git.oblomov.eu Git - linux-2.6/blob - arch/x86/kernel/ptrace.c

   1 /* By Ross Biro 1/23/92 */
   2 /*
   3  * Pentium III FXSR, SSE support
   4  *      Gareth Hughes <gareth@valinux.com>, May 2000
   5  *
   6  * BTS tracing
   7  *      Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/mm.h>
  13 #include <linux/smp.h>
  14 #include <linux/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/regset.h>
  17 #include <linux/tracehook.h>
  18 #include <linux/user.h>
  19 #include <linux/elf.h>
  20 #include <linux/security.h>
  21 #include <linux/audit.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/signal.h>
  24 #include <linux/ftrace.h>
  25
  26 #include <asm/uaccess.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/system.h>
  29 #include <asm/processor.h>
  30 #include <asm/i387.h>
  31 #include <asm/debugreg.h>
  32 #include <asm/ldt.h>
  33 #include <asm/desc.h>
  34 #include <asm/prctl.h>
  35 #include <asm/proto.h>
  36 #include <asm/ds.h>
  37
  38 #include "tls.h"
  39
  40 enum x86_regset {
  41         REGSET_GENERAL,
  42         REGSET_FP,
  43         REGSET_XFP,
  44         REGSET_IOPERM64 = REGSET_XFP,
  45         REGSET_TLS,
  46         REGSET_IOPERM32,
  47 };
  48
  49 /*
  50  * does not yet catch signals sent when the child dies.
  51  * in exit.c or in signal.c.
  52  */
  53
  54 /*
  55  * Determines which flags the user has access to [1 = access, 0 = no access].
  56  */
  57 #define FLAG_MASK_32            ((unsigned long)                        \
  58                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  59                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  60                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  61                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  62                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  63
  64 /*
  65  * Determines whether a value may be installed in a segment register.
  66  */
  67 static inline bool invalid_selector(u16 value)
  68 {
  69         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  70 }
  71
  72 #ifdef CONFIG_X86_32
  73
  74 #define FLAG_MASK               FLAG_MASK_32
  75
  76 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  77 {
  78         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  79         return &regs->bx + (regno >> 2);
  80 }
  81
  82 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  83 {
  84         /*
  85          * Returning the value truncates it to 16 bits.
  86          */
  87         unsigned int retval;
  88         if (offset != offsetof(struct user_regs_struct, gs))
  89                 retval = *pt_regs_access(task_pt_regs(task), offset);
  90         else {
  91                 if (task == current)
  92                         retval = get_user_gs(task_pt_regs(task));
  93                 else
  94                         retval = task_user_gs(task);
  95         }
  96         return retval;
  97 }
  98
  99 static int set_segment_reg(struct task_struct *task,
 100                            unsigned long offset, u16 value)
 101 {
 102         /*
 103          * The value argument was already truncated to 16 bits.
 104          */
 105         if (invalid_selector(value))
 106                 return -EIO;
 107
 108         /*
 109          * For %cs and %ss we cannot permit a null selector.
 110          * We can permit a bogus selector as long as it has USER_RPL.
 111          * Null selectors are fine for other segment registers, but
 112          * we will never get back to user mode with invalid %cs or %ss
 113          * and will take the trap in iret instead.  Much code relies
 114          * on user_mode() to distinguish a user trap frame (which can
 115          * safely use invalid selectors) from a kernel trap frame.
 116          */
 117         switch (offset) {
 118         case offsetof(struct user_regs_struct, cs):
 119         case offsetof(struct user_regs_struct, ss):
 120                 if (unlikely(value == 0))
 121                         return -EIO;
 122
 123         default:
 124                 *pt_regs_access(task_pt_regs(task), offset) = value;
 125                 break;
 126
 127         case offsetof(struct user_regs_struct, gs):
 128                 if (task == current)
 129                         set_user_gs(task_pt_regs(task), value);
 130                 else
 131                         task_user_gs(task) = value;
 132         }
 133
 134         return 0;
 135 }
 136
 137 static unsigned long debugreg_addr_limit(struct task_struct *task)
 138 {
 139         return TASK_SIZE - 3;
 140 }
 141
 142 #else  /* CONFIG_X86_64 */
 143
 144 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 145
 146 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 147 {
 148         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 149         return &regs->r15 + (offset / sizeof(regs->r15));
 150 }
 151
 152 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 153 {
 154         /*
 155          * Returning the value truncates it to 16 bits.
 156          */
 157         unsigned int seg;
 158
 159         switch (offset) {
 160         case offsetof(struct user_regs_struct, fs):
 161                 if (task == current) {
 162                         /* Older gas can't assemble movq %?s,%r?? */
 163                         asm("movl %%fs,%0" : "=r" (seg));
 164                         return seg;
 165                 }
 166                 return task->thread.fsindex;
 167         case offsetof(struct user_regs_struct, gs):
 168                 if (task == current) {
 169                         asm("movl %%gs,%0" : "=r" (seg));
 170                         return seg;
 171                 }
 172                 return task->thread.gsindex;
 173         case offsetof(struct user_regs_struct, ds):
 174                 if (task == current) {
 175                         asm("movl %%ds,%0" : "=r" (seg));
 176                         return seg;
 177                 }
 178                 return task->thread.ds;
 179         case offsetof(struct user_regs_struct, es):
 180                 if (task == current) {
 181                         asm("movl %%es,%0" : "=r" (seg));
 182                         return seg;
 183                 }
 184                 return task->thread.es;
 185
 186         case offsetof(struct user_regs_struct, cs):
 187         case offsetof(struct user_regs_struct, ss):
 188                 break;
 189         }
 190         return *pt_regs_access(task_pt_regs(task), offset);
 191 }
 192
 193 static int set_segment_reg(struct task_struct *task,
 194                            unsigned long offset, u16 value)
 195 {
 196         /*
 197          * The value argument was already truncated to 16 bits.
 198          */
 199         if (invalid_selector(value))
 200                 return -EIO;
 201
 202         switch (offset) {
 203         case offsetof(struct user_regs_struct,fs):
 204                 /*
 205                  * If this is setting fs as for normal 64-bit use but
 206                  * setting fs_base has implicitly changed it, leave it.
 207                  */
 208                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 209                      task->thread.fs != 0) ||
 210                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 211                      task->thread.fs == 0))
 212                         break;
 213                 task->thread.fsindex = value;
 214                 if (task == current)
 215                         loadsegment(fs, task->thread.fsindex);
 216                 break;
 217         case offsetof(struct user_regs_struct,gs):
 218                 /*
 219                  * If this is setting gs as for normal 64-bit use but
 220                  * setting gs_base has implicitly changed it, leave it.
 221                  */
 222                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 223                      task->thread.gs != 0) ||
 224                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 225                      task->thread.gs == 0))
 226                         break;
 227                 task->thread.gsindex = value;
 228                 if (task == current)
 229                         load_gs_index(task->thread.gsindex);
 230                 break;
 231         case offsetof(struct user_regs_struct,ds):
 232                 task->thread.ds = value;
 233                 if (task == current)
 234                         loadsegment(ds, task->thread.ds);
 235                 break;
 236         case offsetof(struct user_regs_struct,es):
 237                 task->thread.es = value;
 238                 if (task == current)
 239                         loadsegment(es, task->thread.es);
 240                 break;
 241
 242                 /*
 243                  * Can't actually change these in 64-bit mode.
 244                  */
 245         case offsetof(struct user_regs_struct,cs):
 246                 if (unlikely(value == 0))
 247                         return -EIO;
 248 #ifdef CONFIG_IA32_EMULATION
 249                 if (test_tsk_thread_flag(task, TIF_IA32))
 250                         task_pt_regs(task)->cs = value;
 251 #endif
 252                 break;
 253         case offsetof(struct user_regs_struct,ss):
 254                 if (unlikely(value == 0))
 255                         return -EIO;
 256 #ifdef CONFIG_IA32_EMULATION
 257                 if (test_tsk_thread_flag(task, TIF_IA32))
 258                         task_pt_regs(task)->ss = value;
 259 #endif
 260                 break;
 261         }
 262
 263         return 0;
 264 }
 265
 266 static unsigned long debugreg_addr_limit(struct task_struct *task)
 267 {
 268 #ifdef CONFIG_IA32_EMULATION
 269         if (test_tsk_thread_flag(task, TIF_IA32))
 270                 return IA32_PAGE_OFFSET - 3;
 271 #endif
 272         return TASK_SIZE_MAX - 7;
 273 }
 274
 275 #endif  /* CONFIG_X86_32 */
 276
 277 static unsigned long get_flags(struct task_struct *task)
 278 {
 279         unsigned long retval = task_pt_regs(task)->flags;
 280
 281         /*
 282          * If the debugger set TF, hide it from the readout.
 283          */
 284         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 285                 retval &= ~X86_EFLAGS_TF;
 286
 287         return retval;
 288 }
 289
 290 static int set_flags(struct task_struct *task, unsigned long value)
 291 {
 292         struct pt_regs *regs = task_pt_regs(task);
 293
 294         /*
 295          * If the user value contains TF, mark that
 296          * it was not "us" (the debugger) that set it.
 297          * If not, make sure it stays set if we had.
 298          */
 299         if (value & X86_EFLAGS_TF)
 300                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 301         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 302                 value |= X86_EFLAGS_TF;
 303
 304         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 305
 306         return 0;
 307 }
 308
 309 static int putreg(struct task_struct *child,
 310                   unsigned long offset, unsigned long value)
 311 {
 312         switch (offset) {
 313         case offsetof(struct user_regs_struct, cs):
 314         case offsetof(struct user_regs_struct, ds):
 315         case offsetof(struct user_regs_struct, es):
 316         case offsetof(struct user_regs_struct, fs):
 317         case offsetof(struct user_regs_struct, gs):
 318         case offsetof(struct user_regs_struct, ss):
 319                 return set_segment_reg(child, offset, value);
 320
 321         case offsetof(struct user_regs_struct, flags):
 322                 return set_flags(child, value);
 323
 324 #ifdef CONFIG_X86_64
 325         /*
 326          * Orig_ax is really just a flag with small positive and
 327          * negative values, so make sure to always sign-extend it
 328          * from 32 bits so that it works correctly regardless of
 329          * whether we come from a 32-bit environment or not.
 330          */
 331         case offsetof(struct user_regs_struct, orig_ax):
 332                 value = (long) (s32) value;
 333                 break;
 334
 335         case offsetof(struct user_regs_struct,fs_base):
 336                 if (value >= TASK_SIZE_OF(child))
 337                         return -EIO;
 338                 /*
 339                  * When changing the segment base, use do_arch_prctl
 340                  * to set either thread.fs or thread.fsindex and the
 341                  * corresponding GDT slot.
 342                  */
 343                 if (child->thread.fs != value)
 344                         return do_arch_prctl(child, ARCH_SET_FS, value);
 345                 return 0;
 346         case offsetof(struct user_regs_struct,gs_base):
 347                 /*
 348                  * Exactly the same here as the %fs handling above.
 349                  */
 350                 if (value >= TASK_SIZE_OF(child))
 351                         return -EIO;
 352                 if (child->thread.gs != value)
 353                         return do_arch_prctl(child, ARCH_SET_GS, value);
 354                 return 0;
 355 #endif
 356         }
 357
 358         *pt_regs_access(task_pt_regs(child), offset) = value;
 359         return 0;
 360 }
 361
 362 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 363 {
 364         switch (offset) {
 365         case offsetof(struct user_regs_struct, cs):
 366         case offsetof(struct user_regs_struct, ds):
 367         case offsetof(struct user_regs_struct, es):
 368         case offsetof(struct user_regs_struct, fs):
 369         case offsetof(struct user_regs_struct, gs):
 370         case offsetof(struct user_regs_struct, ss):
 371                 return get_segment_reg(task, offset);
 372
 373         case offsetof(struct user_regs_struct, flags):
 374                 return get_flags(task);
 375
 376 #ifdef CONFIG_X86_64
 377         case offsetof(struct user_regs_struct, fs_base): {
 378                 /*
 379                  * do_arch_prctl may have used a GDT slot instead of
 380                  * the MSR.  To userland, it appears the same either
 381                  * way, except the %fs segment selector might not be 0.
 382                  */
 383                 unsigned int seg = task->thread.fsindex;
 384                 if (task->thread.fs != 0)
 385                         return task->thread.fs;
 386                 if (task == current)
 387                         asm("movl %%fs,%0" : "=r" (seg));
 388                 if (seg != FS_TLS_SEL)
 389                         return 0;
 390                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 391         }
 392         case offsetof(struct user_regs_struct, gs_base): {
 393                 /*
 394                  * Exactly the same here as the %fs handling above.
 395                  */
 396                 unsigned int seg = task->thread.gsindex;
 397                 if (task->thread.gs != 0)
 398                         return task->thread.gs;
 399                 if (task == current)
 400                         asm("movl %%gs,%0" : "=r" (seg));
 401                 if (seg != GS_TLS_SEL)
 402                         return 0;
 403                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 404         }
 405 #endif
 406         }
 407
 408         return *pt_regs_access(task_pt_regs(task), offset);
 409 }
 410
 411 static int genregs_get(struct task_struct *target,
 412                        const struct user_regset *regset,
 413                        unsigned int pos, unsigned int count,
 414                        void *kbuf, void __user *ubuf)
 415 {
 416         if (kbuf) {
 417                 unsigned long *k = kbuf;
 418                 while (count > 0) {
 419                         *k++ = getreg(target, pos);
 420                         count -= sizeof(*k);
 421                         pos += sizeof(*k);
 422                 }
 423         } else {
 424                 unsigned long __user *u = ubuf;
 425                 while (count > 0) {
 426                         if (__put_user(getreg(target, pos), u++))
 427                                 return -EFAULT;
 428                         count -= sizeof(*u);
 429                         pos += sizeof(*u);
 430                 }
 431         }
 432
 433         return 0;
 434 }
 435
 436 static int genregs_set(struct task_struct *target,
 437                        const struct user_regset *regset,
 438                        unsigned int pos, unsigned int count,
 439                        const void *kbuf, const void __user *ubuf)
 440 {
 441         int ret = 0;
 442         if (kbuf) {
 443                 const unsigned long *k = kbuf;
 444                 while (count > 0 && !ret) {
 445                         ret = putreg(target, pos, *k++);
 446                         count -= sizeof(*k);
 447                         pos += sizeof(*k);
 448                 }
 449         } else {
 450                 const unsigned long  __user *u = ubuf;
 451                 while (count > 0 && !ret) {
 452                         unsigned long word;
 453                         ret = __get_user(word, u++);
 454                         if (ret)
 455                                 break;
 456                         ret = putreg(target, pos, word);
 457                         count -= sizeof(*u);
 458                         pos += sizeof(*u);
 459                 }
 460         }
 461         return ret;
 462 }
 463
 464 /*
 465  * This function is trivial and will be inlined by the compiler.
 466  * Having it separates the implementation details of debug
 467  * registers from the interface details of ptrace.
 468  */
 469 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 470 {
 471         switch (n) {
 472         case 0:         return child->thread.debugreg0;
 473         case 1:         return child->thread.debugreg1;
 474         case 2:         return child->thread.debugreg2;
 475         case 3:         return child->thread.debugreg3;
 476         case 6:         return child->thread.debugreg6;
 477         case 7:         return child->thread.debugreg7;
 478         }
 479         return 0;
 480 }
 481
 482 static int ptrace_set_debugreg(struct task_struct *child,
 483                                int n, unsigned long data)
 484 {
 485         int i;
 486
 487         if (unlikely(n == 4 || n == 5))
 488                 return -EIO;
 489
 490         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 491                 return -EIO;
 492
 493         switch (n) {
 494         case 0:         child->thread.debugreg0 = data; break;
 495         case 1:         child->thread.debugreg1 = data; break;
 496         case 2:         child->thread.debugreg2 = data; break;
 497         case 3:         child->thread.debugreg3 = data; break;
 498
 499         case 6:
 500                 if ((data & ~0xffffffffUL) != 0)
 501                         return -EIO;
 502                 child->thread.debugreg6 = data;
 503                 break;
 504
 505         case 7:
 506                 /*
 507                  * Sanity-check data. Take one half-byte at once with
 508                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 509                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 510                  * 2 and 3 are LENi. Given a list of invalid values,
 511                  * we do mask |= 1 << invalid_value, so that
 512                  * (mask >> check) & 1 is a correct test for invalid
 513                  * values.
 514                  *
 515                  * R/Wi contains the type of the breakpoint /
 516                  * watchpoint, LENi contains the length of the watched
 517                  * data in the watchpoint case.
 518                  *
 519                  * The invalid values are:
 520                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 521                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 522                  *   mask |= 0x4444.
 523                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 524                  *   0x1110.
 525                  *
 526                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 527                  *
 528                  * See the Intel Manual "System Programming Guide",
 529                  * 15.2.4
 530                  *
 531                  * Note that LENi == 0x10 is defined on x86_64 in long
 532                  * mode (i.e. even for 32-bit userspace software, but
 533                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 534                  * See the AMD manual no. 24593 (AMD64 System Programming)
 535                  */
 536 #ifdef CONFIG_X86_32
 537 #define DR7_MASK        0x5f54
 538 #else
 539 #define DR7_MASK        0x5554
 540 #endif
 541                 data &= ~DR_CONTROL_RESERVED;
 542                 for (i = 0; i < 4; i++)
 543                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 544                                 return -EIO;
 545                 child->thread.debugreg7 = data;
 546                 if (data)
 547                         set_tsk_thread_flag(child, TIF_DEBUG);
 548                 else
 549                         clear_tsk_thread_flag(child, TIF_DEBUG);
 550                 break;
 551         }
 552
 553         return 0;
 554 }
 555
 556 /*
 557  * These access the current or another (stopped) task's io permission
 558  * bitmap for debugging or core dump.
 559  */
 560 static int ioperm_active(struct task_struct *target,
 561                          const struct user_regset *regset)
 562 {
 563         return target->thread.io_bitmap_max / regset->size;
 564 }
 565
 566 static int ioperm_get(struct task_struct *target,
 567                       const struct user_regset *regset,
 568                       unsigned int pos, unsigned int count,
 569                       void *kbuf, void __user *ubuf)
 570 {
 571         if (!target->thread.io_bitmap_ptr)
 572                 return -ENXIO;
 573
 574         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 575                                    target->thread.io_bitmap_ptr,
 576                                    0, IO_BITMAP_BYTES);
 577 }
 578
 579 #ifdef CONFIG_X86_PTRACE_BTS
 580 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 581                                   struct bts_struct __user *out)
 582 {
 583         const struct bts_trace *trace;
 584         struct bts_struct bts;
 585         const unsigned char *at;
 586         int error;
 587
 588         trace = ds_read_bts(child->bts);
 589         if (!trace)
 590                 return -EPERM;
 591
 592         at = trace->ds.top - ((index + 1) * trace->ds.size);
 593         if ((void *)at < trace->ds.begin)
 594                 at += (trace->ds.n * trace->ds.size);
 595
 596         if (!trace->read)
 597                 return -EOPNOTSUPP;
 598
 599         error = trace->read(child->bts, at, &bts);
 600         if (error < 0)
 601                 return error;
 602
 603         if (copy_to_user(out, &bts, sizeof(bts)))
 604                 return -EFAULT;
 605
 606         return sizeof(bts);
 607 }
 608
 609 static int ptrace_bts_drain(struct task_struct *child,
 610                             long size,
 611                             struct bts_struct __user *out)
 612 {
 613         const struct bts_trace *trace;
 614         const unsigned char *at;
 615         int error, drained = 0;
 616
 617         trace = ds_read_bts(child->bts);
 618         if (!trace)
 619                 return -EPERM;
 620
 621         if (!trace->read)
 622                 return -EOPNOTSUPP;
 623
 624         if (size < (trace->ds.top - trace->ds.begin))
 625                 return -EIO;
 626
 627         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 628              out++, drained++, at += trace->ds.size) {
 629                 struct bts_struct bts;
 630                 int error;
 631
 632                 error = trace->read(child->bts, at, &bts);
 633                 if (error < 0)
 634                         return error;
 635
 636                 if (copy_to_user(out, &bts, sizeof(bts)))
 637                         return -EFAULT;
 638         }
 639
 640         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 641
 642         error = ds_reset_bts(child->bts);
 643         if (error < 0)
 644                 return error;
 645
 646         return drained;
 647 }
 648
 649 static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
 650 {
 651         child->bts_buffer = alloc_locked_buffer(size);
 652         if (!child->bts_buffer)
 653                 return -ENOMEM;
 654
 655         child->bts_size = size;
 656
 657         return 0;
 658 }
 659
 660 static void ptrace_bts_free_buffer(struct task_struct *child)
 661 {
 662         free_locked_buffer(child->bts_buffer, child->bts_size);
 663         child->bts_buffer = NULL;
 664         child->bts_size = 0;
 665 }
 666
 667 static int ptrace_bts_config(struct task_struct *child,
 668                              long cfg_size,
 669                              const struct ptrace_bts_config __user *ucfg)
 670 {
 671         struct ptrace_bts_config cfg;
 672         unsigned int flags = 0;
 673
 674         if (cfg_size < sizeof(cfg))
 675                 return -EIO;
 676
 677         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 678                 return -EFAULT;
 679
 680         if (child->bts) {
 681                 ds_release_bts(child->bts);
 682                 child->bts = NULL;
 683         }
 684
 685         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 686                 if (!cfg.signal)
 687                         return -EINVAL;
 688
 689                 child->thread.bts_ovfl_signal = cfg.signal;
 690                 return -EOPNOTSUPP;
 691         }
 692
 693         if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
 694             (cfg.size != child->bts_size)) {
 695                 int error;
 696
 697                 ptrace_bts_free_buffer(child);
 698
 699                 error = ptrace_bts_allocate_buffer(child, cfg.size);
 700                 if (error < 0)
 701                         return error;
 702         }
 703
 704         if (cfg.flags & PTRACE_BTS_O_TRACE)
 705                 flags |= BTS_USER;
 706
 707         if (cfg.flags & PTRACE_BTS_O_SCHED)
 708                 flags |= BTS_TIMESTAMPS;
 709
 710         child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
 711                                     /* ovfl = */ NULL, /* th = */ (size_t)-1,
 712                                     flags);
 713         if (IS_ERR(child->bts)) {
 714                 int error = PTR_ERR(child->bts);
 715
 716                 ptrace_bts_free_buffer(child);
 717                 child->bts = NULL;
 718
 719                 return error;
 720         }
 721
 722         return sizeof(cfg);
 723 }
 724
 725 static int ptrace_bts_status(struct task_struct *child,
 726                              long cfg_size,
 727                              struct ptrace_bts_config __user *ucfg)
 728 {
 729         const struct bts_trace *trace;
 730         struct ptrace_bts_config cfg;
 731
 732         if (cfg_size < sizeof(cfg))
 733                 return -EIO;
 734
 735         trace = ds_read_bts(child->bts);
 736         if (!trace)
 737                 return -EPERM;
 738
 739         memset(&cfg, 0, sizeof(cfg));
 740         cfg.size = trace->ds.end - trace->ds.begin;
 741         cfg.signal = child->thread.bts_ovfl_signal;
 742         cfg.bts_size = sizeof(struct bts_struct);
 743
 744         if (cfg.signal)
 745                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 746
 747         if (trace->ds.flags & BTS_USER)
 748                 cfg.flags |= PTRACE_BTS_O_TRACE;
 749
 750         if (trace->ds.flags & BTS_TIMESTAMPS)
 751                 cfg.flags |= PTRACE_BTS_O_SCHED;
 752
 753         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 754                 return -EFAULT;
 755
 756         return sizeof(cfg);
 757 }
 758
 759 static int ptrace_bts_clear(struct task_struct *child)
 760 {
 761         const struct bts_trace *trace;
 762
 763         trace = ds_read_bts(child->bts);
 764         if (!trace)
 765                 return -EPERM;
 766
 767         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 768
 769         return ds_reset_bts(child->bts);
 770 }
 771
 772 static int ptrace_bts_size(struct task_struct *child)
 773 {
 774         const struct bts_trace *trace;
 775
 776         trace = ds_read_bts(child->bts);
 777         if (!trace)
 778                 return -EPERM;
 779
 780         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 781 }
 782
 783 static void ptrace_bts_fork(struct task_struct *tsk)
 784 {
 785         tsk->bts = NULL;
 786         tsk->bts_buffer = NULL;
 787         tsk->bts_size = 0;
 788         tsk->thread.bts_ovfl_signal = 0;
 789 }
 790
 791 static void ptrace_bts_untrace(struct task_struct *child)
 792 {
 793         if (unlikely(child->bts)) {
 794                 ds_release_bts(child->bts);
 795                 child->bts = NULL;
 796
 797                 /* We cannot update total_vm and locked_vm since
 798                    child's mm is already gone. But we can reclaim the
 799                    memory. */
 800                 kfree(child->bts_buffer);
 801                 child->bts_buffer = NULL;
 802                 child->bts_size = 0;
 803         }
 804 }
 805
 806 static void ptrace_bts_detach(struct task_struct *child)
 807 {
 808         /*
 809          * Ptrace_detach() races with ptrace_untrace() in case
 810          * the child dies and is reaped by another thread.
 811          *
 812          * We only do the memory accounting at this point and
 813          * leave the buffer deallocation and the bts tracer
 814          * release to ptrace_bts_untrace() which will be called
 815          * later on with tasklist_lock held.
 816          */
 817         release_locked_buffer(child->bts_buffer, child->bts_size);
 818 }
 819 #else
 820 static inline void ptrace_bts_fork(struct task_struct *tsk) {}
 821 static inline void ptrace_bts_detach(struct task_struct *child) {}
 822 static inline void ptrace_bts_untrace(struct task_struct *child) {}
 823 #endif /* CONFIG_X86_PTRACE_BTS */
 824
 825 void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
 826 {
 827         ptrace_bts_fork(child);
 828 }
 829
 830 void x86_ptrace_untrace(struct task_struct *child)
 831 {
 832         ptrace_bts_untrace(child);
 833 }
 834
 835 /*
 836  * Called by kernel/ptrace.c when detaching..
 837  *
 838  * Make sure the single step bit is not set.
 839  */
 840 void ptrace_disable(struct task_struct *child)
 841 {
 842         user_disable_single_step(child);
 843 #ifdef TIF_SYSCALL_EMU
 844         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 845 #endif
 846         ptrace_bts_detach(child);
 847 }
 848
 849 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 850 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 851 #endif
 852
 853 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 854 {
 855         int ret;
 856         unsigned long __user *datap = (unsigned long __user *)data;
 857
 858         switch (request) {
 859         /* read the word at location addr in the USER area. */
 860         case PTRACE_PEEKUSR: {
 861                 unsigned long tmp;
 862
 863                 ret = -EIO;
 864                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 865                     addr >= sizeof(struct user))
 866                         break;
 867
 868                 tmp = 0;  /* Default return condition */
 869                 if (addr < sizeof(struct user_regs_struct))
 870                         tmp = getreg(child, addr);
 871                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 872                          addr <= offsetof(struct user, u_debugreg[7])) {
 873                         addr -= offsetof(struct user, u_debugreg[0]);
 874                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 875                 }
 876                 ret = put_user(tmp, datap);
 877                 break;
 878         }
 879
 880         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 881                 ret = -EIO;
 882                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 883                     addr >= sizeof(struct user))
 884                         break;
 885
 886                 if (addr < sizeof(struct user_regs_struct))
 887                         ret = putreg(child, addr, data);
 888                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 889                          addr <= offsetof(struct user, u_debugreg[7])) {
 890                         addr -= offsetof(struct user, u_debugreg[0]);
 891                         ret = ptrace_set_debugreg(child,
 892                                                   addr / sizeof(data), data);
 893                 }
 894                 break;
 895
 896         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 897                 return copy_regset_to_user(child,
 898                                            task_user_regset_view(current),
 899                                            REGSET_GENERAL,
 900                                            0, sizeof(struct user_regs_struct),
 901                                            datap);
 902
 903         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 904                 return copy_regset_from_user(child,
 905                                              task_user_regset_view(current),
 906                                              REGSET_GENERAL,
 907                                              0, sizeof(struct user_regs_struct),
 908                                              datap);
 909
 910         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 911                 return copy_regset_to_user(child,
 912                                            task_user_regset_view(current),
 913                                            REGSET_FP,
 914                                            0, sizeof(struct user_i387_struct),
 915                                            datap);
 916
 917         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 918                 return copy_regset_from_user(child,
 919                                              task_user_regset_view(current),
 920                                              REGSET_FP,
 921                                              0, sizeof(struct user_i387_struct),
 922                                              datap);
 923
 924 #ifdef CONFIG_X86_32
 925         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 926                 return copy_regset_to_user(child, &user_x86_32_view,
 927                                            REGSET_XFP,
 928                                            0, sizeof(struct user_fxsr_struct),
 929                                            datap) ? -EIO : 0;
 930
 931         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 932                 return copy_regset_from_user(child, &user_x86_32_view,
 933                                              REGSET_XFP,
 934                                              0, sizeof(struct user_fxsr_struct),
 935                                              datap) ? -EIO : 0;
 936 #endif
 937
 938 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 939         case PTRACE_GET_THREAD_AREA:
 940                 if (addr < 0)
 941                         return -EIO;
 942                 ret = do_get_thread_area(child, addr,
 943                                          (struct user_desc __user *) data);
 944                 break;
 945
 946         case PTRACE_SET_THREAD_AREA:
 947                 if (addr < 0)
 948                         return -EIO;
 949                 ret = do_set_thread_area(child, addr,
 950                                          (struct user_desc __user *) data, 0);
 951                 break;
 952 #endif
 953
 954 #ifdef CONFIG_X86_64
 955                 /* normal 64bit interface to access TLS data.
 956                    Works just like arch_prctl, except that the arguments
 957                    are reversed. */
 958         case PTRACE_ARCH_PRCTL:
 959                 ret = do_arch_prctl(child, data, addr);
 960                 break;
 961 #endif
 962
 963         /*
 964          * These bits need more cooking - not enabled yet:
 965          */
 966 #ifdef CONFIG_X86_PTRACE_BTS
 967         case PTRACE_BTS_CONFIG:
 968                 ret = ptrace_bts_config
 969                         (child, data, (struct ptrace_bts_config __user *)addr);
 970                 break;
 971
 972         case PTRACE_BTS_STATUS:
 973                 ret = ptrace_bts_status
 974                         (child, data, (struct ptrace_bts_config __user *)addr);
 975                 break;
 976
 977         case PTRACE_BTS_SIZE:
 978                 ret = ptrace_bts_size(child);
 979                 break;
 980
 981         case PTRACE_BTS_GET:
 982                 ret = ptrace_bts_read_record
 983                         (child, data, (struct bts_struct __user *) addr);
 984                 break;
 985
 986         case PTRACE_BTS_CLEAR:
 987                 ret = ptrace_bts_clear(child);
 988                 break;
 989
 990         case PTRACE_BTS_DRAIN:
 991                 ret = ptrace_bts_drain
 992                         (child, data, (struct bts_struct __user *) addr);
 993                 break;
 994 #endif /* CONFIG_X86_PTRACE_BTS */
 995
 996         default:
 997                 ret = ptrace_request(child, request, addr, data);
 998                 break;
 999         }
1000
1001         return ret;
1002 }
1003
1004 #ifdef CONFIG_IA32_EMULATION
1005
1006 #include <linux/compat.h>
1007 #include <linux/syscalls.h>
1008 #include <asm/ia32.h>
1009 #include <asm/user32.h>
1010
1011 #define R32(l,q)                                                        \
1012         case offsetof(struct user32, regs.l):                           \
1013                 regs->q = value; break
1014
1015 #define SEG32(rs)                                                       \
1016         case offsetof(struct user32, regs.rs):                          \
1017                 return set_segment_reg(child,                           \
1018                                        offsetof(struct user_regs_struct, rs), \
1019                                        value);                          \
1020                 break
1021
1022 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1023 {
1024         struct pt_regs *regs = task_pt_regs(child);
1025
1026         switch (regno) {
1027
1028         SEG32(cs);
1029         SEG32(ds);
1030         SEG32(es);
1031         SEG32(fs);
1032         SEG32(gs);
1033         SEG32(ss);
1034
1035         R32(ebx, bx);
1036         R32(ecx, cx);
1037         R32(edx, dx);
1038         R32(edi, di);
1039         R32(esi, si);
1040         R32(ebp, bp);
1041         R32(eax, ax);
1042         R32(eip, ip);
1043         R32(esp, sp);
1044
1045         case offsetof(struct user32, regs.orig_eax):
1046                 /*
1047                  * Sign-extend the value so that orig_eax = -1
1048                  * causes (long)orig_ax < 0 tests to fire correctly.
1049                  */
1050                 regs->orig_ax = (long) (s32) value;
1051                 break;
1052
1053         case offsetof(struct user32, regs.eflags):
1054                 return set_flags(child, value);
1055
1056         case offsetof(struct user32, u_debugreg[0]) ...
1057                 offsetof(struct user32, u_debugreg[7]):
1058                 regno -= offsetof(struct user32, u_debugreg[0]);
1059                 return ptrace_set_debugreg(child, regno / 4, value);
1060
1061         default:
1062                 if (regno > sizeof(struct user32) || (regno & 3))
1063                         return -EIO;
1064
1065                 /*
1066                  * Other dummy fields in the virtual user structure
1067                  * are ignored
1068                  */
1069                 break;
1070         }
1071         return 0;
1072 }
1073
1074 #undef R32
1075 #undef SEG32
1076
1077 #define R32(l,q)                                                        \
1078         case offsetof(struct user32, regs.l):                           \
1079                 *val = regs->q; break
1080
1081 #define SEG32(rs)                                                       \
1082         case offsetof(struct user32, regs.rs):                          \
1083                 *val = get_segment_reg(child,                           \
1084                                        offsetof(struct user_regs_struct, rs)); \
1085                 break
1086
1087 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1088 {
1089         struct pt_regs *regs = task_pt_regs(child);
1090
1091         switch (regno) {
1092
1093         SEG32(ds);
1094         SEG32(es);
1095         SEG32(fs);
1096         SEG32(gs);
1097
1098         R32(cs, cs);
1099         R32(ss, ss);
1100         R32(ebx, bx);
1101         R32(ecx, cx);
1102         R32(edx, dx);
1103         R32(edi, di);
1104         R32(esi, si);
1105         R32(ebp, bp);
1106         R32(eax, ax);
1107         R32(orig_eax, orig_ax);
1108         R32(eip, ip);
1109         R32(esp, sp);
1110
1111         case offsetof(struct user32, regs.eflags):
1112                 *val = get_flags(child);
1113                 break;
1114
1115         case offsetof(struct user32, u_debugreg[0]) ...
1116                 offsetof(struct user32, u_debugreg[7]):
1117                 regno -= offsetof(struct user32, u_debugreg[0]);
1118                 *val = ptrace_get_debugreg(child, regno / 4);
1119                 break;
1120
1121         default:
1122                 if (regno > sizeof(struct user32) || (regno & 3))
1123                         return -EIO;
1124
1125                 /*
1126                  * Other dummy fields in the virtual user structure
1127                  * are ignored
1128                  */
1129                 *val = 0;
1130                 break;
1131         }
1132         return 0;
1133 }
1134
1135 #undef R32
1136 #undef SEG32
1137
1138 static int genregs32_get(struct task_struct *target,
1139                          const struct user_regset *regset,
1140                          unsigned int pos, unsigned int count,
1141                          void *kbuf, void __user *ubuf)
1142 {
1143         if (kbuf) {
1144                 compat_ulong_t *k = kbuf;
1145                 while (count > 0) {
1146                         getreg32(target, pos, k++);
1147                         count -= sizeof(*k);
1148                         pos += sizeof(*k);
1149                 }
1150         } else {
1151                 compat_ulong_t __user *u = ubuf;
1152                 while (count > 0) {
1153                         compat_ulong_t word;
1154                         getreg32(target, pos, &word);
1155                         if (__put_user(word, u++))
1156                                 return -EFAULT;
1157                         count -= sizeof(*u);
1158                         pos += sizeof(*u);
1159                 }
1160         }
1161
1162         return 0;
1163 }
1164
1165 static int genregs32_set(struct task_struct *target,
1166                          const struct user_regset *regset,
1167                          unsigned int pos, unsigned int count,
1168                          const void *kbuf, const void __user *ubuf)
1169 {
1170         int ret = 0;
1171         if (kbuf) {
1172                 const compat_ulong_t *k = kbuf;
1173                 while (count > 0 && !ret) {
1174                         ret = putreg32(target, pos, *k++);
1175                         count -= sizeof(*k);
1176                         pos += sizeof(*k);
1177                 }
1178         } else {
1179                 const compat_ulong_t __user *u = ubuf;
1180                 while (count > 0 && !ret) {
1181                         compat_ulong_t word;
1182                         ret = __get_user(word, u++);
1183                         if (ret)
1184                                 break;
1185                         ret = putreg32(target, pos, word);
1186                         count -= sizeof(*u);
1187                         pos += sizeof(*u);
1188                 }
1189         }
1190         return ret;
1191 }
1192
1193 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1194                         compat_ulong_t caddr, compat_ulong_t cdata)
1195 {
1196         unsigned long addr = caddr;
1197         unsigned long data = cdata;
1198         void __user *datap = compat_ptr(data);
1199         int ret;
1200         __u32 val;
1201
1202         switch (request) {
1203         case PTRACE_PEEKUSR:
1204                 ret = getreg32(child, addr, &val);
1205                 if (ret == 0)
1206                         ret = put_user(val, (__u32 __user *)datap);
1207                 break;
1208
1209         case PTRACE_POKEUSR:
1210                 ret = putreg32(child, addr, data);
1211                 break;
1212
1213         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1214                 return copy_regset_to_user(child, &user_x86_32_view,
1215                                            REGSET_GENERAL,
1216                                            0, sizeof(struct user_regs_struct32),
1217                                            datap);
1218
1219         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1220                 return copy_regset_from_user(child, &user_x86_32_view,
1221                                              REGSET_GENERAL, 0,
1222                                              sizeof(struct user_regs_struct32),
1223                                              datap);
1224
1225         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1226                 return copy_regset_to_user(child, &user_x86_32_view,
1227                                            REGSET_FP, 0,
1228                                            sizeof(struct user_i387_ia32_struct),
1229                                            datap);
1230
1231         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1232                 return copy_regset_from_user(
1233                         child, &user_x86_32_view, REGSET_FP,
1234                         0, sizeof(struct user_i387_ia32_struct), datap);
1235
1236         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1237                 return copy_regset_to_user(child, &user_x86_32_view,
1238                                            REGSET_XFP, 0,
1239                                            sizeof(struct user32_fxsr_struct),
1240                                            datap);
1241
1242         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1243                 return copy_regset_from_user(child, &user_x86_32_view,
1244                                              REGSET_XFP, 0,
1245                                              sizeof(struct user32_fxsr_struct),
1246                                              datap);
1247
1248         case PTRACE_GET_THREAD_AREA:
1249         case PTRACE_SET_THREAD_AREA:
1250 #ifdef CONFIG_X86_PTRACE_BTS
1251         case PTRACE_BTS_CONFIG:
1252         case PTRACE_BTS_STATUS:
1253         case PTRACE_BTS_SIZE:
1254         case PTRACE_BTS_GET:
1255         case PTRACE_BTS_CLEAR:
1256         case PTRACE_BTS_DRAIN:
1257 #endif /* CONFIG_X86_PTRACE_BTS */
1258                 return arch_ptrace(child, request, addr, data);
1259
1260         default:
1261                 return compat_ptrace_request(child, request, addr, data);
1262         }
1263
1264         return ret;
1265 }
1266
1267 #endif  /* CONFIG_IA32_EMULATION */
1268
1269 #ifdef CONFIG_X86_64
1270
1271 static const struct user_regset x86_64_regsets[] = {
1272         [REGSET_GENERAL] = {
1273                 .core_note_type = NT_PRSTATUS,
1274                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1275                 .size = sizeof(long), .align = sizeof(long),
1276                 .get = genregs_get, .set = genregs_set
1277         },
1278         [REGSET_FP] = {
1279                 .core_note_type = NT_PRFPREG,
1280                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1281                 .size = sizeof(long), .align = sizeof(long),
1282                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1283         },
1284         [REGSET_IOPERM64] = {
1285                 .core_note_type = NT_386_IOPERM,
1286                 .n = IO_BITMAP_LONGS,
1287                 .size = sizeof(long), .align = sizeof(long),
1288                 .active = ioperm_active, .get = ioperm_get
1289         },
1290 };
1291
1292 static const struct user_regset_view user_x86_64_view = {
1293         .name = "x86_64", .e_machine = EM_X86_64,
1294         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1295 };
1296
1297 #else  /* CONFIG_X86_32 */
1298
1299 #define user_regs_struct32      user_regs_struct
1300 #define genregs32_get           genregs_get
1301 #define genregs32_set           genregs_set
1302
1303 #define user_i387_ia32_struct   user_i387_struct
1304 #define user32_fxsr_struct      user_fxsr_struct
1305
1306 #endif  /* CONFIG_X86_64 */
1307
1308 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1309 static const struct user_regset x86_32_regsets[] = {
1310         [REGSET_GENERAL] = {
1311                 .core_note_type = NT_PRSTATUS,
1312                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1313                 .size = sizeof(u32), .align = sizeof(u32),
1314                 .get = genregs32_get, .set = genregs32_set
1315         },
1316         [REGSET_FP] = {
1317                 .core_note_type = NT_PRFPREG,
1318                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1319                 .size = sizeof(u32), .align = sizeof(u32),
1320                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1321         },
1322         [REGSET_XFP] = {
1323                 .core_note_type = NT_PRXFPREG,
1324                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1325                 .size = sizeof(u32), .align = sizeof(u32),
1326                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1327         },
1328         [REGSET_TLS] = {
1329                 .core_note_type = NT_386_TLS,
1330                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1331                 .size = sizeof(struct user_desc),
1332                 .align = sizeof(struct user_desc),
1333                 .active = regset_tls_active,
1334                 .get = regset_tls_get, .set = regset_tls_set
1335         },
1336         [REGSET_IOPERM32] = {
1337                 .core_note_type = NT_386_IOPERM,
1338                 .n = IO_BITMAP_BYTES / sizeof(u32),
1339                 .size = sizeof(u32), .align = sizeof(u32),
1340                 .active = ioperm_active, .get = ioperm_get
1341         },
1342 };
1343
1344 static const struct user_regset_view user_x86_32_view = {
1345         .name = "i386", .e_machine = EM_386,
1346         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1347 };
1348 #endif
1349
1350 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1351 {
1352 #ifdef CONFIG_IA32_EMULATION
1353         if (test_tsk_thread_flag(task, TIF_IA32))
1354 #endif
1355 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1356                 return &user_x86_32_view;
1357 #endif
1358 #ifdef CONFIG_X86_64
1359         return &user_x86_64_view;
1360 #endif
1361 }
1362
1363 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1364                                          int error_code, int si_code)
1365 {
1366         struct siginfo info;
1367
1368         tsk->thread.trap_no = 1;
1369         tsk->thread.error_code = error_code;
1370
1371         memset(&info, 0, sizeof(info));
1372         info.si_signo = SIGTRAP;
1373         info.si_code = si_code;
1374
1375         /* User-mode ip? */
1376         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1377
1378         /* Send us the fake SIGTRAP */
1379         force_sig_info(SIGTRAP, &info, tsk);
1380 }
1381
1382
1383 #ifdef CONFIG_X86_32
1384 # define IS_IA32        1
1385 #elif defined CONFIG_IA32_EMULATION
1386 # define IS_IA32        is_compat_task()
1387 #else
1388 # define IS_IA32        0
1389 #endif
1390
1391 /*
1392  * We must return the syscall number to actually look up in the table.
1393  * This can be -1L to skip running any syscall at all.
1394  */
1395 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1396 {
1397         long ret = 0;
1398
1399         /*
1400          * If we stepped into a sysenter/syscall insn, it trapped in
1401          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1402          * If user-mode had set TF itself, then it's still clear from
1403          * do_debug() and we need to set it again to restore the user
1404          * state.  If we entered on the slow path, TF was already set.
1405          */
1406         if (test_thread_flag(TIF_SINGLESTEP))
1407                 regs->flags |= X86_EFLAGS_TF;
1408
1409         /* do the secure computing check first */
1410         secure_computing(regs->orig_ax);
1411
1412         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1413                 ret = -1L;
1414
1415         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1416             tracehook_report_syscall_entry(regs))
1417                 ret = -1L;
1418
1419         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1420                 ftrace_syscall_enter(regs);
1421
1422         if (unlikely(current->audit_context)) {
1423                 if (IS_IA32)
1424                         audit_syscall_entry(AUDIT_ARCH_I386,
1425                                             regs->orig_ax,
1426                                             regs->bx, regs->cx,
1427                                             regs->dx, regs->si);
1428 #ifdef CONFIG_X86_64
1429                 else
1430                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1431                                             regs->orig_ax,
1432                                             regs->di, regs->si,
1433                                             regs->dx, regs->r10);
1434 #endif
1435         }
1436
1437         return ret ?: regs->orig_ax;
1438 }
1439
1440 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1441 {
1442         if (unlikely(current->audit_context))
1443                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1444
1445         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1446                 ftrace_syscall_exit(regs);
1447
1448         if (test_thread_flag(TIF_SYSCALL_TRACE))
1449                 tracehook_report_syscall_exit(regs, 0);
1450
1451         /*
1452          * If TIF_SYSCALL_EMU is set, we only get here because of
1453          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1454          * We already reported this syscall instruction in
1455          * syscall_trace_enter(), so don't do any more now.
1456          */
1457         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1458                 return;
1459
1460         /*
1461          * If we are single-stepping, synthesize a trap to follow the
1462          * system call instruction.
1463          */
1464         if (test_thread_flag(TIF_SINGLESTEP) &&
1465             tracehook_consider_fatal_signal(current, SIGTRAP))
1466                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1467 }