arch/i386/kernel/ptrace.c

   1 /* ptrace.c */
   2 /* By Ross Biro 1/23/92 */
   3 /*
   4  * Pentium III FXSR, SSE support
   5  *      Gareth Hughes <gareth@valinux.com>, May 2000
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/sched.h>
  10 #include <linux/mm.h>
  11 #include <linux/smp.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/errno.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/user.h>
  16 #include <linux/security.h>
  17 #include <linux/audit.h>
  18 #include <linux/seccomp.h>
  19 #include <linux/signal.h>
  20
  21 #include <asm/uaccess.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/system.h>
  24 #include <asm/processor.h>
  25 #include <asm/i387.h>
  26 #include <asm/debugreg.h>
  27 #include <asm/ldt.h>
  28 #include <asm/desc.h>
  29
  30 /*
  31  * does not yet catch signals sent when the child dies.
  32  * in exit.c or in signal.c.
  33  */
  34
  35 /* determines which flags the user has access to. */
  36 /* 1 = access 0 = no access */
  37 #define FLAG_MASK 0x00044dd5
  38
  39 /* set's the trap flag. */
  40 #define TRAP_FLAG 0x100
  41
  42 /*
  43  * Offset of eflags on child stack..
  44  */
  45 #define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
  46
  47 static inline struct pt_regs *get_child_regs(struct task_struct *task)
  48 {
  49         void *stack_top = (void *)task->thread.esp0;
  50         return stack_top - sizeof(struct pt_regs);
  51 }
  52
  53 /*
  54  * this routine will get a word off of the processes privileged stack.
  55  * the offset is how far from the base addr as stored in the TSS.
  56  * this routine assumes that all the privileged stacks are in our
  57  * data space.
  58  */
  59 static inline int get_stack_long(struct task_struct *task, int offset)
  60 {
  61         unsigned char *stack;
  62
  63         stack = (unsigned char *)task->thread.esp0;
  64         stack += offset;
  65         return (*((int *)stack));
  66 }
  67
  68 /*
  69  * this routine will put a word on the processes privileged stack.
  70  * the offset is how far from the base addr as stored in the TSS.
  71  * this routine assumes that all the privileged stacks are in our
  72  * data space.
  73  */
  74 static inline int put_stack_long(struct task_struct *task, int offset,
  75         unsigned long data)
  76 {
  77         unsigned char * stack;
  78
  79         stack = (unsigned char *) task->thread.esp0;
  80         stack += offset;
  81         *(unsigned long *) stack = data;
  82         return 0;
  83 }
  84
  85 static int putreg(struct task_struct *child,
  86         unsigned long regno, unsigned long value)
  87 {
  88         switch (regno >> 2) {
  89                 case FS:
  90                         if (value && (value & 3) != 3)
  91                                 return -EIO;
  92                         child->thread.fs = value;
  93                         return 0;
  94                 case GS:
  95                         if (value && (value & 3) != 3)
  96                                 return -EIO;
  97                         child->thread.gs = value;
  98                         return 0;
  99                 case DS:
 100                 case ES:
 101                         if (value && (value & 3) != 3)
 102                                 return -EIO;
 103                         value &= 0xffff;
 104                         break;
 105                 case SS:
 106                 case CS:
 107                         if ((value & 3) != 3)
 108                                 return -EIO;
 109                         value &= 0xffff;
 110                         break;
 111                 case EFL:
 112                         value &= FLAG_MASK;
 113                         value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
 114                         break;
 115         }
 116         if (regno > GS*4)
 117                 regno -= 2*4;
 118         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 119         return 0;
 120 }
 121
 122 static unsigned long getreg(struct task_struct *child,
 123         unsigned long regno)
 124 {
 125         unsigned long retval = ~0UL;
 126
 127         switch (regno >> 2) {
 128                 case FS:
 129                         retval = child->thread.fs;
 130                         break;
 131                 case GS:
 132                         retval = child->thread.gs;
 133                         break;
 134                 case DS:
 135                 case ES:
 136                 case SS:
 137                 case CS:
 138                         retval = 0xffff;
 139                         /* fall through */
 140                 default:
 141                         if (regno > GS*4)
 142                                 regno -= 2*4;
 143                         regno = regno - sizeof(struct pt_regs);
 144                         retval &= get_stack_long(child, regno);
 145         }
 146         return retval;
 147 }
 148
 149 #define LDT_SEGMENT 4
 150
 151 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
 152 {
 153         unsigned long addr, seg;
 154
 155         addr = regs->eip;
 156         seg = regs->xcs & 0xffff;
 157         if (regs->eflags & VM_MASK) {
 158                 addr = (addr & 0xffff) + (seg << 4);
 159                 return addr;
 160         }
 161
 162         /*
 163          * We'll assume that the code segments in the GDT
 164          * are all zero-based. That is largely true: the
 165          * TLS segments are used for data, and the PNPBIOS
 166          * and APM bios ones we just ignore here.
 167          */
 168         if (seg & LDT_SEGMENT) {
 169                 u32 *desc;
 170                 unsigned long base;
 171
 172                 down(&child->mm->context.sem);
 173                 desc = child->mm->context.ldt + (seg & ~7);
 174                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 175
 176                 /* 16-bit code segment? */
 177                 if (!((desc[1] >> 22) & 1))
 178                         addr &= 0xffff;
 179                 addr += base;
 180                 up(&child->mm->context.sem);
 181         }
 182         return addr;
 183 }
 184
 185 static inline int is_at_popf(struct task_struct *child, struct pt_regs *regs)
 186 {
 187         int i, copied;
 188         unsigned char opcode[16];
 189         unsigned long addr = convert_eip_to_linear(child, regs);
 190
 191         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 192         for (i = 0; i < copied; i++) {
 193                 switch (opcode[i]) {
 194                 /* popf */
 195                 case 0x9d:
 196                         return 1;
 197                 /* opcode and address size prefixes */
 198                 case 0x66: case 0x67:
 199                         continue;
 200                 /* irrelevant prefixes (segment overrides and repeats) */
 201                 case 0x26: case 0x2e:
 202                 case 0x36: case 0x3e:
 203                 case 0x64: case 0x65:
 204                 case 0xf0: case 0xf2: case 0xf3:
 205                         continue;
 206
 207                 /*
 208                  * pushf: NOTE! We should probably not let
 209                  * the user see the TF bit being set. But
 210                  * it's more pain than it's worth to avoid
 211                  * it, and a debugger could emulate this
 212                  * all in user space if it _really_ cares.
 213                  */
 214                 case 0x9c:
 215                 default:
 216                         return 0;
 217                 }
 218         }
 219         return 0;
 220 }
 221
 222 static void set_singlestep(struct task_struct *child)
 223 {
 224         struct pt_regs *regs = get_child_regs(child);
 225
 226         /*
 227          * Always set TIF_SINGLESTEP - this guarantees that
 228          * we single-step system calls etc..  This will also
 229          * cause us to set TF when returning to user mode.
 230          */
 231         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 232
 233         /*
 234          * If TF was already set, don't do anything else
 235          */
 236         if (regs->eflags & TRAP_FLAG)
 237                 return;
 238
 239         /* Set TF on the kernel stack.. */
 240         regs->eflags |= TRAP_FLAG;
 241
 242         /*
 243          * ..but if TF is changed by the instruction we will trace,
 244          * don't mark it as being "us" that set it, so that we
 245          * won't clear it by hand later.
 246          */
 247         if (is_at_popf(child, regs))
 248                 return;
 249
 250         child->ptrace |= PT_DTRACE;
 251 }
 252
 253 static void clear_singlestep(struct task_struct *child)
 254 {
 255         /* Always clear TIF_SINGLESTEP... */
 256         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 257
 258         /* But touch TF only if it was set by us.. */
 259         if (child->ptrace & PT_DTRACE) {
 260                 struct pt_regs *regs = get_child_regs(child);
 261                 regs->eflags &= ~TRAP_FLAG;
 262                 child->ptrace &= ~PT_DTRACE;
 263         }
 264 }
 265
 266 /*
 267  * Called by kernel/ptrace.c when detaching..
 268  *
 269  * Make sure the single step bit is not set.
 270  */
 271 void ptrace_disable(struct task_struct *child)
 272 {
 273         clear_singlestep(child);
 274         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 275         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 276 }
 277
 278 /*
 279  * Perform get_thread_area on behalf of the traced child.
 280  */
 281 static int
 282 ptrace_get_thread_area(struct task_struct *child,
 283                        int idx, struct user_desc __user *user_desc)
 284 {
 285         struct user_desc info;
 286         struct desc_struct *desc;
 287
 288 /*
 289  * Get the current Thread-Local Storage area:
 290  */
 291
 292 #define GET_BASE(desc) ( \
 293         (((desc)->a >> 16) & 0x0000ffff) | \
 294         (((desc)->b << 16) & 0x00ff0000) | \
 295         ( (desc)->b        & 0xff000000)   )
 296
 297 #define GET_LIMIT(desc) ( \
 298         ((desc)->a & 0x0ffff) | \
 299          ((desc)->b & 0xf0000) )
 300
 301 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 302 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 303 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 304 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 305 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 306 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 307
 308         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 309                 return -EINVAL;
 310
 311         desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 312
 313         info.entry_number = idx;
 314         info.base_addr = GET_BASE(desc);
 315         info.limit = GET_LIMIT(desc);
 316         info.seg_32bit = GET_32BIT(desc);
 317         info.contents = GET_CONTENTS(desc);
 318         info.read_exec_only = !GET_WRITABLE(desc);
 319         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 320         info.seg_not_present = !GET_PRESENT(desc);
 321         info.useable = GET_USEABLE(desc);
 322
 323         if (copy_to_user(user_desc, &info, sizeof(info)))
 324                 return -EFAULT;
 325
 326         return 0;
 327 }
 328
 329 /*
 330  * Perform set_thread_area on behalf of the traced child.
 331  */
 332 static int
 333 ptrace_set_thread_area(struct task_struct *child,
 334                        int idx, struct user_desc __user *user_desc)
 335 {
 336         struct user_desc info;
 337         struct desc_struct *desc;
 338
 339         if (copy_from_user(&info, user_desc, sizeof(info)))
 340                 return -EFAULT;
 341
 342         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 343                 return -EINVAL;
 344
 345         desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 346         if (LDT_empty(&info)) {
 347                 desc->a = 0;
 348                 desc->b = 0;
 349         } else {
 350                 desc->a = LDT_entry_a(&info);
 351                 desc->b = LDT_entry_b(&info);
 352         }
 353
 354         return 0;
 355 }
 356
 357 asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 358 {
 359         struct task_struct *child;
 360         struct user * dummy = NULL;
 361         int i, ret;
 362         unsigned long __user *datap = (unsigned long __user *)data;
 363
 364         lock_kernel();
 365         ret = -EPERM;
 366         if (request == PTRACE_TRACEME) {
 367                 /* are we already being traced? */
 368                 if (current->ptrace & PT_PTRACED)
 369                         goto out;
 370                 ret = security_ptrace(current->parent, current);
 371                 if (ret)
 372                         goto out;
 373                 /* set the ptrace bit in the process flags. */
 374                 current->ptrace |= PT_PTRACED;
 375                 ret = 0;
 376                 goto out;
 377         }
 378         ret = -ESRCH;
 379         read_lock(&tasklist_lock);
 380         child = find_task_by_pid(pid);
 381         if (child)
 382                 get_task_struct(child);
 383         read_unlock(&tasklist_lock);
 384         if (!child)
 385                 goto out;
 386
 387         ret = -EPERM;
 388         if (pid == 1)           /* you may not mess with init */
 389                 goto out_tsk;
 390
 391         if (request == PTRACE_ATTACH) {
 392                 ret = ptrace_attach(child);
 393                 goto out_tsk;
 394         }
 395
 396         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 397         if (ret < 0)
 398                 goto out_tsk;
 399
 400         switch (request) {
 401         /* when I and D space are separate, these will need to be fixed. */
 402         case PTRACE_PEEKTEXT: /* read word at location addr. */
 403         case PTRACE_PEEKDATA: {
 404                 unsigned long tmp;
 405                 int copied;
 406
 407                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 408                 ret = -EIO;
 409                 if (copied != sizeof(tmp))
 410                         break;
 411                 ret = put_user(tmp, datap);
 412                 break;
 413         }
 414
 415         /* read the word at location addr in the USER area. */
 416         case PTRACE_PEEKUSR: {
 417                 unsigned long tmp;
 418
 419                 ret = -EIO;
 420                 if ((addr & 3) || addr < 0 ||
 421                     addr > sizeof(struct user) - 3)
 422                         break;
 423
 424                 tmp = 0;  /* Default return condition */
 425                 if(addr < FRAME_SIZE*sizeof(long))
 426                         tmp = getreg(child, addr);
 427                 if(addr >= (long) &dummy->u_debugreg[0] &&
 428                    addr <= (long) &dummy->u_debugreg[7]){
 429                         addr -= (long) &dummy->u_debugreg[0];
 430                         addr = addr >> 2;
 431                         tmp = child->thread.debugreg[addr];
 432                 }
 433                 ret = put_user(tmp, datap);
 434                 break;
 435         }
 436
 437         /* when I and D space are separate, this will have to be fixed. */
 438         case PTRACE_POKETEXT: /* write the word at location addr. */
 439         case PTRACE_POKEDATA:
 440                 ret = 0;
 441                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 442                         break;
 443                 ret = -EIO;
 444                 break;
 445
 446         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 447                 ret = -EIO;
 448                 if ((addr & 3) || addr < 0 ||
 449                     addr > sizeof(struct user) - 3)
 450                         break;
 451
 452                 if (addr < FRAME_SIZE*sizeof(long)) {
 453                         ret = putreg(child, addr, data);
 454                         break;
 455                 }
 456                 /* We need to be very careful here.  We implicitly
 457                    want to modify a portion of the task_struct, and we
 458                    have to be selective about what portions we allow someone
 459                    to modify. */
 460
 461                   ret = -EIO;
 462                   if(addr >= (long) &dummy->u_debugreg[0] &&
 463                      addr <= (long) &dummy->u_debugreg[7]){
 464
 465                           if(addr == (long) &dummy->u_debugreg[4]) break;
 466                           if(addr == (long) &dummy->u_debugreg[5]) break;
 467                           if(addr < (long) &dummy->u_debugreg[4] &&
 468                              ((unsigned long) data) >= TASK_SIZE-3) break;
 469
 470                           /* Sanity-check data. Take one half-byte at once with
 471                            * check = (val >> (16 + 4*i)) & 0xf. It contains the
 472                            * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 473                            * 2 and 3 are LENi. Given a list of invalid values,
 474                            * we do mask |= 1 << invalid_value, so that
 475                            * (mask >> check) & 1 is a correct test for invalid
 476                            * values.
 477                            *
 478                            * R/Wi contains the type of the breakpoint /
 479                            * watchpoint, LENi contains the length of the watched
 480                            * data in the watchpoint case.
 481                            *
 482                            * The invalid values are:
 483                            * - LENi == 0x10 (undefined), so mask |= 0x0f00.
 484                            * - R/Wi == 0x10 (break on I/O reads or writes), so
 485                            *   mask |= 0x4444.
 486                            * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 487                            *   0x1110.
 488                            *
 489                            * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 490                            *
 491                            * See the Intel Manual "System Programming Guide",
 492                            * 15.2.4
 493                            *
 494                            * Note that LENi == 0x10 is defined on x86_64 in long
 495                            * mode (i.e. even for 32-bit userspace software, but
 496                            * 64-bit kernel), so the x86_64 mask value is 0x5454.
 497                            * See the AMD manual no. 24593 (AMD64 System
 498                            * Programming)*/
 499
 500                           if(addr == (long) &dummy->u_debugreg[7]) {
 501                                   data &= ~DR_CONTROL_RESERVED;
 502                                   for(i=0; i<4; i++)
 503                                           if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 504                                                   goto out_tsk;
 505                           }
 506
 507                           addr -= (long) &dummy->u_debugreg;
 508                           addr = addr >> 2;
 509                           child->thread.debugreg[addr] = data;
 510                           ret = 0;
 511                   }
 512                   break;
 513
 514         case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
 515         case PTRACE_SYSCALL:    /* continue and stop at next (return from) syscall */
 516         case PTRACE_CONT:       /* restart after signal. */
 517                 ret = -EIO;
 518                 if (!valid_signal(data))
 519                         break;
 520                 if (request == PTRACE_SYSEMU) {
 521                         set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 522                         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 523                 } else if (request == PTRACE_SYSCALL) {
 524                         set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 525                         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 526                 } else {
 527                         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 528                         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 529                 }
 530                 child->exit_code = data;
 531                 /* make sure the single step bit is not set. */
 532                 clear_singlestep(child);
 533                 wake_up_process(child);
 534                 ret = 0;
 535                 break;
 536
 537 /*
 538  * make the child exit.  Best I can do is send it a sigkill.
 539  * perhaps it should be put in the status that it wants to
 540  * exit.
 541  */
 542         case PTRACE_KILL:
 543                 ret = 0;
 544                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 545                         break;
 546                 child->exit_code = SIGKILL;
 547                 /* make sure the single step bit is not set. */
 548                 clear_singlestep(child);
 549                 wake_up_process(child);
 550                 break;
 551
 552         case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
 553         case PTRACE_SINGLESTEP: /* set the trap flag. */
 554                 ret = -EIO;
 555                 if (!valid_signal(data))
 556                         break;
 557
 558                 if (request == PTRACE_SYSEMU_SINGLESTEP)
 559                         set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 560                 else
 561                         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 562
 563                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 564                 set_singlestep(child);
 565                 child->exit_code = data;
 566                 /* give it a chance to run. */
 567                 wake_up_process(child);
 568                 ret = 0;
 569                 break;
 570
 571         case PTRACE_DETACH:
 572                 /* detach a process that was attached. */
 573                 ret = ptrace_detach(child, data);
 574                 break;
 575
 576         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 577                 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
 578                         ret = -EIO;
 579                         break;
 580                 }
 581                 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
 582                         __put_user(getreg(child, i), datap);
 583                         datap++;
 584                 }
 585                 ret = 0;
 586                 break;
 587         }
 588
 589         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 590                 unsigned long tmp;
 591                 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
 592                         ret = -EIO;
 593                         break;
 594                 }
 595                 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
 596                         __get_user(tmp, datap);
 597                         putreg(child, i, tmp);
 598                         datap++;
 599                 }
 600                 ret = 0;
 601                 break;
 602         }
 603
 604         case PTRACE_GETFPREGS: { /* Get the child FPU state. */
 605                 if (!access_ok(VERIFY_WRITE, datap,
 606                                sizeof(struct user_i387_struct))) {
 607                         ret = -EIO;
 608                         break;
 609                 }
 610                 ret = 0;
 611                 if (!tsk_used_math(child))
 612                         init_fpu(child);
 613                 get_fpregs((struct user_i387_struct __user *)data, child);
 614                 break;
 615         }
 616
 617         case PTRACE_SETFPREGS: { /* Set the child FPU state. */
 618                 if (!access_ok(VERIFY_READ, datap,
 619                                sizeof(struct user_i387_struct))) {
 620                         ret = -EIO;
 621                         break;
 622                 }
 623                 set_stopped_child_used_math(child);
 624                 set_fpregs(child, (struct user_i387_struct __user *)data);
 625                 ret = 0;
 626                 break;
 627         }
 628
 629         case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
 630                 if (!access_ok(VERIFY_WRITE, datap,
 631                                sizeof(struct user_fxsr_struct))) {
 632                         ret = -EIO;
 633                         break;
 634                 }
 635                 if (!tsk_used_math(child))
 636                         init_fpu(child);
 637                 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
 638                 break;
 639         }
 640
 641         case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
 642                 if (!access_ok(VERIFY_READ, datap,
 643                                sizeof(struct user_fxsr_struct))) {
 644                         ret = -EIO;
 645                         break;
 646                 }
 647                 set_stopped_child_used_math(child);
 648                 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
 649                 break;
 650         }
 651
 652         case PTRACE_GET_THREAD_AREA:
 653                 ret = ptrace_get_thread_area(child, addr,
 654                                         (struct user_desc __user *) data);
 655                 break;
 656
 657         case PTRACE_SET_THREAD_AREA:
 658                 ret = ptrace_set_thread_area(child, addr,
 659                                         (struct user_desc __user *) data);
 660                 break;
 661
 662         default:
 663                 ret = ptrace_request(child, request, addr, data);
 664                 break;
 665         }
 666 out_tsk:
 667         put_task_struct(child);
 668 out:
 669         unlock_kernel();
 670         return ret;
 671 }
 672
 673 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
 674 {
 675         struct siginfo info;
 676
 677         tsk->thread.trap_no = 1;
 678         tsk->thread.error_code = error_code;
 679
 680         memset(&info, 0, sizeof(info));
 681         info.si_signo = SIGTRAP;
 682         info.si_code = TRAP_BRKPT;
 683
 684         /* User-mode eip? */
 685         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
 686
 687         /* Send us the fakey SIGTRAP */
 688         force_sig_info(SIGTRAP, &info, tsk);
 689 }
 690
 691 /* notification of system call entry/exit
 692  * - triggered by current->work.syscall_trace
 693  */
 694 __attribute__((regparm(3)))
 695 int do_syscall_trace(struct pt_regs *regs, int entryexit)
 696 {
 697         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
 698         /*
 699          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
 700          * interception
 701          */
 702         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
 703         int ret = 0;
 704
 705         /* do the secure computing check first */
 706         if (!entryexit)
 707                 secure_computing(regs->orig_eax);
 708
 709         if (unlikely(current->audit_context)) {
 710                 if (entryexit)
 711                         audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),
 712                                                 regs->eax);
 713                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
 714                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
 715                  * not used, entry.S will call us only on syscall exit, not
 716                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
 717                  * calling send_sigtrap() on syscall entry.
 718                  *
 719                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
 720                  * is_singlestep is false, despite his name, so we will still do
 721                  * the correct thing.
 722                  */
 723                 else if (is_singlestep)
 724                         goto out;
 725         }
 726
 727         if (!(current->ptrace & PT_PTRACED))
 728                 goto out;
 729
 730         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
 731          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
 732          * here. We have to check this and return */
 733         if (is_sysemu && entryexit)
 734                 return 0;
 735
 736         /* Fake a debug trap */
 737         if (is_singlestep)
 738                 send_sigtrap(current, regs, 0);
 739
 740         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
 741                 goto out;
 742
 743         /* the 0x80 provides a way for the tracing parent to distinguish
 744            between a syscall stop and SIGTRAP delivery */
 745         /* Note that the debugger could change the result of test_thread_flag!*/
 746         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
 747
 748         /*
 749          * this isn't the same as continuing with a signal, but it will do
 750          * for normal use.  strace only continues with a signal if the
 751          * stopping signal is not SIGTRAP.  -brl
 752          */
 753         if (current->exit_code) {
 754                 send_sig(current->exit_code, current, 1);
 755                 current->exit_code = 0;
 756         }
 757         ret = is_sysemu;
 758 out:
 759         if (unlikely(current->audit_context) && !entryexit)
 760                 audit_syscall_entry(current, AUDIT_ARCH_I386, regs->orig_eax,
 761                                     regs->ebx, regs->ecx, regs->edx, regs->esi);
 762         if (ret == 0)
 763                 return 0;
 764
 765         regs->orig_eax = -1; /* force skip of syscall restarting */
 766         if (unlikely(current->audit_context))
 767                 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),
 768                                 regs->eax);
 769         return 1;
 770 }