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
   7 #include <linux/kernel.h>
   8 #include <linux/sched.h>
   9 #include <linux/mm.h>
  10 #include <linux/smp.h>
  11 #include <linux/errno.h>
  12 #include <linux/slab.h>
  13 #include <linux/ptrace.h>
  14 #include <linux/regset.h>
  15 #include <linux/tracehook.h>
  16 #include <linux/user.h>
  17 #include <linux/elf.h>
  18 #include <linux/security.h>
  19 #include <linux/audit.h>
  20 #include <linux/seccomp.h>
  21 #include <linux/signal.h>
  22 #include <linux/perf_event.h>
  23 #include <linux/hw_breakpoint.h>
  24 #include <linux/rcupdate.h>
  25
  26 #include <asm/uaccess.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/processor.h>
  29 #include <asm/i387.h>
  30 #include <asm/fpu-internal.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/hw_breakpoint.h>
  37 #include <asm/traps.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_XSTATE,
  50         REGSET_TLS,
  51         REGSET_IOPERM32,
  52 };
  53
  54 struct pt_regs_offset {
  55         const char *name;
  56         int offset;
  57 };
  58
  59 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  60 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  61
  62 static const struct pt_regs_offset regoffset_table[] = {
  63 #ifdef CONFIG_X86_64
  64         REG_OFFSET_NAME(r15),
  65         REG_OFFSET_NAME(r14),
  66         REG_OFFSET_NAME(r13),
  67         REG_OFFSET_NAME(r12),
  68         REG_OFFSET_NAME(r11),
  69         REG_OFFSET_NAME(r10),
  70         REG_OFFSET_NAME(r9),
  71         REG_OFFSET_NAME(r8),
  72 #endif
  73         REG_OFFSET_NAME(bx),
  74         REG_OFFSET_NAME(cx),
  75         REG_OFFSET_NAME(dx),
  76         REG_OFFSET_NAME(si),
  77         REG_OFFSET_NAME(di),
  78         REG_OFFSET_NAME(bp),
  79         REG_OFFSET_NAME(ax),
  80 #ifdef CONFIG_X86_32
  81         REG_OFFSET_NAME(ds),
  82         REG_OFFSET_NAME(es),
  83         REG_OFFSET_NAME(fs),
  84         REG_OFFSET_NAME(gs),
  85 #endif
  86         REG_OFFSET_NAME(orig_ax),
  87         REG_OFFSET_NAME(ip),
  88         REG_OFFSET_NAME(cs),
  89         REG_OFFSET_NAME(flags),
  90         REG_OFFSET_NAME(sp),
  91         REG_OFFSET_NAME(ss),
  92         REG_OFFSET_END,
  93 };
  94
  95 /**
  96  * regs_query_register_offset() - query register offset from its name
  97  * @name:       the name of a register
  98  *
  99  * regs_query_register_offset() returns the offset of a register in struct
 100  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
 101  */
 102 int regs_query_register_offset(const char *name)
 103 {
 104         const struct pt_regs_offset *roff;
 105         for (roff = regoffset_table; roff->name != NULL; roff++)
 106                 if (!strcmp(roff->name, name))
 107                         return roff->offset;
 108         return -EINVAL;
 109 }
 110
 111 /**
 112  * regs_query_register_name() - query register name from its offset
 113  * @offset:     the offset of a register in struct pt_regs.
 114  *
 115  * regs_query_register_name() returns the name of a register from its
 116  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 117  */
 118 const char *regs_query_register_name(unsigned int offset)
 119 {
 120         const struct pt_regs_offset *roff;
 121         for (roff = regoffset_table; roff->name != NULL; roff++)
 122                 if (roff->offset == offset)
 123                         return roff->name;
 124         return NULL;
 125 }
 126
 127 static const int arg_offs_table[] = {
 128 #ifdef CONFIG_X86_32
 129         [0] = offsetof(struct pt_regs, ax),
 130         [1] = offsetof(struct pt_regs, dx),
 131         [2] = offsetof(struct pt_regs, cx)
 132 #else /* CONFIG_X86_64 */
 133         [0] = offsetof(struct pt_regs, di),
 134         [1] = offsetof(struct pt_regs, si),
 135         [2] = offsetof(struct pt_regs, dx),
 136         [3] = offsetof(struct pt_regs, cx),
 137         [4] = offsetof(struct pt_regs, r8),
 138         [5] = offsetof(struct pt_regs, r9)
 139 #endif
 140 };
 141
 142 /*
 143  * does not yet catch signals sent when the child dies.
 144  * in exit.c or in signal.c.
 145  */
 146
 147 /*
 148  * Determines which flags the user has access to [1 = access, 0 = no access].
 149  */
 150 #define FLAG_MASK_32            ((unsigned long)                        \
 151                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 152                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 153                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 154                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 155                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 156
 157 /*
 158  * Determines whether a value may be installed in a segment register.
 159  */
 160 static inline bool invalid_selector(u16 value)
 161 {
 162         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 163 }
 164
 165 #ifdef CONFIG_X86_32
 166
 167 #define FLAG_MASK               FLAG_MASK_32
 168
 169 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 170 {
 171         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 172         return &regs->bx + (regno >> 2);
 173 }
 174
 175 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 176 {
 177         /*
 178          * Returning the value truncates it to 16 bits.
 179          */
 180         unsigned int retval;
 181         if (offset != offsetof(struct user_regs_struct, gs))
 182                 retval = *pt_regs_access(task_pt_regs(task), offset);
 183         else {
 184                 if (task == current)
 185                         retval = get_user_gs(task_pt_regs(task));
 186                 else
 187                         retval = task_user_gs(task);
 188         }
 189         return retval;
 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         /*
 202          * For %cs and %ss we cannot permit a null selector.
 203          * We can permit a bogus selector as long as it has USER_RPL.
 204          * Null selectors are fine for other segment registers, but
 205          * we will never get back to user mode with invalid %cs or %ss
 206          * and will take the trap in iret instead.  Much code relies
 207          * on user_mode() to distinguish a user trap frame (which can
 208          * safely use invalid selectors) from a kernel trap frame.
 209          */
 210         switch (offset) {
 211         case offsetof(struct user_regs_struct, cs):
 212         case offsetof(struct user_regs_struct, ss):
 213                 if (unlikely(value == 0))
 214                         return -EIO;
 215
 216         default:
 217                 *pt_regs_access(task_pt_regs(task), offset) = value;
 218                 break;
 219
 220         case offsetof(struct user_regs_struct, gs):
 221                 if (task == current)
 222                         set_user_gs(task_pt_regs(task), value);
 223                 else
 224                         task_user_gs(task) = value;
 225         }
 226
 227         return 0;
 228 }
 229
 230 #else  /* CONFIG_X86_64 */
 231
 232 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 233
 234 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 235 {
 236         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 237         return &regs->r15 + (offset / sizeof(regs->r15));
 238 }
 239
 240 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 241 {
 242         /*
 243          * Returning the value truncates it to 16 bits.
 244          */
 245         unsigned int seg;
 246
 247         switch (offset) {
 248         case offsetof(struct user_regs_struct, fs):
 249                 if (task == current) {
 250                         /* Older gas can't assemble movq %?s,%r?? */
 251                         asm("movl %%fs,%0" : "=r" (seg));
 252                         return seg;
 253                 }
 254                 return task->thread.fsindex;
 255         case offsetof(struct user_regs_struct, gs):
 256                 if (task == current) {
 257                         asm("movl %%gs,%0" : "=r" (seg));
 258                         return seg;
 259                 }
 260                 return task->thread.gsindex;
 261         case offsetof(struct user_regs_struct, ds):
 262                 if (task == current) {
 263                         asm("movl %%ds,%0" : "=r" (seg));
 264                         return seg;
 265                 }
 266                 return task->thread.ds;
 267         case offsetof(struct user_regs_struct, es):
 268                 if (task == current) {
 269                         asm("movl %%es,%0" : "=r" (seg));
 270                         return seg;
 271                 }
 272                 return task->thread.es;
 273
 274         case offsetof(struct user_regs_struct, cs):
 275         case offsetof(struct user_regs_struct, ss):
 276                 break;
 277         }
 278         return *pt_regs_access(task_pt_regs(task), offset);
 279 }
 280
 281 static int set_segment_reg(struct task_struct *task,
 282                            unsigned long offset, u16 value)
 283 {
 284         /*
 285          * The value argument was already truncated to 16 bits.
 286          */
 287         if (invalid_selector(value))
 288                 return -EIO;
 289
 290         switch (offset) {
 291         case offsetof(struct user_regs_struct,fs):
 292                 /*
 293                  * If this is setting fs as for normal 64-bit use but
 294                  * setting fs_base has implicitly changed it, leave it.
 295                  */
 296                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 297                      task->thread.fs != 0) ||
 298                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 299                      task->thread.fs == 0))
 300                         break;
 301                 task->thread.fsindex = value;
 302                 if (task == current)
 303                         loadsegment(fs, task->thread.fsindex);
 304                 break;
 305         case offsetof(struct user_regs_struct,gs):
 306                 /*
 307                  * If this is setting gs as for normal 64-bit use but
 308                  * setting gs_base has implicitly changed it, leave it.
 309                  */
 310                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 311                      task->thread.gs != 0) ||
 312                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 313                      task->thread.gs == 0))
 314                         break;
 315                 task->thread.gsindex = value;
 316                 if (task == current)
 317                         load_gs_index(task->thread.gsindex);
 318                 break;
 319         case offsetof(struct user_regs_struct,ds):
 320                 task->thread.ds = value;
 321                 if (task == current)
 322                         loadsegment(ds, task->thread.ds);
 323                 break;
 324         case offsetof(struct user_regs_struct,es):
 325                 task->thread.es = value;
 326                 if (task == current)
 327                         loadsegment(es, task->thread.es);
 328                 break;
 329
 330                 /*
 331                  * Can't actually change these in 64-bit mode.
 332                  */
 333         case offsetof(struct user_regs_struct,cs):
 334                 if (unlikely(value == 0))
 335                         return -EIO;
 336 #ifdef CONFIG_IA32_EMULATION
 337                 if (test_tsk_thread_flag(task, TIF_IA32))
 338                         task_pt_regs(task)->cs = value;
 339 #endif
 340                 break;
 341         case offsetof(struct user_regs_struct,ss):
 342                 if (unlikely(value == 0))
 343                         return -EIO;
 344 #ifdef CONFIG_IA32_EMULATION
 345                 if (test_tsk_thread_flag(task, TIF_IA32))
 346                         task_pt_regs(task)->ss = value;
 347 #endif
 348                 break;
 349         }
 350
 351         return 0;
 352 }
 353
 354 #endif  /* CONFIG_X86_32 */
 355
 356 static unsigned long get_flags(struct task_struct *task)
 357 {
 358         unsigned long retval = task_pt_regs(task)->flags;
 359
 360         /*
 361          * If the debugger set TF, hide it from the readout.
 362          */
 363         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 364                 retval &= ~X86_EFLAGS_TF;
 365
 366         return retval;
 367 }
 368
 369 static int set_flags(struct task_struct *task, unsigned long value)
 370 {
 371         struct pt_regs *regs = task_pt_regs(task);
 372
 373         /*
 374          * If the user value contains TF, mark that
 375          * it was not "us" (the debugger) that set it.
 376          * If not, make sure it stays set if we had.
 377          */
 378         if (value & X86_EFLAGS_TF)
 379                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 380         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 381                 value |= X86_EFLAGS_TF;
 382
 383         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 384
 385         return 0;
 386 }
 387
 388 static int putreg(struct task_struct *child,
 389                   unsigned long offset, unsigned long value)
 390 {
 391         switch (offset) {
 392         case offsetof(struct user_regs_struct, cs):
 393         case offsetof(struct user_regs_struct, ds):
 394         case offsetof(struct user_regs_struct, es):
 395         case offsetof(struct user_regs_struct, fs):
 396         case offsetof(struct user_regs_struct, gs):
 397         case offsetof(struct user_regs_struct, ss):
 398                 return set_segment_reg(child, offset, value);
 399
 400         case offsetof(struct user_regs_struct, flags):
 401                 return set_flags(child, value);
 402
 403 #ifdef CONFIG_X86_64
 404         case offsetof(struct user_regs_struct,fs_base):
 405                 if (value >= TASK_SIZE_OF(child))
 406                         return -EIO;
 407                 /*
 408                  * When changing the segment base, use do_arch_prctl
 409                  * to set either thread.fs or thread.fsindex and the
 410                  * corresponding GDT slot.
 411                  */
 412                 if (child->thread.fs != value)
 413                         return do_arch_prctl(child, ARCH_SET_FS, value);
 414                 return 0;
 415         case offsetof(struct user_regs_struct,gs_base):
 416                 /*
 417                  * Exactly the same here as the %fs handling above.
 418                  */
 419                 if (value >= TASK_SIZE_OF(child))
 420                         return -EIO;
 421                 if (child->thread.gs != value)
 422                         return do_arch_prctl(child, ARCH_SET_GS, value);
 423                 return 0;
 424 #endif
 425         }
 426
 427         *pt_regs_access(task_pt_regs(child), offset) = value;
 428         return 0;
 429 }
 430
 431 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 432 {
 433         switch (offset) {
 434         case offsetof(struct user_regs_struct, cs):
 435         case offsetof(struct user_regs_struct, ds):
 436         case offsetof(struct user_regs_struct, es):
 437         case offsetof(struct user_regs_struct, fs):
 438         case offsetof(struct user_regs_struct, gs):
 439         case offsetof(struct user_regs_struct, ss):
 440                 return get_segment_reg(task, offset);
 441
 442         case offsetof(struct user_regs_struct, flags):
 443                 return get_flags(task);
 444
 445 #ifdef CONFIG_X86_64
 446         case offsetof(struct user_regs_struct, fs_base): {
 447                 /*
 448                  * do_arch_prctl may have used a GDT slot instead of
 449                  * the MSR.  To userland, it appears the same either
 450                  * way, except the %fs segment selector might not be 0.
 451                  */
 452                 unsigned int seg = task->thread.fsindex;
 453                 if (task->thread.fs != 0)
 454                         return task->thread.fs;
 455                 if (task == current)
 456                         asm("movl %%fs,%0" : "=r" (seg));
 457                 if (seg != FS_TLS_SEL)
 458                         return 0;
 459                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 460         }
 461         case offsetof(struct user_regs_struct, gs_base): {
 462                 /*
 463                  * Exactly the same here as the %fs handling above.
 464                  */
 465                 unsigned int seg = task->thread.gsindex;
 466                 if (task->thread.gs != 0)
 467                         return task->thread.gs;
 468                 if (task == current)
 469                         asm("movl %%gs,%0" : "=r" (seg));
 470                 if (seg != GS_TLS_SEL)
 471                         return 0;
 472                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 473         }
 474 #endif
 475         }
 476
 477         return *pt_regs_access(task_pt_regs(task), offset);
 478 }
 479
 480 static int genregs_get(struct task_struct *target,
 481                        const struct user_regset *regset,
 482                        unsigned int pos, unsigned int count,
 483                        void *kbuf, void __user *ubuf)
 484 {
 485         if (kbuf) {
 486                 unsigned long *k = kbuf;
 487                 while (count >= sizeof(*k)) {
 488                         *k++ = getreg(target, pos);
 489                         count -= sizeof(*k);
 490                         pos += sizeof(*k);
 491                 }
 492         } else {
 493                 unsigned long __user *u = ubuf;
 494                 while (count >= sizeof(*u)) {
 495                         if (__put_user(getreg(target, pos), u++))
 496                                 return -EFAULT;
 497                         count -= sizeof(*u);
 498                         pos += sizeof(*u);
 499                 }
 500         }
 501
 502         return 0;
 503 }
 504
 505 static int genregs_set(struct task_struct *target,
 506                        const struct user_regset *regset,
 507                        unsigned int pos, unsigned int count,
 508                        const void *kbuf, const void __user *ubuf)
 509 {
 510         int ret = 0;
 511         if (kbuf) {
 512                 const unsigned long *k = kbuf;
 513                 while (count >= sizeof(*k) && !ret) {
 514                         ret = putreg(target, pos, *k++);
 515                         count -= sizeof(*k);
 516                         pos += sizeof(*k);
 517                 }
 518         } else {
 519                 const unsigned long  __user *u = ubuf;
 520                 while (count >= sizeof(*u) && !ret) {
 521                         unsigned long word;
 522                         ret = __get_user(word, u++);
 523                         if (ret)
 524                                 break;
 525                         ret = putreg(target, pos, word);
 526                         count -= sizeof(*u);
 527                         pos += sizeof(*u);
 528                 }
 529         }
 530         return ret;
 531 }
 532
 533 static void ptrace_triggered(struct perf_event *bp,
 534                              struct perf_sample_data *data,
 535                              struct pt_regs *regs)
 536 {
 537         int i;
 538         struct thread_struct *thread = &(current->thread);
 539
 540         /*
 541          * Store in the virtual DR6 register the fact that the breakpoint
 542          * was hit so the thread's debugger will see it.
 543          */
 544         for (i = 0; i < HBP_NUM; i++) {
 545                 if (thread->ptrace_bps[i] == bp)
 546                         break;
 547         }
 548
 549         thread->debugreg6 |= (DR_TRAP0 << i);
 550 }
 551
 552 /*
 553  * Walk through every ptrace breakpoints for this thread and
 554  * build the dr7 value on top of their attributes.
 555  *
 556  */
 557 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 558 {
 559         int i;
 560         int dr7 = 0;
 561         struct arch_hw_breakpoint *info;
 562
 563         for (i = 0; i < HBP_NUM; i++) {
 564                 if (bp[i] && !bp[i]->attr.disabled) {
 565                         info = counter_arch_bp(bp[i]);
 566                         dr7 |= encode_dr7(i, info->len, info->type);
 567                 }
 568         }
 569
 570         return dr7;
 571 }
 572
 573 static int
 574 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 575                          struct task_struct *tsk, int disabled)
 576 {
 577         int err;
 578         int gen_len, gen_type;
 579         struct perf_event_attr attr;
 580
 581         /*
 582          * We should have at least an inactive breakpoint at this
 583          * slot. It means the user is writing dr7 without having
 584          * written the address register first
 585          */
 586         if (!bp)
 587                 return -EINVAL;
 588
 589         err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
 590         if (err)
 591                 return err;
 592
 593         attr = bp->attr;
 594         attr.bp_len = gen_len;
 595         attr.bp_type = gen_type;
 596         attr.disabled = disabled;
 597
 598         return modify_user_hw_breakpoint(bp, &attr);
 599 }
 600
 601 /*
 602  * Handle ptrace writes to debug register 7.
 603  */
 604 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 605 {
 606         struct thread_struct *thread = &(tsk->thread);
 607         unsigned long old_dr7;
 608         int i, orig_ret = 0, rc = 0;
 609         int enabled, second_pass = 0;
 610         unsigned len, type;
 611         struct perf_event *bp;
 612
 613         if (ptrace_get_breakpoints(tsk) < 0)
 614                 return -ESRCH;
 615
 616         data &= ~DR_CONTROL_RESERVED;
 617         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 618 restore:
 619         /*
 620          * Loop through all the hardware breakpoints, making the
 621          * appropriate changes to each.
 622          */
 623         for (i = 0; i < HBP_NUM; i++) {
 624                 enabled = decode_dr7(data, i, &len, &type);
 625                 bp = thread->ptrace_bps[i];
 626
 627                 if (!enabled) {
 628                         if (bp) {
 629                                 /*
 630                                  * Don't unregister the breakpoints right-away,
 631                                  * unless all register_user_hw_breakpoint()
 632                                  * requests have succeeded. This prevents
 633                                  * any window of opportunity for debug
 634                                  * register grabbing by other users.
 635                                  */
 636                                 if (!second_pass)
 637                                         continue;
 638
 639                                 rc = ptrace_modify_breakpoint(bp, len, type,
 640                                                               tsk, 1);
 641                                 if (rc)
 642                                         break;
 643                         }
 644                         continue;
 645                 }
 646
 647                 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
 648                 if (rc)
 649                         break;
 650         }
 651         /*
 652          * Make a second pass to free the remaining unused breakpoints
 653          * or to restore the original breakpoints if an error occurred.
 654          */
 655         if (!second_pass) {
 656                 second_pass = 1;
 657                 if (rc < 0) {
 658                         orig_ret = rc;
 659                         data = old_dr7;
 660                 }
 661                 goto restore;
 662         }
 663
 664         ptrace_put_breakpoints(tsk);
 665
 666         return ((orig_ret < 0) ? orig_ret : rc);
 667 }
 668
 669 /*
 670  * Handle PTRACE_PEEKUSR calls for the debug register area.
 671  */
 672 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 673 {
 674         struct thread_struct *thread = &(tsk->thread);
 675         unsigned long val = 0;
 676
 677         if (n < HBP_NUM) {
 678                 struct perf_event *bp;
 679
 680                 if (ptrace_get_breakpoints(tsk) < 0)
 681                         return -ESRCH;
 682
 683                 bp = thread->ptrace_bps[n];
 684                 if (!bp)
 685                         val = 0;
 686                 else
 687                         val = bp->hw.info.address;
 688
 689                 ptrace_put_breakpoints(tsk);
 690         } else if (n == 6) {
 691                 val = thread->debugreg6;
 692          } else if (n == 7) {
 693                 val = thread->ptrace_dr7;
 694         }
 695         return val;
 696 }
 697
 698 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 699                                       unsigned long addr)
 700 {
 701         struct perf_event *bp;
 702         struct thread_struct *t = &tsk->thread;
 703         struct perf_event_attr attr;
 704         int err = 0;
 705
 706         if (ptrace_get_breakpoints(tsk) < 0)
 707                 return -ESRCH;
 708
 709         if (!t->ptrace_bps[nr]) {
 710                 ptrace_breakpoint_init(&attr);
 711                 /*
 712                  * Put stub len and type to register (reserve) an inactive but
 713                  * correct bp
 714                  */
 715                 attr.bp_addr = addr;
 716                 attr.bp_len = HW_BREAKPOINT_LEN_1;
 717                 attr.bp_type = HW_BREAKPOINT_W;
 718                 attr.disabled = 1;
 719
 720                 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
 721                                                  NULL, tsk);
 722
 723                 /*
 724                  * CHECKME: the previous code returned -EIO if the addr wasn't
 725                  * a valid task virtual addr. The new one will return -EINVAL in
 726                  *  this case.
 727                  * -EINVAL may be what we want for in-kernel breakpoints users,
 728                  * but -EIO looks better for ptrace, since we refuse a register
 729                  * writing for the user. And anyway this is the previous
 730                  * behaviour.
 731                  */
 732                 if (IS_ERR(bp)) {
 733                         err = PTR_ERR(bp);
 734                         goto put;
 735                 }
 736
 737                 t->ptrace_bps[nr] = bp;
 738         } else {
 739                 bp = t->ptrace_bps[nr];
 740
 741                 attr = bp->attr;
 742                 attr.bp_addr = addr;
 743                 err = modify_user_hw_breakpoint(bp, &attr);
 744         }
 745
 746 put:
 747         ptrace_put_breakpoints(tsk);
 748         return err;
 749 }
 750
 751 /*
 752  * Handle PTRACE_POKEUSR calls for the debug register area.
 753  */
 754 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
 755                                unsigned long val)
 756 {
 757         struct thread_struct *thread = &(tsk->thread);
 758         int rc = 0;
 759
 760         /* There are no DR4 or DR5 registers */
 761         if (n == 4 || n == 5)
 762                 return -EIO;
 763
 764         if (n == 6) {
 765                 thread->debugreg6 = val;
 766                 goto ret_path;
 767         }
 768         if (n < HBP_NUM) {
 769                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 770                 if (rc)
 771                         return rc;
 772         }
 773         /* All that's left is DR7 */
 774         if (n == 7) {
 775                 rc = ptrace_write_dr7(tsk, val);
 776                 if (!rc)
 777                         thread->ptrace_dr7 = val;
 778         }
 779
 780 ret_path:
 781         return rc;
 782 }
 783
 784 /*
 785  * These access the current or another (stopped) task's io permission
 786  * bitmap for debugging or core dump.
 787  */
 788 static int ioperm_active(struct task_struct *target,
 789                          const struct user_regset *regset)
 790 {
 791         return target->thread.io_bitmap_max / regset->size;
 792 }
 793
 794 static int ioperm_get(struct task_struct *target,
 795                       const struct user_regset *regset,
 796                       unsigned int pos, unsigned int count,
 797                       void *kbuf, void __user *ubuf)
 798 {
 799         if (!target->thread.io_bitmap_ptr)
 800                 return -ENXIO;
 801
 802         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 803                                    target->thread.io_bitmap_ptr,
 804                                    0, IO_BITMAP_BYTES);
 805 }
 806
 807 /*
 808  * Called by kernel/ptrace.c when detaching..
 809  *
 810  * Make sure the single step bit is not set.
 811  */
 812 void ptrace_disable(struct task_struct *child)
 813 {
 814         user_disable_single_step(child);
 815 #ifdef TIF_SYSCALL_EMU
 816         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 817 #endif
 818 }
 819
 820 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 821 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 822 #endif
 823
 824 long arch_ptrace(struct task_struct *child, long request,
 825                  unsigned long addr, unsigned long data)
 826 {
 827         int ret;
 828         unsigned long __user *datap = (unsigned long __user *)data;
 829
 830         switch (request) {
 831         /* read the word at location addr in the USER area. */
 832         case PTRACE_PEEKUSR: {
 833                 unsigned long tmp;
 834
 835                 ret = -EIO;
 836                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 837                         break;
 838
 839                 tmp = 0;  /* Default return condition */
 840                 if (addr < sizeof(struct user_regs_struct))
 841                         tmp = getreg(child, addr);
 842                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 843                          addr <= offsetof(struct user, u_debugreg[7])) {
 844                         addr -= offsetof(struct user, u_debugreg[0]);
 845                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 846                 }
 847                 ret = put_user(tmp, datap);
 848                 break;
 849         }
 850
 851         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 852                 ret = -EIO;
 853                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 854                         break;
 855
 856                 if (addr < sizeof(struct user_regs_struct))
 857                         ret = putreg(child, addr, data);
 858                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 859                          addr <= offsetof(struct user, u_debugreg[7])) {
 860                         addr -= offsetof(struct user, u_debugreg[0]);
 861                         ret = ptrace_set_debugreg(child,
 862                                                   addr / sizeof(data), data);
 863                 }
 864                 break;
 865
 866         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 867                 return copy_regset_to_user(child,
 868                                            task_user_regset_view(current),
 869                                            REGSET_GENERAL,
 870                                            0, sizeof(struct user_regs_struct),
 871                                            datap);
 872
 873         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 874                 return copy_regset_from_user(child,
 875                                              task_user_regset_view(current),
 876                                              REGSET_GENERAL,
 877                                              0, sizeof(struct user_regs_struct),
 878                                              datap);
 879
 880         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 881                 return copy_regset_to_user(child,
 882                                            task_user_regset_view(current),
 883                                            REGSET_FP,
 884                                            0, sizeof(struct user_i387_struct),
 885                                            datap);
 886
 887         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 888                 return copy_regset_from_user(child,
 889                                              task_user_regset_view(current),
 890                                              REGSET_FP,
 891                                              0, sizeof(struct user_i387_struct),
 892                                              datap);
 893
 894 #ifdef CONFIG_X86_32
 895         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 896                 return copy_regset_to_user(child, &user_x86_32_view,
 897                                            REGSET_XFP,
 898                                            0, sizeof(struct user_fxsr_struct),
 899                                            datap) ? -EIO : 0;
 900
 901         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 902                 return copy_regset_from_user(child, &user_x86_32_view,
 903                                              REGSET_XFP,
 904                                              0, sizeof(struct user_fxsr_struct),
 905                                              datap) ? -EIO : 0;
 906 #endif
 907
 908 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 909         case PTRACE_GET_THREAD_AREA:
 910                 if ((int) addr < 0)
 911                         return -EIO;
 912                 ret = do_get_thread_area(child, addr,
 913                                         (struct user_desc __user *)data);
 914                 break;
 915
 916         case PTRACE_SET_THREAD_AREA:
 917                 if ((int) addr < 0)
 918                         return -EIO;
 919                 ret = do_set_thread_area(child, addr,
 920                                         (struct user_desc __user *)data, 0);
 921                 break;
 922 #endif
 923
 924 #ifdef CONFIG_X86_64
 925                 /* normal 64bit interface to access TLS data.
 926                    Works just like arch_prctl, except that the arguments
 927                    are reversed. */
 928         case PTRACE_ARCH_PRCTL:
 929                 ret = do_arch_prctl(child, data, addr);
 930                 break;
 931 #endif
 932
 933         default:
 934                 ret = ptrace_request(child, request, addr, data);
 935                 break;
 936         }
 937
 938         return ret;
 939 }
 940
 941 #ifdef CONFIG_IA32_EMULATION
 942
 943 #include <linux/compat.h>
 944 #include <linux/syscalls.h>
 945 #include <asm/ia32.h>
 946 #include <asm/user32.h>
 947
 948 #define R32(l,q)                                                        \
 949         case offsetof(struct user32, regs.l):                           \
 950                 regs->q = value; break
 951
 952 #define SEG32(rs)                                                       \
 953         case offsetof(struct user32, regs.rs):                          \
 954                 return set_segment_reg(child,                           \
 955                                        offsetof(struct user_regs_struct, rs), \
 956                                        value);                          \
 957                 break
 958
 959 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 960 {
 961         struct pt_regs *regs = task_pt_regs(child);
 962
 963         switch (regno) {
 964
 965         SEG32(cs);
 966         SEG32(ds);
 967         SEG32(es);
 968         SEG32(fs);
 969         SEG32(gs);
 970         SEG32(ss);
 971
 972         R32(ebx, bx);
 973         R32(ecx, cx);
 974         R32(edx, dx);
 975         R32(edi, di);
 976         R32(esi, si);
 977         R32(ebp, bp);
 978         R32(eax, ax);
 979         R32(eip, ip);
 980         R32(esp, sp);
 981
 982         case offsetof(struct user32, regs.orig_eax):
 983                 /*
 984                  * A 32-bit debugger setting orig_eax means to restore
 985                  * the state of the task restarting a 32-bit syscall.
 986                  * Make sure we interpret the -ERESTART* codes correctly
 987                  * in case the task is not actually still sitting at the
 988                  * exit from a 32-bit syscall with TS_COMPAT still set.
 989                  */
 990                 regs->orig_ax = value;
 991                 if (syscall_get_nr(child, regs) >= 0)
 992                         task_thread_info(child)->status |= TS_COMPAT;
 993                 break;
 994
 995         case offsetof(struct user32, regs.eflags):
 996                 return set_flags(child, value);
 997
 998         case offsetof(struct user32, u_debugreg[0]) ...
 999                 offsetof(struct user32, u_debugreg[7]):
1000                 regno -= offsetof(struct user32, u_debugreg[0]);
1001                 return ptrace_set_debugreg(child, regno / 4, value);
1002
1003         default:
1004                 if (regno > sizeof(struct user32) || (regno & 3))
1005                         return -EIO;
1006
1007                 /*
1008                  * Other dummy fields in the virtual user structure
1009                  * are ignored
1010                  */
1011                 break;
1012         }
1013         return 0;
1014 }
1015
1016 #undef R32
1017 #undef SEG32
1018
1019 #define R32(l,q)                                                        \
1020         case offsetof(struct user32, regs.l):                           \
1021                 *val = regs->q; break
1022
1023 #define SEG32(rs)                                                       \
1024         case offsetof(struct user32, regs.rs):                          \
1025                 *val = get_segment_reg(child,                           \
1026                                        offsetof(struct user_regs_struct, rs)); \
1027                 break
1028
1029 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1030 {
1031         struct pt_regs *regs = task_pt_regs(child);
1032
1033         switch (regno) {
1034
1035         SEG32(ds);
1036         SEG32(es);
1037         SEG32(fs);
1038         SEG32(gs);
1039
1040         R32(cs, cs);
1041         R32(ss, ss);
1042         R32(ebx, bx);
1043         R32(ecx, cx);
1044         R32(edx, dx);
1045         R32(edi, di);
1046         R32(esi, si);
1047         R32(ebp, bp);
1048         R32(eax, ax);
1049         R32(orig_eax, orig_ax);
1050         R32(eip, ip);
1051         R32(esp, sp);
1052
1053         case offsetof(struct user32, regs.eflags):
1054                 *val = get_flags(child);
1055                 break;
1056
1057         case offsetof(struct user32, u_debugreg[0]) ...
1058                 offsetof(struct user32, u_debugreg[7]):
1059                 regno -= offsetof(struct user32, u_debugreg[0]);
1060                 *val = ptrace_get_debugreg(child, regno / 4);
1061                 break;
1062
1063         default:
1064                 if (regno > sizeof(struct user32) || (regno & 3))
1065                         return -EIO;
1066
1067                 /*
1068                  * Other dummy fields in the virtual user structure
1069                  * are ignored
1070                  */
1071                 *val = 0;
1072                 break;
1073         }
1074         return 0;
1075 }
1076
1077 #undef R32
1078 #undef SEG32
1079
1080 static int genregs32_get(struct task_struct *target,
1081                          const struct user_regset *regset,
1082                          unsigned int pos, unsigned int count,
1083                          void *kbuf, void __user *ubuf)
1084 {
1085         if (kbuf) {
1086                 compat_ulong_t *k = kbuf;
1087                 while (count >= sizeof(*k)) {
1088                         getreg32(target, pos, k++);
1089                         count -= sizeof(*k);
1090                         pos += sizeof(*k);
1091                 }
1092         } else {
1093                 compat_ulong_t __user *u = ubuf;
1094                 while (count >= sizeof(*u)) {
1095                         compat_ulong_t word;
1096                         getreg32(target, pos, &word);
1097                         if (__put_user(word, u++))
1098                                 return -EFAULT;
1099                         count -= sizeof(*u);
1100                         pos += sizeof(*u);
1101                 }
1102         }
1103
1104         return 0;
1105 }
1106
1107 static int genregs32_set(struct task_struct *target,
1108                          const struct user_regset *regset,
1109                          unsigned int pos, unsigned int count,
1110                          const void *kbuf, const void __user *ubuf)
1111 {
1112         int ret = 0;
1113         if (kbuf) {
1114                 const compat_ulong_t *k = kbuf;
1115                 while (count >= sizeof(*k) && !ret) {
1116                         ret = putreg32(target, pos, *k++);
1117                         count -= sizeof(*k);
1118                         pos += sizeof(*k);
1119                 }
1120         } else {
1121                 const compat_ulong_t __user *u = ubuf;
1122                 while (count >= sizeof(*u) && !ret) {
1123                         compat_ulong_t word;
1124                         ret = __get_user(word, u++);
1125                         if (ret)
1126                                 break;
1127                         ret = putreg32(target, pos, word);
1128                         count -= sizeof(*u);
1129                         pos += sizeof(*u);
1130                 }
1131         }
1132         return ret;
1133 }
1134
1135 #ifdef CONFIG_X86_X32_ABI
1136 static long x32_arch_ptrace(struct task_struct *child,
1137                             compat_long_t request, compat_ulong_t caddr,
1138                             compat_ulong_t cdata)
1139 {
1140         unsigned long addr = caddr;
1141         unsigned long data = cdata;
1142         void __user *datap = compat_ptr(data);
1143         int ret;
1144
1145         switch (request) {
1146         /* Read 32bits at location addr in the USER area.  Only allow
1147            to return the lower 32bits of segment and debug registers.  */
1148         case PTRACE_PEEKUSR: {
1149                 u32 tmp;
1150
1151                 ret = -EIO;
1152                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1153                     addr < offsetof(struct user_regs_struct, cs))
1154                         break;
1155
1156                 tmp = 0;  /* Default return condition */
1157                 if (addr < sizeof(struct user_regs_struct))
1158                         tmp = getreg(child, addr);
1159                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1160                          addr <= offsetof(struct user, u_debugreg[7])) {
1161                         addr -= offsetof(struct user, u_debugreg[0]);
1162                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1163                 }
1164                 ret = put_user(tmp, (__u32 __user *)datap);
1165                 break;
1166         }
1167
1168         /* Write the word at location addr in the USER area.  Only allow
1169            to update segment and debug registers with the upper 32bits
1170            zero-extended. */
1171         case PTRACE_POKEUSR:
1172                 ret = -EIO;
1173                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1174                     addr < offsetof(struct user_regs_struct, cs))
1175                         break;
1176
1177                 if (addr < sizeof(struct user_regs_struct))
1178                         ret = putreg(child, addr, data);
1179                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1180                          addr <= offsetof(struct user, u_debugreg[7])) {
1181                         addr -= offsetof(struct user, u_debugreg[0]);
1182                         ret = ptrace_set_debugreg(child,
1183                                                   addr / sizeof(data), data);
1184                 }
1185                 break;
1186
1187         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1188                 return copy_regset_to_user(child,
1189                                            task_user_regset_view(current),
1190                                            REGSET_GENERAL,
1191                                            0, sizeof(struct user_regs_struct),
1192                                            datap);
1193
1194         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1195                 return copy_regset_from_user(child,
1196                                              task_user_regset_view(current),
1197                                              REGSET_GENERAL,
1198                                              0, sizeof(struct user_regs_struct),
1199                                              datap);
1200
1201         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1202                 return copy_regset_to_user(child,
1203                                            task_user_regset_view(current),
1204                                            REGSET_FP,
1205                                            0, sizeof(struct user_i387_struct),
1206                                            datap);
1207
1208         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1209                 return copy_regset_from_user(child,
1210                                              task_user_regset_view(current),
1211                                              REGSET_FP,
1212                                              0, sizeof(struct user_i387_struct),
1213                                              datap);
1214
1215         default:
1216                 return compat_ptrace_request(child, request, addr, data);
1217         }
1218
1219         return ret;
1220 }
1221 #endif
1222
1223 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1224                         compat_ulong_t caddr, compat_ulong_t cdata)
1225 {
1226         unsigned long addr = caddr;
1227         unsigned long data = cdata;
1228         void __user *datap = compat_ptr(data);
1229         int ret;
1230         __u32 val;
1231
1232 #ifdef CONFIG_X86_X32_ABI
1233         if (!is_ia32_task())
1234                 return x32_arch_ptrace(child, request, caddr, cdata);
1235 #endif
1236
1237         switch (request) {
1238         case PTRACE_PEEKUSR:
1239                 ret = getreg32(child, addr, &val);
1240                 if (ret == 0)
1241                         ret = put_user(val, (__u32 __user *)datap);
1242                 break;
1243
1244         case PTRACE_POKEUSR:
1245                 ret = putreg32(child, addr, data);
1246                 break;
1247
1248         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1249                 return copy_regset_to_user(child, &user_x86_32_view,
1250                                            REGSET_GENERAL,
1251                                            0, sizeof(struct user_regs_struct32),
1252                                            datap);
1253
1254         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1255                 return copy_regset_from_user(child, &user_x86_32_view,
1256                                              REGSET_GENERAL, 0,
1257                                              sizeof(struct user_regs_struct32),
1258                                              datap);
1259
1260         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1261                 return copy_regset_to_user(child, &user_x86_32_view,
1262                                            REGSET_FP, 0,
1263                                            sizeof(struct user_i387_ia32_struct),
1264                                            datap);
1265
1266         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1267                 return copy_regset_from_user(
1268                         child, &user_x86_32_view, REGSET_FP,
1269                         0, sizeof(struct user_i387_ia32_struct), datap);
1270
1271         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1272                 return copy_regset_to_user(child, &user_x86_32_view,
1273                                            REGSET_XFP, 0,
1274                                            sizeof(struct user32_fxsr_struct),
1275                                            datap);
1276
1277         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1278                 return copy_regset_from_user(child, &user_x86_32_view,
1279                                              REGSET_XFP, 0,
1280                                              sizeof(struct user32_fxsr_struct),
1281                                              datap);
1282
1283         case PTRACE_GET_THREAD_AREA:
1284         case PTRACE_SET_THREAD_AREA:
1285                 return arch_ptrace(child, request, addr, data);
1286
1287         default:
1288                 return compat_ptrace_request(child, request, addr, data);
1289         }
1290
1291         return ret;
1292 }
1293
1294 #endif  /* CONFIG_IA32_EMULATION */
1295
1296 #ifdef CONFIG_X86_64
1297
1298 static struct user_regset x86_64_regsets[] __read_mostly = {
1299         [REGSET_GENERAL] = {
1300                 .core_note_type = NT_PRSTATUS,
1301                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1302                 .size = sizeof(long), .align = sizeof(long),
1303                 .get = genregs_get, .set = genregs_set
1304         },
1305         [REGSET_FP] = {
1306                 .core_note_type = NT_PRFPREG,
1307                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1308                 .size = sizeof(long), .align = sizeof(long),
1309                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1310         },
1311         [REGSET_XSTATE] = {
1312                 .core_note_type = NT_X86_XSTATE,
1313                 .size = sizeof(u64), .align = sizeof(u64),
1314                 .active = xstateregs_active, .get = xstateregs_get,
1315                 .set = xstateregs_set
1316         },
1317         [REGSET_IOPERM64] = {
1318                 .core_note_type = NT_386_IOPERM,
1319                 .n = IO_BITMAP_LONGS,
1320                 .size = sizeof(long), .align = sizeof(long),
1321                 .active = ioperm_active, .get = ioperm_get
1322         },
1323 };
1324
1325 static const struct user_regset_view user_x86_64_view = {
1326         .name = "x86_64", .e_machine = EM_X86_64,
1327         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1328 };
1329
1330 #else  /* CONFIG_X86_32 */
1331
1332 #define user_regs_struct32      user_regs_struct
1333 #define genregs32_get           genregs_get
1334 #define genregs32_set           genregs_set
1335
1336 #endif  /* CONFIG_X86_64 */
1337
1338 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1339 static struct user_regset x86_32_regsets[] __read_mostly = {
1340         [REGSET_GENERAL] = {
1341                 .core_note_type = NT_PRSTATUS,
1342                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1343                 .size = sizeof(u32), .align = sizeof(u32),
1344                 .get = genregs32_get, .set = genregs32_set
1345         },
1346         [REGSET_FP] = {
1347                 .core_note_type = NT_PRFPREG,
1348                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1349                 .size = sizeof(u32), .align = sizeof(u32),
1350                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1351         },
1352         [REGSET_XFP] = {
1353                 .core_note_type = NT_PRXFPREG,
1354                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1355                 .size = sizeof(u32), .align = sizeof(u32),
1356                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1357         },
1358         [REGSET_XSTATE] = {
1359                 .core_note_type = NT_X86_XSTATE,
1360                 .size = sizeof(u64), .align = sizeof(u64),
1361                 .active = xstateregs_active, .get = xstateregs_get,
1362                 .set = xstateregs_set
1363         },
1364         [REGSET_TLS] = {
1365                 .core_note_type = NT_386_TLS,
1366                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1367                 .size = sizeof(struct user_desc),
1368                 .align = sizeof(struct user_desc),
1369                 .active = regset_tls_active,
1370                 .get = regset_tls_get, .set = regset_tls_set
1371         },
1372         [REGSET_IOPERM32] = {
1373                 .core_note_type = NT_386_IOPERM,
1374                 .n = IO_BITMAP_BYTES / sizeof(u32),
1375                 .size = sizeof(u32), .align = sizeof(u32),
1376                 .active = ioperm_active, .get = ioperm_get
1377         },
1378 };
1379
1380 static const struct user_regset_view user_x86_32_view = {
1381         .name = "i386", .e_machine = EM_386,
1382         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1383 };
1384 #endif
1385
1386 /*
1387  * This represents bytes 464..511 in the memory layout exported through
1388  * the REGSET_XSTATE interface.
1389  */
1390 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1391
1392 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1393 {
1394 #ifdef CONFIG_X86_64
1395         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1396 #endif
1397 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1398         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1399 #endif
1400         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1401 }
1402
1403 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1404 {
1405 #ifdef CONFIG_IA32_EMULATION
1406         if (test_tsk_thread_flag(task, TIF_IA32))
1407 #endif
1408 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1409                 return &user_x86_32_view;
1410 #endif
1411 #ifdef CONFIG_X86_64
1412         return &user_x86_64_view;
1413 #endif
1414 }
1415
1416 static void fill_sigtrap_info(struct task_struct *tsk,
1417                                 struct pt_regs *regs,
1418                                 int error_code, int si_code,
1419                                 struct siginfo *info)
1420 {
1421         tsk->thread.trap_nr = X86_TRAP_DB;
1422         tsk->thread.error_code = error_code;
1423
1424         memset(info, 0, sizeof(*info));
1425         info->si_signo = SIGTRAP;
1426         info->si_code = si_code;
1427         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1428 }
1429
1430 void user_single_step_siginfo(struct task_struct *tsk,
1431                                 struct pt_regs *regs,
1432                                 struct siginfo *info)
1433 {
1434         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1435 }
1436
1437 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1438                                          int error_code, int si_code)
1439 {
1440         struct siginfo info;
1441
1442         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1443         /* Send us the fake SIGTRAP */
1444         force_sig_info(SIGTRAP, &info, tsk);
1445 }
1446
1447
1448 #ifdef CONFIG_X86_32
1449 # define IS_IA32        1
1450 #elif defined CONFIG_IA32_EMULATION
1451 # define IS_IA32        is_compat_task()
1452 #else
1453 # define IS_IA32        0
1454 #endif
1455
1456 /*
1457  * We must return the syscall number to actually look up in the table.
1458  * This can be -1L to skip running any syscall at all.
1459  */
1460 long syscall_trace_enter(struct pt_regs *regs)
1461 {
1462         long ret = 0;
1463
1464         rcu_user_exit();
1465
1466         /*
1467          * If we stepped into a sysenter/syscall insn, it trapped in
1468          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1469          * If user-mode had set TF itself, then it's still clear from
1470          * do_debug() and we need to set it again to restore the user
1471          * state.  If we entered on the slow path, TF was already set.
1472          */
1473         if (test_thread_flag(TIF_SINGLESTEP))
1474                 regs->flags |= X86_EFLAGS_TF;
1475
1476         /* do the secure computing check first */
1477         if (secure_computing(regs->orig_ax)) {
1478                 /* seccomp failures shouldn't expose any additional code. */
1479                 ret = -1L;
1480                 goto out;
1481         }
1482
1483         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1484                 ret = -1L;
1485
1486         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1487             tracehook_report_syscall_entry(regs))
1488                 ret = -1L;
1489
1490         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1491                 trace_sys_enter(regs, regs->orig_ax);
1492
1493         if (IS_IA32)
1494                 audit_syscall_entry(AUDIT_ARCH_I386,
1495                                     regs->orig_ax,
1496                                     regs->bx, regs->cx,
1497                                     regs->dx, regs->si);
1498 #ifdef CONFIG_X86_64
1499         else
1500                 audit_syscall_entry(AUDIT_ARCH_X86_64,
1501                                     regs->orig_ax,
1502                                     regs->di, regs->si,
1503                                     regs->dx, regs->r10);
1504 #endif
1505
1506 out:
1507         return ret ?: regs->orig_ax;
1508 }
1509
1510 void syscall_trace_leave(struct pt_regs *regs)
1511 {
1512         bool step;
1513
1514         audit_syscall_exit(regs);
1515
1516         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1517                 trace_sys_exit(regs, regs->ax);
1518
1519         /*
1520          * If TIF_SYSCALL_EMU is set, we only get here because of
1521          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1522          * We already reported this syscall instruction in
1523          * syscall_trace_enter().
1524          */
1525         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1526                         !test_thread_flag(TIF_SYSCALL_EMU);
1527         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1528                 tracehook_report_syscall_exit(regs, step);
1529
1530         rcu_user_enter();
1531 }