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