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
  25 #include <asm/uaccess.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/system.h>
  28 #include <asm/processor.h>
  29 #include <asm/i387.h>
  30 #include <asm/debugreg.h>
  31 #include <asm/ldt.h>
  32 #include <asm/desc.h>
  33 #include <asm/prctl.h>
  34 #include <asm/proto.h>
  35 #include <asm/hw_breakpoint.h>
  36
  37 #include "tls.h"
  38
  39 #define CREATE_TRACE_POINTS
  40 #include <trace/events/syscalls.h>
  41
  42 enum x86_regset {
  43         REGSET_GENERAL,
  44         REGSET_FP,
  45         REGSET_XFP,
  46         REGSET_IOPERM64 = REGSET_XFP,
  47         REGSET_XSTATE,
  48         REGSET_TLS,
  49         REGSET_IOPERM32,
  50 };
  51
  52 struct pt_regs_offset {
  53         const char *name;
  54         int offset;
  55 };
  56
  57 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  58 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  59
  60 static const struct pt_regs_offset regoffset_table[] = {
  61 #ifdef CONFIG_X86_64
  62         REG_OFFSET_NAME(r15),
  63         REG_OFFSET_NAME(r14),
  64         REG_OFFSET_NAME(r13),
  65         REG_OFFSET_NAME(r12),
  66         REG_OFFSET_NAME(r11),
  67         REG_OFFSET_NAME(r10),
  68         REG_OFFSET_NAME(r9),
  69         REG_OFFSET_NAME(r8),
  70 #endif
  71         REG_OFFSET_NAME(bx),
  72         REG_OFFSET_NAME(cx),
  73         REG_OFFSET_NAME(dx),
  74         REG_OFFSET_NAME(si),
  75         REG_OFFSET_NAME(di),
  76         REG_OFFSET_NAME(bp),
  77         REG_OFFSET_NAME(ax),
  78 #ifdef CONFIG_X86_32
  79         REG_OFFSET_NAME(ds),
  80         REG_OFFSET_NAME(es),
  81         REG_OFFSET_NAME(fs),
  82         REG_OFFSET_NAME(gs),
  83 #endif
  84         REG_OFFSET_NAME(orig_ax),
  85         REG_OFFSET_NAME(ip),
  86         REG_OFFSET_NAME(cs),
  87         REG_OFFSET_NAME(flags),
  88         REG_OFFSET_NAME(sp),
  89         REG_OFFSET_NAME(ss),
  90         REG_OFFSET_END,
  91 };
  92
  93 /**
  94  * regs_query_register_offset() - query register offset from its name
  95  * @name:       the name of a register
  96  *
  97  * regs_query_register_offset() returns the offset of a register in struct
  98  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
  99  */
 100 int regs_query_register_offset(const char *name)
 101 {
 102         const struct pt_regs_offset *roff;
 103         for (roff = regoffset_table; roff->name != NULL; roff++)
 104                 if (!strcmp(roff->name, name))
 105                         return roff->offset;
 106         return -EINVAL;
 107 }
 108
 109 /**
 110  * regs_query_register_name() - query register name from its offset
 111  * @offset:     the offset of a register in struct pt_regs.
 112  *
 113  * regs_query_register_name() returns the name of a register from its
 114  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 115  */
 116 const char *regs_query_register_name(unsigned int offset)
 117 {
 118         const struct pt_regs_offset *roff;
 119         for (roff = regoffset_table; roff->name != NULL; roff++)
 120                 if (roff->offset == offset)
 121                         return roff->name;
 122         return NULL;
 123 }
 124
 125 static const int arg_offs_table[] = {
 126 #ifdef CONFIG_X86_32
 127         [0] = offsetof(struct pt_regs, ax),
 128         [1] = offsetof(struct pt_regs, dx),
 129         [2] = offsetof(struct pt_regs, cx)
 130 #else /* CONFIG_X86_64 */
 131         [0] = offsetof(struct pt_regs, di),
 132         [1] = offsetof(struct pt_regs, si),
 133         [2] = offsetof(struct pt_regs, dx),
 134         [3] = offsetof(struct pt_regs, cx),
 135         [4] = offsetof(struct pt_regs, r8),
 136         [5] = offsetof(struct pt_regs, r9)
 137 #endif
 138 };
 139
 140 /*
 141  * does not yet catch signals sent when the child dies.
 142  * in exit.c or in signal.c.
 143  */
 144
 145 /*
 146  * Determines which flags the user has access to [1 = access, 0 = no access].
 147  */
 148 #define FLAG_MASK_32            ((unsigned long)                        \
 149                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 150                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 151                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 152                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 153                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 154
 155 /*
 156  * Determines whether a value may be installed in a segment register.
 157  */
 158 static inline bool invalid_selector(u16 value)
 159 {
 160         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 161 }
 162
 163 #ifdef CONFIG_X86_32
 164
 165 #define FLAG_MASK               FLAG_MASK_32
 166
 167 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 168 {
 169         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 170         return &regs->bx + (regno >> 2);
 171 }
 172
 173 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 174 {
 175         /*
 176          * Returning the value truncates it to 16 bits.
 177          */
 178         unsigned int retval;
 179         if (offset != offsetof(struct user_regs_struct, gs))
 180                 retval = *pt_regs_access(task_pt_regs(task), offset);
 181         else {
 182                 if (task == current)
 183                         retval = get_user_gs(task_pt_regs(task));
 184                 else
 185                         retval = task_user_gs(task);
 186         }
 187         return retval;
 188 }
 189
 190 static int set_segment_reg(struct task_struct *task,
 191                            unsigned long offset, u16 value)
 192 {
 193         /*
 194          * The value argument was already truncated to 16 bits.
 195          */
 196         if (invalid_selector(value))
 197                 return -EIO;
 198
 199         /*
 200          * For %cs and %ss we cannot permit a null selector.
 201          * We can permit a bogus selector as long as it has USER_RPL.
 202          * Null selectors are fine for other segment registers, but
 203          * we will never get back to user mode with invalid %cs or %ss
 204          * and will take the trap in iret instead.  Much code relies
 205          * on user_mode() to distinguish a user trap frame (which can
 206          * safely use invalid selectors) from a kernel trap frame.
 207          */
 208         switch (offset) {
 209         case offsetof(struct user_regs_struct, cs):
 210         case offsetof(struct user_regs_struct, ss):
 211                 if (unlikely(value == 0))
 212                         return -EIO;
 213
 214         default:
 215                 *pt_regs_access(task_pt_regs(task), offset) = value;
 216                 break;
 217
 218         case offsetof(struct user_regs_struct, gs):
 219                 if (task == current)
 220                         set_user_gs(task_pt_regs(task), value);
 221                 else
 222                         task_user_gs(task) = value;
 223         }
 224
 225         return 0;
 226 }
 227
 228 #else  /* CONFIG_X86_64 */
 229
 230 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 231
 232 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 233 {
 234         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 235         return &regs->r15 + (offset / sizeof(regs->r15));
 236 }
 237
 238 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 239 {
 240         /*
 241          * Returning the value truncates it to 16 bits.
 242          */
 243         unsigned int seg;
 244
 245         switch (offset) {
 246         case offsetof(struct user_regs_struct, fs):
 247                 if (task == current) {
 248                         /* Older gas can't assemble movq %?s,%r?? */
 249                         asm("movl %%fs,%0" : "=r" (seg));
 250                         return seg;
 251                 }
 252                 return task->thread.fsindex;
 253         case offsetof(struct user_regs_struct, gs):
 254                 if (task == current) {
 255                         asm("movl %%gs,%0" : "=r" (seg));
 256                         return seg;
 257                 }
 258                 return task->thread.gsindex;
 259         case offsetof(struct user_regs_struct, ds):
 260                 if (task == current) {
 261                         asm("movl %%ds,%0" : "=r" (seg));
 262                         return seg;
 263                 }
 264                 return task->thread.ds;
 265         case offsetof(struct user_regs_struct, es):
 266                 if (task == current) {
 267                         asm("movl %%es,%0" : "=r" (seg));
 268                         return seg;
 269                 }
 270                 return task->thread.es;
 271
 272         case offsetof(struct user_regs_struct, cs):
 273         case offsetof(struct user_regs_struct, ss):
 274                 break;
 275         }
 276         return *pt_regs_access(task_pt_regs(task), offset);
 277 }
 278
 279 static int set_segment_reg(struct task_struct *task,
 280                            unsigned long offset, u16 value)
 281 {
 282         /*
 283          * The value argument was already truncated to 16 bits.
 284          */
 285         if (invalid_selector(value))
 286                 return -EIO;
 287
 288         switch (offset) {
 289         case offsetof(struct user_regs_struct,fs):
 290                 /*
 291                  * If this is setting fs as for normal 64-bit use but
 292                  * setting fs_base has implicitly changed it, leave it.
 293                  */
 294                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 295                      task->thread.fs != 0) ||
 296                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 297                      task->thread.fs == 0))
 298                         break;
 299                 task->thread.fsindex = value;
 300                 if (task == current)
 301                         loadsegment(fs, task->thread.fsindex);
 302                 break;
 303         case offsetof(struct user_regs_struct,gs):
 304                 /*
 305                  * If this is setting gs as for normal 64-bit use but
 306                  * setting gs_base has implicitly changed it, leave it.
 307                  */
 308                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 309                      task->thread.gs != 0) ||
 310                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 311                      task->thread.gs == 0))
 312                         break;
 313                 task->thread.gsindex = value;
 314                 if (task == current)
 315                         load_gs_index(task->thread.gsindex);
 316                 break;
 317         case offsetof(struct user_regs_struct,ds):
 318                 task->thread.ds = value;
 319                 if (task == current)
 320                         loadsegment(ds, task->thread.ds);
 321                 break;
 322         case offsetof(struct user_regs_struct,es):
 323                 task->thread.es = value;
 324                 if (task == current)
 325                         loadsegment(es, task->thread.es);
 326                 break;
 327
 328                 /*
 329                  * Can't actually change these in 64-bit mode.
 330                  */
 331         case offsetof(struct user_regs_struct,cs):
 332                 if (unlikely(value == 0))
 333                         return -EIO;
 334 #ifdef CONFIG_IA32_EMULATION
 335                 if (test_tsk_thread_flag(task, TIF_IA32))
 336                         task_pt_regs(task)->cs = value;
 337 #endif
 338                 break;
 339         case offsetof(struct user_regs_struct,ss):
 340                 if (unlikely(value == 0))
 341                         return -EIO;
 342 #ifdef CONFIG_IA32_EMULATION
 343                 if (test_tsk_thread_flag(task, TIF_IA32))
 344                         task_pt_regs(task)->ss = value;
 345 #endif
 346                 break;
 347         }
 348
 349         return 0;
 350 }
 351
 352 #endif  /* CONFIG_X86_32 */
 353
 354 static unsigned long get_flags(struct task_struct *task)
 355 {
 356         unsigned long retval = task_pt_regs(task)->flags;
 357
 358         /*
 359          * If the debugger set TF, hide it from the readout.
 360          */
 361         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 362                 retval &= ~X86_EFLAGS_TF;
 363
 364         return retval;
 365 }
 366
 367 static int set_flags(struct task_struct *task, unsigned long value)
 368 {
 369         struct pt_regs *regs = task_pt_regs(task);
 370
 371         /*
 372          * If the user value contains TF, mark that
 373          * it was not "us" (the debugger) that set it.
 374          * If not, make sure it stays set if we had.
 375          */
 376         if (value & X86_EFLAGS_TF)
 377                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 378         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 379                 value |= X86_EFLAGS_TF;
 380
 381         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 382
 383         return 0;
 384 }
 385
 386 static int putreg(struct task_struct *child,
 387                   unsigned long offset, unsigned long value)
 388 {
 389         switch (offset) {
 390         case offsetof(struct user_regs_struct, cs):
 391         case offsetof(struct user_regs_struct, ds):
 392         case offsetof(struct user_regs_struct, es):
 393         case offsetof(struct user_regs_struct, fs):
 394         case offsetof(struct user_regs_struct, gs):
 395         case offsetof(struct user_regs_struct, ss):
 396                 return set_segment_reg(child, offset, value);
 397
 398         case offsetof(struct user_regs_struct, flags):
 399                 return set_flags(child, value);
 400
 401 #ifdef CONFIG_X86_64
 402         case offsetof(struct user_regs_struct,fs_base):
 403                 if (value >= TASK_SIZE_OF(child))
 404                         return -EIO;
 405                 /*
 406                  * When changing the segment base, use do_arch_prctl
 407                  * to set either thread.fs or thread.fsindex and the
 408                  * corresponding GDT slot.
 409                  */
 410                 if (child->thread.fs != value)
 411                         return do_arch_prctl(child, ARCH_SET_FS, value);
 412                 return 0;
 413         case offsetof(struct user_regs_struct,gs_base):
 414                 /*
 415                  * Exactly the same here as the %fs handling above.
 416                  */
 417                 if (value >= TASK_SIZE_OF(child))
 418                         return -EIO;
 419                 if (child->thread.gs != value)
 420                         return do_arch_prctl(child, ARCH_SET_GS, value);
 421                 return 0;
 422 #endif
 423         }
 424
 425         *pt_regs_access(task_pt_regs(child), offset) = value;
 426         return 0;
 427 }
 428
 429 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 430 {
 431         switch (offset) {
 432         case offsetof(struct user_regs_struct, cs):
 433         case offsetof(struct user_regs_struct, ds):
 434         case offsetof(struct user_regs_struct, es):
 435         case offsetof(struct user_regs_struct, fs):
 436         case offsetof(struct user_regs_struct, gs):
 437         case offsetof(struct user_regs_struct, ss):
 438                 return get_segment_reg(task, offset);
 439
 440         case offsetof(struct user_regs_struct, flags):
 441                 return get_flags(task);
 442
 443 #ifdef CONFIG_X86_64
 444         case offsetof(struct user_regs_struct, fs_base): {
 445                 /*
 446                  * do_arch_prctl may have used a GDT slot instead of
 447                  * the MSR.  To userland, it appears the same either
 448                  * way, except the %fs segment selector might not be 0.
 449                  */
 450                 unsigned int seg = task->thread.fsindex;
 451                 if (task->thread.fs != 0)
 452                         return task->thread.fs;
 453                 if (task == current)
 454                         asm("movl %%fs,%0" : "=r" (seg));
 455                 if (seg != FS_TLS_SEL)
 456                         return 0;
 457                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 458         }
 459         case offsetof(struct user_regs_struct, gs_base): {
 460                 /*
 461                  * Exactly the same here as the %fs handling above.
 462                  */
 463                 unsigned int seg = task->thread.gsindex;
 464                 if (task->thread.gs != 0)
 465                         return task->thread.gs;
 466                 if (task == current)
 467                         asm("movl %%gs,%0" : "=r" (seg));
 468                 if (seg != GS_TLS_SEL)
 469                         return 0;
 470                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 471         }
 472 #endif
 473         }
 474
 475         return *pt_regs_access(task_pt_regs(task), offset);
 476 }
 477
 478 static int genregs_get(struct task_struct *target,
 479                        const struct user_regset *regset,
 480                        unsigned int pos, unsigned int count,
 481                        void *kbuf, void __user *ubuf)
 482 {
 483         if (kbuf) {
 484                 unsigned long *k = kbuf;
 485                 while (count >= sizeof(*k)) {
 486                         *k++ = getreg(target, pos);
 487                         count -= sizeof(*k);
 488                         pos += sizeof(*k);
 489                 }
 490         } else {
 491                 unsigned long __user *u = ubuf;
 492                 while (count >= sizeof(*u)) {
 493                         if (__put_user(getreg(target, pos), u++))
 494                                 return -EFAULT;
 495                         count -= sizeof(*u);
 496                         pos += sizeof(*u);
 497                 }
 498         }
 499
 500         return 0;
 501 }
 502
 503 static int genregs_set(struct task_struct *target,
 504                        const struct user_regset *regset,
 505                        unsigned int pos, unsigned int count,
 506                        const void *kbuf, const void __user *ubuf)
 507 {
 508         int ret = 0;
 509         if (kbuf) {
 510                 const unsigned long *k = kbuf;
 511                 while (count >= sizeof(*k) && !ret) {
 512                         ret = putreg(target, pos, *k++);
 513                         count -= sizeof(*k);
 514                         pos += sizeof(*k);
 515                 }
 516         } else {
 517                 const unsigned long  __user *u = ubuf;
 518                 while (count >= sizeof(*u) && !ret) {
 519                         unsigned long word;
 520                         ret = __get_user(word, u++);
 521                         if (ret)
 522                                 break;
 523                         ret = putreg(target, pos, word);
 524                         count -= sizeof(*u);
 525                         pos += sizeof(*u);
 526                 }
 527         }
 528         return ret;
 529 }
 530
 531 static void ptrace_triggered(struct perf_event *bp, int nmi,
 532                              struct perf_sample_data *data,
 533                              struct pt_regs *regs)
 534 {
 535         int i;
 536         struct thread_struct *thread = &(current->thread);
 537
 538         /*
 539          * Store in the virtual DR6 register the fact that the breakpoint
 540          * was hit so the thread's debugger will see it.
 541          */
 542         for (i = 0; i < HBP_NUM; i++) {
 543                 if (thread->ptrace_bps[i] == bp)
 544                         break;
 545         }
 546
 547         thread->debugreg6 |= (DR_TRAP0 << i);
 548 }
 549
 550 /*
 551  * Walk through every ptrace breakpoints for this thread and
 552  * build the dr7 value on top of their attributes.
 553  *
 554  */
 555 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 556 {
 557         int i;
 558         int dr7 = 0;
 559         struct arch_hw_breakpoint *info;
 560
 561         for (i = 0; i < HBP_NUM; i++) {
 562                 if (bp[i] && !bp[i]->attr.disabled) {
 563                         info = counter_arch_bp(bp[i]);
 564                         dr7 |= encode_dr7(i, info->len, info->type);
 565                 }
 566         }
 567
 568         return dr7;
 569 }
 570
 571 static int
 572 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 573                          struct task_struct *tsk, int disabled)
 574 {
 575         int err;
 576         int gen_len, gen_type;
 577         struct perf_event_attr attr;
 578
 579         /*
 580          * We should have at least an inactive breakpoint at this
 581          * slot. It means the user is writing dr7 without having
 582          * written the address register first
 583          */
 584         if (!bp)
 585                 return -EINVAL;
 586
 587         err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
 588         if (err)
 589                 return err;
 590
 591         attr = bp->attr;
 592         attr.bp_len = gen_len;
 593         attr.bp_type = gen_type;
 594         attr.disabled = disabled;
 595
 596         return modify_user_hw_breakpoint(bp, &attr);
 597 }
 598
 599 /*
 600  * Handle ptrace writes to debug register 7.
 601  */
 602 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 603 {
 604         struct thread_struct *thread = &(tsk->thread);
 605         unsigned long old_dr7;
 606         int i, orig_ret = 0, rc = 0;
 607         int enabled, second_pass = 0;
 608         unsigned len, type;
 609         struct perf_event *bp;
 610
 611         data &= ~DR_CONTROL_RESERVED;
 612         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 613 restore:
 614         /*
 615          * Loop through all the hardware breakpoints, making the
 616          * appropriate changes to each.
 617          */
 618         for (i = 0; i < HBP_NUM; i++) {
 619                 enabled = decode_dr7(data, i, &len, &type);
 620                 bp = thread->ptrace_bps[i];
 621
 622                 if (!enabled) {
 623                         if (bp) {
 624                                 /*
 625                                  * Don't unregister the breakpoints right-away,
 626                                  * unless all register_user_hw_breakpoint()
 627                                  * requests have succeeded. This prevents
 628                                  * any window of opportunity for debug
 629                                  * register grabbing by other users.
 630                                  */
 631                                 if (!second_pass)
 632                                         continue;
 633
 634                                 rc = ptrace_modify_breakpoint(bp, len, type,
 635                                                               tsk, 1);
 636                                 if (rc)
 637                                         break;
 638                         }
 639                         continue;
 640                 }
 641
 642                 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
 643                 if (rc)
 644                         break;
 645         }
 646         /*
 647          * Make a second pass to free the remaining unused breakpoints
 648          * or to restore the original breakpoints if an error occurred.
 649          */
 650         if (!second_pass) {
 651                 second_pass = 1;
 652                 if (rc < 0) {
 653                         orig_ret = rc;
 654                         data = old_dr7;
 655                 }
 656                 goto restore;
 657         }
 658         return ((orig_ret < 0) ? orig_ret : rc);
 659 }
 660
 661 /*
 662  * Handle PTRACE_PEEKUSR calls for the debug register area.
 663  */
 664 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 665 {
 666         struct thread_struct *thread = &(tsk->thread);
 667         unsigned long val = 0;
 668
 669         if (n < HBP_NUM) {
 670                 struct perf_event *bp;
 671                 bp = thread->ptrace_bps[n];
 672                 if (!bp)
 673                         return 0;
 674                 val = bp->hw.info.address;
 675         } else if (n == 6) {
 676                 val = thread->debugreg6;
 677          } else if (n == 7) {
 678                 val = thread->ptrace_dr7;
 679         }
 680         return val;
 681 }
 682
 683 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 684                                       unsigned long addr)
 685 {
 686         struct perf_event *bp;
 687         struct thread_struct *t = &tsk->thread;
 688         struct perf_event_attr attr;
 689
 690         if (!t->ptrace_bps[nr]) {
 691                 ptrace_breakpoint_init(&attr);
 692                 /*
 693                  * Put stub len and type to register (reserve) an inactive but
 694                  * correct bp
 695                  */
 696                 attr.bp_addr = addr;
 697                 attr.bp_len = HW_BREAKPOINT_LEN_1;
 698                 attr.bp_type = HW_BREAKPOINT_W;
 699                 attr.disabled = 1;
 700
 701                 bp = register_user_hw_breakpoint(&attr, ptrace_triggered, tsk);
 702
 703                 /*
 704                  * CHECKME: the previous code returned -EIO if the addr wasn't
 705                  * a valid task virtual addr. The new one will return -EINVAL in
 706                  *  this case.
 707                  * -EINVAL may be what we want for in-kernel breakpoints users,
 708                  * but -EIO looks better for ptrace, since we refuse a register
 709                  * writing for the user. And anyway this is the previous
 710                  * behaviour.
 711                  */
 712                 if (IS_ERR(bp))
 713                         return PTR_ERR(bp);
 714
 715                 t->ptrace_bps[nr] = bp;
 716         } else {
 717                 int err;
 718
 719                 bp = t->ptrace_bps[nr];
 720
 721                 attr = bp->attr;
 722                 attr.bp_addr = addr;
 723                 err = modify_user_hw_breakpoint(bp, &attr);
 724                 if (err)
 725                         return err;
 726         }
 727
 728
 729         return 0;
 730 }
 731
 732 /*
 733  * Handle PTRACE_POKEUSR calls for the debug register area.
 734  */
 735 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
 736 {
 737         struct thread_struct *thread = &(tsk->thread);
 738         int rc = 0;
 739
 740         /* There are no DR4 or DR5 registers */
 741         if (n == 4 || n == 5)
 742                 return -EIO;
 743
 744         if (n == 6) {
 745                 thread->debugreg6 = val;
 746                 goto ret_path;
 747         }
 748         if (n < HBP_NUM) {
 749                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 750                 if (rc)
 751                         return rc;
 752         }
 753         /* All that's left is DR7 */
 754         if (n == 7) {
 755                 rc = ptrace_write_dr7(tsk, val);
 756                 if (!rc)
 757                         thread->ptrace_dr7 = val;
 758         }
 759
 760 ret_path:
 761         return rc;
 762 }
 763
 764 /*
 765  * These access the current or another (stopped) task's io permission
 766  * bitmap for debugging or core dump.
 767  */
 768 static int ioperm_active(struct task_struct *target,
 769                          const struct user_regset *regset)
 770 {
 771         return target->thread.io_bitmap_max / regset->size;
 772 }
 773
 774 static int ioperm_get(struct task_struct *target,
 775                       const struct user_regset *regset,
 776                       unsigned int pos, unsigned int count,
 777                       void *kbuf, void __user *ubuf)
 778 {
 779         if (!target->thread.io_bitmap_ptr)
 780                 return -ENXIO;
 781
 782         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 783                                    target->thread.io_bitmap_ptr,
 784                                    0, IO_BITMAP_BYTES);
 785 }
 786
 787 /*
 788  * Called by kernel/ptrace.c when detaching..
 789  *
 790  * Make sure the single step bit is not set.
 791  */
 792 void ptrace_disable(struct task_struct *child)
 793 {
 794         user_disable_single_step(child);
 795 #ifdef TIF_SYSCALL_EMU
 796         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 797 #endif
 798 }
 799
 800 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 801 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 802 #endif
 803
 804 long arch_ptrace(struct task_struct *child, long request,
 805                  unsigned long addr, unsigned long data)
 806 {
 807         int ret;
 808         unsigned long __user *datap = (unsigned long __user *)data;
 809
 810         switch (request) {
 811         /* read the word at location addr in the USER area. */
 812         case PTRACE_PEEKUSR: {
 813                 unsigned long tmp;
 814
 815                 ret = -EIO;
 816                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 817                         break;
 818
 819                 tmp = 0;  /* Default return condition */
 820                 if (addr < sizeof(struct user_regs_struct))
 821                         tmp = getreg(child, addr);
 822                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 823                          addr <= offsetof(struct user, u_debugreg[7])) {
 824                         addr -= offsetof(struct user, u_debugreg[0]);
 825                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 826                 }
 827                 ret = put_user(tmp, datap);
 828                 break;
 829         }
 830
 831         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 832                 ret = -EIO;
 833                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 834                         break;
 835
 836                 if (addr < sizeof(struct user_regs_struct))
 837                         ret = putreg(child, addr, data);
 838                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 839                          addr <= offsetof(struct user, u_debugreg[7])) {
 840                         addr -= offsetof(struct user, u_debugreg[0]);
 841                         ret = ptrace_set_debugreg(child,
 842                                                   addr / sizeof(data), data);
 843                 }
 844                 break;
 845
 846         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 847                 return copy_regset_to_user(child,
 848                                            task_user_regset_view(current),
 849                                            REGSET_GENERAL,
 850                                            0, sizeof(struct user_regs_struct),
 851                                            datap);
 852
 853         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 854                 return copy_regset_from_user(child,
 855                                              task_user_regset_view(current),
 856                                              REGSET_GENERAL,
 857                                              0, sizeof(struct user_regs_struct),
 858                                              datap);
 859
 860         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 861                 return copy_regset_to_user(child,
 862                                            task_user_regset_view(current),
 863                                            REGSET_FP,
 864                                            0, sizeof(struct user_i387_struct),
 865                                            datap);
 866
 867         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 868                 return copy_regset_from_user(child,
 869                                              task_user_regset_view(current),
 870                                              REGSET_FP,
 871                                              0, sizeof(struct user_i387_struct),
 872                                              datap);
 873
 874 #ifdef CONFIG_X86_32
 875         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 876                 return copy_regset_to_user(child, &user_x86_32_view,
 877                                            REGSET_XFP,
 878                                            0, sizeof(struct user_fxsr_struct),
 879                                            datap) ? -EIO : 0;
 880
 881         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 882                 return copy_regset_from_user(child, &user_x86_32_view,
 883                                              REGSET_XFP,
 884                                              0, sizeof(struct user_fxsr_struct),
 885                                              datap) ? -EIO : 0;
 886 #endif
 887
 888 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 889         case PTRACE_GET_THREAD_AREA:
 890                 if ((int) addr < 0)
 891                         return -EIO;
 892                 ret = do_get_thread_area(child, addr,
 893                                         (struct user_desc __user *)data);
 894                 break;
 895
 896         case PTRACE_SET_THREAD_AREA:
 897                 if ((int) addr < 0)
 898                         return -EIO;
 899                 ret = do_set_thread_area(child, addr,
 900                                         (struct user_desc __user *)data, 0);
 901                 break;
 902 #endif
 903
 904 #ifdef CONFIG_X86_64
 905                 /* normal 64bit interface to access TLS data.
 906                    Works just like arch_prctl, except that the arguments
 907                    are reversed. */
 908         case PTRACE_ARCH_PRCTL:
 909                 ret = do_arch_prctl(child, data, addr);
 910                 break;
 911 #endif
 912
 913         default:
 914                 ret = ptrace_request(child, request, addr, data);
 915                 break;
 916         }
 917
 918         return ret;
 919 }
 920
 921 #ifdef CONFIG_IA32_EMULATION
 922
 923 #include <linux/compat.h>
 924 #include <linux/syscalls.h>
 925 #include <asm/ia32.h>
 926 #include <asm/user32.h>
 927
 928 #define R32(l,q)                                                        \
 929         case offsetof(struct user32, regs.l):                           \
 930                 regs->q = value; break
 931
 932 #define SEG32(rs)                                                       \
 933         case offsetof(struct user32, regs.rs):                          \
 934                 return set_segment_reg(child,                           \
 935                                        offsetof(struct user_regs_struct, rs), \
 936                                        value);                          \
 937                 break
 938
 939 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 940 {
 941         struct pt_regs *regs = task_pt_regs(child);
 942
 943         switch (regno) {
 944
 945         SEG32(cs);
 946         SEG32(ds);
 947         SEG32(es);
 948         SEG32(fs);
 949         SEG32(gs);
 950         SEG32(ss);
 951
 952         R32(ebx, bx);
 953         R32(ecx, cx);
 954         R32(edx, dx);
 955         R32(edi, di);
 956         R32(esi, si);
 957         R32(ebp, bp);
 958         R32(eax, ax);
 959         R32(eip, ip);
 960         R32(esp, sp);
 961
 962         case offsetof(struct user32, regs.orig_eax):
 963                 /*
 964                  * A 32-bit debugger setting orig_eax means to restore
 965                  * the state of the task restarting a 32-bit syscall.
 966                  * Make sure we interpret the -ERESTART* codes correctly
 967                  * in case the task is not actually still sitting at the
 968                  * exit from a 32-bit syscall with TS_COMPAT still set.
 969                  */
 970                 regs->orig_ax = value;
 971                 if (syscall_get_nr(child, regs) >= 0)
 972                         task_thread_info(child)->status |= TS_COMPAT;
 973                 break;
 974
 975         case offsetof(struct user32, regs.eflags):
 976                 return set_flags(child, value);
 977
 978         case offsetof(struct user32, u_debugreg[0]) ...
 979                 offsetof(struct user32, u_debugreg[7]):
 980                 regno -= offsetof(struct user32, u_debugreg[0]);
 981                 return ptrace_set_debugreg(child, regno / 4, value);
 982
 983         default:
 984                 if (regno > sizeof(struct user32) || (regno & 3))
 985                         return -EIO;
 986
 987                 /*
 988                  * Other dummy fields in the virtual user structure
 989                  * are ignored
 990                  */
 991                 break;
 992         }
 993         return 0;
 994 }
 995
 996 #undef R32
 997 #undef SEG32
 998
 999 #define R32(l,q)                                                        \
1000         case offsetof(struct user32, regs.l):                           \
1001                 *val = regs->q; break
1002
1003 #define SEG32(rs)                                                       \
1004         case offsetof(struct user32, regs.rs):                          \
1005                 *val = get_segment_reg(child,                           \
1006                                        offsetof(struct user_regs_struct, rs)); \
1007                 break
1008
1009 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1010 {
1011         struct pt_regs *regs = task_pt_regs(child);
1012
1013         switch (regno) {
1014
1015         SEG32(ds);
1016         SEG32(es);
1017         SEG32(fs);
1018         SEG32(gs);
1019
1020         R32(cs, cs);
1021         R32(ss, ss);
1022         R32(ebx, bx);
1023         R32(ecx, cx);
1024         R32(edx, dx);
1025         R32(edi, di);
1026         R32(esi, si);
1027         R32(ebp, bp);
1028         R32(eax, ax);
1029         R32(orig_eax, orig_ax);
1030         R32(eip, ip);
1031         R32(esp, sp);
1032
1033         case offsetof(struct user32, regs.eflags):
1034                 *val = get_flags(child);
1035                 break;
1036
1037         case offsetof(struct user32, u_debugreg[0]) ...
1038                 offsetof(struct user32, u_debugreg[7]):
1039                 regno -= offsetof(struct user32, u_debugreg[0]);
1040                 *val = ptrace_get_debugreg(child, regno / 4);
1041                 break;
1042
1043         default:
1044                 if (regno > sizeof(struct user32) || (regno & 3))
1045                         return -EIO;
1046
1047                 /*
1048                  * Other dummy fields in the virtual user structure
1049                  * are ignored
1050                  */
1051                 *val = 0;
1052                 break;
1053         }
1054         return 0;
1055 }
1056
1057 #undef R32
1058 #undef SEG32
1059
1060 static int genregs32_get(struct task_struct *target,
1061                          const struct user_regset *regset,
1062                          unsigned int pos, unsigned int count,
1063                          void *kbuf, void __user *ubuf)
1064 {
1065         if (kbuf) {
1066                 compat_ulong_t *k = kbuf;
1067                 while (count >= sizeof(*k)) {
1068                         getreg32(target, pos, k++);
1069                         count -= sizeof(*k);
1070                         pos += sizeof(*k);
1071                 }
1072         } else {
1073                 compat_ulong_t __user *u = ubuf;
1074                 while (count >= sizeof(*u)) {
1075                         compat_ulong_t word;
1076                         getreg32(target, pos, &word);
1077                         if (__put_user(word, u++))
1078                                 return -EFAULT;
1079                         count -= sizeof(*u);
1080                         pos += sizeof(*u);
1081                 }
1082         }
1083
1084         return 0;
1085 }
1086
1087 static int genregs32_set(struct task_struct *target,
1088                          const struct user_regset *regset,
1089                          unsigned int pos, unsigned int count,
1090                          const void *kbuf, const void __user *ubuf)
1091 {
1092         int ret = 0;
1093         if (kbuf) {
1094                 const compat_ulong_t *k = kbuf;
1095                 while (count >= sizeof(*k) && !ret) {
1096                         ret = putreg32(target, pos, *k++);
1097                         count -= sizeof(*k);
1098                         pos += sizeof(*k);
1099                 }
1100         } else {
1101                 const compat_ulong_t __user *u = ubuf;
1102                 while (count >= sizeof(*u) && !ret) {
1103                         compat_ulong_t word;
1104                         ret = __get_user(word, u++);
1105                         if (ret)
1106                                 break;
1107                         ret = putreg32(target, pos, word);
1108                         count -= sizeof(*u);
1109                         pos += sizeof(*u);
1110                 }
1111         }
1112         return ret;
1113 }
1114
1115 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1116                         compat_ulong_t caddr, compat_ulong_t cdata)
1117 {
1118         unsigned long addr = caddr;
1119         unsigned long data = cdata;
1120         void __user *datap = compat_ptr(data);
1121         int ret;
1122         __u32 val;
1123
1124         switch (request) {
1125         case PTRACE_PEEKUSR:
1126                 ret = getreg32(child, addr, &val);
1127                 if (ret == 0)
1128                         ret = put_user(val, (__u32 __user *)datap);
1129                 break;
1130
1131         case PTRACE_POKEUSR:
1132                 ret = putreg32(child, addr, data);
1133                 break;
1134
1135         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1136                 return copy_regset_to_user(child, &user_x86_32_view,
1137                                            REGSET_GENERAL,
1138                                            0, sizeof(struct user_regs_struct32),
1139                                            datap);
1140
1141         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1142                 return copy_regset_from_user(child, &user_x86_32_view,
1143                                              REGSET_GENERAL, 0,
1144                                              sizeof(struct user_regs_struct32),
1145                                              datap);
1146
1147         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1148                 return copy_regset_to_user(child, &user_x86_32_view,
1149                                            REGSET_FP, 0,
1150                                            sizeof(struct user_i387_ia32_struct),
1151                                            datap);
1152
1153         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1154                 return copy_regset_from_user(
1155                         child, &user_x86_32_view, REGSET_FP,
1156                         0, sizeof(struct user_i387_ia32_struct), datap);
1157
1158         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1159                 return copy_regset_to_user(child, &user_x86_32_view,
1160                                            REGSET_XFP, 0,
1161                                            sizeof(struct user32_fxsr_struct),
1162                                            datap);
1163
1164         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1165                 return copy_regset_from_user(child, &user_x86_32_view,
1166                                              REGSET_XFP, 0,
1167                                              sizeof(struct user32_fxsr_struct),
1168                                              datap);
1169
1170         case PTRACE_GET_THREAD_AREA:
1171         case PTRACE_SET_THREAD_AREA:
1172                 return arch_ptrace(child, request, addr, data);
1173
1174         default:
1175                 return compat_ptrace_request(child, request, addr, data);
1176         }
1177
1178         return ret;
1179 }
1180
1181 #endif  /* CONFIG_IA32_EMULATION */
1182
1183 #ifdef CONFIG_X86_64
1184
1185 static struct user_regset x86_64_regsets[] __read_mostly = {
1186         [REGSET_GENERAL] = {
1187                 .core_note_type = NT_PRSTATUS,
1188                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1189                 .size = sizeof(long), .align = sizeof(long),
1190                 .get = genregs_get, .set = genregs_set
1191         },
1192         [REGSET_FP] = {
1193                 .core_note_type = NT_PRFPREG,
1194                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1195                 .size = sizeof(long), .align = sizeof(long),
1196                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1197         },
1198         [REGSET_XSTATE] = {
1199                 .core_note_type = NT_X86_XSTATE,
1200                 .size = sizeof(u64), .align = sizeof(u64),
1201                 .active = xstateregs_active, .get = xstateregs_get,
1202                 .set = xstateregs_set
1203         },
1204         [REGSET_IOPERM64] = {
1205                 .core_note_type = NT_386_IOPERM,
1206                 .n = IO_BITMAP_LONGS,
1207                 .size = sizeof(long), .align = sizeof(long),
1208                 .active = ioperm_active, .get = ioperm_get
1209         },
1210 };
1211
1212 static const struct user_regset_view user_x86_64_view = {
1213         .name = "x86_64", .e_machine = EM_X86_64,
1214         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1215 };
1216
1217 #else  /* CONFIG_X86_32 */
1218
1219 #define user_regs_struct32      user_regs_struct
1220 #define genregs32_get           genregs_get
1221 #define genregs32_set           genregs_set
1222
1223 #define user_i387_ia32_struct   user_i387_struct
1224 #define user32_fxsr_struct      user_fxsr_struct
1225
1226 #endif  /* CONFIG_X86_64 */
1227
1228 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1229 static struct user_regset x86_32_regsets[] __read_mostly = {
1230         [REGSET_GENERAL] = {
1231                 .core_note_type = NT_PRSTATUS,
1232                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1233                 .size = sizeof(u32), .align = sizeof(u32),
1234                 .get = genregs32_get, .set = genregs32_set
1235         },
1236         [REGSET_FP] = {
1237                 .core_note_type = NT_PRFPREG,
1238                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1239                 .size = sizeof(u32), .align = sizeof(u32),
1240                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1241         },
1242         [REGSET_XFP] = {
1243                 .core_note_type = NT_PRXFPREG,
1244                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1245                 .size = sizeof(u32), .align = sizeof(u32),
1246                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1247         },
1248         [REGSET_XSTATE] = {
1249                 .core_note_type = NT_X86_XSTATE,
1250                 .size = sizeof(u64), .align = sizeof(u64),
1251                 .active = xstateregs_active, .get = xstateregs_get,
1252                 .set = xstateregs_set
1253         },
1254         [REGSET_TLS] = {
1255                 .core_note_type = NT_386_TLS,
1256                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1257                 .size = sizeof(struct user_desc),
1258                 .align = sizeof(struct user_desc),
1259                 .active = regset_tls_active,
1260                 .get = regset_tls_get, .set = regset_tls_set
1261         },
1262         [REGSET_IOPERM32] = {
1263                 .core_note_type = NT_386_IOPERM,
1264                 .n = IO_BITMAP_BYTES / sizeof(u32),
1265                 .size = sizeof(u32), .align = sizeof(u32),
1266                 .active = ioperm_active, .get = ioperm_get
1267         },
1268 };
1269
1270 static const struct user_regset_view user_x86_32_view = {
1271         .name = "i386", .e_machine = EM_386,
1272         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1273 };
1274 #endif
1275
1276 /*
1277  * This represents bytes 464..511 in the memory layout exported through
1278  * the REGSET_XSTATE interface.
1279  */
1280 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1281
1282 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1283 {
1284 #ifdef CONFIG_X86_64
1285         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1286 #endif
1287 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1288         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1289 #endif
1290         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1291 }
1292
1293 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1294 {
1295 #ifdef CONFIG_IA32_EMULATION
1296         if (test_tsk_thread_flag(task, TIF_IA32))
1297 #endif
1298 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1299                 return &user_x86_32_view;
1300 #endif
1301 #ifdef CONFIG_X86_64
1302         return &user_x86_64_view;
1303 #endif
1304 }
1305
1306 static void fill_sigtrap_info(struct task_struct *tsk,
1307                                 struct pt_regs *regs,
1308                                 int error_code, int si_code,
1309                                 struct siginfo *info)
1310 {
1311         tsk->thread.trap_no = 1;
1312         tsk->thread.error_code = error_code;
1313
1314         memset(info, 0, sizeof(*info));
1315         info->si_signo = SIGTRAP;
1316         info->si_code = si_code;
1317         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1318 }
1319
1320 void user_single_step_siginfo(struct task_struct *tsk,
1321                                 struct pt_regs *regs,
1322                                 struct siginfo *info)
1323 {
1324         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1325 }
1326
1327 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1328                                          int error_code, int si_code)
1329 {
1330         struct siginfo info;
1331
1332         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1333         /* Send us the fake SIGTRAP */
1334         force_sig_info(SIGTRAP, &info, tsk);
1335 }
1336
1337
1338 #ifdef CONFIG_X86_32
1339 # define IS_IA32        1
1340 #elif defined CONFIG_IA32_EMULATION
1341 # define IS_IA32        is_compat_task()
1342 #else
1343 # define IS_IA32        0
1344 #endif
1345
1346 /*
1347  * We must return the syscall number to actually look up in the table.
1348  * This can be -1L to skip running any syscall at all.
1349  */
1350 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1351 {
1352         long ret = 0;
1353
1354         /*
1355          * If we stepped into a sysenter/syscall insn, it trapped in
1356          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1357          * If user-mode had set TF itself, then it's still clear from
1358          * do_debug() and we need to set it again to restore the user
1359          * state.  If we entered on the slow path, TF was already set.
1360          */
1361         if (test_thread_flag(TIF_SINGLESTEP))
1362                 regs->flags |= X86_EFLAGS_TF;
1363
1364         /* do the secure computing check first */
1365         secure_computing(regs->orig_ax);
1366
1367         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1368                 ret = -1L;
1369
1370         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1371             tracehook_report_syscall_entry(regs))
1372                 ret = -1L;
1373
1374         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1375                 trace_sys_enter(regs, regs->orig_ax);
1376
1377         if (unlikely(current->audit_context)) {
1378                 if (IS_IA32)
1379                         audit_syscall_entry(AUDIT_ARCH_I386,
1380                                             regs->orig_ax,
1381                                             regs->bx, regs->cx,
1382                                             regs->dx, regs->si);
1383 #ifdef CONFIG_X86_64
1384                 else
1385                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1386                                             regs->orig_ax,
1387                                             regs->di, regs->si,
1388                                             regs->dx, regs->r10);
1389 #endif
1390         }
1391
1392         return ret ?: regs->orig_ax;
1393 }
1394
1395 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1396 {
1397         bool step;
1398
1399         if (unlikely(current->audit_context))
1400                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1401
1402         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1403                 trace_sys_exit(regs, regs->ax);
1404
1405         /*
1406          * If TIF_SYSCALL_EMU is set, we only get here because of
1407          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1408          * We already reported this syscall instruction in
1409          * syscall_trace_enter().
1410          */
1411         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1412                         !test_thread_flag(TIF_SYSCALL_EMU);
1413         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1414                 tracehook_report_syscall_exit(regs, step);
1415 }