git.oblomov.eu Git - linux-2.6/blob - 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  * BTS tracing
   7  *      Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/mm.h>
  13 #include <linux/smp.h>
  14 #include <linux/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/regset.h>
  17 #include <linux/tracehook.h>
  18 #include <linux/user.h>
  19 #include <linux/elf.h>
  20 #include <linux/security.h>
  21 #include <linux/audit.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/signal.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/ds.h>
  36
  37 #include "tls.h"
  38
  39 enum x86_regset {
  40         REGSET_GENERAL,
  41         REGSET_FP,
  42         REGSET_XFP,
  43         REGSET_IOPERM64 = REGSET_XFP,
  44         REGSET_TLS,
  45         REGSET_IOPERM32,
  46 };
  47
  48 /*
  49  * does not yet catch signals sent when the child dies.
  50  * in exit.c or in signal.c.
  51  */
  52
  53 /*
  54  * Determines which flags the user has access to [1 = access, 0 = no access].
  55  */
  56 #define FLAG_MASK_32            ((unsigned long)                        \
  57                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  58                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  59                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  60                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  61                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  62
  63 /*
  64  * Determines whether a value may be installed in a segment register.
  65  */
  66 static inline bool invalid_selector(u16 value)
  67 {
  68         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  69 }
  70
  71 #ifdef CONFIG_X86_32
  72
  73 #define FLAG_MASK               FLAG_MASK_32
  74
  75 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  76 {
  77         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  78         regno >>= 2;
  79         if (regno > FS)
  80                 --regno;
  81         return &regs->bx + regno;
  82 }
  83
  84 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  85 {
  86         /*
  87          * Returning the value truncates it to 16 bits.
  88          */
  89         unsigned int retval;
  90         if (offset != offsetof(struct user_regs_struct, gs))
  91                 retval = *pt_regs_access(task_pt_regs(task), offset);
  92         else {
  93                 retval = task->thread.gs;
  94                 if (task == current)
  95                         savesegment(gs, retval);
  96         }
  97         return retval;
  98 }
  99
 100 static int set_segment_reg(struct task_struct *task,
 101                            unsigned long offset, u16 value)
 102 {
 103         /*
 104          * The value argument was already truncated to 16 bits.
 105          */
 106         if (invalid_selector(value))
 107                 return -EIO;
 108
 109         /*
 110          * For %cs and %ss we cannot permit a null selector.
 111          * We can permit a bogus selector as long as it has USER_RPL.
 112          * Null selectors are fine for other segment registers, but
 113          * we will never get back to user mode with invalid %cs or %ss
 114          * and will take the trap in iret instead.  Much code relies
 115          * on user_mode() to distinguish a user trap frame (which can
 116          * safely use invalid selectors) from a kernel trap frame.
 117          */
 118         switch (offset) {
 119         case offsetof(struct user_regs_struct, cs):
 120         case offsetof(struct user_regs_struct, ss):
 121                 if (unlikely(value == 0))
 122                         return -EIO;
 123
 124         default:
 125                 *pt_regs_access(task_pt_regs(task), offset) = value;
 126                 break;
 127
 128         case offsetof(struct user_regs_struct, gs):
 129                 task->thread.gs = value;
 130                 if (task == current)
 131                         /*
 132                          * The user-mode %gs is not affected by
 133                          * kernel entry, so we must update the CPU.
 134                          */
 135                         loadsegment(gs, value);
 136         }
 137
 138         return 0;
 139 }
 140
 141 static unsigned long debugreg_addr_limit(struct task_struct *task)
 142 {
 143         return TASK_SIZE - 3;
 144 }
 145
 146 #else  /* CONFIG_X86_64 */
 147
 148 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 149
 150 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 151 {
 152         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 153         return &regs->r15 + (offset / sizeof(regs->r15));
 154 }
 155
 156 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 157 {
 158         /*
 159          * Returning the value truncates it to 16 bits.
 160          */
 161         unsigned int seg;
 162
 163         switch (offset) {
 164         case offsetof(struct user_regs_struct, fs):
 165                 if (task == current) {
 166                         /* Older gas can't assemble movq %?s,%r?? */
 167                         asm("movl %%fs,%0" : "=r" (seg));
 168                         return seg;
 169                 }
 170                 return task->thread.fsindex;
 171         case offsetof(struct user_regs_struct, gs):
 172                 if (task == current) {
 173                         asm("movl %%gs,%0" : "=r" (seg));
 174                         return seg;
 175                 }
 176                 return task->thread.gsindex;
 177         case offsetof(struct user_regs_struct, ds):
 178                 if (task == current) {
 179                         asm("movl %%ds,%0" : "=r" (seg));
 180                         return seg;
 181                 }
 182                 return task->thread.ds;
 183         case offsetof(struct user_regs_struct, es):
 184                 if (task == current) {
 185                         asm("movl %%es,%0" : "=r" (seg));
 186                         return seg;
 187                 }
 188                 return task->thread.es;
 189
 190         case offsetof(struct user_regs_struct, cs):
 191         case offsetof(struct user_regs_struct, ss):
 192                 break;
 193         }
 194         return *pt_regs_access(task_pt_regs(task), offset);
 195 }
 196
 197 static int set_segment_reg(struct task_struct *task,
 198                            unsigned long offset, u16 value)
 199 {
 200         /*
 201          * The value argument was already truncated to 16 bits.
 202          */
 203         if (invalid_selector(value))
 204                 return -EIO;
 205
 206         switch (offset) {
 207         case offsetof(struct user_regs_struct,fs):
 208                 /*
 209                  * If this is setting fs as for normal 64-bit use but
 210                  * setting fs_base has implicitly changed it, leave it.
 211                  */
 212                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 213                      task->thread.fs != 0) ||
 214                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 215                      task->thread.fs == 0))
 216                         break;
 217                 task->thread.fsindex = value;
 218                 if (task == current)
 219                         loadsegment(fs, task->thread.fsindex);
 220                 break;
 221         case offsetof(struct user_regs_struct,gs):
 222                 /*
 223                  * If this is setting gs as for normal 64-bit use but
 224                  * setting gs_base has implicitly changed it, leave it.
 225                  */
 226                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 227                      task->thread.gs != 0) ||
 228                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 229                      task->thread.gs == 0))
 230                         break;
 231                 task->thread.gsindex = value;
 232                 if (task == current)
 233                         load_gs_index(task->thread.gsindex);
 234                 break;
 235         case offsetof(struct user_regs_struct,ds):
 236                 task->thread.ds = value;
 237                 if (task == current)
 238                         loadsegment(ds, task->thread.ds);
 239                 break;
 240         case offsetof(struct user_regs_struct,es):
 241                 task->thread.es = value;
 242                 if (task == current)
 243                         loadsegment(es, task->thread.es);
 244                 break;
 245
 246                 /*
 247                  * Can't actually change these in 64-bit mode.
 248                  */
 249         case offsetof(struct user_regs_struct,cs):
 250                 if (unlikely(value == 0))
 251                         return -EIO;
 252 #ifdef CONFIG_IA32_EMULATION
 253                 if (test_tsk_thread_flag(task, TIF_IA32))
 254                         task_pt_regs(task)->cs = value;
 255 #endif
 256                 break;
 257         case offsetof(struct user_regs_struct,ss):
 258                 if (unlikely(value == 0))
 259                         return -EIO;
 260 #ifdef CONFIG_IA32_EMULATION
 261                 if (test_tsk_thread_flag(task, TIF_IA32))
 262                         task_pt_regs(task)->ss = value;
 263 #endif
 264                 break;
 265         }
 266
 267         return 0;
 268 }
 269
 270 static unsigned long debugreg_addr_limit(struct task_struct *task)
 271 {
 272 #ifdef CONFIG_IA32_EMULATION
 273         if (test_tsk_thread_flag(task, TIF_IA32))
 274                 return IA32_PAGE_OFFSET - 3;
 275 #endif
 276         return TASK_SIZE64 - 7;
 277 }
 278
 279 #endif  /* CONFIG_X86_32 */
 280
 281 static unsigned long get_flags(struct task_struct *task)
 282 {
 283         unsigned long retval = task_pt_regs(task)->flags;
 284
 285         /*
 286          * If the debugger set TF, hide it from the readout.
 287          */
 288         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 289                 retval &= ~X86_EFLAGS_TF;
 290
 291         return retval;
 292 }
 293
 294 static int set_flags(struct task_struct *task, unsigned long value)
 295 {
 296         struct pt_regs *regs = task_pt_regs(task);
 297
 298         /*
 299          * If the user value contains TF, mark that
 300          * it was not "us" (the debugger) that set it.
 301          * If not, make sure it stays set if we had.
 302          */
 303         if (value & X86_EFLAGS_TF)
 304                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 305         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 306                 value |= X86_EFLAGS_TF;
 307
 308         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 309
 310         return 0;
 311 }
 312
 313 static int putreg(struct task_struct *child,
 314                   unsigned long offset, unsigned long value)
 315 {
 316         switch (offset) {
 317         case offsetof(struct user_regs_struct, cs):
 318         case offsetof(struct user_regs_struct, ds):
 319         case offsetof(struct user_regs_struct, es):
 320         case offsetof(struct user_regs_struct, fs):
 321         case offsetof(struct user_regs_struct, gs):
 322         case offsetof(struct user_regs_struct, ss):
 323                 return set_segment_reg(child, offset, value);
 324
 325         case offsetof(struct user_regs_struct, flags):
 326                 return set_flags(child, value);
 327
 328 #ifdef CONFIG_X86_64
 329         /*
 330          * Orig_ax is really just a flag with small positive and
 331          * negative values, so make sure to always sign-extend it
 332          * from 32 bits so that it works correctly regardless of
 333          * whether we come from a 32-bit environment or not.
 334          */
 335         case offsetof(struct user_regs_struct, orig_ax):
 336                 value = (long) (s32) value;
 337                 break;
 338
 339         case offsetof(struct user_regs_struct,fs_base):
 340                 if (value >= TASK_SIZE_OF(child))
 341                         return -EIO;
 342                 /*
 343                  * When changing the segment base, use do_arch_prctl
 344                  * to set either thread.fs or thread.fsindex and the
 345                  * corresponding GDT slot.
 346                  */
 347                 if (child->thread.fs != value)
 348                         return do_arch_prctl(child, ARCH_SET_FS, value);
 349                 return 0;
 350         case offsetof(struct user_regs_struct,gs_base):
 351                 /*
 352                  * Exactly the same here as the %fs handling above.
 353                  */
 354                 if (value >= TASK_SIZE_OF(child))
 355                         return -EIO;
 356                 if (child->thread.gs != value)
 357                         return do_arch_prctl(child, ARCH_SET_GS, value);
 358                 return 0;
 359 #endif
 360         }
 361
 362         *pt_regs_access(task_pt_regs(child), offset) = value;
 363         return 0;
 364 }
 365
 366 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 367 {
 368         switch (offset) {
 369         case offsetof(struct user_regs_struct, cs):
 370         case offsetof(struct user_regs_struct, ds):
 371         case offsetof(struct user_regs_struct, es):
 372         case offsetof(struct user_regs_struct, fs):
 373         case offsetof(struct user_regs_struct, gs):
 374         case offsetof(struct user_regs_struct, ss):
 375                 return get_segment_reg(task, offset);
 376
 377         case offsetof(struct user_regs_struct, flags):
 378                 return get_flags(task);
 379
 380 #ifdef CONFIG_X86_64
 381         case offsetof(struct user_regs_struct, fs_base): {
 382                 /*
 383                  * do_arch_prctl may have used a GDT slot instead of
 384                  * the MSR.  To userland, it appears the same either
 385                  * way, except the %fs segment selector might not be 0.
 386                  */
 387                 unsigned int seg = task->thread.fsindex;
 388                 if (task->thread.fs != 0)
 389                         return task->thread.fs;
 390                 if (task == current)
 391                         asm("movl %%fs,%0" : "=r" (seg));
 392                 if (seg != FS_TLS_SEL)
 393                         return 0;
 394                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 395         }
 396         case offsetof(struct user_regs_struct, gs_base): {
 397                 /*
 398                  * Exactly the same here as the %fs handling above.
 399                  */
 400                 unsigned int seg = task->thread.gsindex;
 401                 if (task->thread.gs != 0)
 402                         return task->thread.gs;
 403                 if (task == current)
 404                         asm("movl %%gs,%0" : "=r" (seg));
 405                 if (seg != GS_TLS_SEL)
 406                         return 0;
 407                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 408         }
 409 #endif
 410         }
 411
 412         return *pt_regs_access(task_pt_regs(task), offset);
 413 }
 414
 415 static int genregs_get(struct task_struct *target,
 416                        const struct user_regset *regset,
 417                        unsigned int pos, unsigned int count,
 418                        void *kbuf, void __user *ubuf)
 419 {
 420         if (kbuf) {
 421                 unsigned long *k = kbuf;
 422                 while (count > 0) {
 423                         *k++ = getreg(target, pos);
 424                         count -= sizeof(*k);
 425                         pos += sizeof(*k);
 426                 }
 427         } else {
 428                 unsigned long __user *u = ubuf;
 429                 while (count > 0) {
 430                         if (__put_user(getreg(target, pos), u++))
 431                                 return -EFAULT;
 432                         count -= sizeof(*u);
 433                         pos += sizeof(*u);
 434                 }
 435         }
 436
 437         return 0;
 438 }
 439
 440 static int genregs_set(struct task_struct *target,
 441                        const struct user_regset *regset,
 442                        unsigned int pos, unsigned int count,
 443                        const void *kbuf, const void __user *ubuf)
 444 {
 445         int ret = 0;
 446         if (kbuf) {
 447                 const unsigned long *k = kbuf;
 448                 while (count > 0 && !ret) {
 449                         ret = putreg(target, pos, *k++);
 450                         count -= sizeof(*k);
 451                         pos += sizeof(*k);
 452                 }
 453         } else {
 454                 const unsigned long  __user *u = ubuf;
 455                 while (count > 0 && !ret) {
 456                         unsigned long word;
 457                         ret = __get_user(word, u++);
 458                         if (ret)
 459                                 break;
 460                         ret = putreg(target, pos, word);
 461                         count -= sizeof(*u);
 462                         pos += sizeof(*u);
 463                 }
 464         }
 465         return ret;
 466 }
 467
 468 /*
 469  * This function is trivial and will be inlined by the compiler.
 470  * Having it separates the implementation details of debug
 471  * registers from the interface details of ptrace.
 472  */
 473 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 474 {
 475         switch (n) {
 476         case 0:         return child->thread.debugreg0;
 477         case 1:         return child->thread.debugreg1;
 478         case 2:         return child->thread.debugreg2;
 479         case 3:         return child->thread.debugreg3;
 480         case 6:         return child->thread.debugreg6;
 481         case 7:         return child->thread.debugreg7;
 482         }
 483         return 0;
 484 }
 485
 486 static int ptrace_set_debugreg(struct task_struct *child,
 487                                int n, unsigned long data)
 488 {
 489         int i;
 490
 491         if (unlikely(n == 4 || n == 5))
 492                 return -EIO;
 493
 494         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 495                 return -EIO;
 496
 497         switch (n) {
 498         case 0:         child->thread.debugreg0 = data; break;
 499         case 1:         child->thread.debugreg1 = data; break;
 500         case 2:         child->thread.debugreg2 = data; break;
 501         case 3:         child->thread.debugreg3 = data; break;
 502
 503         case 6:
 504                 if ((data & ~0xffffffffUL) != 0)
 505                         return -EIO;
 506                 child->thread.debugreg6 = data;
 507                 break;
 508
 509         case 7:
 510                 /*
 511                  * Sanity-check data. Take one half-byte at once with
 512                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 513                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 514                  * 2 and 3 are LENi. Given a list of invalid values,
 515                  * we do mask |= 1 << invalid_value, so that
 516                  * (mask >> check) & 1 is a correct test for invalid
 517                  * values.
 518                  *
 519                  * R/Wi contains the type of the breakpoint /
 520                  * watchpoint, LENi contains the length of the watched
 521                  * data in the watchpoint case.
 522                  *
 523                  * The invalid values are:
 524                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 525                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 526                  *   mask |= 0x4444.
 527                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 528                  *   0x1110.
 529                  *
 530                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 531                  *
 532                  * See the Intel Manual "System Programming Guide",
 533                  * 15.2.4
 534                  *
 535                  * Note that LENi == 0x10 is defined on x86_64 in long
 536                  * mode (i.e. even for 32-bit userspace software, but
 537                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 538                  * See the AMD manual no. 24593 (AMD64 System Programming)
 539                  */
 540 #ifdef CONFIG_X86_32
 541 #define DR7_MASK        0x5f54
 542 #else
 543 #define DR7_MASK        0x5554
 544 #endif
 545                 data &= ~DR_CONTROL_RESERVED;
 546                 for (i = 0; i < 4; i++)
 547                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 548                                 return -EIO;
 549                 child->thread.debugreg7 = data;
 550                 if (data)
 551                         set_tsk_thread_flag(child, TIF_DEBUG);
 552                 else
 553                         clear_tsk_thread_flag(child, TIF_DEBUG);
 554                 break;
 555         }
 556
 557         return 0;
 558 }
 559
 560 /*
 561  * These access the current or another (stopped) task's io permission
 562  * bitmap for debugging or core dump.
 563  */
 564 static int ioperm_active(struct task_struct *target,
 565                          const struct user_regset *regset)
 566 {
 567         return target->thread.io_bitmap_max / regset->size;
 568 }
 569
 570 static int ioperm_get(struct task_struct *target,
 571                       const struct user_regset *regset,
 572                       unsigned int pos, unsigned int count,
 573                       void *kbuf, void __user *ubuf)
 574 {
 575         if (!target->thread.io_bitmap_ptr)
 576                 return -ENXIO;
 577
 578         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 579                                    target->thread.io_bitmap_ptr,
 580                                    0, IO_BITMAP_BYTES);
 581 }
 582
 583 #ifdef CONFIG_X86_PTRACE_BTS
 584 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 585                                   struct bts_struct __user *out)
 586 {
 587         const struct bts_trace *trace;
 588         struct bts_struct bts;
 589         const unsigned char *at;
 590         int error;
 591
 592         trace = ds_read_bts(child->bts);
 593         if (!trace)
 594                 return -EPERM;
 595
 596         at = trace->ds.top - ((index + 1) * trace->ds.size);
 597         if ((void *)at < trace->ds.begin)
 598                 at += (trace->ds.n * trace->ds.size);
 599
 600         if (!trace->read)
 601                 return -EOPNOTSUPP;
 602
 603         error = trace->read(child->bts, at, &bts);
 604         if (error < 0)
 605                 return error;
 606
 607         if (copy_to_user(out, &bts, sizeof(bts)))
 608                 return -EFAULT;
 609
 610         return sizeof(bts);
 611 }
 612
 613 static int ptrace_bts_drain(struct task_struct *child,
 614                             long size,
 615                             struct bts_struct __user *out)
 616 {
 617         const struct bts_trace *trace;
 618         const unsigned char *at;
 619         int error, drained = 0;
 620
 621         trace = ds_read_bts(child->bts);
 622         if (!trace)
 623                 return -EPERM;
 624
 625         if (!trace->read)
 626                 return -EOPNOTSUPP;
 627
 628         if (size < (trace->ds.top - trace->ds.begin))
 629                 return -EIO;
 630
 631         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 632              out++, drained++, at += trace->ds.size) {
 633                 struct bts_struct bts;
 634                 int error;
 635
 636                 error = trace->read(child->bts, at, &bts);
 637                 if (error < 0)
 638                         return error;
 639
 640                 if (copy_to_user(out, &bts, sizeof(bts)))
 641                         return -EFAULT;
 642         }
 643
 644         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 645
 646         error = ds_reset_bts(child->bts);
 647         if (error < 0)
 648                 return error;
 649
 650         return drained;
 651 }
 652
 653 static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
 654 {
 655         child->bts_buffer = alloc_locked_buffer(size);
 656         if (!child->bts_buffer)
 657                 return -ENOMEM;
 658
 659         child->bts_size = size;
 660
 661         return 0;
 662 }
 663
 664 static void ptrace_bts_free_buffer(struct task_struct *child)
 665 {
 666         free_locked_buffer(child->bts_buffer, child->bts_size);
 667         child->bts_buffer = NULL;
 668         child->bts_size = 0;
 669 }
 670
 671 static int ptrace_bts_config(struct task_struct *child,
 672                              long cfg_size,
 673                              const struct ptrace_bts_config __user *ucfg)
 674 {
 675         struct ptrace_bts_config cfg;
 676         unsigned int flags = 0;
 677
 678         if (cfg_size < sizeof(cfg))
 679                 return -EIO;
 680
 681         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 682                 return -EFAULT;
 683
 684         if (child->bts) {
 685                 ds_release_bts(child->bts);
 686                 child->bts = NULL;
 687         }
 688
 689         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 690                 if (!cfg.signal)
 691                         return -EINVAL;
 692
 693                 return -EOPNOTSUPP;
 694
 695                 child->thread.bts_ovfl_signal = cfg.signal;
 696         }
 697
 698         if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
 699             (cfg.size != child->bts_size)) {
 700                 int error;
 701
 702                 ptrace_bts_free_buffer(child);
 703
 704                 error = ptrace_bts_allocate_buffer(child, cfg.size);
 705                 if (error < 0)
 706                         return error;
 707         }
 708
 709         if (cfg.flags & PTRACE_BTS_O_TRACE)
 710                 flags |= BTS_USER;
 711
 712         if (cfg.flags & PTRACE_BTS_O_SCHED)
 713                 flags |= BTS_TIMESTAMPS;
 714
 715         child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
 716                                     /* ovfl = */ NULL, /* th = */ (size_t)-1,
 717                                     flags);
 718         if (IS_ERR(child->bts)) {
 719                 int error = PTR_ERR(child->bts);
 720
 721                 ptrace_bts_free_buffer(child);
 722                 child->bts = NULL;
 723
 724                 return error;
 725         }
 726
 727         return sizeof(cfg);
 728 }
 729
 730 static int ptrace_bts_status(struct task_struct *child,
 731                              long cfg_size,
 732                              struct ptrace_bts_config __user *ucfg)
 733 {
 734         const struct bts_trace *trace;
 735         struct ptrace_bts_config cfg;
 736
 737         if (cfg_size < sizeof(cfg))
 738                 return -EIO;
 739
 740         trace = ds_read_bts(child->bts);
 741         if (!trace)
 742                 return -EPERM;
 743
 744         memset(&cfg, 0, sizeof(cfg));
 745         cfg.size = trace->ds.end - trace->ds.begin;
 746         cfg.signal = child->thread.bts_ovfl_signal;
 747         cfg.bts_size = sizeof(struct bts_struct);
 748
 749         if (cfg.signal)
 750                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 751
 752         if (trace->ds.flags & BTS_USER)
 753                 cfg.flags |= PTRACE_BTS_O_TRACE;
 754
 755         if (trace->ds.flags & BTS_TIMESTAMPS)
 756                 cfg.flags |= PTRACE_BTS_O_SCHED;
 757
 758         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 759                 return -EFAULT;
 760
 761         return sizeof(cfg);
 762 }
 763
 764 static int ptrace_bts_clear(struct task_struct *child)
 765 {
 766         const struct bts_trace *trace;
 767
 768         trace = ds_read_bts(child->bts);
 769         if (!trace)
 770                 return -EPERM;
 771
 772         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 773
 774         return ds_reset_bts(child->bts);
 775 }
 776
 777 static int ptrace_bts_size(struct task_struct *child)
 778 {
 779         const struct bts_trace *trace;
 780
 781         trace = ds_read_bts(child->bts);
 782         if (!trace)
 783                 return -EPERM;
 784
 785         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 786 }
 787
 788 static void ptrace_bts_fork(struct task_struct *tsk)
 789 {
 790         tsk->bts = NULL;
 791         tsk->bts_buffer = NULL;
 792         tsk->bts_size = 0;
 793         tsk->thread.bts_ovfl_signal = 0;
 794 }
 795
 796 static void ptrace_bts_untrace(struct task_struct *child)
 797 {
 798         if (unlikely(child->bts)) {
 799                 ds_release_bts(child->bts);
 800                 child->bts = NULL;
 801
 802                 /* We cannot update total_vm and locked_vm since
 803                    child's mm is already gone. But we can reclaim the
 804                    memory. */
 805                 kfree(child->bts_buffer);
 806                 child->bts_buffer = NULL;
 807                 child->bts_size = 0;
 808         }
 809 }
 810
 811 static void ptrace_bts_detach(struct task_struct *child)
 812 {
 813         if (unlikely(child->bts)) {
 814                 ds_release_bts(child->bts);
 815                 child->bts = NULL;
 816
 817                 ptrace_bts_free_buffer(child);
 818         }
 819 }
 820 #else
 821 static inline void ptrace_bts_fork(struct task_struct *tsk) {}
 822 static inline void ptrace_bts_detach(struct task_struct *child) {}
 823 static inline void ptrace_bts_untrace(struct task_struct *child) {}
 824 #endif /* CONFIG_X86_PTRACE_BTS */
 825
 826 void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
 827 {
 828         ptrace_bts_fork(child);
 829 }
 830
 831 void x86_ptrace_untrace(struct task_struct *child)
 832 {
 833         ptrace_bts_untrace(child);
 834 }
 835
 836 /*
 837  * Called by kernel/ptrace.c when detaching..
 838  *
 839  * Make sure the single step bit is not set.
 840  */
 841 void ptrace_disable(struct task_struct *child)
 842 {
 843         user_disable_single_step(child);
 844 #ifdef TIF_SYSCALL_EMU
 845         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 846 #endif
 847         ptrace_bts_detach(child);
 848 }
 849
 850 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 851 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 852 #endif
 853
 854 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 855 {
 856         int ret;
 857         unsigned long __user *datap = (unsigned long __user *)data;
 858
 859         switch (request) {
 860         /* read the word at location addr in the USER area. */
 861         case PTRACE_PEEKUSR: {
 862                 unsigned long tmp;
 863
 864                 ret = -EIO;
 865                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 866                     addr >= sizeof(struct user))
 867                         break;
 868
 869                 tmp = 0;  /* Default return condition */
 870                 if (addr < sizeof(struct user_regs_struct))
 871                         tmp = getreg(child, addr);
 872                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 873                          addr <= offsetof(struct user, u_debugreg[7])) {
 874                         addr -= offsetof(struct user, u_debugreg[0]);
 875                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 876                 }
 877                 ret = put_user(tmp, datap);
 878                 break;
 879         }
 880
 881         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 882                 ret = -EIO;
 883                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 884                     addr >= sizeof(struct user))
 885                         break;
 886
 887                 if (addr < sizeof(struct user_regs_struct))
 888                         ret = putreg(child, addr, data);
 889                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 890                          addr <= offsetof(struct user, u_debugreg[7])) {
 891                         addr -= offsetof(struct user, u_debugreg[0]);
 892                         ret = ptrace_set_debugreg(child,
 893                                                   addr / sizeof(data), data);
 894                 }
 895                 break;
 896
 897         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 898                 return copy_regset_to_user(child,
 899                                            task_user_regset_view(current),
 900                                            REGSET_GENERAL,
 901                                            0, sizeof(struct user_regs_struct),
 902                                            datap);
 903
 904         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 905                 return copy_regset_from_user(child,
 906                                              task_user_regset_view(current),
 907                                              REGSET_GENERAL,
 908                                              0, sizeof(struct user_regs_struct),
 909                                              datap);
 910
 911         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 912                 return copy_regset_to_user(child,
 913                                            task_user_regset_view(current),
 914                                            REGSET_FP,
 915                                            0, sizeof(struct user_i387_struct),
 916                                            datap);
 917
 918         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 919                 return copy_regset_from_user(child,
 920                                              task_user_regset_view(current),
 921                                              REGSET_FP,
 922                                              0, sizeof(struct user_i387_struct),
 923                                              datap);
 924
 925 #ifdef CONFIG_X86_32
 926         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 927                 return copy_regset_to_user(child, &user_x86_32_view,
 928                                            REGSET_XFP,
 929                                            0, sizeof(struct user_fxsr_struct),
 930                                            datap) ? -EIO : 0;
 931
 932         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 933                 return copy_regset_from_user(child, &user_x86_32_view,
 934                                              REGSET_XFP,
 935                                              0, sizeof(struct user_fxsr_struct),
 936                                              datap) ? -EIO : 0;
 937 #endif
 938
 939 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 940         case PTRACE_GET_THREAD_AREA:
 941                 if (addr < 0)
 942                         return -EIO;
 943                 ret = do_get_thread_area(child, addr,
 944                                          (struct user_desc __user *) data);
 945                 break;
 946
 947         case PTRACE_SET_THREAD_AREA:
 948                 if (addr < 0)
 949                         return -EIO;
 950                 ret = do_set_thread_area(child, addr,
 951                                          (struct user_desc __user *) data, 0);
 952                 break;
 953 #endif
 954
 955 #ifdef CONFIG_X86_64
 956                 /* normal 64bit interface to access TLS data.
 957                    Works just like arch_prctl, except that the arguments
 958                    are reversed. */
 959         case PTRACE_ARCH_PRCTL:
 960                 ret = do_arch_prctl(child, data, addr);
 961                 break;
 962 #endif
 963
 964         /*
 965          * These bits need more cooking - not enabled yet:
 966          */
 967 #ifdef CONFIG_X86_PTRACE_BTS
 968         case PTRACE_BTS_CONFIG:
 969                 ret = ptrace_bts_config
 970                         (child, data, (struct ptrace_bts_config __user *)addr);
 971                 break;
 972
 973         case PTRACE_BTS_STATUS:
 974                 ret = ptrace_bts_status
 975                         (child, data, (struct ptrace_bts_config __user *)addr);
 976                 break;
 977
 978         case PTRACE_BTS_SIZE:
 979                 ret = ptrace_bts_size(child);
 980                 break;
 981
 982         case PTRACE_BTS_GET:
 983                 ret = ptrace_bts_read_record
 984                         (child, data, (struct bts_struct __user *) addr);
 985                 break;
 986
 987         case PTRACE_BTS_CLEAR:
 988                 ret = ptrace_bts_clear(child);
 989                 break;
 990
 991         case PTRACE_BTS_DRAIN:
 992                 ret = ptrace_bts_drain
 993                         (child, data, (struct bts_struct __user *) addr);
 994                 break;
 995 #endif /* CONFIG_X86_PTRACE_BTS */
 996
 997         default:
 998                 ret = ptrace_request(child, request, addr, data);
 999                 break;
1000         }
1001
1002         return ret;
1003 }
1004
1005 #ifdef CONFIG_IA32_EMULATION
1006
1007 #include <linux/compat.h>
1008 #include <linux/syscalls.h>
1009 #include <asm/ia32.h>
1010 #include <asm/user32.h>
1011
1012 #define R32(l,q)                                                        \
1013         case offsetof(struct user32, regs.l):                           \
1014                 regs->q = value; break
1015
1016 #define SEG32(rs)                                                       \
1017         case offsetof(struct user32, regs.rs):                          \
1018                 return set_segment_reg(child,                           \
1019                                        offsetof(struct user_regs_struct, rs), \
1020                                        value);                          \
1021                 break
1022
1023 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1024 {
1025         struct pt_regs *regs = task_pt_regs(child);
1026
1027         switch (regno) {
1028
1029         SEG32(cs);
1030         SEG32(ds);
1031         SEG32(es);
1032         SEG32(fs);
1033         SEG32(gs);
1034         SEG32(ss);
1035
1036         R32(ebx, bx);
1037         R32(ecx, cx);
1038         R32(edx, dx);
1039         R32(edi, di);
1040         R32(esi, si);
1041         R32(ebp, bp);
1042         R32(eax, ax);
1043         R32(eip, ip);
1044         R32(esp, sp);
1045
1046         case offsetof(struct user32, regs.orig_eax):
1047                 /*
1048                  * Sign-extend the value so that orig_eax = -1
1049                  * causes (long)orig_ax < 0 tests to fire correctly.
1050                  */
1051                 regs->orig_ax = (long) (s32) value;
1052                 break;
1053
1054         case offsetof(struct user32, regs.eflags):
1055                 return set_flags(child, value);
1056
1057         case offsetof(struct user32, u_debugreg[0]) ...
1058                 offsetof(struct user32, u_debugreg[7]):
1059                 regno -= offsetof(struct user32, u_debugreg[0]);
1060                 return ptrace_set_debugreg(child, regno / 4, value);
1061
1062         default:
1063                 if (regno > sizeof(struct user32) || (regno & 3))
1064                         return -EIO;
1065
1066                 /*
1067                  * Other dummy fields in the virtual user structure
1068                  * are ignored
1069                  */
1070                 break;
1071         }
1072         return 0;
1073 }
1074
1075 #undef R32
1076 #undef SEG32
1077
1078 #define R32(l,q)                                                        \
1079         case offsetof(struct user32, regs.l):                           \
1080                 *val = regs->q; break
1081
1082 #define SEG32(rs)                                                       \
1083         case offsetof(struct user32, regs.rs):                          \
1084                 *val = get_segment_reg(child,                           \
1085                                        offsetof(struct user_regs_struct, rs)); \
1086                 break
1087
1088 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1089 {
1090         struct pt_regs *regs = task_pt_regs(child);
1091
1092         switch (regno) {
1093
1094         SEG32(ds);
1095         SEG32(es);
1096         SEG32(fs);
1097         SEG32(gs);
1098
1099         R32(cs, cs);
1100         R32(ss, ss);
1101         R32(ebx, bx);
1102         R32(ecx, cx);
1103         R32(edx, dx);
1104         R32(edi, di);
1105         R32(esi, si);
1106         R32(ebp, bp);
1107         R32(eax, ax);
1108         R32(orig_eax, orig_ax);
1109         R32(eip, ip);
1110         R32(esp, sp);
1111
1112         case offsetof(struct user32, regs.eflags):
1113                 *val = get_flags(child);
1114                 break;
1115
1116         case offsetof(struct user32, u_debugreg[0]) ...
1117                 offsetof(struct user32, u_debugreg[7]):
1118                 regno -= offsetof(struct user32, u_debugreg[0]);
1119                 *val = ptrace_get_debugreg(child, regno / 4);
1120                 break;
1121
1122         default:
1123                 if (regno > sizeof(struct user32) || (regno & 3))
1124                         return -EIO;
1125
1126                 /*
1127                  * Other dummy fields in the virtual user structure
1128                  * are ignored
1129                  */
1130                 *val = 0;
1131                 break;
1132         }
1133         return 0;
1134 }
1135
1136 #undef R32
1137 #undef SEG32
1138
1139 static int genregs32_get(struct task_struct *target,
1140                          const struct user_regset *regset,
1141                          unsigned int pos, unsigned int count,
1142                          void *kbuf, void __user *ubuf)
1143 {
1144         if (kbuf) {
1145                 compat_ulong_t *k = kbuf;
1146                 while (count > 0) {
1147                         getreg32(target, pos, k++);
1148                         count -= sizeof(*k);
1149                         pos += sizeof(*k);
1150                 }
1151         } else {
1152                 compat_ulong_t __user *u = ubuf;
1153                 while (count > 0) {
1154                         compat_ulong_t word;
1155                         getreg32(target, pos, &word);
1156                         if (__put_user(word, u++))
1157                                 return -EFAULT;
1158                         count -= sizeof(*u);
1159                         pos += sizeof(*u);
1160                 }
1161         }
1162
1163         return 0;
1164 }
1165
1166 static int genregs32_set(struct task_struct *target,
1167                          const struct user_regset *regset,
1168                          unsigned int pos, unsigned int count,
1169                          const void *kbuf, const void __user *ubuf)
1170 {
1171         int ret = 0;
1172         if (kbuf) {
1173                 const compat_ulong_t *k = kbuf;
1174                 while (count > 0 && !ret) {
1175                         ret = putreg32(target, pos, *k++);
1176                         count -= sizeof(*k);
1177                         pos += sizeof(*k);
1178                 }
1179         } else {
1180                 const compat_ulong_t __user *u = ubuf;
1181                 while (count > 0 && !ret) {
1182                         compat_ulong_t word;
1183                         ret = __get_user(word, u++);
1184                         if (ret)
1185                                 break;
1186                         ret = putreg32(target, pos, word);
1187                         count -= sizeof(*u);
1188                         pos += sizeof(*u);
1189                 }
1190         }
1191         return ret;
1192 }
1193
1194 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1195                         compat_ulong_t caddr, compat_ulong_t cdata)
1196 {
1197         unsigned long addr = caddr;
1198         unsigned long data = cdata;
1199         void __user *datap = compat_ptr(data);
1200         int ret;
1201         __u32 val;
1202
1203         switch (request) {
1204         case PTRACE_PEEKUSR:
1205                 ret = getreg32(child, addr, &val);
1206                 if (ret == 0)
1207                         ret = put_user(val, (__u32 __user *)datap);
1208                 break;
1209
1210         case PTRACE_POKEUSR:
1211                 ret = putreg32(child, addr, data);
1212                 break;
1213
1214         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1215                 return copy_regset_to_user(child, &user_x86_32_view,
1216                                            REGSET_GENERAL,
1217                                            0, sizeof(struct user_regs_struct32),
1218                                            datap);
1219
1220         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1221                 return copy_regset_from_user(child, &user_x86_32_view,
1222                                              REGSET_GENERAL, 0,
1223                                              sizeof(struct user_regs_struct32),
1224                                              datap);
1225
1226         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1227                 return copy_regset_to_user(child, &user_x86_32_view,
1228                                            REGSET_FP, 0,
1229                                            sizeof(struct user_i387_ia32_struct),
1230                                            datap);
1231
1232         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1233                 return copy_regset_from_user(
1234                         child, &user_x86_32_view, REGSET_FP,
1235                         0, sizeof(struct user_i387_ia32_struct), datap);
1236
1237         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1238                 return copy_regset_to_user(child, &user_x86_32_view,
1239                                            REGSET_XFP, 0,
1240                                            sizeof(struct user32_fxsr_struct),
1241                                            datap);
1242
1243         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1244                 return copy_regset_from_user(child, &user_x86_32_view,
1245                                              REGSET_XFP, 0,
1246                                              sizeof(struct user32_fxsr_struct),
1247                                              datap);
1248
1249         case PTRACE_GET_THREAD_AREA:
1250         case PTRACE_SET_THREAD_AREA:
1251 #ifdef CONFIG_X86_PTRACE_BTS
1252         case PTRACE_BTS_CONFIG:
1253         case PTRACE_BTS_STATUS:
1254         case PTRACE_BTS_SIZE:
1255         case PTRACE_BTS_GET:
1256         case PTRACE_BTS_CLEAR:
1257         case PTRACE_BTS_DRAIN:
1258 #endif /* CONFIG_X86_PTRACE_BTS */
1259                 return arch_ptrace(child, request, addr, data);
1260
1261         default:
1262                 return compat_ptrace_request(child, request, addr, data);
1263         }
1264
1265         return ret;
1266 }
1267
1268 #endif  /* CONFIG_IA32_EMULATION */
1269
1270 #ifdef CONFIG_X86_64
1271
1272 static const struct user_regset x86_64_regsets[] = {
1273         [REGSET_GENERAL] = {
1274                 .core_note_type = NT_PRSTATUS,
1275                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1276                 .size = sizeof(long), .align = sizeof(long),
1277                 .get = genregs_get, .set = genregs_set
1278         },
1279         [REGSET_FP] = {
1280                 .core_note_type = NT_PRFPREG,
1281                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1282                 .size = sizeof(long), .align = sizeof(long),
1283                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1284         },
1285         [REGSET_IOPERM64] = {
1286                 .core_note_type = NT_386_IOPERM,
1287                 .n = IO_BITMAP_LONGS,
1288                 .size = sizeof(long), .align = sizeof(long),
1289                 .active = ioperm_active, .get = ioperm_get
1290         },
1291 };
1292
1293 static const struct user_regset_view user_x86_64_view = {
1294         .name = "x86_64", .e_machine = EM_X86_64,
1295         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1296 };
1297
1298 #else  /* CONFIG_X86_32 */
1299
1300 #define user_regs_struct32      user_regs_struct
1301 #define genregs32_get           genregs_get
1302 #define genregs32_set           genregs_set
1303
1304 #define user_i387_ia32_struct   user_i387_struct
1305 #define user32_fxsr_struct      user_fxsr_struct
1306
1307 #endif  /* CONFIG_X86_64 */
1308
1309 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1310 static const struct user_regset x86_32_regsets[] = {
1311         [REGSET_GENERAL] = {
1312                 .core_note_type = NT_PRSTATUS,
1313                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1314                 .size = sizeof(u32), .align = sizeof(u32),
1315                 .get = genregs32_get, .set = genregs32_set
1316         },
1317         [REGSET_FP] = {
1318                 .core_note_type = NT_PRFPREG,
1319                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1320                 .size = sizeof(u32), .align = sizeof(u32),
1321                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1322         },
1323         [REGSET_XFP] = {
1324                 .core_note_type = NT_PRXFPREG,
1325                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1326                 .size = sizeof(u32), .align = sizeof(u32),
1327                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1328         },
1329         [REGSET_TLS] = {
1330                 .core_note_type = NT_386_TLS,
1331                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1332                 .size = sizeof(struct user_desc),
1333                 .align = sizeof(struct user_desc),
1334                 .active = regset_tls_active,
1335                 .get = regset_tls_get, .set = regset_tls_set
1336         },
1337         [REGSET_IOPERM32] = {
1338                 .core_note_type = NT_386_IOPERM,
1339                 .n = IO_BITMAP_BYTES / sizeof(u32),
1340                 .size = sizeof(u32), .align = sizeof(u32),
1341                 .active = ioperm_active, .get = ioperm_get
1342         },
1343 };
1344
1345 static const struct user_regset_view user_x86_32_view = {
1346         .name = "i386", .e_machine = EM_386,
1347         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1348 };
1349 #endif
1350
1351 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1352 {
1353 #ifdef CONFIG_IA32_EMULATION
1354         if (test_tsk_thread_flag(task, TIF_IA32))
1355 #endif
1356 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1357                 return &user_x86_32_view;
1358 #endif
1359 #ifdef CONFIG_X86_64
1360         return &user_x86_64_view;
1361 #endif
1362 }
1363
1364 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1365                                          int error_code, int si_code)
1366 {
1367         struct siginfo info;
1368
1369         tsk->thread.trap_no = 1;
1370         tsk->thread.error_code = error_code;
1371
1372         memset(&info, 0, sizeof(info));
1373         info.si_signo = SIGTRAP;
1374         info.si_code = si_code;
1375
1376         /* User-mode ip? */
1377         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1378
1379         /* Send us the fake SIGTRAP */
1380         force_sig_info(SIGTRAP, &info, tsk);
1381 }
1382
1383
1384 #ifdef CONFIG_X86_32
1385 # define IS_IA32        1
1386 #elif defined CONFIG_IA32_EMULATION
1387 # define IS_IA32        test_thread_flag(TIF_IA32)
1388 #else
1389 # define IS_IA32        0
1390 #endif
1391
1392 /*
1393  * We must return the syscall number to actually look up in the table.
1394  * This can be -1L to skip running any syscall at all.
1395  */
1396 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1397 {
1398         long ret = 0;
1399
1400         /*
1401          * If we stepped into a sysenter/syscall insn, it trapped in
1402          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1403          * If user-mode had set TF itself, then it's still clear from
1404          * do_debug() and we need to set it again to restore the user
1405          * state.  If we entered on the slow path, TF was already set.
1406          */
1407         if (test_thread_flag(TIF_SINGLESTEP))
1408                 regs->flags |= X86_EFLAGS_TF;
1409
1410         /* do the secure computing check first */
1411         secure_computing(regs->orig_ax);
1412
1413         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1414                 ret = -1L;
1415
1416         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1417             tracehook_report_syscall_entry(regs))
1418                 ret = -1L;
1419
1420         if (unlikely(current->audit_context)) {
1421                 if (IS_IA32)
1422                         audit_syscall_entry(AUDIT_ARCH_I386,
1423                                             regs->orig_ax,
1424                                             regs->bx, regs->cx,
1425                                             regs->dx, regs->si);
1426 #ifdef CONFIG_X86_64
1427                 else
1428                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1429                                             regs->orig_ax,
1430                                             regs->di, regs->si,
1431                                             regs->dx, regs->r10);
1432 #endif
1433         }
1434
1435         return ret ?: regs->orig_ax;
1436 }
1437
1438 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1439 {
1440         if (unlikely(current->audit_context))
1441                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1442
1443         if (test_thread_flag(TIF_SYSCALL_TRACE))
1444                 tracehook_report_syscall_exit(regs, 0);
1445
1446         /*
1447          * If TIF_SYSCALL_EMU is set, we only get here because of
1448          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1449          * We already reported this syscall instruction in
1450          * syscall_trace_enter(), so don't do any more now.
1451          */
1452         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1453                 return;
1454
1455         /*
1456          * If we are single-stepping, synthesize a trap to follow the
1457          * system call instruction.
1458          */
1459         if (test_thread_flag(TIF_SINGLESTEP) &&
1460             tracehook_consider_fatal_signal(current, SIGTRAP, SIG_DFL))
1461                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1462 }