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/user.h>
  18 #include <linux/elf.h>
  19 #include <linux/security.h>
  20 #include <linux/audit.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/signal.h>
  23
  24 #include <asm/uaccess.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/system.h>
  27 #include <asm/processor.h>
  28 #include <asm/i387.h>
  29 #include <asm/debugreg.h>
  30 #include <asm/ldt.h>
  31 #include <asm/desc.h>
  32 #include <asm/prctl.h>
  33 #include <asm/proto.h>
  34 #include <asm/ds.h>
  35
  36 #include "tls.h"
  37
  38 enum x86_regset {
  39         REGSET_GENERAL,
  40         REGSET_FP,
  41         REGSET_XFP,
  42         REGSET_TLS,
  43 };
  44
  45 /*
  46  * does not yet catch signals sent when the child dies.
  47  * in exit.c or in signal.c.
  48  */
  49
  50 /*
  51  * Determines which flags the user has access to [1 = access, 0 = no access].
  52  */
  53 #define FLAG_MASK_32            ((unsigned long)                        \
  54                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  55                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  56                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  57                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  58                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  59
  60 /*
  61  * Determines whether a value may be installed in a segment register.
  62  */
  63 static inline bool invalid_selector(u16 value)
  64 {
  65         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  66 }
  67
  68 #ifdef CONFIG_X86_32
  69
  70 #define FLAG_MASK               FLAG_MASK_32
  71
  72 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  73 {
  74         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  75         regno >>= 2;
  76         if (regno > FS)
  77                 --regno;
  78         return &regs->bx + regno;
  79 }
  80
  81 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  82 {
  83         /*
  84          * Returning the value truncates it to 16 bits.
  85          */
  86         unsigned int retval;
  87         if (offset != offsetof(struct user_regs_struct, gs))
  88                 retval = *pt_regs_access(task_pt_regs(task), offset);
  89         else {
  90                 retval = task->thread.gs;
  91                 if (task == current)
  92                         savesegment(gs, retval);
  93         }
  94         return retval;
  95 }
  96
  97 static int set_segment_reg(struct task_struct *task,
  98                            unsigned long offset, u16 value)
  99 {
 100         /*
 101          * The value argument was already truncated to 16 bits.
 102          */
 103         if (invalid_selector(value))
 104                 return -EIO;
 105
 106         /*
 107          * For %cs and %ss we cannot permit a null selector.
 108          * We can permit a bogus selector as long as it has USER_RPL.
 109          * Null selectors are fine for other segment registers, but
 110          * we will never get back to user mode with invalid %cs or %ss
 111          * and will take the trap in iret instead.  Much code relies
 112          * on user_mode() to distinguish a user trap frame (which can
 113          * safely use invalid selectors) from a kernel trap frame.
 114          */
 115         switch (offset) {
 116         case offsetof(struct user_regs_struct, cs):
 117         case offsetof(struct user_regs_struct, ss):
 118                 if (unlikely(value == 0))
 119                         return -EIO;
 120
 121         default:
 122                 *pt_regs_access(task_pt_regs(task), offset) = value;
 123                 break;
 124
 125         case offsetof(struct user_regs_struct, gs):
 126                 task->thread.gs = value;
 127                 if (task == current)
 128                         /*
 129                          * The user-mode %gs is not affected by
 130                          * kernel entry, so we must update the CPU.
 131                          */
 132                         loadsegment(gs, value);
 133         }
 134
 135         return 0;
 136 }
 137
 138 static unsigned long debugreg_addr_limit(struct task_struct *task)
 139 {
 140         return TASK_SIZE - 3;
 141 }
 142
 143 #else  /* CONFIG_X86_64 */
 144
 145 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 146
 147 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 148 {
 149         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 150         return &regs->r15 + (offset / sizeof(regs->r15));
 151 }
 152
 153 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 154 {
 155         /*
 156          * Returning the value truncates it to 16 bits.
 157          */
 158         unsigned int seg;
 159
 160         switch (offset) {
 161         case offsetof(struct user_regs_struct, fs):
 162                 if (task == current) {
 163                         /* Older gas can't assemble movq %?s,%r?? */
 164                         asm("movl %%fs,%0" : "=r" (seg));
 165                         return seg;
 166                 }
 167                 return task->thread.fsindex;
 168         case offsetof(struct user_regs_struct, gs):
 169                 if (task == current) {
 170                         asm("movl %%gs,%0" : "=r" (seg));
 171                         return seg;
 172                 }
 173                 return task->thread.gsindex;
 174         case offsetof(struct user_regs_struct, ds):
 175                 if (task == current) {
 176                         asm("movl %%ds,%0" : "=r" (seg));
 177                         return seg;
 178                 }
 179                 return task->thread.ds;
 180         case offsetof(struct user_regs_struct, es):
 181                 if (task == current) {
 182                         asm("movl %%es,%0" : "=r" (seg));
 183                         return seg;
 184                 }
 185                 return task->thread.es;
 186
 187         case offsetof(struct user_regs_struct, cs):
 188         case offsetof(struct user_regs_struct, ss):
 189                 break;
 190         }
 191         return *pt_regs_access(task_pt_regs(task), offset);
 192 }
 193
 194 static int set_segment_reg(struct task_struct *task,
 195                            unsigned long offset, u16 value)
 196 {
 197         /*
 198          * The value argument was already truncated to 16 bits.
 199          */
 200         if (invalid_selector(value))
 201                 return -EIO;
 202
 203         switch (offset) {
 204         case offsetof(struct user_regs_struct,fs):
 205                 /*
 206                  * If this is setting fs as for normal 64-bit use but
 207                  * setting fs_base has implicitly changed it, leave it.
 208                  */
 209                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 210                      task->thread.fs != 0) ||
 211                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 212                      task->thread.fs == 0))
 213                         break;
 214                 task->thread.fsindex = value;
 215                 if (task == current)
 216                         loadsegment(fs, task->thread.fsindex);
 217                 break;
 218         case offsetof(struct user_regs_struct,gs):
 219                 /*
 220                  * If this is setting gs as for normal 64-bit use but
 221                  * setting gs_base has implicitly changed it, leave it.
 222                  */
 223                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 224                      task->thread.gs != 0) ||
 225                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 226                      task->thread.gs == 0))
 227                         break;
 228                 task->thread.gsindex = value;
 229                 if (task == current)
 230                         load_gs_index(task->thread.gsindex);
 231                 break;
 232         case offsetof(struct user_regs_struct,ds):
 233                 task->thread.ds = value;
 234                 if (task == current)
 235                         loadsegment(ds, task->thread.ds);
 236                 break;
 237         case offsetof(struct user_regs_struct,es):
 238                 task->thread.es = value;
 239                 if (task == current)
 240                         loadsegment(es, task->thread.es);
 241                 break;
 242
 243                 /*
 244                  * Can't actually change these in 64-bit mode.
 245                  */
 246         case offsetof(struct user_regs_struct,cs):
 247                 if (unlikely(value == 0))
 248                         return -EIO;
 249 #ifdef CONFIG_IA32_EMULATION
 250                 if (test_tsk_thread_flag(task, TIF_IA32))
 251                         task_pt_regs(task)->cs = value;
 252 #endif
 253                 break;
 254         case offsetof(struct user_regs_struct,ss):
 255                 if (unlikely(value == 0))
 256                         return -EIO;
 257 #ifdef CONFIG_IA32_EMULATION
 258                 if (test_tsk_thread_flag(task, TIF_IA32))
 259                         task_pt_regs(task)->ss = value;
 260 #endif
 261                 break;
 262         }
 263
 264         return 0;
 265 }
 266
 267 static unsigned long debugreg_addr_limit(struct task_struct *task)
 268 {
 269 #ifdef CONFIG_IA32_EMULATION
 270         if (test_tsk_thread_flag(task, TIF_IA32))
 271                 return IA32_PAGE_OFFSET - 3;
 272 #endif
 273         return TASK_SIZE64 - 7;
 274 }
 275
 276 #endif  /* CONFIG_X86_32 */
 277
 278 static unsigned long get_flags(struct task_struct *task)
 279 {
 280         unsigned long retval = task_pt_regs(task)->flags;
 281
 282         /*
 283          * If the debugger set TF, hide it from the readout.
 284          */
 285         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 286                 retval &= ~X86_EFLAGS_TF;
 287
 288         return retval;
 289 }
 290
 291 static int set_flags(struct task_struct *task, unsigned long value)
 292 {
 293         struct pt_regs *regs = task_pt_regs(task);
 294
 295         /*
 296          * If the user value contains TF, mark that
 297          * it was not "us" (the debugger) that set it.
 298          * If not, make sure it stays set if we had.
 299          */
 300         if (value & X86_EFLAGS_TF)
 301                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 302         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 303                 value |= X86_EFLAGS_TF;
 304
 305         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 306
 307         return 0;
 308 }
 309
 310 static int putreg(struct task_struct *child,
 311                   unsigned long offset, unsigned long value)
 312 {
 313         switch (offset) {
 314         case offsetof(struct user_regs_struct, cs):
 315         case offsetof(struct user_regs_struct, ds):
 316         case offsetof(struct user_regs_struct, es):
 317         case offsetof(struct user_regs_struct, fs):
 318         case offsetof(struct user_regs_struct, gs):
 319         case offsetof(struct user_regs_struct, ss):
 320                 return set_segment_reg(child, offset, value);
 321
 322         case offsetof(struct user_regs_struct, flags):
 323                 return set_flags(child, value);
 324
 325 #ifdef CONFIG_X86_64
 326         /*
 327          * Orig_ax is really just a flag with small positive and
 328          * negative values, so make sure to always sign-extend it
 329          * from 32 bits so that it works correctly regardless of
 330          * whether we come from a 32-bit environment or not.
 331          */
 332         case offsetof(struct user_regs_struct, orig_ax):
 333                 value = (long) (s32) value;
 334                 break;
 335
 336         case offsetof(struct user_regs_struct,fs_base):
 337                 if (value >= TASK_SIZE_OF(child))
 338                         return -EIO;
 339                 /*
 340                  * When changing the segment base, use do_arch_prctl
 341                  * to set either thread.fs or thread.fsindex and the
 342                  * corresponding GDT slot.
 343                  */
 344                 if (child->thread.fs != value)
 345                         return do_arch_prctl(child, ARCH_SET_FS, value);
 346                 return 0;
 347         case offsetof(struct user_regs_struct,gs_base):
 348                 /*
 349                  * Exactly the same here as the %fs handling above.
 350                  */
 351                 if (value >= TASK_SIZE_OF(child))
 352                         return -EIO;
 353                 if (child->thread.gs != value)
 354                         return do_arch_prctl(child, ARCH_SET_GS, value);
 355                 return 0;
 356 #endif
 357         }
 358
 359         *pt_regs_access(task_pt_regs(child), offset) = value;
 360         return 0;
 361 }
 362
 363 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 364 {
 365         switch (offset) {
 366         case offsetof(struct user_regs_struct, cs):
 367         case offsetof(struct user_regs_struct, ds):
 368         case offsetof(struct user_regs_struct, es):
 369         case offsetof(struct user_regs_struct, fs):
 370         case offsetof(struct user_regs_struct, gs):
 371         case offsetof(struct user_regs_struct, ss):
 372                 return get_segment_reg(task, offset);
 373
 374         case offsetof(struct user_regs_struct, flags):
 375                 return get_flags(task);
 376
 377 #ifdef CONFIG_X86_64
 378         case offsetof(struct user_regs_struct, fs_base): {
 379                 /*
 380                  * do_arch_prctl may have used a GDT slot instead of
 381                  * the MSR.  To userland, it appears the same either
 382                  * way, except the %fs segment selector might not be 0.
 383                  */
 384                 unsigned int seg = task->thread.fsindex;
 385                 if (task->thread.fs != 0)
 386                         return task->thread.fs;
 387                 if (task == current)
 388                         asm("movl %%fs,%0" : "=r" (seg));
 389                 if (seg != FS_TLS_SEL)
 390                         return 0;
 391                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 392         }
 393         case offsetof(struct user_regs_struct, gs_base): {
 394                 /*
 395                  * Exactly the same here as the %fs handling above.
 396                  */
 397                 unsigned int seg = task->thread.gsindex;
 398                 if (task->thread.gs != 0)
 399                         return task->thread.gs;
 400                 if (task == current)
 401                         asm("movl %%gs,%0" : "=r" (seg));
 402                 if (seg != GS_TLS_SEL)
 403                         return 0;
 404                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 405         }
 406 #endif
 407         }
 408
 409         return *pt_regs_access(task_pt_regs(task), offset);
 410 }
 411
 412 static int genregs_get(struct task_struct *target,
 413                        const struct user_regset *regset,
 414                        unsigned int pos, unsigned int count,
 415                        void *kbuf, void __user *ubuf)
 416 {
 417         if (kbuf) {
 418                 unsigned long *k = kbuf;
 419                 while (count > 0) {
 420                         *k++ = getreg(target, pos);
 421                         count -= sizeof(*k);
 422                         pos += sizeof(*k);
 423                 }
 424         } else {
 425                 unsigned long __user *u = ubuf;
 426                 while (count > 0) {
 427                         if (__put_user(getreg(target, pos), u++))
 428                                 return -EFAULT;
 429                         count -= sizeof(*u);
 430                         pos += sizeof(*u);
 431                 }
 432         }
 433
 434         return 0;
 435 }
 436
 437 static int genregs_set(struct task_struct *target,
 438                        const struct user_regset *regset,
 439                        unsigned int pos, unsigned int count,
 440                        const void *kbuf, const void __user *ubuf)
 441 {
 442         int ret = 0;
 443         if (kbuf) {
 444                 const unsigned long *k = kbuf;
 445                 while (count > 0 && !ret) {
 446                         ret = putreg(target, pos, *k++);
 447                         count -= sizeof(*k);
 448                         pos += sizeof(*k);
 449                 }
 450         } else {
 451                 const unsigned long  __user *u = ubuf;
 452                 while (count > 0 && !ret) {
 453                         unsigned long word;
 454                         ret = __get_user(word, u++);
 455                         if (ret)
 456                                 break;
 457                         ret = putreg(target, pos, word);
 458                         count -= sizeof(*u);
 459                         pos += sizeof(*u);
 460                 }
 461         }
 462         return ret;
 463 }
 464
 465 /*
 466  * This function is trivial and will be inlined by the compiler.
 467  * Having it separates the implementation details of debug
 468  * registers from the interface details of ptrace.
 469  */
 470 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 471 {
 472         switch (n) {
 473         case 0:         return child->thread.debugreg0;
 474         case 1:         return child->thread.debugreg1;
 475         case 2:         return child->thread.debugreg2;
 476         case 3:         return child->thread.debugreg3;
 477         case 6:         return child->thread.debugreg6;
 478         case 7:         return child->thread.debugreg7;
 479         }
 480         return 0;
 481 }
 482
 483 static int ptrace_set_debugreg(struct task_struct *child,
 484                                int n, unsigned long data)
 485 {
 486         int i;
 487
 488         if (unlikely(n == 4 || n == 5))
 489                 return -EIO;
 490
 491         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 492                 return -EIO;
 493
 494         switch (n) {
 495         case 0:         child->thread.debugreg0 = data; break;
 496         case 1:         child->thread.debugreg1 = data; break;
 497         case 2:         child->thread.debugreg2 = data; break;
 498         case 3:         child->thread.debugreg3 = data; break;
 499
 500         case 6:
 501                 if ((data & ~0xffffffffUL) != 0)
 502                         return -EIO;
 503                 child->thread.debugreg6 = data;
 504                 break;
 505
 506         case 7:
 507                 /*
 508                  * Sanity-check data. Take one half-byte at once with
 509                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 510                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 511                  * 2 and 3 are LENi. Given a list of invalid values,
 512                  * we do mask |= 1 << invalid_value, so that
 513                  * (mask >> check) & 1 is a correct test for invalid
 514                  * values.
 515                  *
 516                  * R/Wi contains the type of the breakpoint /
 517                  * watchpoint, LENi contains the length of the watched
 518                  * data in the watchpoint case.
 519                  *
 520                  * The invalid values are:
 521                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 522                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 523                  *   mask |= 0x4444.
 524                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 525                  *   0x1110.
 526                  *
 527                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 528                  *
 529                  * See the Intel Manual "System Programming Guide",
 530                  * 15.2.4
 531                  *
 532                  * Note that LENi == 0x10 is defined on x86_64 in long
 533                  * mode (i.e. even for 32-bit userspace software, but
 534                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 535                  * See the AMD manual no. 24593 (AMD64 System Programming)
 536                  */
 537 #ifdef CONFIG_X86_32
 538 #define DR7_MASK        0x5f54
 539 #else
 540 #define DR7_MASK        0x5554
 541 #endif
 542                 data &= ~DR_CONTROL_RESERVED;
 543                 for (i = 0; i < 4; i++)
 544                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 545                                 return -EIO;
 546                 child->thread.debugreg7 = data;
 547                 if (data)
 548                         set_tsk_thread_flag(child, TIF_DEBUG);
 549                 else
 550                         clear_tsk_thread_flag(child, TIF_DEBUG);
 551                 break;
 552         }
 553
 554         return 0;
 555 }
 556
 557 #ifdef X86_BTS
 558
 559 static int ptrace_bts_get_size(struct task_struct *child)
 560 {
 561         if (!child->thread.ds_area_msr)
 562                 return -ENXIO;
 563
 564         return ds_get_bts_index((void *)child->thread.ds_area_msr);
 565 }
 566
 567 static int ptrace_bts_read_record(struct task_struct *child,
 568                                   long index,
 569                                   struct bts_struct __user *out)
 570 {
 571         struct bts_struct ret;
 572         int retval;
 573         int bts_end;
 574         int bts_index;
 575
 576         if (!child->thread.ds_area_msr)
 577                 return -ENXIO;
 578
 579         if (index < 0)
 580                 return -EINVAL;
 581
 582         bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
 583         if (bts_end <= index)
 584                 return -EINVAL;
 585
 586         /* translate the ptrace bts index into the ds bts index */
 587         bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
 588         bts_index -= (index + 1);
 589         if (bts_index < 0)
 590                 bts_index += bts_end;
 591
 592         retval = ds_read_bts((void *)child->thread.ds_area_msr,
 593                              bts_index, &ret);
 594         if (retval < 0)
 595                 return retval;
 596
 597         if (copy_to_user(out, &ret, sizeof(ret)))
 598                 return -EFAULT;
 599
 600         return sizeof(ret);
 601 }
 602
 603 static int ptrace_bts_clear(struct task_struct *child)
 604 {
 605         if (!child->thread.ds_area_msr)
 606                 return -ENXIO;
 607
 608         return ds_clear((void *)child->thread.ds_area_msr);
 609 }
 610
 611 static int ptrace_bts_drain(struct task_struct *child,
 612                             long size,
 613                             struct bts_struct __user *out)
 614 {
 615         int end, i;
 616         void *ds = (void *)child->thread.ds_area_msr;
 617
 618         if (!ds)
 619                 return -ENXIO;
 620
 621         end = ds_get_bts_index(ds);
 622         if (end <= 0)
 623                 return end;
 624
 625         if (size < (end * sizeof(struct bts_struct)))
 626                 return -EIO;
 627
 628         for (i = 0; i < end; i++, out++) {
 629                 struct bts_struct ret;
 630                 int retval;
 631
 632                 retval = ds_read_bts(ds, i, &ret);
 633                 if (retval < 0)
 634                         return retval;
 635
 636                 if (copy_to_user(out, &ret, sizeof(ret)))
 637                         return -EFAULT;
 638         }
 639
 640         ds_clear(ds);
 641
 642         return end;
 643 }
 644
 645 static int ptrace_bts_config(struct task_struct *child,
 646                              long cfg_size,
 647                              const struct ptrace_bts_config __user *ucfg)
 648 {
 649         struct ptrace_bts_config cfg;
 650         int bts_size, ret = 0;
 651         void *ds;
 652
 653         if (cfg_size < sizeof(cfg))
 654                 return -EIO;
 655
 656         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 657                 return -EFAULT;
 658
 659         if ((int)cfg.size < 0)
 660                 return -EINVAL;
 661
 662         bts_size = 0;
 663         ds = (void *)child->thread.ds_area_msr;
 664         if (ds) {
 665                 bts_size = ds_get_bts_size(ds);
 666                 if (bts_size < 0)
 667                         return bts_size;
 668         }
 669         cfg.size = PAGE_ALIGN(cfg.size);
 670
 671         if (bts_size != cfg.size) {
 672                 ret = ptrace_bts_realloc(child, cfg.size,
 673                                          cfg.flags & PTRACE_BTS_O_CUT_SIZE);
 674                 if (ret < 0)
 675                         goto errout;
 676
 677                 ds = (void *)child->thread.ds_area_msr;
 678         }
 679
 680         if (cfg.flags & PTRACE_BTS_O_SIGNAL)
 681                 ret = ds_set_overflow(ds, DS_O_SIGNAL);
 682         else
 683                 ret = ds_set_overflow(ds, DS_O_WRAP);
 684         if (ret < 0)
 685                 goto errout;
 686
 687         if (cfg.flags & PTRACE_BTS_O_TRACE)
 688                 child->thread.debugctlmsr |= ds_debugctl_mask();
 689         else
 690                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 691
 692         if (cfg.flags & PTRACE_BTS_O_SCHED)
 693                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 694         else
 695                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 696
 697         ret = sizeof(cfg);
 698
 699 out:
 700         if (child->thread.debugctlmsr)
 701                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 702         else
 703                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 704
 705         return ret;
 706
 707 errout:
 708         child->thread.debugctlmsr &= ~ds_debugctl_mask();
 709         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 710         goto out;
 711 }
 712
 713 static int ptrace_bts_status(struct task_struct *child,
 714                              long cfg_size,
 715                              struct ptrace_bts_config __user *ucfg)
 716 {
 717         void *ds = (void *)child->thread.ds_area_msr;
 718         struct ptrace_bts_config cfg;
 719
 720         if (cfg_size < sizeof(cfg))
 721                 return -EIO;
 722
 723         memset(&cfg, 0, sizeof(cfg));
 724
 725         if (ds) {
 726                 cfg.size = ds_get_bts_size(ds);
 727
 728                 if (ds_get_overflow(ds) == DS_O_SIGNAL)
 729                         cfg.flags |= PTRACE_BTS_O_SIGNAL;
 730
 731                 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
 732                     child->thread.debugctlmsr & ds_debugctl_mask())
 733                         cfg.flags |= PTRACE_BTS_O_TRACE;
 734
 735                 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
 736                         cfg.flags |= PTRACE_BTS_O_SCHED;
 737         }
 738
 739         cfg.bts_size = sizeof(struct bts_struct);
 740
 741         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 742                 return -EFAULT;
 743
 744         return sizeof(cfg);
 745 }
 746
 747
 748 static int ptrace_bts_write_record(struct task_struct *child,
 749                                    const struct bts_struct *in)
 750 {
 751         int retval;
 752
 753         if (!child->thread.ds_area_msr)
 754                 return -ENXIO;
 755
 756         retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
 757         if (retval)
 758                 return retval;
 759
 760         return sizeof(*in);
 761 }
 762
 763 static int ptrace_bts_realloc(struct task_struct *child,
 764                               int size, int reduce_size)
 765 {
 766         unsigned long rlim, vm;
 767         int ret, old_size;
 768
 769         if (size < 0)
 770                 return -EINVAL;
 771
 772         old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
 773         if (old_size < 0)
 774                 return old_size;
 775
 776         ret = ds_free((void **)&child->thread.ds_area_msr);
 777         if (ret < 0)
 778                 goto out;
 779
 780         size >>= PAGE_SHIFT;
 781         old_size >>= PAGE_SHIFT;
 782
 783         current->mm->total_vm  -= old_size;
 784         current->mm->locked_vm -= old_size;
 785
 786         if (size == 0)
 787                 goto out;
 788
 789         rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
 790         vm = current->mm->total_vm  + size;
 791         if (rlim < vm) {
 792                 ret = -ENOMEM;
 793
 794                 if (!reduce_size)
 795                         goto out;
 796
 797                 size = rlim - current->mm->total_vm;
 798                 if (size <= 0)
 799                         goto out;
 800         }
 801
 802         rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
 803         vm = current->mm->locked_vm  + size;
 804         if (rlim < vm) {
 805                 ret = -ENOMEM;
 806
 807                 if (!reduce_size)
 808                         goto out;
 809
 810                 size = rlim - current->mm->locked_vm;
 811                 if (size <= 0)
 812                         goto out;
 813         }
 814
 815         ret = ds_allocate((void **)&child->thread.ds_area_msr,
 816                           size << PAGE_SHIFT);
 817         if (ret < 0)
 818                 goto out;
 819
 820         current->mm->total_vm  += size;
 821         current->mm->locked_vm += size;
 822
 823 out:
 824         if (child->thread.ds_area_msr)
 825                 set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 826         else
 827                 clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 828
 829         return ret;
 830 }
 831
 832 void ptrace_bts_take_timestamp(struct task_struct *tsk,
 833                                enum bts_qualifier qualifier)
 834 {
 835         struct bts_struct rec = {
 836                 .qualifier = qualifier,
 837                 .variant.jiffies = jiffies_64
 838         };
 839
 840         ptrace_bts_write_record(tsk, &rec);
 841 }
 842 #endif /* X86_BTS */
 843
 844 /*
 845  * Called by kernel/ptrace.c when detaching..
 846  *
 847  * Make sure the single step bit is not set.
 848  */
 849 void ptrace_disable(struct task_struct *child)
 850 {
 851         user_disable_single_step(child);
 852 #ifdef TIF_SYSCALL_EMU
 853         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 854 #endif
 855         if (child->thread.ds_area_msr) {
 856 #ifdef X86_BTS
 857                 ptrace_bts_realloc(child, 0, 0);
 858 #endif
 859                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 860                 if (!child->thread.debugctlmsr)
 861                         clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 862                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 863         }
 864 }
 865
 866 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 867 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 868 #endif
 869
 870 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 871 {
 872         int ret;
 873         unsigned long __user *datap = (unsigned long __user *)data;
 874
 875         switch (request) {
 876         /* read the word at location addr in the USER area. */
 877         case PTRACE_PEEKUSR: {
 878                 unsigned long tmp;
 879
 880                 ret = -EIO;
 881                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 882                     addr >= sizeof(struct user))
 883                         break;
 884
 885                 tmp = 0;  /* Default return condition */
 886                 if (addr < sizeof(struct user_regs_struct))
 887                         tmp = getreg(child, addr);
 888                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 889                          addr <= offsetof(struct user, u_debugreg[7])) {
 890                         addr -= offsetof(struct user, u_debugreg[0]);
 891                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 892                 }
 893                 ret = put_user(tmp, datap);
 894                 break;
 895         }
 896
 897         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 898                 ret = -EIO;
 899                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 900                     addr >= sizeof(struct user))
 901                         break;
 902
 903                 if (addr < sizeof(struct user_regs_struct))
 904                         ret = putreg(child, addr, data);
 905                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 906                          addr <= offsetof(struct user, u_debugreg[7])) {
 907                         addr -= offsetof(struct user, u_debugreg[0]);
 908                         ret = ptrace_set_debugreg(child,
 909                                                   addr / sizeof(data), data);
 910                 }
 911                 break;
 912
 913         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 914                 return copy_regset_to_user(child,
 915                                            task_user_regset_view(current),
 916                                            REGSET_GENERAL,
 917                                            0, sizeof(struct user_regs_struct),
 918                                            datap);
 919
 920         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 921                 return copy_regset_from_user(child,
 922                                              task_user_regset_view(current),
 923                                              REGSET_GENERAL,
 924                                              0, sizeof(struct user_regs_struct),
 925                                              datap);
 926
 927         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 928                 return copy_regset_to_user(child,
 929                                            task_user_regset_view(current),
 930                                            REGSET_FP,
 931                                            0, sizeof(struct user_i387_struct),
 932                                            datap);
 933
 934         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 935                 return copy_regset_from_user(child,
 936                                              task_user_regset_view(current),
 937                                              REGSET_FP,
 938                                              0, sizeof(struct user_i387_struct),
 939                                              datap);
 940
 941 #ifdef CONFIG_X86_32
 942         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 943                 return copy_regset_to_user(child, &user_x86_32_view,
 944                                            REGSET_XFP,
 945                                            0, sizeof(struct user_fxsr_struct),
 946                                            datap) ? -EIO : 0;
 947
 948         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 949                 return copy_regset_from_user(child, &user_x86_32_view,
 950                                              REGSET_XFP,
 951                                              0, sizeof(struct user_fxsr_struct),
 952                                              datap) ? -EIO : 0;
 953 #endif
 954
 955 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 956         case PTRACE_GET_THREAD_AREA:
 957                 if (addr < 0)
 958                         return -EIO;
 959                 ret = do_get_thread_area(child, addr,
 960                                          (struct user_desc __user *) data);
 961                 break;
 962
 963         case PTRACE_SET_THREAD_AREA:
 964                 if (addr < 0)
 965                         return -EIO;
 966                 ret = do_set_thread_area(child, addr,
 967                                          (struct user_desc __user *) data, 0);
 968                 break;
 969 #endif
 970
 971 #ifdef CONFIG_X86_64
 972                 /* normal 64bit interface to access TLS data.
 973                    Works just like arch_prctl, except that the arguments
 974                    are reversed. */
 975         case PTRACE_ARCH_PRCTL:
 976                 ret = do_arch_prctl(child, data, addr);
 977                 break;
 978 #endif
 979
 980         /*
 981          * These bits need more cooking - not enabled yet:
 982          */
 983 #ifdef X86_BTS
 984         case PTRACE_BTS_CONFIG:
 985                 ret = ptrace_bts_config
 986                         (child, data, (struct ptrace_bts_config __user *)addr);
 987                 break;
 988
 989         case PTRACE_BTS_STATUS:
 990                 ret = ptrace_bts_status
 991                         (child, data, (struct ptrace_bts_config __user *)addr);
 992                 break;
 993
 994         case PTRACE_BTS_SIZE:
 995                 ret = ptrace_bts_get_size(child);
 996                 break;
 997
 998         case PTRACE_BTS_GET:
 999                 ret = ptrace_bts_read_record
1000                         (child, data, (struct bts_struct __user *) addr);
1001                 break;
1002
1003         case PTRACE_BTS_CLEAR:
1004                 ret = ptrace_bts_clear(child);
1005                 break;
1006
1007         case PTRACE_BTS_DRAIN:
1008                 ret = ptrace_bts_drain
1009                         (child, data, (struct bts_struct __user *) addr);
1010                 break;
1011 #endif
1012
1013         default:
1014                 ret = ptrace_request(child, request, addr, data);
1015                 break;
1016         }
1017
1018         return ret;
1019 }
1020
1021 #ifdef CONFIG_IA32_EMULATION
1022
1023 #include <linux/compat.h>
1024 #include <linux/syscalls.h>
1025 #include <asm/ia32.h>
1026 #include <asm/user32.h>
1027
1028 #define R32(l,q)                                                        \
1029         case offsetof(struct user32, regs.l):                           \
1030                 regs->q = value; break
1031
1032 #define SEG32(rs)                                                       \
1033         case offsetof(struct user32, regs.rs):                          \
1034                 return set_segment_reg(child,                           \
1035                                        offsetof(struct user_regs_struct, rs), \
1036                                        value);                          \
1037                 break
1038
1039 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1040 {
1041         struct pt_regs *regs = task_pt_regs(child);
1042
1043         switch (regno) {
1044
1045         SEG32(cs);
1046         SEG32(ds);
1047         SEG32(es);
1048         SEG32(fs);
1049         SEG32(gs);
1050         SEG32(ss);
1051
1052         R32(ebx, bx);
1053         R32(ecx, cx);
1054         R32(edx, dx);
1055         R32(edi, di);
1056         R32(esi, si);
1057         R32(ebp, bp);
1058         R32(eax, ax);
1059         R32(eip, ip);
1060         R32(esp, sp);
1061
1062         case offsetof(struct user32, regs.orig_eax):
1063                 /*
1064                  * Sign-extend the value so that orig_eax = -1
1065                  * causes (long)orig_ax < 0 tests to fire correctly.
1066                  */
1067                 regs->orig_ax = (long) (s32) value;
1068                 break;
1069
1070         case offsetof(struct user32, regs.eflags):
1071                 return set_flags(child, value);
1072
1073         case offsetof(struct user32, u_debugreg[0]) ...
1074                 offsetof(struct user32, u_debugreg[7]):
1075                 regno -= offsetof(struct user32, u_debugreg[0]);
1076                 return ptrace_set_debugreg(child, regno / 4, value);
1077
1078         default:
1079                 if (regno > sizeof(struct user32) || (regno & 3))
1080                         return -EIO;
1081
1082                 /*
1083                  * Other dummy fields in the virtual user structure
1084                  * are ignored
1085                  */
1086                 break;
1087         }
1088         return 0;
1089 }
1090
1091 #undef R32
1092 #undef SEG32
1093
1094 #define R32(l,q)                                                        \
1095         case offsetof(struct user32, regs.l):                           \
1096                 *val = regs->q; break
1097
1098 #define SEG32(rs)                                                       \
1099         case offsetof(struct user32, regs.rs):                          \
1100                 *val = get_segment_reg(child,                           \
1101                                        offsetof(struct user_regs_struct, rs)); \
1102                 break
1103
1104 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1105 {
1106         struct pt_regs *regs = task_pt_regs(child);
1107
1108         switch (regno) {
1109
1110         SEG32(ds);
1111         SEG32(es);
1112         SEG32(fs);
1113         SEG32(gs);
1114
1115         R32(cs, cs);
1116         R32(ss, ss);
1117         R32(ebx, bx);
1118         R32(ecx, cx);
1119         R32(edx, dx);
1120         R32(edi, di);
1121         R32(esi, si);
1122         R32(ebp, bp);
1123         R32(eax, ax);
1124         R32(orig_eax, orig_ax);
1125         R32(eip, ip);
1126         R32(esp, sp);
1127
1128         case offsetof(struct user32, regs.eflags):
1129                 *val = get_flags(child);
1130                 break;
1131
1132         case offsetof(struct user32, u_debugreg[0]) ...
1133                 offsetof(struct user32, u_debugreg[7]):
1134                 regno -= offsetof(struct user32, u_debugreg[0]);
1135                 *val = ptrace_get_debugreg(child, regno / 4);
1136                 break;
1137
1138         default:
1139                 if (regno > sizeof(struct user32) || (regno & 3))
1140                         return -EIO;
1141
1142                 /*
1143                  * Other dummy fields in the virtual user structure
1144                  * are ignored
1145                  */
1146                 *val = 0;
1147                 break;
1148         }
1149         return 0;
1150 }
1151
1152 #undef R32
1153 #undef SEG32
1154
1155 static int genregs32_get(struct task_struct *target,
1156                          const struct user_regset *regset,
1157                          unsigned int pos, unsigned int count,
1158                          void *kbuf, void __user *ubuf)
1159 {
1160         if (kbuf) {
1161                 compat_ulong_t *k = kbuf;
1162                 while (count > 0) {
1163                         getreg32(target, pos, k++);
1164                         count -= sizeof(*k);
1165                         pos += sizeof(*k);
1166                 }
1167         } else {
1168                 compat_ulong_t __user *u = ubuf;
1169                 while (count > 0) {
1170                         compat_ulong_t word;
1171                         getreg32(target, pos, &word);
1172                         if (__put_user(word, u++))
1173                                 return -EFAULT;
1174                         count -= sizeof(*u);
1175                         pos += sizeof(*u);
1176                 }
1177         }
1178
1179         return 0;
1180 }
1181
1182 static int genregs32_set(struct task_struct *target,
1183                          const struct user_regset *regset,
1184                          unsigned int pos, unsigned int count,
1185                          const void *kbuf, const void __user *ubuf)
1186 {
1187         int ret = 0;
1188         if (kbuf) {
1189                 const compat_ulong_t *k = kbuf;
1190                 while (count > 0 && !ret) {
1191                         ret = putreg32(target, pos, *k++);
1192                         count -= sizeof(*k);
1193                         pos += sizeof(*k);
1194                 }
1195         } else {
1196                 const compat_ulong_t __user *u = ubuf;
1197                 while (count > 0 && !ret) {
1198                         compat_ulong_t word;
1199                         ret = __get_user(word, u++);
1200                         if (ret)
1201                                 break;
1202                         ret = putreg32(target, pos, word);
1203                         count -= sizeof(*u);
1204                         pos += sizeof(*u);
1205                 }
1206         }
1207         return ret;
1208 }
1209
1210 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1211                         compat_ulong_t caddr, compat_ulong_t cdata)
1212 {
1213         unsigned long addr = caddr;
1214         unsigned long data = cdata;
1215         void __user *datap = compat_ptr(data);
1216         int ret;
1217         __u32 val;
1218
1219         switch (request) {
1220         case PTRACE_PEEKUSR:
1221                 ret = getreg32(child, addr, &val);
1222                 if (ret == 0)
1223                         ret = put_user(val, (__u32 __user *)datap);
1224                 break;
1225
1226         case PTRACE_POKEUSR:
1227                 ret = putreg32(child, addr, data);
1228                 break;
1229
1230         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1231                 return copy_regset_to_user(child, &user_x86_32_view,
1232                                            REGSET_GENERAL,
1233                                            0, sizeof(struct user_regs_struct32),
1234                                            datap);
1235
1236         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1237                 return copy_regset_from_user(child, &user_x86_32_view,
1238                                              REGSET_GENERAL, 0,
1239                                              sizeof(struct user_regs_struct32),
1240                                              datap);
1241
1242         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1243                 return copy_regset_to_user(child, &user_x86_32_view,
1244                                            REGSET_FP, 0,
1245                                            sizeof(struct user_i387_ia32_struct),
1246                                            datap);
1247
1248         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1249                 return copy_regset_from_user(
1250                         child, &user_x86_32_view, REGSET_FP,
1251                         0, sizeof(struct user_i387_ia32_struct), datap);
1252
1253         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1254                 return copy_regset_to_user(child, &user_x86_32_view,
1255                                            REGSET_XFP, 0,
1256                                            sizeof(struct user32_fxsr_struct),
1257                                            datap);
1258
1259         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1260                 return copy_regset_from_user(child, &user_x86_32_view,
1261                                              REGSET_XFP, 0,
1262                                              sizeof(struct user32_fxsr_struct),
1263                                              datap);
1264
1265         case PTRACE_GET_THREAD_AREA:
1266         case PTRACE_SET_THREAD_AREA:
1267                 return arch_ptrace(child, request, addr, data);
1268
1269         default:
1270                 return compat_ptrace_request(child, request, addr, data);
1271         }
1272
1273         return ret;
1274 }
1275
1276 #endif  /* CONFIG_IA32_EMULATION */
1277
1278 #ifdef CONFIG_X86_64
1279
1280 static const struct user_regset x86_64_regsets[] = {
1281         [REGSET_GENERAL] = {
1282                 .core_note_type = NT_PRSTATUS,
1283                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1284                 .size = sizeof(long), .align = sizeof(long),
1285                 .get = genregs_get, .set = genregs_set
1286         },
1287         [REGSET_FP] = {
1288                 .core_note_type = NT_PRFPREG,
1289                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1290                 .size = sizeof(long), .align = sizeof(long),
1291                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1292         },
1293 };
1294
1295 static const struct user_regset_view user_x86_64_view = {
1296         .name = "x86_64", .e_machine = EM_X86_64,
1297         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1298 };
1299
1300 #else  /* CONFIG_X86_32 */
1301
1302 #define user_regs_struct32      user_regs_struct
1303 #define genregs32_get           genregs_get
1304 #define genregs32_set           genregs_set
1305
1306 #define user_i387_ia32_struct   user_i387_struct
1307 #define user32_fxsr_struct      user_fxsr_struct
1308
1309 #endif  /* CONFIG_X86_64 */
1310
1311 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1312 static const struct user_regset x86_32_regsets[] = {
1313         [REGSET_GENERAL] = {
1314                 .core_note_type = NT_PRSTATUS,
1315                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1316                 .size = sizeof(u32), .align = sizeof(u32),
1317                 .get = genregs32_get, .set = genregs32_set
1318         },
1319         [REGSET_FP] = {
1320                 .core_note_type = NT_PRFPREG,
1321                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1322                 .size = sizeof(u32), .align = sizeof(u32),
1323                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1324         },
1325         [REGSET_XFP] = {
1326                 .core_note_type = NT_PRXFPREG,
1327                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1328                 .size = sizeof(u32), .align = sizeof(u32),
1329                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1330         },
1331         [REGSET_TLS] = {
1332                 .core_note_type = NT_386_TLS,
1333                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1334                 .size = sizeof(struct user_desc),
1335                 .align = sizeof(struct user_desc),
1336                 .active = regset_tls_active,
1337                 .get = regset_tls_get, .set = regset_tls_set
1338         },
1339 };
1340
1341 static const struct user_regset_view user_x86_32_view = {
1342         .name = "i386", .e_machine = EM_386,
1343         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1344 };
1345 #endif
1346
1347 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1348 {
1349 #ifdef CONFIG_IA32_EMULATION
1350         if (test_tsk_thread_flag(task, TIF_IA32))
1351 #endif
1352 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1353                 return &user_x86_32_view;
1354 #endif
1355 #ifdef CONFIG_X86_64
1356         return &user_x86_64_view;
1357 #endif
1358 }
1359
1360 #ifdef CONFIG_X86_32
1361
1362 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1363 {
1364         struct siginfo info;
1365
1366         tsk->thread.trap_no = 1;
1367         tsk->thread.error_code = error_code;
1368
1369         memset(&info, 0, sizeof(info));
1370         info.si_signo = SIGTRAP;
1371         info.si_code = TRAP_BRKPT;
1372
1373         /* User-mode ip? */
1374         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1375
1376         /* Send us the fake SIGTRAP */
1377         force_sig_info(SIGTRAP, &info, tsk);
1378 }
1379
1380 /* notification of system call entry/exit
1381  * - triggered by current->work.syscall_trace
1382  */
1383 int do_syscall_trace(struct pt_regs *regs, int entryexit)
1384 {
1385         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
1386         /*
1387          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
1388          * interception
1389          */
1390         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
1391         int ret = 0;
1392
1393         /* do the secure computing check first */
1394         if (!entryexit)
1395                 secure_computing(regs->orig_ax);
1396
1397         if (unlikely(current->audit_context)) {
1398                 if (entryexit)
1399                         audit_syscall_exit(AUDITSC_RESULT(regs->ax),
1400                                                 regs->ax);
1401                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
1402                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
1403                  * not used, entry.S will call us only on syscall exit, not
1404                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
1405                  * calling send_sigtrap() on syscall entry.
1406                  *
1407                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
1408                  * is_singlestep is false, despite his name, so we will still do
1409                  * the correct thing.
1410                  */
1411                 else if (is_singlestep)
1412                         goto out;
1413         }
1414
1415         if (!(current->ptrace & PT_PTRACED))
1416                 goto out;
1417
1418         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
1419          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
1420          * here. We have to check this and return */
1421         if (is_sysemu && entryexit)
1422                 return 0;
1423
1424         /* Fake a debug trap */
1425         if (is_singlestep)
1426                 send_sigtrap(current, regs, 0);
1427
1428         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
1429                 goto out;
1430
1431         /* the 0x80 provides a way for the tracing parent to distinguish
1432            between a syscall stop and SIGTRAP delivery */
1433         /* Note that the debugger could change the result of test_thread_flag!*/
1434         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1435
1436         /*
1437          * this isn't the same as continuing with a signal, but it will do
1438          * for normal use.  strace only continues with a signal if the
1439          * stopping signal is not SIGTRAP.  -brl
1440          */
1441         if (current->exit_code) {
1442                 send_sig(current->exit_code, current, 1);
1443                 current->exit_code = 0;
1444         }
1445         ret = is_sysemu;
1446 out:
1447         if (unlikely(current->audit_context) && !entryexit)
1448                 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
1449                                     regs->bx, regs->cx, regs->dx, regs->si);
1450         if (ret == 0)
1451                 return 0;
1452
1453         regs->orig_ax = -1; /* force skip of syscall restarting */
1454         if (unlikely(current->audit_context))
1455                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1456         return 1;
1457 }
1458
1459 #else  /* CONFIG_X86_64 */
1460
1461 static void syscall_trace(struct pt_regs *regs)
1462 {
1463
1464 #if 0
1465         printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
1466                current->comm,
1467                regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0),
1468                current_thread_info()->flags, current->ptrace);
1469 #endif
1470
1471         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
1472                                 ? 0x80 : 0));
1473         /*
1474          * this isn't the same as continuing with a signal, but it will do
1475          * for normal use.  strace only continues with a signal if the
1476          * stopping signal is not SIGTRAP.  -brl
1477          */
1478         if (current->exit_code) {
1479                 send_sig(current->exit_code, current, 1);
1480                 current->exit_code = 0;
1481         }
1482 }
1483
1484 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
1485 {
1486         /* do the secure computing check first */
1487         secure_computing(regs->orig_ax);
1488
1489         if (test_thread_flag(TIF_SYSCALL_TRACE)
1490             && (current->ptrace & PT_PTRACED))
1491                 syscall_trace(regs);
1492
1493         if (unlikely(current->audit_context)) {
1494                 if (test_thread_flag(TIF_IA32)) {
1495                         audit_syscall_entry(AUDIT_ARCH_I386,
1496                                             regs->orig_ax,
1497                                             regs->bx, regs->cx,
1498                                             regs->dx, regs->si);
1499                 } else {
1500                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1501                                             regs->orig_ax,
1502                                             regs->di, regs->si,
1503                                             regs->dx, regs->r10);
1504                 }
1505         }
1506 }
1507
1508 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
1509 {
1510         if (unlikely(current->audit_context))
1511                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1512
1513         if ((test_thread_flag(TIF_SYSCALL_TRACE)
1514              || test_thread_flag(TIF_SINGLESTEP))
1515             && (current->ptrace & PT_PTRACED))
1516                 syscall_trace(regs);
1517 }
1518
1519 #endif  /* CONFIG_X86_32 */