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 #include <linux/seccomp.h>
  21 #include <linux/signal.h>
  22
  23 #include <asm/uaccess.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/system.h>
  26 #include <asm/processor.h>
  27 #include <asm/i387.h>
  28 #include <asm/debugreg.h>
  29 #include <asm/ldt.h>
  30 #include <asm/desc.h>
  31 #include <asm/proto.h>
  32 #include <asm/ia32.h>
  33
  34 /*
  35  * does not yet catch signals sent when the child dies.
  36  * in exit.c or in signal.c.
  37  */
  38
  39 /*
  40  * Determines which flags the user has access to [1 = access, 0 = no access].
  41  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
  42  * Also masks reserved bits (63-22, 15, 5, 3, 1).
  43  */
  44 #define FLAG_MASK 0x54dd5UL
  45
  46 /* set's the trap flag. */
  47 #define TRAP_FLAG 0x100UL
  48
  49 /*
  50  * eflags and offset of eflags on child stack..
  51  */
  52 #define EFLAGS offsetof(struct pt_regs, eflags)
  53 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  54
  55 /*
  56  * this routine will get a word off of the processes privileged stack.
  57  * the offset is how far from the base addr as stored in the TSS.
  58  * this routine assumes that all the privileged stacks are in our
  59  * data space.
  60  */
  61 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  62 {
  63         unsigned char *stack;
  64
  65         stack = (unsigned char *)task->thread.rsp0;
  66         stack += offset;
  67         return (*((unsigned long *)stack));
  68 }
  69
  70 /*
  71  * this routine will put a word on the processes privileged stack.
  72  * the offset is how far from the base addr as stored in the TSS.
  73  * this routine assumes that all the privileged stacks are in our
  74  * data space.
  75  */
  76 static inline long put_stack_long(struct task_struct *task, int offset,
  77         unsigned long data)
  78 {
  79         unsigned char * stack;
  80
  81         stack = (unsigned char *) task->thread.rsp0;
  82         stack += offset;
  83         *(unsigned long *) stack = data;
  84         return 0;
  85 }
  86
  87 #define LDT_SEGMENT 4
  88
  89 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  90 {
  91         unsigned long addr, seg;
  92
  93         addr = regs->rip;
  94         seg = regs->cs & 0xffff;
  95
  96         /*
  97          * We'll assume that the code segments in the GDT
  98          * are all zero-based. That is largely true: the
  99          * TLS segments are used for data, and the PNPBIOS
 100          * and APM bios ones we just ignore here.
 101          */
 102         if (seg & LDT_SEGMENT) {
 103                 u32 *desc;
 104                 unsigned long base;
 105
 106                 down(&child->mm->context.sem);
 107                 desc = child->mm->context.ldt + (seg & ~7);
 108                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 109
 110                 /* 16-bit code segment? */
 111                 if (!((desc[1] >> 22) & 1))
 112                         addr &= 0xffff;
 113                 addr += base;
 114                 up(&child->mm->context.sem);
 115         }
 116         return addr;
 117 }
 118
 119 static int is_at_popf(struct task_struct *child, struct pt_regs *regs)
 120 {
 121         int i, copied;
 122         unsigned char opcode[16];
 123         unsigned long addr = convert_rip_to_linear(child, regs);
 124
 125         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 126         for (i = 0; i < copied; i++) {
 127                 switch (opcode[i]) {
 128                 /* popf */
 129                 case 0x9d:
 130                         return 1;
 131
 132                         /* CHECKME: 64 65 */
 133
 134                 /* opcode and address size prefixes */
 135                 case 0x66: case 0x67:
 136                         continue;
 137                 /* irrelevant prefixes (segment overrides and repeats) */
 138                 case 0x26: case 0x2e:
 139                 case 0x36: case 0x3e:
 140                 case 0x64: case 0x65:
 141                 case 0xf0: case 0xf2: case 0xf3:
 142                         continue;
 143
 144                 /* REX prefixes */
 145                 case 0x40 ... 0x4f:
 146                         continue;
 147
 148                         /* CHECKME: f0, f2, f3 */
 149
 150                 /*
 151                  * pushf: NOTE! We should probably not let
 152                  * the user see the TF bit being set. But
 153                  * it's more pain than it's worth to avoid
 154                  * it, and a debugger could emulate this
 155                  * all in user space if it _really_ cares.
 156                  */
 157                 case 0x9c:
 158                 default:
 159                         return 0;
 160                 }
 161         }
 162         return 0;
 163 }
 164
 165 static void set_singlestep(struct task_struct *child)
 166 {
 167         struct pt_regs *regs = task_pt_regs(child);
 168
 169         /*
 170          * Always set TIF_SINGLESTEP - this guarantees that
 171          * we single-step system calls etc..  This will also
 172          * cause us to set TF when returning to user mode.
 173          */
 174         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 175
 176         /*
 177          * If TF was already set, don't do anything else
 178          */
 179         if (regs->eflags & TRAP_FLAG)
 180                 return;
 181
 182         /* Set TF on the kernel stack.. */
 183         regs->eflags |= TRAP_FLAG;
 184
 185         /*
 186          * ..but if TF is changed by the instruction we will trace,
 187          * don't mark it as being "us" that set it, so that we
 188          * won't clear it by hand later.
 189          *
 190          * AK: this is not enough, LAHF and IRET can change TF in user space too.
 191          */
 192         if (is_at_popf(child, regs))
 193                 return;
 194
 195         child->ptrace |= PT_DTRACE;
 196 }
 197
 198 static void clear_singlestep(struct task_struct *child)
 199 {
 200         /* Always clear TIF_SINGLESTEP... */
 201         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 202
 203         /* But touch TF only if it was set by us.. */
 204         if (child->ptrace & PT_DTRACE) {
 205                 struct pt_regs *regs = task_pt_regs(child);
 206                 regs->eflags &= ~TRAP_FLAG;
 207                 child->ptrace &= ~PT_DTRACE;
 208         }
 209 }
 210
 211 /*
 212  * Called by kernel/ptrace.c when detaching..
 213  *
 214  * Make sure the single step bit is not set.
 215  */
 216 void ptrace_disable(struct task_struct *child)
 217 {
 218         clear_singlestep(child);
 219 }
 220
 221 static int putreg(struct task_struct *child,
 222         unsigned long regno, unsigned long value)
 223 {
 224         unsigned long tmp;
 225
 226         /* Some code in the 64bit emulation may not be 64bit clean.
 227            Don't take any chances. */
 228         if (test_tsk_thread_flag(child, TIF_IA32))
 229                 value &= 0xffffffff;
 230         switch (regno) {
 231                 case offsetof(struct user_regs_struct,fs):
 232                         if (value && (value & 3) != 3)
 233                                 return -EIO;
 234                         child->thread.fsindex = value & 0xffff;
 235                         return 0;
 236                 case offsetof(struct user_regs_struct,gs):
 237                         if (value && (value & 3) != 3)
 238                                 return -EIO;
 239                         child->thread.gsindex = value & 0xffff;
 240                         return 0;
 241                 case offsetof(struct user_regs_struct,ds):
 242                         if (value && (value & 3) != 3)
 243                                 return -EIO;
 244                         child->thread.ds = value & 0xffff;
 245                         return 0;
 246                 case offsetof(struct user_regs_struct,es):
 247                         if (value && (value & 3) != 3)
 248                                 return -EIO;
 249                         child->thread.es = value & 0xffff;
 250                         return 0;
 251                 case offsetof(struct user_regs_struct,ss):
 252                         if ((value & 3) != 3)
 253                                 return -EIO;
 254                         value &= 0xffff;
 255                         return 0;
 256                 case offsetof(struct user_regs_struct,fs_base):
 257                         if (value >= TASK_SIZE_OF(child))
 258                                 return -EIO;
 259                         child->thread.fs = value;
 260                         return 0;
 261                 case offsetof(struct user_regs_struct,gs_base):
 262                         if (value >= TASK_SIZE_OF(child))
 263                                 return -EIO;
 264                         child->thread.gs = value;
 265                         return 0;
 266                 case offsetof(struct user_regs_struct, eflags):
 267                         value &= FLAG_MASK;
 268                         tmp = get_stack_long(child, EFL_OFFSET);
 269                         tmp &= ~FLAG_MASK;
 270                         value |= tmp;
 271                         break;
 272                 case offsetof(struct user_regs_struct,cs):
 273                         if ((value & 3) != 3)
 274                                 return -EIO;
 275                         value &= 0xffff;
 276                         break;
 277         }
 278         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 279         return 0;
 280 }
 281
 282 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 283 {
 284         unsigned long val;
 285         switch (regno) {
 286                 case offsetof(struct user_regs_struct, fs):
 287                         return child->thread.fsindex;
 288                 case offsetof(struct user_regs_struct, gs):
 289                         return child->thread.gsindex;
 290                 case offsetof(struct user_regs_struct, ds):
 291                         return child->thread.ds;
 292                 case offsetof(struct user_regs_struct, es):
 293                         return child->thread.es;
 294                 case offsetof(struct user_regs_struct, fs_base):
 295                         return child->thread.fs;
 296                 case offsetof(struct user_regs_struct, gs_base):
 297                         return child->thread.gs;
 298                 default:
 299                         regno = regno - sizeof(struct pt_regs);
 300                         val = get_stack_long(child, regno);
 301                         if (test_tsk_thread_flag(child, TIF_IA32))
 302                                 val &= 0xffffffff;
 303                         return val;
 304         }
 305
 306 }
 307
 308 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 309 {
 310         long i, ret;
 311         unsigned ui;
 312
 313         switch (request) {
 314         /* when I and D space are separate, these will need to be fixed. */
 315         case PTRACE_PEEKTEXT: /* read word at location addr. */
 316         case PTRACE_PEEKDATA: {
 317                 unsigned long tmp;
 318                 int copied;
 319
 320                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 321                 ret = -EIO;
 322                 if (copied != sizeof(tmp))
 323                         break;
 324                 ret = put_user(tmp,(unsigned long __user *) data);
 325                 break;
 326         }
 327
 328         /* read the word at location addr in the USER area. */
 329         case PTRACE_PEEKUSR: {
 330                 unsigned long tmp;
 331
 332                 ret = -EIO;
 333                 if ((addr & 7) ||
 334                     addr > sizeof(struct user) - 7)
 335                         break;
 336
 337                 switch (addr) {
 338                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 339                         tmp = getreg(child, addr);
 340                         break;
 341                 case offsetof(struct user, u_debugreg[0]):
 342                         tmp = child->thread.debugreg0;
 343                         break;
 344                 case offsetof(struct user, u_debugreg[1]):
 345                         tmp = child->thread.debugreg1;
 346                         break;
 347                 case offsetof(struct user, u_debugreg[2]):
 348                         tmp = child->thread.debugreg2;
 349                         break;
 350                 case offsetof(struct user, u_debugreg[3]):
 351                         tmp = child->thread.debugreg3;
 352                         break;
 353                 case offsetof(struct user, u_debugreg[6]):
 354                         tmp = child->thread.debugreg6;
 355                         break;
 356                 case offsetof(struct user, u_debugreg[7]):
 357                         tmp = child->thread.debugreg7;
 358                         break;
 359                 default:
 360                         tmp = 0;
 361                         break;
 362                 }
 363                 ret = put_user(tmp,(unsigned long __user *) data);
 364                 break;
 365         }
 366
 367         /* when I and D space are separate, this will have to be fixed. */
 368         case PTRACE_POKETEXT: /* write the word at location addr. */
 369         case PTRACE_POKEDATA:
 370                 ret = 0;
 371                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 372                         break;
 373                 ret = -EIO;
 374                 break;
 375
 376         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 377         {
 378                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 379                 ret = -EIO;
 380                 if ((addr & 7) ||
 381                     addr > sizeof(struct user) - 7)
 382                         break;
 383
 384                 switch (addr) {
 385                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 386                         ret = putreg(child, addr, data);
 387                         break;
 388                 /* Disallows to set a breakpoint into the vsyscall */
 389                 case offsetof(struct user, u_debugreg[0]):
 390                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 391                         child->thread.debugreg0 = data;
 392                         ret = 0;
 393                         break;
 394                 case offsetof(struct user, u_debugreg[1]):
 395                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 396                         child->thread.debugreg1 = data;
 397                         ret = 0;
 398                         break;
 399                 case offsetof(struct user, u_debugreg[2]):
 400                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 401                         child->thread.debugreg2 = data;
 402                         ret = 0;
 403                         break;
 404                 case offsetof(struct user, u_debugreg[3]):
 405                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 406                         child->thread.debugreg3 = data;
 407                         ret = 0;
 408                         break;
 409                 case offsetof(struct user, u_debugreg[6]):
 410                                   if (data >> 32)
 411                                 break;
 412                         child->thread.debugreg6 = data;
 413                         ret = 0;
 414                         break;
 415                 case offsetof(struct user, u_debugreg[7]):
 416                         /* See arch/i386/kernel/ptrace.c for an explanation of
 417                          * this awkward check.*/
 418                         data &= ~DR_CONTROL_RESERVED;
 419                         for(i=0; i<4; i++)
 420                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 421                                         break;
 422                         if (i == 4) {
 423                                 child->thread.debugreg7 = data;
 424                           ret = 0;
 425                   }
 426                   break;
 427                 }
 428                 break;
 429         }
 430         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 431         case PTRACE_CONT:    /* restart after signal. */
 432
 433                 ret = -EIO;
 434                 if (!valid_signal(data))
 435                         break;
 436                 if (request == PTRACE_SYSCALL)
 437                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 438                 else
 439                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 440                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 441                 child->exit_code = data;
 442                 /* make sure the single step bit is not set. */
 443                 clear_singlestep(child);
 444                 wake_up_process(child);
 445                 ret = 0;
 446                 break;
 447
 448 #ifdef CONFIG_IA32_EMULATION
 449                 /* This makes only sense with 32bit programs. Allow a
 450                    64bit debugger to fully examine them too. Better
 451                    don't use it against 64bit processes, use
 452                    PTRACE_ARCH_PRCTL instead. */
 453         case PTRACE_SET_THREAD_AREA: {
 454                 struct user_desc __user *p;
 455                 int old;
 456                 p = (struct user_desc __user *)data;
 457                 get_user(old,  &p->entry_number);
 458                 put_user(addr, &p->entry_number);
 459                 ret = do_set_thread_area(&child->thread, p);
 460                 put_user(old,  &p->entry_number);
 461                 break;
 462         case PTRACE_GET_THREAD_AREA:
 463                 p = (struct user_desc __user *)data;
 464                 get_user(old,  &p->entry_number);
 465                 put_user(addr, &p->entry_number);
 466                 ret = do_get_thread_area(&child->thread, p);
 467                 put_user(old,  &p->entry_number);
 468                 break;
 469         }
 470 #endif
 471                 /* normal 64bit interface to access TLS data.
 472                    Works just like arch_prctl, except that the arguments
 473                    are reversed. */
 474         case PTRACE_ARCH_PRCTL:
 475                 ret = do_arch_prctl(child, data, addr);
 476                 break;
 477
 478 /*
 479  * make the child exit.  Best I can do is send it a sigkill.
 480  * perhaps it should be put in the status that it wants to
 481  * exit.
 482  */
 483         case PTRACE_KILL:
 484                 ret = 0;
 485                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 486                         break;
 487                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 488                 child->exit_code = SIGKILL;
 489                 /* make sure the single step bit is not set. */
 490                 clear_singlestep(child);
 491                 wake_up_process(child);
 492                 break;
 493
 494         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 495                 ret = -EIO;
 496                 if (!valid_signal(data))
 497                         break;
 498                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 499                 set_singlestep(child);
 500                 child->exit_code = data;
 501                 /* give it a chance to run. */
 502                 wake_up_process(child);
 503                 ret = 0;
 504                 break;
 505
 506         case PTRACE_DETACH:
 507                 /* detach a process that was attached. */
 508                 ret = ptrace_detach(child, data);
 509                 break;
 510
 511         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 512                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 513                                sizeof(struct user_regs_struct))) {
 514                         ret = -EIO;
 515                         break;
 516                 }
 517                 ret = 0;
 518                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 519                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 520                         data += sizeof(long);
 521                 }
 522                 break;
 523         }
 524
 525         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 526                 unsigned long tmp;
 527                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 528                                sizeof(struct user_regs_struct))) {
 529                         ret = -EIO;
 530                         break;
 531                 }
 532                 ret = 0;
 533                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 534                         ret |= __get_user(tmp, (unsigned long __user *) data);
 535                         putreg(child, ui, tmp);
 536                         data += sizeof(long);
 537                 }
 538                 break;
 539         }
 540
 541         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 542                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 543                                sizeof(struct user_i387_struct))) {
 544                         ret = -EIO;
 545                         break;
 546                 }
 547                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 548                 break;
 549         }
 550
 551         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 552                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 553                                sizeof(struct user_i387_struct))) {
 554                         ret = -EIO;
 555                         break;
 556                 }
 557                 set_stopped_child_used_math(child);
 558                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 559                 break;
 560         }
 561
 562         default:
 563                 ret = ptrace_request(child, request, addr, data);
 564                 break;
 565         }
 566         return ret;
 567 }
 568
 569 static void syscall_trace(struct pt_regs *regs)
 570 {
 571
 572 #if 0
 573         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 574                current->comm,
 575                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 576                current_thread_info()->flags, current->ptrace);
 577 #endif
 578
 579         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 580                                 ? 0x80 : 0));
 581         /*
 582          * this isn't the same as continuing with a signal, but it will do
 583          * for normal use.  strace only continues with a signal if the
 584          * stopping signal is not SIGTRAP.  -brl
 585          */
 586         if (current->exit_code) {
 587                 send_sig(current->exit_code, current, 1);
 588                 current->exit_code = 0;
 589         }
 590 }
 591
 592 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 593 {
 594         /* do the secure computing check first */
 595         secure_computing(regs->orig_rax);
 596
 597         if (test_thread_flag(TIF_SYSCALL_TRACE)
 598             && (current->ptrace & PT_PTRACED))
 599                 syscall_trace(regs);
 600
 601         if (unlikely(current->audit_context)) {
 602                 if (test_thread_flag(TIF_IA32)) {
 603                         audit_syscall_entry(AUDIT_ARCH_I386,
 604                                             regs->orig_rax,
 605                                             regs->rbx, regs->rcx,
 606                                             regs->rdx, regs->rsi);
 607                 } else {
 608                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 609                                             regs->orig_rax,
 610                                             regs->rdi, regs->rsi,
 611                                             regs->rdx, regs->r10);
 612                 }
 613         }
 614 }
 615
 616 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 617 {
 618         if (unlikely(current->audit_context))
 619                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 620
 621         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 622              || test_thread_flag(TIF_SINGLESTEP))
 623             && (current->ptrace & PT_PTRACED))
 624                 syscall_trace(regs);
 625 }