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_TLS,
  44 };
  45
  46 /*
  47  * does not yet catch signals sent when the child dies.
  48  * in exit.c or in signal.c.
  49  */
  50
  51 /*
  52  * Determines which flags the user has access to [1 = access, 0 = no access].
  53  */
  54 #define FLAG_MASK_32            ((unsigned long)                        \
  55                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  56                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  57                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  58                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  59                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  60
  61 /*
  62  * Determines whether a value may be installed in a segment register.
  63  */
  64 static inline bool invalid_selector(u16 value)
  65 {
  66         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  67 }
  68
  69 #ifdef CONFIG_X86_32
  70
  71 #define FLAG_MASK               FLAG_MASK_32
  72
  73 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  74 {
  75         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  76         regno >>= 2;
  77         if (regno > FS)
  78                 --regno;
  79         return &regs->bx + regno;
  80 }
  81
  82 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  83 {
  84         /*
  85          * Returning the value truncates it to 16 bits.
  86          */
  87         unsigned int retval;
  88         if (offset != offsetof(struct user_regs_struct, gs))
  89                 retval = *pt_regs_access(task_pt_regs(task), offset);
  90         else {
  91                 retval = task->thread.gs;
  92                 if (task == current)
  93                         savesegment(gs, retval);
  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                 task->thread.gs = value;
 128                 if (task == current)
 129                         /*
 130                          * The user-mode %gs is not affected by
 131                          * kernel entry, so we must update the CPU.
 132                          */
 133                         loadsegment(gs, value);
 134         }
 135
 136         return 0;
 137 }
 138
 139 static unsigned long debugreg_addr_limit(struct task_struct *task)
 140 {
 141         return TASK_SIZE - 3;
 142 }
 143
 144 #else  /* CONFIG_X86_64 */
 145
 146 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 147
 148 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 149 {
 150         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 151         return &regs->r15 + (offset / sizeof(regs->r15));
 152 }
 153
 154 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 155 {
 156         /*
 157          * Returning the value truncates it to 16 bits.
 158          */
 159         unsigned int seg;
 160
 161         switch (offset) {
 162         case offsetof(struct user_regs_struct, fs):
 163                 if (task == current) {
 164                         /* Older gas can't assemble movq %?s,%r?? */
 165                         asm("movl %%fs,%0" : "=r" (seg));
 166                         return seg;
 167                 }
 168                 return task->thread.fsindex;
 169         case offsetof(struct user_regs_struct, gs):
 170                 if (task == current) {
 171                         asm("movl %%gs,%0" : "=r" (seg));
 172                         return seg;
 173                 }
 174                 return task->thread.gsindex;
 175         case offsetof(struct user_regs_struct, ds):
 176                 if (task == current) {
 177                         asm("movl %%ds,%0" : "=r" (seg));
 178                         return seg;
 179                 }
 180                 return task->thread.ds;
 181         case offsetof(struct user_regs_struct, es):
 182                 if (task == current) {
 183                         asm("movl %%es,%0" : "=r" (seg));
 184                         return seg;
 185                 }
 186                 return task->thread.es;
 187
 188         case offsetof(struct user_regs_struct, cs):
 189         case offsetof(struct user_regs_struct, ss):
 190                 break;
 191         }
 192         return *pt_regs_access(task_pt_regs(task), offset);
 193 }
 194
 195 static int set_segment_reg(struct task_struct *task,
 196                            unsigned long offset, u16 value)
 197 {
 198         /*
 199          * The value argument was already truncated to 16 bits.
 200          */
 201         if (invalid_selector(value))
 202                 return -EIO;
 203
 204         switch (offset) {
 205         case offsetof(struct user_regs_struct,fs):
 206                 /*
 207                  * If this is setting fs as for normal 64-bit use but
 208                  * setting fs_base has implicitly changed it, leave it.
 209                  */
 210                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 211                      task->thread.fs != 0) ||
 212                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 213                      task->thread.fs == 0))
 214                         break;
 215                 task->thread.fsindex = value;
 216                 if (task == current)
 217                         loadsegment(fs, task->thread.fsindex);
 218                 break;
 219         case offsetof(struct user_regs_struct,gs):
 220                 /*
 221                  * If this is setting gs as for normal 64-bit use but
 222                  * setting gs_base has implicitly changed it, leave it.
 223                  */
 224                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 225                      task->thread.gs != 0) ||
 226                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 227                      task->thread.gs == 0))
 228                         break;
 229                 task->thread.gsindex = value;
 230                 if (task == current)
 231                         load_gs_index(task->thread.gsindex);
 232                 break;
 233         case offsetof(struct user_regs_struct,ds):
 234                 task->thread.ds = value;
 235                 if (task == current)
 236                         loadsegment(ds, task->thread.ds);
 237                 break;
 238         case offsetof(struct user_regs_struct,es):
 239                 task->thread.es = value;
 240                 if (task == current)
 241                         loadsegment(es, task->thread.es);
 242                 break;
 243
 244                 /*
 245                  * Can't actually change these in 64-bit mode.
 246                  */
 247         case offsetof(struct user_regs_struct,cs):
 248                 if (unlikely(value == 0))
 249                         return -EIO;
 250 #ifdef CONFIG_IA32_EMULATION
 251                 if (test_tsk_thread_flag(task, TIF_IA32))
 252                         task_pt_regs(task)->cs = value;
 253 #endif
 254                 break;
 255         case offsetof(struct user_regs_struct,ss):
 256                 if (unlikely(value == 0))
 257                         return -EIO;
 258 #ifdef CONFIG_IA32_EMULATION
 259                 if (test_tsk_thread_flag(task, TIF_IA32))
 260                         task_pt_regs(task)->ss = value;
 261 #endif
 262                 break;
 263         }
 264
 265         return 0;
 266 }
 267
 268 static unsigned long debugreg_addr_limit(struct task_struct *task)
 269 {
 270 #ifdef CONFIG_IA32_EMULATION
 271         if (test_tsk_thread_flag(task, TIF_IA32))
 272                 return IA32_PAGE_OFFSET - 3;
 273 #endif
 274         return TASK_SIZE64 - 7;
 275 }
 276
 277 #endif  /* CONFIG_X86_32 */
 278
 279 static unsigned long get_flags(struct task_struct *task)
 280 {
 281         unsigned long retval = task_pt_regs(task)->flags;
 282
 283         /*
 284          * If the debugger set TF, hide it from the readout.
 285          */
 286         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 287                 retval &= ~X86_EFLAGS_TF;
 288
 289         return retval;
 290 }
 291
 292 static int set_flags(struct task_struct *task, unsigned long value)
 293 {
 294         struct pt_regs *regs = task_pt_regs(task);
 295
 296         /*
 297          * If the user value contains TF, mark that
 298          * it was not "us" (the debugger) that set it.
 299          * If not, make sure it stays set if we had.
 300          */
 301         if (value & X86_EFLAGS_TF)
 302                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 303         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 304                 value |= X86_EFLAGS_TF;
 305
 306         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 307
 308         return 0;
 309 }
 310
 311 static int putreg(struct task_struct *child,
 312                   unsigned long offset, unsigned long value)
 313 {
 314         switch (offset) {
 315         case offsetof(struct user_regs_struct, cs):
 316         case offsetof(struct user_regs_struct, ds):
 317         case offsetof(struct user_regs_struct, es):
 318         case offsetof(struct user_regs_struct, fs):
 319         case offsetof(struct user_regs_struct, gs):
 320         case offsetof(struct user_regs_struct, ss):
 321                 return set_segment_reg(child, offset, value);
 322
 323         case offsetof(struct user_regs_struct, flags):
 324                 return set_flags(child, value);
 325
 326 #ifdef CONFIG_X86_64
 327         /*
 328          * Orig_ax is really just a flag with small positive and
 329          * negative values, so make sure to always sign-extend it
 330          * from 32 bits so that it works correctly regardless of
 331          * whether we come from a 32-bit environment or not.
 332          */
 333         case offsetof(struct user_regs_struct, orig_ax):
 334                 value = (long) (s32) value;
 335                 break;
 336
 337         case offsetof(struct user_regs_struct,fs_base):
 338                 if (value >= TASK_SIZE_OF(child))
 339                         return -EIO;
 340                 /*
 341                  * When changing the segment base, use do_arch_prctl
 342                  * to set either thread.fs or thread.fsindex and the
 343                  * corresponding GDT slot.
 344                  */
 345                 if (child->thread.fs != value)
 346                         return do_arch_prctl(child, ARCH_SET_FS, value);
 347                 return 0;
 348         case offsetof(struct user_regs_struct,gs_base):
 349                 /*
 350                  * Exactly the same here as the %fs handling above.
 351                  */
 352                 if (value >= TASK_SIZE_OF(child))
 353                         return -EIO;
 354                 if (child->thread.gs != value)
 355                         return do_arch_prctl(child, ARCH_SET_GS, value);
 356                 return 0;
 357 #endif
 358         }
 359
 360         *pt_regs_access(task_pt_regs(child), offset) = value;
 361         return 0;
 362 }
 363
 364 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 365 {
 366         switch (offset) {
 367         case offsetof(struct user_regs_struct, cs):
 368         case offsetof(struct user_regs_struct, ds):
 369         case offsetof(struct user_regs_struct, es):
 370         case offsetof(struct user_regs_struct, fs):
 371         case offsetof(struct user_regs_struct, gs):
 372         case offsetof(struct user_regs_struct, ss):
 373                 return get_segment_reg(task, offset);
 374
 375         case offsetof(struct user_regs_struct, flags):
 376                 return get_flags(task);
 377
 378 #ifdef CONFIG_X86_64
 379         case offsetof(struct user_regs_struct, fs_base): {
 380                 /*
 381                  * do_arch_prctl may have used a GDT slot instead of
 382                  * the MSR.  To userland, it appears the same either
 383                  * way, except the %fs segment selector might not be 0.
 384                  */
 385                 unsigned int seg = task->thread.fsindex;
 386                 if (task->thread.fs != 0)
 387                         return task->thread.fs;
 388                 if (task == current)
 389                         asm("movl %%fs,%0" : "=r" (seg));
 390                 if (seg != FS_TLS_SEL)
 391                         return 0;
 392                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 393         }
 394         case offsetof(struct user_regs_struct, gs_base): {
 395                 /*
 396                  * Exactly the same here as the %fs handling above.
 397                  */
 398                 unsigned int seg = task->thread.gsindex;
 399                 if (task->thread.gs != 0)
 400                         return task->thread.gs;
 401                 if (task == current)
 402                         asm("movl %%gs,%0" : "=r" (seg));
 403                 if (seg != GS_TLS_SEL)
 404                         return 0;
 405                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 406         }
 407 #endif
 408         }
 409
 410         return *pt_regs_access(task_pt_regs(task), offset);
 411 }
 412
 413 static int genregs_get(struct task_struct *target,
 414                        const struct user_regset *regset,
 415                        unsigned int pos, unsigned int count,
 416                        void *kbuf, void __user *ubuf)
 417 {
 418         if (kbuf) {
 419                 unsigned long *k = kbuf;
 420                 while (count > 0) {
 421                         *k++ = getreg(target, pos);
 422                         count -= sizeof(*k);
 423                         pos += sizeof(*k);
 424                 }
 425         } else {
 426                 unsigned long __user *u = ubuf;
 427                 while (count > 0) {
 428                         if (__put_user(getreg(target, pos), u++))
 429                                 return -EFAULT;
 430                         count -= sizeof(*u);
 431                         pos += sizeof(*u);
 432                 }
 433         }
 434
 435         return 0;
 436 }
 437
 438 static int genregs_set(struct task_struct *target,
 439                        const struct user_regset *regset,
 440                        unsigned int pos, unsigned int count,
 441                        const void *kbuf, const void __user *ubuf)
 442 {
 443         int ret = 0;
 444         if (kbuf) {
 445                 const unsigned long *k = kbuf;
 446                 while (count > 0 && !ret) {
 447                         ret = putreg(target, pos, *k++);
 448                         count -= sizeof(*k);
 449                         pos += sizeof(*k);
 450                 }
 451         } else {
 452                 const unsigned long  __user *u = ubuf;
 453                 while (count > 0 && !ret) {
 454                         unsigned long word;
 455                         ret = __get_user(word, u++);
 456                         if (ret)
 457                                 break;
 458                         ret = putreg(target, pos, word);
 459                         count -= sizeof(*u);
 460                         pos += sizeof(*u);
 461                 }
 462         }
 463         return ret;
 464 }
 465
 466 /*
 467  * This function is trivial and will be inlined by the compiler.
 468  * Having it separates the implementation details of debug
 469  * registers from the interface details of ptrace.
 470  */
 471 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 472 {
 473         switch (n) {
 474         case 0:         return child->thread.debugreg0;
 475         case 1:         return child->thread.debugreg1;
 476         case 2:         return child->thread.debugreg2;
 477         case 3:         return child->thread.debugreg3;
 478         case 6:         return child->thread.debugreg6;
 479         case 7:         return child->thread.debugreg7;
 480         }
 481         return 0;
 482 }
 483
 484 static int ptrace_set_debugreg(struct task_struct *child,
 485                                int n, unsigned long data)
 486 {
 487         int i;
 488
 489         if (unlikely(n == 4 || n == 5))
 490                 return -EIO;
 491
 492         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 493                 return -EIO;
 494
 495         switch (n) {
 496         case 0:         child->thread.debugreg0 = data; break;
 497         case 1:         child->thread.debugreg1 = data; break;
 498         case 2:         child->thread.debugreg2 = data; break;
 499         case 3:         child->thread.debugreg3 = data; break;
 500
 501         case 6:
 502                 if ((data & ~0xffffffffUL) != 0)
 503                         return -EIO;
 504                 child->thread.debugreg6 = data;
 505                 break;
 506
 507         case 7:
 508                 /*
 509                  * Sanity-check data. Take one half-byte at once with
 510                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 511                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 512                  * 2 and 3 are LENi. Given a list of invalid values,
 513                  * we do mask |= 1 << invalid_value, so that
 514                  * (mask >> check) & 1 is a correct test for invalid
 515                  * values.
 516                  *
 517                  * R/Wi contains the type of the breakpoint /
 518                  * watchpoint, LENi contains the length of the watched
 519                  * data in the watchpoint case.
 520                  *
 521                  * The invalid values are:
 522                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 523                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 524                  *   mask |= 0x4444.
 525                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 526                  *   0x1110.
 527                  *
 528                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 529                  *
 530                  * See the Intel Manual "System Programming Guide",
 531                  * 15.2.4
 532                  *
 533                  * Note that LENi == 0x10 is defined on x86_64 in long
 534                  * mode (i.e. even for 32-bit userspace software, but
 535                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 536                  * See the AMD manual no. 24593 (AMD64 System Programming)
 537                  */
 538 #ifdef CONFIG_X86_32
 539 #define DR7_MASK        0x5f54
 540 #else
 541 #define DR7_MASK        0x5554
 542 #endif
 543                 data &= ~DR_CONTROL_RESERVED;
 544                 for (i = 0; i < 4; i++)
 545                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 546                                 return -EIO;
 547                 child->thread.debugreg7 = data;
 548                 if (data)
 549                         set_tsk_thread_flag(child, TIF_DEBUG);
 550                 else
 551                         clear_tsk_thread_flag(child, TIF_DEBUG);
 552                 break;
 553         }
 554
 555         return 0;
 556 }
 557
 558 #ifdef X86_BTS
 559
 560 static int ptrace_bts_get_size(struct task_struct *child)
 561 {
 562         if (!child->thread.ds_area_msr)
 563                 return -ENXIO;
 564
 565         return ds_get_bts_index((void *)child->thread.ds_area_msr);
 566 }
 567
 568 static int ptrace_bts_read_record(struct task_struct *child,
 569                                   long index,
 570                                   struct bts_struct __user *out)
 571 {
 572         struct bts_struct ret;
 573         int retval;
 574         int bts_end;
 575         int bts_index;
 576
 577         if (!child->thread.ds_area_msr)
 578                 return -ENXIO;
 579
 580         if (index < 0)
 581                 return -EINVAL;
 582
 583         bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
 584         if (bts_end <= index)
 585                 return -EINVAL;
 586
 587         /* translate the ptrace bts index into the ds bts index */
 588         bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
 589         bts_index -= (index + 1);
 590         if (bts_index < 0)
 591                 bts_index += bts_end;
 592
 593         retval = ds_read_bts((void *)child->thread.ds_area_msr,
 594                              bts_index, &ret);
 595         if (retval < 0)
 596                 return retval;
 597
 598         if (copy_to_user(out, &ret, sizeof(ret)))
 599                 return -EFAULT;
 600
 601         return sizeof(ret);
 602 }
 603
 604 static int ptrace_bts_clear(struct task_struct *child)
 605 {
 606         if (!child->thread.ds_area_msr)
 607                 return -ENXIO;
 608
 609         return ds_clear((void *)child->thread.ds_area_msr);
 610 }
 611
 612 static int ptrace_bts_drain(struct task_struct *child,
 613                             long size,
 614                             struct bts_struct __user *out)
 615 {
 616         int end, i;
 617         void *ds = (void *)child->thread.ds_area_msr;
 618
 619         if (!ds)
 620                 return -ENXIO;
 621
 622         end = ds_get_bts_index(ds);
 623         if (end <= 0)
 624                 return end;
 625
 626         if (size < (end * sizeof(struct bts_struct)))
 627                 return -EIO;
 628
 629         for (i = 0; i < end; i++, out++) {
 630                 struct bts_struct ret;
 631                 int retval;
 632
 633                 retval = ds_read_bts(ds, i, &ret);
 634                 if (retval < 0)
 635                         return retval;
 636
 637                 if (copy_to_user(out, &ret, sizeof(ret)))
 638                         return -EFAULT;
 639         }
 640
 641         ds_clear(ds);
 642
 643         return end;
 644 }
 645
 646 static int ptrace_bts_config(struct task_struct *child,
 647                              long cfg_size,
 648                              const struct ptrace_bts_config __user *ucfg)
 649 {
 650         struct ptrace_bts_config cfg;
 651         int bts_size, ret = 0;
 652         void *ds;
 653
 654         if (cfg_size < sizeof(cfg))
 655                 return -EIO;
 656
 657         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 658                 return -EFAULT;
 659
 660         if ((int)cfg.size < 0)
 661                 return -EINVAL;
 662
 663         bts_size = 0;
 664         ds = (void *)child->thread.ds_area_msr;
 665         if (ds) {
 666                 bts_size = ds_get_bts_size(ds);
 667                 if (bts_size < 0)
 668                         return bts_size;
 669         }
 670         cfg.size = PAGE_ALIGN(cfg.size);
 671
 672         if (bts_size != cfg.size) {
 673                 ret = ptrace_bts_realloc(child, cfg.size,
 674                                          cfg.flags & PTRACE_BTS_O_CUT_SIZE);
 675                 if (ret < 0)
 676                         goto errout;
 677
 678                 ds = (void *)child->thread.ds_area_msr;
 679         }
 680
 681         if (cfg.flags & PTRACE_BTS_O_SIGNAL)
 682                 ret = ds_set_overflow(ds, DS_O_SIGNAL);
 683         else
 684                 ret = ds_set_overflow(ds, DS_O_WRAP);
 685         if (ret < 0)
 686                 goto errout;
 687
 688         if (cfg.flags & PTRACE_BTS_O_TRACE)
 689                 child->thread.debugctlmsr |= ds_debugctl_mask();
 690         else
 691                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 692
 693         if (cfg.flags & PTRACE_BTS_O_SCHED)
 694                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 695         else
 696                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 697
 698         ret = sizeof(cfg);
 699
 700 out:
 701         if (child->thread.debugctlmsr)
 702                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 703         else
 704                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 705
 706         return ret;
 707
 708 errout:
 709         child->thread.debugctlmsr &= ~ds_debugctl_mask();
 710         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 711         goto out;
 712 }
 713
 714 static int ptrace_bts_status(struct task_struct *child,
 715                              long cfg_size,
 716                              struct ptrace_bts_config __user *ucfg)
 717 {
 718         void *ds = (void *)child->thread.ds_area_msr;
 719         struct ptrace_bts_config cfg;
 720
 721         if (cfg_size < sizeof(cfg))
 722                 return -EIO;
 723
 724         memset(&cfg, 0, sizeof(cfg));
 725
 726         if (ds) {
 727                 cfg.size = ds_get_bts_size(ds);
 728
 729                 if (ds_get_overflow(ds) == DS_O_SIGNAL)
 730                         cfg.flags |= PTRACE_BTS_O_SIGNAL;
 731
 732                 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
 733                     child->thread.debugctlmsr & ds_debugctl_mask())
 734                         cfg.flags |= PTRACE_BTS_O_TRACE;
 735
 736                 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
 737                         cfg.flags |= PTRACE_BTS_O_SCHED;
 738         }
 739
 740         cfg.bts_size = sizeof(struct bts_struct);
 741
 742         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 743                 return -EFAULT;
 744
 745         return sizeof(cfg);
 746 }
 747
 748
 749 static int ptrace_bts_write_record(struct task_struct *child,
 750                                    const struct bts_struct *in)
 751 {
 752         int retval;
 753
 754         if (!child->thread.ds_area_msr)
 755                 return -ENXIO;
 756
 757         retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
 758         if (retval)
 759                 return retval;
 760
 761         return sizeof(*in);
 762 }
 763
 764 static int ptrace_bts_realloc(struct task_struct *child,
 765                               int size, int reduce_size)
 766 {
 767         unsigned long rlim, vm;
 768         int ret, old_size;
 769
 770         if (size < 0)
 771                 return -EINVAL;
 772
 773         old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
 774         if (old_size < 0)
 775                 return old_size;
 776
 777         ret = ds_free((void **)&child->thread.ds_area_msr);
 778         if (ret < 0)
 779                 goto out;
 780
 781         size >>= PAGE_SHIFT;
 782         old_size >>= PAGE_SHIFT;
 783
 784         current->mm->total_vm  -= old_size;
 785         current->mm->locked_vm -= old_size;
 786
 787         if (size == 0)
 788                 goto out;
 789
 790         rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
 791         vm = current->mm->total_vm  + size;
 792         if (rlim < vm) {
 793                 ret = -ENOMEM;
 794
 795                 if (!reduce_size)
 796                         goto out;
 797
 798                 size = rlim - current->mm->total_vm;
 799                 if (size <= 0)
 800                         goto out;
 801         }
 802
 803         rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
 804         vm = current->mm->locked_vm  + size;
 805         if (rlim < vm) {
 806                 ret = -ENOMEM;
 807
 808                 if (!reduce_size)
 809                         goto out;
 810
 811                 size = rlim - current->mm->locked_vm;
 812                 if (size <= 0)
 813                         goto out;
 814         }
 815
 816         ret = ds_allocate((void **)&child->thread.ds_area_msr,
 817                           size << PAGE_SHIFT);
 818         if (ret < 0)
 819                 goto out;
 820
 821         current->mm->total_vm  += size;
 822         current->mm->locked_vm += size;
 823
 824 out:
 825         if (child->thread.ds_area_msr)
 826                 set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 827         else
 828                 clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 829
 830         return ret;
 831 }
 832
 833 void ptrace_bts_take_timestamp(struct task_struct *tsk,
 834                                enum bts_qualifier qualifier)
 835 {
 836         struct bts_struct rec = {
 837                 .qualifier = qualifier,
 838                 .variant.jiffies = jiffies_64
 839         };
 840
 841         ptrace_bts_write_record(tsk, &rec);
 842 }
 843 #endif /* X86_BTS */
 844
 845 /*
 846  * Called by kernel/ptrace.c when detaching..
 847  *
 848  * Make sure the single step bit is not set.
 849  */
 850 void ptrace_disable(struct task_struct *child)
 851 {
 852         user_disable_single_step(child);
 853 #ifdef TIF_SYSCALL_EMU
 854         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 855 #endif
 856         if (child->thread.ds_area_msr) {
 857 #ifdef X86_BTS
 858                 ptrace_bts_realloc(child, 0, 0);
 859 #endif
 860                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 861                 if (!child->thread.debugctlmsr)
 862                         clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 863                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 864         }
 865 }
 866
 867 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 868 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 869 #endif
 870
 871 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 872 {
 873         int ret;
 874         unsigned long __user *datap = (unsigned long __user *)data;
 875
 876         switch (request) {
 877         /* read the word at location addr in the USER area. */
 878         case PTRACE_PEEKUSR: {
 879                 unsigned long tmp;
 880
 881                 ret = -EIO;
 882                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 883                     addr >= sizeof(struct user))
 884                         break;
 885
 886                 tmp = 0;  /* Default return condition */
 887                 if (addr < sizeof(struct user_regs_struct))
 888                         tmp = getreg(child, addr);
 889                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 890                          addr <= offsetof(struct user, u_debugreg[7])) {
 891                         addr -= offsetof(struct user, u_debugreg[0]);
 892                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 893                 }
 894                 ret = put_user(tmp, datap);
 895                 break;
 896         }
 897
 898         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 899                 ret = -EIO;
 900                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 901                     addr >= sizeof(struct user))
 902                         break;
 903
 904                 if (addr < sizeof(struct user_regs_struct))
 905                         ret = putreg(child, addr, data);
 906                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 907                          addr <= offsetof(struct user, u_debugreg[7])) {
 908                         addr -= offsetof(struct user, u_debugreg[0]);
 909                         ret = ptrace_set_debugreg(child,
 910                                                   addr / sizeof(data), data);
 911                 }
 912                 break;
 913
 914         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 915                 return copy_regset_to_user(child,
 916                                            task_user_regset_view(current),
 917                                            REGSET_GENERAL,
 918                                            0, sizeof(struct user_regs_struct),
 919                                            datap);
 920
 921         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 922                 return copy_regset_from_user(child,
 923                                              task_user_regset_view(current),
 924                                              REGSET_GENERAL,
 925                                              0, sizeof(struct user_regs_struct),
 926                                              datap);
 927
 928         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 929                 return copy_regset_to_user(child,
 930                                            task_user_regset_view(current),
 931                                            REGSET_FP,
 932                                            0, sizeof(struct user_i387_struct),
 933                                            datap);
 934
 935         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 936                 return copy_regset_from_user(child,
 937                                              task_user_regset_view(current),
 938                                              REGSET_FP,
 939                                              0, sizeof(struct user_i387_struct),
 940                                              datap);
 941
 942 #ifdef CONFIG_X86_32
 943         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 944                 return copy_regset_to_user(child, &user_x86_32_view,
 945                                            REGSET_XFP,
 946                                            0, sizeof(struct user_fxsr_struct),
 947                                            datap) ? -EIO : 0;
 948
 949         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 950                 return copy_regset_from_user(child, &user_x86_32_view,
 951                                              REGSET_XFP,
 952                                              0, sizeof(struct user_fxsr_struct),
 953                                              datap) ? -EIO : 0;
 954 #endif
 955
 956 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 957         case PTRACE_GET_THREAD_AREA:
 958                 if (addr < 0)
 959                         return -EIO;
 960                 ret = do_get_thread_area(child, addr,
 961                                          (struct user_desc __user *) data);
 962                 break;
 963
 964         case PTRACE_SET_THREAD_AREA:
 965                 if (addr < 0)
 966                         return -EIO;
 967                 ret = do_set_thread_area(child, addr,
 968                                          (struct user_desc __user *) data, 0);
 969                 break;
 970 #endif
 971
 972 #ifdef CONFIG_X86_64
 973                 /* normal 64bit interface to access TLS data.
 974                    Works just like arch_prctl, except that the arguments
 975                    are reversed. */
 976         case PTRACE_ARCH_PRCTL:
 977                 ret = do_arch_prctl(child, data, addr);
 978                 break;
 979 #endif
 980
 981         /*
 982          * These bits need more cooking - not enabled yet:
 983          */
 984 #ifdef X86_BTS
 985         case PTRACE_BTS_CONFIG:
 986                 ret = ptrace_bts_config
 987                         (child, data, (struct ptrace_bts_config __user *)addr);
 988                 break;
 989
 990         case PTRACE_BTS_STATUS:
 991                 ret = ptrace_bts_status
 992                         (child, data, (struct ptrace_bts_config __user *)addr);
 993                 break;
 994
 995         case PTRACE_BTS_SIZE:
 996                 ret = ptrace_bts_get_size(child);
 997                 break;
 998
 999         case PTRACE_BTS_GET:
1000                 ret = ptrace_bts_read_record
1001                         (child, data, (struct bts_struct __user *) addr);
1002                 break;
1003
1004         case PTRACE_BTS_CLEAR:
1005                 ret = ptrace_bts_clear(child);
1006                 break;
1007
1008         case PTRACE_BTS_DRAIN:
1009                 ret = ptrace_bts_drain
1010                         (child, data, (struct bts_struct __user *) addr);
1011                 break;
1012 #endif
1013
1014         default:
1015                 ret = ptrace_request(child, request, addr, data);
1016                 break;
1017         }
1018
1019         return ret;
1020 }
1021
1022 #ifdef CONFIG_IA32_EMULATION
1023
1024 #include <linux/compat.h>
1025 #include <linux/syscalls.h>
1026 #include <asm/ia32.h>
1027 #include <asm/user32.h>
1028
1029 #define R32(l,q)                                                        \
1030         case offsetof(struct user32, regs.l):                           \
1031                 regs->q = value; break
1032
1033 #define SEG32(rs)                                                       \
1034         case offsetof(struct user32, regs.rs):                          \
1035                 return set_segment_reg(child,                           \
1036                                        offsetof(struct user_regs_struct, rs), \
1037                                        value);                          \
1038                 break
1039
1040 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1041 {
1042         struct pt_regs *regs = task_pt_regs(child);
1043
1044         switch (regno) {
1045
1046         SEG32(cs);
1047         SEG32(ds);
1048         SEG32(es);
1049         SEG32(fs);
1050         SEG32(gs);
1051         SEG32(ss);
1052
1053         R32(ebx, bx);
1054         R32(ecx, cx);
1055         R32(edx, dx);
1056         R32(edi, di);
1057         R32(esi, si);
1058         R32(ebp, bp);
1059         R32(eax, ax);
1060         R32(eip, ip);
1061         R32(esp, sp);
1062
1063         case offsetof(struct user32, regs.orig_eax):
1064                 /*
1065                  * Sign-extend the value so that orig_eax = -1
1066                  * causes (long)orig_ax < 0 tests to fire correctly.
1067                  */
1068                 regs->orig_ax = (long) (s32) value;
1069                 break;
1070
1071         case offsetof(struct user32, regs.eflags):
1072                 return set_flags(child, value);
1073
1074         case offsetof(struct user32, u_debugreg[0]) ...
1075                 offsetof(struct user32, u_debugreg[7]):
1076                 regno -= offsetof(struct user32, u_debugreg[0]);
1077                 return ptrace_set_debugreg(child, regno / 4, value);
1078
1079         default:
1080                 if (regno > sizeof(struct user32) || (regno & 3))
1081                         return -EIO;
1082
1083                 /*
1084                  * Other dummy fields in the virtual user structure
1085                  * are ignored
1086                  */
1087                 break;
1088         }
1089         return 0;
1090 }
1091
1092 #undef R32
1093 #undef SEG32
1094
1095 #define R32(l,q)                                                        \
1096         case offsetof(struct user32, regs.l):                           \
1097                 *val = regs->q; break
1098
1099 #define SEG32(rs)                                                       \
1100         case offsetof(struct user32, regs.rs):                          \
1101                 *val = get_segment_reg(child,                           \
1102                                        offsetof(struct user_regs_struct, rs)); \
1103                 break
1104
1105 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1106 {
1107         struct pt_regs *regs = task_pt_regs(child);
1108
1109         switch (regno) {
1110
1111         SEG32(ds);
1112         SEG32(es);
1113         SEG32(fs);
1114         SEG32(gs);
1115
1116         R32(cs, cs);
1117         R32(ss, ss);
1118         R32(ebx, bx);
1119         R32(ecx, cx);
1120         R32(edx, dx);
1121         R32(edi, di);
1122         R32(esi, si);
1123         R32(ebp, bp);
1124         R32(eax, ax);
1125         R32(orig_eax, orig_ax);
1126         R32(eip, ip);
1127         R32(esp, sp);
1128
1129         case offsetof(struct user32, regs.eflags):
1130                 *val = get_flags(child);
1131                 break;
1132
1133         case offsetof(struct user32, u_debugreg[0]) ...
1134                 offsetof(struct user32, u_debugreg[7]):
1135                 regno -= offsetof(struct user32, u_debugreg[0]);
1136                 *val = ptrace_get_debugreg(child, regno / 4);
1137                 break;
1138
1139         default:
1140                 if (regno > sizeof(struct user32) || (regno & 3))
1141                         return -EIO;
1142
1143                 /*
1144                  * Other dummy fields in the virtual user structure
1145                  * are ignored
1146                  */
1147                 *val = 0;
1148                 break;
1149         }
1150         return 0;
1151 }
1152
1153 #undef R32
1154 #undef SEG32
1155
1156 static int genregs32_get(struct task_struct *target,
1157                          const struct user_regset *regset,
1158                          unsigned int pos, unsigned int count,
1159                          void *kbuf, void __user *ubuf)
1160 {
1161         if (kbuf) {
1162                 compat_ulong_t *k = kbuf;
1163                 while (count > 0) {
1164                         getreg32(target, pos, k++);
1165                         count -= sizeof(*k);
1166                         pos += sizeof(*k);
1167                 }
1168         } else {
1169                 compat_ulong_t __user *u = ubuf;
1170                 while (count > 0) {
1171                         compat_ulong_t word;
1172                         getreg32(target, pos, &word);
1173                         if (__put_user(word, u++))
1174                                 return -EFAULT;
1175                         count -= sizeof(*u);
1176                         pos += sizeof(*u);
1177                 }
1178         }
1179
1180         return 0;
1181 }
1182
1183 static int genregs32_set(struct task_struct *target,
1184                          const struct user_regset *regset,
1185                          unsigned int pos, unsigned int count,
1186                          const void *kbuf, const void __user *ubuf)
1187 {
1188         int ret = 0;
1189         if (kbuf) {
1190                 const compat_ulong_t *k = kbuf;
1191                 while (count > 0 && !ret) {
1192                         ret = putreg32(target, pos, *k++);
1193                         count -= sizeof(*k);
1194                         pos += sizeof(*k);
1195                 }
1196         } else {
1197                 const compat_ulong_t __user *u = ubuf;
1198                 while (count > 0 && !ret) {
1199                         compat_ulong_t word;
1200                         ret = __get_user(word, u++);
1201                         if (ret)
1202                                 break;
1203                         ret = putreg32(target, pos, word);
1204                         count -= sizeof(*u);
1205                         pos += sizeof(*u);
1206                 }
1207         }
1208         return ret;
1209 }
1210
1211 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1212                         compat_ulong_t caddr, compat_ulong_t cdata)
1213 {
1214         unsigned long addr = caddr;
1215         unsigned long data = cdata;
1216         void __user *datap = compat_ptr(data);
1217         int ret;
1218         __u32 val;
1219
1220         switch (request) {
1221         case PTRACE_PEEKUSR:
1222                 ret = getreg32(child, addr, &val);
1223                 if (ret == 0)
1224                         ret = put_user(val, (__u32 __user *)datap);
1225                 break;
1226
1227         case PTRACE_POKEUSR:
1228                 ret = putreg32(child, addr, data);
1229                 break;
1230
1231         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1232                 return copy_regset_to_user(child, &user_x86_32_view,
1233                                            REGSET_GENERAL,
1234                                            0, sizeof(struct user_regs_struct32),
1235                                            datap);
1236
1237         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1238                 return copy_regset_from_user(child, &user_x86_32_view,
1239                                              REGSET_GENERAL, 0,
1240                                              sizeof(struct user_regs_struct32),
1241                                              datap);
1242
1243         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1244                 return copy_regset_to_user(child, &user_x86_32_view,
1245                                            REGSET_FP, 0,
1246                                            sizeof(struct user_i387_ia32_struct),
1247                                            datap);
1248
1249         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1250                 return copy_regset_from_user(
1251                         child, &user_x86_32_view, REGSET_FP,
1252                         0, sizeof(struct user_i387_ia32_struct), datap);
1253
1254         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1255                 return copy_regset_to_user(child, &user_x86_32_view,
1256                                            REGSET_XFP, 0,
1257                                            sizeof(struct user32_fxsr_struct),
1258                                            datap);
1259
1260         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1261                 return copy_regset_from_user(child, &user_x86_32_view,
1262                                              REGSET_XFP, 0,
1263                                              sizeof(struct user32_fxsr_struct),
1264                                              datap);
1265
1266         case PTRACE_GET_THREAD_AREA:
1267         case PTRACE_SET_THREAD_AREA:
1268                 return arch_ptrace(child, request, addr, data);
1269
1270         default:
1271                 return compat_ptrace_request(child, request, addr, data);
1272         }
1273
1274         return ret;
1275 }
1276
1277 #endif  /* CONFIG_IA32_EMULATION */
1278
1279 #ifdef CONFIG_X86_64
1280
1281 static const struct user_regset x86_64_regsets[] = {
1282         [REGSET_GENERAL] = {
1283                 .core_note_type = NT_PRSTATUS,
1284                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1285                 .size = sizeof(long), .align = sizeof(long),
1286                 .get = genregs_get, .set = genregs_set
1287         },
1288         [REGSET_FP] = {
1289                 .core_note_type = NT_PRFPREG,
1290                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1291                 .size = sizeof(long), .align = sizeof(long),
1292                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1293         },
1294 };
1295
1296 static const struct user_regset_view user_x86_64_view = {
1297         .name = "x86_64", .e_machine = EM_X86_64,
1298         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1299 };
1300
1301 #else  /* CONFIG_X86_32 */
1302
1303 #define user_regs_struct32      user_regs_struct
1304 #define genregs32_get           genregs_get
1305 #define genregs32_set           genregs_set
1306
1307 #define user_i387_ia32_struct   user_i387_struct
1308 #define user32_fxsr_struct      user_fxsr_struct
1309
1310 #endif  /* CONFIG_X86_64 */
1311
1312 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1313 static const struct user_regset x86_32_regsets[] = {
1314         [REGSET_GENERAL] = {
1315                 .core_note_type = NT_PRSTATUS,
1316                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1317                 .size = sizeof(u32), .align = sizeof(u32),
1318                 .get = genregs32_get, .set = genregs32_set
1319         },
1320         [REGSET_FP] = {
1321                 .core_note_type = NT_PRFPREG,
1322                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1323                 .size = sizeof(u32), .align = sizeof(u32),
1324                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1325         },
1326         [REGSET_XFP] = {
1327                 .core_note_type = NT_PRXFPREG,
1328                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1329                 .size = sizeof(u32), .align = sizeof(u32),
1330                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1331         },
1332         [REGSET_TLS] = {
1333                 .core_note_type = NT_386_TLS,
1334                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1335                 .size = sizeof(struct user_desc),
1336                 .align = sizeof(struct user_desc),
1337                 .active = regset_tls_active,
1338                 .get = regset_tls_get, .set = regset_tls_set
1339         },
1340 };
1341
1342 static const struct user_regset_view user_x86_32_view = {
1343         .name = "i386", .e_machine = EM_386,
1344         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1345 };
1346 #endif
1347
1348 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1349 {
1350 #ifdef CONFIG_IA32_EMULATION
1351         if (test_tsk_thread_flag(task, TIF_IA32))
1352 #endif
1353 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1354                 return &user_x86_32_view;
1355 #endif
1356 #ifdef CONFIG_X86_64
1357         return &user_x86_64_view;
1358 #endif
1359 }
1360
1361 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1362 {
1363         struct siginfo info;
1364
1365         tsk->thread.trap_no = 1;
1366         tsk->thread.error_code = error_code;
1367
1368         memset(&info, 0, sizeof(info));
1369         info.si_signo = SIGTRAP;
1370         info.si_code = TRAP_BRKPT;
1371
1372         /* User-mode ip? */
1373         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1374
1375         /* Send us the fake SIGTRAP */
1376         force_sig_info(SIGTRAP, &info, tsk);
1377 }
1378
1379
1380 #ifdef CONFIG_X86_32
1381 # define IS_IA32        1
1382 #elif defined CONFIG_IA32_EMULATION
1383 # define IS_IA32        test_thread_flag(TIF_IA32)
1384 #else
1385 # define IS_IA32        0
1386 #endif
1387
1388 /*
1389  * We must return the syscall number to actually look up in the table.
1390  * This can be -1L to skip running any syscall at all.
1391  */
1392 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1393 {
1394         long ret = 0;
1395
1396         /*
1397          * If we stepped into a sysenter/syscall insn, it trapped in
1398          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1399          * If user-mode had set TF itself, then it's still clear from
1400          * do_debug() and we need to set it again to restore the user
1401          * state.  If we entered on the slow path, TF was already set.
1402          */
1403         if (test_thread_flag(TIF_SINGLESTEP))
1404                 regs->flags |= X86_EFLAGS_TF;
1405
1406         /* do the secure computing check first */
1407         secure_computing(regs->orig_ax);
1408
1409         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1410                 ret = -1L;
1411
1412         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1413             tracehook_report_syscall_entry(regs))
1414                 ret = -1L;
1415
1416         if (unlikely(current->audit_context)) {
1417                 if (IS_IA32)
1418                         audit_syscall_entry(AUDIT_ARCH_I386,
1419                                             regs->orig_ax,
1420                                             regs->bx, regs->cx,
1421                                             regs->dx, regs->si);
1422 #ifdef CONFIG_X86_64
1423                 else
1424                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1425                                             regs->orig_ax,
1426                                             regs->di, regs->si,
1427                                             regs->dx, regs->r10);
1428 #endif
1429         }
1430
1431         return ret ?: regs->orig_ax;
1432 }
1433
1434 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1435 {
1436         if (unlikely(current->audit_context))
1437                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1438
1439         if (test_thread_flag(TIF_SYSCALL_TRACE))
1440                 tracehook_report_syscall_exit(regs, 0);
1441
1442         /*
1443          * If TIF_SYSCALL_EMU is set, we only get here because of
1444          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1445          * We already reported this syscall instruction in
1446          * syscall_trace_enter(), so don't do any more now.
1447          */
1448         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1449                 return;
1450
1451         /*
1452          * If we are single-stepping, synthesize a trap to follow the
1453          * system call instruction.
1454          */
1455         if (test_thread_flag(TIF_SINGLESTEP) &&
1456             tracehook_consider_fatal_signal(current, SIGTRAP, SIG_DFL))
1457                 send_sigtrap(current, regs, 0);
1458 }