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 >= TASK_SIZE)
 133                                 return -EIO;
 134                         child->thread.fs = value;
 135                         return 0;
 136                 case offsetof(struct user_regs_struct,gs_base):
 137                         if (value >= TASK_SIZE)
 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                 case offsetof(struct user_regs_struct, rip):
 153                         /* Check if the new RIP address is canonical */
 154                         if (value >= TASK_SIZE)
 155                                 return -EIO;
 156                         break;
 157         }
 158         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 159         return 0;
 160 }
 161
 162 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 163 {
 164         unsigned long val;
 165         switch (regno) {
 166                 case offsetof(struct user_regs_struct, fs):
 167                         return child->thread.fsindex;
 168                 case offsetof(struct user_regs_struct, gs):
 169                         return child->thread.gsindex;
 170                 case offsetof(struct user_regs_struct, ds):
 171                         return child->thread.ds;
 172                 case offsetof(struct user_regs_struct, es):
 173                         return child->thread.es;
 174                 case offsetof(struct user_regs_struct, fs_base):
 175                         return child->thread.fs;
 176                 case offsetof(struct user_regs_struct, gs_base):
 177                         return child->thread.gs;
 178                 default:
 179                         regno = regno - sizeof(struct pt_regs);
 180                         val = get_stack_long(child, regno);
 181                         if (test_tsk_thread_flag(child, TIF_IA32))
 182                                 val &= 0xffffffff;
 183                         return val;
 184         }
 185
 186 }
 187
 188 asmlinkage long sys_ptrace(long request, long pid, unsigned long addr, long data)
 189 {
 190         struct task_struct *child;
 191         long i, ret;
 192         unsigned ui;
 193
 194         /* This lock_kernel fixes a subtle race with suid exec */
 195         lock_kernel();
 196         ret = -EPERM;
 197         if (request == PTRACE_TRACEME) {
 198                 /* are we already being traced? */
 199                 if (current->ptrace & PT_PTRACED)
 200                         goto out;
 201                 ret = security_ptrace(current->parent, current);
 202                 if (ret)
 203                         goto out;
 204                 /* set the ptrace bit in the process flags. */
 205                 current->ptrace |= PT_PTRACED;
 206                 ret = 0;
 207                 goto out;
 208         }
 209         ret = -ESRCH;
 210         read_lock(&tasklist_lock);
 211         child = find_task_by_pid(pid);
 212         if (child)
 213                 get_task_struct(child);
 214         read_unlock(&tasklist_lock);
 215         if (!child)
 216                 goto out;
 217         if (!vx_check(vx_task_xid(child), VX_WATCH|VX_IDENT))
 218                 goto out_tsk;
 219
 220         ret = -EPERM;
 221         if (pid == 1)           /* you may not mess with init */
 222                 goto out_tsk;
 223
 224         if (request == PTRACE_ATTACH) {
 225                 ret = ptrace_attach(child);
 226                 goto out_tsk;
 227         }
 228         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 229         if (ret < 0)
 230                 goto out_tsk;
 231
 232         switch (request) {
 233         /* when I and D space are separate, these will need to be fixed. */
 234         case PTRACE_PEEKTEXT: /* read word at location addr. */
 235         case PTRACE_PEEKDATA: {
 236                 unsigned long tmp;
 237                 int copied;
 238
 239                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 240                 ret = -EIO;
 241                 if (copied != sizeof(tmp))
 242                         break;
 243                 ret = put_user(tmp,(unsigned long __user *) data);
 244                 break;
 245         }
 246
 247         /* read the word at location addr in the USER area. */
 248         case PTRACE_PEEKUSR: {
 249                 unsigned long tmp;
 250
 251                 ret = -EIO;
 252                 if ((addr & 7) ||
 253                     addr > sizeof(struct user) - 7)
 254                         break;
 255
 256                 switch (addr) {
 257                 case 0 ... sizeof(struct user_regs_struct):
 258                         tmp = getreg(child, addr);
 259                         break;
 260                 case offsetof(struct user, u_debugreg[0]):
 261                         tmp = child->thread.debugreg0;
 262                         break;
 263                 case offsetof(struct user, u_debugreg[1]):
 264                         tmp = child->thread.debugreg1;
 265                         break;
 266                 case offsetof(struct user, u_debugreg[2]):
 267                         tmp = child->thread.debugreg2;
 268                         break;
 269                 case offsetof(struct user, u_debugreg[3]):
 270                         tmp = child->thread.debugreg3;
 271                         break;
 272                 case offsetof(struct user, u_debugreg[6]):
 273                         tmp = child->thread.debugreg6;
 274                         break;
 275                 case offsetof(struct user, u_debugreg[7]):
 276                         tmp = child->thread.debugreg7;
 277                         break;
 278                 default:
 279                         tmp = 0;
 280                         break;
 281                 }
 282                 ret = put_user(tmp,(unsigned long __user *) data);
 283                 break;
 284         }
 285
 286         /* when I and D space are separate, this will have to be fixed. */
 287         case PTRACE_POKETEXT: /* write the word at location addr. */
 288         case PTRACE_POKEDATA:
 289                 ret = 0;
 290                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 291                         break;
 292                 ret = -EIO;
 293                 break;
 294
 295         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 296                 ret = -EIO;
 297                 if ((addr & 7) ||
 298                     addr > sizeof(struct user) - 7)
 299                         break;
 300
 301                 switch (addr) {
 302                 case 0 ... sizeof(struct user_regs_struct):
 303                         ret = putreg(child, addr, data);
 304                         break;
 305                 /* Disallows to set a breakpoint into the vsyscall */
 306                 case offsetof(struct user, u_debugreg[0]):
 307                         if (data >= TASK_SIZE-7) break;
 308                         child->thread.debugreg0 = data;
 309                         ret = 0;
 310                         break;
 311                 case offsetof(struct user, u_debugreg[1]):
 312                         if (data >= TASK_SIZE-7) break;
 313                         child->thread.debugreg1 = data;
 314                         ret = 0;
 315                         break;
 316                 case offsetof(struct user, u_debugreg[2]):
 317                         if (data >= TASK_SIZE-7) break;
 318                         child->thread.debugreg2 = data;
 319                         ret = 0;
 320                         break;
 321                 case offsetof(struct user, u_debugreg[3]):
 322                         if (data >= TASK_SIZE-7) break;
 323                         child->thread.debugreg3 = data;
 324                         ret = 0;
 325                         break;
 326                 case offsetof(struct user, u_debugreg[6]):
 327                                   if (data >> 32)
 328                                 break;
 329                         child->thread.debugreg6 = data;
 330                         ret = 0;
 331                         break;
 332                 case offsetof(struct user, u_debugreg[7]):
 333                         /* See arch/i386/kernel/ptrace.c for an explanation of
 334                          * this awkward check.*/
 335                                   data &= ~DR_CONTROL_RESERVED;
 336                                   for(i=0; i<4; i++)
 337                                           if ((0x5454 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 338                                         break;
 339                         if (i == 4) {
 340                                 child->thread.debugreg7 = data;
 341                           ret = 0;
 342                   }
 343                   break;
 344                 }
 345                 break;
 346         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 347         case PTRACE_CONT: { /* restart after signal. */
 348                 long tmp;
 349
 350                 ret = -EIO;
 351                 if ((unsigned long) data > _NSIG)
 352                         break;
 353                 if (request == PTRACE_SYSCALL)
 354                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 355                 else
 356                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 357                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 358                 child->exit_code = data;
 359         /* make sure the single step bit is not set. */
 360                 tmp = get_stack_long(child, EFL_OFFSET);
 361                 tmp &= ~TRAP_FLAG;
 362                 put_stack_long(child, EFL_OFFSET,tmp);
 363                 wake_up_process(child);
 364                 ret = 0;
 365                 break;
 366         }
 367
 368 #ifdef CONFIG_IA32_EMULATION
 369                 /* This makes only sense with 32bit programs. Allow a
 370                    64bit debugger to fully examine them too. Better
 371                    don't use it against 64bit processes, use
 372                    PTRACE_ARCH_PRCTL instead. */
 373         case PTRACE_SET_THREAD_AREA: {
 374                 struct user_desc __user *p;
 375                 int old;
 376                 p = (struct user_desc __user *)data;
 377                 get_user(old,  &p->entry_number);
 378                 put_user(addr, &p->entry_number);
 379                 ret = do_set_thread_area(&child->thread, p);
 380                 put_user(old,  &p->entry_number);
 381                 break;
 382         case PTRACE_GET_THREAD_AREA:
 383                 p = (struct user_desc __user *)data;
 384                 get_user(old,  &p->entry_number);
 385                 put_user(addr, &p->entry_number);
 386                 ret = do_get_thread_area(&child->thread, p);
 387                 put_user(old,  &p->entry_number);
 388                 break;
 389         }
 390 #endif
 391                 /* normal 64bit interface to access TLS data.
 392                    Works just like arch_prctl, except that the arguments
 393                    are reversed. */
 394         case PTRACE_ARCH_PRCTL:
 395                 ret = do_arch_prctl(child, data, addr);
 396                 break;
 397
 398 /*
 399  * make the child exit.  Best I can do is send it a sigkill.
 400  * perhaps it should be put in the status that it wants to
 401  * exit.
 402  */
 403         case PTRACE_KILL: {
 404                 long tmp;
 405
 406                 ret = 0;
 407                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 408                         break;
 409                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 410                 child->exit_code = SIGKILL;
 411                 /* make sure the single step bit is not set. */
 412                 tmp = get_stack_long(child, EFL_OFFSET) & ~TRAP_FLAG;
 413                 put_stack_long(child, EFL_OFFSET, tmp);
 414                 wake_up_process(child);
 415                 break;
 416         }
 417
 418         case PTRACE_SINGLESTEP: {  /* set the trap flag. */
 419                 long tmp;
 420
 421                 ret = -EIO;
 422                 if ((unsigned long) data > _NSIG)
 423                         break;
 424                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 425                 if ((child->ptrace & PT_DTRACE) == 0) {
 426                         /* Spurious delayed TF traps may occur */
 427                         child->ptrace |= PT_DTRACE;
 428                 }
 429                 tmp = get_stack_long(child, EFL_OFFSET) | TRAP_FLAG;
 430                 put_stack_long(child, EFL_OFFSET, tmp);
 431                 set_tsk_thread_flag(child, TIF_SINGLESTEP);
 432                 child->exit_code = data;
 433                 /* give it a chance to run. */
 434                 wake_up_process(child);
 435                 ret = 0;
 436                 break;
 437         }
 438
 439         case PTRACE_DETACH:
 440                 /* detach a process that was attached. */
 441                 ret = ptrace_detach(child, data);
 442                 break;
 443
 444         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 445                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 446                                sizeof(struct user_regs_struct))) {
 447                         ret = -EIO;
 448                         break;
 449                 }
 450                 ret = 0;
 451                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 452                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 453                         data += sizeof(long);
 454                 }
 455                 break;
 456         }
 457
 458         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 459                 unsigned long tmp;
 460                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 461                                sizeof(struct user_regs_struct))) {
 462                         ret = -EIO;
 463                         break;
 464                 }
 465                 ret = 0;
 466                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 467                         ret |= __get_user(tmp, (unsigned long __user *) data);
 468                         putreg(child, ui, tmp);
 469                         data += sizeof(long);
 470                 }
 471                 break;
 472         }
 473
 474         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 475                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 476                                sizeof(struct user_i387_struct))) {
 477                         ret = -EIO;
 478                         break;
 479                 }
 480                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 481                 break;
 482         }
 483
 484         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 485                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 486                                sizeof(struct user_i387_struct))) {
 487                         ret = -EIO;
 488                         break;
 489                 }
 490                 set_stopped_child_used_math(child);
 491                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 492                 break;
 493         }
 494
 495         default:
 496                 ret = ptrace_request(child, request, addr, data);
 497                 break;
 498         }
 499 out_tsk:
 500         put_task_struct(child);
 501 out:
 502         unlock_kernel();
 503         return ret;
 504 }
 505
 506 static void syscall_trace(struct pt_regs *regs)
 507 {
 508
 509 #if 0
 510         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 511                current->comm,
 512                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 513                current_thread_info()->flags, current->ptrace);
 514 #endif
 515
 516         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 517                                 ? 0x80 : 0));
 518         /*
 519          * this isn't the same as continuing with a signal, but it will do
 520          * for normal use.  strace only continues with a signal if the
 521          * stopping signal is not SIGTRAP.  -brl
 522          */
 523         if (current->exit_code) {
 524                 send_sig(current->exit_code, current, 1);
 525                 current->exit_code = 0;
 526         }
 527 }
 528
 529 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 530 {
 531         if (unlikely(current->audit_context))
 532                 audit_syscall_entry(current, regs->orig_rax,
 533                                     regs->rdi, regs->rsi,
 534                                     regs->rdx, regs->r10);
 535
 536         if (test_thread_flag(TIF_SYSCALL_TRACE)
 537             && (current->ptrace & PT_PTRACED))
 538                 syscall_trace(regs);
 539 }
 540
 541 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 542 {
 543         if (unlikely(current->audit_context))
 544                 audit_syscall_exit(current, regs->rax);
 545
 546         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 547              || test_thread_flag(TIF_SINGLESTEP))
 548             && (current->ptrace & PT_PTRACED))
 549                 syscall_trace(regs);
 550 }