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