try to fix it first.
 897                  */
 898                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 899                         return 1;
 900
 901                 /*
 902                  * In case the user-specified fault handler returned
 903                  * zero, try to fix up.
 904                  */
 905                 if (fixup_exception(regs))
 906                         return 1;
 907
 908                 /*
 909                  * fixup routine could not handle it,
 910                  * Let do_page_fault() fix it.
 911                  */
 912                 break;
 913         default:
 914                 break;
 915         }
 916         return 0;
 917 }
 918
 919 /*
 920  * Wrapper routine for handling exceptions.
 921  */
 922 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 923                                        unsigned long val, void *data)
 924 {
 925         struct die_args *args = data;
 926         int ret = NOTIFY_DONE;
 927
 928         if (args->regs && user_mode_vm(args->regs))
 929                 return ret;
 930
 931         switch (val) {
 932         case DIE_INT3:
 933                 if (kprobe_handler(args->regs))
 934                         ret = NOTIFY_STOP;
 935                 break;
 936         case DIE_DEBUG:
 937                 if (post_kprobe_handler(args->regs)) {
 938                         /*
 939                          * Reset the BS bit in dr6 (pointed by args->err) to
 940                          * denote completion of processing
 941                          */
 942                         (*(unsigned long *)ERR_PTR(args->err)) &= ~DR_STEP;
 943                         ret = NOTIFY_STOP;
 944                 }
 945                 break;
 946         case DIE_GPF:
 947                 /*
 948                  * To be potentially processing a kprobe fault and to
 949                  * trust the result from kprobe_running(), we have
 950                  * be non-preemptible.
 951                  */
 952                 if (!preemptible() && kprobe_running() &&
 953                     kprobe_fault_handler(args->regs, args->trapnr))
 954                         ret = NOTIFY_STOP;
 955                 break;
 956         default:
 957                 break;
 958         }
 959         return ret;
 960 }
 961
 962 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 963 {
 964         struct jprobe *jp = container_of(p, struct jprobe, kp);
 965         unsigned long addr;
 966         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 967
 968         kcb->jprobe_saved_regs = *regs;
 969         kcb->jprobe_saved_sp = stack_addr(regs);
 970         addr = (unsigned long)(kcb->jprobe_saved_sp);
 971
 972         /*
 973          * As Linus pointed out, gcc assumes that the callee
 974          * owns the argument space and could overwrite it, e.g.
 975          * tailcall optimization. So, to be absolutely safe
 976          * we also save and restore enough stack bytes to cover
 977          * the argument area.
 978          */
 979         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
 980                MIN_STACK_SIZE(addr));
 981         regs->flags &= ~X86_EFLAGS_IF;
 982         trace_hardirqs_off();
 983         regs->ip = (unsigned long)(jp->entry);
 984         return 1;
 985 }
 986
 987 void __kprobes jprobe_return(void)
 988 {
 989         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 990
 991         asm volatile (
 992 #ifdef CONFIG_X86_64
 993                         "       xchg   %%rbx,%%rsp      \n"
 994 #else
 995                         "       xchgl   %%ebx,%%esp     \n"
 996 #endif
 997                         "       int3                    \n"
 998                         "       .globl jprobe_return_end\n"
 999                         "       jprobe_return_end:      \n"
1000                         "       nop                     \n"::"b"
1001                         (kcb->jprobe_saved_sp):"memory");
1002 }
1003
1004 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1005 {
1006         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1007         u8 *addr = (u8 *) (regs->ip - 1);
1008         struct jprobe *jp = container_of(p, struct jprobe, kp);
1009
1010         if ((addr > (u8 *) jprobe_return) &&
1011             (addr < (u8 *) jprobe_return_end)) {
1012                 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
1013                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
1014                         printk(KERN_ERR
1015                                "current sp %p does not match saved sp %p\n",
1016                                stack_addr(regs), kcb->jprobe_saved_sp);
1017                         printk(KERN_ERR "Saved registers for jprobe %p\n", jp);
1018                         show_registers(saved_regs);
1019                         printk(KERN_ERR "Current registers\n");
1020                         show_registers(regs);
1021                         BUG();
1022                 }
1023                 *regs = kcb->jprobe_saved_regs;
1024                 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
1025                        kcb->jprobes_stack,
1026                        MIN_STACK_SIZE(kcb->jprobe_saved_sp));
1027                 preempt_enable_no_resched();
1028                 return 1;
1029         }
1030         return 0;
1031 }
1032
1033 int __init arch_init_kprobes(void)
1034 {
1035         return 0;
1036 }
1037
1038 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
1039 {
1040         return 0;
1041 }