arch/x86/kernel/kprobes_32.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  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  26  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  27  *              <prasanna@in.ibm.com> added function-return probes.
  28  */
  29
  30 #include <linux/kprobes.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/string.h>
  33 #include <linux/slab.h>
  34 #include <linux/preempt.h>
  35 #include <linux/module.h>
  36 #include <linux/kdebug.h>
  37
  38 #include <asm/cacheflush.h>
  39 #include <asm/desc.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/alternative.h>
  43
  44 void jprobe_return_end(void);
  45
  46 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
  47 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
  48
  49 /*
  50  * "&regs->sp" looks wrong, but it's correct for x86_32.  x86_32 CPUs
  51  * don't save the ss and esp registers if the CPU is already in kernel
  52  * mode when it traps.  So for kprobes, regs->sp and regs->ss are not
  53  * the [nonexistent] saved stack pointer and ss register, but rather
  54  * the top 8 bytes of the pre-int3 stack.  So &regs->sp happens to
  55  * point to the top of the pre-int3 stack.
  56  */
  57 #define stack_addr(regs) ((unsigned long *)&regs->sp)
  58
  59 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
  60         (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
  61           (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
  62           (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
  63           (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
  64          << (row % 32))
  65         /*
  66          * Undefined/reserved opcodes, conditional jump, Opcode Extension
  67          * Groups, and some special opcodes can not boost.
  68          */
  69 static const u32 twobyte_is_boostable[256 / 32] = {
  70         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  71         /*      ----------------------------------------------          */
  72         W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
  73         W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
  74         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
  75         W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
  76         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
  77         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
  78         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
  79         W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
  80         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
  81         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
  82         W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
  83         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
  84         W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
  85         W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
  86         W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
  87         W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0)   /* f0 */
  88         /*      -----------------------------------------------         */
  89         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  90 };
  91 static const u32 onebyte_has_modrm[256 / 32] = {
  92         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
  93         /*      -----------------------------------------------         */
  94         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
  95         W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
  96         W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
  97         W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
  98         W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
  99         W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
 100         W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
 101         W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
 102         W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
 103         W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
 104         W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
 105         W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
 106         W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
 107         W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
 108         W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
 109         W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1)   /* f0 */
 110         /*      -----------------------------------------------         */
 111         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 112 };
 113 static const u32 twobyte_has_modrm[256 / 32] = {
 114         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 115         /*      -----------------------------------------------         */
 116         W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
 117         W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
 118         W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
 119         W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
 120         W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
 121         W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
 122         W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
 123         W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
 124         W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
 125         W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
 126         W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
 127         W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
 128         W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
 129         W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
 130         W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
 131         W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0)   /* ff */
 132         /*      -----------------------------------------------         */
 133         /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
 134 };
 135 #undef W
 136
 137 struct kretprobe_blackpoint kretprobe_blacklist[] = {
 138         {"__switch_to", }, /* This function switches only current task, but
 139                               doesn't switch kernel stack.*/
 140         {NULL, NULL}    /* Terminator */
 141 };
 142 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
 143
 144 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
 145 static __always_inline void set_jmp_op(void *from, void *to)
 146 {
 147         struct __arch_jmp_op {
 148                 char op;
 149                 s32 raddr;
 150         } __attribute__((packed)) * jop;
 151         jop = (struct __arch_jmp_op *)from;
 152         jop->raddr = (s32)((long)(to) - ((long)(from) + 5));
 153         jop->op = RELATIVEJUMP_INSTRUCTION;
 154 }
 155
 156 /*
 157  * returns non-zero if opcode is boostable.
 158  */
 159 static __always_inline int can_boost(kprobe_opcode_t *opcodes)
 160 {
 161         kprobe_opcode_t opcode;
 162         kprobe_opcode_t *orig_opcodes = opcodes;
 163
 164 retry:
 165         if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 166                 return 0;
 167         opcode = *(opcodes++);
 168
 169         /* 2nd-byte opcode */
 170         if (opcode == 0x0f) {
 171                 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
 172                         return 0;
 173                 return test_bit(*opcodes,
 174                                 (unsigned long *)twobyte_is_boostable);
 175         }
 176
 177         switch (opcode & 0xf0) {
 178         case 0x60:
 179                 if (0x63 < opcode && opcode < 0x67)
 180                         goto retry; /* prefixes */
 181                 /* can't boost Address-size override and bound */
 182                 return (opcode != 0x62 && opcode != 0x67);
 183         case 0x70:
 184                 return 0; /* can't boost conditional jump */
 185         case 0xc0:
 186                 /* can't boost software-interruptions */
 187                 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
 188         case 0xd0:
 189                 /* can boost AA* and XLAT */
 190                 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
 191         case 0xe0:
 192                 /* can boost in/out and absolute jmps */
 193                 return ((opcode & 0x04) || opcode == 0xea);
 194         case 0xf0:
 195                 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
 196                         goto retry; /* lock/rep(ne) prefix */
 197                 /* clear and set flags are boostable */
 198                 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
 199         default:
 200                 /* segment override prefixes are boostable */
 201                 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
 202                         goto retry; /* prefixes */
 203                 /* CS override prefix and call are not boostable */
 204                 return (opcode != 0x2e && opcode != 0x9a);
 205         }
 206 }
 207
 208 /*
 209  * returns non-zero if opcode modifies the interrupt flag.
 210  */
 211 static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
 212 {
 213         switch (*insn) {
 214         case 0xfa:              /* cli */
 215         case 0xfb:              /* sti */
 216         case 0xcf:              /* iret/iretd */
 217         case 0x9d:              /* popf/popfd */
 218                 return 1;
 219         }
 220         return 0;
 221 }
 222
 223 static void __kprobes arch_copy_kprobe(struct kprobe *p)
 224 {
 225         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 226         if (can_boost(p->addr))
 227                 p->ainsn.boostable = 0;
 228         else
 229                 p->ainsn.boostable = -1;
 230
 231         p->opcode = *p->addr;
 232 }
 233
 234 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 235 {
 236         /* insn: must be on special executable page on x86. */
 237         p->ainsn.insn = get_insn_slot();
 238         if (!p->ainsn.insn)
 239                 return -ENOMEM;
 240         arch_copy_kprobe(p);
 241         return 0;
 242 }
 243
 244 void __kprobes arch_arm_kprobe(struct kprobe *p)
 245 {
 246         text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1);
 247 }
 248
 249 void __kprobes arch_disarm_kprobe(struct kprobe *p)
 250 {
 251         text_poke(p->addr, &p->opcode, 1);
 252 }
 253
 254 void __kprobes arch_remove_kprobe(struct kprobe *p)
 255 {
 256         mutex_lock(&kprobe_mutex);
 257         free_insn_slot(p->ainsn.insn, (p->ainsn.boostable == 1));
 258         mutex_unlock(&kprobe_mutex);
 259 }
 260
 261 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
 262 {
 263         kcb->prev_kprobe.kp = kprobe_running();
 264         kcb->prev_kprobe.status = kcb->kprobe_status;
 265         kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags;
 266         kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags;
 267 }
 268
 269 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
 270 {
 271         __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
 272         kcb->kprobe_status = kcb->prev_kprobe.status;
 273         kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags;
 274         kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags;
 275 }
 276
 277 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 278                                 struct kprobe_ctlblk *kcb)
 279 {
 280         __get_cpu_var(current_kprobe) = p;
 281         kcb->kprobe_saved_flags = kcb->kprobe_old_flags
 282                 = (regs->flags & (TF_MASK | IF_MASK));
 283         if (is_IF_modifier(p->ainsn.insn))
 284                 kcb->kprobe_saved_flags &= ~IF_MASK;
 285 }
 286
 287 static __always_inline void clear_btf(void)
 288 {
 289         if (test_thread_flag(TIF_DEBUGCTLMSR))
 290                 wrmsr(MSR_IA32_DEBUGCTLMSR, 0, 0);
 291 }
 292
 293 static __always_inline void restore_btf(void)
 294 {
 295         if (test_thread_flag(TIF_DEBUGCTLMSR))
 296                 wrmsr(MSR_IA32_DEBUGCTLMSR, current->thread.debugctlmsr, 0);
 297 }
 298
 299 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 300 {
 301         clear_btf();
 302         regs->flags |= TF_MASK;
 303         regs->flags &= ~IF_MASK;
 304         /*single step inline if the instruction is an int3*/
 305         if (p->opcode == BREAKPOINT_INSTRUCTION)
 306                 regs->ip = (unsigned long)p->addr;
 307         else
 308                 regs->ip = (unsigned long)p->ainsn.insn;
 309 }
 310
 311 /* Called with kretprobe_lock held */
 312 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
 313                                       struct pt_regs *regs)
 314 {
 315         unsigned long *sara = stack_addr(regs);
 316
 317         ri->ret_addr = (kprobe_opcode_t *) *sara;
 318
 319         /* Replace the return addr with trampoline addr */
 320         *sara = (unsigned long) &kretprobe_trampoline;
 321 }
 322
 323 /*
 324  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 325  * remain disabled thorough out this function.
 326  */
 327 static int __kprobes kprobe_handler(struct pt_regs *regs)
 328 {
 329         struct kprobe *p;
 330         int ret = 0;
 331         kprobe_opcode_t *addr;
 332         struct kprobe_ctlblk *kcb;
 333
 334         addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t));
 335
 336         /*
 337          * We don't want to be preempted for the entire
 338          * duration of kprobe processing
 339          */
 340         preempt_disable();
 341         kcb = get_kprobe_ctlblk();
 342
 343         /* Check we're not actually recursing */
 344         if (kprobe_running()) {
 345                 p = get_kprobe(addr);
 346                 if (p) {
 347                         if (kcb->kprobe_status == KPROBE_HIT_SS &&
 348                                 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
 349                                 regs->flags &= ~TF_MASK;
 350                                 regs->flags |= kcb->kprobe_saved_flags;
 351                                 goto no_kprobe;
 352                         }
 353                         /* We have reentered the kprobe_handler(), since
 354                          * another probe was hit while within the handler.
 355                          * We here save the original kprobes variables and
 356                          * just single step on the instruction of the new probe
 357                          * without calling any user handlers.
 358                          */
 359                         save_previous_kprobe(kcb);
 360                         set_current_kprobe(p, regs, kcb);
 361                         kprobes_inc_nmissed_count(p);
 362                         prepare_singlestep(p, regs);
 363                         kcb->kprobe_status = KPROBE_REENTER;
 364                         return 1;
 365                 } else {
 366                         if (*addr != BREAKPOINT_INSTRUCTION) {
 367                         /* The breakpoint instruction was removed by
 368                          * another cpu right after we hit, no further
 369                          * handling of this interrupt is appropriate
 370                          */
 371                                 regs->ip = (unsigned long)addr;
 372                                 ret = 1;
 373                                 goto no_kprobe;
 374                         }
 375                         p = __get_cpu_var(current_kprobe);
 376                         if (p->break_handler && p->break_handler(p, regs)) {
 377                                 goto ss_probe;
 378                         }
 379                 }
 380                 goto no_kprobe;
 381         }
 382
 383         p = get_kprobe(addr);
 384         if (!p) {
 385                 if (*addr != BREAKPOINT_INSTRUCTION) {
 386                         /*
 387                          * The breakpoint instruction was removed right
 388                          * after we hit it.  Another cpu has removed
 389                          * either a probepoint or a debugger breakpoint
 390                          * at this address.  In either case, no further
 391                          * handling of this interrupt is appropriate.
 392                          * Back up over the (now missing) int3 and run
 393                          * the original instruction.
 394                          */
 395                         regs->ip = (unsigned long)addr;
 396                         ret = 1;
 397                 }
 398                 /* Not one of ours: let kernel handle it */
 399                 goto no_kprobe;
 400         }
 401
 402         set_current_kprobe(p, regs, kcb);
 403         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 404
 405         if (p->pre_handler && p->pre_handler(p, regs))
 406                 /* handler has already set things up, so skip ss setup */
 407                 return 1;
 408
 409 ss_probe:
 410 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
 411         if (p->ainsn.boostable == 1 && !p->post_handler) {
 412                 /* Boost up -- we can execute copied instructions directly */
 413                 reset_current_kprobe();
 414                 regs->ip = (unsigned long)p->ainsn.insn;
 415                 preempt_enable_no_resched();
 416                 return 1;
 417         }
 418 #endif
 419         prepare_singlestep(p, regs);
 420         kcb->kprobe_status = KPROBE_HIT_SS;
 421         return 1;
 422
 423 no_kprobe:
 424         preempt_enable_no_resched();
 425         return ret;
 426 }
 427
 428 /*
 429  * When a retprobed function returns, this code saves registers and
 430  * calls trampoline_handler() runs, which calls the kretprobe's handler.
 431  */
 432  void __kprobes kretprobe_trampoline_holder(void)
 433  {
 434         asm volatile ( ".global kretprobe_trampoline\n"
 435                         "kretprobe_trampoline: \n"
 436                         "       pushf\n"
 437                         /*
 438                          * Skip cs, ip, orig_ax.
 439                          * trampoline_handler() will plug in these values
 440                          */
 441                         "       subl $12, %esp\n"
 442                         "       pushl %fs\n"
 443                         "       pushl %ds\n"
 444                         "       pushl %es\n"
 445                         "       pushl %eax\n"
 446                         "       pushl %ebp\n"
 447                         "       pushl %edi\n"
 448                         "       pushl %esi\n"
 449                         "       pushl %edx\n"
 450                         "       pushl %ecx\n"
 451                         "       pushl %ebx\n"
 452                         "       movl %esp, %eax\n"
 453                         "       call trampoline_handler\n"
 454                         /* Move flags to cs */
 455                         "       movl 52(%esp), %edx\n"
 456                         "       movl %edx, 48(%esp)\n"
 457                         /* Replace saved flags with true return address. */
 458                         "       movl %eax, 52(%esp)\n"
 459                         "       popl %ebx\n"
 460                         "       popl %ecx\n"
 461                         "       popl %edx\n"
 462                         "       popl %esi\n"
 463                         "       popl %edi\n"
 464                         "       popl %ebp\n"
 465                         "       popl %eax\n"
 466                         /* Skip ip, orig_ax, es, ds, fs */
 467                         "       addl $20, %esp\n"
 468                         "       popf\n"
 469                         "       ret\n");
 470  }
 471
 472 /*
 473  * Called from kretprobe_trampoline
 474  */
 475 void * __kprobes trampoline_handler(struct pt_regs *regs)
 476 {
 477         struct kretprobe_instance *ri = NULL;
 478         struct hlist_head *head, empty_rp;
 479         struct hlist_node *node, *tmp;
 480         unsigned long flags, orig_ret_address = 0;
 481         unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
 482
 483         INIT_HLIST_HEAD(&empty_rp);
 484         spin_lock_irqsave(&kretprobe_lock, flags);
 485         head = kretprobe_inst_table_head(current);
 486         /* fixup registers */
 487         regs->cs = __KERNEL_CS | get_kernel_rpl();
 488         regs->ip = trampoline_address;
 489         regs->orig_ax = ~0UL;
 490
 491         /*
 492          * It is possible to have multiple instances associated with a given
 493          * task either because multiple functions in the call path have
 494          * return probes installed on them, and/or more then one
 495          * return probe was registered for a target function.
 496          *
 497          * We can handle this because:
 498          *     - instances are always pushed into the head of the list
 499          *     - when multiple return probes are registered for the same
 500          *       function, the (chronologically) first instance's ret_addr
 501          *       will be the real return address, and all the rest will
 502          *       point to kretprobe_trampoline.
 503          */
 504         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 505                 if (ri->task != current)
 506                         /* another task is sharing our hash bucket */
 507                         continue;
 508
 509                 if (ri->rp && ri->rp->handler) {
 510                         __get_cpu_var(current_kprobe) = &ri->rp->kp;
 511                         get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
 512                         ri->rp->handler(ri, regs);
 513                         __get_cpu_var(current_kprobe) = NULL;
 514                 }
 515
 516                 orig_ret_address = (unsigned long)ri->ret_addr;
 517                 recycle_rp_inst(ri, &empty_rp);
 518
 519                 if (orig_ret_address != trampoline_address)
 520                         /*
 521                          * This is the real return address. Any other
 522                          * instances associated with this task are for
 523                          * other calls deeper on the call stack
 524                          */
 525                         break;
 526         }
 527
 528         kretprobe_assert(ri, orig_ret_address, trampoline_address);
 529
 530         spin_unlock_irqrestore(&kretprobe_lock, flags);
 531
 532         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 533                 hlist_del(&ri->hlist);
 534                 kfree(ri);
 535         }
 536         return (void *)orig_ret_address;
 537 }
 538
 539 /*
 540  * Called after single-stepping.  p->addr is the address of the
 541  * instruction whose first byte has been replaced by the "int 3"
 542  * instruction.  To avoid the SMP problems that can occur when we
 543  * temporarily put back the original opcode to single-step, we
 544  * single-stepped a copy of the instruction.  The address of this
 545  * copy is p->ainsn.insn.
 546  *
 547  * This function prepares to return from the post-single-step
 548  * interrupt.  We have to fix up the stack as follows:
 549  *
 550  * 0) Except in the case of absolute or indirect jump or call instructions,
 551  * the new ip is relative to the copied instruction.  We need to make
 552  * it relative to the original instruction.
 553  *
 554  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 555  * flags are set in the just-pushed flags, and may need to be cleared.
 556  *
 557  * 2) If the single-stepped instruction was a call, the return address
 558  * that is atop the stack is the address following the copied instruction.
 559  * We need to make it the address following the original instruction.
 560  *
 561  * If this is the first time we've single-stepped the instruction at
 562  * this probepoint, and the instruction is boostable, boost it: add a
 563  * jump instruction after the copied instruction, that jumps to the next
 564  * instruction after the probepoint.
 565  */
 566 static void __kprobes resume_execution(struct kprobe *p,
 567                 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
 568 {
 569         unsigned long *tos = stack_addr(regs);
 570         unsigned long copy_ip = (unsigned long)p->ainsn.insn;
 571         unsigned long orig_ip = (unsigned long)p->addr;
 572         kprobe_opcode_t *insn = p->ainsn.insn;
 573
 574         regs->flags &= ~TF_MASK;
 575         switch (*insn) {
 576         case 0x9c:      /* pushfl */
 577                 *tos &= ~(TF_MASK | IF_MASK);
 578                 *tos |= kcb->kprobe_old_flags;
 579                 break;
 580         case 0xc2:      /* iret/ret/lret */
 581         case 0xc3:
 582         case 0xca:
 583         case 0xcb:
 584         case 0xcf:
 585         case 0xea:      /* jmp absolute -- ip is correct */
 586                 /* ip is already adjusted, no more changes required */
 587                 p->ainsn.boostable = 1;
 588                 goto no_change;
 589         case 0xe8:      /* call relative - Fix return addr */
 590                 *tos = orig_ip + (*tos - copy_ip);
 591                 break;
 592         case 0x9a:      /* call absolute -- same as call absolute, indirect */
 593                 *tos = orig_ip + (*tos - copy_ip);
 594                 goto no_change;
 595         case 0xff:
 596                 if ((insn[1] & 0x30) == 0x10) {
 597                         /*
 598                          * call absolute, indirect
 599                          * Fix return addr; ip is correct.
 600                          * But this is not boostable
 601                          */
 602                         *tos = orig_ip + (*tos - copy_ip);
 603                         goto no_change;
 604                 } else if (((insn[1] & 0x31) == 0x20) ||
 605                            ((insn[1] & 0x31) == 0x21)) {
 606                         /*
 607                          * jmp near and far, absolute indirect
 608                          * ip is correct. And this is boostable
 609                          */
 610                         p->ainsn.boostable = 1;
 611                         goto no_change;
 612                 }
 613         default:
 614                 break;
 615         }
 616
 617         if (p->ainsn.boostable == 0) {
 618                 if ((regs->ip > copy_ip) &&
 619                     (regs->ip - copy_ip) + 5 < MAX_INSN_SIZE) {
 620                         /*
 621                          * These instructions can be executed directly if it
 622                          * jumps back to correct address.
 623                          */
 624                         set_jmp_op((void *)regs->ip,
 625                                    (void *)orig_ip + (regs->ip - copy_ip));
 626                         p->ainsn.boostable = 1;
 627                 } else {
 628                         p->ainsn.boostable = -1;
 629                 }
 630         }
 631
 632         regs->ip += orig_ip - copy_ip;
 633
 634 no_change:
 635         restore_btf();
 636
 637         return;
 638 }
 639
 640 /*
 641  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 642  * remain disabled thoroughout this function.
 643  */
 644 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
 645 {
 646         struct kprobe *cur = kprobe_running();
 647         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 648
 649         if (!cur)
 650                 return 0;
 651
 652         if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
 653                 kcb->kprobe_status = KPROBE_HIT_SSDONE;
 654                 cur->post_handler(cur, regs, 0);
 655         }
 656
 657         resume_execution(cur, regs, kcb);
 658         regs->flags |= kcb->kprobe_saved_flags;
 659         trace_hardirqs_fixup_flags(regs->flags);
 660
 661         /* Restore back the original saved kprobes variables and continue. */
 662         if (kcb->kprobe_status == KPROBE_REENTER) {
 663                 restore_previous_kprobe(kcb);
 664                 goto out;
 665         }
 666         reset_current_kprobe();
 667 out:
 668         preempt_enable_no_resched();
 669
 670         /*
 671          * if somebody else is singlestepping across a probe point, flags
 672          * will have TF set, in which case, continue the remaining processing
 673          * of do_debug, as if this is not a probe hit.
 674          */
 675         if (regs->flags & TF_MASK)
 676                 return 0;
 677
 678         return 1;
 679 }
 680
 681 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 682 {
 683         struct kprobe *cur = kprobe_running();
 684         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 685
 686         switch(kcb->kprobe_status) {
 687         case KPROBE_HIT_SS:
 688         case KPROBE_REENTER:
 689                 /*
 690                  * We are here because the instruction being single
 691                  * stepped caused a page fault. We reset the current
 692                  * kprobe and the ip points back to the probe address
 693                  * and allow the page fault handler to continue as a
 694                  * normal page fault.
 695                  */
 696                 regs->ip = (unsigned long)cur->addr;
 697                 regs->flags |= kcb->kprobe_old_flags;
 698                 if (kcb->kprobe_status == KPROBE_REENTER)
 699                         restore_previous_kprobe(kcb);
 700                 else
 701                         reset_current_kprobe();
 702                 preempt_enable_no_resched();
 703                 break;
 704         case KPROBE_HIT_ACTIVE:
 705         case KPROBE_HIT_SSDONE:
 706                 /*
 707                  * We increment the nmissed count for accounting,
 708                  * we can also use npre/npostfault count for accounting
 709                  * these specific fault cases.
 710                  */
 711                 kprobes_inc_nmissed_count(cur);
 712
 713                 /*
 714                  * We come here because instructions in the pre/post
 715                  * handler caused the page_fault, this could happen
 716                  * if handler tries to access user space by
 717                  * copy_from_user(), get_user() etc. Let the
 718                  * user-specified handler try to fix it first.
 719                  */
 720                 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 721                         return 1;
 722
 723                 /*
 724                  * In case the user-specified fault handler returned
 725                  * zero, try to fix up.
 726                  */
 727                 if (fixup_exception(regs))
 728                         return 1;
 729
 730                 /*
 731                  * fixup routine could not handle it,
 732                  * Let do_page_fault() fix it.
 733                  */
 734                 break;
 735         default:
 736                 break;
 737         }
 738         return 0;
 739 }
 740
 741 /*
 742  * Wrapper routine for handling exceptions.
 743  */
 744 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 745                                        unsigned long val, void *data)
 746 {
 747         struct die_args *args = (struct die_args *)data;
 748         int ret = NOTIFY_DONE;
 749
 750         if (args->regs && user_mode_vm(args->regs))
 751                 return ret;
 752
 753         switch (val) {
 754         case DIE_INT3:
 755                 if (kprobe_handler(args->regs))
 756                         ret = NOTIFY_STOP;
 757                 break;
 758         case DIE_DEBUG:
 759                 if (post_kprobe_handler(args->regs))
 760                         ret = NOTIFY_STOP;
 761                 break;
 762         case DIE_GPF:
 763                 /* kprobe_running() needs smp_processor_id() */
 764                 preempt_disable();
 765                 if (kprobe_running() &&
 766                     kprobe_fault_handler(args->regs, args->trapnr))
 767                         ret = NOTIFY_STOP;
 768                 preempt_enable();
 769                 break;
 770         default:
 771                 break;
 772         }
 773         return ret;
 774 }
 775
 776 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 777 {
 778         struct jprobe *jp = container_of(p, struct jprobe, kp);
 779         unsigned long addr;
 780         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 781
 782         kcb->jprobe_saved_regs = *regs;
 783         kcb->jprobe_saved_sp = stack_addr(regs);
 784         addr = (unsigned long)(kcb->jprobe_saved_sp);
 785
 786         /*
 787          * As Linus pointed out, gcc assumes that the callee
 788          * owns the argument space and could overwrite it, e.g.
 789          * tailcall optimization. So, to be absolutely safe
 790          * we also save and restore enough stack bytes to cover
 791          * the argument area.
 792          */
 793         memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
 794                         MIN_STACK_SIZE(addr));
 795         regs->flags &= ~IF_MASK;
 796         trace_hardirqs_off();
 797         regs->ip = (unsigned long)(jp->entry);
 798         return 1;
 799 }
 800
 801 void __kprobes jprobe_return(void)
 802 {
 803         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 804
 805         asm volatile ("       xchgl   %%ebx,%%esp     \n"
 806                       "       int3                      \n"
 807                       "       .globl jprobe_return_end  \n"
 808                       "       jprobe_return_end:        \n"
 809                       "       nop                       \n"::"b"
 810                       (kcb->jprobe_saved_sp):"memory");
 811 }
 812
 813 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 814 {
 815         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 816         u8 *addr = (u8 *) (regs->ip - 1);
 817         struct jprobe *jp = container_of(p, struct jprobe, kp);
 818
 819         if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
 820                 if (stack_addr(regs) != kcb->jprobe_saved_sp) {
 821                         struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
 822                         printk("current sp %p does not match saved sp %p\n",
 823                                stack_addr(regs), kcb->jprobe_saved_sp);
 824                         printk("Saved registers for jprobe %p\n", jp);
 825                         show_registers(saved_regs);
 826                         printk("Current registers\n");
 827                         show_registers(regs);
 828                         BUG();
 829                 }
 830                 *regs = kcb->jprobe_saved_regs;
 831                 memcpy((kprobe_opcode_t *)(kcb->jprobe_saved_sp),
 832                        kcb->jprobes_stack,
 833                        MIN_STACK_SIZE(kcb->jprobe_saved_sp));
 834                 preempt_enable_no_resched();
 835                 return 1;
 836         }
 837         return 0;
 838 }
 839
 840 int __init arch_init_kprobes(void)
 841 {
 842         return 0;
 843 }
 844
 845 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
 846 {
 847         return 0;
 848 }