arch/x86_64/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  * x86-64 port 2000-2002 Andi Kleen
   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/smp_lock.h>
  15 #include <linux/errno.h>
  16 #include <linux/ptrace.h>
  17 #include <linux/user.h>
  18 #include <linux/security.h>
  19 #include <linux/audit.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 #include <asm/proto.h>
  30 #include <asm/ia32.h>
  31
  32 /*
  33  * does not yet catch signals sent when the child dies.
  34  * in exit.c or in signal.c.
  35  */
  36
  37 /* determines which flags the user has access to. */
  38 /* 1 = access 0 = no access */
  39 #define FLAG_MASK 0x44dd5UL
  40
  41 /* set's the trap flag. */
  42 #define TRAP_FLAG 0x100UL
  43
  44 /*
  45  * eflags and offset of eflags on child stack..
  46  */
  47 #define EFLAGS offsetof(struct pt_regs, eflags)
  48 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  49
  50 /*
  51  * this routine will get a word off of the processes privileged stack.
  52  * the offset is how far from the base addr as stored in the TSS.
  53  * this routine assumes that all the privileged stacks are in our
  54  * data space.
  55  */
  56 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  57 {
  58         unsigned char *stack;
  59
  60         stack = (unsigned char *)task->thread.rsp0;
  61         stack += offset;
  62         return (*((unsigned long *)stack));
  63 }
  64
  65 /*
  66  * this routine will put a word on the processes privileged stack.
  67  * the offset is how far from the base addr as stored in the TSS.
  68  * this routine assumes that all the privileged stacks are in our
  69  * data space.
  70  */
  71 static inline long put_stack_long(struct task_struct *task, int offset,
  72         unsigned long data)
  73 {
  74         unsigned char * stack;
  75
  76         stack = (unsigned char *) task->thread.rsp0;
  77         stack += offset;
  78         *(unsigned long *) stack = data;
  79         return 0;
  80 }
  81
  82 /*
  83  * Called by kernel/ptrace.c when detaching..
  84  *
  85  * Make sure the single step bit is not set.
  86  */
  87 void ptrace_disable(struct task_struct *child)
  88 {
  89         long tmp;
  90
  91         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
  92         tmp = get_stack_long(child, EFL_OFFSET) & ~TRAP_FLAG;
  93         put_stack_long(child, EFL_OFFSET, tmp);
  94 }
  95
  96 static int putreg(struct task_struct *child,
  97         unsigned long regno, unsigned long value)
  98 {
  99         unsigned long tmp;
 100
 101         /* Some code in the 64bit emulation may not be 64bit clean.
 102            Don't take any chances. */
 103         if (test_tsk_thread_flag(child, TIF_IA32))
 104                 value &= 0xffffffff;
 105         switch (regno) {
 106                 case offsetof(struct user_regs_struct,fs):
 107                         if (value && (value & 3) != 3)
 108                                 return -EIO;
 109                         child->thread.fsindex = value & 0xffff;
 110                         return 0;
 111                 case offsetof(struct user_regs_struct,gs):
 112                         if (value && (value & 3) != 3)
 113                                 return -EIO;
 114                         child->thread.gsindex = value & 0xffff;
 115                         return 0;
 116                 case offsetof(struct user_regs_struct,ds):
 117                         if (value && (value & 3) != 3)
 118                                 return -EIO;
 119                         child->thread.ds = value & 0xffff;
 120                         return 0;
 121                 case offsetof(struct user_regs_struct,es):
 122                         if (value && (value & 3) != 3)
 123                                 return -EIO;
 124                         child->thread.es = value & 0xffff;
 125                         return 0;
 126                 case offsetof(struct user_regs_struct,ss):
 127                         if ((value & 3) != 3)
 128                                 return -EIO;
 129                         value &= 0xffff;
 130                         return 0;
 131                 case offsetof(struct user_regs_struct,fs_base):
 132                         if (!((value >> 48) == 0 || (value >> 48) == 0xffff))
 133                                 return -EIO;
 134                         child->thread.fs = value;
 135                         return 0;
 136                 case offsetof(struct user_regs_struct,gs_base):
 137                         if (!((value >> 48) == 0 || (value >> 48) == 0xffff))
 138                                 return -EIO;
 139                         child->thread.gs = value;
 140                         return 0;
 141                 case offsetof(struct user_regs_struct, eflags):
 142                         value &= FLAG_MASK;
 143                         tmp = get_stack_long(child, EFL_OFFSET);
 144                         tmp &= ~FLAG_MASK;
 145                         value |= tmp;
 146                         break;
 147                 case offsetof(struct user_regs_struct,cs):
 148                         if ((value & 3) != 3)
 149                                 return -EIO;
 150                         value &= 0xffff;
 151                         break;
 152         }
 153         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 154         return 0;
 155 }
 156
 157 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 158 {
 159         unsigned long val;
 160         switch (regno) {
 161                 case offsetof(struct user_regs_struct, fs):
 162                         return child->thread.fsindex;
 163                 case offsetof(struct user_regs_struct, gs):
 164                         return child->thread.gsindex;
 165                 case offsetof(struct user_regs_struct, ds):
 166                         return child->thread.ds;
 167                 case offsetof(struct user_regs_struct, es):
 168                         return child->thread.es;
 169                 case offsetof(struct user_regs_struct, fs_base):
 170                         return child->thread.fs;
 171                 case offsetof(struct user_regs_struct, gs_base):
 172                         return child->thread.gs;
 173                 default:
 174                         regno = regno - sizeof(struct pt_regs);
 175                         val = get_stack_long(child, regno);
 176                         if (test_tsk_thread_flag(child, TIF_IA32))
 177                                 val &= 0xffffffff;
 178                         return val;
 179         }
 180
 181 }
 182
 183 asmlinkage long sys_ptrace(long request, long pid, unsigned long addr, long data)
 184 {
 185         struct task_struct *child;
 186         long i, ret;
 187         unsigned ui;
 188
 189         /* This lock_kernel fixes a subtle race with suid exec */
 190         lock_kernel();
 191         ret = -EPERM;
 192         if (request == PTRACE_TRACEME) {
 193                 /* are we already being traced? */
 194                 if (current->ptrace & PT_PTRACED)
 195                         goto out;
 196                 ret = security_ptrace(current->parent, current);
 197                 if (ret)
 198                         goto out;
 199                 /* set the ptrace bit in the process flags. */
 200                 current->ptrace |= PT_PTRACED;
 201                 ret = 0;
 202                 goto out;
 203         }
 204         ret = -ESRCH;
 205         read_lock(&tasklist_lock);
 206         child = find_task_by_pid(pid);
 207         if (child)
 208                 get_task_struct(child);
 209         read_unlock(&tasklist_lock);
 210         if (!child)
 211                 goto out;
 212         if (!vx_check(vx_task_xid(child), VX_WATCH|VX_IDENT))
 213                 goto out_tsk;
 214
 215         ret = -EPERM;
 216         if (pid == 1)           /* you may not mess with init */
 217                 goto out_tsk;
 218
 219         if (request == PTRACE_ATTACH) {
 220                 ret = ptrace_attach(child);
 221                 goto out_tsk;
 222         }
 223         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 224         if (ret < 0)
 225                 goto out_tsk;
 226
 227         switch (request) {
 228         /* when I and D space are separate, these will need to be fixed. */
 229         case PTRACE_PEEKTEXT: /* read word at location addr. */
 230         case PTRACE_PEEKDATA: {
 231                 unsigned long tmp;
 232                 int copied;
 233
 234                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 235                 ret = -EIO;
 236                 if (copied != sizeof(tmp))
 237                         break;
 238                 ret = put_user(tmp,(unsigned long __user *) data);
 239                 break;
 240         }
 241
 242         /* read the word at location addr in the USER area. */
 243         case PTRACE_PEEKUSR: {
 244                 unsigned long tmp;
 245
 246                 ret = -EIO;
 247                 if ((addr & 7) ||
 248                     addr > sizeof(struct user) - 7)
 249                         break;
 250
 251                 switch (addr) {
 252                 case 0 ... sizeof(struct user_regs_struct):
 253                         tmp = getreg(child, addr);
 254                         break;
 255                 case offsetof(struct user, u_debugreg[0]):
 256                         tmp = child->thread.debugreg0;
 257                         break;
 258                 case offsetof(struct user, u_debugreg[1]):
 259                         tmp = child->thread.debugreg1;
 260                         break;
 261                 case offsetof(struct user, u_debugreg[2]):
 262                         tmp = child->thread.debugreg2;
 263                         break;
 264                 case offsetof(struct user, u_debugreg[3]):
 265                         tmp = child->thread.debugreg3;
 266                         break;
 267                 case offsetof(struct user, u_debugreg[6]):
 268                         tmp = child->thread.debugreg6;
 269                         break;
 270                 case offsetof(struct user, u_debugreg[7]):
 271                         tmp = child->thread.debugreg7;
 272                         break;
 273                 default:
 274                         tmp = 0;
 275                         break;
 276                 }
 277                 ret = put_user(tmp,(unsigned long __user *) data);
 278                 break;
 279         }
 280
 281         /* when I and D space are separate, this will have to be fixed. */
 282         case PTRACE_POKETEXT: /* write the word at location addr. */
 283         case PTRACE_POKEDATA:
 284                 ret = 0;
 285                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 286                         break;
 287                 ret = -EIO;
 288                 break;
 289
 290         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 291                 ret = -EIO;
 292                 if ((addr & 7) ||
 293                     addr > sizeof(struct user) - 7)
 294                         break;
 295
 296                 switch (addr) {
 297                 case 0 ... sizeof(struct user_regs_struct):
 298                         ret = putreg(child, addr, data);
 299                         break;
 300                 /* Disallows to set a breakpoint into the vsyscall */
 301                 case offsetof(struct user, u_debugreg[0]):
 302                         if (data >= TASK_SIZE-7) break;
 303                         child->thread.debugreg0 = data;
 304                         ret = 0;
 305                         break;
 306                 case offsetof(struct user, u_debugreg[1]):
 307                         if (data >= TASK_SIZE-7) break;
 308                         child->thread.debugreg1 = data;
 309                         ret = 0;
 310                         break;
 311                 case offsetof(struct user, u_debugreg[2]):
 312                         if (data >= TASK_SIZE-7) break;
 313                         child->thread.debugreg2 = data;
 314                         ret = 0;
 315                         break;
 316                 case offsetof(struct user, u_debugreg[3]):
 317                         if (data >= TASK_SIZE-7) break;
 318                         child->thread.debugreg3 = data;
 319                         ret = 0;
 320                         break;
 321                 case offsetof(struct user, u_debugreg[6]):
 322                                   if (data >> 32)
 323                                 break;
 324                         child->thread.debugreg6 = data;
 325                         ret = 0;
 326                         break;
 327                 case offsetof(struct user, u_debugreg[7]):
 328                         /* See arch/i386/kernel/ptrace.c for an explanation of
 329                          * this awkward check.*/
 330                                   data &= ~DR_CONTROL_RESERVED;
 331                                   for(i=0; i<4; i++)
 332                                           if ((0x5454 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 333                                         break;
 334                         if (i == 4) {
 335                                 child->thread.debugreg7 = data;
 336                           ret = 0;
 337                   }
 338                   break;
 339                 }
 340                 break;
 341         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 342         case PTRACE_CONT: { /* restart after signal. */
 343                 long tmp;
 344
 345                 ret = -EIO;
 346                 if ((unsigned long) data > _NSIG)
 347                         break;
 348                 if (request == PTRACE_SYSCALL)
 349                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 350                 else
 351                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 352                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 353                 child->exit_code = data;
 354         /* make sure the single step bit is not set. */
 355                 tmp = get_stack_long(child, EFL_OFFSET);
 356                 tmp &= ~TRAP_FLAG;
 357                 put_stack_long(child, EFL_OFFSET,tmp);
 358                 wake_up_process(child);
 359                 ret = 0;
 360                 break;
 361         }
 362
 363 #ifdef CONFIG_IA32_EMULATION
 364                 /* This makes only sense with 32bit programs. Allow a
 365                    64bit debugger to fully examine them too. Better
 366                    don't use it against 64bit processes, use
 367                    PTRACE_ARCH_PRCTL instead. */
 368         case PTRACE_SET_THREAD_AREA: {
 369                 struct user_desc __user *p;
 370                 int old;
 371                 p = (struct user_desc __user *)data;
 372                 get_user(old,  &p->entry_number);
 373                 put_user(addr, &p->entry_number);
 374                 ret = do_set_thread_area(&child->thread, p);
 375                 put_user(old,  &p->entry_number);
 376                 break;
 377         case PTRACE_GET_THREAD_AREA:
 378                 p = (struct user_desc __user *)data;
 379                 get_user(old,  &p->entry_number);
 380                 put_user(addr, &p->entry_number);
 381                 ret = do_get_thread_area(&child->thread, p);
 382                 put_user(old,  &p->entry_number);
 383                 break;
 384         }
 385 #endif
 386                 /* normal 64bit interface to access TLS data.
 387                    Works just like arch_prctl, except that the arguments
 388                    are reversed. */
 389         case PTRACE_ARCH_PRCTL:
 390                 ret = do_arch_prctl(child, data, addr);
 391                 break;
 392
 393 /*
 394  * make the child exit.  Best I can do is send it a sigkill.
 395  * perhaps it should be put in the status that it wants to
 396  * exit.
 397  */
 398         case PTRACE_KILL: {
 399                 long tmp;
 400
 401                 ret = 0;
 402                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 403                         break;
 404                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 405                 child->exit_code = SIGKILL;
 406                 /* make sure the single step bit is not set. */
 407                 tmp = get_stack_long(child, EFL_OFFSET) & ~TRAP_FLAG;
 408                 put_stack_long(child, EFL_OFFSET, tmp);
 409                 wake_up_process(child);
 410                 break;
 411         }
 412
 413         case PTRACE_SINGLESTEP: {  /* set the trap flag. */
 414                 long tmp;
 415
 416                 ret = -EIO;
 417                 if ((unsigned long) data > _NSIG)
 418                         break;
 419                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 420                 if ((child->ptrace & PT_DTRACE) == 0) {
 421                         /* Spurious delayed TF traps may occur */
 422                         child->ptrace |= PT_DTRACE;
 423                 }
 424                 tmp = get_stack_long(child, EFL_OFFSET) | TRAP_FLAG;
 425                 put_stack_long(child, EFL_OFFSET, tmp);
 426                 set_tsk_thread_flag(child, TIF_SINGLESTEP);
 427                 child->exit_code = data;
 428                 /* give it a chance to run. */
 429                 wake_up_process(child);
 430                 ret = 0;
 431                 break;
 432         }
 433
 434         case PTRACE_DETACH:
 435                 /* detach a process that was attached. */
 436                 ret = ptrace_detach(child, data);
 437                 break;
 438
 439         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 440                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 441                                sizeof(struct user_regs_struct))) {
 442                         ret = -EIO;
 443                         break;
 444                 }
 445                 ret = 0;
 446                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 447                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 448                         data += sizeof(long);
 449                 }
 450                 break;
 451         }
 452
 453         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 454                 unsigned long tmp;
 455                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 456                                sizeof(struct user_regs_struct))) {
 457                         ret = -EIO;
 458                         break;
 459                 }
 460                 ret = 0;
 461                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 462                         ret |= __get_user(tmp, (unsigned long __user *) data);
 463                         putreg(child, ui, tmp);
 464                         data += sizeof(long);
 465                 }
 466                 break;
 467         }
 468
 469         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 470                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 471                                sizeof(struct user_i387_struct))) {
 472                         ret = -EIO;
 473                         break;
 474                 }
 475                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 476                 break;
 477         }
 478
 479         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 480                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 481                                sizeof(struct user_i387_struct))) {
 482                         ret = -EIO;
 483                         break;
 484                 }
 485                 set_stopped_child_used_math(child);
 486                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 487                 break;
 488         }
 489
 490         default:
 491                 ret = ptrace_request(child, request, addr, data);
 492                 break;
 493         }
 494 out_tsk:
 495         put_task_struct(child);
 496 out:
 497         unlock_kernel();
 498         return ret;
 499 }
 500
 501 static void syscall_trace(struct pt_regs *regs)
 502 {
 503
 504 #if 0
 505         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 506                current->comm,
 507                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 508                current_thread_info()->flags, current->ptrace);
 509 #endif
 510
 511         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 512                                 ? 0x80 : 0));
 513         /*
 514          * this isn't the same as continuing with a signal, but it will do
 515          * for normal use.  strace only continues with a signal if the
 516          * stopping signal is not SIGTRAP.  -brl
 517          */
 518         if (current->exit_code) {
 519                 send_sig(current->exit_code, current, 1);
 520                 current->exit_code = 0;
 521         }
 522 }
 523
 524 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 525 {
 526         if (unlikely(current->audit_context))
 527                 audit_syscall_entry(current, regs->orig_rax,
 528                                     regs->rdi, regs->rsi,
 529                                     regs->rdx, regs->r10);
 530
 531         if (test_thread_flag(TIF_SYSCALL_TRACE)
 532             && (current->ptrace & PT_PTRACED))
 533                 syscall_trace(regs);
 534 }
 535
 536 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 537 {
 538         if (unlikely(current->audit_context))
 539                 audit_syscall_exit(current, regs->rax);
 540
 541         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 542              || test_thread_flag(TIF_SINGLESTEP))
 543             && (current->ptrace & PT_PTRACED))
 544                 syscall_trace(regs);
 545 }