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 #include <linux/perf_event.h>
  26 #include <linux/hw_breakpoint.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/pgtable.h>
  30 #include <asm/system.h>
  31 #include <asm/processor.h>
  32 #include <asm/i387.h>
  33 #include <asm/debugreg.h>
  34 #include <asm/ldt.h>
  35 #include <asm/desc.h>
  36 #include <asm/prctl.h>
  37 #include <asm/proto.h>
  38 #include <asm/ds.h>
  39 #include <asm/hw_breakpoint.h>
  40
  41 #include "tls.h"
  42
  43 #define CREATE_TRACE_POINTS
  44 #include <trace/events/syscalls.h>
  45
  46 enum x86_regset {
  47         REGSET_GENERAL,
  48         REGSET_FP,
  49         REGSET_XFP,
  50         REGSET_IOPERM64 = REGSET_XFP,
  51         REGSET_TLS,
  52         REGSET_IOPERM32,
  53 };
  54
  55 struct pt_regs_offset {
  56         const char *name;
  57         int offset;
  58 };
  59
  60 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  61 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  62
  63 static const struct pt_regs_offset regoffset_table[] = {
  64 #ifdef CONFIG_X86_64
  65         REG_OFFSET_NAME(r15),
  66         REG_OFFSET_NAME(r14),
  67         REG_OFFSET_NAME(r13),
  68         REG_OFFSET_NAME(r12),
  69         REG_OFFSET_NAME(r11),
  70         REG_OFFSET_NAME(r10),
  71         REG_OFFSET_NAME(r9),
  72         REG_OFFSET_NAME(r8),
  73 #endif
  74         REG_OFFSET_NAME(bx),
  75         REG_OFFSET_NAME(cx),
  76         REG_OFFSET_NAME(dx),
  77         REG_OFFSET_NAME(si),
  78         REG_OFFSET_NAME(di),
  79         REG_OFFSET_NAME(bp),
  80         REG_OFFSET_NAME(ax),
  81 #ifdef CONFIG_X86_32
  82         REG_OFFSET_NAME(ds),
  83         REG_OFFSET_NAME(es),
  84         REG_OFFSET_NAME(fs),
  85         REG_OFFSET_NAME(gs),
  86 #endif
  87         REG_OFFSET_NAME(orig_ax),
  88         REG_OFFSET_NAME(ip),
  89         REG_OFFSET_NAME(cs),
  90         REG_OFFSET_NAME(flags),
  91         REG_OFFSET_NAME(sp),
  92         REG_OFFSET_NAME(ss),
  93         REG_OFFSET_END,
  94 };
  95
  96 /**
  97  * regs_query_register_offset() - query register offset from its name
  98  * @name:       the name of a register
  99  *
 100  * regs_query_register_offset() returns the offset of a register in struct
 101  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
 102  */
 103 int regs_query_register_offset(const char *name)
 104 {
 105         const struct pt_regs_offset *roff;
 106         for (roff = regoffset_table; roff->name != NULL; roff++)
 107                 if (!strcmp(roff->name, name))
 108                         return roff->offset;
 109         return -EINVAL;
 110 }
 111
 112 /**
 113  * regs_query_register_name() - query register name from its offset
 114  * @offset:     the offset of a register in struct pt_regs.
 115  *
 116  * regs_query_register_name() returns the name of a register from its
 117  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 118  */
 119 const char *regs_query_register_name(unsigned int offset)
 120 {
 121         const struct pt_regs_offset *roff;
 122         for (roff = regoffset_table; roff->name != NULL; roff++)
 123                 if (roff->offset == offset)
 124                         return roff->name;
 125         return NULL;
 126 }
 127
 128 static const int arg_offs_table[] = {
 129 #ifdef CONFIG_X86_32
 130         [0] = offsetof(struct pt_regs, ax),
 131         [1] = offsetof(struct pt_regs, dx),
 132         [2] = offsetof(struct pt_regs, cx)
 133 #else /* CONFIG_X86_64 */
 134         [0] = offsetof(struct pt_regs, di),
 135         [1] = offsetof(struct pt_regs, si),
 136         [2] = offsetof(struct pt_regs, dx),
 137         [3] = offsetof(struct pt_regs, cx),
 138         [4] = offsetof(struct pt_regs, r8),
 139         [5] = offsetof(struct pt_regs, r9)
 140 #endif
 141 };
 142
 143 /*
 144  * does not yet catch signals sent when the child dies.
 145  * in exit.c or in signal.c.
 146  */
 147
 148 /*
 149  * Determines which flags the user has access to [1 = access, 0 = no access].
 150  */
 151 #define FLAG_MASK_32            ((unsigned long)                        \
 152                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 153                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 154                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 155                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 156                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 157
 158 /*
 159  * Determines whether a value may be installed in a segment register.
 160  */
 161 static inline bool invalid_selector(u16 value)
 162 {
 163         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 164 }
 165
 166 #ifdef CONFIG_X86_32
 167
 168 #define FLAG_MASK               FLAG_MASK_32
 169
 170 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 171 {
 172         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 173         return &regs->bx + (regno >> 2);
 174 }
 175
 176 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 177 {
 178         /*
 179          * Returning the value truncates it to 16 bits.
 180          */
 181         unsigned int retval;
 182         if (offset != offsetof(struct user_regs_struct, gs))
 183                 retval = *pt_regs_access(task_pt_regs(task), offset);
 184         else {
 185                 if (task == current)
 186                         retval = get_user_gs(task_pt_regs(task));
 187                 else
 188                         retval = task_user_gs(task);
 189         }
 190         return retval;
 191 }
 192
 193 static int set_segment_reg(struct task_struct *task,
 194                            unsigned long offset, u16 value)
 195 {
 196         /*
 197          * The value argument was already truncated to 16 bits.
 198          */
 199         if (invalid_selector(value))
 200                 return -EIO;
 201
 202         /*
 203          * For %cs and %ss we cannot permit a null selector.
 204          * We can permit a bogus selector as long as it has USER_RPL.
 205          * Null selectors are fine for other segment registers, but
 206          * we will never get back to user mode with invalid %cs or %ss
 207          * and will take the trap in iret instead.  Much code relies
 208          * on user_mode() to distinguish a user trap frame (which can
 209          * safely use invalid selectors) from a kernel trap frame.
 210          */
 211         switch (offset) {
 212         case offsetof(struct user_regs_struct, cs):
 213         case offsetof(struct user_regs_struct, ss):
 214                 if (unlikely(value == 0))
 215                         return -EIO;
 216
 217         default:
 218                 *pt_regs_access(task_pt_regs(task), offset) = value;
 219                 break;
 220
 221         case offsetof(struct user_regs_struct, gs):
 222                 if (task == current)
 223                         set_user_gs(task_pt_regs(task), value);
 224                 else
 225                         task_user_gs(task) = value;
 226         }
 227
 228         return 0;
 229 }
 230
 231 #else  /* CONFIG_X86_64 */
 232
 233 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 234
 235 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 236 {
 237         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 238         return &regs->r15 + (offset / sizeof(regs->r15));
 239 }
 240
 241 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 242 {
 243         /*
 244          * Returning the value truncates it to 16 bits.
 245          */
 246         unsigned int seg;
 247
 248         switch (offset) {
 249         case offsetof(struct user_regs_struct, fs):
 250                 if (task == current) {
 251                         /* Older gas can't assemble movq %?s,%r?? */
 252                         asm("movl %%fs,%0" : "=r" (seg));
 253                         return seg;
 254                 }
 255                 return task->thread.fsindex;
 256         case offsetof(struct user_regs_struct, gs):
 257                 if (task == current) {
 258                         asm("movl %%gs,%0" : "=r" (seg));
 259                         return seg;
 260                 }
 261                 return task->thread.gsindex;
 262         case offsetof(struct user_regs_struct, ds):
 263                 if (task == current) {
 264                         asm("movl %%ds,%0" : "=r" (seg));
 265                         return seg;
 266                 }
 267                 return task->thread.ds;
 268         case offsetof(struct user_regs_struct, es):
 269                 if (task == current) {
 270                         asm("movl %%es,%0" : "=r" (seg));
 271                         return seg;
 272                 }
 273                 return task->thread.es;
 274
 275         case offsetof(struct user_regs_struct, cs):
 276         case offsetof(struct user_regs_struct, ss):
 277                 break;
 278         }
 279         return *pt_regs_access(task_pt_regs(task), offset);
 280 }
 281
 282 static int set_segment_reg(struct task_struct *task,
 283                            unsigned long offset, u16 value)
 284 {
 285         /*
 286          * The value argument was already truncated to 16 bits.
 287          */
 288         if (invalid_selector(value))
 289                 return -EIO;
 290
 291         switch (offset) {
 292         case offsetof(struct user_regs_struct,fs):
 293                 /*
 294                  * If this is setting fs as for normal 64-bit use but
 295                  * setting fs_base has implicitly changed it, leave it.
 296                  */
 297                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 298                      task->thread.fs != 0) ||
 299                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 300                      task->thread.fs == 0))
 301                         break;
 302                 task->thread.fsindex = value;
 303                 if (task == current)
 304                         loadsegment(fs, task->thread.fsindex);
 305                 break;
 306         case offsetof(struct user_regs_struct,gs):
 307                 /*
 308                  * If this is setting gs as for normal 64-bit use but
 309                  * setting gs_base has implicitly changed it, leave it.
 310                  */
 311                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 312                      task->thread.gs != 0) ||
 313                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 314                      task->thread.gs == 0))
 315                         break;
 316                 task->thread.gsindex = value;
 317                 if (task == current)
 318                         load_gs_index(task->thread.gsindex);
 319                 break;
 320         case offsetof(struct user_regs_struct,ds):
 321                 task->thread.ds = value;
 322                 if (task == current)
 323                         loadsegment(ds, task->thread.ds);
 324                 break;
 325         case offsetof(struct user_regs_struct,es):
 326                 task->thread.es = value;
 327                 if (task == current)
 328                         loadsegment(es, task->thread.es);
 329                 break;
 330
 331                 /*
 332                  * Can't actually change these in 64-bit mode.
 333                  */
 334         case offsetof(struct user_regs_struct,cs):
 335                 if (unlikely(value == 0))
 336                         return -EIO;
 337 #ifdef CONFIG_IA32_EMULATION
 338                 if (test_tsk_thread_flag(task, TIF_IA32))
 339                         task_pt_regs(task)->cs = value;
 340 #endif
 341                 break;
 342         case offsetof(struct user_regs_struct,ss):
 343                 if (unlikely(value == 0))
 344                         return -EIO;
 345 #ifdef CONFIG_IA32_EMULATION
 346                 if (test_tsk_thread_flag(task, TIF_IA32))
 347                         task_pt_regs(task)->ss = value;
 348 #endif
 349                 break;
 350         }
 351
 352         return 0;
 353 }
 354
 355 #endif  /* CONFIG_X86_32 */
 356
 357 static unsigned long get_flags(struct task_struct *task)
 358 {
 359         unsigned long retval = task_pt_regs(task)->flags;
 360
 361         /*
 362          * If the debugger set TF, hide it from the readout.
 363          */
 364         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 365                 retval &= ~X86_EFLAGS_TF;
 366
 367         return retval;
 368 }
 369
 370 static int set_flags(struct task_struct *task, unsigned long value)
 371 {
 372         struct pt_regs *regs = task_pt_regs(task);
 373
 374         /*
 375          * If the user value contains TF, mark that
 376          * it was not "us" (the debugger) that set it.
 377          * If not, make sure it stays set if we had.
 378          */
 379         if (value & X86_EFLAGS_TF)
 380                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 381         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 382                 value |= X86_EFLAGS_TF;
 383
 384         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 385
 386         return 0;
 387 }
 388
 389 static int putreg(struct task_struct *child,
 390                   unsigned long offset, unsigned long value)
 391 {
 392         switch (offset) {
 393         case offsetof(struct user_regs_struct, cs):
 394         case offsetof(struct user_regs_struct, ds):
 395         case offsetof(struct user_regs_struct, es):
 396         case offsetof(struct user_regs_struct, fs):
 397         case offsetof(struct user_regs_struct, gs):
 398         case offsetof(struct user_regs_struct, ss):
 399                 return set_segment_reg(child, offset, value);
 400
 401         case offsetof(struct user_regs_struct, flags):
 402                 return set_flags(child, value);
 403
 404 #ifdef CONFIG_X86_64
 405         case offsetof(struct user_regs_struct,fs_base):
 406                 if (value >= TASK_SIZE_OF(child))
 407                         return -EIO;
 408                 /*
 409                  * When changing the segment base, use do_arch_prctl
 410                  * to set either thread.fs or thread.fsindex and the
 411                  * corresponding GDT slot.
 412                  */
 413                 if (child->thread.fs != value)
 414                         return do_arch_prctl(child, ARCH_SET_FS, value);
 415                 return 0;
 416         case offsetof(struct user_regs_struct,gs_base):
 417                 /*
 418                  * Exactly the same here as the %fs handling above.
 419                  */
 420                 if (value >= TASK_SIZE_OF(child))
 421                         return -EIO;
 422                 if (child->thread.gs != value)
 423                         return do_arch_prctl(child, ARCH_SET_GS, value);
 424                 return 0;
 425 #endif
 426         }
 427
 428         *pt_regs_access(task_pt_regs(child), offset) = value;
 429         return 0;
 430 }
 431
 432 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 433 {
 434         switch (offset) {
 435         case offsetof(struct user_regs_struct, cs):
 436         case offsetof(struct user_regs_struct, ds):
 437         case offsetof(struct user_regs_struct, es):
 438         case offsetof(struct user_regs_struct, fs):
 439         case offsetof(struct user_regs_struct, gs):
 440         case offsetof(struct user_regs_struct, ss):
 441                 return get_segment_reg(task, offset);
 442
 443         case offsetof(struct user_regs_struct, flags):
 444                 return get_flags(task);
 445
 446 #ifdef CONFIG_X86_64
 447         case offsetof(struct user_regs_struct, fs_base): {
 448                 /*
 449                  * do_arch_prctl may have used a GDT slot instead of
 450                  * the MSR.  To userland, it appears the same either
 451                  * way, except the %fs segment selector might not be 0.
 452                  */
 453                 unsigned int seg = task->thread.fsindex;
 454                 if (task->thread.fs != 0)
 455                         return task->thread.fs;
 456                 if (task == current)
 457                         asm("movl %%fs,%0" : "=r" (seg));
 458                 if (seg != FS_TLS_SEL)
 459                         return 0;
 460                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 461         }
 462         case offsetof(struct user_regs_struct, gs_base): {
 463                 /*
 464                  * Exactly the same here as the %fs handling above.
 465                  */
 466                 unsigned int seg = task->thread.gsindex;
 467                 if (task->thread.gs != 0)
 468                         return task->thread.gs;
 469                 if (task == current)
 470                         asm("movl %%gs,%0" : "=r" (seg));
 471                 if (seg != GS_TLS_SEL)
 472                         return 0;
 473                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 474         }
 475 #endif
 476         }
 477
 478         return *pt_regs_access(task_pt_regs(task), offset);
 479 }
 480
 481 static int genregs_get(struct task_struct *target,
 482                        const struct user_regset *regset,
 483                        unsigned int pos, unsigned int count,
 484                        void *kbuf, void __user *ubuf)
 485 {
 486         if (kbuf) {
 487                 unsigned long *k = kbuf;
 488                 while (count >= sizeof(*k)) {
 489                         *k++ = getreg(target, pos);
 490                         count -= sizeof(*k);
 491                         pos += sizeof(*k);
 492                 }
 493         } else {
 494                 unsigned long __user *u = ubuf;
 495                 while (count >= sizeof(*u)) {
 496                         if (__put_user(getreg(target, pos), u++))
 497                                 return -EFAULT;
 498                         count -= sizeof(*u);
 499                         pos += sizeof(*u);
 500                 }
 501         }
 502
 503         return 0;
 504 }
 505
 506 static int genregs_set(struct task_struct *target,
 507                        const struct user_regset *regset,
 508                        unsigned int pos, unsigned int count,
 509                        const void *kbuf, const void __user *ubuf)
 510 {
 511         int ret = 0;
 512         if (kbuf) {
 513                 const unsigned long *k = kbuf;
 514                 while (count >= sizeof(*k) && !ret) {
 515                         ret = putreg(target, pos, *k++);
 516                         count -= sizeof(*k);
 517                         pos += sizeof(*k);
 518                 }
 519         } else {
 520                 const unsigned long  __user *u = ubuf;
 521                 while (count >= sizeof(*u) && !ret) {
 522                         unsigned long word;
 523                         ret = __get_user(word, u++);
 524                         if (ret)
 525                                 break;
 526                         ret = putreg(target, pos, word);
 527                         count -= sizeof(*u);
 528                         pos += sizeof(*u);
 529                 }
 530         }
 531         return ret;
 532 }
 533
 534 static void ptrace_triggered(struct perf_event *bp, int nmi,
 535                              struct perf_sample_data *data,
 536                              struct pt_regs *regs)
 537 {
 538         int i;
 539         struct thread_struct *thread = &(current->thread);
 540
 541         /*
 542          * Store in the virtual DR6 register the fact that the breakpoint
 543          * was hit so the thread's debugger will see it.
 544          */
 545         for (i = 0; i < HBP_NUM; i++) {
 546                 if (thread->ptrace_bps[i] == bp)
 547                         break;
 548         }
 549
 550         thread->debugreg6 |= (DR_TRAP0 << i);
 551 }
 552
 553 /*
 554  * Walk through every ptrace breakpoints for this thread and
 555  * build the dr7 value on top of their attributes.
 556  *
 557  */
 558 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 559 {
 560         int i;
 561         int dr7 = 0;
 562         struct arch_hw_breakpoint *info;
 563
 564         for (i = 0; i < HBP_NUM; i++) {
 565                 if (bp[i] && !bp[i]->attr.disabled) {
 566                         info = counter_arch_bp(bp[i]);
 567                         dr7 |= encode_dr7(i, info->len, info->type);
 568                 }
 569         }
 570
 571         return dr7;
 572 }
 573
 574 static int
 575 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 576                          struct task_struct *tsk, int disabled)
 577 {
 578         int err;
 579         int gen_len, gen_type;
 580         struct perf_event_attr attr;
 581
 582         /*
 583          * We shoud have at least an inactive breakpoint at this
 584          * slot. It means the user is writing dr7 without having
 585          * written the address register first
 586          */
 587         if (!bp)
 588                 return -EINVAL;
 589
 590         err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
 591         if (err)
 592                 return err;
 593
 594         attr = bp->attr;
 595         attr.bp_len = gen_len;
 596         attr.bp_type = gen_type;
 597         attr.disabled = disabled;
 598
 599         return modify_user_hw_breakpoint(bp, &attr);
 600 }
 601
 602 /*
 603  * Handle ptrace writes to debug register 7.
 604  */
 605 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 606 {
 607         struct thread_struct *thread = &(tsk->thread);
 608         unsigned long old_dr7;
 609         int i, orig_ret = 0, rc = 0;
 610         int enabled, second_pass = 0;
 611         unsigned len, type;
 612         struct perf_event *bp;
 613
 614         data &= ~DR_CONTROL_RESERVED;
 615         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 616 restore:
 617         /*
 618          * Loop through all the hardware breakpoints, making the
 619          * appropriate changes to each.
 620          */
 621         for (i = 0; i < HBP_NUM; i++) {
 622                 enabled = decode_dr7(data, i, &len, &type);
 623                 bp = thread->ptrace_bps[i];
 624
 625                 if (!enabled) {
 626                         if (bp) {
 627                                 /*
 628                                  * Don't unregister the breakpoints right-away,
 629                                  * unless all register_user_hw_breakpoint()
 630                                  * requests have succeeded. This prevents
 631                                  * any window of opportunity for debug
 632                                  * register grabbing by other users.
 633                                  */
 634                                 if (!second_pass)
 635                                         continue;
 636
 637                                 rc = ptrace_modify_breakpoint(bp, len, type,
 638                                                               tsk, 1);
 639                                 if (rc)
 640                                         break;
 641                         }
 642                         continue;
 643                 }
 644
 645                 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
 646                 if (rc)
 647                         break;
 648         }
 649         /*
 650          * Make a second pass to free the remaining unused breakpoints
 651          * or to restore the original breakpoints if an error occurred.
 652          */
 653         if (!second_pass) {
 654                 second_pass = 1;
 655                 if (rc < 0) {
 656                         orig_ret = rc;
 657                         data = old_dr7;
 658                 }
 659                 goto restore;
 660         }
 661         return ((orig_ret < 0) ? orig_ret : rc);
 662 }
 663
 664 /*
 665  * Handle PTRACE_PEEKUSR calls for the debug register area.
 666  */
 667 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 668 {
 669         struct thread_struct *thread = &(tsk->thread);
 670         unsigned long val = 0;
 671
 672         if (n < HBP_NUM) {
 673                 struct perf_event *bp;
 674                 bp = thread->ptrace_bps[n];
 675                 if (!bp)
 676                         return 0;
 677                 val = bp->hw.info.address;
 678         } else if (n == 6) {
 679                 val = thread->debugreg6;
 680          } else if (n == 7) {
 681                 val = thread->ptrace_dr7;
 682         }
 683         return val;
 684 }
 685
 686 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 687                                       unsigned long addr)
 688 {
 689         struct perf_event *bp;
 690         struct thread_struct *t = &tsk->thread;
 691         struct perf_event_attr attr;
 692
 693         if (!t->ptrace_bps[nr]) {
 694                 hw_breakpoint_init(&attr);
 695                 /*
 696                  * Put stub len and type to register (reserve) an inactive but
 697                  * correct bp
 698                  */
 699                 attr.bp_addr = addr;
 700                 attr.bp_len = HW_BREAKPOINT_LEN_1;
 701                 attr.bp_type = HW_BREAKPOINT_W;
 702                 attr.disabled = 1;
 703
 704                 bp = register_user_hw_breakpoint(&attr, ptrace_triggered, tsk);
 705
 706                 /*
 707                  * CHECKME: the previous code returned -EIO if the addr wasn't
 708                  * a valid task virtual addr. The new one will return -EINVAL in
 709                  *  this case.
 710                  * -EINVAL may be what we want for in-kernel breakpoints users,
 711                  * but -EIO looks better for ptrace, since we refuse a register
 712                  * writing for the user. And anyway this is the previous
 713                  * behaviour.
 714                  */
 715                 if (IS_ERR(bp))
 716                         return PTR_ERR(bp);
 717
 718                 t->ptrace_bps[nr] = bp;
 719         } else {
 720                 int err;
 721
 722                 bp = t->ptrace_bps[nr];
 723
 724                 attr = bp->attr;
 725                 attr.bp_addr = addr;
 726                 err = modify_user_hw_breakpoint(bp, &attr);
 727                 if (err)
 728                         return err;
 729         }
 730
 731
 732         return 0;
 733 }
 734
 735 /*
 736  * Handle PTRACE_POKEUSR calls for the debug register area.
 737  */
 738 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
 739 {
 740         struct thread_struct *thread = &(tsk->thread);
 741         int rc = 0;
 742
 743         /* There are no DR4 or DR5 registers */
 744         if (n == 4 || n == 5)
 745                 return -EIO;
 746
 747         if (n == 6) {
 748                 thread->debugreg6 = val;
 749                 goto ret_path;
 750         }
 751         if (n < HBP_NUM) {
 752                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 753                 if (rc)
 754                         return rc;
 755         }
 756         /* All that's left is DR7 */
 757         if (n == 7) {
 758                 rc = ptrace_write_dr7(tsk, val);
 759                 if (!rc)
 760                         thread->ptrace_dr7 = val;
 761         }
 762
 763 ret_path:
 764         return rc;
 765 }
 766
 767 /*
 768  * These access the current or another (stopped) task's io permission
 769  * bitmap for debugging or core dump.
 770  */
 771 static int ioperm_active(struct task_struct *target,
 772                          const struct user_regset *regset)
 773 {
 774         return target->thread.io_bitmap_max / regset->size;
 775 }
 776
 777 static int ioperm_get(struct task_struct *target,
 778                       const struct user_regset *regset,
 779                       unsigned int pos, unsigned int count,
 780                       void *kbuf, void __user *ubuf)
 781 {
 782         if (!target->thread.io_bitmap_ptr)
 783                 return -ENXIO;
 784
 785         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 786                                    target->thread.io_bitmap_ptr,
 787                                    0, IO_BITMAP_BYTES);
 788 }
 789
 790 #ifdef CONFIG_X86_PTRACE_BTS
 791 /*
 792  * A branch trace store context.
 793  *
 794  * Contexts may only be installed by ptrace_bts_config() and only for
 795  * ptraced tasks.
 796  *
 797  * Contexts are destroyed when the tracee is detached from the tracer.
 798  * The actual destruction work requires interrupts enabled, so the
 799  * work is deferred and will be scheduled during __ptrace_unlink().
 800  *
 801  * Contexts hold an additional task_struct reference on the traced
 802  * task, as well as a reference on the tracer's mm.
 803  *
 804  * Ptrace already holds a task_struct for the duration of ptrace operations,
 805  * but since destruction is deferred, it may be executed after both
 806  * tracer and tracee exited.
 807  */
 808 struct bts_context {
 809         /* The branch trace handle. */
 810         struct bts_tracer       *tracer;
 811
 812         /* The buffer used to store the branch trace and its size. */
 813         void                    *buffer;
 814         unsigned int            size;
 815
 816         /* The mm that paid for the above buffer. */
 817         struct mm_struct        *mm;
 818
 819         /* The task this context belongs to. */
 820         struct task_struct      *task;
 821
 822         /* The signal to send on a bts buffer overflow. */
 823         unsigned int            bts_ovfl_signal;
 824
 825         /* The work struct to destroy a context. */
 826         struct work_struct      work;
 827 };
 828
 829 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
 830 {
 831         void *buffer = NULL;
 832         int err = -ENOMEM;
 833
 834         err = account_locked_memory(current->mm, current->signal->rlim, size);
 835         if (err < 0)
 836                 return err;
 837
 838         buffer = kzalloc(size, GFP_KERNEL);
 839         if (!buffer)
 840                 goto out_refund;
 841
 842         context->buffer = buffer;
 843         context->size = size;
 844         context->mm = get_task_mm(current);
 845
 846         return 0;
 847
 848  out_refund:
 849         refund_locked_memory(current->mm, size);
 850         return err;
 851 }
 852
 853 static inline void free_bts_buffer(struct bts_context *context)
 854 {
 855         if (!context->buffer)
 856                 return;
 857
 858         kfree(context->buffer);
 859         context->buffer = NULL;
 860
 861         refund_locked_memory(context->mm, context->size);
 862         context->size = 0;
 863
 864         mmput(context->mm);
 865         context->mm = NULL;
 866 }
 867
 868 static void free_bts_context_work(struct work_struct *w)
 869 {
 870         struct bts_context *context;
 871
 872         context = container_of(w, struct bts_context, work);
 873
 874         ds_release_bts(context->tracer);
 875         put_task_struct(context->task);
 876         free_bts_buffer(context);
 877         kfree(context);
 878 }
 879
 880 static inline void free_bts_context(struct bts_context *context)
 881 {
 882         INIT_WORK(&context->work, free_bts_context_work);
 883         schedule_work(&context->work);
 884 }
 885
 886 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
 887 {
 888         struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
 889         if (context) {
 890                 context->task = task;
 891                 task->bts = context;
 892
 893                 get_task_struct(task);
 894         }
 895
 896         return context;
 897 }
 898
 899 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 900                                   struct bts_struct __user *out)
 901 {
 902         struct bts_context *context;
 903         const struct bts_trace *trace;
 904         struct bts_struct bts;
 905         const unsigned char *at;
 906         int error;
 907
 908         context = child->bts;
 909         if (!context)
 910                 return -ESRCH;
 911
 912         trace = ds_read_bts(context->tracer);
 913         if (!trace)
 914                 return -ESRCH;
 915
 916         at = trace->ds.top - ((index + 1) * trace->ds.size);
 917         if ((void *)at < trace->ds.begin)
 918                 at += (trace->ds.n * trace->ds.size);
 919
 920         if (!trace->read)
 921                 return -EOPNOTSUPP;
 922
 923         error = trace->read(context->tracer, at, &bts);
 924         if (error < 0)
 925                 return error;
 926
 927         if (copy_to_user(out, &bts, sizeof(bts)))
 928                 return -EFAULT;
 929
 930         return sizeof(bts);
 931 }
 932
 933 static int ptrace_bts_drain(struct task_struct *child,
 934                             long size,
 935                             struct bts_struct __user *out)
 936 {
 937         struct bts_context *context;
 938         const struct bts_trace *trace;
 939         const unsigned char *at;
 940         int error, drained = 0;
 941
 942         context = child->bts;
 943         if (!context)
 944                 return -ESRCH;
 945
 946         trace = ds_read_bts(context->tracer);
 947         if (!trace)
 948                 return -ESRCH;
 949
 950         if (!trace->read)
 951                 return -EOPNOTSUPP;
 952
 953         if (size < (trace->ds.top - trace->ds.begin))
 954                 return -EIO;
 955
 956         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 957              out++, drained++, at += trace->ds.size) {
 958                 struct bts_struct bts;
 959
 960                 error = trace->read(context->tracer, at, &bts);
 961                 if (error < 0)
 962                         return error;
 963
 964                 if (copy_to_user(out, &bts, sizeof(bts)))
 965                         return -EFAULT;
 966         }
 967
 968         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 969
 970         error = ds_reset_bts(context->tracer);
 971         if (error < 0)
 972                 return error;
 973
 974         return drained;
 975 }
 976
 977 static int ptrace_bts_config(struct task_struct *child,
 978                              long cfg_size,
 979                              const struct ptrace_bts_config __user *ucfg)
 980 {
 981         struct bts_context *context;
 982         struct ptrace_bts_config cfg;
 983         unsigned int flags = 0;
 984
 985         if (cfg_size < sizeof(cfg))
 986                 return -EIO;
 987
 988         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 989                 return -EFAULT;
 990
 991         context = child->bts;
 992         if (!context)
 993                 context = alloc_bts_context(child);
 994         if (!context)
 995                 return -ENOMEM;
 996
 997         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 998                 if (!cfg.signal)
 999                         return -EINVAL;
1000
1001                 return -EOPNOTSUPP;
1002                 context->bts_ovfl_signal = cfg.signal;
1003         }
1004
1005         ds_release_bts(context->tracer);
1006         context->tracer = NULL;
1007
1008         if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
1009                 int err;
1010
1011                 free_bts_buffer(context);
1012                 if (!cfg.size)
1013                         return 0;
1014
1015                 err = alloc_bts_buffer(context, cfg.size);
1016                 if (err < 0)
1017                         return err;
1018         }
1019
1020         if (cfg.flags & PTRACE_BTS_O_TRACE)
1021                 flags |= BTS_USER;
1022
1023         if (cfg.flags & PTRACE_BTS_O_SCHED)
1024                 flags |= BTS_TIMESTAMPS;
1025
1026         context->tracer =
1027                 ds_request_bts_task(child, context->buffer, context->size,
1028                                     NULL, (size_t)-1, flags);
1029         if (unlikely(IS_ERR(context->tracer))) {
1030                 int error = PTR_ERR(context->tracer);
1031
1032                 free_bts_buffer(context);
1033                 context->tracer = NULL;
1034                 return error;
1035         }
1036
1037         return sizeof(cfg);
1038 }
1039
1040 static int ptrace_bts_status(struct task_struct *child,
1041                              long cfg_size,
1042                              struct ptrace_bts_config __user *ucfg)
1043 {
1044         struct bts_context *context;
1045         const struct bts_trace *trace;
1046         struct ptrace_bts_config cfg;
1047
1048         context = child->bts;
1049         if (!context)
1050                 return -ESRCH;
1051
1052         if (cfg_size < sizeof(cfg))
1053                 return -EIO;
1054
1055         trace = ds_read_bts(context->tracer);
1056         if (!trace)
1057                 return -ESRCH;
1058
1059         memset(&cfg, 0, sizeof(cfg));
1060         cfg.size        = trace->ds.end - trace->ds.begin;
1061         cfg.signal      = context->bts_ovfl_signal;
1062         cfg.bts_size    = sizeof(struct bts_struct);
1063
1064         if (cfg.signal)
1065                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
1066
1067         if (trace->ds.flags & BTS_USER)
1068                 cfg.flags |= PTRACE_BTS_O_TRACE;
1069
1070         if (trace->ds.flags & BTS_TIMESTAMPS)
1071                 cfg.flags |= PTRACE_BTS_O_SCHED;
1072
1073         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
1074                 return -EFAULT;
1075
1076         return sizeof(cfg);
1077 }
1078
1079 static int ptrace_bts_clear(struct task_struct *child)
1080 {
1081         struct bts_context *context;
1082         const struct bts_trace *trace;
1083
1084         context = child->bts;
1085         if (!context)
1086                 return -ESRCH;
1087
1088         trace = ds_read_bts(context->tracer);
1089         if (!trace)
1090                 return -ESRCH;
1091
1092         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
1093
1094         return ds_reset_bts(context->tracer);
1095 }
1096
1097 static int ptrace_bts_size(struct task_struct *child)
1098 {
1099         struct bts_context *context;
1100         const struct bts_trace *trace;
1101
1102         context = child->bts;
1103         if (!context)
1104                 return -ESRCH;
1105
1106         trace = ds_read_bts(context->tracer);
1107         if (!trace)
1108                 return -ESRCH;
1109
1110         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
1111 }
1112
1113 /*
1114  * Called from __ptrace_unlink() after the child has been moved back
1115  * to its original parent.
1116  */
1117 void ptrace_bts_untrace(struct task_struct *child)
1118 {
1119         if (unlikely(child->bts)) {
1120                 free_bts_context(child->bts);
1121                 child->bts = NULL;
1122         }
1123 }
1124 #endif /* CONFIG_X86_PTRACE_BTS */
1125
1126 /*
1127  * Called by kernel/ptrace.c when detaching..
1128  *
1129  * Make sure the single step bit is not set.
1130  */
1131 void ptrace_disable(struct task_struct *child)
1132 {
1133         user_disable_single_step(child);
1134 #ifdef TIF_SYSCALL_EMU
1135         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
1136 #endif
1137 }
1138
1139 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1140 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
1141 #endif
1142
1143 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1144 {
1145         int ret;
1146         unsigned long __user *datap = (unsigned long __user *)data;
1147
1148         switch (request) {
1149         /* read the word at location addr in the USER area. */
1150         case PTRACE_PEEKUSR: {
1151                 unsigned long tmp;
1152
1153                 ret = -EIO;
1154                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1155                     addr >= sizeof(struct user))
1156                         break;
1157
1158                 tmp = 0;  /* Default return condition */
1159                 if (addr < sizeof(struct user_regs_struct))
1160                         tmp = getreg(child, addr);
1161                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1162                          addr <= offsetof(struct user, u_debugreg[7])) {
1163                         addr -= offsetof(struct user, u_debugreg[0]);
1164                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1165                 }
1166                 ret = put_user(tmp, datap);
1167                 break;
1168         }
1169
1170         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1171                 ret = -EIO;
1172                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1173                     addr >= sizeof(struct user))
1174                         break;
1175
1176                 if (addr < sizeof(struct user_regs_struct))
1177                         ret = putreg(child, addr, data);
1178                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1179                          addr <= offsetof(struct user, u_debugreg[7])) {
1180                         addr -= offsetof(struct user, u_debugreg[0]);
1181                         ret = ptrace_set_debugreg(child,
1182                                                   addr / sizeof(data), data);
1183                 }
1184                 break;
1185
1186         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1187                 return copy_regset_to_user(child,
1188                                            task_user_regset_view(current),
1189                                            REGSET_GENERAL,
1190                                            0, sizeof(struct user_regs_struct),
1191                                            datap);
1192
1193         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1194                 return copy_regset_from_user(child,
1195                                              task_user_regset_view(current),
1196                                              REGSET_GENERAL,
1197                                              0, sizeof(struct user_regs_struct),
1198                                              datap);
1199
1200         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1201                 return copy_regset_to_user(child,
1202                                            task_user_regset_view(current),
1203                                            REGSET_FP,
1204                                            0, sizeof(struct user_i387_struct),
1205                                            datap);
1206
1207         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1208                 return copy_regset_from_user(child,
1209                                              task_user_regset_view(current),
1210                                              REGSET_FP,
1211                                              0, sizeof(struct user_i387_struct),
1212                                              datap);
1213
1214 #ifdef CONFIG_X86_32
1215         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1216                 return copy_regset_to_user(child, &user_x86_32_view,
1217                                            REGSET_XFP,
1218                                            0, sizeof(struct user_fxsr_struct),
1219                                            datap) ? -EIO : 0;
1220
1221         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1222                 return copy_regset_from_user(child, &user_x86_32_view,
1223                                              REGSET_XFP,
1224                                              0, sizeof(struct user_fxsr_struct),
1225                                              datap) ? -EIO : 0;
1226 #endif
1227
1228 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1229         case PTRACE_GET_THREAD_AREA:
1230                 if (addr < 0)
1231                         return -EIO;
1232                 ret = do_get_thread_area(child, addr,
1233                                          (struct user_desc __user *) data);
1234                 break;
1235
1236         case PTRACE_SET_THREAD_AREA:
1237                 if (addr < 0)
1238                         return -EIO;
1239                 ret = do_set_thread_area(child, addr,
1240                                          (struct user_desc __user *) data, 0);
1241                 break;
1242 #endif
1243
1244 #ifdef CONFIG_X86_64
1245                 /* normal 64bit interface to access TLS data.
1246                    Works just like arch_prctl, except that the arguments
1247                    are reversed. */
1248         case PTRACE_ARCH_PRCTL:
1249                 ret = do_arch_prctl(child, data, addr);
1250                 break;
1251 #endif
1252
1253         /*
1254          * These bits need more cooking - not enabled yet:
1255          */
1256 #ifdef CONFIG_X86_PTRACE_BTS
1257         case PTRACE_BTS_CONFIG:
1258                 ret = ptrace_bts_config
1259                         (child, data, (struct ptrace_bts_config __user *)addr);
1260                 break;
1261
1262         case PTRACE_BTS_STATUS:
1263                 ret = ptrace_bts_status
1264                         (child, data, (struct ptrace_bts_config __user *)addr);
1265                 break;
1266
1267         case PTRACE_BTS_SIZE:
1268                 ret = ptrace_bts_size(child);
1269                 break;
1270
1271         case PTRACE_BTS_GET:
1272                 ret = ptrace_bts_read_record
1273                         (child, data, (struct bts_struct __user *) addr);
1274                 break;
1275
1276         case PTRACE_BTS_CLEAR:
1277                 ret = ptrace_bts_clear(child);
1278                 break;
1279
1280         case PTRACE_BTS_DRAIN:
1281                 ret = ptrace_bts_drain
1282                         (child, data, (struct bts_struct __user *) addr);
1283                 break;
1284 #endif /* CONFIG_X86_PTRACE_BTS */
1285
1286         default:
1287                 ret = ptrace_request(child, request, addr, data);
1288                 break;
1289         }
1290
1291         return ret;
1292 }
1293
1294 #ifdef CONFIG_IA32_EMULATION
1295
1296 #include <linux/compat.h>
1297 #include <linux/syscalls.h>
1298 #include <asm/ia32.h>
1299 #include <asm/user32.h>
1300
1301 #define R32(l,q)                                                        \
1302         case offsetof(struct user32, regs.l):                           \
1303                 regs->q = value; break
1304
1305 #define SEG32(rs)                                                       \
1306         case offsetof(struct user32, regs.rs):                          \
1307                 return set_segment_reg(child,                           \
1308                                        offsetof(struct user_regs_struct, rs), \
1309                                        value);                          \
1310                 break
1311
1312 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1313 {
1314         struct pt_regs *regs = task_pt_regs(child);
1315
1316         switch (regno) {
1317
1318         SEG32(cs);
1319         SEG32(ds);
1320         SEG32(es);
1321         SEG32(fs);
1322         SEG32(gs);
1323         SEG32(ss);
1324
1325         R32(ebx, bx);
1326         R32(ecx, cx);
1327         R32(edx, dx);
1328         R32(edi, di);
1329         R32(esi, si);
1330         R32(ebp, bp);
1331         R32(eax, ax);
1332         R32(eip, ip);
1333         R32(esp, sp);
1334
1335         case offsetof(struct user32, regs.orig_eax):
1336                 /*
1337                  * A 32-bit debugger setting orig_eax means to restore
1338                  * the state of the task restarting a 32-bit syscall.
1339                  * Make sure we interpret the -ERESTART* codes correctly
1340                  * in case the task is not actually still sitting at the
1341                  * exit from a 32-bit syscall with TS_COMPAT still set.
1342                  */
1343                 regs->orig_ax = value;
1344                 if (syscall_get_nr(child, regs) >= 0)
1345                         task_thread_info(child)->status |= TS_COMPAT;
1346                 break;
1347
1348         case offsetof(struct user32, regs.eflags):
1349                 return set_flags(child, value);
1350
1351         case offsetof(struct user32, u_debugreg[0]) ...
1352                 offsetof(struct user32, u_debugreg[7]):
1353                 regno -= offsetof(struct user32, u_debugreg[0]);
1354                 return ptrace_set_debugreg(child, regno / 4, value);
1355
1356         default:
1357                 if (regno > sizeof(struct user32) || (regno & 3))
1358                         return -EIO;
1359
1360                 /*
1361                  * Other dummy fields in the virtual user structure
1362                  * are ignored
1363                  */
1364                 break;
1365         }
1366         return 0;
1367 }
1368
1369 #undef R32
1370 #undef SEG32
1371
1372 #define R32(l,q)                                                        \
1373         case offsetof(struct user32, regs.l):                           \
1374                 *val = regs->q; break
1375
1376 #define SEG32(rs)                                                       \
1377         case offsetof(struct user32, regs.rs):                          \
1378                 *val = get_segment_reg(child,                           \
1379                                        offsetof(struct user_regs_struct, rs)); \
1380                 break
1381
1382 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1383 {
1384         struct pt_regs *regs = task_pt_regs(child);
1385
1386         switch (regno) {
1387
1388         SEG32(ds);
1389         SEG32(es);
1390         SEG32(fs);
1391         SEG32(gs);
1392
1393         R32(cs, cs);
1394         R32(ss, ss);
1395         R32(ebx, bx);
1396         R32(ecx, cx);
1397         R32(edx, dx);
1398         R32(edi, di);
1399         R32(esi, si);
1400         R32(ebp, bp);
1401         R32(eax, ax);
1402         R32(orig_eax, orig_ax);
1403         R32(eip, ip);
1404         R32(esp, sp);
1405
1406         case offsetof(struct user32, regs.eflags):
1407                 *val = get_flags(child);
1408                 break;
1409
1410         case offsetof(struct user32, u_debugreg[0]) ...
1411                 offsetof(struct user32, u_debugreg[7]):
1412                 regno -= offsetof(struct user32, u_debugreg[0]);
1413                 *val = ptrace_get_debugreg(child, regno / 4);
1414                 break;
1415
1416         default:
1417                 if (regno > sizeof(struct user32) || (regno & 3))
1418                         return -EIO;
1419
1420                 /*
1421                  * Other dummy fields in the virtual user structure
1422                  * are ignored
1423                  */
1424                 *val = 0;
1425                 break;
1426         }
1427         return 0;
1428 }
1429
1430 #undef R32
1431 #undef SEG32
1432
1433 static int genregs32_get(struct task_struct *target,
1434                          const struct user_regset *regset,
1435                          unsigned int pos, unsigned int count,
1436                          void *kbuf, void __user *ubuf)
1437 {
1438         if (kbuf) {
1439                 compat_ulong_t *k = kbuf;
1440                 while (count >= sizeof(*k)) {
1441                         getreg32(target, pos, k++);
1442                         count -= sizeof(*k);
1443                         pos += sizeof(*k);
1444                 }
1445         } else {
1446                 compat_ulong_t __user *u = ubuf;
1447                 while (count >= sizeof(*u)) {
1448                         compat_ulong_t word;
1449                         getreg32(target, pos, &word);
1450                         if (__put_user(word, u++))
1451                                 return -EFAULT;
1452                         count -= sizeof(*u);
1453                         pos += sizeof(*u);
1454                 }
1455         }
1456
1457         return 0;
1458 }
1459
1460 static int genregs32_set(struct task_struct *target,
1461                          const struct user_regset *regset,
1462                          unsigned int pos, unsigned int count,
1463                          const void *kbuf, const void __user *ubuf)
1464 {
1465         int ret = 0;
1466         if (kbuf) {
1467                 const compat_ulong_t *k = kbuf;
1468                 while (count >= sizeof(*k) && !ret) {
1469                         ret = putreg32(target, pos, *k++);
1470                         count -= sizeof(*k);
1471                         pos += sizeof(*k);
1472                 }
1473         } else {
1474                 const compat_ulong_t __user *u = ubuf;
1475                 while (count >= sizeof(*u) && !ret) {
1476                         compat_ulong_t word;
1477                         ret = __get_user(word, u++);
1478                         if (ret)
1479                                 break;
1480                         ret = putreg32(target, pos, word);
1481                         count -= sizeof(*u);
1482                         pos += sizeof(*u);
1483                 }
1484         }
1485         return ret;
1486 }
1487
1488 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1489                         compat_ulong_t caddr, compat_ulong_t cdata)
1490 {
1491         unsigned long addr = caddr;
1492         unsigned long data = cdata;
1493         void __user *datap = compat_ptr(data);
1494         int ret;
1495         __u32 val;
1496
1497         switch (request) {
1498         case PTRACE_PEEKUSR:
1499                 ret = getreg32(child, addr, &val);
1500                 if (ret == 0)
1501                         ret = put_user(val, (__u32 __user *)datap);
1502                 break;
1503
1504         case PTRACE_POKEUSR:
1505                 ret = putreg32(child, addr, data);
1506                 break;
1507
1508         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1509                 return copy_regset_to_user(child, &user_x86_32_view,
1510                                            REGSET_GENERAL,
1511                                            0, sizeof(struct user_regs_struct32),
1512                                            datap);
1513
1514         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1515                 return copy_regset_from_user(child, &user_x86_32_view,
1516                                              REGSET_GENERAL, 0,
1517                                              sizeof(struct user_regs_struct32),
1518                                              datap);
1519
1520         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1521                 return copy_regset_to_user(child, &user_x86_32_view,
1522                                            REGSET_FP, 0,
1523                                            sizeof(struct user_i387_ia32_struct),
1524                                            datap);
1525
1526         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1527                 return copy_regset_from_user(
1528                         child, &user_x86_32_view, REGSET_FP,
1529                         0, sizeof(struct user_i387_ia32_struct), datap);
1530
1531         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1532                 return copy_regset_to_user(child, &user_x86_32_view,
1533                                            REGSET_XFP, 0,
1534                                            sizeof(struct user32_fxsr_struct),
1535                                            datap);
1536
1537         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1538                 return copy_regset_from_user(child, &user_x86_32_view,
1539                                              REGSET_XFP, 0,
1540                                              sizeof(struct user32_fxsr_struct),
1541                                              datap);
1542
1543         case PTRACE_GET_THREAD_AREA:
1544         case PTRACE_SET_THREAD_AREA:
1545 #ifdef CONFIG_X86_PTRACE_BTS
1546         case PTRACE_BTS_CONFIG:
1547         case PTRACE_BTS_STATUS:
1548         case PTRACE_BTS_SIZE:
1549         case PTRACE_BTS_GET:
1550         case PTRACE_BTS_CLEAR:
1551         case PTRACE_BTS_DRAIN:
1552 #endif /* CONFIG_X86_PTRACE_BTS */
1553                 return arch_ptrace(child, request, addr, data);
1554
1555         default:
1556                 return compat_ptrace_request(child, request, addr, data);
1557         }
1558
1559         return ret;
1560 }
1561
1562 #endif  /* CONFIG_IA32_EMULATION */
1563
1564 #ifdef CONFIG_X86_64
1565
1566 static const struct user_regset x86_64_regsets[] = {
1567         [REGSET_GENERAL] = {
1568                 .core_note_type = NT_PRSTATUS,
1569                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1570                 .size = sizeof(long), .align = sizeof(long),
1571                 .get = genregs_get, .set = genregs_set
1572         },
1573         [REGSET_FP] = {
1574                 .core_note_type = NT_PRFPREG,
1575                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1576                 .size = sizeof(long), .align = sizeof(long),
1577                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1578         },
1579         [REGSET_IOPERM64] = {
1580                 .core_note_type = NT_386_IOPERM,
1581                 .n = IO_BITMAP_LONGS,
1582                 .size = sizeof(long), .align = sizeof(long),
1583                 .active = ioperm_active, .get = ioperm_get
1584         },
1585 };
1586
1587 static const struct user_regset_view user_x86_64_view = {
1588         .name = "x86_64", .e_machine = EM_X86_64,
1589         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1590 };
1591
1592 #else  /* CONFIG_X86_32 */
1593
1594 #define user_regs_struct32      user_regs_struct
1595 #define genregs32_get           genregs_get
1596 #define genregs32_set           genregs_set
1597
1598 #define user_i387_ia32_struct   user_i387_struct
1599 #define user32_fxsr_struct      user_fxsr_struct
1600
1601 #endif  /* CONFIG_X86_64 */
1602
1603 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1604 static const struct user_regset x86_32_regsets[] = {
1605         [REGSET_GENERAL] = {
1606                 .core_note_type = NT_PRSTATUS,
1607                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1608                 .size = sizeof(u32), .align = sizeof(u32),
1609                 .get = genregs32_get, .set = genregs32_set
1610         },
1611         [REGSET_FP] = {
1612                 .core_note_type = NT_PRFPREG,
1613                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1614                 .size = sizeof(u32), .align = sizeof(u32),
1615                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1616         },
1617         [REGSET_XFP] = {
1618                 .core_note_type = NT_PRXFPREG,
1619                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1620                 .size = sizeof(u32), .align = sizeof(u32),
1621                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1622         },
1623         [REGSET_TLS] = {
1624                 .core_note_type = NT_386_TLS,
1625                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1626                 .size = sizeof(struct user_desc),
1627                 .align = sizeof(struct user_desc),
1628                 .active = regset_tls_active,
1629                 .get = regset_tls_get, .set = regset_tls_set
1630         },
1631         [REGSET_IOPERM32] = {
1632                 .core_note_type = NT_386_IOPERM,
1633                 .n = IO_BITMAP_BYTES / sizeof(u32),
1634                 .size = sizeof(u32), .align = sizeof(u32),
1635                 .active = ioperm_active, .get = ioperm_get
1636         },
1637 };
1638
1639 static const struct user_regset_view user_x86_32_view = {
1640         .name = "i386", .e_machine = EM_386,
1641         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1642 };
1643 #endif
1644
1645 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1646 {
1647 #ifdef CONFIG_IA32_EMULATION
1648         if (test_tsk_thread_flag(task, TIF_IA32))
1649 #endif
1650 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1651                 return &user_x86_32_view;
1652 #endif
1653 #ifdef CONFIG_X86_64
1654         return &user_x86_64_view;
1655 #endif
1656 }
1657
1658 static void fill_sigtrap_info(struct task_struct *tsk,
1659                                 struct pt_regs *regs,
1660                                 int error_code, int si_code,
1661                                 struct siginfo *info)
1662 {
1663         tsk->thread.trap_no = 1;
1664         tsk->thread.error_code = error_code;
1665
1666         memset(info, 0, sizeof(*info));
1667         info->si_signo = SIGTRAP;
1668         info->si_code = si_code;
1669         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1670 }
1671
1672 void user_single_step_siginfo(struct task_struct *tsk,
1673                                 struct pt_regs *regs,
1674                                 struct siginfo *info)
1675 {
1676         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1677 }
1678
1679 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1680                                          int error_code, int si_code)
1681 {
1682         struct siginfo info;
1683
1684         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1685         /* Send us the fake SIGTRAP */
1686         force_sig_info(SIGTRAP, &info, tsk);
1687 }
1688
1689
1690 #ifdef CONFIG_X86_32
1691 # define IS_IA32        1
1692 #elif defined CONFIG_IA32_EMULATION
1693 # define IS_IA32        is_compat_task()
1694 #else
1695 # define IS_IA32        0
1696 #endif
1697
1698 /*
1699  * We must return the syscall number to actually look up in the table.
1700  * This can be -1L to skip running any syscall at all.
1701  */
1702 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1703 {
1704         long ret = 0;
1705
1706         /*
1707          * If we stepped into a sysenter/syscall insn, it trapped in
1708          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1709          * If user-mode had set TF itself, then it's still clear from
1710          * do_debug() and we need to set it again to restore the user
1711          * state.  If we entered on the slow path, TF was already set.
1712          */
1713         if (test_thread_flag(TIF_SINGLESTEP))
1714                 regs->flags |= X86_EFLAGS_TF;
1715
1716         /* do the secure computing check first */
1717         secure_computing(regs->orig_ax);
1718
1719         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1720                 ret = -1L;
1721
1722         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1723             tracehook_report_syscall_entry(regs))
1724                 ret = -1L;
1725
1726         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1727                 trace_sys_enter(regs, regs->orig_ax);
1728
1729         if (unlikely(current->audit_context)) {
1730                 if (IS_IA32)
1731                         audit_syscall_entry(AUDIT_ARCH_I386,
1732                                             regs->orig_ax,
1733                                             regs->bx, regs->cx,
1734                                             regs->dx, regs->si);
1735 #ifdef CONFIG_X86_64
1736                 else
1737                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1738                                             regs->orig_ax,
1739                                             regs->di, regs->si,
1740                                             regs->dx, regs->r10);
1741 #endif
1742         }
1743
1744         return ret ?: regs->orig_ax;
1745 }
1746
1747 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1748 {
1749         bool step;
1750
1751         if (unlikely(current->audit_context))
1752                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1753
1754         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1755                 trace_sys_exit(regs, regs->ax);
1756
1757         /*
1758          * If TIF_SYSCALL_EMU is set, we only get here because of
1759          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1760          * We already reported this syscall instruction in
1761          * syscall_trace_enter().
1762          */
1763         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1764                         !test_thread_flag(TIF_SYSCALL_EMU);
1765         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1766                 tracehook_report_syscall_exit(regs, step);
1767 }