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/i387.h>
  32 #include <asm/fpu-internal.h>
  33 #include <asm/debugreg.h>
  34 #include <asm/ldt.h>
  35 #include <asm/desc.h>
  36 #include <asm/prctl.h>
  37 #include <asm/proto.h>
  38 #include <asm/hw_breakpoint.h>
  39 #include <asm/traps.h>
  40
  41 #include "tls.h"
  42
  43 #define CREATE_TRACE_POINTS
  44 #include <trace/events/syscalls.h>
  45
  46 enum x86_regset {
  47         REGSET_GENERAL,
  48         REGSET_FP,
  49         REGSET_XFP,
  50         REGSET_IOPERM64 = REGSET_XFP,
  51         REGSET_XSTATE,
  52         REGSET_TLS,
  53         REGSET_IOPERM32,
  54 };
  55
  56 struct pt_regs_offset {
  57         const char *name;
  58         int offset;
  59 };
  60
  61 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  62 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  63
  64 static const struct pt_regs_offset regoffset_table[] = {
  65 #ifdef CONFIG_X86_64
  66         REG_OFFSET_NAME(r15),
  67         REG_OFFSET_NAME(r14),
  68         REG_OFFSET_NAME(r13),
  69         REG_OFFSET_NAME(r12),
  70         REG_OFFSET_NAME(r11),
  71         REG_OFFSET_NAME(r10),
  72         REG_OFFSET_NAME(r9),
  73         REG_OFFSET_NAME(r8),
  74 #endif
  75         REG_OFFSET_NAME(bx),
  76         REG_OFFSET_NAME(cx),
  77         REG_OFFSET_NAME(dx),
  78         REG_OFFSET_NAME(si),
  79         REG_OFFSET_NAME(di),
  80         REG_OFFSET_NAME(bp),
  81         REG_OFFSET_NAME(ax),
  82 #ifdef CONFIG_X86_32
  83         REG_OFFSET_NAME(ds),
  84         REG_OFFSET_NAME(es),
  85         REG_OFFSET_NAME(fs),
  86         REG_OFFSET_NAME(gs),
  87 #endif
  88         REG_OFFSET_NAME(orig_ax),
  89         REG_OFFSET_NAME(ip),
  90         REG_OFFSET_NAME(cs),
  91         REG_OFFSET_NAME(flags),
  92         REG_OFFSET_NAME(sp),
  93         REG_OFFSET_NAME(ss),
  94         REG_OFFSET_END,
  95 };
  96
  97 /**
  98  * regs_query_register_offset() - query register offset from its name
  99  * @name:       the name of a register
 100  *
 101  * regs_query_register_offset() returns the offset of a register in struct
 102  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
 103  */
 104 int regs_query_register_offset(const char *name)
 105 {
 106         const struct pt_regs_offset *roff;
 107         for (roff = regoffset_table; roff->name != NULL; roff++)
 108                 if (!strcmp(roff->name, name))
 109                         return roff->offset;
 110         return -EINVAL;
 111 }
 112
 113 /**
 114  * regs_query_register_name() - query register name from its offset
 115  * @offset:     the offset of a register in struct pt_regs.
 116  *
 117  * regs_query_register_name() returns the name of a register from its
 118  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 119  */
 120 const char *regs_query_register_name(unsigned int offset)
 121 {
 122         const struct pt_regs_offset *roff;
 123         for (roff = regoffset_table; roff->name != NULL; roff++)
 124                 if (roff->offset == offset)
 125                         return roff->name;
 126         return NULL;
 127 }
 128
 129 static const int arg_offs_table[] = {
 130 #ifdef CONFIG_X86_32
 131         [0] = offsetof(struct pt_regs, ax),
 132         [1] = offsetof(struct pt_regs, dx),
 133         [2] = offsetof(struct pt_regs, cx)
 134 #else /* CONFIG_X86_64 */
 135         [0] = offsetof(struct pt_regs, di),
 136         [1] = offsetof(struct pt_regs, si),
 137         [2] = offsetof(struct pt_regs, dx),
 138         [3] = offsetof(struct pt_regs, cx),
 139         [4] = offsetof(struct pt_regs, r8),
 140         [5] = offsetof(struct pt_regs, r9)
 141 #endif
 142 };
 143
 144 /*
 145  * does not yet catch signals sent when the child dies.
 146  * in exit.c or in signal.c.
 147  */
 148
 149 /*
 150  * Determines which flags the user has access to [1 = access, 0 = no access].
 151  */
 152 #define FLAG_MASK_32            ((unsigned long)                        \
 153                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 154                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 155                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 156                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 157                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 158
 159 /*
 160  * Determines whether a value may be installed in a segment register.
 161  */
 162 static inline bool invalid_selector(u16 value)
 163 {
 164         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 165 }
 166
 167 #ifdef CONFIG_X86_32
 168
 169 #define FLAG_MASK               FLAG_MASK_32
 170
 171 /*
 172  * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
 173  * when it traps.  The previous stack will be directly underneath the saved
 174  * registers, and 'sp/ss' won't even have been saved. Thus the '&regs->sp'.
 175  *
 176  * Now, if the stack is empty, '&regs->sp' is out of range. In this
 177  * case we try to take the previous stack. To always return a non-null
 178  * stack pointer we fall back to regs as stack if no previous stack
 179  * exists.
 180  *
 181  * This is valid only for kernel mode traps.
 182  */
 183 unsigned long kernel_stack_pointer(struct pt_regs *regs)
 184 {
 185         unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
 186         unsigned long sp = (unsigned long)&regs->sp;
 187         u32 *prev_esp;
 188
 189         if (context == (sp & ~(THREAD_SIZE - 1)))
 190                 return sp;
 191
 192         prev_esp = (u32 *)(context);
 193         if (prev_esp)
 194                 return (unsigned long)prev_esp;
 195
 196         return (unsigned long)regs;
 197 }
 198 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
 199
 200 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 201 {
 202         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 203         return &regs->bx + (regno >> 2);
 204 }
 205
 206 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 207 {
 208         /*
 209          * Returning the value truncates it to 16 bits.
 210          */
 211         unsigned int retval;
 212         if (offset != offsetof(struct user_regs_struct, gs))
 213                 retval = *pt_regs_access(task_pt_regs(task), offset);
 214         else {
 215                 if (task == current)
 216                         retval = get_user_gs(task_pt_regs(task));
 217                 else
 218                         retval = task_user_gs(task);
 219         }
 220         return retval;
 221 }
 222
 223 static int set_segment_reg(struct task_struct *task,
 224                            unsigned long offset, u16 value)
 225 {
 226         /*
 227          * The value argument was already truncated to 16 bits.
 228          */
 229         if (invalid_selector(value))
 230                 return -EIO;
 231
 232         /*
 233          * For %cs and %ss we cannot permit a null selector.
 234          * We can permit a bogus selector as long as it has USER_RPL.
 235          * Null selectors are fine for other segment registers, but
 236          * we will never get back to user mode with invalid %cs or %ss
 237          * and will take the trap in iret instead.  Much code relies
 238          * on user_mode() to distinguish a user trap frame (which can
 239          * safely use invalid selectors) from a kernel trap frame.
 240          */
 241         switch (offset) {
 242         case offsetof(struct user_regs_struct, cs):
 243         case offsetof(struct user_regs_struct, ss):
 244                 if (unlikely(value == 0))
 245                         return -EIO;
 246
 247         default:
 248                 *pt_regs_access(task_pt_regs(task), offset) = value;
 249                 break;
 250
 251         case offsetof(struct user_regs_struct, gs):
 252                 if (task == current)
 253                         set_user_gs(task_pt_regs(task), value);
 254                 else
 255                         task_user_gs(task) = value;
 256         }
 257
 258         return 0;
 259 }
 260
 261 #else  /* CONFIG_X86_64 */
 262
 263 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 264
 265 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 266 {
 267         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 268         return &regs->r15 + (offset / sizeof(regs->r15));
 269 }
 270
 271 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 272 {
 273         /*
 274          * Returning the value truncates it to 16 bits.
 275          */
 276         unsigned int seg;
 277
 278         switch (offset) {
 279         case offsetof(struct user_regs_struct, fs):
 280                 if (task == current) {
 281                         /* Older gas can't assemble movq %?s,%r?? */
 282                         asm("movl %%fs,%0" : "=r" (seg));
 283                         return seg;
 284                 }
 285                 return task->thread.fsindex;
 286         case offsetof(struct user_regs_struct, gs):
 287                 if (task == current) {
 288                         asm("movl %%gs,%0" : "=r" (seg));
 289                         return seg;
 290                 }
 291                 return task->thread.gsindex;
 292         case offsetof(struct user_regs_struct, ds):
 293                 if (task == current) {
 294                         asm("movl %%ds,%0" : "=r" (seg));
 295                         return seg;
 296                 }
 297                 return task->thread.ds;
 298         case offsetof(struct user_regs_struct, es):
 299                 if (task == current) {
 300                         asm("movl %%es,%0" : "=r" (seg));
 301                         return seg;
 302                 }
 303                 return task->thread.es;
 304
 305         case offsetof(struct user_regs_struct, cs):
 306         case offsetof(struct user_regs_struct, ss):
 307                 break;
 308         }
 309         return *pt_regs_access(task_pt_regs(task), offset);
 310 }
 311
 312 static int set_segment_reg(struct task_struct *task,
 313                            unsigned long offset, u16 value)
 314 {
 315         /*
 316          * The value argument was already truncated to 16 bits.
 317          */
 318         if (invalid_selector(value))
 319                 return -EIO;
 320
 321         switch (offset) {
 322         case offsetof(struct user_regs_struct,fs):
 323                 /*
 324                  * If this is setting fs as for normal 64-bit use but
 325                  * setting fs_base has implicitly changed it, leave it.
 326                  */
 327                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 328                      task->thread.fs != 0) ||
 329                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 330                      task->thread.fs == 0))
 331                         break;
 332                 task->thread.fsindex = value;
 333                 if (task == current)
 334                         loadsegment(fs, task->thread.fsindex);
 335                 break;
 336         case offsetof(struct user_regs_struct,gs):
 337                 /*
 338                  * If this is setting gs as for normal 64-bit use but
 339                  * setting gs_base has implicitly changed it, leave it.
 340                  */
 341                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 342                      task->thread.gs != 0) ||
 343                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 344                      task->thread.gs == 0))
 345                         break;
 346                 task->thread.gsindex = value;
 347                 if (task == current)
 348                         load_gs_index(task->thread.gsindex);
 349                 break;
 350         case offsetof(struct user_regs_struct,ds):
 351                 task->thread.ds = value;
 352                 if (task == current)
 353                         loadsegment(ds, task->thread.ds);
 354                 break;
 355         case offsetof(struct user_regs_struct,es):
 356                 task->thread.es = value;
 357                 if (task == current)
 358                         loadsegment(es, task->thread.es);
 359                 break;
 360
 361                 /*
 362                  * Can't actually change these in 64-bit mode.
 363                  */
 364         case offsetof(struct user_regs_struct,cs):
 365                 if (unlikely(value == 0))
 366                         return -EIO;
 367 #ifdef CONFIG_IA32_EMULATION
 368                 if (test_tsk_thread_flag(task, TIF_IA32))
 369                         task_pt_regs(task)->cs = value;
 370 #endif
 371                 break;
 372         case offsetof(struct user_regs_struct,ss):
 373                 if (unlikely(value == 0))
 374                         return -EIO;
 375 #ifdef CONFIG_IA32_EMULATION
 376                 if (test_tsk_thread_flag(task, TIF_IA32))
 377                         task_pt_regs(task)->ss = value;
 378 #endif
 379                 break;
 380         }
 381
 382         return 0;
 383 }
 384
 385 #endif  /* CONFIG_X86_32 */
 386
 387 static unsigned long get_flags(struct task_struct *task)
 388 {
 389         unsigned long retval = task_pt_regs(task)->flags;
 390
 391         /*
 392          * If the debugger set TF, hide it from the readout.
 393          */
 394         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 395                 retval &= ~X86_EFLAGS_TF;
 396
 397         return retval;
 398 }
 399
 400 static int set_flags(struct task_struct *task, unsigned long value)
 401 {
 402         struct pt_regs *regs = task_pt_regs(task);
 403
 404         /*
 405          * If the user value contains TF, mark that
 406          * it was not "us" (the debugger) that set it.
 407          * If not, make sure it stays set if we had.
 408          */
 409         if (value & X86_EFLAGS_TF)
 410                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 411         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 412                 value |= X86_EFLAGS_TF;
 413
 414         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 415
 416         return 0;
 417 }
 418
 419 static int putreg(struct task_struct *child,
 420                   unsigned long offset, unsigned long value)
 421 {
 422         switch (offset) {
 423         case offsetof(struct user_regs_struct, cs):
 424         case offsetof(struct user_regs_struct, ds):
 425         case offsetof(struct user_regs_struct, es):
 426         case offsetof(struct user_regs_struct, fs):
 427         case offsetof(struct user_regs_struct, gs):
 428         case offsetof(struct user_regs_struct, ss):
 429                 return set_segment_reg(child, offset, value);
 430
 431         case offsetof(struct user_regs_struct, flags):
 432                 return set_flags(child, value);
 433
 434 #ifdef CONFIG_X86_64
 435         case offsetof(struct user_regs_struct,fs_base):
 436                 if (value >= TASK_SIZE_OF(child))
 437                         return -EIO;
 438                 /*
 439                  * When changing the segment base, use do_arch_prctl
 440                  * to set either thread.fs or thread.fsindex and the
 441                  * corresponding GDT slot.
 442                  */
 443                 if (child->thread.fs != value)
 444                         return do_arch_prctl(child, ARCH_SET_FS, value);
 445                 return 0;
 446         case offsetof(struct user_regs_struct,gs_base):
 447                 /*
 448                  * Exactly the same here as the %fs handling above.
 449                  */
 450                 if (value >= TASK_SIZE_OF(child))
 451                         return -EIO;
 452                 if (child->thread.gs != value)
 453                         return do_arch_prctl(child, ARCH_SET_GS, value);
 454                 return 0;
 455 #endif
 456         }
 457
 458         *pt_regs_access(task_pt_regs(child), offset) = value;
 459         return 0;
 460 }
 461
 462 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 463 {
 464         switch (offset) {
 465         case offsetof(struct user_regs_struct, cs):
 466         case offsetof(struct user_regs_struct, ds):
 467         case offsetof(struct user_regs_struct, es):
 468         case offsetof(struct user_regs_struct, fs):
 469         case offsetof(struct user_regs_struct, gs):
 470         case offsetof(struct user_regs_struct, ss):
 471                 return get_segment_reg(task, offset);
 472
 473         case offsetof(struct user_regs_struct, flags):
 474                 return get_flags(task);
 475
 476 #ifdef CONFIG_X86_64
 477         case offsetof(struct user_regs_struct, fs_base): {
 478                 /*
 479                  * do_arch_prctl may have used a GDT slot instead of
 480                  * the MSR.  To userland, it appears the same either
 481                  * way, except the %fs segment selector might not be 0.
 482                  */
 483                 unsigned int seg = task->thread.fsindex;
 484                 if (task->thread.fs != 0)
 485                         return task->thread.fs;
 486                 if (task == current)
 487                         asm("movl %%fs,%0" : "=r" (seg));
 488                 if (seg != FS_TLS_SEL)
 489                         return 0;
 490                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 491         }
 492         case offsetof(struct user_regs_struct, gs_base): {
 493                 /*
 494                  * Exactly the same here as the %fs handling above.
 495                  */
 496                 unsigned int seg = task->thread.gsindex;
 497                 if (task->thread.gs != 0)
 498                         return task->thread.gs;
 499                 if (task == current)
 500                         asm("movl %%gs,%0" : "=r" (seg));
 501                 if (seg != GS_TLS_SEL)
 502                         return 0;
 503                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 504         }
 505 #endif
 506         }
 507
 508         return *pt_regs_access(task_pt_regs(task), offset);
 509 }
 510
 511 static int genregs_get(struct task_struct *target,
 512                        const struct user_regset *regset,
 513                        unsigned int pos, unsigned int count,
 514                        void *kbuf, void __user *ubuf)
 515 {
 516         if (kbuf) {
 517                 unsigned long *k = kbuf;
 518                 while (count >= sizeof(*k)) {
 519                         *k++ = getreg(target, pos);
 520                         count -= sizeof(*k);
 521                         pos += sizeof(*k);
 522                 }
 523         } else {
 524                 unsigned long __user *u = ubuf;
 525                 while (count >= sizeof(*u)) {
 526                         if (__put_user(getreg(target, pos), u++))
 527                                 return -EFAULT;
 528                         count -= sizeof(*u);
 529                         pos += sizeof(*u);
 530                 }
 531         }
 532
 533         return 0;
 534 }
 535
 536 static int genregs_set(struct task_struct *target,
 537                        const struct user_regset *regset,
 538                        unsigned int pos, unsigned int count,
 539                        const void *kbuf, const void __user *ubuf)
 540 {
 541         int ret = 0;
 542         if (kbuf) {
 543                 const unsigned long *k = kbuf;
 544                 while (count >= sizeof(*k) && !ret) {
 545                         ret = putreg(target, pos, *k++);
 546                         count -= sizeof(*k);
 547                         pos += sizeof(*k);
 548                 }
 549         } else {
 550                 const unsigned long  __user *u = ubuf;
 551                 while (count >= sizeof(*u) && !ret) {
 552                         unsigned long word;
 553                         ret = __get_user(word, u++);
 554                         if (ret)
 555                                 break;
 556                         ret = putreg(target, pos, word);
 557                         count -= sizeof(*u);
 558                         pos += sizeof(*u);
 559                 }
 560         }
 561         return ret;
 562 }
 563
 564 static void ptrace_triggered(struct perf_event *bp,
 565                              struct perf_sample_data *data,
 566                              struct pt_regs *regs)
 567 {
 568         int i;
 569         struct thread_struct *thread = &(current->thread);
 570
 571         /*
 572          * Store in the virtual DR6 register the fact that the breakpoint
 573          * was hit so the thread's debugger will see it.
 574          */
 575         for (i = 0; i < HBP_NUM; i++) {
 576                 if (thread->ptrace_bps[i] == bp)
 577                         break;
 578         }
 579
 580         thread->debugreg6 |= (DR_TRAP0 << i);
 581 }
 582
 583 /*
 584  * Walk through every ptrace breakpoints for this thread and
 585  * build the dr7 value on top of their attributes.
 586  *
 587  */
 588 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 589 {
 590         int i;
 591         int dr7 = 0;
 592         struct arch_hw_breakpoint *info;
 593
 594         for (i = 0; i < HBP_NUM; i++) {
 595                 if (bp[i] && !bp[i]->attr.disabled) {
 596                         info = counter_arch_bp(bp[i]);
 597                         dr7 |= encode_dr7(i, info->len, info->type);
 598                 }
 599         }
 600
 601         return dr7;
 602 }
 603
 604 static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
 605                                         int len, int type, bool disabled)
 606 {
 607         int err, bp_len, bp_type;
 608
 609         err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
 610         if (!err) {
 611                 attr->bp_len = bp_len;
 612                 attr->bp_type = bp_type;
 613                 attr->disabled = disabled;
 614         }
 615
 616         return err;
 617 }
 618
 619 static struct perf_event *
 620 ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
 621                                 unsigned long addr, bool disabled)
 622 {
 623         struct perf_event_attr attr;
 624         int err;
 625
 626         ptrace_breakpoint_init(&attr);
 627         attr.bp_addr = addr;
 628
 629         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
 630         if (err)
 631                 return ERR_PTR(err);
 632
 633         return register_user_hw_breakpoint(&attr, ptrace_triggered,
 634                                                  NULL, tsk);
 635 }
 636
 637 static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 638                                         int disabled)
 639 {
 640         struct perf_event_attr attr = bp->attr;
 641         int err;
 642
 643         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
 644         if (err)
 645                 return err;
 646
 647         return modify_user_hw_breakpoint(bp, &attr);
 648 }
 649
 650 /*
 651  * Handle ptrace writes to debug register 7.
 652  */
 653 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 654 {
 655         struct thread_struct *thread = &tsk->thread;
 656         unsigned long old_dr7;
 657         bool second_pass = false;
 658         int i, rc, ret = 0;
 659
 660         data &= ~DR_CONTROL_RESERVED;
 661         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 662
 663 restore:
 664         rc = 0;
 665         for (i = 0; i < HBP_NUM; i++) {
 666                 unsigned len, type;
 667                 bool disabled = !decode_dr7(data, i, &len, &type);
 668                 struct perf_event *bp = thread->ptrace_bps[i];
 669
 670                 if (!bp) {
 671                         if (disabled)
 672                                 continue;
 673
 674                         bp = ptrace_register_breakpoint(tsk,
 675                                         len, type, 0, disabled);
 676                         if (IS_ERR(bp)) {
 677                                 rc = PTR_ERR(bp);
 678                                 break;
 679                         }
 680
 681                         thread->ptrace_bps[i] = bp;
 682                         continue;
 683                 }
 684
 685                 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
 686                 if (rc)
 687                         break;
 688         }
 689
 690         /* Restore if the first pass failed, second_pass shouldn't fail. */
 691         if (rc && !WARN_ON(second_pass)) {
 692                 ret = rc;
 693                 data = old_dr7;
 694                 second_pass = true;
 695                 goto restore;
 696         }
 697
 698         return ret;
 699 }
 700
 701 /*
 702  * Handle PTRACE_PEEKUSR calls for the debug register area.
 703  */
 704 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 705 {
 706         struct thread_struct *thread = &tsk->thread;
 707         unsigned long val = 0;
 708
 709         if (n < HBP_NUM) {
 710                 struct perf_event *bp = thread->ptrace_bps[n];
 711
 712                 if (bp)
 713                         val = bp->hw.info.address;
 714         } else if (n == 6) {
 715                 val = thread->debugreg6;
 716         } else if (n == 7) {
 717                 val = thread->ptrace_dr7;
 718         }
 719         return val;
 720 }
 721
 722 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 723                                       unsigned long addr)
 724 {
 725         struct thread_struct *t = &tsk->thread;
 726         struct perf_event *bp = t->ptrace_bps[nr];
 727         int err = 0;
 728
 729         if (!bp) {
 730                 /*
 731                  * Put stub len and type to create an inactive but correct bp.
 732                  *
 733                  * CHECKME: the previous code returned -EIO if the addr wasn't
 734                  * a valid task virtual addr. The new one will return -EINVAL in
 735                  *  this case.
 736                  * -EINVAL may be what we want for in-kernel breakpoints users,
 737                  * but -EIO looks better for ptrace, since we refuse a register
 738                  * writing for the user. And anyway this is the previous
 739                  * behaviour.
 740                  */
 741                 bp = ptrace_register_breakpoint(tsk,
 742                                 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
 743                                 addr, true);
 744                 if (IS_ERR(bp))
 745                         err = PTR_ERR(bp);
 746                 else
 747                         t->ptrace_bps[nr] = bp;
 748         } else {
 749                 struct perf_event_attr attr = bp->attr;
 750
 751                 attr.bp_addr = addr;
 752                 err = modify_user_hw_breakpoint(bp, &attr);
 753         }
 754
 755         return err;
 756 }
 757
 758 /*
 759  * Handle PTRACE_POKEUSR calls for the debug register area.
 760  */
 761 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
 762                                unsigned long val)
 763 {
 764         struct thread_struct *thread = &tsk->thread;
 765         /* There are no DR4 or DR5 registers */
 766         int rc = -EIO;
 767
 768         if (n < HBP_NUM) {
 769                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 770         } else if (n == 6) {
 771                 thread->debugreg6 = val;
 772                 rc = 0;
 773         } else if (n == 7) {
 774                 rc = ptrace_write_dr7(tsk, val);
 775                 if (!rc)
 776                         thread->ptrace_dr7 = val;
 777         }
 778         return rc;
 779 }
 780
 781 /*
 782  * These access the current or another (stopped) task's io permission
 783  * bitmap for debugging or core dump.
 784  */
 785 static int ioperm_active(struct task_struct *target,
 786                          const struct user_regset *regset)
 787 {
 788         return target->thread.io_bitmap_max / regset->size;
 789 }
 790
 791 static int ioperm_get(struct task_struct *target,
 792                       const struct user_regset *regset,
 793                       unsigned int pos, unsigned int count,
 794                       void *kbuf, void __user *ubuf)
 795 {
 796         if (!target->thread.io_bitmap_ptr)
 797                 return -ENXIO;
 798
 799         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 800                                    target->thread.io_bitmap_ptr,
 801                                    0, IO_BITMAP_BYTES);
 802 }
 803
 804 /*
 805  * Called by kernel/ptrace.c when detaching..
 806  *
 807  * Make sure the single step bit is not set.
 808  */
 809 void ptrace_disable(struct task_struct *child)
 810 {
 811         user_disable_single_step(child);
 812 #ifdef TIF_SYSCALL_EMU
 813         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 814 #endif
 815 }
 816
 817 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 818 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 819 #endif
 820
 821 long arch_ptrace(struct task_struct *child, long request,
 822                  unsigned long addr, unsigned long data)
 823 {
 824         int ret;
 825         unsigned long __user *datap = (unsigned long __user *)data;
 826
 827         switch (request) {
 828         /* read the word at location addr in the USER area. */
 829         case PTRACE_PEEKUSR: {
 830                 unsigned long tmp;
 831
 832                 ret = -EIO;
 833                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 834                         break;
 835
 836                 tmp = 0;  /* Default return condition */
 837                 if (addr < sizeof(struct user_regs_struct))
 838                         tmp = getreg(child, addr);
 839                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 840                          addr <= offsetof(struct user, u_debugreg[7])) {
 841                         addr -= offsetof(struct user, u_debugreg[0]);
 842                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 843                 }
 844                 ret = put_user(tmp, datap);
 845                 break;
 846         }
 847
 848         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 849                 ret = -EIO;
 850                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 851                         break;
 852
 853                 if (addr < sizeof(struct user_regs_struct))
 854                         ret = putreg(child, addr, data);
 855                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 856                          addr <= offsetof(struct user, u_debugreg[7])) {
 857                         addr -= offsetof(struct user, u_debugreg[0]);
 858                         ret = ptrace_set_debugreg(child,
 859                                                   addr / sizeof(data), data);
 860                 }
 861                 break;
 862
 863         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 864                 return copy_regset_to_user(child,
 865                                            task_user_regset_view(current),
 866                                            REGSET_GENERAL,
 867                                            0, sizeof(struct user_regs_struct),
 868                                            datap);
 869
 870         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 871                 return copy_regset_from_user(child,
 872                                              task_user_regset_view(current),
 873                                              REGSET_GENERAL,
 874                                              0, sizeof(struct user_regs_struct),
 875                                              datap);
 876
 877         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 878                 return copy_regset_to_user(child,
 879                                            task_user_regset_view(current),
 880                                            REGSET_FP,
 881                                            0, sizeof(struct user_i387_struct),
 882                                            datap);
 883
 884         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 885                 return copy_regset_from_user(child,
 886                                              task_user_regset_view(current),
 887                                              REGSET_FP,
 888                                              0, sizeof(struct user_i387_struct),
 889                                              datap);
 890
 891 #ifdef CONFIG_X86_32
 892         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 893                 return copy_regset_to_user(child, &user_x86_32_view,
 894                                            REGSET_XFP,
 895                                            0, sizeof(struct user_fxsr_struct),
 896                                            datap) ? -EIO : 0;
 897
 898         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 899                 return copy_regset_from_user(child, &user_x86_32_view,
 900                                              REGSET_XFP,
 901                                              0, sizeof(struct user_fxsr_struct),
 902                                              datap) ? -EIO : 0;
 903 #endif
 904
 905 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 906         case PTRACE_GET_THREAD_AREA:
 907                 if ((int) addr < 0)
 908                         return -EIO;
 909                 ret = do_get_thread_area(child, addr,
 910                                         (struct user_desc __user *)data);
 911                 break;
 912
 913         case PTRACE_SET_THREAD_AREA:
 914                 if ((int) addr < 0)
 915                         return -EIO;
 916                 ret = do_set_thread_area(child, addr,
 917                                         (struct user_desc __user *)data, 0);
 918                 break;
 919 #endif
 920
 921 #ifdef CONFIG_X86_64
 922                 /* normal 64bit interface to access TLS data.
 923                    Works just like arch_prctl, except that the arguments
 924                    are reversed. */
 925         case PTRACE_ARCH_PRCTL:
 926                 ret = do_arch_prctl(child, data, addr);
 927                 break;
 928 #endif
 929
 930         default:
 931                 ret = ptrace_request(child, request, addr, data);
 932                 break;
 933         }
 934
 935         return ret;
 936 }
 937
 938 #ifdef CONFIG_IA32_EMULATION
 939
 940 #include <linux/compat.h>
 941 #include <linux/syscalls.h>
 942 #include <asm/ia32.h>
 943 #include <asm/user32.h>
 944
 945 #define R32(l,q)                                                        \
 946         case offsetof(struct user32, regs.l):                           \
 947                 regs->q = value; break
 948
 949 #define SEG32(rs)                                                       \
 950         case offsetof(struct user32, regs.rs):                          \
 951                 return set_segment_reg(child,                           \
 952                                        offsetof(struct user_regs_struct, rs), \
 953                                        value);                          \
 954                 break
 955
 956 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 957 {
 958         struct pt_regs *regs = task_pt_regs(child);
 959
 960         switch (regno) {
 961
 962         SEG32(cs);
 963         SEG32(ds);
 964         SEG32(es);
 965         SEG32(fs);
 966         SEG32(gs);
 967         SEG32(ss);
 968
 969         R32(ebx, bx);
 970         R32(ecx, cx);
 971         R32(edx, dx);
 972         R32(edi, di);
 973         R32(esi, si);
 974         R32(ebp, bp);
 975         R32(eax, ax);
 976         R32(eip, ip);
 977         R32(esp, sp);
 978
 979         case offsetof(struct user32, regs.orig_eax):
 980                 /*
 981                  * A 32-bit debugger setting orig_eax means to restore
 982                  * the state of the task restarting a 32-bit syscall.
 983                  * Make sure we interpret the -ERESTART* codes correctly
 984                  * in case the task is not actually still sitting at the
 985                  * exit from a 32-bit syscall with TS_COMPAT still set.
 986                  */
 987                 regs->orig_ax = value;
 988                 if (syscall_get_nr(child, regs) >= 0)
 989                         task_thread_info(child)->status |= TS_COMPAT;
 990                 break;
 991
 992         case offsetof(struct user32, regs.eflags):
 993                 return set_flags(child, value);
 994
 995         case offsetof(struct user32, u_debugreg[0]) ...
 996                 offsetof(struct user32, u_debugreg[7]):
 997                 regno -= offsetof(struct user32, u_debugreg[0]);
 998                 return ptrace_set_debugreg(child, regno / 4, value);
 999
1000         default:
1001                 if (regno > sizeof(struct user32) || (regno & 3))
1002                         return -EIO;
1003
1004                 /*
1005                  * Other dummy fields in the virtual user structure
1006                  * are ignored
1007                  */
1008                 break;
1009         }
1010         return 0;
1011 }
1012
1013 #undef R32
1014 #undef SEG32
1015
1016 #define R32(l,q)                                                        \
1017         case offsetof(struct user32, regs.l):                           \
1018                 *val = regs->q; break
1019
1020 #define SEG32(rs)                                                       \
1021         case offsetof(struct user32, regs.rs):                          \
1022                 *val = get_segment_reg(child,                           \
1023                                        offsetof(struct user_regs_struct, rs)); \
1024                 break
1025
1026 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1027 {
1028         struct pt_regs *regs = task_pt_regs(child);
1029
1030         switch (regno) {
1031
1032         SEG32(ds);
1033         SEG32(es);
1034         SEG32(fs);
1035         SEG32(gs);
1036
1037         R32(cs, cs);
1038         R32(ss, ss);
1039         R32(ebx, bx);
1040         R32(ecx, cx);
1041         R32(edx, dx);
1042         R32(edi, di);
1043         R32(esi, si);
1044         R32(ebp, bp);
1045         R32(eax, ax);
1046         R32(orig_eax, orig_ax);
1047         R32(eip, ip);
1048         R32(esp, sp);
1049
1050         case offsetof(struct user32, regs.eflags):
1051                 *val = get_flags(child);
1052                 break;
1053
1054         case offsetof(struct user32, u_debugreg[0]) ...
1055                 offsetof(struct user32, u_debugreg[7]):
1056                 regno -= offsetof(struct user32, u_debugreg[0]);
1057                 *val = ptrace_get_debugreg(child, regno / 4);
1058                 break;
1059
1060         default:
1061                 if (regno > sizeof(struct user32) || (regno & 3))
1062                         return -EIO;
1063
1064                 /*
1065                  * Other dummy fields in the virtual user structure
1066                  * are ignored
1067                  */
1068                 *val = 0;
1069                 break;
1070         }
1071         return 0;
1072 }
1073
1074 #undef R32
1075 #undef SEG32
1076
1077 static int genregs32_get(struct task_struct *target,
1078                          const struct user_regset *regset,
1079                          unsigned int pos, unsigned int count,
1080                          void *kbuf, void __user *ubuf)
1081 {
1082         if (kbuf) {
1083                 compat_ulong_t *k = kbuf;
1084                 while (count >= sizeof(*k)) {
1085                         getreg32(target, pos, k++);
1086                         count -= sizeof(*k);
1087                         pos += sizeof(*k);
1088                 }
1089         } else {
1090                 compat_ulong_t __user *u = ubuf;
1091                 while (count >= sizeof(*u)) {
1092                         compat_ulong_t word;
1093                         getreg32(target, pos, &word);
1094                         if (__put_user(word, u++))
1095                                 return -EFAULT;
1096                         count -= sizeof(*u);
1097                         pos += sizeof(*u);
1098                 }
1099         }
1100
1101         return 0;
1102 }
1103
1104 static int genregs32_set(struct task_struct *target,
1105                          const struct user_regset *regset,
1106                          unsigned int pos, unsigned int count,
1107                          const void *kbuf, const void __user *ubuf)
1108 {
1109         int ret = 0;
1110         if (kbuf) {
1111                 const compat_ulong_t *k = kbuf;
1112                 while (count >= sizeof(*k) && !ret) {
1113                         ret = putreg32(target, pos, *k++);
1114                         count -= sizeof(*k);
1115                         pos += sizeof(*k);
1116                 }
1117         } else {
1118                 const compat_ulong_t __user *u = ubuf;
1119                 while (count >= sizeof(*u) && !ret) {
1120                         compat_ulong_t word;
1121                         ret = __get_user(word, u++);
1122                         if (ret)
1123                                 break;
1124                         ret = putreg32(target, pos, word);
1125                         count -= sizeof(*u);
1126                         pos += sizeof(*u);
1127                 }
1128         }
1129         return ret;
1130 }
1131
1132 #ifdef CONFIG_X86_X32_ABI
1133 static long x32_arch_ptrace(struct task_struct *child,
1134                             compat_long_t request, compat_ulong_t caddr,
1135                             compat_ulong_t cdata)
1136 {
1137         unsigned long addr = caddr;
1138         unsigned long data = cdata;
1139         void __user *datap = compat_ptr(data);
1140         int ret;
1141
1142         switch (request) {
1143         /* Read 32bits at location addr in the USER area.  Only allow
1144            to return the lower 32bits of segment and debug registers.  */
1145         case PTRACE_PEEKUSR: {
1146                 u32 tmp;
1147
1148                 ret = -EIO;
1149                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1150                     addr < offsetof(struct user_regs_struct, cs))
1151                         break;
1152
1153                 tmp = 0;  /* Default return condition */
1154                 if (addr < sizeof(struct user_regs_struct))
1155                         tmp = getreg(child, addr);
1156                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1157                          addr <= offsetof(struct user, u_debugreg[7])) {
1158                         addr -= offsetof(struct user, u_debugreg[0]);
1159                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1160                 }
1161                 ret = put_user(tmp, (__u32 __user *)datap);
1162                 break;
1163         }
1164
1165         /* Write the word at location addr in the USER area.  Only allow
1166            to update segment and debug registers with the upper 32bits
1167            zero-extended. */
1168         case PTRACE_POKEUSR:
1169                 ret = -EIO;
1170                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1171                     addr < offsetof(struct user_regs_struct, cs))
1172                         break;
1173
1174                 if (addr < sizeof(struct user_regs_struct))
1175                         ret = putreg(child, addr, data);
1176                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1177                          addr <= offsetof(struct user, u_debugreg[7])) {
1178                         addr -= offsetof(struct user, u_debugreg[0]);
1179                         ret = ptrace_set_debugreg(child,
1180                                                   addr / sizeof(data), data);
1181                 }
1182                 break;
1183
1184         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1185                 return copy_regset_to_user(child,
1186                                            task_user_regset_view(current),
1187                                            REGSET_GENERAL,
1188                                            0, sizeof(struct user_regs_struct),
1189                                            datap);
1190
1191         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1192                 return copy_regset_from_user(child,
1193                                              task_user_regset_view(current),
1194                                              REGSET_GENERAL,
1195                                              0, sizeof(struct user_regs_struct),
1196                                              datap);
1197
1198         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1199                 return copy_regset_to_user(child,
1200                                            task_user_regset_view(current),
1201                                            REGSET_FP,
1202                                            0, sizeof(struct user_i387_struct),
1203                                            datap);
1204
1205         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1206                 return copy_regset_from_user(child,
1207                                              task_user_regset_view(current),
1208                                              REGSET_FP,
1209                                              0, sizeof(struct user_i387_struct),
1210                                              datap);
1211
1212         default:
1213                 return compat_ptrace_request(child, request, addr, data);
1214         }
1215
1216         return ret;
1217 }
1218 #endif
1219
1220 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1221                         compat_ulong_t caddr, compat_ulong_t cdata)
1222 {
1223         unsigned long addr = caddr;
1224         unsigned long data = cdata;
1225         void __user *datap = compat_ptr(data);
1226         int ret;
1227         __u32 val;
1228
1229 #ifdef CONFIG_X86_X32_ABI
1230         if (!is_ia32_task())
1231                 return x32_arch_ptrace(child, request, caddr, cdata);
1232 #endif
1233
1234         switch (request) {
1235         case PTRACE_PEEKUSR:
1236                 ret = getreg32(child, addr, &val);
1237                 if (ret == 0)
1238                         ret = put_user(val, (__u32 __user *)datap);
1239                 break;
1240
1241         case PTRACE_POKEUSR:
1242                 ret = putreg32(child, addr, data);
1243                 break;
1244
1245         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1246                 return copy_regset_to_user(child, &user_x86_32_view,
1247                                            REGSET_GENERAL,
1248                                            0, sizeof(struct user_regs_struct32),
1249                                            datap);
1250
1251         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1252                 return copy_regset_from_user(child, &user_x86_32_view,
1253                                              REGSET_GENERAL, 0,
1254                                              sizeof(struct user_regs_struct32),
1255                                              datap);
1256
1257         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1258                 return copy_regset_to_user(child, &user_x86_32_view,
1259                                            REGSET_FP, 0,
1260                                            sizeof(struct user_i387_ia32_struct),
1261                                            datap);
1262
1263         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1264                 return copy_regset_from_user(
1265                         child, &user_x86_32_view, REGSET_FP,
1266                         0, sizeof(struct user_i387_ia32_struct), datap);
1267
1268         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1269                 return copy_regset_to_user(child, &user_x86_32_view,
1270                                            REGSET_XFP, 0,
1271                                            sizeof(struct user32_fxsr_struct),
1272                                            datap);
1273
1274         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1275                 return copy_regset_from_user(child, &user_x86_32_view,
1276                                              REGSET_XFP, 0,
1277                                              sizeof(struct user32_fxsr_struct),
1278                                              datap);
1279
1280         case PTRACE_GET_THREAD_AREA:
1281         case PTRACE_SET_THREAD_AREA:
1282                 return arch_ptrace(child, request, addr, data);
1283
1284         default:
1285                 return compat_ptrace_request(child, request, addr, data);
1286         }
1287
1288         return ret;
1289 }
1290
1291 #endif  /* CONFIG_IA32_EMULATION */
1292
1293 #ifdef CONFIG_X86_64
1294
1295 static struct user_regset x86_64_regsets[] __read_mostly = {
1296         [REGSET_GENERAL] = {
1297                 .core_note_type = NT_PRSTATUS,
1298                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1299                 .size = sizeof(long), .align = sizeof(long),
1300                 .get = genregs_get, .set = genregs_set
1301         },
1302         [REGSET_FP] = {
1303                 .core_note_type = NT_PRFPREG,
1304                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1305                 .size = sizeof(long), .align = sizeof(long),
1306                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1307         },
1308         [REGSET_XSTATE] = {
1309                 .core_note_type = NT_X86_XSTATE,
1310                 .size = sizeof(u64), .align = sizeof(u64),
1311                 .active = xstateregs_active, .get = xstateregs_get,
1312                 .set = xstateregs_set
1313         },
1314         [REGSET_IOPERM64] = {
1315                 .core_note_type = NT_386_IOPERM,
1316                 .n = IO_BITMAP_LONGS,
1317                 .size = sizeof(long), .align = sizeof(long),
1318                 .active = ioperm_active, .get = ioperm_get
1319         },
1320 };
1321
1322 static const struct user_regset_view user_x86_64_view = {
1323         .name = "x86_64", .e_machine = EM_X86_64,
1324         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1325 };
1326
1327 #else  /* CONFIG_X86_32 */
1328
1329 #define user_regs_struct32      user_regs_struct
1330 #define genregs32_get           genregs_get
1331 #define genregs32_set           genregs_set
1332
1333 #endif  /* CONFIG_X86_64 */
1334
1335 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1336 static struct user_regset x86_32_regsets[] __read_mostly = {
1337         [REGSET_GENERAL] = {
1338                 .core_note_type = NT_PRSTATUS,
1339                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1340                 .size = sizeof(u32), .align = sizeof(u32),
1341                 .get = genregs32_get, .set = genregs32_set
1342         },
1343         [REGSET_FP] = {
1344                 .core_note_type = NT_PRFPREG,
1345                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1346                 .size = sizeof(u32), .align = sizeof(u32),
1347                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1348         },
1349         [REGSET_XFP] = {
1350                 .core_note_type = NT_PRXFPREG,
1351                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1352                 .size = sizeof(u32), .align = sizeof(u32),
1353                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1354         },
1355         [REGSET_XSTATE] = {
1356                 .core_note_type = NT_X86_XSTATE,
1357                 .size = sizeof(u64), .align = sizeof(u64),
1358                 .active = xstateregs_active, .get = xstateregs_get,
1359                 .set = xstateregs_set
1360         },
1361         [REGSET_TLS] = {
1362                 .core_note_type = NT_386_TLS,
1363                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1364                 .size = sizeof(struct user_desc),
1365                 .align = sizeof(struct user_desc),
1366                 .active = regset_tls_active,
1367                 .get = regset_tls_get, .set = regset_tls_set
1368         },
1369         [REGSET_IOPERM32] = {
1370                 .core_note_type = NT_386_IOPERM,
1371                 .n = IO_BITMAP_BYTES / sizeof(u32),
1372                 .size = sizeof(u32), .align = sizeof(u32),
1373                 .active = ioperm_active, .get = ioperm_get
1374         },
1375 };
1376
1377 static const struct user_regset_view user_x86_32_view = {
1378         .name = "i386", .e_machine = EM_386,
1379         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1380 };
1381 #endif
1382
1383 /*
1384  * This represents bytes 464..511 in the memory layout exported through
1385  * the REGSET_XSTATE interface.
1386  */
1387 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1388
1389 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1390 {
1391 #ifdef CONFIG_X86_64
1392         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1393 #endif
1394 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1395         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1396 #endif
1397         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1398 }
1399
1400 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1401 {
1402 #ifdef CONFIG_IA32_EMULATION
1403         if (test_tsk_thread_flag(task, TIF_IA32))
1404 #endif
1405 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1406                 return &user_x86_32_view;
1407 #endif
1408 #ifdef CONFIG_X86_64
1409         return &user_x86_64_view;
1410 #endif
1411 }
1412
1413 static void fill_sigtrap_info(struct task_struct *tsk,
1414                                 struct pt_regs *regs,
1415                                 int error_code, int si_code,
1416                                 struct siginfo *info)
1417 {
1418         tsk->thread.trap_nr = X86_TRAP_DB;
1419         tsk->thread.error_code = error_code;
1420
1421         memset(info, 0, sizeof(*info));
1422         info->si_signo = SIGTRAP;
1423         info->si_code = si_code;
1424         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1425 }
1426
1427 void user_single_step_siginfo(struct task_struct *tsk,
1428                                 struct pt_regs *regs,
1429                                 struct siginfo *info)
1430 {
1431         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1432 }
1433
1434 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1435                                          int error_code, int si_code)
1436 {
1437         struct siginfo info;
1438
1439         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1440         /* Send us the fake SIGTRAP */
1441         force_sig_info(SIGTRAP, &info, tsk);
1442 }
1443
1444 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
1445 {
1446 #ifdef CONFIG_X86_64
1447         if (arch == AUDIT_ARCH_X86_64) {
1448                 audit_syscall_entry(arch, regs->orig_ax, regs->di,
1449                                     regs->si, regs->dx, regs->r10);
1450         } else
1451 #endif
1452         {
1453                 audit_syscall_entry(arch, regs->orig_ax, regs->bx,
1454                                     regs->cx, regs->dx, regs->si);
1455         }
1456 }
1457
1458 /*
1459  * We can return 0 to resume the syscall or anything else to go to phase
1460  * 2.  If we resume the syscall, we need to put something appropriate in
1461  * regs->orig_ax.
1462  *
1463  * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
1464  * are fully functional.
1465  *
1466  * For phase 2's benefit, our return value is:
1467  * 0:                   resume the syscall
1468  * 1:                   go to phase 2; no seccomp phase 2 needed
1469  * anything else:       go to phase 2; pass return value to seccomp
1470  */
1471 unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
1472 {
1473         unsigned long ret = 0;
1474         u32 work;
1475
1476         BUG_ON(regs != task_pt_regs(current));
1477
1478         work = ACCESS_ONCE(current_thread_info()->flags) &
1479                 _TIF_WORK_SYSCALL_ENTRY;
1480
1481         /*
1482          * If TIF_NOHZ is set, we are required to call user_exit() before
1483          * doing anything that could touch RCU.
1484          */
1485         if (work & _TIF_NOHZ) {
1486                 user_exit();
1487                 work &= ~TIF_NOHZ;
1488         }
1489
1490 #ifdef CONFIG_SECCOMP
1491         /*
1492          * Do seccomp first -- it should minimize exposure of other
1493          * code, and keeping seccomp fast is probably more valuable
1494          * than the rest of this.
1495          */
1496         if (work & _TIF_SECCOMP) {
1497                 struct seccomp_data sd;
1498
1499                 sd.arch = arch;
1500                 sd.nr = regs->orig_ax;
1501                 sd.instruction_pointer = regs->ip;
1502 #ifdef CONFIG_X86_64
1503                 if (arch == AUDIT_ARCH_X86_64) {
1504                         sd.args[0] = regs->di;
1505                         sd.args[1] = regs->si;
1506                         sd.args[2] = regs->dx;
1507                         sd.args[3] = regs->r10;
1508                         sd.args[4] = regs->r8;
1509                         sd.args[5] = regs->r9;
1510                 } else
1511 #endif
1512                 {
1513                         sd.args[0] = regs->bx;
1514                         sd.args[1] = regs->cx;
1515                         sd.args[2] = regs->dx;
1516                         sd.args[3] = regs->si;
1517                         sd.args[4] = regs->di;
1518                         sd.args[5] = regs->bp;
1519                 }
1520
1521                 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
1522                 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
1523
1524                 ret = seccomp_phase1(&sd);
1525                 if (ret == SECCOMP_PHASE1_SKIP) {
1526                         regs->orig_ax = -1;
1527                         ret = 0;
1528                 } else if (ret != SECCOMP_PHASE1_OK) {
1529                         return ret;  /* Go directly to phase 2 */
1530                 }
1531
1532                 work &= ~_TIF_SECCOMP;
1533         }
1534 #endif
1535
1536         /* Do our best to finish without phase 2. */
1537         if (work == 0)
1538                 return ret;  /* seccomp and/or nohz only (ret == 0 here) */
1539
1540 #ifdef CONFIG_AUDITSYSCALL
1541         if (work == _TIF_SYSCALL_AUDIT) {
1542                 /*
1543                  * If there is no more work to be done except auditing,
1544                  * then audit in phase 1.  Phase 2 always audits, so, if
1545                  * we audit here, then we can't go on to phase 2.
1546                  */
1547                 do_audit_syscall_entry(regs, arch);
1548                 return 0;
1549         }
1550 #endif
1551
1552         return 1;  /* Something is enabled that we can't handle in phase 1 */
1553 }
1554
1555 /* Returns the syscall nr to run (which should match regs->orig_ax). */
1556 long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
1557                                 unsigned long phase1_result)
1558 {
1559         long ret = 0;
1560         u32 work = ACCESS_ONCE(current_thread_info()->flags) &
1561                 _TIF_WORK_SYSCALL_ENTRY;
1562
1563         BUG_ON(regs != task_pt_regs(current));
1564
1565         /*
1566          * If we stepped into a sysenter/syscall insn, it trapped in
1567          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1568          * If user-mode had set TF itself, then it's still clear from
1569          * do_debug() and we need to set it again to restore the user
1570          * state.  If we entered on the slow path, TF was already set.
1571          */
1572         if (work & _TIF_SINGLESTEP)
1573                 regs->flags |= X86_EFLAGS_TF;
1574
1575 #ifdef CONFIG_SECCOMP
1576         /*
1577          * Call seccomp_phase2 before running the other hooks so that
1578          * they can see any changes made by a seccomp tracer.
1579          */
1580         if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
1581                 /* seccomp failures shouldn't expose any additional code. */
1582                 return -1;
1583         }
1584 #endif
1585
1586         if (unlikely(work & _TIF_SYSCALL_EMU))
1587                 ret = -1L;
1588
1589         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1590             tracehook_report_syscall_entry(regs))
1591                 ret = -1L;
1592
1593         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1594                 trace_sys_enter(regs, regs->orig_ax);
1595
1596         do_audit_syscall_entry(regs, arch);
1597
1598         return ret ?: regs->orig_ax;
1599 }
1600
1601 long syscall_trace_enter(struct pt_regs *regs)
1602 {
1603         u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
1604         unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
1605
1606         if (phase1_result == 0)
1607                 return regs->orig_ax;
1608         else
1609                 return syscall_trace_enter_phase2(regs, arch, phase1_result);
1610 }
1611
1612 void syscall_trace_leave(struct pt_regs *regs)
1613 {
1614         bool step;
1615
1616         /*
1617          * We may come here right after calling schedule_user()
1618          * or do_notify_resume(), in which case we can be in RCU
1619          * user mode.
1620          */
1621         user_exit();
1622
1623         audit_syscall_exit(regs);
1624
1625         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1626                 trace_sys_exit(regs, regs->ax);
1627
1628         /*
1629          * If TIF_SYSCALL_EMU is set, we only get here because of
1630          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1631          * We already reported this syscall instruction in
1632          * syscall_trace_enter().
1633          */
1634         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1635                         !test_thread_flag(TIF_SYSCALL_EMU);
1636         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1637                 tracehook_report_syscall_exit(regs, step);
1638
1639         user_enter();
1640 }