SAFE public projects git trees. - safe/jmp/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/user.h>
  18 #include <linux/elf.h>
  19 #include <linux/security.h>
  20 #include <linux/audit.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/signal.h>
  23
  24 #include <asm/uaccess.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/system.h>
  27 #include <asm/processor.h>
  28 #include <asm/i387.h>
  29 #include <asm/debugreg.h>
  30 #include <asm/ldt.h>
  31 #include <asm/desc.h>
  32 #include <asm/prctl.h>
  33 #include <asm/proto.h>
  34 #include <asm/ds.h>
  35
  36 #include "tls.h"
  37
  38 enum x86_regset {
  39         REGSET_GENERAL,
  40         REGSET_FP,
  41         REGSET_XFP,
  42         REGSET_TLS,
  43 };
  44
  45 /*
  46  * does not yet catch signals sent when the child dies.
  47  * in exit.c or in signal.c.
  48  */
  49
  50 /*
  51  * Determines which flags the user has access to [1 = access, 0 = no access].
  52  */
  53 #define FLAG_MASK_32            ((unsigned long)                        \
  54                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  55                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  56                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  57                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  58                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  59
  60 /*
  61  * Determines whether a value may be installed in a segment register.
  62  */
  63 static inline bool invalid_selector(u16 value)
  64 {
  65         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  66 }
  67
  68 #ifdef CONFIG_X86_32
  69
  70 #define FLAG_MASK               FLAG_MASK_32
  71
  72 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  73 {
  74         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  75         regno >>= 2;
  76         if (regno > FS)
  77                 --regno;
  78         return &regs->bx + regno;
  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                 retval = task->thread.gs;
  91                 if (task == current)
  92                         savesegment(gs, retval);
  93         }
  94         return retval;
  95 }
  96
  97 static int set_segment_reg(struct task_struct *task,
  98                            unsigned long offset, u16 value)
  99 {
 100         /*
 101          * The value argument was already truncated to 16 bits.
 102          */
 103         if (invalid_selector(value))
 104                 return -EIO;
 105
 106         if (offset != offsetof(struct user_regs_struct, gs))
 107                 *pt_regs_access(task_pt_regs(task), offset) = value;
 108         else {
 109                 task->thread.gs = value;
 110                 if (task == current)
 111                         /*
 112                          * The user-mode %gs is not affected by
 113                          * kernel entry, so we must update the CPU.
 114                          */
 115                         loadsegment(gs, value);
 116         }
 117
 118         return 0;
 119 }
 120
 121 static unsigned long debugreg_addr_limit(struct task_struct *task)
 122 {
 123         return TASK_SIZE - 3;
 124 }
 125
 126 #else  /* CONFIG_X86_64 */
 127
 128 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 129
 130 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 131 {
 132         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 133         return &regs->r15 + (offset / sizeof(regs->r15));
 134 }
 135
 136 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 137 {
 138         /*
 139          * Returning the value truncates it to 16 bits.
 140          */
 141         unsigned int seg;
 142
 143         switch (offset) {
 144         case offsetof(struct user_regs_struct, fs):
 145                 if (task == current) {
 146                         /* Older gas can't assemble movq %?s,%r?? */
 147                         asm("movl %%fs,%0" : "=r" (seg));
 148                         return seg;
 149                 }
 150                 return task->thread.fsindex;
 151         case offsetof(struct user_regs_struct, gs):
 152                 if (task == current) {
 153                         asm("movl %%gs,%0" : "=r" (seg));
 154                         return seg;
 155                 }
 156                 return task->thread.gsindex;
 157         case offsetof(struct user_regs_struct, ds):
 158                 if (task == current) {
 159                         asm("movl %%ds,%0" : "=r" (seg));
 160                         return seg;
 161                 }
 162                 return task->thread.ds;
 163         case offsetof(struct user_regs_struct, es):
 164                 if (task == current) {
 165                         asm("movl %%es,%0" : "=r" (seg));
 166                         return seg;
 167                 }
 168                 return task->thread.es;
 169
 170         case offsetof(struct user_regs_struct, cs):
 171         case offsetof(struct user_regs_struct, ss):
 172                 break;
 173         }
 174         return *pt_regs_access(task_pt_regs(task), offset);
 175 }
 176
 177 static int set_segment_reg(struct task_struct *task,
 178                            unsigned long offset, u16 value)
 179 {
 180         /*
 181          * The value argument was already truncated to 16 bits.
 182          */
 183         if (invalid_selector(value))
 184                 return -EIO;
 185
 186         switch (offset) {
 187         case offsetof(struct user_regs_struct,fs):
 188                 /*
 189                  * If this is setting fs as for normal 64-bit use but
 190                  * setting fs_base has implicitly changed it, leave it.
 191                  */
 192                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 193                      task->thread.fs != 0) ||
 194                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 195                      task->thread.fs == 0))
 196                         break;
 197                 task->thread.fsindex = value;
 198                 if (task == current)
 199                         loadsegment(fs, task->thread.fsindex);
 200                 break;
 201         case offsetof(struct user_regs_struct,gs):
 202                 /*
 203                  * If this is setting gs as for normal 64-bit use but
 204                  * setting gs_base has implicitly changed it, leave it.
 205                  */
 206                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 207                      task->thread.gs != 0) ||
 208                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 209                      task->thread.gs == 0))
 210                         break;
 211                 task->thread.gsindex = value;
 212                 if (task == current)
 213                         load_gs_index(task->thread.gsindex);
 214                 break;
 215         case offsetof(struct user_regs_struct,ds):
 216                 task->thread.ds = value;
 217                 if (task == current)
 218                         loadsegment(ds, task->thread.ds);
 219                 break;
 220         case offsetof(struct user_regs_struct,es):
 221                 task->thread.es = value;
 222                 if (task == current)
 223                         loadsegment(es, task->thread.es);
 224                 break;
 225
 226                 /*
 227                  * Can't actually change these in 64-bit mode.
 228                  */
 229         case offsetof(struct user_regs_struct,cs):
 230 #ifdef CONFIG_IA32_EMULATION
 231                 if (test_tsk_thread_flag(task, TIF_IA32))
 232                         task_pt_regs(task)->cs = value;
 233 #endif
 234                 break;
 235         case offsetof(struct user_regs_struct,ss):
 236 #ifdef CONFIG_IA32_EMULATION
 237                 if (test_tsk_thread_flag(task, TIF_IA32))
 238                         task_pt_regs(task)->ss = value;
 239 #endif
 240                 break;
 241         }
 242
 243         return 0;
 244 }
 245
 246 static unsigned long debugreg_addr_limit(struct task_struct *task)
 247 {
 248 #ifdef CONFIG_IA32_EMULATION
 249         if (test_tsk_thread_flag(task, TIF_IA32))
 250                 return IA32_PAGE_OFFSET - 3;
 251 #endif
 252         return TASK_SIZE64 - 7;
 253 }
 254
 255 #endif  /* CONFIG_X86_32 */
 256
 257 static unsigned long get_flags(struct task_struct *task)
 258 {
 259         unsigned long retval = task_pt_regs(task)->flags;
 260
 261         /*
 262          * If the debugger set TF, hide it from the readout.
 263          */
 264         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 265                 retval &= ~X86_EFLAGS_TF;
 266
 267         return retval;
 268 }
 269
 270 static int set_flags(struct task_struct *task, unsigned long value)
 271 {
 272         struct pt_regs *regs = task_pt_regs(task);
 273
 274         /*
 275          * If the user value contains TF, mark that
 276          * it was not "us" (the debugger) that set it.
 277          * If not, make sure it stays set if we had.
 278          */
 279         if (value & X86_EFLAGS_TF)
 280                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 281         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 282                 value |= X86_EFLAGS_TF;
 283
 284         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 285
 286         return 0;
 287 }
 288
 289 static int putreg(struct task_struct *child,
 290                   unsigned long offset, unsigned long value)
 291 {
 292         switch (offset) {
 293         case offsetof(struct user_regs_struct, cs):
 294         case offsetof(struct user_regs_struct, ds):
 295         case offsetof(struct user_regs_struct, es):
 296         case offsetof(struct user_regs_struct, fs):
 297         case offsetof(struct user_regs_struct, gs):
 298         case offsetof(struct user_regs_struct, ss):
 299                 return set_segment_reg(child, offset, value);
 300
 301         case offsetof(struct user_regs_struct, flags):
 302                 return set_flags(child, value);
 303
 304 #ifdef CONFIG_X86_64
 305         case offsetof(struct user_regs_struct,fs_base):
 306                 if (value >= TASK_SIZE_OF(child))
 307                         return -EIO;
 308                 /*
 309                  * When changing the segment base, use do_arch_prctl
 310                  * to set either thread.fs or thread.fsindex and the
 311                  * corresponding GDT slot.
 312                  */
 313                 if (child->thread.fs != value)
 314                         return do_arch_prctl(child, ARCH_SET_FS, value);
 315                 return 0;
 316         case offsetof(struct user_regs_struct,gs_base):
 317                 /*
 318                  * Exactly the same here as the %fs handling above.
 319                  */
 320                 if (value >= TASK_SIZE_OF(child))
 321                         return -EIO;
 322                 if (child->thread.gs != value)
 323                         return do_arch_prctl(child, ARCH_SET_GS, value);
 324                 return 0;
 325 #endif
 326         }
 327
 328         *pt_regs_access(task_pt_regs(child), offset) = value;
 329         return 0;
 330 }
 331
 332 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 333 {
 334         switch (offset) {
 335         case offsetof(struct user_regs_struct, cs):
 336         case offsetof(struct user_regs_struct, ds):
 337         case offsetof(struct user_regs_struct, es):
 338         case offsetof(struct user_regs_struct, fs):
 339         case offsetof(struct user_regs_struct, gs):
 340         case offsetof(struct user_regs_struct, ss):
 341                 return get_segment_reg(task, offset);
 342
 343         case offsetof(struct user_regs_struct, flags):
 344                 return get_flags(task);
 345
 346 #ifdef CONFIG_X86_64
 347         case offsetof(struct user_regs_struct, fs_base): {
 348                 /*
 349                  * do_arch_prctl may have used a GDT slot instead of
 350                  * the MSR.  To userland, it appears the same either
 351                  * way, except the %fs segment selector might not be 0.
 352                  */
 353                 unsigned int seg = task->thread.fsindex;
 354                 if (task->thread.fs != 0)
 355                         return task->thread.fs;
 356                 if (task == current)
 357                         asm("movl %%fs,%0" : "=r" (seg));
 358                 if (seg != FS_TLS_SEL)
 359                         return 0;
 360                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 361         }
 362         case offsetof(struct user_regs_struct, gs_base): {
 363                 /*
 364                  * Exactly the same here as the %fs handling above.
 365                  */
 366                 unsigned int seg = task->thread.gsindex;
 367                 if (task->thread.gs != 0)
 368                         return task->thread.gs;
 369                 if (task == current)
 370                         asm("movl %%gs,%0" : "=r" (seg));
 371                 if (seg != GS_TLS_SEL)
 372                         return 0;
 373                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 374         }
 375 #endif
 376         }
 377
 378         return *pt_regs_access(task_pt_regs(task), offset);
 379 }
 380
 381 static int genregs_get(struct task_struct *target,
 382                        const struct user_regset *regset,
 383                        unsigned int pos, unsigned int count,
 384                        void *kbuf, void __user *ubuf)
 385 {
 386         if (kbuf) {
 387                 unsigned long *k = kbuf;
 388                 while (count > 0) {
 389                         *k++ = getreg(target, pos);
 390                         count -= sizeof(*k);
 391                         pos += sizeof(*k);
 392                 }
 393         } else {
 394                 unsigned long __user *u = ubuf;
 395                 while (count > 0) {
 396                         if (__put_user(getreg(target, pos), u++))
 397                                 return -EFAULT;
 398                         count -= sizeof(*u);
 399                         pos += sizeof(*u);
 400                 }
 401         }
 402
 403         return 0;
 404 }
 405
 406 static int genregs_set(struct task_struct *target,
 407                        const struct user_regset *regset,
 408                        unsigned int pos, unsigned int count,
 409                        const void *kbuf, const void __user *ubuf)
 410 {
 411         int ret = 0;
 412         if (kbuf) {
 413                 const unsigned long *k = kbuf;
 414                 while (count > 0 && !ret) {
 415                         ret = putreg(target, pos, *k++);
 416                         count -= sizeof(*k);
 417                         pos += sizeof(*k);
 418                 }
 419         } else {
 420                 const unsigned long  __user *u = ubuf;
 421                 while (count > 0 && !ret) {
 422                         unsigned long word;
 423                         ret = __get_user(word, u++);
 424                         if (ret)
 425                                 break;
 426                         ret = putreg(target, pos, word);
 427                         count -= sizeof(*u);
 428                         pos += sizeof(*u);
 429                 }
 430         }
 431         return ret;
 432 }
 433
 434 /*
 435  * This function is trivial and will be inlined by the compiler.
 436  * Having it separates the implementation details of debug
 437  * registers from the interface details of ptrace.
 438  */
 439 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 440 {
 441         switch (n) {
 442         case 0:         return child->thread.debugreg0;
 443         case 1:         return child->thread.debugreg1;
 444         case 2:         return child->thread.debugreg2;
 445         case 3:         return child->thread.debugreg3;
 446         case 6:         return child->thread.debugreg6;
 447         case 7:         return child->thread.debugreg7;
 448         }
 449         return 0;
 450 }
 451
 452 static int ptrace_set_debugreg(struct task_struct *child,
 453                                int n, unsigned long data)
 454 {
 455         int i;
 456
 457         if (unlikely(n == 4 || n == 5))
 458                 return -EIO;
 459
 460         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 461                 return -EIO;
 462
 463         switch (n) {
 464         case 0:         child->thread.debugreg0 = data; break;
 465         case 1:         child->thread.debugreg1 = data; break;
 466         case 2:         child->thread.debugreg2 = data; break;
 467         case 3:         child->thread.debugreg3 = data; break;
 468
 469         case 6:
 470                 if ((data & ~0xffffffffUL) != 0)
 471                         return -EIO;
 472                 child->thread.debugreg6 = data;
 473                 break;
 474
 475         case 7:
 476                 /*
 477                  * Sanity-check data. Take one half-byte at once with
 478                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 479                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 480                  * 2 and 3 are LENi. Given a list of invalid values,
 481                  * we do mask |= 1 << invalid_value, so that
 482                  * (mask >> check) & 1 is a correct test for invalid
 483                  * values.
 484                  *
 485                  * R/Wi contains the type of the breakpoint /
 486                  * watchpoint, LENi contains the length of the watched
 487                  * data in the watchpoint case.
 488                  *
 489                  * The invalid values are:
 490                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 491                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 492                  *   mask |= 0x4444.
 493                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 494                  *   0x1110.
 495                  *
 496                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 497                  *
 498                  * See the Intel Manual "System Programming Guide",
 499                  * 15.2.4
 500                  *
 501                  * Note that LENi == 0x10 is defined on x86_64 in long
 502                  * mode (i.e. even for 32-bit userspace software, but
 503                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 504                  * See the AMD manual no. 24593 (AMD64 System Programming)
 505                  */
 506 #ifdef CONFIG_X86_32
 507 #define DR7_MASK        0x5f54
 508 #else
 509 #define DR7_MASK        0x5554
 510 #endif
 511                 data &= ~DR_CONTROL_RESERVED;
 512                 for (i = 0; i < 4; i++)
 513                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 514                                 return -EIO;
 515                 child->thread.debugreg7 = data;
 516                 if (data)
 517                         set_tsk_thread_flag(child, TIF_DEBUG);
 518                 else
 519                         clear_tsk_thread_flag(child, TIF_DEBUG);
 520                 break;
 521         }
 522
 523         return 0;
 524 }
 525
 526 static int ptrace_bts_get_size(struct task_struct *child)
 527 {
 528         if (!child->thread.ds_area_msr)
 529                 return -ENXIO;
 530
 531         return ds_get_bts_index((void *)child->thread.ds_area_msr);
 532 }
 533
 534 static int ptrace_bts_read_record(struct task_struct *child,
 535                                   long index,
 536                                   struct bts_struct __user *out)
 537 {
 538         struct bts_struct ret;
 539         int retval;
 540         int bts_end;
 541         int bts_index;
 542
 543         if (!child->thread.ds_area_msr)
 544                 return -ENXIO;
 545
 546         if (index < 0)
 547                 return -EINVAL;
 548
 549         bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
 550         if (bts_end <= index)
 551                 return -EINVAL;
 552
 553         /* translate the ptrace bts index into the ds bts index */
 554         bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
 555         bts_index -= (index + 1);
 556         if (bts_index < 0)
 557                 bts_index += bts_end;
 558
 559         retval = ds_read_bts((void *)child->thread.ds_area_msr,
 560                              bts_index, &ret);
 561         if (retval)
 562                 return retval;
 563
 564         if (copy_to_user(out, &ret, sizeof(ret)))
 565                 return -EFAULT;
 566
 567         return sizeof(ret);
 568 }
 569
 570 static int ptrace_bts_write_record(struct task_struct *child,
 571                                    const struct bts_struct *in)
 572 {
 573         int retval;
 574
 575         if (!child->thread.ds_area_msr)
 576                 return -ENXIO;
 577
 578         retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
 579         if (retval)
 580                 return retval;
 581
 582         return sizeof(*in);
 583 }
 584
 585 static int ptrace_bts_clear(struct task_struct *child)
 586 {
 587         if (!child->thread.ds_area_msr)
 588                 return -ENXIO;
 589
 590         return ds_clear((void *)child->thread.ds_area_msr);
 591 }
 592
 593 static int ptrace_bts_drain(struct task_struct *child,
 594                             struct bts_struct __user *out)
 595 {
 596         int end, i;
 597         void *ds = (void *)child->thread.ds_area_msr;
 598
 599         if (!ds)
 600                 return -ENXIO;
 601
 602         end = ds_get_bts_index(ds);
 603         if (end <= 0)
 604                 return end;
 605
 606         for (i = 0; i < end; i++, out++) {
 607                 struct bts_struct ret;
 608                 int retval;
 609
 610                 retval = ds_read_bts(ds, i, &ret);
 611                 if (retval < 0)
 612                         return retval;
 613
 614                 if (copy_to_user(out, &ret, sizeof(ret)))
 615                         return -EFAULT;
 616         }
 617
 618         ds_clear(ds);
 619
 620         return i;
 621 }
 622
 623 static int ptrace_bts_config(struct task_struct *child,
 624                              const struct ptrace_bts_config __user *ucfg)
 625 {
 626         struct ptrace_bts_config cfg;
 627         unsigned long debugctl_mask;
 628         int bts_size, ret;
 629         void *ds;
 630
 631         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 632                 return -EFAULT;
 633
 634         bts_size = 0;
 635         ds = (void *)child->thread.ds_area_msr;
 636         if (ds) {
 637                 bts_size = ds_get_bts_size(ds);
 638                 if (bts_size < 0)
 639                         return bts_size;
 640         }
 641
 642         if (bts_size != cfg.size) {
 643                 ret = ds_free((void **)&child->thread.ds_area_msr);
 644                 if (ret < 0)
 645                         return ret;
 646
 647                 if (cfg.size > 0)
 648                         ret = ds_allocate((void **)&child->thread.ds_area_msr,
 649                                           cfg.size);
 650                 ds = (void *)child->thread.ds_area_msr;
 651                 if (ds)
 652                         set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 653                 else
 654                         clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 655
 656                 if (ret < 0)
 657                         return ret;
 658
 659                 bts_size = ds_get_bts_size(ds);
 660                 if (bts_size <= 0)
 661                         return bts_size;
 662         }
 663
 664         if (ds) {
 665                 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 666                         ret = ds_set_overflow(ds, DS_O_SIGNAL);
 667                 } else {
 668                         ret = ds_set_overflow(ds, DS_O_WRAP);
 669                 }
 670                 if (ret < 0)
 671                         return ret;
 672         }
 673
 674         debugctl_mask = ds_debugctl_mask();
 675         if (ds && (cfg.flags & PTRACE_BTS_O_TRACE)) {
 676                 child->thread.debugctlmsr |= debugctl_mask;
 677                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 678         } else {
 679                 /* there is no way for us to check whether we 'own'
 680                  * the respective bits in the DEBUGCTL MSR, we're
 681                  * about to clear */
 682                 child->thread.debugctlmsr &= ~debugctl_mask;
 683
 684                 if (!child->thread.debugctlmsr)
 685                         clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 686         }
 687
 688         if (ds && (cfg.flags & PTRACE_BTS_O_SCHED))
 689                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 690         else
 691                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 692
 693         return 0;
 694 }
 695
 696 static int ptrace_bts_status(struct task_struct *child,
 697                              struct ptrace_bts_config __user *ucfg)
 698 {
 699         void *ds = (void *)child->thread.ds_area_msr;
 700         struct ptrace_bts_config cfg;
 701
 702         memset(&cfg, 0, sizeof(cfg));
 703
 704         if (ds) {
 705                 cfg.size = ds_get_bts_size(ds);
 706
 707                 if (ds_get_overflow(ds) == DS_O_SIGNAL)
 708                         cfg.flags |= PTRACE_BTS_O_SIGNAL;
 709
 710                 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
 711                     child->thread.debugctlmsr & ds_debugctl_mask())
 712                         cfg.flags |= PTRACE_BTS_O_TRACE;
 713
 714                 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
 715                         cfg.flags |= PTRACE_BTS_O_SCHED;
 716         }
 717
 718         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 719                 return -EFAULT;
 720
 721         return sizeof(cfg);
 722 }
 723
 724 void ptrace_bts_take_timestamp(struct task_struct *tsk,
 725                                enum bts_qualifier qualifier)
 726 {
 727         struct bts_struct rec = {
 728                 .qualifier = qualifier,
 729                 .variant.jiffies = jiffies
 730         };
 731
 732         ptrace_bts_write_record(tsk, &rec);
 733 }
 734
 735 /*
 736  * Called by kernel/ptrace.c when detaching..
 737  *
 738  * Make sure the single step bit is not set.
 739  */
 740 void ptrace_disable(struct task_struct *child)
 741 {
 742         user_disable_single_step(child);
 743 #ifdef TIF_SYSCALL_EMU
 744         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 745 #endif
 746         ptrace_bts_config(child, /* options = */ 0);
 747         if (child->thread.ds_area_msr) {
 748             ds_free((void **)&child->thread.ds_area_msr);
 749             clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 750         }
 751 }
 752
 753 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 754 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 755 #endif
 756
 757 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 758 {
 759         int ret;
 760         unsigned long __user *datap = (unsigned long __user *)data;
 761
 762         switch (request) {
 763         /* read the word at location addr in the USER area. */
 764         case PTRACE_PEEKUSR: {
 765                 unsigned long tmp;
 766
 767                 ret = -EIO;
 768                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 769                     addr >= sizeof(struct user))
 770                         break;
 771
 772                 tmp = 0;  /* Default return condition */
 773                 if (addr < sizeof(struct user_regs_struct))
 774                         tmp = getreg(child, addr);
 775                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 776                          addr <= offsetof(struct user, u_debugreg[7])) {
 777                         addr -= offsetof(struct user, u_debugreg[0]);
 778                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 779                 }
 780                 ret = put_user(tmp, datap);
 781                 break;
 782         }
 783
 784         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 785                 ret = -EIO;
 786                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 787                     addr >= sizeof(struct user))
 788                         break;
 789
 790                 if (addr < sizeof(struct user_regs_struct))
 791                         ret = putreg(child, addr, data);
 792                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 793                          addr <= offsetof(struct user, u_debugreg[7])) {
 794                         addr -= offsetof(struct user, u_debugreg[0]);
 795                         ret = ptrace_set_debugreg(child,
 796                                                   addr / sizeof(data), data);
 797                 }
 798                 break;
 799
 800         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 801                 return copy_regset_to_user(child,
 802                                            task_user_regset_view(current),
 803                                            REGSET_GENERAL,
 804                                            0, sizeof(struct user_regs_struct),
 805                                            datap);
 806
 807         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 808                 return copy_regset_from_user(child,
 809                                              task_user_regset_view(current),
 810                                              REGSET_GENERAL,
 811                                              0, sizeof(struct user_regs_struct),
 812                                              datap);
 813
 814         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 815                 return copy_regset_to_user(child,
 816                                            task_user_regset_view(current),
 817                                            REGSET_FP,
 818                                            0, sizeof(struct user_i387_struct),
 819                                            datap);
 820
 821         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 822                 return copy_regset_from_user(child,
 823                                              task_user_regset_view(current),
 824                                              REGSET_FP,
 825                                              0, sizeof(struct user_i387_struct),
 826                                              datap);
 827
 828 #ifdef CONFIG_X86_32
 829         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 830                 return copy_regset_to_user(child, &user_x86_32_view,
 831                                            REGSET_XFP,
 832                                            0, sizeof(struct user_fxsr_struct),
 833                                            datap);
 834
 835         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 836                 return copy_regset_from_user(child, &user_x86_32_view,
 837                                              REGSET_XFP,
 838                                              0, sizeof(struct user_fxsr_struct),
 839                                              datap);
 840 #endif
 841
 842 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 843         case PTRACE_GET_THREAD_AREA:
 844                 if (addr < 0)
 845                         return -EIO;
 846                 ret = do_get_thread_area(child, addr,
 847                                          (struct user_desc __user *) data);
 848                 break;
 849
 850         case PTRACE_SET_THREAD_AREA:
 851                 if (addr < 0)
 852                         return -EIO;
 853                 ret = do_set_thread_area(child, addr,
 854                                          (struct user_desc __user *) data, 0);
 855                 break;
 856 #endif
 857
 858 #ifdef CONFIG_X86_64
 859                 /* normal 64bit interface to access TLS data.
 860                    Works just like arch_prctl, except that the arguments
 861                    are reversed. */
 862         case PTRACE_ARCH_PRCTL:
 863                 ret = do_arch_prctl(child, data, addr);
 864                 break;
 865 #endif
 866
 867         case PTRACE_BTS_CONFIG:
 868                 ret = ptrace_bts_config
 869                         (child, (struct ptrace_bts_config __user *)addr);
 870                 break;
 871
 872         case PTRACE_BTS_STATUS:
 873                 ret = ptrace_bts_status
 874                         (child, (struct ptrace_bts_config __user *)addr);
 875                 break;
 876
 877         case PTRACE_BTS_SIZE:
 878                 ret = ptrace_bts_get_size(child);
 879                 break;
 880
 881         case PTRACE_BTS_GET:
 882                 ret = ptrace_bts_read_record
 883                         (child, data, (struct bts_struct __user *) addr);
 884                 break;
 885
 886         case PTRACE_BTS_CLEAR:
 887                 ret = ptrace_bts_clear(child);
 888                 break;
 889
 890         case PTRACE_BTS_DRAIN:
 891                 ret = ptrace_bts_drain
 892                         (child, (struct bts_struct __user *) addr);
 893                 break;
 894
 895         default:
 896                 ret = ptrace_request(child, request, addr, data);
 897                 break;
 898         }
 899
 900         return ret;
 901 }
 902
 903 #ifdef CONFIG_IA32_EMULATION
 904
 905 #include <linux/compat.h>
 906 #include <linux/syscalls.h>
 907 #include <asm/ia32.h>
 908 #include <asm/user32.h>
 909
 910 #define R32(l,q)                                                        \
 911         case offsetof(struct user32, regs.l):                           \
 912                 regs->q = value; break
 913
 914 #define SEG32(rs)                                                       \
 915         case offsetof(struct user32, regs.rs):                          \
 916                 return set_segment_reg(child,                           \
 917                                        offsetof(struct user_regs_struct, rs), \
 918                                        value);                          \
 919                 break
 920
 921 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 922 {
 923         struct pt_regs *regs = task_pt_regs(child);
 924
 925         switch (regno) {
 926
 927         SEG32(cs);
 928         SEG32(ds);
 929         SEG32(es);
 930         SEG32(fs);
 931         SEG32(gs);
 932         SEG32(ss);
 933
 934         R32(ebx, bx);
 935         R32(ecx, cx);
 936         R32(edx, dx);
 937         R32(edi, di);
 938         R32(esi, si);
 939         R32(ebp, bp);
 940         R32(eax, ax);
 941         R32(orig_eax, orig_ax);
 942         R32(eip, ip);
 943         R32(esp, sp);
 944
 945         case offsetof(struct user32, regs.eflags):
 946                 return set_flags(child, value);
 947
 948         case offsetof(struct user32, u_debugreg[0]) ...
 949                 offsetof(struct user32, u_debugreg[7]):
 950                 regno -= offsetof(struct user32, u_debugreg[0]);
 951                 return ptrace_set_debugreg(child, regno / 4, value);
 952
 953         default:
 954                 if (regno > sizeof(struct user32) || (regno & 3))
 955                         return -EIO;
 956
 957                 /*
 958                  * Other dummy fields in the virtual user structure
 959                  * are ignored
 960                  */
 961                 break;
 962         }
 963         return 0;
 964 }
 965
 966 #undef R32
 967 #undef SEG32
 968
 969 #define R32(l,q)                                                        \
 970         case offsetof(struct user32, regs.l):                           \
 971                 *val = regs->q; break
 972
 973 #define SEG32(rs)                                                       \
 974         case offsetof(struct user32, regs.rs):                          \
 975                 *val = get_segment_reg(child,                           \
 976                                        offsetof(struct user_regs_struct, rs)); \
 977                 break
 978
 979 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
 980 {
 981         struct pt_regs *regs = task_pt_regs(child);
 982
 983         switch (regno) {
 984
 985         SEG32(ds);
 986         SEG32(es);
 987         SEG32(fs);
 988         SEG32(gs);
 989
 990         R32(cs, cs);
 991         R32(ss, ss);
 992         R32(ebx, bx);
 993         R32(ecx, cx);
 994         R32(edx, dx);
 995         R32(edi, di);
 996         R32(esi, si);
 997         R32(ebp, bp);
 998         R32(eax, ax);
 999         R32(orig_eax, orig_ax);
1000         R32(eip, ip);
1001         R32(esp, sp);
1002
1003         case offsetof(struct user32, regs.eflags):
1004                 *val = get_flags(child);
1005                 break;
1006
1007         case offsetof(struct user32, u_debugreg[0]) ...
1008                 offsetof(struct user32, u_debugreg[7]):
1009                 regno -= offsetof(struct user32, u_debugreg[0]);
1010                 *val = ptrace_get_debugreg(child, regno / 4);
1011                 break;
1012
1013         default:
1014                 if (regno > sizeof(struct user32) || (regno & 3))
1015                         return -EIO;
1016
1017                 /*
1018                  * Other dummy fields in the virtual user structure
1019                  * are ignored
1020                  */
1021                 *val = 0;
1022                 break;
1023         }
1024         return 0;
1025 }
1026
1027 #undef R32
1028 #undef SEG32
1029
1030 static int genregs32_get(struct task_struct *target,
1031                          const struct user_regset *regset,
1032                          unsigned int pos, unsigned int count,
1033                          void *kbuf, void __user *ubuf)
1034 {
1035         if (kbuf) {
1036                 compat_ulong_t *k = kbuf;
1037                 while (count > 0) {
1038                         getreg32(target, pos, k++);
1039                         count -= sizeof(*k);
1040                         pos += sizeof(*k);
1041                 }
1042         } else {
1043                 compat_ulong_t __user *u = ubuf;
1044                 while (count > 0) {
1045                         compat_ulong_t word;
1046                         getreg32(target, pos, &word);
1047                         if (__put_user(word, u++))
1048                                 return -EFAULT;
1049                         count -= sizeof(*u);
1050                         pos += sizeof(*u);
1051                 }
1052         }
1053
1054         return 0;
1055 }
1056
1057 static int genregs32_set(struct task_struct *target,
1058                          const struct user_regset *regset,
1059                          unsigned int pos, unsigned int count,
1060                          const void *kbuf, const void __user *ubuf)
1061 {
1062         int ret = 0;
1063         if (kbuf) {
1064                 const compat_ulong_t *k = kbuf;
1065                 while (count > 0 && !ret) {
1066                         ret = putreg(target, pos, *k++);
1067                         count -= sizeof(*k);
1068                         pos += sizeof(*k);
1069                 }
1070         } else {
1071                 const compat_ulong_t __user *u = ubuf;
1072                 while (count > 0 && !ret) {
1073                         compat_ulong_t word;
1074                         ret = __get_user(word, u++);
1075                         if (ret)
1076                                 break;
1077                         ret = putreg(target, pos, word);
1078                         count -= sizeof(*u);
1079                         pos += sizeof(*u);
1080                 }
1081         }
1082         return ret;
1083 }
1084
1085 static long ptrace32_siginfo(unsigned request, u32 pid, u32 addr, u32 data)
1086 {
1087         siginfo_t __user *si = compat_alloc_user_space(sizeof(siginfo_t));
1088         compat_siginfo_t __user *si32 = compat_ptr(data);
1089         siginfo_t ssi;
1090         int ret;
1091
1092         if (request == PTRACE_SETSIGINFO) {
1093                 memset(&ssi, 0, sizeof(siginfo_t));
1094                 ret = copy_siginfo_from_user32(&ssi, si32);
1095                 if (ret)
1096                         return ret;
1097                 if (copy_to_user(si, &ssi, sizeof(siginfo_t)))
1098                         return -EFAULT;
1099         }
1100         ret = sys_ptrace(request, pid, addr, (unsigned long)si);
1101         if (ret)
1102                 return ret;
1103         if (request == PTRACE_GETSIGINFO) {
1104                 if (copy_from_user(&ssi, si, sizeof(siginfo_t)))
1105                         return -EFAULT;
1106                 ret = copy_siginfo_to_user32(si32, &ssi);
1107         }
1108         return ret;
1109 }
1110
1111 asmlinkage long sys32_ptrace(long request, u32 pid, u32 addr, u32 data)
1112 {
1113         struct task_struct *child;
1114         struct pt_regs *childregs;
1115         void __user *datap = compat_ptr(data);
1116         int ret;
1117         __u32 val;
1118
1119         switch (request) {
1120         case PTRACE_TRACEME:
1121         case PTRACE_ATTACH:
1122         case PTRACE_KILL:
1123         case PTRACE_CONT:
1124         case PTRACE_SINGLESTEP:
1125         case PTRACE_SINGLEBLOCK:
1126         case PTRACE_DETACH:
1127         case PTRACE_SYSCALL:
1128         case PTRACE_OLDSETOPTIONS:
1129         case PTRACE_SETOPTIONS:
1130         case PTRACE_SET_THREAD_AREA:
1131         case PTRACE_GET_THREAD_AREA:
1132         case PTRACE_BTS_CONFIG:
1133         case PTRACE_BTS_STATUS:
1134         case PTRACE_BTS_SIZE:
1135         case PTRACE_BTS_GET:
1136         case PTRACE_BTS_CLEAR:
1137         case PTRACE_BTS_DRAIN:
1138                 return sys_ptrace(request, pid, addr, data);
1139
1140         default:
1141                 return -EINVAL;
1142
1143         case PTRACE_PEEKTEXT:
1144         case PTRACE_PEEKDATA:
1145         case PTRACE_POKEDATA:
1146         case PTRACE_POKETEXT:
1147         case PTRACE_POKEUSR:
1148         case PTRACE_PEEKUSR:
1149         case PTRACE_GETREGS:
1150         case PTRACE_SETREGS:
1151         case PTRACE_SETFPREGS:
1152         case PTRACE_GETFPREGS:
1153         case PTRACE_SETFPXREGS:
1154         case PTRACE_GETFPXREGS:
1155         case PTRACE_GETEVENTMSG:
1156                 break;
1157
1158         case PTRACE_SETSIGINFO:
1159         case PTRACE_GETSIGINFO:
1160                 return ptrace32_siginfo(request, pid, addr, data);
1161         }
1162
1163         child = ptrace_get_task_struct(pid);
1164         if (IS_ERR(child))
1165                 return PTR_ERR(child);
1166
1167         ret = ptrace_check_attach(child, request == PTRACE_KILL);
1168         if (ret < 0)
1169                 goto out;
1170
1171         childregs = task_pt_regs(child);
1172
1173         switch (request) {
1174         case PTRACE_PEEKUSR:
1175                 ret = getreg32(child, addr, &val);
1176                 if (ret == 0)
1177                         ret = put_user(val, (__u32 __user *)datap);
1178                 break;
1179
1180         case PTRACE_POKEUSR:
1181                 ret = putreg32(child, addr, data);
1182                 break;
1183
1184         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1185                 return copy_regset_to_user(child, &user_x86_32_view,
1186                                            REGSET_GENERAL,
1187                                            0, sizeof(struct user_regs_struct32),
1188                                            datap);
1189
1190         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1191                 return copy_regset_from_user(child, &user_x86_32_view,
1192                                              REGSET_GENERAL, 0,
1193                                              sizeof(struct user_regs_struct32),
1194                                              datap);
1195
1196         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1197                 return copy_regset_to_user(child, &user_x86_32_view,
1198                                            REGSET_FP, 0,
1199                                            sizeof(struct user_i387_ia32_struct),
1200                                            datap);
1201
1202         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1203                 return copy_regset_from_user(
1204                         child, &user_x86_32_view, REGSET_FP,
1205                         0, sizeof(struct user_i387_ia32_struct), datap);
1206
1207         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1208                 return copy_regset_to_user(child, &user_x86_32_view,
1209                                            REGSET_XFP, 0,
1210                                            sizeof(struct user32_fxsr_struct),
1211                                            datap);
1212
1213         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1214                 return copy_regset_from_user(child, &user_x86_32_view,
1215                                              REGSET_XFP, 0,
1216                                              sizeof(struct user32_fxsr_struct),
1217                                              datap);
1218
1219         default:
1220                 return compat_ptrace_request(child, request, addr, data);
1221         }
1222
1223  out:
1224         put_task_struct(child);
1225         return ret;
1226 }
1227
1228 #endif  /* CONFIG_IA32_EMULATION */
1229
1230 #ifdef CONFIG_X86_64
1231
1232 static const struct user_regset x86_64_regsets[] = {
1233         [REGSET_GENERAL] = {
1234                 .core_note_type = NT_PRSTATUS,
1235                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1236                 .size = sizeof(long), .align = sizeof(long),
1237                 .get = genregs_get, .set = genregs_set
1238         },
1239         [REGSET_FP] = {
1240                 .core_note_type = NT_PRFPREG,
1241                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1242                 .size = sizeof(long), .align = sizeof(long),
1243                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1244         },
1245 };
1246
1247 static const struct user_regset_view user_x86_64_view = {
1248         .name = "x86_64", .e_machine = EM_X86_64,
1249         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1250 };
1251
1252 #else  /* CONFIG_X86_32 */
1253
1254 #define user_regs_struct32      user_regs_struct
1255 #define genregs32_get           genregs_get
1256 #define genregs32_set           genregs_set
1257
1258 #endif  /* CONFIG_X86_64 */
1259
1260 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1261 static const struct user_regset x86_32_regsets[] = {
1262         [REGSET_GENERAL] = {
1263                 .core_note_type = NT_PRSTATUS,
1264                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1265                 .size = sizeof(u32), .align = sizeof(u32),
1266                 .get = genregs32_get, .set = genregs32_set
1267         },
1268         [REGSET_FP] = {
1269                 .core_note_type = NT_PRFPREG,
1270                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1271                 .size = sizeof(u32), .align = sizeof(u32),
1272                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1273         },
1274         [REGSET_XFP] = {
1275                 .core_note_type = NT_PRXFPREG,
1276                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1277                 .size = sizeof(u32), .align = sizeof(u32),
1278                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1279         },
1280         [REGSET_TLS] = {
1281                 .core_note_type = NT_386_TLS,
1282                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1283                 .size = sizeof(struct user_desc),
1284                 .align = sizeof(struct user_desc),
1285                 .active = regset_tls_active,
1286                 .get = regset_tls_get, .set = regset_tls_set
1287         },
1288 };
1289
1290 static const struct user_regset_view user_x86_32_view = {
1291         .name = "i386", .e_machine = EM_386,
1292         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1293 };
1294 #endif
1295
1296 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1297 {
1298 #ifdef CONFIG_IA32_EMULATION
1299         if (test_tsk_thread_flag(task, TIF_IA32))
1300 #endif
1301 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1302                 return &user_x86_32_view;
1303 #endif
1304 #ifdef CONFIG_X86_64
1305         return &user_x86_64_view;
1306 #endif
1307 }
1308
1309 #ifdef CONFIG_X86_32
1310
1311 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1312 {
1313         struct siginfo info;
1314
1315         tsk->thread.trap_no = 1;
1316         tsk->thread.error_code = error_code;
1317
1318         memset(&info, 0, sizeof(info));
1319         info.si_signo = SIGTRAP;
1320         info.si_code = TRAP_BRKPT;
1321
1322         /* User-mode ip? */
1323         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1324
1325         /* Send us the fake SIGTRAP */
1326         force_sig_info(SIGTRAP, &info, tsk);
1327 }
1328
1329 /* notification of system call entry/exit
1330  * - triggered by current->work.syscall_trace
1331  */
1332 __attribute__((regparm(3)))
1333 int do_syscall_trace(struct pt_regs *regs, int entryexit)
1334 {
1335         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
1336         /*
1337          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
1338          * interception
1339          */
1340         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
1341         int ret = 0;
1342
1343         /* do the secure computing check first */
1344         if (!entryexit)
1345                 secure_computing(regs->orig_ax);
1346
1347         if (unlikely(current->audit_context)) {
1348                 if (entryexit)
1349                         audit_syscall_exit(AUDITSC_RESULT(regs->ax),
1350                                                 regs->ax);
1351                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
1352                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
1353                  * not used, entry.S will call us only on syscall exit, not
1354                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
1355                  * calling send_sigtrap() on syscall entry.
1356                  *
1357                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
1358                  * is_singlestep is false, despite his name, so we will still do
1359                  * the correct thing.
1360                  */
1361                 else if (is_singlestep)
1362                         goto out;
1363         }
1364
1365         if (!(current->ptrace & PT_PTRACED))
1366                 goto out;
1367
1368         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
1369          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
1370          * here. We have to check this and return */
1371         if (is_sysemu && entryexit)
1372                 return 0;
1373
1374         /* Fake a debug trap */
1375         if (is_singlestep)
1376                 send_sigtrap(current, regs, 0);
1377
1378         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
1379                 goto out;
1380
1381         /* the 0x80 provides a way for the tracing parent to distinguish
1382            between a syscall stop and SIGTRAP delivery */
1383         /* Note that the debugger could change the result of test_thread_flag!*/
1384         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1385
1386         /*
1387          * this isn't the same as continuing with a signal, but it will do
1388          * for normal use.  strace only continues with a signal if the
1389          * stopping signal is not SIGTRAP.  -brl
1390          */
1391         if (current->exit_code) {
1392                 send_sig(current->exit_code, current, 1);
1393                 current->exit_code = 0;
1394         }
1395         ret = is_sysemu;
1396 out:
1397         if (unlikely(current->audit_context) && !entryexit)
1398                 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
1399                                     regs->bx, regs->cx, regs->dx, regs->si);
1400         if (ret == 0)
1401                 return 0;
1402
1403         regs->orig_ax = -1; /* force skip of syscall restarting */
1404         if (unlikely(current->audit_context))
1405                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1406         return 1;
1407 }
1408
1409 #else  /* CONFIG_X86_64 */
1410
1411 static void syscall_trace(struct pt_regs *regs)
1412 {
1413
1414 #if 0
1415         printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
1416                current->comm,
1417                regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0),
1418                current_thread_info()->flags, current->ptrace);
1419 #endif
1420
1421         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
1422                                 ? 0x80 : 0));
1423         /*
1424          * this isn't the same as continuing with a signal, but it will do
1425          * for normal use.  strace only continues with a signal if the
1426          * stopping signal is not SIGTRAP.  -brl
1427          */
1428         if (current->exit_code) {
1429                 send_sig(current->exit_code, current, 1);
1430                 current->exit_code = 0;
1431         }
1432 }
1433
1434 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
1435 {
1436         /* do the secure computing check first */
1437         secure_computing(regs->orig_ax);
1438
1439         if (test_thread_flag(TIF_SYSCALL_TRACE)
1440             && (current->ptrace & PT_PTRACED))
1441                 syscall_trace(regs);
1442
1443         if (unlikely(current->audit_context)) {
1444                 if (test_thread_flag(TIF_IA32)) {
1445                         audit_syscall_entry(AUDIT_ARCH_I386,
1446                                             regs->orig_ax,
1447                                             regs->bx, regs->cx,
1448                                             regs->dx, regs->si);
1449                 } else {
1450                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1451                                             regs->orig_ax,
1452                                             regs->di, regs->si,
1453                                             regs->dx, regs->r10);
1454                 }
1455         }
1456 }
1457
1458 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
1459 {
1460         if (unlikely(current->audit_context))
1461                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1462
1463         if ((test_thread_flag(TIF_SYSCALL_TRACE)
1464              || test_thread_flag(TIF_SINGLESTEP))
1465             && (current->ptrace & PT_PTRACED))
1466                 syscall_trace(regs);
1467 }
1468
1469 #endif  /* CONFIG_X86_32 */