arch/x86/kernel/ptrace.c

   1 /* By Ross Biro 1/23/92 */
   2 /*
   3  * Pentium III FXSR, SSE support
   4  *      Gareth Hughes <gareth@valinux.com>, May 2000
   5  *
   6  * BTS tracing
   7  *      Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/mm.h>
  13 #include <linux/smp.h>
  14 #include <linux/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/regset.h>
  17 #include <linux/tracehook.h>
  18 #include <linux/user.h>
  19 #include <linux/elf.h>
  20 #include <linux/security.h>
  21 #include <linux/audit.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/signal.h>
  24 #include <linux/workqueue.h>
  25
  26 #include <asm/uaccess.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/system.h>
  29 #include <asm/processor.h>
  30 #include <asm/i387.h>
  31 #include <asm/debugreg.h>
  32 #include <asm/ldt.h>
  33 #include <asm/desc.h>
  34 #include <asm/prctl.h>
  35 #include <asm/proto.h>
  36 #include <asm/ds.h>
  37 #include <asm/hw_breakpoint.h>
  38
  39 #include "tls.h"
  40
  41 #define CREATE_TRACE_POINTS
  42 #include <trace/events/syscalls.h>
  43
  44 enum x86_regset {
  45         REGSET_GENERAL,
  46         REGSET_FP,
  47         REGSET_XFP,
  48         REGSET_IOPERM64 = REGSET_XFP,
  49         REGSET_TLS,
  50         REGSET_IOPERM32,
  51 };
  52
  53 /*
  54  * does not yet catch signals sent when the child dies.
  55  * in exit.c or in signal.c.
  56  */
  57
  58 /*
  59  * Determines which flags the user has access to [1 = access, 0 = no access].
  60  */
  61 #define FLAG_MASK_32            ((unsigned long)                        \
  62                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  63                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  64                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  65                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  66                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  67
  68 /*
  69  * Determines whether a value may be installed in a segment register.
  70  */
  71 static inline bool invalid_selector(u16 value)
  72 {
  73         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  74 }
  75
  76 #ifdef CONFIG_X86_32
  77
  78 #define FLAG_MASK               FLAG_MASK_32
  79
  80 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  81 {
  82         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  83         return &regs->bx + (regno >> 2);
  84 }
  85
  86 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  87 {
  88         /*
  89          * Returning the value truncates it to 16 bits.
  90          */
  91         unsigned int retval;
  92         if (offset != offsetof(struct user_regs_struct, gs))
  93                 retval = *pt_regs_access(task_pt_regs(task), offset);
  94         else {
  95                 if (task == current)
  96                         retval = get_user_gs(task_pt_regs(task));
  97                 else
  98                         retval = task_user_gs(task);
  99         }
 100         return retval;
 101 }
 102
 103 static int set_segment_reg(struct task_struct *task,
 104                            unsigned long offset, u16 value)
 105 {
 106         /*
 107          * The value argument was already truncated to 16 bits.
 108          */
 109         if (invalid_selector(value))
 110                 return -EIO;
 111
 112         /*
 113          * For %cs and %ss we cannot permit a null selector.
 114          * We can permit a bogus selector as long as it has USER_RPL.
 115          * Null selectors are fine for other segment registers, but
 116          * we will never get back to user mode with invalid %cs or %ss
 117          * and will take the trap in iret instead.  Much code relies
 118          * on user_mode() to distinguish a user trap frame (which can
 119          * safely use invalid selectors) from a kernel trap frame.
 120          */
 121         switch (offset) {
 122         case offsetof(struct user_regs_struct, cs):
 123         case offsetof(struct user_regs_struct, ss):
 124                 if (unlikely(value == 0))
 125                         return -EIO;
 126
 127         default:
 128                 *pt_regs_access(task_pt_regs(task), offset) = value;
 129                 break;
 130
 131         case offsetof(struct user_regs_struct, gs):
 132                 if (task == current)
 133                         set_user_gs(task_pt_regs(task), value);
 134                 else
 135                         task_user_gs(task) = value;
 136         }
 137
 138         return 0;
 139 }
 140
 141 #else  /* CONFIG_X86_64 */
 142
 143 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 144
 145 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 146 {
 147         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 148         return &regs->r15 + (offset / sizeof(regs->r15));
 149 }
 150
 151 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 152 {
 153         /*
 154          * Returning the value truncates it to 16 bits.
 155          */
 156         unsigned int seg;
 157
 158         switch (offset) {
 159         case offsetof(struct user_regs_struct, fs):
 160                 if (task == current) {
 161                         /* Older gas can't assemble movq %?s,%r?? */
 162                         asm("movl %%fs,%0" : "=r" (seg));
 163                         return seg;
 164                 }
 165                 return task->thread.fsindex;
 166         case offsetof(struct user_regs_struct, gs):
 167                 if (task == current) {
 168                         asm("movl %%gs,%0" : "=r" (seg));
 169                         return seg;
 170                 }
 171                 return task->thread.gsindex;
 172         case offsetof(struct user_regs_struct, ds):
 173                 if (task == current) {
 174                         asm("movl %%ds,%0" : "=r" (seg));
 175                         return seg;
 176                 }
 177                 return task->thread.ds;
 178         case offsetof(struct user_regs_struct, es):
 179                 if (task == current) {
 180                         asm("movl %%es,%0" : "=r" (seg));
 181                         return seg;
 182                 }
 183                 return task->thread.es;
 184
 185         case offsetof(struct user_regs_struct, cs):
 186         case offsetof(struct user_regs_struct, ss):
 187                 break;
 188         }
 189         return *pt_regs_access(task_pt_regs(task), offset);
 190 }
 191
 192 static int set_segment_reg(struct task_struct *task,
 193                            unsigned long offset, u16 value)
 194 {
 195         /*
 196          * The value argument was already truncated to 16 bits.
 197          */
 198         if (invalid_selector(value))
 199                 return -EIO;
 200
 201         switch (offset) {
 202         case offsetof(struct user_regs_struct,fs):
 203                 /*
 204                  * If this is setting fs as for normal 64-bit use but
 205                  * setting fs_base has implicitly changed it, leave it.
 206                  */
 207                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 208                      task->thread.fs != 0) ||
 209                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 210                      task->thread.fs == 0))
 211                         break;
 212                 task->thread.fsindex = value;
 213                 if (task == current)
 214                         loadsegment(fs, task->thread.fsindex);
 215                 break;
 216         case offsetof(struct user_regs_struct,gs):
 217                 /*
 218                  * If this is setting gs as for normal 64-bit use but
 219                  * setting gs_base has implicitly changed it, leave it.
 220                  */
 221                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 222                      task->thread.gs != 0) ||
 223                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 224                      task->thread.gs == 0))
 225                         break;
 226                 task->thread.gsindex = value;
 227                 if (task == current)
 228                         load_gs_index(task->thread.gsindex);
 229                 break;
 230         case offsetof(struct user_regs_struct,ds):
 231                 task->thread.ds = value;
 232                 if (task == current)
 233                         loadsegment(ds, task->thread.ds);
 234                 break;
 235         case offsetof(struct user_regs_struct,es):
 236                 task->thread.es = value;
 237                 if (task == current)
 238                         loadsegment(es, task->thread.es);
 239                 break;
 240
 241                 /*
 242                  * Can't actually change these in 64-bit mode.
 243                  */
 244         case offsetof(struct user_regs_struct,cs):
 245                 if (unlikely(value == 0))
 246                         return -EIO;
 247 #ifdef CONFIG_IA32_EMULATION
 248                 if (test_tsk_thread_flag(task, TIF_IA32))
 249                         task_pt_regs(task)->cs = value;
 250 #endif
 251                 break;
 252         case offsetof(struct user_regs_struct,ss):
 253                 if (unlikely(value == 0))
 254                         return -EIO;
 255 #ifdef CONFIG_IA32_EMULATION
 256                 if (test_tsk_thread_flag(task, TIF_IA32))
 257                         task_pt_regs(task)->ss = value;
 258 #endif
 259                 break;
 260         }
 261
 262         return 0;
 263 }
 264
 265 #endif  /* CONFIG_X86_32 */
 266
 267 static unsigned long get_flags(struct task_struct *task)
 268 {
 269         unsigned long retval = task_pt_regs(task)->flags;
 270
 271         /*
 272          * If the debugger set TF, hide it from the readout.
 273          */
 274         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 275                 retval &= ~X86_EFLAGS_TF;
 276
 277         return retval;
 278 }
 279
 280 static int set_flags(struct task_struct *task, unsigned long value)
 281 {
 282         struct pt_regs *regs = task_pt_regs(task);
 283
 284         /*
 285          * If the user value contains TF, mark that
 286          * it was not "us" (the debugger) that set it.
 287          * If not, make sure it stays set if we had.
 288          */
 289         if (value & X86_EFLAGS_TF)
 290                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 291         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 292                 value |= X86_EFLAGS_TF;
 293
 294         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 295
 296         return 0;
 297 }
 298
 299 static int putreg(struct task_struct *child,
 300                   unsigned long offset, unsigned long value)
 301 {
 302         switch (offset) {
 303         case offsetof(struct user_regs_struct, cs):
 304         case offsetof(struct user_regs_struct, ds):
 305         case offsetof(struct user_regs_struct, es):
 306         case offsetof(struct user_regs_struct, fs):
 307         case offsetof(struct user_regs_struct, gs):
 308         case offsetof(struct user_regs_struct, ss):
 309                 return set_segment_reg(child, offset, value);
 310
 311         case offsetof(struct user_regs_struct, flags):
 312                 return set_flags(child, value);
 313
 314 #ifdef CONFIG_X86_64
 315         case offsetof(struct user_regs_struct,fs_base):
 316                 if (value >= TASK_SIZE_OF(child))
 317                         return -EIO;
 318                 /*
 319                  * When changing the segment base, use do_arch_prctl
 320                  * to set either thread.fs or thread.fsindex and the
 321                  * corresponding GDT slot.
 322                  */
 323                 if (child->thread.fs != value)
 324                         return do_arch_prctl(child, ARCH_SET_FS, value);
 325                 return 0;
 326         case offsetof(struct user_regs_struct,gs_base):
 327                 /*
 328                  * Exactly the same here as the %fs handling above.
 329                  */
 330                 if (value >= TASK_SIZE_OF(child))
 331                         return -EIO;
 332                 if (child->thread.gs != value)
 333                         return do_arch_prctl(child, ARCH_SET_GS, value);
 334                 return 0;
 335 #endif
 336         }
 337
 338         *pt_regs_access(task_pt_regs(child), offset) = value;
 339         return 0;
 340 }
 341
 342 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 343 {
 344         switch (offset) {
 345         case offsetof(struct user_regs_struct, cs):
 346         case offsetof(struct user_regs_struct, ds):
 347         case offsetof(struct user_regs_struct, es):
 348         case offsetof(struct user_regs_struct, fs):
 349         case offsetof(struct user_regs_struct, gs):
 350         case offsetof(struct user_regs_struct, ss):
 351                 return get_segment_reg(task, offset);
 352
 353         case offsetof(struct user_regs_struct, flags):
 354                 return get_flags(task);
 355
 356 #ifdef CONFIG_X86_64
 357         case offsetof(struct user_regs_struct, fs_base): {
 358                 /*
 359                  * do_arch_prctl may have used a GDT slot instead of
 360                  * the MSR.  To userland, it appears the same either
 361                  * way, except the %fs segment selector might not be 0.
 362                  */
 363                 unsigned int seg = task->thread.fsindex;
 364                 if (task->thread.fs != 0)
 365                         return task->thread.fs;
 366                 if (task == current)
 367                         asm("movl %%fs,%0" : "=r" (seg));
 368                 if (seg != FS_TLS_SEL)
 369                         return 0;
 370                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 371         }
 372         case offsetof(struct user_regs_struct, gs_base): {
 373                 /*
 374                  * Exactly the same here as the %fs handling above.
 375                  */
 376                 unsigned int seg = task->thread.gsindex;
 377                 if (task->thread.gs != 0)
 378                         return task->thread.gs;
 379                 if (task == current)
 380                         asm("movl %%gs,%0" : "=r" (seg));
 381                 if (seg != GS_TLS_SEL)
 382                         return 0;
 383                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 384         }
 385 #endif
 386         }
 387
 388         return *pt_regs_access(task_pt_regs(task), offset);
 389 }
 390
 391 static int genregs_get(struct task_struct *target,
 392                        const struct user_regset *regset,
 393                        unsigned int pos, unsigned int count,
 394                        void *kbuf, void __user *ubuf)
 395 {
 396         if (kbuf) {
 397                 unsigned long *k = kbuf;
 398                 while (count > 0) {
 399                         *k++ = getreg(target, pos);
 400                         count -= sizeof(*k);
 401                         pos += sizeof(*k);
 402                 }
 403         } else {
 404                 unsigned long __user *u = ubuf;
 405                 while (count > 0) {
 406                         if (__put_user(getreg(target, pos), u++))
 407                                 return -EFAULT;
 408                         count -= sizeof(*u);
 409                         pos += sizeof(*u);
 410                 }
 411         }
 412
 413         return 0;
 414 }
 415
 416 static int genregs_set(struct task_struct *target,
 417                        const struct user_regset *regset,
 418                        unsigned int pos, unsigned int count,
 419                        const void *kbuf, const void __user *ubuf)
 420 {
 421         int ret = 0;
 422         if (kbuf) {
 423                 const unsigned long *k = kbuf;
 424                 while (count > 0 && !ret) {
 425                         ret = putreg(target, pos, *k++);
 426                         count -= sizeof(*k);
 427                         pos += sizeof(*k);
 428                 }
 429         } else {
 430                 const unsigned long  __user *u = ubuf;
 431                 while (count > 0 && !ret) {
 432                         unsigned long word;
 433                         ret = __get_user(word, u++);
 434                         if (ret)
 435                                 break;
 436                         ret = putreg(target, pos, word);
 437                         count -= sizeof(*u);
 438                         pos += sizeof(*u);
 439                 }
 440         }
 441         return ret;
 442 }
 443
 444 /*
 445  * Decode the length and type bits for a particular breakpoint as
 446  * stored in debug register 7.  Return the "enabled" status.
 447  */
 448 static int decode_dr7(unsigned long dr7, int bpnum, unsigned *len,
 449                 unsigned *type)
 450 {
 451         int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE);
 452
 453         *len = (bp_info & 0xc) | 0x40;
 454         *type = (bp_info & 0x3) | 0x80;
 455         return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3;
 456 }
 457
 458 static void ptrace_triggered(struct hw_breakpoint *bp, struct pt_regs *regs)
 459 {
 460         struct thread_struct *thread = &(current->thread);
 461         int i;
 462
 463         /*
 464          * Store in the virtual DR6 register the fact that the breakpoint
 465          * was hit so the thread's debugger will see it.
 466          */
 467         for (i = 0; i < hbp_kernel_pos; i++)
 468                 /*
 469                  * We will check bp->info.address against the address stored in
 470                  * thread's hbp structure and not debugreg[i]. This is to ensure
 471                  * that the corresponding bit for 'i' in DR7 register is enabled
 472                  */
 473                 if (bp->info.address == thread->hbp[i]->info.address)
 474                         break;
 475
 476         thread->debugreg6 |= (DR_TRAP0 << i);
 477 }
 478
 479 /*
 480  * Handle ptrace writes to debug register 7.
 481  */
 482 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 483 {
 484         struct thread_struct *thread = &(tsk->thread);
 485         unsigned long old_dr7 = thread->debugreg7;
 486         int i, orig_ret = 0, rc = 0;
 487         int enabled, second_pass = 0;
 488         unsigned len, type;
 489         struct hw_breakpoint *bp;
 490
 491         data &= ~DR_CONTROL_RESERVED;
 492 restore:
 493         /*
 494          * Loop through all the hardware breakpoints, making the
 495          * appropriate changes to each.
 496          */
 497         for (i = 0; i < HBP_NUM; i++) {
 498                 enabled = decode_dr7(data, i, &len, &type);
 499                 bp = thread->hbp[i];
 500
 501                 if (!enabled) {
 502                         if (bp) {
 503                                 /* Don't unregister the breakpoints right-away,
 504                                  * unless all register_user_hw_breakpoint()
 505                                  * requests have succeeded. This prevents
 506                                  * any window of opportunity for debug
 507                                  * register grabbing by other users.
 508                                  */
 509                                 if (!second_pass)
 510                                         continue;
 511                                 unregister_user_hw_breakpoint(tsk, bp);
 512                                 kfree(bp);
 513                         }
 514                         continue;
 515                 }
 516                 if (!bp) {
 517                         rc = -ENOMEM;
 518                         bp = kzalloc(sizeof(struct hw_breakpoint), GFP_KERNEL);
 519                         if (bp) {
 520                                 bp->info.address = thread->debugreg[i];
 521                                 bp->triggered = ptrace_triggered;
 522                                 bp->info.len = len;
 523                                 bp->info.type = type;
 524                                 rc = register_user_hw_breakpoint(tsk, bp);
 525                                 if (rc)
 526                                         kfree(bp);
 527                         }
 528                 } else
 529                         rc = modify_user_hw_breakpoint(tsk, bp);
 530                 if (rc)
 531                         break;
 532         }
 533         /*
 534          * Make a second pass to free the remaining unused breakpoints
 535          * or to restore the original breakpoints if an error occurred.
 536          */
 537         if (!second_pass) {
 538                 second_pass = 1;
 539                 if (rc < 0) {
 540                         orig_ret = rc;
 541                         data = old_dr7;
 542                 }
 543                 goto restore;
 544         }
 545         return ((orig_ret < 0) ? orig_ret : rc);
 546 }
 547
 548 /*
 549  * Handle PTRACE_PEEKUSR calls for the debug register area.
 550  */
 551 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 552 {
 553         struct thread_struct *thread = &(tsk->thread);
 554         unsigned long val = 0;
 555
 556         if (n < HBP_NUM)
 557                 val = thread->debugreg[n];
 558         else if (n == 6)
 559                 val = thread->debugreg6;
 560         else if (n == 7)
 561                 val = thread->debugreg7;
 562         return val;
 563 }
 564
 565 /*
 566  * Handle PTRACE_POKEUSR calls for the debug register area.
 567  */
 568 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
 569 {
 570         struct thread_struct *thread = &(tsk->thread);
 571         int rc = 0;
 572
 573         /* There are no DR4 or DR5 registers */
 574         if (n == 4 || n == 5)
 575                 return -EIO;
 576
 577         if (n == 6) {
 578                 tsk->thread.debugreg6 = val;
 579                 goto ret_path;
 580         }
 581         if (n < HBP_NUM) {
 582                 if (thread->hbp[n]) {
 583                         if (arch_check_va_in_userspace(val,
 584                                         thread->hbp[n]->info.len) == 0) {
 585                                 rc = -EIO;
 586                                 goto ret_path;
 587                         }
 588                         thread->hbp[n]->info.address = val;
 589                 }
 590                 thread->debugreg[n] = val;
 591         }
 592         /* All that's left is DR7 */
 593         if (n == 7)
 594                 rc = ptrace_write_dr7(tsk, val);
 595
 596 ret_path:
 597         return rc;
 598 }
 599
 600 /*
 601  * These access the current or another (stopped) task's io permission
 602  * bitmap for debugging or core dump.
 603  */
 604 static int ioperm_active(struct task_struct *target,
 605                          const struct user_regset *regset)
 606 {
 607         return target->thread.io_bitmap_max / regset->size;
 608 }
 609
 610 static int ioperm_get(struct task_struct *target,
 611                       const struct user_regset *regset,
 612                       unsigned int pos, unsigned int count,
 613                       void *kbuf, void __user *ubuf)
 614 {
 615         if (!target->thread.io_bitmap_ptr)
 616                 return -ENXIO;
 617
 618         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 619                                    target->thread.io_bitmap_ptr,
 620                                    0, IO_BITMAP_BYTES);
 621 }
 622
 623 #ifdef CONFIG_X86_PTRACE_BTS
 624 /*
 625  * A branch trace store context.
 626  *
 627  * Contexts may only be installed by ptrace_bts_config() and only for
 628  * ptraced tasks.
 629  *
 630  * Contexts are destroyed when the tracee is detached from the tracer.
 631  * The actual destruction work requires interrupts enabled, so the
 632  * work is deferred and will be scheduled during __ptrace_unlink().
 633  *
 634  * Contexts hold an additional task_struct reference on the traced
 635  * task, as well as a reference on the tracer's mm.
 636  *
 637  * Ptrace already holds a task_struct for the duration of ptrace operations,
 638  * but since destruction is deferred, it may be executed after both
 639  * tracer and tracee exited.
 640  */
 641 struct bts_context {
 642         /* The branch trace handle. */
 643         struct bts_tracer       *tracer;
 644
 645         /* The buffer used to store the branch trace and its size. */
 646         void                    *buffer;
 647         unsigned int            size;
 648
 649         /* The mm that paid for the above buffer. */
 650         struct mm_struct        *mm;
 651
 652         /* The task this context belongs to. */
 653         struct task_struct      *task;
 654
 655         /* The signal to send on a bts buffer overflow. */
 656         unsigned int            bts_ovfl_signal;
 657
 658         /* The work struct to destroy a context. */
 659         struct work_struct      work;
 660 };
 661
 662 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
 663 {
 664         void *buffer = NULL;
 665         int err = -ENOMEM;
 666
 667         err = account_locked_memory(current->mm, current->signal->rlim, size);
 668         if (err < 0)
 669                 return err;
 670
 671         buffer = kzalloc(size, GFP_KERNEL);
 672         if (!buffer)
 673                 goto out_refund;
 674
 675         context->buffer = buffer;
 676         context->size = size;
 677         context->mm = get_task_mm(current);
 678
 679         return 0;
 680
 681  out_refund:
 682         refund_locked_memory(current->mm, size);
 683         return err;
 684 }
 685
 686 static inline void free_bts_buffer(struct bts_context *context)
 687 {
 688         if (!context->buffer)
 689                 return;
 690
 691         kfree(context->buffer);
 692         context->buffer = NULL;
 693
 694         refund_locked_memory(context->mm, context->size);
 695         context->size = 0;
 696
 697         mmput(context->mm);
 698         context->mm = NULL;
 699 }
 700
 701 static void free_bts_context_work(struct work_struct *w)
 702 {
 703         struct bts_context *context;
 704
 705         context = container_of(w, struct bts_context, work);
 706
 707         ds_release_bts(context->tracer);
 708         put_task_struct(context->task);
 709         free_bts_buffer(context);
 710         kfree(context);
 711 }
 712
 713 static inline void free_bts_context(struct bts_context *context)
 714 {
 715         INIT_WORK(&context->work, free_bts_context_work);
 716         schedule_work(&context->work);
 717 }
 718
 719 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
 720 {
 721         struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
 722         if (context) {
 723                 context->task = task;
 724                 task->bts = context;
 725
 726                 get_task_struct(task);
 727         }
 728
 729         return context;
 730 }
 731
 732 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 733                                   struct bts_struct __user *out)
 734 {
 735         struct bts_context *context;
 736         const struct bts_trace *trace;
 737         struct bts_struct bts;
 738         const unsigned char *at;
 739         int error;
 740
 741         context = child->bts;
 742         if (!context)
 743                 return -ESRCH;
 744
 745         trace = ds_read_bts(context->tracer);
 746         if (!trace)
 747                 return -ESRCH;
 748
 749         at = trace->ds.top - ((index + 1) * trace->ds.size);
 750         if ((void *)at < trace->ds.begin)
 751                 at += (trace->ds.n * trace->ds.size);
 752
 753         if (!trace->read)
 754                 return -EOPNOTSUPP;
 755
 756         error = trace->read(context->tracer, at, &bts);
 757         if (error < 0)
 758                 return error;
 759
 760         if (copy_to_user(out, &bts, sizeof(bts)))
 761                 return -EFAULT;
 762
 763         return sizeof(bts);
 764 }
 765
 766 static int ptrace_bts_drain(struct task_struct *child,
 767                             long size,
 768                             struct bts_struct __user *out)
 769 {
 770         struct bts_context *context;
 771         const struct bts_trace *trace;
 772         const unsigned char *at;
 773         int error, drained = 0;
 774
 775         context = child->bts;
 776         if (!context)
 777                 return -ESRCH;
 778
 779         trace = ds_read_bts(context->tracer);
 780         if (!trace)
 781                 return -ESRCH;
 782
 783         if (!trace->read)
 784                 return -EOPNOTSUPP;
 785
 786         if (size < (trace->ds.top - trace->ds.begin))
 787                 return -EIO;
 788
 789         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 790              out++, drained++, at += trace->ds.size) {
 791                 struct bts_struct bts;
 792
 793                 error = trace->read(context->tracer, at, &bts);
 794                 if (error < 0)
 795                         return error;
 796
 797                 if (copy_to_user(out, &bts, sizeof(bts)))
 798                         return -EFAULT;
 799         }
 800
 801         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 802
 803         error = ds_reset_bts(context->tracer);
 804         if (error < 0)
 805                 return error;
 806
 807         return drained;
 808 }
 809
 810 static int ptrace_bts_config(struct task_struct *child,
 811                              long cfg_size,
 812                              const struct ptrace_bts_config __user *ucfg)
 813 {
 814         struct bts_context *context;
 815         struct ptrace_bts_config cfg;
 816         unsigned int flags = 0;
 817
 818         if (cfg_size < sizeof(cfg))
 819                 return -EIO;
 820
 821         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 822                 return -EFAULT;
 823
 824         context = child->bts;
 825         if (!context)
 826                 context = alloc_bts_context(child);
 827         if (!context)
 828                 return -ENOMEM;
 829
 830         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 831                 if (!cfg.signal)
 832                         return -EINVAL;
 833
 834                 return -EOPNOTSUPP;
 835                 context->bts_ovfl_signal = cfg.signal;
 836         }
 837
 838         ds_release_bts(context->tracer);
 839         context->tracer = NULL;
 840
 841         if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
 842                 int err;
 843
 844                 free_bts_buffer(context);
 845                 if (!cfg.size)
 846                         return 0;
 847
 848                 err = alloc_bts_buffer(context, cfg.size);
 849                 if (err < 0)
 850                         return err;
 851         }
 852
 853         if (cfg.flags & PTRACE_BTS_O_TRACE)
 854                 flags |= BTS_USER;
 855
 856         if (cfg.flags & PTRACE_BTS_O_SCHED)
 857                 flags |= BTS_TIMESTAMPS;
 858
 859         context->tracer =
 860                 ds_request_bts_task(child, context->buffer, context->size,
 861                                     NULL, (size_t)-1, flags);
 862         if (unlikely(IS_ERR(context->tracer))) {
 863                 int error = PTR_ERR(context->tracer);
 864
 865                 free_bts_buffer(context);
 866                 context->tracer = NULL;
 867                 return error;
 868         }
 869
 870         return sizeof(cfg);
 871 }
 872
 873 static int ptrace_bts_status(struct task_struct *child,
 874                              long cfg_size,
 875                              struct ptrace_bts_config __user *ucfg)
 876 {
 877         struct bts_context *context;
 878         const struct bts_trace *trace;
 879         struct ptrace_bts_config cfg;
 880
 881         context = child->bts;
 882         if (!context)
 883                 return -ESRCH;
 884
 885         if (cfg_size < sizeof(cfg))
 886                 return -EIO;
 887
 888         trace = ds_read_bts(context->tracer);
 889         if (!trace)
 890                 return -ESRCH;
 891
 892         memset(&cfg, 0, sizeof(cfg));
 893         cfg.size        = trace->ds.end - trace->ds.begin;
 894         cfg.signal      = context->bts_ovfl_signal;
 895         cfg.bts_size    = sizeof(struct bts_struct);
 896
 897         if (cfg.signal)
 898                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 899
 900         if (trace->ds.flags & BTS_USER)
 901                 cfg.flags |= PTRACE_BTS_O_TRACE;
 902
 903         if (trace->ds.flags & BTS_TIMESTAMPS)
 904                 cfg.flags |= PTRACE_BTS_O_SCHED;
 905
 906         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 907                 return -EFAULT;
 908
 909         return sizeof(cfg);
 910 }
 911
 912 static int ptrace_bts_clear(struct task_struct *child)
 913 {
 914         struct bts_context *context;
 915         const struct bts_trace *trace;
 916
 917         context = child->bts;
 918         if (!context)
 919                 return -ESRCH;
 920
 921         trace = ds_read_bts(context->tracer);
 922         if (!trace)
 923                 return -ESRCH;
 924
 925         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 926
 927         return ds_reset_bts(context->tracer);
 928 }
 929
 930 static int ptrace_bts_size(struct task_struct *child)
 931 {
 932         struct bts_context *context;
 933         const struct bts_trace *trace;
 934
 935         context = child->bts;
 936         if (!context)
 937                 return -ESRCH;
 938
 939         trace = ds_read_bts(context->tracer);
 940         if (!trace)
 941                 return -ESRCH;
 942
 943         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 944 }
 945
 946 /*
 947  * Called from __ptrace_unlink() after the child has been moved back
 948  * to its original parent.
 949  */
 950 void ptrace_bts_untrace(struct task_struct *child)
 951 {
 952         if (unlikely(child->bts)) {
 953                 free_bts_context(child->bts);
 954                 child->bts = NULL;
 955         }
 956 }
 957 #endif /* CONFIG_X86_PTRACE_BTS */
 958
 959 /*
 960  * Called by kernel/ptrace.c when detaching..
 961  *
 962  * Make sure the single step bit is not set.
 963  */
 964 void ptrace_disable(struct task_struct *child)
 965 {
 966         user_disable_single_step(child);
 967 #ifdef TIF_SYSCALL_EMU
 968         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 969 #endif
 970 }
 971
 972 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 973 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 974 #endif
 975
 976 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 977 {
 978         int ret;
 979         unsigned long __user *datap = (unsigned long __user *)data;
 980
 981         switch (request) {
 982         /* read the word at location addr in the USER area. */
 983         case PTRACE_PEEKUSR: {
 984                 unsigned long tmp;
 985
 986                 ret = -EIO;
 987                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 988                     addr >= sizeof(struct user))
 989                         break;
 990
 991                 tmp = 0;  /* Default return condition */
 992                 if (addr < sizeof(struct user_regs_struct))
 993                         tmp = getreg(child, addr);
 994                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 995                          addr <= offsetof(struct user, u_debugreg[7])) {
 996                         addr -= offsetof(struct user, u_debugreg[0]);
 997                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 998                 }
 999                 ret = put_user(tmp, datap);
1000                 break;
1001         }
1002
1003         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1004                 ret = -EIO;
1005                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1006                     addr >= sizeof(struct user))
1007                         break;
1008
1009                 if (addr < sizeof(struct user_regs_struct))
1010                         ret = putreg(child, addr, data);
1011                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1012                          addr <= offsetof(struct user, u_debugreg[7])) {
1013                         addr -= offsetof(struct user, u_debugreg[0]);
1014                         ret = ptrace_set_debugreg(child,
1015                                                   addr / sizeof(data), data);
1016                 }
1017                 break;
1018
1019         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1020                 return copy_regset_to_user(child,
1021                                            task_user_regset_view(current),
1022                                            REGSET_GENERAL,
1023                                            0, sizeof(struct user_regs_struct),
1024                                            datap);
1025
1026         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1027                 return copy_regset_from_user(child,
1028                                              task_user_regset_view(current),
1029                                              REGSET_GENERAL,
1030                                              0, sizeof(struct user_regs_struct),
1031                                              datap);
1032
1033         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1034                 return copy_regset_to_user(child,
1035                                            task_user_regset_view(current),
1036                                            REGSET_FP,
1037                                            0, sizeof(struct user_i387_struct),
1038                                            datap);
1039
1040         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1041                 return copy_regset_from_user(child,
1042                                              task_user_regset_view(current),
1043                                              REGSET_FP,
1044                                              0, sizeof(struct user_i387_struct),
1045                                              datap);
1046
1047 #ifdef CONFIG_X86_32
1048         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1049                 return copy_regset_to_user(child, &user_x86_32_view,
1050                                            REGSET_XFP,
1051                                            0, sizeof(struct user_fxsr_struct),
1052                                            datap) ? -EIO : 0;
1053
1054         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1055                 return copy_regset_from_user(child, &user_x86_32_view,
1056                                              REGSET_XFP,
1057                                              0, sizeof(struct user_fxsr_struct),
1058                                              datap) ? -EIO : 0;
1059 #endif
1060
1061 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1062         case PTRACE_GET_THREAD_AREA:
1063                 if (addr < 0)
1064                         return -EIO;
1065                 ret = do_get_thread_area(child, addr,
1066                                          (struct user_desc __user *) data);
1067                 break;
1068
1069         case PTRACE_SET_THREAD_AREA:
1070                 if (addr < 0)
1071                         return -EIO;
1072                 ret = do_set_thread_area(child, addr,
1073                                          (struct user_desc __user *) data, 0);
1074                 break;
1075 #endif
1076
1077 #ifdef CONFIG_X86_64
1078                 /* normal 64bit interface to access TLS data.
1079                    Works just like arch_prctl, except that the arguments
1080                    are reversed. */
1081         case PTRACE_ARCH_PRCTL:
1082                 ret = do_arch_prctl(child, data, addr);
1083                 break;
1084 #endif
1085
1086         /*
1087          * These bits need more cooking - not enabled yet:
1088          */
1089 #ifdef CONFIG_X86_PTRACE_BTS
1090         case PTRACE_BTS_CONFIG:
1091                 ret = ptrace_bts_config
1092                         (child, data, (struct ptrace_bts_config __user *)addr);
1093                 break;
1094
1095         case PTRACE_BTS_STATUS:
1096                 ret = ptrace_bts_status
1097                         (child, data, (struct ptrace_bts_config __user *)addr);
1098                 break;
1099
1100         case PTRACE_BTS_SIZE:
1101                 ret = ptrace_bts_size(child);
1102                 break;
1103
1104         case PTRACE_BTS_GET:
1105                 ret = ptrace_bts_read_record
1106                         (child, data, (struct bts_struct __user *) addr);
1107                 break;
1108
1109         case PTRACE_BTS_CLEAR:
1110                 ret = ptrace_bts_clear(child);
1111                 break;
1112
1113         case PTRACE_BTS_DRAIN:
1114                 ret = ptrace_bts_drain
1115                         (child, data, (struct bts_struct __user *) addr);
1116                 break;
1117 #endif /* CONFIG_X86_PTRACE_BTS */
1118
1119         default:
1120                 ret = ptrace_request(child, request, addr, data);
1121                 break;
1122         }
1123
1124         return ret;
1125 }
1126
1127 #ifdef CONFIG_IA32_EMULATION
1128
1129 #include <linux/compat.h>
1130 #include <linux/syscalls.h>
1131 #include <asm/ia32.h>
1132 #include <asm/user32.h>
1133
1134 #define R32(l,q)                                                        \
1135         case offsetof(struct user32, regs.l):                           \
1136                 regs->q = value; break
1137
1138 #define SEG32(rs)                                                       \
1139         case offsetof(struct user32, regs.rs):                          \
1140                 return set_segment_reg(child,                           \
1141                                        offsetof(struct user_regs_struct, rs), \
1142                                        value);                          \
1143                 break
1144
1145 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1146 {
1147         struct pt_regs *regs = task_pt_regs(child);
1148
1149         switch (regno) {
1150
1151         SEG32(cs);
1152         SEG32(ds);
1153         SEG32(es);
1154         SEG32(fs);
1155         SEG32(gs);
1156         SEG32(ss);
1157
1158         R32(ebx, bx);
1159         R32(ecx, cx);
1160         R32(edx, dx);
1161         R32(edi, di);
1162         R32(esi, si);
1163         R32(ebp, bp);
1164         R32(eax, ax);
1165         R32(eip, ip);
1166         R32(esp, sp);
1167
1168         case offsetof(struct user32, regs.orig_eax):
1169                 /*
1170                  * A 32-bit debugger setting orig_eax means to restore
1171                  * the state of the task restarting a 32-bit syscall.
1172                  * Make sure we interpret the -ERESTART* codes correctly
1173                  * in case the task is not actually still sitting at the
1174                  * exit from a 32-bit syscall with TS_COMPAT still set.
1175                  */
1176                 regs->orig_ax = value;
1177                 if (syscall_get_nr(child, regs) >= 0)
1178                         task_thread_info(child)->status |= TS_COMPAT;
1179                 break;
1180
1181         case offsetof(struct user32, regs.eflags):
1182                 return set_flags(child, value);
1183
1184         case offsetof(struct user32, u_debugreg[0]) ...
1185                 offsetof(struct user32, u_debugreg[7]):
1186                 regno -= offsetof(struct user32, u_debugreg[0]);
1187                 return ptrace_set_debugreg(child, regno / 4, value);
1188
1189         default:
1190                 if (regno > sizeof(struct user32) || (regno & 3))
1191                         return -EIO;
1192
1193                 /*
1194                  * Other dummy fields in the virtual user structure
1195                  * are ignored
1196                  */
1197                 break;
1198         }
1199         return 0;
1200 }
1201
1202 #undef R32
1203 #undef SEG32
1204
1205 #define R32(l,q)                                                        \
1206         case offsetof(struct user32, regs.l):                           \
1207                 *val = regs->q; break
1208
1209 #define SEG32(rs)                                                       \
1210         case offsetof(struct user32, regs.rs):                          \
1211                 *val = get_segment_reg(child,                           \
1212                                        offsetof(struct user_regs_struct, rs)); \
1213                 break
1214
1215 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1216 {
1217         struct pt_regs *regs = task_pt_regs(child);
1218
1219         switch (regno) {
1220
1221         SEG32(ds);
1222         SEG32(es);
1223         SEG32(fs);
1224         SEG32(gs);
1225
1226         R32(cs, cs);
1227         R32(ss, ss);
1228         R32(ebx, bx);
1229         R32(ecx, cx);
1230         R32(edx, dx);
1231         R32(edi, di);
1232         R32(esi, si);
1233         R32(ebp, bp);
1234         R32(eax, ax);
1235         R32(orig_eax, orig_ax);
1236         R32(eip, ip);
1237         R32(esp, sp);
1238
1239         case offsetof(struct user32, regs.eflags):
1240                 *val = get_flags(child);
1241                 break;
1242
1243         case offsetof(struct user32, u_debugreg[0]) ...
1244                 offsetof(struct user32, u_debugreg[7]):
1245                 regno -= offsetof(struct user32, u_debugreg[0]);
1246                 *val = ptrace_get_debugreg(child, regno / 4);
1247                 break;
1248
1249         default:
1250                 if (regno > sizeof(struct user32) || (regno & 3))
1251                         return -EIO;
1252
1253                 /*
1254                  * Other dummy fields in the virtual user structure
1255                  * are ignored
1256                  */
1257                 *val = 0;
1258                 break;
1259         }
1260         return 0;
1261 }
1262
1263 #undef R32
1264 #undef SEG32
1265
1266 static int genregs32_get(struct task_struct *target,
1267                          const struct user_regset *regset,
1268                          unsigned int pos, unsigned int count,
1269                          void *kbuf, void __user *ubuf)
1270 {
1271         if (kbuf) {
1272                 compat_ulong_t *k = kbuf;
1273                 while (count > 0) {
1274                         getreg32(target, pos, k++);
1275                         count -= sizeof(*k);
1276                         pos += sizeof(*k);
1277                 }
1278         } else {
1279                 compat_ulong_t __user *u = ubuf;
1280                 while (count > 0) {
1281                         compat_ulong_t word;
1282                         getreg32(target, pos, &word);
1283                         if (__put_user(word, u++))
1284                                 return -EFAULT;
1285                         count -= sizeof(*u);
1286                         pos += sizeof(*u);
1287                 }
1288         }
1289
1290         return 0;
1291 }
1292
1293 static int genregs32_set(struct task_struct *target,
1294                          const struct user_regset *regset,
1295                          unsigned int pos, unsigned int count,
1296                          const void *kbuf, const void __user *ubuf)
1297 {
1298         int ret = 0;
1299         if (kbuf) {
1300                 const compat_ulong_t *k = kbuf;
1301                 while (count > 0 && !ret) {
1302                         ret = putreg32(target, pos, *k++);
1303                         count -= sizeof(*k);
1304                         pos += sizeof(*k);
1305                 }
1306         } else {
1307                 const compat_ulong_t __user *u = ubuf;
1308                 while (count > 0 && !ret) {
1309                         compat_ulong_t word;
1310                         ret = __get_user(word, u++);
1311                         if (ret)
1312                                 break;
1313                         ret = putreg32(target, pos, word);
1314                         count -= sizeof(*u);
1315                         pos += sizeof(*u);
1316                 }
1317         }
1318         return ret;
1319 }
1320
1321 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1322                         compat_ulong_t caddr, compat_ulong_t cdata)
1323 {
1324         unsigned long addr = caddr;
1325         unsigned long data = cdata;
1326         void __user *datap = compat_ptr(data);
1327         int ret;
1328         __u32 val;
1329
1330         switch (request) {
1331         case PTRACE_PEEKUSR:
1332                 ret = getreg32(child, addr, &val);
1333                 if (ret == 0)
1334                         ret = put_user(val, (__u32 __user *)datap);
1335                 break;
1336
1337         case PTRACE_POKEUSR:
1338                 ret = putreg32(child, addr, data);
1339                 break;
1340
1341         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1342                 return copy_regset_to_user(child, &user_x86_32_view,
1343                                            REGSET_GENERAL,
1344                                            0, sizeof(struct user_regs_struct32),
1345                                            datap);
1346
1347         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1348                 return copy_regset_from_user(child, &user_x86_32_view,
1349                                              REGSET_GENERAL, 0,
1350                                              sizeof(struct user_regs_struct32),
1351                                              datap);
1352
1353         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1354                 return copy_regset_to_user(child, &user_x86_32_view,
1355                                            REGSET_FP, 0,
1356                                            sizeof(struct user_i387_ia32_struct),
1357                                            datap);
1358
1359         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1360                 return copy_regset_from_user(
1361                         child, &user_x86_32_view, REGSET_FP,
1362                         0, sizeof(struct user_i387_ia32_struct), datap);
1363
1364         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1365                 return copy_regset_to_user(child, &user_x86_32_view,
1366                                            REGSET_XFP, 0,
1367                                            sizeof(struct user32_fxsr_struct),
1368                                            datap);
1369
1370         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1371                 return copy_regset_from_user(child, &user_x86_32_view,
1372                                              REGSET_XFP, 0,
1373                                              sizeof(struct user32_fxsr_struct),
1374                                              datap);
1375
1376         case PTRACE_GET_THREAD_AREA:
1377         case PTRACE_SET_THREAD_AREA:
1378 #ifdef CONFIG_X86_PTRACE_BTS
1379         case PTRACE_BTS_CONFIG:
1380         case PTRACE_BTS_STATUS:
1381         case PTRACE_BTS_SIZE:
1382         case PTRACE_BTS_GET:
1383         case PTRACE_BTS_CLEAR:
1384         case PTRACE_BTS_DRAIN:
1385 #endif /* CONFIG_X86_PTRACE_BTS */
1386                 return arch_ptrace(child, request, addr, data);
1387
1388         default:
1389                 return compat_ptrace_request(child, request, addr, data);
1390         }
1391
1392         return ret;
1393 }
1394
1395 #endif  /* CONFIG_IA32_EMULATION */
1396
1397 #ifdef CONFIG_X86_64
1398
1399 static const struct user_regset x86_64_regsets[] = {
1400         [REGSET_GENERAL] = {
1401                 .core_note_type = NT_PRSTATUS,
1402                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1403                 .size = sizeof(long), .align = sizeof(long),
1404                 .get = genregs_get, .set = genregs_set
1405         },
1406         [REGSET_FP] = {
1407                 .core_note_type = NT_PRFPREG,
1408                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1409                 .size = sizeof(long), .align = sizeof(long),
1410                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1411         },
1412         [REGSET_IOPERM64] = {
1413                 .core_note_type = NT_386_IOPERM,
1414                 .n = IO_BITMAP_LONGS,
1415                 .size = sizeof(long), .align = sizeof(long),
1416                 .active = ioperm_active, .get = ioperm_get
1417         },
1418 };
1419
1420 static const struct user_regset_view user_x86_64_view = {
1421         .name = "x86_64", .e_machine = EM_X86_64,
1422         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1423 };
1424
1425 #else  /* CONFIG_X86_32 */
1426
1427 #define user_regs_struct32      user_regs_struct
1428 #define genregs32_get           genregs_get
1429 #define genregs32_set           genregs_set
1430
1431 #define user_i387_ia32_struct   user_i387_struct
1432 #define user32_fxsr_struct      user_fxsr_struct
1433
1434 #endif  /* CONFIG_X86_64 */
1435
1436 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1437 static const struct user_regset x86_32_regsets[] = {
1438         [REGSET_GENERAL] = {
1439                 .core_note_type = NT_PRSTATUS,
1440                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1441                 .size = sizeof(u32), .align = sizeof(u32),
1442                 .get = genregs32_get, .set = genregs32_set
1443         },
1444         [REGSET_FP] = {
1445                 .core_note_type = NT_PRFPREG,
1446                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1447                 .size = sizeof(u32), .align = sizeof(u32),
1448                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1449         },
1450         [REGSET_XFP] = {
1451                 .core_note_type = NT_PRXFPREG,
1452                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1453                 .size = sizeof(u32), .align = sizeof(u32),
1454                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1455         },
1456         [REGSET_TLS] = {
1457                 .core_note_type = NT_386_TLS,
1458                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1459                 .size = sizeof(struct user_desc),
1460                 .align = sizeof(struct user_desc),
1461                 .active = regset_tls_active,
1462                 .get = regset_tls_get, .set = regset_tls_set
1463         },
1464         [REGSET_IOPERM32] = {
1465                 .core_note_type = NT_386_IOPERM,
1466                 .n = IO_BITMAP_BYTES / sizeof(u32),
1467                 .size = sizeof(u32), .align = sizeof(u32),
1468                 .active = ioperm_active, .get = ioperm_get
1469         },
1470 };
1471
1472 static const struct user_regset_view user_x86_32_view = {
1473         .name = "i386", .e_machine = EM_386,
1474         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1475 };
1476 #endif
1477
1478 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1479 {
1480 #ifdef CONFIG_IA32_EMULATION
1481         if (test_tsk_thread_flag(task, TIF_IA32))
1482 #endif
1483 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1484                 return &user_x86_32_view;
1485 #endif
1486 #ifdef CONFIG_X86_64
1487         return &user_x86_64_view;
1488 #endif
1489 }
1490
1491 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1492                                          int error_code, int si_code)
1493 {
1494         struct siginfo info;
1495
1496         tsk->thread.trap_no = 1;
1497         tsk->thread.error_code = error_code;
1498
1499         memset(&info, 0, sizeof(info));
1500         info.si_signo = SIGTRAP;
1501         info.si_code = si_code;
1502
1503         /* User-mode ip? */
1504         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1505
1506         /* Send us the fake SIGTRAP */
1507         force_sig_info(SIGTRAP, &info, tsk);
1508 }
1509
1510
1511 #ifdef CONFIG_X86_32
1512 # define IS_IA32        1
1513 #elif defined CONFIG_IA32_EMULATION
1514 # define IS_IA32        is_compat_task()
1515 #else
1516 # define IS_IA32        0
1517 #endif
1518
1519 /*
1520  * We must return the syscall number to actually look up in the table.
1521  * This can be -1L to skip running any syscall at all.
1522  */
1523 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1524 {
1525         long ret = 0;
1526
1527         /*
1528          * If we stepped into a sysenter/syscall insn, it trapped in
1529          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1530          * If user-mode had set TF itself, then it's still clear from
1531          * do_debug() and we need to set it again to restore the user
1532          * state.  If we entered on the slow path, TF was already set.
1533          */
1534         if (test_thread_flag(TIF_SINGLESTEP))
1535                 regs->flags |= X86_EFLAGS_TF;
1536
1537         /* do the secure computing check first */
1538         secure_computing(regs->orig_ax);
1539
1540         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1541                 ret = -1L;
1542
1543         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1544             tracehook_report_syscall_entry(regs))
1545                 ret = -1L;
1546
1547         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1548                 trace_sys_enter(regs, regs->orig_ax);
1549
1550         if (unlikely(current->audit_context)) {
1551                 if (IS_IA32)
1552                         audit_syscall_entry(AUDIT_ARCH_I386,
1553                                             regs->orig_ax,
1554                                             regs->bx, regs->cx,
1555                                             regs->dx, regs->si);
1556 #ifdef CONFIG_X86_64
1557                 else
1558                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1559                                             regs->orig_ax,
1560                                             regs->di, regs->si,
1561                                             regs->dx, regs->r10);
1562 #endif
1563         }
1564
1565         return ret ?: regs->orig_ax;
1566 }
1567
1568 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1569 {
1570         if (unlikely(current->audit_context))
1571                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1572
1573         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1574                 trace_sys_exit(regs, regs->ax);
1575
1576         if (test_thread_flag(TIF_SYSCALL_TRACE))
1577                 tracehook_report_syscall_exit(regs, 0);
1578
1579         /*
1580          * If TIF_SYSCALL_EMU is set, we only get here because of
1581          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1582          * We already reported this syscall instruction in
1583          * syscall_trace_enter(), so don't do any more now.
1584          */
1585         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1586                 return;
1587
1588         /*
1589          * If we are single-stepping, synthesize a trap to follow the
1590          * system call instruction.
1591          */
1592         if (test_thread_flag(TIF_SINGLESTEP) &&
1593             tracehook_consider_fatal_signal(current, SIGTRAP))
1594                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1595 }