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 <trace/syscall.h>
  40
  41 #include "tls.h"
  42
  43 enum x86_regset {
  44         REGSET_GENERAL,
  45         REGSET_FP,
  46         REGSET_XFP,
  47         REGSET_IOPERM64 = REGSET_XFP,
  48         REGSET_TLS,
  49         REGSET_IOPERM32,
  50 };
  51
  52 /*
  53  * does not yet catch signals sent when the child dies.
  54  * in exit.c or in signal.c.
  55  */
  56
  57 /*
  58  * Determines which flags the user has access to [1 = access, 0 = no access].
  59  */
  60 #define FLAG_MASK_32            ((unsigned long)                        \
  61                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  62                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  63                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  64                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  65                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  66
  67 /*
  68  * Determines whether a value may be installed in a segment register.
  69  */
  70 static inline bool invalid_selector(u16 value)
  71 {
  72         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  73 }
  74
  75 #ifdef CONFIG_X86_32
  76
  77 #define FLAG_MASK               FLAG_MASK_32
  78
  79 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  80 {
  81         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  82         return &regs->bx + (regno >> 2);
  83 }
  84
  85 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  86 {
  87         /*
  88          * Returning the value truncates it to 16 bits.
  89          */
  90         unsigned int retval;
  91         if (offset != offsetof(struct user_regs_struct, gs))
  92                 retval = *pt_regs_access(task_pt_regs(task), offset);
  93         else {
  94                 if (task == current)
  95                         retval = get_user_gs(task_pt_regs(task));
  96                 else
  97                         retval = task_user_gs(task);
  98         }
  99         return retval;
 100 }
 101
 102 static int set_segment_reg(struct task_struct *task,
 103                            unsigned long offset, u16 value)
 104 {
 105         /*
 106          * The value argument was already truncated to 16 bits.
 107          */
 108         if (invalid_selector(value))
 109                 return -EIO;
 110
 111         /*
 112          * For %cs and %ss we cannot permit a null selector.
 113          * We can permit a bogus selector as long as it has USER_RPL.
 114          * Null selectors are fine for other segment registers, but
 115          * we will never get back to user mode with invalid %cs or %ss
 116          * and will take the trap in iret instead.  Much code relies
 117          * on user_mode() to distinguish a user trap frame (which can
 118          * safely use invalid selectors) from a kernel trap frame.
 119          */
 120         switch (offset) {
 121         case offsetof(struct user_regs_struct, cs):
 122         case offsetof(struct user_regs_struct, ss):
 123                 if (unlikely(value == 0))
 124                         return -EIO;
 125
 126         default:
 127                 *pt_regs_access(task_pt_regs(task), offset) = value;
 128                 break;
 129
 130         case offsetof(struct user_regs_struct, gs):
 131                 if (task == current)
 132                         set_user_gs(task_pt_regs(task), value);
 133                 else
 134                         task_user_gs(task) = value;
 135         }
 136
 137         return 0;
 138 }
 139
 140 #else  /* CONFIG_X86_64 */
 141
 142 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 143
 144 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 145 {
 146         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 147         return &regs->r15 + (offset / sizeof(regs->r15));
 148 }
 149
 150 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 151 {
 152         /*
 153          * Returning the value truncates it to 16 bits.
 154          */
 155         unsigned int seg;
 156
 157         switch (offset) {
 158         case offsetof(struct user_regs_struct, fs):
 159                 if (task == current) {
 160                         /* Older gas can't assemble movq %?s,%r?? */
 161                         asm("movl %%fs,%0" : "=r" (seg));
 162                         return seg;
 163                 }
 164                 return task->thread.fsindex;
 165         case offsetof(struct user_regs_struct, gs):
 166                 if (task == current) {
 167                         asm("movl %%gs,%0" : "=r" (seg));
 168                         return seg;
 169                 }
 170                 return task->thread.gsindex;
 171         case offsetof(struct user_regs_struct, ds):
 172                 if (task == current) {
 173                         asm("movl %%ds,%0" : "=r" (seg));
 174                         return seg;
 175                 }
 176                 return task->thread.ds;
 177         case offsetof(struct user_regs_struct, es):
 178                 if (task == current) {
 179                         asm("movl %%es,%0" : "=r" (seg));
 180                         return seg;
 181                 }
 182                 return task->thread.es;
 183
 184         case offsetof(struct user_regs_struct, cs):
 185         case offsetof(struct user_regs_struct, ss):
 186                 break;
 187         }
 188         return *pt_regs_access(task_pt_regs(task), offset);
 189 }
 190
 191 static int set_segment_reg(struct task_struct *task,
 192                            unsigned long offset, u16 value)
 193 {
 194         /*
 195          * The value argument was already truncated to 16 bits.
 196          */
 197         if (invalid_selector(value))
 198                 return -EIO;
 199
 200         switch (offset) {
 201         case offsetof(struct user_regs_struct,fs):
 202                 /*
 203                  * If this is setting fs as for normal 64-bit use but
 204                  * setting fs_base has implicitly changed it, leave it.
 205                  */
 206                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 207                      task->thread.fs != 0) ||
 208                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 209                      task->thread.fs == 0))
 210                         break;
 211                 task->thread.fsindex = value;
 212                 if (task == current)
 213                         loadsegment(fs, task->thread.fsindex);
 214                 break;
 215         case offsetof(struct user_regs_struct,gs):
 216                 /*
 217                  * If this is setting gs as for normal 64-bit use but
 218                  * setting gs_base has implicitly changed it, leave it.
 219                  */
 220                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 221                      task->thread.gs != 0) ||
 222                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 223                      task->thread.gs == 0))
 224                         break;
 225                 task->thread.gsindex = value;
 226                 if (task == current)
 227                         load_gs_index(task->thread.gsindex);
 228                 break;
 229         case offsetof(struct user_regs_struct,ds):
 230                 task->thread.ds = value;
 231                 if (task == current)
 232                         loadsegment(ds, task->thread.ds);
 233                 break;
 234         case offsetof(struct user_regs_struct,es):
 235                 task->thread.es = value;
 236                 if (task == current)
 237                         loadsegment(es, task->thread.es);
 238                 break;
 239
 240                 /*
 241                  * Can't actually change these in 64-bit mode.
 242                  */
 243         case offsetof(struct user_regs_struct,cs):
 244                 if (unlikely(value == 0))
 245                         return -EIO;
 246 #ifdef CONFIG_IA32_EMULATION
 247                 if (test_tsk_thread_flag(task, TIF_IA32))
 248                         task_pt_regs(task)->cs = value;
 249 #endif
 250                 break;
 251         case offsetof(struct user_regs_struct,ss):
 252                 if (unlikely(value == 0))
 253                         return -EIO;
 254 #ifdef CONFIG_IA32_EMULATION
 255                 if (test_tsk_thread_flag(task, TIF_IA32))
 256                         task_pt_regs(task)->ss = value;
 257 #endif
 258                 break;
 259         }
 260
 261         return 0;
 262 }
 263
 264 #endif  /* CONFIG_X86_32 */
 265
 266 static unsigned long get_flags(struct task_struct *task)
 267 {
 268         unsigned long retval = task_pt_regs(task)->flags;
 269
 270         /*
 271          * If the debugger set TF, hide it from the readout.
 272          */
 273         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 274                 retval &= ~X86_EFLAGS_TF;
 275
 276         return retval;
 277 }
 278
 279 static int set_flags(struct task_struct *task, unsigned long value)
 280 {
 281         struct pt_regs *regs = task_pt_regs(task);
 282
 283         /*
 284          * If the user value contains TF, mark that
 285          * it was not "us" (the debugger) that set it.
 286          * If not, make sure it stays set if we had.
 287          */
 288         if (value & X86_EFLAGS_TF)
 289                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 290         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 291                 value |= X86_EFLAGS_TF;
 292
 293         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 294
 295         return 0;
 296 }
 297
 298 static int putreg(struct task_struct *child,
 299                   unsigned long offset, unsigned long value)
 300 {
 301         switch (offset) {
 302         case offsetof(struct user_regs_struct, cs):
 303         case offsetof(struct user_regs_struct, ds):
 304         case offsetof(struct user_regs_struct, es):
 305         case offsetof(struct user_regs_struct, fs):
 306         case offsetof(struct user_regs_struct, gs):
 307         case offsetof(struct user_regs_struct, ss):
 308                 return set_segment_reg(child, offset, value);
 309
 310         case offsetof(struct user_regs_struct, flags):
 311                 return set_flags(child, value);
 312
 313 #ifdef CONFIG_X86_64
 314         /*
 315          * Orig_ax is really just a flag with small positive and
 316          * negative values, so make sure to always sign-extend it
 317          * from 32 bits so that it works correctly regardless of
 318          * whether we come from a 32-bit environment or not.
 319          */
 320         case offsetof(struct user_regs_struct, orig_ax):
 321                 value = (long) (s32) value;
 322                 break;
 323
 324         case offsetof(struct user_regs_struct,fs_base):
 325                 if (value >= TASK_SIZE_OF(child))
 326                         return -EIO;
 327                 /*
 328                  * When changing the segment base, use do_arch_prctl
 329                  * to set either thread.fs or thread.fsindex and the
 330                  * corresponding GDT slot.
 331                  */
 332                 if (child->thread.fs != value)
 333                         return do_arch_prctl(child, ARCH_SET_FS, value);
 334                 return 0;
 335         case offsetof(struct user_regs_struct,gs_base):
 336                 /*
 337                  * Exactly the same here as the %fs handling above.
 338                  */
 339                 if (value >= TASK_SIZE_OF(child))
 340                         return -EIO;
 341                 if (child->thread.gs != value)
 342                         return do_arch_prctl(child, ARCH_SET_GS, value);
 343                 return 0;
 344 #endif
 345         }
 346
 347         *pt_regs_access(task_pt_regs(child), offset) = value;
 348         return 0;
 349 }
 350
 351 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 352 {
 353         switch (offset) {
 354         case offsetof(struct user_regs_struct, cs):
 355         case offsetof(struct user_regs_struct, ds):
 356         case offsetof(struct user_regs_struct, es):
 357         case offsetof(struct user_regs_struct, fs):
 358         case offsetof(struct user_regs_struct, gs):
 359         case offsetof(struct user_regs_struct, ss):
 360                 return get_segment_reg(task, offset);
 361
 362         case offsetof(struct user_regs_struct, flags):
 363                 return get_flags(task);
 364
 365 #ifdef CONFIG_X86_64
 366         case offsetof(struct user_regs_struct, fs_base): {
 367                 /*
 368                  * do_arch_prctl may have used a GDT slot instead of
 369                  * the MSR.  To userland, it appears the same either
 370                  * way, except the %fs segment selector might not be 0.
 371                  */
 372                 unsigned int seg = task->thread.fsindex;
 373                 if (task->thread.fs != 0)
 374                         return task->thread.fs;
 375                 if (task == current)
 376                         asm("movl %%fs,%0" : "=r" (seg));
 377                 if (seg != FS_TLS_SEL)
 378                         return 0;
 379                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 380         }
 381         case offsetof(struct user_regs_struct, gs_base): {
 382                 /*
 383                  * Exactly the same here as the %fs handling above.
 384                  */
 385                 unsigned int seg = task->thread.gsindex;
 386                 if (task->thread.gs != 0)
 387                         return task->thread.gs;
 388                 if (task == current)
 389                         asm("movl %%gs,%0" : "=r" (seg));
 390                 if (seg != GS_TLS_SEL)
 391                         return 0;
 392                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 393         }
 394 #endif
 395         }
 396
 397         return *pt_regs_access(task_pt_regs(task), offset);
 398 }
 399
 400 static int genregs_get(struct task_struct *target,
 401                        const struct user_regset *regset,
 402                        unsigned int pos, unsigned int count,
 403                        void *kbuf, void __user *ubuf)
 404 {
 405         if (kbuf) {
 406                 unsigned long *k = kbuf;
 407                 while (count > 0) {
 408                         *k++ = getreg(target, pos);
 409                         count -= sizeof(*k);
 410                         pos += sizeof(*k);
 411                 }
 412         } else {
 413                 unsigned long __user *u = ubuf;
 414                 while (count > 0) {
 415                         if (__put_user(getreg(target, pos), u++))
 416                                 return -EFAULT;
 417                         count -= sizeof(*u);
 418                         pos += sizeof(*u);
 419                 }
 420         }
 421
 422         return 0;
 423 }
 424
 425 static int genregs_set(struct task_struct *target,
 426                        const struct user_regset *regset,
 427                        unsigned int pos, unsigned int count,
 428                        const void *kbuf, const void __user *ubuf)
 429 {
 430         int ret = 0;
 431         if (kbuf) {
 432                 const unsigned long *k = kbuf;
 433                 while (count > 0 && !ret) {
 434                         ret = putreg(target, pos, *k++);
 435                         count -= sizeof(*k);
 436                         pos += sizeof(*k);
 437                 }
 438         } else {
 439                 const unsigned long  __user *u = ubuf;
 440                 while (count > 0 && !ret) {
 441                         unsigned long word;
 442                         ret = __get_user(word, u++);
 443                         if (ret)
 444                                 break;
 445                         ret = putreg(target, pos, word);
 446                         count -= sizeof(*u);
 447                         pos += sizeof(*u);
 448                 }
 449         }
 450         return ret;
 451 }
 452
 453 /*
 454  * Decode the length and type bits for a particular breakpoint as
 455  * stored in debug register 7.  Return the "enabled" status.
 456  */
 457 static int decode_dr7(unsigned long dr7, int bpnum, unsigned *len,
 458                 unsigned *type)
 459 {
 460         int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE);
 461
 462         *len = (bp_info & 0xc) | 0x40;
 463         *type = (bp_info & 0x3) | 0x80;
 464         return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3;
 465 }
 466
 467 static void ptrace_triggered(struct hw_breakpoint *bp, struct pt_regs *regs)
 468 {
 469         struct thread_struct *thread = &(current->thread);
 470         int i;
 471
 472         /*
 473          * Store in the virtual DR6 register the fact that the breakpoint
 474          * was hit so the thread's debugger will see it.
 475          */
 476         for (i = 0; i < hbp_kernel_pos; i++)
 477                 /*
 478                  * We will check bp->info.address against the address stored in
 479                  * thread's hbp structure and not debugreg[i]. This is to ensure
 480                  * that the corresponding bit for 'i' in DR7 register is enabled
 481                  */
 482                 if (bp->info.address == thread->hbp[i]->info.address)
 483                         break;
 484
 485         thread->debugreg6 |= (DR_TRAP0 << i);
 486 }
 487
 488 /*
 489  * Handle ptrace writes to debug register 7.
 490  */
 491 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 492 {
 493         struct thread_struct *thread = &(tsk->thread);
 494         unsigned long old_dr7 = thread->debugreg7;
 495         int i, orig_ret = 0, rc = 0;
 496         int enabled, second_pass = 0;
 497         unsigned len, type;
 498         struct hw_breakpoint *bp;
 499
 500         data &= ~DR_CONTROL_RESERVED;
 501 restore:
 502         /*
 503          * Loop through all the hardware breakpoints, making the
 504          * appropriate changes to each.
 505          */
 506         for (i = 0; i < HBP_NUM; i++) {
 507                 enabled = decode_dr7(data, i, &len, &type);
 508                 bp = thread->hbp[i];
 509
 510                 if (!enabled) {
 511                         if (bp) {
 512                                 /* Don't unregister the breakpoints right-away,
 513                                  * unless all register_user_hw_breakpoint()
 514                                  * requests have succeeded. This prevents
 515                                  * any window of opportunity for debug
 516                                  * register grabbing by other users.
 517                                  */
 518                                 if (!second_pass)
 519                                         continue;
 520                                 unregister_user_hw_breakpoint(tsk, bp);
 521                                 kfree(bp);
 522                         }
 523                         continue;
 524                 }
 525                 if (!bp) {
 526                         rc = -ENOMEM;
 527                         bp = kzalloc(sizeof(struct hw_breakpoint), GFP_KERNEL);
 528                         if (bp) {
 529                                 bp->info.address = thread->debugreg[i];
 530                                 bp->triggered = ptrace_triggered;
 531                                 bp->info.len = len;
 532                                 bp->info.type = type;
 533                                 rc = register_user_hw_breakpoint(tsk, bp);
 534                                 if (rc)
 535                                         kfree(bp);
 536                         }
 537                 } else
 538                         rc = modify_user_hw_breakpoint(tsk, bp);
 539                 if (rc)
 540                         break;
 541         }
 542         /*
 543          * Make a second pass to free the remaining unused breakpoints
 544          * or to restore the original breakpoints if an error occurred.
 545          */
 546         if (!second_pass) {
 547                 second_pass = 1;
 548                 if (rc < 0) {
 549                         orig_ret = rc;
 550                         data = old_dr7;
 551                 }
 552                 goto restore;
 553         }
 554         return ((orig_ret < 0) ? orig_ret : rc);
 555 }
 556
 557 /*
 558  * Handle PTRACE_PEEKUSR calls for the debug register area.
 559  */
 560 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 561 {
 562         struct thread_struct *thread = &(tsk->thread);
 563         unsigned long val = 0;
 564
 565         if (n < HBP_NUM)
 566                 val = thread->debugreg[n];
 567         else if (n == 6)
 568                 val = thread->debugreg6;
 569         else if (n == 7)
 570                 val = thread->debugreg7;
 571         return val;
 572 }
 573
 574 /*
 575  * Handle PTRACE_POKEUSR calls for the debug register area.
 576  */
 577 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
 578 {
 579         struct thread_struct *thread = &(tsk->thread);
 580         int rc = 0;
 581
 582         /* There are no DR4 or DR5 registers */
 583         if (n == 4 || n == 5)
 584                 return -EIO;
 585
 586         if (n == 6) {
 587                 tsk->thread.debugreg6 = val;
 588                 goto ret_path;
 589         }
 590         if (n < HBP_NUM) {
 591                 if (thread->hbp[n]) {
 592                         if (arch_check_va_in_userspace(val,
 593                                         thread->hbp[n]->info.len) == 0) {
 594                                 rc = -EIO;
 595                                 goto ret_path;
 596                         }
 597                         thread->hbp[n]->info.address = val;
 598                 }
 599                 thread->debugreg[n] = val;
 600         }
 601         /* All that's left is DR7 */
 602         if (n == 7)
 603                 rc = ptrace_write_dr7(tsk, val);
 604
 605 ret_path:
 606         return rc;
 607 }
 608
 609 /*
 610  * These access the current or another (stopped) task's io permission
 611  * bitmap for debugging or core dump.
 612  */
 613 static int ioperm_active(struct task_struct *target,
 614                          const struct user_regset *regset)
 615 {
 616         return target->thread.io_bitmap_max / regset->size;
 617 }
 618
 619 static int ioperm_get(struct task_struct *target,
 620                       const struct user_regset *regset,
 621                       unsigned int pos, unsigned int count,
 622                       void *kbuf, void __user *ubuf)
 623 {
 624         if (!target->thread.io_bitmap_ptr)
 625                 return -ENXIO;
 626
 627         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 628                                    target->thread.io_bitmap_ptr,
 629                                    0, IO_BITMAP_BYTES);
 630 }
 631
 632 #ifdef CONFIG_X86_PTRACE_BTS
 633 /*
 634  * A branch trace store context.
 635  *
 636  * Contexts may only be installed by ptrace_bts_config() and only for
 637  * ptraced tasks.
 638  *
 639  * Contexts are destroyed when the tracee is detached from the tracer.
 640  * The actual destruction work requires interrupts enabled, so the
 641  * work is deferred and will be scheduled during __ptrace_unlink().
 642  *
 643  * Contexts hold an additional task_struct reference on the traced
 644  * task, as well as a reference on the tracer's mm.
 645  *
 646  * Ptrace already holds a task_struct for the duration of ptrace operations,
 647  * but since destruction is deferred, it may be executed after both
 648  * tracer and tracee exited.
 649  */
 650 struct bts_context {
 651         /* The branch trace handle. */
 652         struct bts_tracer       *tracer;
 653
 654         /* The buffer used to store the branch trace and its size. */
 655         void                    *buffer;
 656         unsigned int            size;
 657
 658         /* The mm that paid for the above buffer. */
 659         struct mm_struct        *mm;
 660
 661         /* The task this context belongs to. */
 662         struct task_struct      *task;
 663
 664         /* The signal to send on a bts buffer overflow. */
 665         unsigned int            bts_ovfl_signal;
 666
 667         /* The work struct to destroy a context. */
 668         struct work_struct      work;
 669 };
 670
 671 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
 672 {
 673         void *buffer = NULL;
 674         int err = -ENOMEM;
 675
 676         err = account_locked_memory(current->mm, current->signal->rlim, size);
 677         if (err < 0)
 678                 return err;
 679
 680         buffer = kzalloc(size, GFP_KERNEL);
 681         if (!buffer)
 682                 goto out_refund;
 683
 684         context->buffer = buffer;
 685         context->size = size;
 686         context->mm = get_task_mm(current);
 687
 688         return 0;
 689
 690  out_refund:
 691         refund_locked_memory(current->mm, size);
 692         return err;
 693 }
 694
 695 static inline void free_bts_buffer(struct bts_context *context)
 696 {
 697         if (!context->buffer)
 698                 return;
 699
 700         kfree(context->buffer);
 701         context->buffer = NULL;
 702
 703         refund_locked_memory(context->mm, context->size);
 704         context->size = 0;
 705
 706         mmput(context->mm);
 707         context->mm = NULL;
 708 }
 709
 710 static void free_bts_context_work(struct work_struct *w)
 711 {
 712         struct bts_context *context;
 713
 714         context = container_of(w, struct bts_context, work);
 715
 716         ds_release_bts(context->tracer);
 717         put_task_struct(context->task);
 718         free_bts_buffer(context);
 719         kfree(context);
 720 }
 721
 722 static inline void free_bts_context(struct bts_context *context)
 723 {
 724         INIT_WORK(&context->work, free_bts_context_work);
 725         schedule_work(&context->work);
 726 }
 727
 728 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
 729 {
 730         struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
 731         if (context) {
 732                 context->task = task;
 733                 task->bts = context;
 734
 735                 get_task_struct(task);
 736         }
 737
 738         return context;
 739 }
 740
 741 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 742                                   struct bts_struct __user *out)
 743 {
 744         struct bts_context *context;
 745         const struct bts_trace *trace;
 746         struct bts_struct bts;
 747         const unsigned char *at;
 748         int error;
 749
 750         context = child->bts;
 751         if (!context)
 752                 return -ESRCH;
 753
 754         trace = ds_read_bts(context->tracer);
 755         if (!trace)
 756                 return -ESRCH;
 757
 758         at = trace->ds.top - ((index + 1) * trace->ds.size);
 759         if ((void *)at < trace->ds.begin)
 760                 at += (trace->ds.n * trace->ds.size);
 761
 762         if (!trace->read)
 763                 return -EOPNOTSUPP;
 764
 765         error = trace->read(context->tracer, at, &bts);
 766         if (error < 0)
 767                 return error;
 768
 769         if (copy_to_user(out, &bts, sizeof(bts)))
 770                 return -EFAULT;
 771
 772         return sizeof(bts);
 773 }
 774
 775 static int ptrace_bts_drain(struct task_struct *child,
 776                             long size,
 777                             struct bts_struct __user *out)
 778 {
 779         struct bts_context *context;
 780         const struct bts_trace *trace;
 781         const unsigned char *at;
 782         int error, drained = 0;
 783
 784         context = child->bts;
 785         if (!context)
 786                 return -ESRCH;
 787
 788         trace = ds_read_bts(context->tracer);
 789         if (!trace)
 790                 return -ESRCH;
 791
 792         if (!trace->read)
 793                 return -EOPNOTSUPP;
 794
 795         if (size < (trace->ds.top - trace->ds.begin))
 796                 return -EIO;
 797
 798         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 799              out++, drained++, at += trace->ds.size) {
 800                 struct bts_struct bts;
 801
 802                 error = trace->read(context->tracer, at, &bts);
 803                 if (error < 0)
 804                         return error;
 805
 806                 if (copy_to_user(out, &bts, sizeof(bts)))
 807                         return -EFAULT;
 808         }
 809
 810         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 811
 812         error = ds_reset_bts(context->tracer);
 813         if (error < 0)
 814                 return error;
 815
 816         return drained;
 817 }
 818
 819 static int ptrace_bts_config(struct task_struct *child,
 820                              long cfg_size,
 821                              const struct ptrace_bts_config __user *ucfg)
 822 {
 823         struct bts_context *context;
 824         struct ptrace_bts_config cfg;
 825         unsigned int flags = 0;
 826
 827         if (cfg_size < sizeof(cfg))
 828                 return -EIO;
 829
 830         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 831                 return -EFAULT;
 832
 833         context = child->bts;
 834         if (!context)
 835                 context = alloc_bts_context(child);
 836         if (!context)
 837                 return -ENOMEM;
 838
 839         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 840                 if (!cfg.signal)
 841                         return -EINVAL;
 842
 843                 return -EOPNOTSUPP;
 844                 context->bts_ovfl_signal = cfg.signal;
 845         }
 846
 847         ds_release_bts(context->tracer);
 848         context->tracer = NULL;
 849
 850         if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
 851                 int err;
 852
 853                 free_bts_buffer(context);
 854                 if (!cfg.size)
 855                         return 0;
 856
 857                 err = alloc_bts_buffer(context, cfg.size);
 858                 if (err < 0)
 859                         return err;
 860         }
 861
 862         if (cfg.flags & PTRACE_BTS_O_TRACE)
 863                 flags |= BTS_USER;
 864
 865         if (cfg.flags & PTRACE_BTS_O_SCHED)
 866                 flags |= BTS_TIMESTAMPS;
 867
 868         context->tracer =
 869                 ds_request_bts_task(child, context->buffer, context->size,
 870                                     NULL, (size_t)-1, flags);
 871         if (unlikely(IS_ERR(context->tracer))) {
 872                 int error = PTR_ERR(context->tracer);
 873
 874                 free_bts_buffer(context);
 875                 context->tracer = NULL;
 876                 return error;
 877         }
 878
 879         return sizeof(cfg);
 880 }
 881
 882 static int ptrace_bts_status(struct task_struct *child,
 883                              long cfg_size,
 884                              struct ptrace_bts_config __user *ucfg)
 885 {
 886         struct bts_context *context;
 887         const struct bts_trace *trace;
 888         struct ptrace_bts_config cfg;
 889
 890         context = child->bts;
 891         if (!context)
 892                 return -ESRCH;
 893
 894         if (cfg_size < sizeof(cfg))
 895                 return -EIO;
 896
 897         trace = ds_read_bts(context->tracer);
 898         if (!trace)
 899                 return -ESRCH;
 900
 901         memset(&cfg, 0, sizeof(cfg));
 902         cfg.size        = trace->ds.end - trace->ds.begin;
 903         cfg.signal      = context->bts_ovfl_signal;
 904         cfg.bts_size    = sizeof(struct bts_struct);
 905
 906         if (cfg.signal)
 907                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 908
 909         if (trace->ds.flags & BTS_USER)
 910                 cfg.flags |= PTRACE_BTS_O_TRACE;
 911
 912         if (trace->ds.flags & BTS_TIMESTAMPS)
 913                 cfg.flags |= PTRACE_BTS_O_SCHED;
 914
 915         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 916                 return -EFAULT;
 917
 918         return sizeof(cfg);
 919 }
 920
 921 static int ptrace_bts_clear(struct task_struct *child)
 922 {
 923         struct bts_context *context;
 924         const struct bts_trace *trace;
 925
 926         context = child->bts;
 927         if (!context)
 928                 return -ESRCH;
 929
 930         trace = ds_read_bts(context->tracer);
 931         if (!trace)
 932                 return -ESRCH;
 933
 934         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 935
 936         return ds_reset_bts(context->tracer);
 937 }
 938
 939 static int ptrace_bts_size(struct task_struct *child)
 940 {
 941         struct bts_context *context;
 942         const struct bts_trace *trace;
 943
 944         context = child->bts;
 945         if (!context)
 946                 return -ESRCH;
 947
 948         trace = ds_read_bts(context->tracer);
 949         if (!trace)
 950                 return -ESRCH;
 951
 952         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 953 }
 954
 955 /*
 956  * Called from __ptrace_unlink() after the child has been moved back
 957  * to its original parent.
 958  */
 959 void ptrace_bts_untrace(struct task_struct *child)
 960 {
 961         if (unlikely(child->bts)) {
 962                 free_bts_context(child->bts);
 963                 child->bts = NULL;
 964         }
 965 }
 966 #endif /* CONFIG_X86_PTRACE_BTS */
 967
 968 /*
 969  * Called by kernel/ptrace.c when detaching..
 970  *
 971  * Make sure the single step bit is not set.
 972  */
 973 void ptrace_disable(struct task_struct *child)
 974 {
 975         user_disable_single_step(child);
 976 #ifdef TIF_SYSCALL_EMU
 977         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 978 #endif
 979 }
 980
 981 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 982 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 983 #endif
 984
 985 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 986 {
 987         int ret;
 988         unsigned long __user *datap = (unsigned long __user *)data;
 989
 990         switch (request) {
 991         /* read the word at location addr in the USER area. */
 992         case PTRACE_PEEKUSR: {
 993                 unsigned long tmp;
 994
 995                 ret = -EIO;
 996                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 997                     addr >= sizeof(struct user))
 998                         break;
 999
1000                 tmp = 0;  /* Default return condition */
1001                 if (addr < sizeof(struct user_regs_struct))
1002                         tmp = getreg(child, addr);
1003                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1004                          addr <= offsetof(struct user, u_debugreg[7])) {
1005                         addr -= offsetof(struct user, u_debugreg[0]);
1006                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1007                 }
1008                 ret = put_user(tmp, datap);
1009                 break;
1010         }
1011
1012         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1013                 ret = -EIO;
1014                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1015                     addr >= sizeof(struct user))
1016                         break;
1017
1018                 if (addr < sizeof(struct user_regs_struct))
1019                         ret = putreg(child, addr, data);
1020                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1021                          addr <= offsetof(struct user, u_debugreg[7])) {
1022                         addr -= offsetof(struct user, u_debugreg[0]);
1023                         ret = ptrace_set_debugreg(child,
1024                                                   addr / sizeof(data), data);
1025                 }
1026                 break;
1027
1028         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1029                 return copy_regset_to_user(child,
1030                                            task_user_regset_view(current),
1031                                            REGSET_GENERAL,
1032                                            0, sizeof(struct user_regs_struct),
1033                                            datap);
1034
1035         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1036                 return copy_regset_from_user(child,
1037                                              task_user_regset_view(current),
1038                                              REGSET_GENERAL,
1039                                              0, sizeof(struct user_regs_struct),
1040                                              datap);
1041
1042         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1043                 return copy_regset_to_user(child,
1044                                            task_user_regset_view(current),
1045                                            REGSET_FP,
1046                                            0, sizeof(struct user_i387_struct),
1047                                            datap);
1048
1049         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1050                 return copy_regset_from_user(child,
1051                                              task_user_regset_view(current),
1052                                              REGSET_FP,
1053                                              0, sizeof(struct user_i387_struct),
1054                                              datap);
1055
1056 #ifdef CONFIG_X86_32
1057         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1058                 return copy_regset_to_user(child, &user_x86_32_view,
1059                                            REGSET_XFP,
1060                                            0, sizeof(struct user_fxsr_struct),
1061                                            datap) ? -EIO : 0;
1062
1063         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1064                 return copy_regset_from_user(child, &user_x86_32_view,
1065                                              REGSET_XFP,
1066                                              0, sizeof(struct user_fxsr_struct),
1067                                              datap) ? -EIO : 0;
1068 #endif
1069
1070 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1071         case PTRACE_GET_THREAD_AREA:
1072                 if (addr < 0)
1073                         return -EIO;
1074                 ret = do_get_thread_area(child, addr,
1075                                          (struct user_desc __user *) data);
1076                 break;
1077
1078         case PTRACE_SET_THREAD_AREA:
1079                 if (addr < 0)
1080                         return -EIO;
1081                 ret = do_set_thread_area(child, addr,
1082                                          (struct user_desc __user *) data, 0);
1083                 break;
1084 #endif
1085
1086 #ifdef CONFIG_X86_64
1087                 /* normal 64bit interface to access TLS data.
1088                    Works just like arch_prctl, except that the arguments
1089                    are reversed. */
1090         case PTRACE_ARCH_PRCTL:
1091                 ret = do_arch_prctl(child, data, addr);
1092                 break;
1093 #endif
1094
1095         /*
1096          * These bits need more cooking - not enabled yet:
1097          */
1098 #ifdef CONFIG_X86_PTRACE_BTS
1099         case PTRACE_BTS_CONFIG:
1100                 ret = ptrace_bts_config
1101                         (child, data, (struct ptrace_bts_config __user *)addr);
1102                 break;
1103
1104         case PTRACE_BTS_STATUS:
1105                 ret = ptrace_bts_status
1106                         (child, data, (struct ptrace_bts_config __user *)addr);
1107                 break;
1108
1109         case PTRACE_BTS_SIZE:
1110                 ret = ptrace_bts_size(child);
1111                 break;
1112
1113         case PTRACE_BTS_GET:
1114                 ret = ptrace_bts_read_record
1115                         (child, data, (struct bts_struct __user *) addr);
1116                 break;
1117
1118         case PTRACE_BTS_CLEAR:
1119                 ret = ptrace_bts_clear(child);
1120                 break;
1121
1122         case PTRACE_BTS_DRAIN:
1123                 ret = ptrace_bts_drain
1124                         (child, data, (struct bts_struct __user *) addr);
1125                 break;
1126 #endif /* CONFIG_X86_PTRACE_BTS */
1127
1128         default:
1129                 ret = ptrace_request(child, request, addr, data);
1130                 break;
1131         }
1132
1133         return ret;
1134 }
1135
1136 #ifdef CONFIG_IA32_EMULATION
1137
1138 #include <linux/compat.h>
1139 #include <linux/syscalls.h>
1140 #include <asm/ia32.h>
1141 #include <asm/user32.h>
1142
1143 #define R32(l,q)                                                        \
1144         case offsetof(struct user32, regs.l):                           \
1145                 regs->q = value; break
1146
1147 #define SEG32(rs)                                                       \
1148         case offsetof(struct user32, regs.rs):                          \
1149                 return set_segment_reg(child,                           \
1150                                        offsetof(struct user_regs_struct, rs), \
1151                                        value);                          \
1152                 break
1153
1154 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1155 {
1156         struct pt_regs *regs = task_pt_regs(child);
1157
1158         switch (regno) {
1159
1160         SEG32(cs);
1161         SEG32(ds);
1162         SEG32(es);
1163         SEG32(fs);
1164         SEG32(gs);
1165         SEG32(ss);
1166
1167         R32(ebx, bx);
1168         R32(ecx, cx);
1169         R32(edx, dx);
1170         R32(edi, di);
1171         R32(esi, si);
1172         R32(ebp, bp);
1173         R32(eax, ax);
1174         R32(eip, ip);
1175         R32(esp, sp);
1176
1177         case offsetof(struct user32, regs.orig_eax):
1178                 /*
1179                  * Sign-extend the value so that orig_eax = -1
1180                  * causes (long)orig_ax < 0 tests to fire correctly.
1181                  */
1182                 regs->orig_ax = (long) (s32) value;
1183                 break;
1184
1185         case offsetof(struct user32, regs.eflags):
1186                 return set_flags(child, value);
1187
1188         case offsetof(struct user32, u_debugreg[0]) ...
1189                 offsetof(struct user32, u_debugreg[7]):
1190                 regno -= offsetof(struct user32, u_debugreg[0]);
1191                 return ptrace_set_debugreg(child, regno / 4, value);
1192
1193         default:
1194                 if (regno > sizeof(struct user32) || (regno & 3))
1195                         return -EIO;
1196
1197                 /*
1198                  * Other dummy fields in the virtual user structure
1199                  * are ignored
1200                  */
1201                 break;
1202         }
1203         return 0;
1204 }
1205
1206 #undef R32
1207 #undef SEG32
1208
1209 #define R32(l,q)                                                        \
1210         case offsetof(struct user32, regs.l):                           \
1211                 *val = regs->q; break
1212
1213 #define SEG32(rs)                                                       \
1214         case offsetof(struct user32, regs.rs):                          \
1215                 *val = get_segment_reg(child,                           \
1216                                        offsetof(struct user_regs_struct, rs)); \
1217                 break
1218
1219 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1220 {
1221         struct pt_regs *regs = task_pt_regs(child);
1222
1223         switch (regno) {
1224
1225         SEG32(ds);
1226         SEG32(es);
1227         SEG32(fs);
1228         SEG32(gs);
1229
1230         R32(cs, cs);
1231         R32(ss, ss);
1232         R32(ebx, bx);
1233         R32(ecx, cx);
1234         R32(edx, dx);
1235         R32(edi, di);
1236         R32(esi, si);
1237         R32(ebp, bp);
1238         R32(eax, ax);
1239         R32(orig_eax, orig_ax);
1240         R32(eip, ip);
1241         R32(esp, sp);
1242
1243         case offsetof(struct user32, regs.eflags):
1244                 *val = get_flags(child);
1245                 break;
1246
1247         case offsetof(struct user32, u_debugreg[0]) ...
1248                 offsetof(struct user32, u_debugreg[7]):
1249                 regno -= offsetof(struct user32, u_debugreg[0]);
1250                 *val = ptrace_get_debugreg(child, regno / 4);
1251                 break;
1252
1253         default:
1254                 if (regno > sizeof(struct user32) || (regno & 3))
1255                         return -EIO;
1256
1257                 /*
1258                  * Other dummy fields in the virtual user structure
1259                  * are ignored
1260                  */
1261                 *val = 0;
1262                 break;
1263         }
1264         return 0;
1265 }
1266
1267 #undef R32
1268 #undef SEG32
1269
1270 static int genregs32_get(struct task_struct *target,
1271                          const struct user_regset *regset,
1272                          unsigned int pos, unsigned int count,
1273                          void *kbuf, void __user *ubuf)
1274 {
1275         if (kbuf) {
1276                 compat_ulong_t *k = kbuf;
1277                 while (count > 0) {
1278                         getreg32(target, pos, k++);
1279                         count -= sizeof(*k);
1280                         pos += sizeof(*k);
1281                 }
1282         } else {
1283                 compat_ulong_t __user *u = ubuf;
1284                 while (count > 0) {
1285                         compat_ulong_t word;
1286                         getreg32(target, pos, &word);
1287                         if (__put_user(word, u++))
1288                                 return -EFAULT;
1289                         count -= sizeof(*u);
1290                         pos += sizeof(*u);
1291                 }
1292         }
1293
1294         return 0;
1295 }
1296
1297 static int genregs32_set(struct task_struct *target,
1298                          const struct user_regset *regset,
1299                          unsigned int pos, unsigned int count,
1300                          const void *kbuf, const void __user *ubuf)
1301 {
1302         int ret = 0;
1303         if (kbuf) {
1304                 const compat_ulong_t *k = kbuf;
1305                 while (count > 0 && !ret) {
1306                         ret = putreg32(target, pos, *k++);
1307                         count -= sizeof(*k);
1308                         pos += sizeof(*k);
1309                 }
1310         } else {
1311                 const compat_ulong_t __user *u = ubuf;
1312                 while (count > 0 && !ret) {
1313                         compat_ulong_t word;
1314                         ret = __get_user(word, u++);
1315                         if (ret)
1316                                 break;
1317                         ret = putreg32(target, pos, word);
1318                         count -= sizeof(*u);
1319                         pos += sizeof(*u);
1320                 }
1321         }
1322         return ret;
1323 }
1324
1325 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1326                         compat_ulong_t caddr, compat_ulong_t cdata)
1327 {
1328         unsigned long addr = caddr;
1329         unsigned long data = cdata;
1330         void __user *datap = compat_ptr(data);
1331         int ret;
1332         __u32 val;
1333
1334         switch (request) {
1335         case PTRACE_PEEKUSR:
1336                 ret = getreg32(child, addr, &val);
1337                 if (ret == 0)
1338                         ret = put_user(val, (__u32 __user *)datap);
1339                 break;
1340
1341         case PTRACE_POKEUSR:
1342                 ret = putreg32(child, addr, data);
1343                 break;
1344
1345         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1346                 return copy_regset_to_user(child, &user_x86_32_view,
1347                                            REGSET_GENERAL,
1348                                            0, sizeof(struct user_regs_struct32),
1349                                            datap);
1350
1351         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1352                 return copy_regset_from_user(child, &user_x86_32_view,
1353                                              REGSET_GENERAL, 0,
1354                                              sizeof(struct user_regs_struct32),
1355                                              datap);
1356
1357         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1358                 return copy_regset_to_user(child, &user_x86_32_view,
1359                                            REGSET_FP, 0,
1360                                            sizeof(struct user_i387_ia32_struct),
1361                                            datap);
1362
1363         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1364                 return copy_regset_from_user(
1365                         child, &user_x86_32_view, REGSET_FP,
1366                         0, sizeof(struct user_i387_ia32_struct), datap);
1367
1368         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1369                 return copy_regset_to_user(child, &user_x86_32_view,
1370                                            REGSET_XFP, 0,
1371                                            sizeof(struct user32_fxsr_struct),
1372                                            datap);
1373
1374         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1375                 return copy_regset_from_user(child, &user_x86_32_view,
1376                                              REGSET_XFP, 0,
1377                                              sizeof(struct user32_fxsr_struct),
1378                                              datap);
1379
1380         case PTRACE_GET_THREAD_AREA:
1381         case PTRACE_SET_THREAD_AREA:
1382 #ifdef CONFIG_X86_PTRACE_BTS
1383         case PTRACE_BTS_CONFIG:
1384         case PTRACE_BTS_STATUS:
1385         case PTRACE_BTS_SIZE:
1386         case PTRACE_BTS_GET:
1387         case PTRACE_BTS_CLEAR:
1388         case PTRACE_BTS_DRAIN:
1389 #endif /* CONFIG_X86_PTRACE_BTS */
1390                 return arch_ptrace(child, request, addr, data);
1391
1392         default:
1393                 return compat_ptrace_request(child, request, addr, data);
1394         }
1395
1396         return ret;
1397 }
1398
1399 #endif  /* CONFIG_IA32_EMULATION */
1400
1401 #ifdef CONFIG_X86_64
1402
1403 static const struct user_regset x86_64_regsets[] = {
1404         [REGSET_GENERAL] = {
1405                 .core_note_type = NT_PRSTATUS,
1406                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1407                 .size = sizeof(long), .align = sizeof(long),
1408                 .get = genregs_get, .set = genregs_set
1409         },
1410         [REGSET_FP] = {
1411                 .core_note_type = NT_PRFPREG,
1412                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1413                 .size = sizeof(long), .align = sizeof(long),
1414                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1415         },
1416         [REGSET_IOPERM64] = {
1417                 .core_note_type = NT_386_IOPERM,
1418                 .n = IO_BITMAP_LONGS,
1419                 .size = sizeof(long), .align = sizeof(long),
1420                 .active = ioperm_active, .get = ioperm_get
1421         },
1422 };
1423
1424 static const struct user_regset_view user_x86_64_view = {
1425         .name = "x86_64", .e_machine = EM_X86_64,
1426         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1427 };
1428
1429 #else  /* CONFIG_X86_32 */
1430
1431 #define user_regs_struct32      user_regs_struct
1432 #define genregs32_get           genregs_get
1433 #define genregs32_set           genregs_set
1434
1435 #define user_i387_ia32_struct   user_i387_struct
1436 #define user32_fxsr_struct      user_fxsr_struct
1437
1438 #endif  /* CONFIG_X86_64 */
1439
1440 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1441 static const struct user_regset x86_32_regsets[] = {
1442         [REGSET_GENERAL] = {
1443                 .core_note_type = NT_PRSTATUS,
1444                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1445                 .size = sizeof(u32), .align = sizeof(u32),
1446                 .get = genregs32_get, .set = genregs32_set
1447         },
1448         [REGSET_FP] = {
1449                 .core_note_type = NT_PRFPREG,
1450                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1451                 .size = sizeof(u32), .align = sizeof(u32),
1452                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1453         },
1454         [REGSET_XFP] = {
1455                 .core_note_type = NT_PRXFPREG,
1456                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1457                 .size = sizeof(u32), .align = sizeof(u32),
1458                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1459         },
1460         [REGSET_TLS] = {
1461                 .core_note_type = NT_386_TLS,
1462                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1463                 .size = sizeof(struct user_desc),
1464                 .align = sizeof(struct user_desc),
1465                 .active = regset_tls_active,
1466                 .get = regset_tls_get, .set = regset_tls_set
1467         },
1468         [REGSET_IOPERM32] = {
1469                 .core_note_type = NT_386_IOPERM,
1470                 .n = IO_BITMAP_BYTES / sizeof(u32),
1471                 .size = sizeof(u32), .align = sizeof(u32),
1472                 .active = ioperm_active, .get = ioperm_get
1473         },
1474 };
1475
1476 static const struct user_regset_view user_x86_32_view = {
1477         .name = "i386", .e_machine = EM_386,
1478         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1479 };
1480 #endif
1481
1482 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1483 {
1484 #ifdef CONFIG_IA32_EMULATION
1485         if (test_tsk_thread_flag(task, TIF_IA32))
1486 #endif
1487 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1488                 return &user_x86_32_view;
1489 #endif
1490 #ifdef CONFIG_X86_64
1491         return &user_x86_64_view;
1492 #endif
1493 }
1494
1495 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1496                                          int error_code, int si_code)
1497 {
1498         struct siginfo info;
1499
1500         tsk->thread.trap_no = 1;
1501         tsk->thread.error_code = error_code;
1502
1503         memset(&info, 0, sizeof(info));
1504         info.si_signo = SIGTRAP;
1505         info.si_code = si_code;
1506
1507         /* User-mode ip? */
1508         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1509
1510         /* Send us the fake SIGTRAP */
1511         force_sig_info(SIGTRAP, &info, tsk);
1512 }
1513
1514
1515 #ifdef CONFIG_X86_32
1516 # define IS_IA32        1
1517 #elif defined CONFIG_IA32_EMULATION
1518 # define IS_IA32        is_compat_task()
1519 #else
1520 # define IS_IA32        0
1521 #endif
1522
1523 /*
1524  * We must return the syscall number to actually look up in the table.
1525  * This can be -1L to skip running any syscall at all.
1526  */
1527 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1528 {
1529         long ret = 0;
1530
1531         /*
1532          * If we stepped into a sysenter/syscall insn, it trapped in
1533          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1534          * If user-mode had set TF itself, then it's still clear from
1535          * do_debug() and we need to set it again to restore the user
1536          * state.  If we entered on the slow path, TF was already set.
1537          */
1538         if (test_thread_flag(TIF_SINGLESTEP))
1539                 regs->flags |= X86_EFLAGS_TF;
1540
1541         /* do the secure computing check first */
1542         secure_computing(regs->orig_ax);
1543
1544         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1545                 ret = -1L;
1546
1547         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1548             tracehook_report_syscall_entry(regs))
1549                 ret = -1L;
1550
1551         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1552                 ftrace_syscall_enter(regs);
1553
1554         if (unlikely(current->audit_context)) {
1555                 if (IS_IA32)
1556                         audit_syscall_entry(AUDIT_ARCH_I386,
1557                                             regs->orig_ax,
1558                                             regs->bx, regs->cx,
1559                                             regs->dx, regs->si);
1560 #ifdef CONFIG_X86_64
1561                 else
1562                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1563                                             regs->orig_ax,
1564                                             regs->di, regs->si,
1565                                             regs->dx, regs->r10);
1566 #endif
1567         }
1568
1569         return ret ?: regs->orig_ax;
1570 }
1571
1572 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1573 {
1574         if (unlikely(current->audit_context))
1575                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1576
1577         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1578                 ftrace_syscall_exit(regs);
1579
1580         if (test_thread_flag(TIF_SYSCALL_TRACE))
1581                 tracehook_report_syscall_exit(regs, 0);
1582
1583         /*
1584          * If TIF_SYSCALL_EMU is set, we only get here because of
1585          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1586          * We already reported this syscall instruction in
1587          * syscall_trace_enter(), so don't do any more now.
1588          */
1589         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1590                 return;
1591
1592         /*
1593          * If we are single-stepping, synthesize a trap to follow the
1594          * system call instruction.
1595          */
1596         if (test_thread_flag(TIF_SINGLESTEP) &&
1597             tracehook_consider_fatal_signal(current, SIGTRAP))
1598                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1599 }