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