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 #include <linux/export.h>
  26 #include <linux/context_tracking.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/pgtable.h>
  30 #include <asm/processor.h>
  31 #include <asm/fpu/internal.h>
  32 #include <asm/debugreg.h>
  33 #include <asm/ldt.h>
  34 #include <asm/desc.h>
  35 #include <asm/prctl.h>
  36 #include <asm/proto.h>
  37 #include <asm/hw_breakpoint.h>
  38 #include <asm/traps.h>
  39
  40 #include "tls.h"
  41
  42 #define CREATE_TRACE_POINTS
  43 #include <trace/events/syscalls.h>
  44
  45 enum x86_regset {
  46         REGSET_GENERAL,
  47         REGSET_FP,
  48         REGSET_XFP,
  49         REGSET_IOPERM64 = REGSET_XFP,
  50         REGSET_XSTATE,
  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 /*
 171  * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
 172  * when it traps.  The previous stack will be directly underneath the saved
 173  * registers, and 'sp/ss' won't even have been saved. Thus the '&regs->sp'.
 174  *
 175  * Now, if the stack is empty, '&regs->sp' is out of range. In this
 176  * case we try to take the previous stack. To always return a non-null
 177  * stack pointer we fall back to regs as stack if no previous stack
 178  * exists.
 179  *
 180  * This is valid only for kernel mode traps.
 181  */
 182 unsigned long kernel_stack_pointer(struct pt_regs *regs)
 183 {
 184         unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
 185         unsigned long sp = (unsigned long)&regs->sp;
 186         u32 *prev_esp;
 187
 188         if (context == (sp & ~(THREAD_SIZE - 1)))
 189                 return sp;
 190
 191         prev_esp = (u32 *)(context);
 192         if (prev_esp)
 193                 return (unsigned long)prev_esp;
 194
 195         return (unsigned long)regs;
 196 }
 197 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
 198
 199 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 200 {
 201         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 202         return &regs->bx + (regno >> 2);
 203 }
 204
 205 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 206 {
 207         /*
 208          * Returning the value truncates it to 16 bits.
 209          */
 210         unsigned int retval;
 211         if (offset != offsetof(struct user_regs_struct, gs))
 212                 retval = *pt_regs_access(task_pt_regs(task), offset);
 213         else {
 214                 if (task == current)
 215                         retval = get_user_gs(task_pt_regs(task));
 216                 else
 217                         retval = task_user_gs(task);
 218         }
 219         return retval;
 220 }
 221
 222 static int set_segment_reg(struct task_struct *task,
 223                            unsigned long offset, u16 value)
 224 {
 225         /*
 226          * The value argument was already truncated to 16 bits.
 227          */
 228         if (invalid_selector(value))
 229                 return -EIO;
 230
 231         /*
 232          * For %cs and %ss we cannot permit a null selector.
 233          * We can permit a bogus selector as long as it has USER_RPL.
 234          * Null selectors are fine for other segment registers, but
 235          * we will never get back to user mode with invalid %cs or %ss
 236          * and will take the trap in iret instead.  Much code relies
 237          * on user_mode() to distinguish a user trap frame (which can
 238          * safely use invalid selectors) from a kernel trap frame.
 239          */
 240         switch (offset) {
 241         case offsetof(struct user_regs_struct, cs):
 242         case offsetof(struct user_regs_struct, ss):
 243                 if (unlikely(value == 0))
 244                         return -EIO;
 245
 246         default:
 247                 *pt_regs_access(task_pt_regs(task), offset) = value;
 248                 break;
 249
 250         case offsetof(struct user_regs_struct, gs):
 251                 if (task == current)
 252                         set_user_gs(task_pt_regs(task), value);
 253                 else
 254                         task_user_gs(task) = value;
 255         }
 256
 257         return 0;
 258 }
 259
 260 #else  /* CONFIG_X86_64 */
 261
 262 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 263
 264 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 265 {
 266         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 267         return &regs->r15 + (offset / sizeof(regs->r15));
 268 }
 269
 270 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 271 {
 272         /*
 273          * Returning the value truncates it to 16 bits.
 274          */
 275         unsigned int seg;
 276
 277         switch (offset) {
 278         case offsetof(struct user_regs_struct, fs):
 279                 if (task == current) {
 280                         /* Older gas can't assemble movq %?s,%r?? */
 281                         asm("movl %%fs,%0" : "=r" (seg));
 282                         return seg;
 283                 }
 284                 return task->thread.fsindex;
 285         case offsetof(struct user_regs_struct, gs):
 286                 if (task == current) {
 287                         asm("movl %%gs,%0" : "=r" (seg));
 288                         return seg;
 289                 }
 290                 return task->thread.gsindex;
 291         case offsetof(struct user_regs_struct, ds):
 292                 if (task == current) {
 293                         asm("movl %%ds,%0" : "=r" (seg));
 294                         return seg;
 295                 }
 296                 return task->thread.ds;
 297         case offsetof(struct user_regs_struct, es):
 298                 if (task == current) {
 299                         asm("movl %%es,%0" : "=r" (seg));
 300                         return seg;
 301                 }
 302                 return task->thread.es;
 303
 304         case offsetof(struct user_regs_struct, cs):
 305         case offsetof(struct user_regs_struct, ss):
 306                 break;
 307         }
 308         return *pt_regs_access(task_pt_regs(task), offset);
 309 }
 310
 311 static int set_segment_reg(struct task_struct *task,
 312                            unsigned long offset, u16 value)
 313 {
 314         /*
 315          * The value argument was already truncated to 16 bits.
 316          */
 317         if (invalid_selector(value))
 318                 return -EIO;
 319
 320         switch (offset) {
 321         case offsetof(struct user_regs_struct,fs):
 322                 /*
 323                  * If this is setting fs as for normal 64-bit use but
 324                  * setting fs_base has implicitly changed it, leave it.
 325                  */
 326                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 327                      task->thread.fs != 0) ||
 328                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 329                      task->thread.fs == 0))
 330                         break;
 331                 task->thread.fsindex = value;
 332                 if (task == current)
 333                         loadsegment(fs, task->thread.fsindex);
 334                 break;
 335         case offsetof(struct user_regs_struct,gs):
 336                 /*
 337                  * If this is setting gs as for normal 64-bit use but
 338                  * setting gs_base has implicitly changed it, leave it.
 339                  */
 340                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 341                      task->thread.gs != 0) ||
 342                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 343                      task->thread.gs == 0))
 344                         break;
 345                 task->thread.gsindex = value;
 346                 if (task == current)
 347                         load_gs_index(task->thread.gsindex);
 348                 break;
 349         case offsetof(struct user_regs_struct,ds):
 350                 task->thread.ds = value;
 351                 if (task == current)
 352                         loadsegment(ds, task->thread.ds);
 353                 break;
 354         case offsetof(struct user_regs_struct,es):
 355                 task->thread.es = value;
 356                 if (task == current)
 357                         loadsegment(es, task->thread.es);
 358                 break;
 359
 360                 /*
 361                  * Can't actually change these in 64-bit mode.
 362                  */
 363         case offsetof(struct user_regs_struct,cs):
 364                 if (unlikely(value == 0))
 365                         return -EIO;
 366                 task_pt_regs(task)->cs = value;
 367                 break;
 368         case offsetof(struct user_regs_struct,ss):
 369                 if (unlikely(value == 0))
 370                         return -EIO;
 371                 task_pt_regs(task)->ss = value;
 372                 break;
 373         }
 374
 375         return 0;
 376 }
 377
 378 #endif  /* CONFIG_X86_32 */
 379
 380 static unsigned long get_flags(struct task_struct *task)
 381 {
 382         unsigned long retval = task_pt_regs(task)->flags;
 383
 384         /*
 385          * If the debugger set TF, hide it from the readout.
 386          */
 387         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 388                 retval &= ~X86_EFLAGS_TF;
 389
 390         return retval;
 391 }
 392
 393 static int set_flags(struct task_struct *task, unsigned long value)
 394 {
 395         struct pt_regs *regs = task_pt_regs(task);
 396
 397         /*
 398          * If the user value contains TF, mark that
 399          * it was not "us" (the debugger) that set it.
 400          * If not, make sure it stays set if we had.
 401          */
 402         if (value & X86_EFLAGS_TF)
 403                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 404         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 405                 value |= X86_EFLAGS_TF;
 406
 407         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 408
 409         return 0;
 410 }
 411
 412 static int putreg(struct task_struct *child,
 413                   unsigned long offset, unsigned long value)
 414 {
 415         switch (offset) {
 416         case offsetof(struct user_regs_struct, cs):
 417         case offsetof(struct user_regs_struct, ds):
 418         case offsetof(struct user_regs_struct, es):
 419         case offsetof(struct user_regs_struct, fs):
 420         case offsetof(struct user_regs_struct, gs):
 421         case offsetof(struct user_regs_struct, ss):
 422                 return set_segment_reg(child, offset, value);
 423
 424         case offsetof(struct user_regs_struct, flags):
 425                 return set_flags(child, value);
 426
 427 #ifdef CONFIG_X86_64
 428         case offsetof(struct user_regs_struct,fs_base):
 429                 if (value >= TASK_SIZE_OF(child))
 430                         return -EIO;
 431                 /*
 432                  * When changing the segment base, use do_arch_prctl
 433                  * to set either thread.fs or thread.fsindex and the
 434                  * corresponding GDT slot.
 435                  */
 436                 if (child->thread.fs != value)
 437                         return do_arch_prctl(child, ARCH_SET_FS, value);
 438                 return 0;
 439         case offsetof(struct user_regs_struct,gs_base):
 440                 /*
 441                  * Exactly the same here as the %fs handling above.
 442                  */
 443                 if (value >= TASK_SIZE_OF(child))
 444                         return -EIO;
 445                 if (child->thread.gs != value)
 446                         return do_arch_prctl(child, ARCH_SET_GS, value);
 447                 return 0;
 448 #endif
 449         }
 450
 451         *pt_regs_access(task_pt_regs(child), offset) = value;
 452         return 0;
 453 }
 454
 455 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 456 {
 457         switch (offset) {
 458         case offsetof(struct user_regs_struct, cs):
 459         case offsetof(struct user_regs_struct, ds):
 460         case offsetof(struct user_regs_struct, es):
 461         case offsetof(struct user_regs_struct, fs):
 462         case offsetof(struct user_regs_struct, gs):
 463         case offsetof(struct user_regs_struct, ss):
 464                 return get_segment_reg(task, offset);
 465
 466         case offsetof(struct user_regs_struct, flags):
 467                 return get_flags(task);
 468
 469 #ifdef CONFIG_X86_64
 470         case offsetof(struct user_regs_struct, fs_base): {
 471                 /*
 472                  * do_arch_prctl may have used a GDT slot instead of
 473                  * the MSR.  To userland, it appears the same either
 474                  * way, except the %fs segment selector might not be 0.
 475                  */
 476                 unsigned int seg = task->thread.fsindex;
 477                 if (task->thread.fs != 0)
 478                         return task->thread.fs;
 479                 if (task == current)
 480                         asm("movl %%fs,%0" : "=r" (seg));
 481                 if (seg != FS_TLS_SEL)
 482                         return 0;
 483                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 484         }
 485         case offsetof(struct user_regs_struct, gs_base): {
 486                 /*
 487                  * Exactly the same here as the %fs handling above.
 488                  */
 489                 unsigned int seg = task->thread.gsindex;
 490                 if (task->thread.gs != 0)
 491                         return task->thread.gs;
 492                 if (task == current)
 493                         asm("movl %%gs,%0" : "=r" (seg));
 494                 if (seg != GS_TLS_SEL)
 495                         return 0;
 496                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 497         }
 498 #endif
 499         }
 500
 501         return *pt_regs_access(task_pt_regs(task), offset);
 502 }
 503
 504 static int genregs_get(struct task_struct *target,
 505                        const struct user_regset *regset,
 506                        unsigned int pos, unsigned int count,
 507                        void *kbuf, void __user *ubuf)
 508 {
 509         if (kbuf) {
 510                 unsigned long *k = kbuf;
 511                 while (count >= sizeof(*k)) {
 512                         *k++ = getreg(target, pos);
 513                         count -= sizeof(*k);
 514                         pos += sizeof(*k);
 515                 }
 516         } else {
 517                 unsigned long __user *u = ubuf;
 518                 while (count >= sizeof(*u)) {
 519                         if (__put_user(getreg(target, pos), u++))
 520                                 return -EFAULT;
 521                         count -= sizeof(*u);
 522                         pos += sizeof(*u);
 523                 }
 524         }
 525
 526         return 0;
 527 }
 528
 529 static int genregs_set(struct task_struct *target,
 530                        const struct user_regset *regset,
 531                        unsigned int pos, unsigned int count,
 532                        const void *kbuf, const void __user *ubuf)
 533 {
 534         int ret = 0;
 535         if (kbuf) {
 536                 const unsigned long *k = kbuf;
 537                 while (count >= sizeof(*k) && !ret) {
 538                         ret = putreg(target, pos, *k++);
 539                         count -= sizeof(*k);
 540                         pos += sizeof(*k);
 541                 }
 542         } else {
 543                 const unsigned long  __user *u = ubuf;
 544                 while (count >= sizeof(*u) && !ret) {
 545                         unsigned long word;
 546                         ret = __get_user(word, u++);
 547                         if (ret)
 548                                 break;
 549                         ret = putreg(target, pos, word);
 550                         count -= sizeof(*u);
 551                         pos += sizeof(*u);
 552                 }
 553         }
 554         return ret;
 555 }
 556
 557 static void ptrace_triggered(struct perf_event *bp,
 558                              struct perf_sample_data *data,
 559                              struct pt_regs *regs)
 560 {
 561         int i;
 562         struct thread_struct *thread = &(current->thread);
 563
 564         /*
 565          * Store in the virtual DR6 register the fact that the breakpoint
 566          * was hit so the thread's debugger will see it.
 567          */
 568         for (i = 0; i < HBP_NUM; i++) {
 569                 if (thread->ptrace_bps[i] == bp)
 570                         break;
 571         }
 572
 573         thread->debugreg6 |= (DR_TRAP0 << i);
 574 }
 575
 576 /*
 577  * Walk through every ptrace breakpoints for this thread and
 578  * build the dr7 value on top of their attributes.
 579  *
 580  */
 581 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 582 {
 583         int i;
 584         int dr7 = 0;
 585         struct arch_hw_breakpoint *info;
 586
 587         for (i = 0; i < HBP_NUM; i++) {
 588                 if (bp[i] && !bp[i]->attr.disabled) {
 589                         info = counter_arch_bp(bp[i]);
 590                         dr7 |= encode_dr7(i, info->len, info->type);
 591                 }
 592         }
 593
 594         return dr7;
 595 }
 596
 597 static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
 598                                         int len, int type, bool disabled)
 599 {
 600         int err, bp_len, bp_type;
 601
 602         err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
 603         if (!err) {
 604                 attr->bp_len = bp_len;
 605                 attr->bp_type = bp_type;
 606                 attr->disabled = disabled;
 607         }
 608
 609         return err;
 610 }
 611
 612 static struct perf_event *
 613 ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
 614                                 unsigned long addr, bool disabled)
 615 {
 616         struct perf_event_attr attr;
 617         int err;
 618
 619         ptrace_breakpoint_init(&attr);
 620         attr.bp_addr = addr;
 621
 622         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
 623         if (err)
 624                 return ERR_PTR(err);
 625
 626         return register_user_hw_breakpoint(&attr, ptrace_triggered,
 627                                                  NULL, tsk);
 628 }
 629
 630 static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 631                                         int disabled)
 632 {
 633         struct perf_event_attr attr = bp->attr;
 634         int err;
 635
 636         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
 637         if (err)
 638                 return err;
 639
 640         return modify_user_hw_breakpoint(bp, &attr);
 641 }
 642
 643 /*
 644  * Handle ptrace writes to debug register 7.
 645  */
 646 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 647 {
 648         struct thread_struct *thread = &tsk->thread;
 649         unsigned long old_dr7;
 650         bool second_pass = false;
 651         int i, rc, ret = 0;
 652
 653         data &= ~DR_CONTROL_RESERVED;
 654         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 655
 656 restore:
 657         rc = 0;
 658         for (i = 0; i < HBP_NUM; i++) {
 659                 unsigned len, type;
 660                 bool disabled = !decode_dr7(data, i, &len, &type);
 661                 struct perf_event *bp = thread->ptrace_bps[i];
 662
 663                 if (!bp) {
 664                         if (disabled)
 665                                 continue;
 666
 667                         bp = ptrace_register_breakpoint(tsk,
 668                                         len, type, 0, disabled);
 669                         if (IS_ERR(bp)) {
 670                                 rc = PTR_ERR(bp);
 671                                 break;
 672                         }
 673
 674                         thread->ptrace_bps[i] = bp;
 675                         continue;
 676                 }
 677
 678                 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
 679                 if (rc)
 680                         break;
 681         }
 682
 683         /* Restore if the first pass failed, second_pass shouldn't fail. */
 684         if (rc && !WARN_ON(second_pass)) {
 685                 ret = rc;
 686                 data = old_dr7;
 687                 second_pass = true;
 688                 goto restore;
 689         }
 690
 691         return ret;
 692 }
 693
 694 /*
 695  * Handle PTRACE_PEEKUSR calls for the debug register area.
 696  */
 697 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 698 {
 699         struct thread_struct *thread = &tsk->thread;
 700         unsigned long val = 0;
 701
 702         if (n < HBP_NUM) {
 703                 struct perf_event *bp = thread->ptrace_bps[n];
 704
 705                 if (bp)
 706                         val = bp->hw.info.address;
 707         } else if (n == 6) {
 708                 val = thread->debugreg6;
 709         } else if (n == 7) {
 710                 val = thread->ptrace_dr7;
 711         }
 712         return val;
 713 }
 714
 715 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 716                                       unsigned long addr)
 717 {
 718         struct thread_struct *t = &tsk->thread;
 719         struct perf_event *bp = t->ptrace_bps[nr];
 720         int err = 0;
 721
 722         if (!bp) {
 723                 /*
 724                  * Put stub len and type to create an inactive but correct bp.
 725                  *
 726                  * CHECKME: the previous code returned -EIO if the addr wasn't
 727                  * a valid task virtual addr. The new one will return -EINVAL in
 728                  *  this case.
 729                  * -EINVAL may be what we want for in-kernel breakpoints users,
 730                  * but -EIO looks better for ptrace, since we refuse a register
 731                  * writing for the user. And anyway this is the previous
 732                  * behaviour.
 733                  */
 734                 bp = ptrace_register_breakpoint(tsk,
 735                                 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
 736                                 addr, true);
 737                 if (IS_ERR(bp))
 738                         err = PTR_ERR(bp);
 739                 else
 740                         t->ptrace_bps[nr] = bp;
 741         } else {
 742                 struct perf_event_attr attr = bp->attr;
 743
 744                 attr.bp_addr = addr;
 745                 err = modify_user_hw_breakpoint(bp, &attr);
 746         }
 747
 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         /* There are no DR4 or DR5 registers */
 759         int rc = -EIO;
 760
 761         if (n < HBP_NUM) {
 762                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 763         } else if (n == 6) {
 764                 thread->debugreg6 = val;
 765                 rc = 0;
 766         } else if (n == 7) {
 767                 rc = ptrace_write_dr7(tsk, val);
 768                 if (!rc)
 769                         thread->ptrace_dr7 = val;
 770         }
 771         return rc;
 772 }
 773
 774 /*
 775  * These access the current or another (stopped) task's io permission
 776  * bitmap for debugging or core dump.
 777  */
 778 static int ioperm_active(struct task_struct *target,
 779                          const struct user_regset *regset)
 780 {
 781         return target->thread.io_bitmap_max / regset->size;
 782 }
 783
 784 static int ioperm_get(struct task_struct *target,
 785                       const struct user_regset *regset,
 786                       unsigned int pos, unsigned int count,
 787                       void *kbuf, void __user *ubuf)
 788 {
 789         if (!target->thread.io_bitmap_ptr)
 790                 return -ENXIO;
 791
 792         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 793                                    target->thread.io_bitmap_ptr,
 794                                    0, IO_BITMAP_BYTES);
 795 }
 796
 797 /*
 798  * Called by kernel/ptrace.c when detaching..
 799  *
 800  * Make sure the single step bit is not set.
 801  */
 802 void ptrace_disable(struct task_struct *child)
 803 {
 804         user_disable_single_step(child);
 805 #ifdef TIF_SYSCALL_EMU
 806         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 807 #endif
 808 }
 809
 810 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 811 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 812 #endif
 813
 814 long arch_ptrace(struct task_struct *child, long request,
 815                  unsigned long addr, unsigned long data)
 816 {
 817         int ret;
 818         unsigned long __user *datap = (unsigned long __user *)data;
 819
 820         switch (request) {
 821         /* read the word at location addr in the USER area. */
 822         case PTRACE_PEEKUSR: {
 823                 unsigned long tmp;
 824
 825                 ret = -EIO;
 826                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 827                         break;
 828
 829                 tmp = 0;  /* Default return condition */
 830                 if (addr < sizeof(struct user_regs_struct))
 831                         tmp = getreg(child, addr);
 832                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 833                          addr <= offsetof(struct user, u_debugreg[7])) {
 834                         addr -= offsetof(struct user, u_debugreg[0]);
 835                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 836                 }
 837                 ret = put_user(tmp, datap);
 838                 break;
 839         }
 840
 841         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 842                 ret = -EIO;
 843                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 844                         break;
 845
 846                 if (addr < sizeof(struct user_regs_struct))
 847                         ret = putreg(child, addr, data);
 848                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 849                          addr <= offsetof(struct user, u_debugreg[7])) {
 850                         addr -= offsetof(struct user, u_debugreg[0]);
 851                         ret = ptrace_set_debugreg(child,
 852                                                   addr / sizeof(data), data);
 853                 }
 854                 break;
 855
 856         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 857                 return copy_regset_to_user(child,
 858                                            task_user_regset_view(current),
 859                                            REGSET_GENERAL,
 860                                            0, sizeof(struct user_regs_struct),
 861                                            datap);
 862
 863         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 864                 return copy_regset_from_user(child,
 865                                              task_user_regset_view(current),
 866                                              REGSET_GENERAL,
 867                                              0, sizeof(struct user_regs_struct),
 868                                              datap);
 869
 870         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 871                 return copy_regset_to_user(child,
 872                                            task_user_regset_view(current),
 873                                            REGSET_FP,
 874                                            0, sizeof(struct user_i387_struct),
 875                                            datap);
 876
 877         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 878                 return copy_regset_from_user(child,
 879                                              task_user_regset_view(current),
 880                                              REGSET_FP,
 881                                              0, sizeof(struct user_i387_struct),
 882                                              datap);
 883
 884 #ifdef CONFIG_X86_32
 885         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 886                 return copy_regset_to_user(child, &user_x86_32_view,
 887                                            REGSET_XFP,
 888                                            0, sizeof(struct user_fxsr_struct),
 889                                            datap) ? -EIO : 0;
 890
 891         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 892                 return copy_regset_from_user(child, &user_x86_32_view,
 893                                              REGSET_XFP,
 894                                              0, sizeof(struct user_fxsr_struct),
 895                                              datap) ? -EIO : 0;
 896 #endif
 897
 898 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 899         case PTRACE_GET_THREAD_AREA:
 900                 if ((int) addr < 0)
 901                         return -EIO;
 902                 ret = do_get_thread_area(child, addr,
 903                                         (struct user_desc __user *)data);
 904                 break;
 905
 906         case PTRACE_SET_THREAD_AREA:
 907                 if ((int) addr < 0)
 908                         return -EIO;
 909                 ret = do_set_thread_area(child, addr,
 910                                         (struct user_desc __user *)data, 0);
 911                 break;
 912 #endif
 913
 914 #ifdef CONFIG_X86_64
 915                 /* normal 64bit interface to access TLS data.
 916                    Works just like arch_prctl, except that the arguments
 917                    are reversed. */
 918         case PTRACE_ARCH_PRCTL:
 919                 ret = do_arch_prctl(child, data, addr);
 920                 break;
 921 #endif
 922
 923         default:
 924                 ret = ptrace_request(child, request, addr, data);
 925                 break;
 926         }
 927
 928         return ret;
 929 }
 930
 931 #ifdef CONFIG_IA32_EMULATION
 932
 933 #include <linux/compat.h>
 934 #include <linux/syscalls.h>
 935 #include <asm/ia32.h>
 936 #include <asm/user32.h>
 937
 938 #define R32(l,q)                                                        \
 939         case offsetof(struct user32, regs.l):                           \
 940                 regs->q = value; break
 941
 942 #define SEG32(rs)                                                       \
 943         case offsetof(struct user32, regs.rs):                          \
 944                 return set_segment_reg(child,                           \
 945                                        offsetof(struct user_regs_struct, rs), \
 946                                        value);                          \
 947                 break
 948
 949 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 950 {
 951         struct pt_regs *regs = task_pt_regs(child);
 952
 953         switch (regno) {
 954
 955         SEG32(cs);
 956         SEG32(ds);
 957         SEG32(es);
 958         SEG32(fs);
 959         SEG32(gs);
 960         SEG32(ss);
 961
 962         R32(ebx, bx);
 963         R32(ecx, cx);
 964         R32(edx, dx);
 965         R32(edi, di);
 966         R32(esi, si);
 967         R32(ebp, bp);
 968         R32(eax, ax);
 969         R32(eip, ip);
 970         R32(esp, sp);
 971
 972         case offsetof(struct user32, regs.orig_eax):
 973                 /*
 974                  * A 32-bit debugger setting orig_eax means to restore
 975                  * the state of the task restarting a 32-bit syscall.
 976                  * Make sure we interpret the -ERESTART* codes correctly
 977                  * in case the task is not actually still sitting at the
 978                  * exit from a 32-bit syscall with TS_COMPAT still set.
 979                  */
 980                 regs->orig_ax = value;
 981                 if (syscall_get_nr(child, regs) >= 0)
 982                         task_thread_info(child)->status |= TS_COMPAT;
 983                 break;
 984
 985         case offsetof(struct user32, regs.eflags):
 986                 return set_flags(child, value);
 987
 988         case offsetof(struct user32, u_debugreg[0]) ...
 989                 offsetof(struct user32, u_debugreg[7]):
 990                 regno -= offsetof(struct user32, u_debugreg[0]);
 991                 return ptrace_set_debugreg(child, regno / 4, value);
 992
 993         default:
 994                 if (regno > sizeof(struct user32) || (regno & 3))
 995                         return -EIO;
 996
 997                 /*
 998                  * Other dummy fields in the virtual user structure
 999                  * are ignored
1000                  */
1001                 break;
1002         }
1003         return 0;
1004 }
1005
1006 #undef R32
1007 #undef SEG32
1008
1009 #define R32(l,q)                                                        \
1010         case offsetof(struct user32, regs.l):                           \
1011                 *val = regs->q; break
1012
1013 #define SEG32(rs)                                                       \
1014         case offsetof(struct user32, regs.rs):                          \
1015                 *val = get_segment_reg(child,                           \
1016                                        offsetof(struct user_regs_struct, rs)); \
1017                 break
1018
1019 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1020 {
1021         struct pt_regs *regs = task_pt_regs(child);
1022
1023         switch (regno) {
1024
1025         SEG32(ds);
1026         SEG32(es);
1027         SEG32(fs);
1028         SEG32(gs);
1029
1030         R32(cs, cs);
1031         R32(ss, ss);
1032         R32(ebx, bx);
1033         R32(ecx, cx);
1034         R32(edx, dx);
1035         R32(edi, di);
1036         R32(esi, si);
1037         R32(ebp, bp);
1038         R32(eax, ax);
1039         R32(orig_eax, orig_ax);
1040         R32(eip, ip);
1041         R32(esp, sp);
1042
1043         case offsetof(struct user32, regs.eflags):
1044                 *val = get_flags(child);
1045                 break;
1046
1047         case offsetof(struct user32, u_debugreg[0]) ...
1048                 offsetof(struct user32, u_debugreg[7]):
1049                 regno -= offsetof(struct user32, u_debugreg[0]);
1050                 *val = ptrace_get_debugreg(child, regno / 4);
1051                 break;
1052
1053         default:
1054                 if (regno > sizeof(struct user32) || (regno & 3))
1055                         return -EIO;
1056
1057                 /*
1058                  * Other dummy fields in the virtual user structure
1059                  * are ignored
1060                  */
1061                 *val = 0;
1062                 break;
1063         }
1064         return 0;
1065 }
1066
1067 #undef R32
1068 #undef SEG32
1069
1070 static int genregs32_get(struct task_struct *target,
1071                          const struct user_regset *regset,
1072                          unsigned int pos, unsigned int count,
1073                          void *kbuf, void __user *ubuf)
1074 {
1075         if (kbuf) {
1076                 compat_ulong_t *k = kbuf;
1077                 while (count >= sizeof(*k)) {
1078                         getreg32(target, pos, k++);
1079                         count -= sizeof(*k);
1080                         pos += sizeof(*k);
1081                 }
1082         } else {
1083                 compat_ulong_t __user *u = ubuf;
1084                 while (count >= sizeof(*u)) {
1085                         compat_ulong_t word;
1086                         getreg32(target, pos, &word);
1087                         if (__put_user(word, u++))
1088                                 return -EFAULT;
1089                         count -= sizeof(*u);
1090                         pos += sizeof(*u);
1091                 }
1092         }
1093
1094         return 0;
1095 }
1096
1097 static int genregs32_set(struct task_struct *target,
1098                          const struct user_regset *regset,
1099                          unsigned int pos, unsigned int count,
1100                          const void *kbuf, const void __user *ubuf)
1101 {
1102         int ret = 0;
1103         if (kbuf) {
1104                 const compat_ulong_t *k = kbuf;
1105                 while (count >= sizeof(*k) && !ret) {
1106                         ret = putreg32(target, pos, *k++);
1107                         count -= sizeof(*k);
1108                         pos += sizeof(*k);
1109                 }
1110         } else {
1111                 const compat_ulong_t __user *u = ubuf;
1112                 while (count >= sizeof(*u) && !ret) {
1113                         compat_ulong_t word;
1114                         ret = __get_user(word, u++);
1115                         if (ret)
1116                                 break;
1117                         ret = putreg32(target, pos, word);
1118                         count -= sizeof(*u);
1119                         pos += sizeof(*u);
1120                 }
1121         }
1122         return ret;
1123 }
1124
1125 #ifdef CONFIG_X86_X32_ABI
1126 static long x32_arch_ptrace(struct task_struct *child,
1127                             compat_long_t request, compat_ulong_t caddr,
1128                             compat_ulong_t cdata)
1129 {
1130         unsigned long addr = caddr;
1131         unsigned long data = cdata;
1132         void __user *datap = compat_ptr(data);
1133         int ret;
1134
1135         switch (request) {
1136         /* Read 32bits at location addr in the USER area.  Only allow
1137            to return the lower 32bits of segment and debug registers.  */
1138         case PTRACE_PEEKUSR: {
1139                 u32 tmp;
1140
1141                 ret = -EIO;
1142                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1143                     addr < offsetof(struct user_regs_struct, cs))
1144                         break;
1145
1146                 tmp = 0;  /* Default return condition */
1147                 if (addr < sizeof(struct user_regs_struct))
1148                         tmp = getreg(child, addr);
1149                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1150                          addr <= offsetof(struct user, u_debugreg[7])) {
1151                         addr -= offsetof(struct user, u_debugreg[0]);
1152                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1153                 }
1154                 ret = put_user(tmp, (__u32 __user *)datap);
1155                 break;
1156         }
1157
1158         /* Write the word at location addr in the USER area.  Only allow
1159            to update segment and debug registers with the upper 32bits
1160            zero-extended. */
1161         case PTRACE_POKEUSR:
1162                 ret = -EIO;
1163                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1164                     addr < offsetof(struct user_regs_struct, cs))
1165                         break;
1166
1167                 if (addr < sizeof(struct user_regs_struct))
1168                         ret = putreg(child, addr, data);
1169                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1170                          addr <= offsetof(struct user, u_debugreg[7])) {
1171                         addr -= offsetof(struct user, u_debugreg[0]);
1172                         ret = ptrace_set_debugreg(child,
1173                                                   addr / sizeof(data), data);
1174                 }
1175                 break;
1176
1177         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1178                 return copy_regset_to_user(child,
1179                                            task_user_regset_view(current),
1180                                            REGSET_GENERAL,
1181                                            0, sizeof(struct user_regs_struct),
1182                                            datap);
1183
1184         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1185                 return copy_regset_from_user(child,
1186                                              task_user_regset_view(current),
1187                                              REGSET_GENERAL,
1188                                              0, sizeof(struct user_regs_struct),
1189                                              datap);
1190
1191         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1192                 return copy_regset_to_user(child,
1193                                            task_user_regset_view(current),
1194                                            REGSET_FP,
1195                                            0, sizeof(struct user_i387_struct),
1196                                            datap);
1197
1198         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1199                 return copy_regset_from_user(child,
1200                                              task_user_regset_view(current),
1201                                              REGSET_FP,
1202                                              0, sizeof(struct user_i387_struct),
1203                                              datap);
1204
1205         default:
1206                 return compat_ptrace_request(child, request, addr, data);
1207         }
1208
1209         return ret;
1210 }
1211 #endif
1212
1213 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1214                         compat_ulong_t caddr, compat_ulong_t cdata)
1215 {
1216         unsigned long addr = caddr;
1217         unsigned long data = cdata;
1218         void __user *datap = compat_ptr(data);
1219         int ret;
1220         __u32 val;
1221
1222 #ifdef CONFIG_X86_X32_ABI
1223         if (!is_ia32_task())
1224                 return x32_arch_ptrace(child, request, caddr, cdata);
1225 #endif
1226
1227         switch (request) {
1228         case PTRACE_PEEKUSR:
1229                 ret = getreg32(child, addr, &val);
1230                 if (ret == 0)
1231                         ret = put_user(val, (__u32 __user *)datap);
1232                 break;
1233
1234         case PTRACE_POKEUSR:
1235                 ret = putreg32(child, addr, data);
1236                 break;
1237
1238         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1239                 return copy_regset_to_user(child, &user_x86_32_view,
1240                                            REGSET_GENERAL,
1241                                            0, sizeof(struct user_regs_struct32),
1242                                            datap);
1243
1244         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1245                 return copy_regset_from_user(child, &user_x86_32_view,
1246                                              REGSET_GENERAL, 0,
1247                                              sizeof(struct user_regs_struct32),
1248                                              datap);
1249
1250         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1251                 return copy_regset_to_user(child, &user_x86_32_view,
1252                                            REGSET_FP, 0,
1253                                            sizeof(struct user_i387_ia32_struct),
1254                                            datap);
1255
1256         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1257                 return copy_regset_from_user(
1258                         child, &user_x86_32_view, REGSET_FP,
1259                         0, sizeof(struct user_i387_ia32_struct), datap);
1260
1261         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1262                 return copy_regset_to_user(child, &user_x86_32_view,
1263                                            REGSET_XFP, 0,
1264                                            sizeof(struct user32_fxsr_struct),
1265                                            datap);
1266
1267         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1268                 return copy_regset_from_user(child, &user_x86_32_view,
1269                                              REGSET_XFP, 0,
1270                                              sizeof(struct user32_fxsr_struct),
1271                                              datap);
1272
1273         case PTRACE_GET_THREAD_AREA:
1274         case PTRACE_SET_THREAD_AREA:
1275                 return arch_ptrace(child, request, addr, data);
1276
1277         default:
1278                 return compat_ptrace_request(child, request, addr, data);
1279         }
1280
1281         return ret;
1282 }
1283
1284 #endif  /* CONFIG_IA32_EMULATION */
1285
1286 #ifdef CONFIG_X86_64
1287
1288 static struct user_regset x86_64_regsets[] __read_mostly = {
1289         [REGSET_GENERAL] = {
1290                 .core_note_type = NT_PRSTATUS,
1291                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1292                 .size = sizeof(long), .align = sizeof(long),
1293                 .get = genregs_get, .set = genregs_set
1294         },
1295         [REGSET_FP] = {
1296                 .core_note_type = NT_PRFPREG,
1297                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1298                 .size = sizeof(long), .align = sizeof(long),
1299                 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1300         },
1301         [REGSET_XSTATE] = {
1302                 .core_note_type = NT_X86_XSTATE,
1303                 .size = sizeof(u64), .align = sizeof(u64),
1304                 .active = xstateregs_active, .get = xstateregs_get,
1305                 .set = xstateregs_set
1306         },
1307         [REGSET_IOPERM64] = {
1308                 .core_note_type = NT_386_IOPERM,
1309                 .n = IO_BITMAP_LONGS,
1310                 .size = sizeof(long), .align = sizeof(long),
1311                 .active = ioperm_active, .get = ioperm_get
1312         },
1313 };
1314
1315 static const struct user_regset_view user_x86_64_view = {
1316         .name = "x86_64", .e_machine = EM_X86_64,
1317         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1318 };
1319
1320 #else  /* CONFIG_X86_32 */
1321
1322 #define user_regs_struct32      user_regs_struct
1323 #define genregs32_get           genregs_get
1324 #define genregs32_set           genregs_set
1325
1326 #endif  /* CONFIG_X86_64 */
1327
1328 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1329 static struct user_regset x86_32_regsets[] __read_mostly = {
1330         [REGSET_GENERAL] = {
1331                 .core_note_type = NT_PRSTATUS,
1332                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1333                 .size = sizeof(u32), .align = sizeof(u32),
1334                 .get = genregs32_get, .set = genregs32_set
1335         },
1336         [REGSET_FP] = {
1337                 .core_note_type = NT_PRFPREG,
1338                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1339                 .size = sizeof(u32), .align = sizeof(u32),
1340                 .active = regset_fpregs_active, .get = fpregs_get, .set = fpregs_set
1341         },
1342         [REGSET_XFP] = {
1343                 .core_note_type = NT_PRXFPREG,
1344                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1345                 .size = sizeof(u32), .align = sizeof(u32),
1346                 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1347         },
1348         [REGSET_XSTATE] = {
1349                 .core_note_type = NT_X86_XSTATE,
1350                 .size = sizeof(u64), .align = sizeof(u64),
1351                 .active = xstateregs_active, .get = xstateregs_get,
1352                 .set = xstateregs_set
1353         },
1354         [REGSET_TLS] = {
1355                 .core_note_type = NT_386_TLS,
1356                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1357                 .size = sizeof(struct user_desc),
1358                 .align = sizeof(struct user_desc),
1359                 .active = regset_tls_active,
1360                 .get = regset_tls_get, .set = regset_tls_set
1361         },
1362         [REGSET_IOPERM32] = {
1363                 .core_note_type = NT_386_IOPERM,
1364                 .n = IO_BITMAP_BYTES / sizeof(u32),
1365                 .size = sizeof(u32), .align = sizeof(u32),
1366                 .active = ioperm_active, .get = ioperm_get
1367         },
1368 };
1369
1370 static const struct user_regset_view user_x86_32_view = {
1371         .name = "i386", .e_machine = EM_386,
1372         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1373 };
1374 #endif
1375
1376 /*
1377  * This represents bytes 464..511 in the memory layout exported through
1378  * the REGSET_XSTATE interface.
1379  */
1380 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1381
1382 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1383 {
1384 #ifdef CONFIG_X86_64
1385         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1386 #endif
1387 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1388         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1389 #endif
1390         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1391 }
1392
1393 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1394 {
1395 #ifdef CONFIG_IA32_EMULATION
1396         if (test_tsk_thread_flag(task, TIF_IA32))
1397 #endif
1398 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1399                 return &user_x86_32_view;
1400 #endif
1401 #ifdef CONFIG_X86_64
1402         return &user_x86_64_view;
1403 #endif
1404 }
1405
1406 static void fill_sigtrap_info(struct task_struct *tsk,
1407                                 struct pt_regs *regs,
1408                                 int error_code, int si_code,
1409                                 struct siginfo *info)
1410 {
1411         tsk->thread.trap_nr = X86_TRAP_DB;
1412         tsk->thread.error_code = error_code;
1413
1414         memset(info, 0, sizeof(*info));
1415         info->si_signo = SIGTRAP;
1416         info->si_code = si_code;
1417         info->si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
1418 }
1419
1420 void user_single_step_siginfo(struct task_struct *tsk,
1421                                 struct pt_regs *regs,
1422                                 struct siginfo *info)
1423 {
1424         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1425 }
1426
1427 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1428                                          int error_code, int si_code)
1429 {
1430         struct siginfo info;
1431
1432         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1433         /* Send us the fake SIGTRAP */
1434         force_sig_info(SIGTRAP, &info, tsk);
1435 }
1436
1437 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
1438 {
1439 #ifdef CONFIG_X86_64
1440         if (arch == AUDIT_ARCH_X86_64) {
1441                 audit_syscall_entry(regs->orig_ax, regs->di,
1442                                     regs->si, regs->dx, regs->r10);
1443         } else
1444 #endif
1445         {
1446                 audit_syscall_entry(regs->orig_ax, regs->bx,
1447                                     regs->cx, regs->dx, regs->si);
1448         }
1449 }
1450
1451 /*
1452  * We can return 0 to resume the syscall or anything else to go to phase
1453  * 2.  If we resume the syscall, we need to put something appropriate in
1454  * regs->orig_ax.
1455  *
1456  * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
1457  * are fully functional.
1458  *
1459  * For phase 2's benefit, our return value is:
1460  * 0:                   resume the syscall
1461  * 1:                   go to phase 2; no seccomp phase 2 needed
1462  * anything else:       go to phase 2; pass return value to seccomp
1463  */
1464 unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
1465 {
1466         unsigned long ret = 0;
1467         u32 work;
1468
1469         BUG_ON(regs != task_pt_regs(current));
1470
1471         work = ACCESS_ONCE(current_thread_info()->flags) &
1472                 _TIF_WORK_SYSCALL_ENTRY;
1473
1474         /*
1475          * If TIF_NOHZ is set, we are required to call user_exit() before
1476          * doing anything that could touch RCU.
1477          */
1478         if (work & _TIF_NOHZ) {
1479                 user_exit();
1480                 work &= ~_TIF_NOHZ;
1481         }
1482
1483 #ifdef CONFIG_SECCOMP
1484         /*
1485          * Do seccomp first -- it should minimize exposure of other
1486          * code, and keeping seccomp fast is probably more valuable
1487          * than the rest of this.
1488          */
1489         if (work & _TIF_SECCOMP) {
1490                 struct seccomp_data sd;
1491
1492                 sd.arch = arch;
1493                 sd.nr = regs->orig_ax;
1494                 sd.instruction_pointer = regs->ip;
1495 #ifdef CONFIG_X86_64
1496                 if (arch == AUDIT_ARCH_X86_64) {
1497                         sd.args[0] = regs->di;
1498                         sd.args[1] = regs->si;
1499                         sd.args[2] = regs->dx;
1500                         sd.args[3] = regs->r10;
1501                         sd.args[4] = regs->r8;
1502                         sd.args[5] = regs->r9;
1503                 } else
1504 #endif
1505                 {
1506                         sd.args[0] = regs->bx;
1507                         sd.args[1] = regs->cx;
1508                         sd.args[2] = regs->dx;
1509                         sd.args[3] = regs->si;
1510                         sd.args[4] = regs->di;
1511                         sd.args[5] = regs->bp;
1512                 }
1513
1514                 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
1515                 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
1516
1517                 ret = seccomp_phase1(&sd);
1518                 if (ret == SECCOMP_PHASE1_SKIP) {
1519                         regs->orig_ax = -1;
1520                         ret = 0;
1521                 } else if (ret != SECCOMP_PHASE1_OK) {
1522                         return ret;  /* Go directly to phase 2 */
1523                 }
1524
1525                 work &= ~_TIF_SECCOMP;
1526         }
1527 #endif
1528
1529         /* Do our best to finish without phase 2. */
1530         if (work == 0)
1531                 return ret;  /* seccomp and/or nohz only (ret == 0 here) */
1532
1533 #ifdef CONFIG_AUDITSYSCALL
1534         if (work == _TIF_SYSCALL_AUDIT) {
1535                 /*
1536                  * If there is no more work to be done except auditing,
1537                  * then audit in phase 1.  Phase 2 always audits, so, if
1538                  * we audit here, then we can't go on to phase 2.
1539                  */
1540                 do_audit_syscall_entry(regs, arch);
1541                 return 0;
1542         }
1543 #endif
1544
1545         return 1;  /* Something is enabled that we can't handle in phase 1 */
1546 }
1547
1548 /* Returns the syscall nr to run (which should match regs->orig_ax). */
1549 long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
1550                                 unsigned long phase1_result)
1551 {
1552         long ret = 0;
1553         u32 work = ACCESS_ONCE(current_thread_info()->flags) &
1554                 _TIF_WORK_SYSCALL_ENTRY;
1555
1556         BUG_ON(regs != task_pt_regs(current));
1557
1558         /*
1559          * If we stepped into a sysenter/syscall insn, it trapped in
1560          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1561          * If user-mode had set TF itself, then it's still clear from
1562          * do_debug() and we need to set it again to restore the user
1563          * state.  If we entered on the slow path, TF was already set.
1564          */
1565         if (work & _TIF_SINGLESTEP)
1566                 regs->flags |= X86_EFLAGS_TF;
1567
1568 #ifdef CONFIG_SECCOMP
1569         /*
1570          * Call seccomp_phase2 before running the other hooks so that
1571          * they can see any changes made by a seccomp tracer.
1572          */
1573         if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
1574                 /* seccomp failures shouldn't expose any additional code. */
1575                 return -1;
1576         }
1577 #endif
1578
1579         if (unlikely(work & _TIF_SYSCALL_EMU))
1580                 ret = -1L;
1581
1582         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1583             tracehook_report_syscall_entry(regs))
1584                 ret = -1L;
1585
1586         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1587                 trace_sys_enter(regs, regs->orig_ax);
1588
1589         do_audit_syscall_entry(regs, arch);
1590
1591         return ret ?: regs->orig_ax;
1592 }
1593
1594 long syscall_trace_enter(struct pt_regs *regs)
1595 {
1596         u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
1597         unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
1598
1599         if (phase1_result == 0)
1600                 return regs->orig_ax;
1601         else
1602                 return syscall_trace_enter_phase2(regs, arch, phase1_result);
1603 }
1604
1605 void syscall_trace_leave(struct pt_regs *regs)
1606 {
1607         bool step;
1608
1609         /*
1610          * We may come here right after calling schedule_user()
1611          * or do_notify_resume(), in which case we can be in RCU
1612          * user mode.
1613          */
1614         user_exit();
1615
1616         audit_syscall_exit(regs);
1617
1618         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1619                 trace_sys_exit(regs, regs->ax);
1620
1621         /*
1622          * If TIF_SYSCALL_EMU is set, we only get here because of
1623          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1624          * We already reported this syscall instruction in
1625          * syscall_trace_enter().
1626          */
1627         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1628                         !test_thread_flag(TIF_SYSCALL_EMU);
1629         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1630                 tracehook_report_syscall_exit(regs, step);
1631
1632         user_enter();
1633 }