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         /*
 316          * Orig_ax is really just a flag with small positive and
 317          * negative values, so make sure to always sign-extend it
 318          * from 32 bits so that it works correctly regardless of
 319          * whether we come from a 32-bit environment or not.
 320          */
 321         case offsetof(struct user_regs_struct, orig_ax):
 322                 value = (long) (s32) value;
 323                 break;
 324
 325         case offsetof(struct user_regs_struct,fs_base):
 326                 if (value >= TASK_SIZE_OF(child))
 327                         return -EIO;
 328                 /*
 329                  * When changing the segment base, use do_arch_prctl
 330                  * to set either thread.fs or thread.fsindex and the
 331                  * corresponding GDT slot.
 332                  */
 333                 if (child->thread.fs != value)
 334                         return do_arch_prctl(child, ARCH_SET_FS, value);
 335                 return 0;
 336         case offsetof(struct user_regs_struct,gs_base):
 337                 /*
 338                  * Exactly the same here as the %fs handling above.
 339                  */
 340                 if (value >= TASK_SIZE_OF(child))
 341                         return -EIO;
 342                 if (child->thread.gs != value)
 343                         return do_arch_prctl(child, ARCH_SET_GS, value);
 344                 return 0;
 345 #endif
 346         }
 347
 348         *pt_regs_access(task_pt_regs(child), offset) = value;
 349         return 0;
 350 }
 351
 352 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 353 {
 354         switch (offset) {
 355         case offsetof(struct user_regs_struct, cs):
 356         case offsetof(struct user_regs_struct, ds):
 357         case offsetof(struct user_regs_struct, es):
 358         case offsetof(struct user_regs_struct, fs):
 359         case offsetof(struct user_regs_struct, gs):
 360         case offsetof(struct user_regs_struct, ss):
 361                 return get_segment_reg(task, offset);
 362
 363         case offsetof(struct user_regs_struct, flags):
 364                 return get_flags(task);
 365
 366 #ifdef CONFIG_X86_64
 367         case offsetof(struct user_regs_struct, fs_base): {
 368                 /*
 369                  * do_arch_prctl may have used a GDT slot instead of
 370                  * the MSR.  To userland, it appears the same either
 371                  * way, except the %fs segment selector might not be 0.
 372                  */
 373                 unsigned int seg = task->thread.fsindex;
 374                 if (task->thread.fs != 0)
 375                         return task->thread.fs;
 376                 if (task == current)
 377                         asm("movl %%fs,%0" : "=r" (seg));
 378                 if (seg != FS_TLS_SEL)
 379                         return 0;
 380                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 381         }
 382         case offsetof(struct user_regs_struct, gs_base): {
 383                 /*
 384                  * Exactly the same here as the %fs handling above.
 385                  */
 386                 unsigned int seg = task->thread.gsindex;
 387                 if (task->thread.gs != 0)
 388                         return task->thread.gs;
 389                 if (task == current)
 390                         asm("movl %%gs,%0" : "=r" (seg));
 391                 if (seg != GS_TLS_SEL)
 392                         return 0;
 393                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 394         }
 395 #endif
 396         }
 397
 398         return *pt_regs_access(task_pt_regs(task), offset);
 399 }
 400
 401 static int genregs_get(struct task_struct *target,
 402                        const struct user_regset *regset,
 403                        unsigned int pos, unsigned int count,
 404                        void *kbuf, void __user *ubuf)
 405 {
 406         if (kbuf) {
 407                 unsigned long *k = kbuf;
 408                 while (count > 0) {
 409                         *k++ = getreg(target, pos);
 410                         count -= sizeof(*k);
 411                         pos += sizeof(*k);
 412                 }
 413         } else {
 414                 unsigned long __user *u = ubuf;
 415                 while (count > 0) {
 416                         if (__put_user(getreg(target, pos), u++))
 417                                 return -EFAULT;
 418                         count -= sizeof(*u);
 419                         pos += sizeof(*u);
 420                 }
 421         }
 422
 423         return 0;
 424 }
 425
 426 static int genregs_set(struct task_struct *target,
 427                        const struct user_regset *regset,
 428                        unsigned int pos, unsigned int count,
 429                        const void *kbuf, const void __user *ubuf)
 430 {
 431         int ret = 0;
 432         if (kbuf) {
 433                 const unsigned long *k = kbuf;
 434                 while (count > 0 && !ret) {
 435                         ret = putreg(target, pos, *k++);
 436                         count -= sizeof(*k);
 437                         pos += sizeof(*k);
 438                 }
 439         } else {
 440                 const unsigned long  __user *u = ubuf;
 441                 while (count > 0 && !ret) {
 442                         unsigned long word;
 443                         ret = __get_user(word, u++);
 444                         if (ret)
 445                                 break;
 446                         ret = putreg(target, pos, word);
 447                         count -= sizeof(*u);
 448                         pos += sizeof(*u);
 449                 }
 450         }
 451         return ret;
 452 }
 453
 454 /*
 455  * Decode the length and type bits for a particular breakpoint as
 456  * stored in debug register 7.  Return the "enabled" status.
 457  */
 458 static int decode_dr7(unsigned long dr7, int bpnum, unsigned *len,
 459                 unsigned *type)
 460 {
 461         int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE);
 462
 463         *len = (bp_info & 0xc) | 0x40;
 464         *type = (bp_info & 0x3) | 0x80;
 465         return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3;
 466 }
 467
 468 static void ptrace_triggered(struct hw_breakpoint *bp, struct pt_regs *regs)
 469 {
 470         struct thread_struct *thread = &(current->thread);
 471         int i;
 472
 473         /*
 474          * Store in the virtual DR6 register the fact that the breakpoint
 475          * was hit so the thread's debugger will see it.
 476          */
 477         for (i = 0; i < hbp_kernel_pos; i++)
 478                 /*
 479                  * We will check bp->info.address against the address stored in
 480                  * thread's hbp structure and not debugreg[i]. This is to ensure
 481                  * that the corresponding bit for 'i' in DR7 register is enabled
 482                  */
 483                 if (bp->info.address == thread->hbp[i]->info.address)
 484                         break;
 485
 486         thread->debugreg6 |= (DR_TRAP0 << i);
 487 }
 488
 489 /*
 490  * Handle ptrace writes to debug register 7.
 491  */
 492 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 493 {
 494         struct thread_struct *thread = &(tsk->thread);
 495         unsigned long old_dr7 = thread->debugreg7;
 496         int i, orig_ret = 0, rc = 0;
 497         int enabled, second_pass = 0;
 498         unsigned len, type;
 499         struct hw_breakpoint *bp;
 500
 501         data &= ~DR_CONTROL_RESERVED;
 502 restore:
 503         /*
 504          * Loop through all the hardware breakpoints, making the
 505          * appropriate changes to each.
 506          */
 507         for (i = 0; i < HBP_NUM; i++) {
 508                 enabled = decode_dr7(data, i, &len, &type);
 509                 bp = thread->hbp[i];
 510
 511                 if (!enabled) {
 512                         if (bp) {
 513                                 /* Don't unregister the breakpoints right-away,
 514                                  * unless all register_user_hw_breakpoint()
 515                                  * requests have succeeded. This prevents
 516                                  * any window of opportunity for debug
 517                                  * register grabbing by other users.
 518                                  */
 519                                 if (!second_pass)
 520                                         continue;
 521                                 unregister_user_hw_breakpoint(tsk, bp);
 522                                 kfree(bp);
 523                         }
 524                         continue;
 525                 }
 526                 if (!bp) {
 527                         rc = -ENOMEM;
 528                         bp = kzalloc(sizeof(struct hw_breakpoint), GFP_KERNEL);
 529                         if (bp) {
 530                                 bp->info.address = thread->debugreg[i];
 531                                 bp->triggered = ptrace_triggered;
 532                                 bp->info.len = len;
 533                                 bp->info.type = type;
 534                                 rc = register_user_hw_breakpoint(tsk, bp);
 535                                 if (rc)
 536                                         kfree(bp);
 537                         }
 538                 } else
 539                         rc = modify_user_hw_breakpoint(tsk, bp);
 540                 if (rc)
 541                         break;
 542         }
 543         /*
 544          * Make a second pass to free the remaining unused breakpoints
 545          * or to restore the original breakpoints if an error occurred.
 546          */
 547         if (!second_pass) {
 548                 second_pass = 1;
 549                 if (rc < 0) {
 550                         orig_ret = rc;
 551                         data = old_dr7;
 552                 }
 553                 goto restore;
 554         }
 555         return ((orig_ret < 0) ? orig_ret : rc);
 556 }
 557
 558 /*
 559  * Handle PTRACE_PEEKUSR calls for the debug register area.
 560  */
 561 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 562 {
 563         struct thread_struct *thread = &(tsk->thread);
 564         unsigned long val = 0;
 565
 566         if (n < HBP_NUM)
 567                 val = thread->debugreg[n];
 568         else if (n == 6)
 569                 val = thread->debugreg6;
 570         else if (n == 7)
 571                 val = thread->debugreg7;
 572         return val;
 573 }
 574
 575 /*
 576  * Handle PTRACE_POKEUSR calls for the debug register area.
 577  */
 578 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
 579 {
 580         struct thread_struct *thread = &(tsk->thread);
 581         int rc = 0;
 582
 583         /* There are no DR4 or DR5 registers */
 584         if (n == 4 || n == 5)
 585                 return -EIO;
 586
 587         if (n == 6) {
 588                 tsk->thread.debugreg6 = val;
 589                 goto ret_path;
 590         }
 591         if (n < HBP_NUM) {
 592                 if (thread->hbp[n]) {
 593                         if (arch_check_va_in_userspace(val,
 594                                         thread->hbp[n]->info.len) == 0) {
 595                                 rc = -EIO;
 596                                 goto ret_path;
 597                         }
 598                         thread->hbp[n]->info.address = val;
 599                 }
 600                 thread->debugreg[n] = val;
 601         }
 602         /* All that's left is DR7 */
 603         if (n == 7)
 604                 rc = ptrace_write_dr7(tsk, val);
 605
 606 ret_path:
 607         return rc;
 608 }
 609
 610 /*
 611  * These access the current or another (stopped) task's io permission
 612  * bitmap for debugging or core dump.
 613  */
 614 static int ioperm_active(struct task_struct *target,
 615                          const struct user_regset *regset)
 616 {
 617         return target->thread.io_bitmap_max / regset->size;
 618 }
 619
 620 static int ioperm_get(struct task_struct *target,
 621                       const struct user_regset *regset,
 622                       unsigned int pos, unsigned int count,
 623                       void *kbuf, void __user *ubuf)
 624 {
 625         if (!target->thread.io_bitmap_ptr)
 626                 return -ENXIO;
 627
 628         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 629                                    target->thread.io_bitmap_ptr,
 630                                    0, IO_BITMAP_BYTES);
 631 }
 632
 633 #ifdef CONFIG_X86_PTRACE_BTS
 634 /*
 635  * A branch trace store context.
 636  *
 637  * Contexts may only be installed by ptrace_bts_config() and only for
 638  * ptraced tasks.
 639  *
 640  * Contexts are destroyed when the tracee is detached from the tracer.
 641  * The actual destruction work requires interrupts enabled, so the
 642  * work is deferred and will be scheduled during __ptrace_unlink().
 643  *
 644  * Contexts hold an additional task_struct reference on the traced
 645  * task, as well as a reference on the tracer's mm.
 646  *
 647  * Ptrace already holds a task_struct for the duration of ptrace operations,
 648  * but since destruction is deferred, it may be executed after both
 649  * tracer and tracee exited.
 650  */
 651 struct bts_context {
 652         /* The branch trace handle. */
 653         struct bts_tracer       *tracer;
 654
 655         /* The buffer used to store the branch trace and its size. */
 656         void                    *buffer;
 657         unsigned int            size;
 658
 659         /* The mm that paid for the above buffer. */
 660         struct mm_struct        *mm;
 661
 662         /* The task this context belongs to. */
 663         struct task_struct      *task;
 664
 665         /* The signal to send on a bts buffer overflow. */
 666         unsigned int            bts_ovfl_signal;
 667
 668         /* The work struct to destroy a context. */
 669         struct work_struct      work;
 670 };
 671
 672 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
 673 {
 674         void *buffer = NULL;
 675         int err = -ENOMEM;
 676
 677         err = account_locked_memory(current->mm, current->signal->rlim, size);
 678         if (err < 0)
 679                 return err;
 680
 681         buffer = kzalloc(size, GFP_KERNEL);
 682         if (!buffer)
 683                 goto out_refund;
 684
 685         context->buffer = buffer;
 686         context->size = size;
 687         context->mm = get_task_mm(current);
 688
 689         return 0;
 690
 691  out_refund:
 692         refund_locked_memory(current->mm, size);
 693         return err;
 694 }
 695
 696 static inline void free_bts_buffer(struct bts_context *context)
 697 {
 698         if (!context->buffer)
 699                 return;
 700
 701         kfree(context->buffer);
 702         context->buffer = NULL;
 703
 704         refund_locked_memory(context->mm, context->size);
 705         context->size = 0;
 706
 707         mmput(context->mm);
 708         context->mm = NULL;
 709 }
 710
 711 static void free_bts_context_work(struct work_struct *w)
 712 {
 713         struct bts_context *context;
 714
 715         context = container_of(w, struct bts_context, work);
 716
 717         ds_release_bts(context->tracer);
 718         put_task_struct(context->task);
 719         free_bts_buffer(context);
 720         kfree(context);
 721 }
 722
 723 static inline void free_bts_context(struct bts_context *context)
 724 {
 725         INIT_WORK(&context->work, free_bts_context_work);
 726         schedule_work(&context->work);
 727 }
 728
 729 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
 730 {
 731         struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
 732         if (context) {
 733                 context->task = task;
 734                 task->bts = context;
 735
 736                 get_task_struct(task);
 737         }
 738
 739         return context;
 740 }
 741
 742 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 743                                   struct bts_struct __user *out)
 744 {
 745         struct bts_context *context;
 746         const struct bts_trace *trace;
 747         struct bts_struct bts;
 748         const unsigned char *at;
 749         int error;
 750
 751         context = child->bts;
 752         if (!context)
 753                 return -ESRCH;
 754
 755         trace = ds_read_bts(context->tracer);
 756         if (!trace)
 757                 return -ESRCH;
 758
 759         at = trace->ds.top - ((index + 1) * trace->ds.size);
 760         if ((void *)at < trace->ds.begin)
 761                 at += (trace->ds.n * trace->ds.size);
 762
 763         if (!trace->read)
 764                 return -EOPNOTSUPP;
 765
 766         error = trace->read(context->tracer, at, &bts);
 767         if (error < 0)
 768                 return error;
 769
 770         if (copy_to_user(out, &bts, sizeof(bts)))
 771                 return -EFAULT;
 772
 773         return sizeof(bts);
 774 }
 775
 776 static int ptrace_bts_drain(struct task_struct *child,
 777                             long size,
 778                             struct bts_struct __user *out)
 779 {
 780         struct bts_context *context;
 781         const struct bts_trace *trace;
 782         const unsigned char *at;
 783         int error, drained = 0;
 784
 785         context = child->bts;
 786         if (!context)
 787                 return -ESRCH;
 788
 789         trace = ds_read_bts(context->tracer);
 790         if (!trace)
 791                 return -ESRCH;
 792
 793         if (!trace->read)
 794                 return -EOPNOTSUPP;
 795
 796         if (size < (trace->ds.top - trace->ds.begin))
 797                 return -EIO;
 798
 799         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 800              out++, drained++, at += trace->ds.size) {
 801                 struct bts_struct bts;
 802
 803                 error = trace->read(context->tracer, at, &bts);
 804                 if (error < 0)
 805                         return error;
 806
 807                 if (copy_to_user(out, &bts, sizeof(bts)))
 808                         return -EFAULT;
 809         }
 810
 811         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 812
 813         error = ds_reset_bts(context->tracer);
 814         if (error < 0)
 815                 return error;
 816
 817         return drained;
 818 }
 819
 820 static int ptrace_bts_config(struct task_struct *child,
 821                              long cfg_size,
 822                              const struct ptrace_bts_config __user *ucfg)
 823 {
 824         struct bts_context *context;
 825         struct ptrace_bts_config cfg;
 826         unsigned int flags = 0;
 827
 828         if (cfg_size < sizeof(cfg))
 829                 return -EIO;
 830
 831         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 832                 return -EFAULT;
 833
 834         context = child->bts;
 835         if (!context)
 836                 context = alloc_bts_context(child);
 837         if (!context)
 838                 return -ENOMEM;
 839
 840         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 841                 if (!cfg.signal)
 842                         return -EINVAL;
 843
 844                 return -EOPNOTSUPP;
 845                 context->bts_ovfl_signal = cfg.signal;
 846         }
 847
 848         ds_release_bts(context->tracer);
 849         context->tracer = NULL;
 850
 851         if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
 852                 int err;
 853
 854                 free_bts_buffer(context);
 855                 if (!cfg.size)
 856                         return 0;
 857
 858                 err = alloc_bts_buffer(context, cfg.size);
 859                 if (err < 0)
 860                         return err;
 861         }
 862
 863         if (cfg.flags & PTRACE_BTS_O_TRACE)
 864                 flags |= BTS_USER;
 865
 866         if (cfg.flags & PTRACE_BTS_O_SCHED)
 867                 flags |= BTS_TIMESTAMPS;
 868
 869         context->tracer =
 870                 ds_request_bts_task(child, context->buffer, context->size,
 871                                     NULL, (size_t)-1, flags);
 872         if (unlikely(IS_ERR(context->tracer))) {
 873                 int error = PTR_ERR(context->tracer);
 874
 875                 free_bts_buffer(context);
 876                 context->tracer = NULL;
 877                 return error;
 878         }
 879
 880         return sizeof(cfg);
 881 }
 882
 883 static int ptrace_bts_status(struct task_struct *child,
 884                              long cfg_size,
 885                              struct ptrace_bts_config __user *ucfg)
 886 {
 887         struct bts_context *context;
 888         const struct bts_trace *trace;
 889         struct ptrace_bts_config cfg;
 890
 891         context = child->bts;
 892         if (!context)
 893                 return -ESRCH;
 894
 895         if (cfg_size < sizeof(cfg))
 896                 return -EIO;
 897
 898         trace = ds_read_bts(context->tracer);
 899         if (!trace)
 900                 return -ESRCH;
 901
 902         memset(&cfg, 0, sizeof(cfg));
 903         cfg.size        = trace->ds.end - trace->ds.begin;
 904         cfg.signal      = context->bts_ovfl_signal;
 905         cfg.bts_size    = sizeof(struct bts_struct);
 906
 907         if (cfg.signal)
 908                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 909
 910         if (trace->ds.flags & BTS_USER)
 911                 cfg.flags |= PTRACE_BTS_O_TRACE;
 912
 913         if (trace->ds.flags & BTS_TIMESTAMPS)
 914                 cfg.flags |= PTRACE_BTS_O_SCHED;
 915
 916         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 917                 return -EFAULT;
 918
 919         return sizeof(cfg);
 920 }
 921
 922 static int ptrace_bts_clear(struct task_struct *child)
 923 {
 924         struct bts_context *context;
 925         const struct bts_trace *trace;
 926
 927         context = child->bts;
 928         if (!context)
 929                 return -ESRCH;
 930
 931         trace = ds_read_bts(context->tracer);
 932         if (!trace)
 933                 return -ESRCH;
 934
 935         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 936
 937         return ds_reset_bts(context->tracer);
 938 }
 939
 940 static int ptrace_bts_size(struct task_struct *child)
 941 {
 942         struct bts_context *context;
 943         const struct bts_trace *trace;
 944
 945         context = child->bts;
 946         if (!context)
 947                 return -ESRCH;
 948
 949         trace = ds_read_bts(context->tracer);
 950         if (!trace)
 951                 return -ESRCH;
 952
 953         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 954 }
 955
 956 /*
 957  * Called from __ptrace_unlink() after the child has been moved back
 958  * to its original parent.
 959  */
 960 void ptrace_bts_untrace(struct task_struct *child)
 961 {
 962         if (unlikely(child->bts)) {
 963                 free_bts_context(child->bts);
 964                 child->bts = NULL;
 965         }
 966 }
 967 #endif /* CONFIG_X86_PTRACE_BTS */
 968
 969 /*
 970  * Called by kernel/ptrace.c when detaching..
 971  *
 972  * Make sure the single step bit is not set.
 973  */
 974 void ptrace_disable(struct task_struct *child)
 975 {
 976         user_disable_single_step(child);
 977 #ifdef TIF_SYSCALL_EMU
 978         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 979 #endif
 980 }
 981
 982 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 983 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 984 #endif
 985
 986 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 987 {
 988         int ret;
 989         unsigned long __user *datap = (unsigned long __user *)data;
 990
 991         switch (request) {
 992         /* read the word at location addr in the USER area. */
 993         case PTRACE_PEEKUSR: {
 994                 unsigned long tmp;
 995
 996                 ret = -EIO;
 997                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 998                     addr >= sizeof(struct user))
 999                         break;
1000
1001                 tmp = 0;  /* Default return condition */
1002                 if (addr < sizeof(struct user_regs_struct))
1003                         tmp = getreg(child, addr);
1004                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1005                          addr <= offsetof(struct user, u_debugreg[7])) {
1006                         addr -= offsetof(struct user, u_debugreg[0]);
1007                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1008                 }
1009                 ret = put_user(tmp, datap);
1010                 break;
1011         }
1012
1013         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1014                 ret = -EIO;
1015                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1016                     addr >= sizeof(struct user))
1017                         break;
1018
1019                 if (addr < sizeof(struct user_regs_struct))
1020                         ret = putreg(child, addr, data);
1021                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1022                          addr <= offsetof(struct user, u_debugreg[7])) {
1023                         addr -= offsetof(struct user, u_debugreg[0]);
1024                         ret = ptrace_set_debugreg(child,
1025                                                   addr / sizeof(data), data);
1026                 }
1027                 break;
1028
1029         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1030                 return copy_regset_to_user(child,
1031                                            task_user_regset_view(current),
1032                                            REGSET_GENERAL,
1033                                            0, sizeof(struct user_regs_struct),
1034                                            datap);
1035
1036         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1037                 return copy_regset_from_user(child,
1038                                              task_user_regset_view(current),
1039                                              REGSET_GENERAL,
1040                                              0, sizeof(struct user_regs_struct),
1041                                              datap);
1042
1043         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1044                 return copy_regset_to_user(child,
1045                                            task_user_regset_view(current),
1046                                            REGSET_FP,
1047                                            0, sizeof(struct user_i387_struct),
1048                                            datap);
1049
1050         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1051                 return copy_regset_from_user(child,
1052                                              task_user_regset_view(current),
1053                                              REGSET_FP,
1054                                              0, sizeof(struct user_i387_struct),
1055                                              datap);
1056
1057 #ifdef CONFIG_X86_32
1058         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1059                 return copy_regset_to_user(child, &user_x86_32_view,
1060                                            REGSET_XFP,
1061                                            0, sizeof(struct user_fxsr_struct),
1062                                            datap) ? -EIO : 0;
1063
1064         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1065                 return copy_regset_from_user(child, &user_x86_32_view,
1066                                              REGSET_XFP,
1067                                              0, sizeof(struct user_fxsr_struct),
1068                                              datap) ? -EIO : 0;
1069 #endif
1070
1071 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1072         case PTRACE_GET_THREAD_AREA:
1073                 if (addr < 0)
1074                         return -EIO;
1075                 ret = do_get_thread_area(child, addr,
1076                                          (struct user_desc __user *) data);
1077                 break;
1078
1079         case PTRACE_SET_THREAD_AREA:
1080                 if (addr < 0)
1081                         return -EIO;
1082                 ret = do_set_thread_area(child, addr,
1083                                          (struct user_desc __user *) data, 0);
1084                 break;
1085 #endif
1086
1087 #ifdef CONFIG_X86_64
1088                 /* normal 64bit interface to access TLS data.
1089                    Works just like arch_prctl, except that the arguments
1090                    are reversed. */
1091         case PTRACE_ARCH_PRCTL:
1092                 ret = do_arch_prctl(child, data, addr);
1093                 break;
1094 #endif
1095
1096         /*
1097          * These bits need more cooking - not enabled yet:
1098          */
1099 #ifdef CONFIG_X86_PTRACE_BTS
1100         case PTRACE_BTS_CONFIG:
1101                 ret = ptrace_bts_config
1102                         (child, data, (struct ptrace_bts_config __user *)addr);
1103                 break;
1104
1105         case PTRACE_BTS_STATUS:
1106                 ret = ptrace_bts_status
1107                         (child, data, (struct ptrace_bts_config __user *)addr);
1108                 break;
1109
1110         case PTRACE_BTS_SIZE:
1111                 ret = ptrace_bts_size(child);
1112                 break;
1113
1114         case PTRACE_BTS_GET:
1115                 ret = ptrace_bts_read_record
1116                         (child, data, (struct bts_struct __user *) addr);
1117                 break;
1118
1119         case PTRACE_BTS_CLEAR:
1120                 ret = ptrace_bts_clear(child);
1121                 break;
1122
1123         case PTRACE_BTS_DRAIN:
1124                 ret = ptrace_bts_drain
1125                         (child, data, (struct bts_struct __user *) addr);
1126                 break;
1127 #endif /* CONFIG_X86_PTRACE_BTS */
1128
1129         default:
1130                 ret = ptrace_request(child, request, addr, data);
1131                 break;
1132         }
1133
1134         return ret;
1135 }
1136
1137 #ifdef CONFIG_IA32_EMULATION
1138
1139 #include <linux/compat.h>
1140 #include <linux/syscalls.h>
1141 #include <asm/ia32.h>
1142 #include <asm/user32.h>
1143
1144 #define R32(l,q)                                                        \
1145         case offsetof(struct user32, regs.l):                           \
1146                 regs->q = value; break
1147
1148 #define SEG32(rs)                                                       \
1149         case offsetof(struct user32, regs.rs):                          \
1150                 return set_segment_reg(child,                           \
1151                                        offsetof(struct user_regs_struct, rs), \
1152                                        value);                          \
1153                 break
1154
1155 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1156 {
1157         struct pt_regs *regs = task_pt_regs(child);
1158
1159         switch (regno) {
1160
1161         SEG32(cs);
1162         SEG32(ds);
1163         SEG32(es);
1164         SEG32(fs);
1165         SEG32(gs);
1166         SEG32(ss);
1167
1168         R32(ebx, bx);
1169         R32(ecx, cx);
1170         R32(edx, dx);
1171         R32(edi, di);
1172         R32(esi, si);
1173         R32(ebp, bp);
1174         R32(eax, ax);
1175         R32(eip, ip);
1176         R32(esp, sp);
1177
1178         case offsetof(struct user32, regs.orig_eax):
1179                 /*
1180                  * Sign-extend the value so that orig_eax = -1
1181                  * causes (long)orig_ax < 0 tests to fire correctly.
1182                  */
1183                 regs->orig_ax = (long) (s32) value;
1184                 break;
1185
1186         case offsetof(struct user32, regs.eflags):
1187                 return set_flags(child, value);
1188
1189         case offsetof(struct user32, u_debugreg[0]) ...
1190                 offsetof(struct user32, u_debugreg[7]):
1191                 regno -= offsetof(struct user32, u_debugreg[0]);
1192                 return ptrace_set_debugreg(child, regno / 4, value);
1193
1194         default:
1195                 if (regno > sizeof(struct user32) || (regno & 3))
1196                         return -EIO;
1197
1198                 /*
1199                  * Other dummy fields in the virtual user structure
1200                  * are ignored
1201                  */
1202                 break;
1203         }
1204         return 0;
1205 }
1206
1207 #undef R32
1208 #undef SEG32
1209
1210 #define R32(l,q)                                                        \
1211         case offsetof(struct user32, regs.l):                           \
1212                 *val = regs->q; break
1213
1214 #define SEG32(rs)                                                       \
1215         case offsetof(struct user32, regs.rs):                          \
1216                 *val = get_segment_reg(child,                           \
1217                                        offsetof(struct user_regs_struct, rs)); \
1218                 break
1219
1220 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1221 {
1222         struct pt_regs *regs = task_pt_regs(child);
1223
1224         switch (regno) {
1225
1226         SEG32(ds);
1227         SEG32(es);
1228         SEG32(fs);
1229         SEG32(gs);
1230
1231         R32(cs, cs);
1232         R32(ss, ss);
1233         R32(ebx, bx);
1234         R32(ecx, cx);
1235         R32(edx, dx);
1236         R32(edi, di);
1237         R32(esi, si);
1238         R32(ebp, bp);
1239         R32(eax, ax);
1240         R32(orig_eax, orig_ax);
1241         R32(eip, ip);
1242         R32(esp, sp);
1243
1244         case offsetof(struct user32, regs.eflags):
1245                 *val = get_flags(child);
1246                 break;
1247
1248         case offsetof(struct user32, u_debugreg[0]) ...
1249                 offsetof(struct user32, u_debugreg[7]):
1250                 regno -= offsetof(struct user32, u_debugreg[0]);
1251                 *val = ptrace_get_debugreg(child, regno / 4);
1252                 break;
1253
1254         default:
1255                 if (regno > sizeof(struct user32) || (regno & 3))
1256                         return -EIO;
1257
1258                 /*
1259                  * Other dummy fields in the virtual user structure
1260                  * are ignored
1261                  */
1262                 *val = 0;
1263                 break;
1264         }
1265         return 0;
1266 }
1267
1268 #undef R32
1269 #undef SEG32
1270
1271 static int genregs32_get(struct task_struct *target,
1272                          const struct user_regset *regset,
1273                          unsigned int pos, unsigned int count,
1274                          void *kbuf, void __user *ubuf)
1275 {
1276         if (kbuf) {
1277                 compat_ulong_t *k = kbuf;
1278                 while (count > 0) {
1279                         getreg32(target, pos, k++);
1280                         count -= sizeof(*k);
1281                         pos += sizeof(*k);
1282                 }
1283         } else {
1284                 compat_ulong_t __user *u = ubuf;
1285                 while (count > 0) {
1286                         compat_ulong_t word;
1287                         getreg32(target, pos, &word);
1288                         if (__put_user(word, u++))
1289                                 return -EFAULT;
1290                         count -= sizeof(*u);
1291                         pos += sizeof(*u);
1292                 }
1293         }
1294
1295         return 0;
1296 }
1297
1298 static int genregs32_set(struct task_struct *target,
1299                          const struct user_regset *regset,
1300                          unsigned int pos, unsigned int count,
1301                          const void *kbuf, const void __user *ubuf)
1302 {
1303         int ret = 0;
1304         if (kbuf) {
1305                 const compat_ulong_t *k = kbuf;
1306                 while (count > 0 && !ret) {
1307                         ret = putreg32(target, pos, *k++);
1308                         count -= sizeof(*k);
1309                         pos += sizeof(*k);
1310                 }
1311         } else {
1312                 const compat_ulong_t __user *u = ubuf;
1313                 while (count > 0 && !ret) {
1314                         compat_ulong_t word;
1315                         ret = __get_user(word, u++);
1316                         if (ret)
1317                                 break;
1318                         ret = putreg32(target, pos, word);
1319                         count -= sizeof(*u);
1320                         pos += sizeof(*u);
1321                 }
1322         }
1323         return ret;
1324 }
1325
1326 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1327                         compat_ulong_t caddr, compat_ulong_t cdata)
1328 {
1329         unsigned long addr = caddr;
1330         unsigned long data = cdata;
1331         void __user *datap = compat_ptr(data);
1332         int ret;
1333         __u32 val;
1334
1335         switch (request) {
1336         case PTRACE_PEEKUSR:
1337                 ret = getreg32(child, addr, &val);
1338                 if (ret == 0)
1339                         ret = put_user(val, (__u32 __user *)datap);
1340                 break;
1341
1342         case PTRACE_POKEUSR:
1343                 ret = putreg32(child, addr, data);
1344                 break;
1345
1346         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1347                 return copy_regset_to_user(child, &user_x86_32_view,
1348                                            REGSET_GENERAL,
1349                                            0, sizeof(struct user_regs_struct32),
1350                                            datap);
1351
1352         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1353                 return copy_regset_from_user(child, &user_x86_32_view,
1354                                              REGSET_GENERAL, 0,
1355                                              sizeof(struct user_regs_struct32),
1356                                              datap);
1357
1358         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1359                 return copy_regset_to_user(child, &user_x86_32_view,
1360                                            REGSET_FP, 0,
1361                                            sizeof(struct user_i387_ia32_struct),
1362                                            datap);
1363
1364         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1365                 return copy_regset_from_user(
1366                         child, &user_x86_32_view, REGSET_FP,
1367                         0, sizeof(struct user_i387_ia32_struct), datap);
1368
1369         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1370                 return copy_regset_to_user(child, &user_x86_32_view,
1371                                            REGSET_XFP, 0,
1372                                            sizeof(struct user32_fxsr_struct),
1373                                            datap);
1374
1375         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1376                 return copy_regset_from_user(child, &user_x86_32_view,
1377                                              REGSET_XFP, 0,
1378                                              sizeof(struct user32_fxsr_struct),
1379                                              datap);
1380
1381         case PTRACE_GET_THREAD_AREA:
1382         case PTRACE_SET_THREAD_AREA:
1383 #ifdef CONFIG_X86_PTRACE_BTS
1384         case PTRACE_BTS_CONFIG:
1385         case PTRACE_BTS_STATUS:
1386         case PTRACE_BTS_SIZE:
1387         case PTRACE_BTS_GET:
1388         case PTRACE_BTS_CLEAR:
1389         case PTRACE_BTS_DRAIN:
1390 #endif /* CONFIG_X86_PTRACE_BTS */
1391                 return arch_ptrace(child, request, addr, data);
1392
1393         default:
1394                 return compat_ptrace_request(child, request, addr, data);
1395         }
1396
1397         return ret;
1398 }
1399
1400 #endif  /* CONFIG_IA32_EMULATION */
1401
1402 #ifdef CONFIG_X86_64
1403
1404 static const struct user_regset x86_64_regsets[] = {
1405         [REGSET_GENERAL] = {
1406                 .core_note_type = NT_PRSTATUS,
1407                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1408                 .size = sizeof(long), .align = sizeof(long),
1409                 .get = genregs_get, .set = genregs_set
1410         },
1411         [REGSET_FP] = {
1412                 .core_note_type = NT_PRFPREG,
1413                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1414                 .size = sizeof(long), .align = sizeof(long),
1415                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1416         },
1417         [REGSET_IOPERM64] = {
1418                 .core_note_type = NT_386_IOPERM,
1419                 .n = IO_BITMAP_LONGS,
1420                 .size = sizeof(long), .align = sizeof(long),
1421                 .active = ioperm_active, .get = ioperm_get
1422         },
1423 };
1424
1425 static const struct user_regset_view user_x86_64_view = {
1426         .name = "x86_64", .e_machine = EM_X86_64,
1427         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1428 };
1429
1430 #else  /* CONFIG_X86_32 */
1431
1432 #define user_regs_struct32      user_regs_struct
1433 #define genregs32_get           genregs_get
1434 #define genregs32_set           genregs_set
1435
1436 #define user_i387_ia32_struct   user_i387_struct
1437 #define user32_fxsr_struct      user_fxsr_struct
1438
1439 #endif  /* CONFIG_X86_64 */
1440
1441 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1442 static const struct user_regset x86_32_regsets[] = {
1443         [REGSET_GENERAL] = {
1444                 .core_note_type = NT_PRSTATUS,
1445                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1446                 .size = sizeof(u32), .align = sizeof(u32),
1447                 .get = genregs32_get, .set = genregs32_set
1448         },
1449         [REGSET_FP] = {
1450                 .core_note_type = NT_PRFPREG,
1451                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1452                 .size = sizeof(u32), .align = sizeof(u32),
1453                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1454         },
1455         [REGSET_XFP] = {
1456                 .core_note_type = NT_PRXFPREG,
1457                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1458                 .size = sizeof(u32), .align = sizeof(u32),
1459                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1460         },
1461         [REGSET_TLS] = {
1462                 .core_note_type = NT_386_TLS,
1463                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1464                 .size = sizeof(struct user_desc),
1465                 .align = sizeof(struct user_desc),
1466                 .active = regset_tls_active,
1467                 .get = regset_tls_get, .set = regset_tls_set
1468         },
1469         [REGSET_IOPERM32] = {
1470                 .core_note_type = NT_386_IOPERM,
1471                 .n = IO_BITMAP_BYTES / sizeof(u32),
1472                 .size = sizeof(u32), .align = sizeof(u32),
1473                 .active = ioperm_active, .get = ioperm_get
1474         },
1475 };
1476
1477 static const struct user_regset_view user_x86_32_view = {
1478         .name = "i386", .e_machine = EM_386,
1479         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1480 };
1481 #endif
1482
1483 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1484 {
1485 #ifdef CONFIG_IA32_EMULATION
1486         if (test_tsk_thread_flag(task, TIF_IA32))
1487 #endif
1488 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1489                 return &user_x86_32_view;
1490 #endif
1491 #ifdef CONFIG_X86_64
1492         return &user_x86_64_view;
1493 #endif
1494 }
1495
1496 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1497                                          int error_code, int si_code)
1498 {
1499         struct siginfo info;
1500
1501         tsk->thread.trap_no = 1;
1502         tsk->thread.error_code = error_code;
1503
1504         memset(&info, 0, sizeof(info));
1505         info.si_signo = SIGTRAP;
1506         info.si_code = si_code;
1507
1508         /* User-mode ip? */
1509         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1510
1511         /* Send us the fake SIGTRAP */
1512         force_sig_info(SIGTRAP, &info, tsk);
1513 }
1514
1515
1516 #ifdef CONFIG_X86_32
1517 # define IS_IA32        1
1518 #elif defined CONFIG_IA32_EMULATION
1519 # define IS_IA32        is_compat_task()
1520 #else
1521 # define IS_IA32        0
1522 #endif
1523
1524 /*
1525  * We must return the syscall number to actually look up in the table.
1526  * This can be -1L to skip running any syscall at all.
1527  */
1528 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1529 {
1530         long ret = 0;
1531
1532         /*
1533          * If we stepped into a sysenter/syscall insn, it trapped in
1534          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1535          * If user-mode had set TF itself, then it's still clear from
1536          * do_debug() and we need to set it again to restore the user
1537          * state.  If we entered on the slow path, TF was already set.
1538          */
1539         if (test_thread_flag(TIF_SINGLESTEP))
1540                 regs->flags |= X86_EFLAGS_TF;
1541
1542         /* do the secure computing check first */
1543         secure_computing(regs->orig_ax);
1544
1545         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1546                 ret = -1L;
1547
1548         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1549             tracehook_report_syscall_entry(regs))
1550                 ret = -1L;
1551
1552         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1553                 trace_sys_enter(regs, regs->orig_ax);
1554
1555         if (unlikely(current->audit_context)) {
1556                 if (IS_IA32)
1557                         audit_syscall_entry(AUDIT_ARCH_I386,
1558                                             regs->orig_ax,
1559                                             regs->bx, regs->cx,
1560                                             regs->dx, regs->si);
1561 #ifdef CONFIG_X86_64
1562                 else
1563                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1564                                             regs->orig_ax,
1565                                             regs->di, regs->si,
1566                                             regs->dx, regs->r10);
1567 #endif
1568         }
1569
1570         return ret ?: regs->orig_ax;
1571 }
1572
1573 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1574 {
1575         if (unlikely(current->audit_context))
1576                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1577
1578         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1579                 trace_sys_exit(regs, regs->ax);
1580
1581         if (test_thread_flag(TIF_SYSCALL_TRACE))
1582                 tracehook_report_syscall_exit(regs, 0);
1583
1584         /*
1585          * If TIF_SYSCALL_EMU is set, we only get here because of
1586          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1587          * We already reported this syscall instruction in
1588          * syscall_trace_enter(), so don't do any more now.
1589          */
1590         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1591                 return;
1592
1593         /*
1594          * If we are single-stepping, synthesize a trap to follow the
1595          * system call instruction.
1596          */
1597         if (test_thread_flag(TIF_SINGLESTEP) &&
1598             tracehook_consider_fatal_signal(current, SIGTRAP))
1599                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1600 }