arch/x86/kernel/ptrace_64.c

   1 /* By Ross Biro 1/23/92 */
   2 /*
   3  * Pentium III FXSR, SSE support
   4  *      Gareth Hughes <gareth@valinux.com>, May 2000
   5  *
   6  * x86-64 port 2000-2002 Andi Kleen
   7  */
   8
   9 #include <linux/kernel.h>
  10 #include <linux/sched.h>
  11 #include <linux/mm.h>
  12 #include <linux/smp.h>
  13 #include <linux/errno.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/user.h>
  16 #include <linux/security.h>
  17 #include <linux/audit.h>
  18 #include <linux/seccomp.h>
  19 #include <linux/signal.h>
  20
  21 #include <asm/uaccess.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/system.h>
  24 #include <asm/processor.h>
  25 #include <asm/i387.h>
  26 #include <asm/debugreg.h>
  27 #include <asm/ldt.h>
  28 #include <asm/desc.h>
  29 #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 /*
  38  * Determines which flags the user has access to [1 = access, 0 = no access].
  39  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
  40  * Also masks reserved bits (63-22, 15, 5, 3, 1).
  41  */
  42 #define FLAG_MASK 0x54dd5UL
  43
  44 /* set's the trap flag. */
  45 #define TRAP_FLAG 0x100UL
  46
  47 /*
  48  * eflags and offset of eflags on child stack..
  49  */
  50 #define EFLAGS offsetof(struct pt_regs, eflags)
  51 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  52
  53 /*
  54  * this routine will get a word off of the processes privileged stack.
  55  * the offset is how far from the base addr as stored in the TSS.
  56  * this routine assumes that all the privileged stacks are in our
  57  * data space.
  58  */
  59 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  60 {
  61         unsigned char *stack;
  62
  63         stack = (unsigned char *)task->thread.rsp0;
  64         stack += offset;
  65         return (*((unsigned long *)stack));
  66 }
  67
  68 /*
  69  * this routine will put a word on the processes privileged stack.
  70  * the offset is how far from the base addr as stored in the TSS.
  71  * this routine assumes that all the privileged stacks are in our
  72  * data space.
  73  */
  74 static inline long put_stack_long(struct task_struct *task, int offset,
  75         unsigned long data)
  76 {
  77         unsigned char * stack;
  78
  79         stack = (unsigned char *) task->thread.rsp0;
  80         stack += offset;
  81         *(unsigned long *) stack = data;
  82         return 0;
  83 }
  84
  85 #define LDT_SEGMENT 4
  86
  87 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  88 {
  89         unsigned long addr, seg;
  90
  91         addr = regs->rip;
  92         seg = regs->cs & 0xffff;
  93
  94         /*
  95          * We'll assume that the code segments in the GDT
  96          * are all zero-based. That is largely true: the
  97          * TLS segments are used for data, and the PNPBIOS
  98          * and APM bios ones we just ignore here.
  99          */
 100         if (seg & LDT_SEGMENT) {
 101                 u32 *desc;
 102                 unsigned long base;
 103
 104                 seg &= ~7UL;
 105
 106                 mutex_lock(&child->mm->context.lock);
 107                 if (unlikely((seg >> 3) >= child->mm->context.size))
 108                         addr = -1L; /* bogus selector, access would fault */
 109                 else {
 110                         desc = child->mm->context.ldt + seg;
 111                         base = ((desc[0] >> 16) |
 112                                 ((desc[1] & 0xff) << 16) |
 113                                 (desc[1] & 0xff000000));
 114
 115                         /* 16-bit code segment? */
 116                         if (!((desc[1] >> 22) & 1))
 117                                 addr &= 0xffff;
 118                         addr += base;
 119                 }
 120                 mutex_unlock(&child->mm->context.lock);
 121         }
 122
 123         return addr;
 124 }
 125
 126 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 127 {
 128         int i, copied;
 129         unsigned char opcode[15];
 130         unsigned long addr = convert_rip_to_linear(child, regs);
 131
 132         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 133         for (i = 0; i < copied; i++) {
 134                 switch (opcode[i]) {
 135                 /* popf and iret */
 136                 case 0x9d: case 0xcf:
 137                         return 1;
 138
 139                         /* CHECKME: 64 65 */
 140
 141                 /* opcode and address size prefixes */
 142                 case 0x66: case 0x67:
 143                         continue;
 144                 /* irrelevant prefixes (segment overrides and repeats) */
 145                 case 0x26: case 0x2e:
 146                 case 0x36: case 0x3e:
 147                 case 0x64: case 0x65:
 148                 case 0xf2: case 0xf3:
 149                         continue;
 150
 151                 case 0x40 ... 0x4f:
 152                         if (regs->cs != __USER_CS)
 153                                 /* 32-bit mode: register increment */
 154                                 return 0;
 155                         /* 64-bit mode: REX prefix */
 156                         continue;
 157
 158                         /* CHECKME: f2, f3 */
 159
 160                 /*
 161                  * pushf: NOTE! We should probably not let
 162                  * the user see the TF bit being set. But
 163                  * it's more pain than it's worth to avoid
 164                  * it, and a debugger could emulate this
 165                  * all in user space if it _really_ cares.
 166                  */
 167                 case 0x9c:
 168                 default:
 169                         return 0;
 170                 }
 171         }
 172         return 0;
 173 }
 174
 175 static void set_singlestep(struct task_struct *child)
 176 {
 177         struct pt_regs *regs = task_pt_regs(child);
 178
 179         /*
 180          * Always set TIF_SINGLESTEP - this guarantees that
 181          * we single-step system calls etc..  This will also
 182          * cause us to set TF when returning to user mode.
 183          */
 184         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 185
 186         /*
 187          * If TF was already set, don't do anything else
 188          */
 189         if (regs->eflags & TRAP_FLAG)
 190                 return;
 191
 192         /* Set TF on the kernel stack.. */
 193         regs->eflags |= TRAP_FLAG;
 194
 195         /*
 196          * ..but if TF is changed by the instruction we will trace,
 197          * don't mark it as being "us" that set it, so that we
 198          * won't clear it by hand later.
 199          */
 200         if (is_setting_trap_flag(child, regs))
 201                 return;
 202
 203         child->ptrace |= PT_DTRACE;
 204 }
 205
 206 static void clear_singlestep(struct task_struct *child)
 207 {
 208         /* Always clear TIF_SINGLESTEP... */
 209         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 210
 211         /* But touch TF only if it was set by us.. */
 212         if (child->ptrace & PT_DTRACE) {
 213                 struct pt_regs *regs = task_pt_regs(child);
 214                 regs->eflags &= ~TRAP_FLAG;
 215                 child->ptrace &= ~PT_DTRACE;
 216         }
 217 }
 218
 219 /*
 220  * Called by kernel/ptrace.c when detaching..
 221  *
 222  * Make sure the single step bit is not set.
 223  */
 224 void ptrace_disable(struct task_struct *child)
 225 {
 226         clear_singlestep(child);
 227 }
 228
 229 static int putreg(struct task_struct *child,
 230         unsigned long regno, unsigned long value)
 231 {
 232         unsigned long tmp;
 233
 234         switch (regno) {
 235                 case offsetof(struct user_regs_struct,fs):
 236                         if (value && (value & 3) != 3)
 237                                 return -EIO;
 238                         child->thread.fsindex = value & 0xffff;
 239                         return 0;
 240                 case offsetof(struct user_regs_struct,gs):
 241                         if (value && (value & 3) != 3)
 242                                 return -EIO;
 243                         child->thread.gsindex = value & 0xffff;
 244                         return 0;
 245                 case offsetof(struct user_regs_struct,ds):
 246                         if (value && (value & 3) != 3)
 247                                 return -EIO;
 248                         child->thread.ds = value & 0xffff;
 249                         return 0;
 250                 case offsetof(struct user_regs_struct,es):
 251                         if (value && (value & 3) != 3)
 252                                 return -EIO;
 253                         child->thread.es = value & 0xffff;
 254                         return 0;
 255                 case offsetof(struct user_regs_struct,ss):
 256                         if ((value & 3) != 3)
 257                                 return -EIO;
 258                         value &= 0xffff;
 259                         return 0;
 260                 case offsetof(struct user_regs_struct,fs_base):
 261                         if (value >= TASK_SIZE_OF(child))
 262                                 return -EIO;
 263                         child->thread.fs = value;
 264                         return 0;
 265                 case offsetof(struct user_regs_struct,gs_base):
 266                         if (value >= TASK_SIZE_OF(child))
 267                                 return -EIO;
 268                         child->thread.gs = value;
 269                         return 0;
 270                 case offsetof(struct user_regs_struct, eflags):
 271                         value &= FLAG_MASK;
 272                         tmp = get_stack_long(child, EFL_OFFSET);
 273                         tmp &= ~FLAG_MASK;
 274                         value |= tmp;
 275                         break;
 276                 case offsetof(struct user_regs_struct,cs):
 277                         if ((value & 3) != 3)
 278                                 return -EIO;
 279                         value &= 0xffff;
 280                         break;
 281         }
 282         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 283         return 0;
 284 }
 285
 286 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 287 {
 288         unsigned long val;
 289         switch (regno) {
 290                 case offsetof(struct user_regs_struct, fs):
 291                         return child->thread.fsindex;
 292                 case offsetof(struct user_regs_struct, gs):
 293                         return child->thread.gsindex;
 294                 case offsetof(struct user_regs_struct, ds):
 295                         return child->thread.ds;
 296                 case offsetof(struct user_regs_struct, es):
 297                         return child->thread.es;
 298                 case offsetof(struct user_regs_struct, fs_base):
 299                         return child->thread.fs;
 300                 case offsetof(struct user_regs_struct, gs_base):
 301                         return child->thread.gs;
 302                 default:
 303                         regno = regno - sizeof(struct pt_regs);
 304                         val = get_stack_long(child, regno);
 305                         if (test_tsk_thread_flag(child, TIF_IA32))
 306                                 val &= 0xffffffff;
 307                         return val;
 308         }
 309
 310 }
 311
 312 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 313 {
 314         long i, ret;
 315         unsigned ui;
 316
 317         switch (request) {
 318         /* when I and D space are separate, these will need to be fixed. */
 319         case PTRACE_PEEKTEXT: /* read word at location addr. */
 320         case PTRACE_PEEKDATA:
 321                 ret = generic_ptrace_peekdata(child, addr, data);
 322                 break;
 323
 324         /* read the word at location addr in the USER area. */
 325         case PTRACE_PEEKUSR: {
 326                 unsigned long tmp;
 327
 328                 ret = -EIO;
 329                 if ((addr & 7) ||
 330                     addr > sizeof(struct user) - 7)
 331                         break;
 332
 333                 switch (addr) {
 334                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 335                         tmp = getreg(child, addr);
 336                         break;
 337                 case offsetof(struct user, u_debugreg[0]):
 338                         tmp = child->thread.debugreg0;
 339                         break;
 340                 case offsetof(struct user, u_debugreg[1]):
 341                         tmp = child->thread.debugreg1;
 342                         break;
 343                 case offsetof(struct user, u_debugreg[2]):
 344                         tmp = child->thread.debugreg2;
 345                         break;
 346                 case offsetof(struct user, u_debugreg[3]):
 347                         tmp = child->thread.debugreg3;
 348                         break;
 349                 case offsetof(struct user, u_debugreg[6]):
 350                         tmp = child->thread.debugreg6;
 351                         break;
 352                 case offsetof(struct user, u_debugreg[7]):
 353                         tmp = child->thread.debugreg7;
 354                         break;
 355                 default:
 356                         tmp = 0;
 357                         break;
 358                 }
 359                 ret = put_user(tmp,(unsigned long __user *) data);
 360                 break;
 361         }
 362
 363         /* when I and D space are separate, this will have to be fixed. */
 364         case PTRACE_POKETEXT: /* write the word at location addr. */
 365         case PTRACE_POKEDATA:
 366                 ret = generic_ptrace_pokedata(child, addr, data);
 367                 break;
 368
 369         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 370         {
 371                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 372                 ret = -EIO;
 373                 if ((addr & 7) ||
 374                     addr > sizeof(struct user) - 7)
 375                         break;
 376
 377                 switch (addr) {
 378                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 379                         ret = putreg(child, addr, data);
 380                         break;
 381                 /* Disallows to set a breakpoint into the vsyscall */
 382                 case offsetof(struct user, u_debugreg[0]):
 383                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 384                         child->thread.debugreg0 = data;
 385                         ret = 0;
 386                         break;
 387                 case offsetof(struct user, u_debugreg[1]):
 388                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 389                         child->thread.debugreg1 = data;
 390                         ret = 0;
 391                         break;
 392                 case offsetof(struct user, u_debugreg[2]):
 393                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 394                         child->thread.debugreg2 = data;
 395                         ret = 0;
 396                         break;
 397                 case offsetof(struct user, u_debugreg[3]):
 398                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 399                         child->thread.debugreg3 = data;
 400                         ret = 0;
 401                         break;
 402                 case offsetof(struct user, u_debugreg[6]):
 403                                   if (data >> 32)
 404                                 break;
 405                         child->thread.debugreg6 = data;
 406                         ret = 0;
 407                         break;
 408                 case offsetof(struct user, u_debugreg[7]):
 409                         /* See arch/i386/kernel/ptrace.c for an explanation of
 410                          * this awkward check.*/
 411                         data &= ~DR_CONTROL_RESERVED;
 412                         for(i=0; i<4; i++)
 413                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 414                                         break;
 415                         if (i == 4) {
 416                           child->thread.debugreg7 = data;
 417                           if (data)
 418                                 set_tsk_thread_flag(child, TIF_DEBUG);
 419                           else
 420                                 clear_tsk_thread_flag(child, TIF_DEBUG);
 421                           ret = 0;
 422                         }
 423                   break;
 424                 }
 425                 break;
 426         }
 427         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 428         case PTRACE_CONT:    /* restart after signal. */
 429
 430                 ret = -EIO;
 431                 if (!valid_signal(data))
 432                         break;
 433                 if (request == PTRACE_SYSCALL)
 434                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 435                 else
 436                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 437                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 438                 child->exit_code = data;
 439                 /* make sure the single step bit is not set. */
 440                 clear_singlestep(child);
 441                 wake_up_process(child);
 442                 ret = 0;
 443                 break;
 444
 445 #ifdef CONFIG_IA32_EMULATION
 446                 /* This makes only sense with 32bit programs. Allow a
 447                    64bit debugger to fully examine them too. Better
 448                    don't use it against 64bit processes, use
 449                    PTRACE_ARCH_PRCTL instead. */
 450         case PTRACE_SET_THREAD_AREA: {
 451                 struct user_desc __user *p;
 452                 int old;
 453                 p = (struct user_desc __user *)data;
 454                 get_user(old,  &p->entry_number);
 455                 put_user(addr, &p->entry_number);
 456                 ret = do_set_thread_area(&child->thread, p);
 457                 put_user(old,  &p->entry_number);
 458                 break;
 459         case PTRACE_GET_THREAD_AREA:
 460                 p = (struct user_desc __user *)data;
 461                 get_user(old,  &p->entry_number);
 462                 put_user(addr, &p->entry_number);
 463                 ret = do_get_thread_area(&child->thread, p);
 464                 put_user(old,  &p->entry_number);
 465                 break;
 466         }
 467 #endif
 468                 /* normal 64bit interface to access TLS data.
 469                    Works just like arch_prctl, except that the arguments
 470                    are reversed. */
 471         case PTRACE_ARCH_PRCTL:
 472                 ret = do_arch_prctl(child, data, addr);
 473                 break;
 474
 475 /*
 476  * make the child exit.  Best I can do is send it a sigkill.
 477  * perhaps it should be put in the status that it wants to
 478  * exit.
 479  */
 480         case PTRACE_KILL:
 481                 ret = 0;
 482                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 483                         break;
 484                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 485                 child->exit_code = SIGKILL;
 486                 /* make sure the single step bit is not set. */
 487                 clear_singlestep(child);
 488                 wake_up_process(child);
 489                 break;
 490
 491         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 492                 ret = -EIO;
 493                 if (!valid_signal(data))
 494                         break;
 495                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 496                 set_singlestep(child);
 497                 child->exit_code = data;
 498                 /* give it a chance to run. */
 499                 wake_up_process(child);
 500                 ret = 0;
 501                 break;
 502
 503         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 504                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 505                                sizeof(struct user_regs_struct))) {
 506                         ret = -EIO;
 507                         break;
 508                 }
 509                 ret = 0;
 510                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 511                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 512                         data += sizeof(long);
 513                 }
 514                 break;
 515         }
 516
 517         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 518                 unsigned long tmp;
 519                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 520                                sizeof(struct user_regs_struct))) {
 521                         ret = -EIO;
 522                         break;
 523                 }
 524                 ret = 0;
 525                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 526                         ret = __get_user(tmp, (unsigned long __user *) data);
 527                         if (ret)
 528                                 break;
 529                         ret = putreg(child, ui, tmp);
 530                         if (ret)
 531                                 break;
 532                         data += sizeof(long);
 533                 }
 534                 break;
 535         }
 536
 537         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 538                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 539                                sizeof(struct user_i387_struct))) {
 540                         ret = -EIO;
 541                         break;
 542                 }
 543                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 544                 break;
 545         }
 546
 547         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 548                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 549                                sizeof(struct user_i387_struct))) {
 550                         ret = -EIO;
 551                         break;
 552                 }
 553                 set_stopped_child_used_math(child);
 554                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 555                 break;
 556         }
 557
 558         default:
 559                 ret = ptrace_request(child, request, addr, data);
 560                 break;
 561         }
 562         return ret;
 563 }
 564
 565 static void syscall_trace(struct pt_regs *regs)
 566 {
 567
 568 #if 0
 569         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 570                current->comm,
 571                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 572                current_thread_info()->flags, current->ptrace);
 573 #endif
 574
 575         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 576                                 ? 0x80 : 0));
 577         /*
 578          * this isn't the same as continuing with a signal, but it will do
 579          * for normal use.  strace only continues with a signal if the
 580          * stopping signal is not SIGTRAP.  -brl
 581          */
 582         if (current->exit_code) {
 583                 send_sig(current->exit_code, current, 1);
 584                 current->exit_code = 0;
 585         }
 586 }
 587
 588 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 589 {
 590         /* do the secure computing check first */
 591         secure_computing(regs->orig_rax);
 592
 593         if (test_thread_flag(TIF_SYSCALL_TRACE)
 594             && (current->ptrace & PT_PTRACED))
 595                 syscall_trace(regs);
 596
 597         if (unlikely(current->audit_context)) {
 598                 if (test_thread_flag(TIF_IA32)) {
 599                         audit_syscall_entry(AUDIT_ARCH_I386,
 600                                             regs->orig_rax,
 601                                             regs->rbx, regs->rcx,
 602                                             regs->rdx, regs->rsi);
 603                 } else {
 604                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 605                                             regs->orig_rax,
 606                                             regs->rdi, regs->rsi,
 607                                             regs->rdx, regs->r10);
 608                 }
 609         }
 610 }
 611
 612 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 613 {
 614         if (unlikely(current->audit_context))
 615                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 616
 617         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 618              || test_thread_flag(TIF_SINGLESTEP))
 619             && (current->ptrace & PT_PTRACED))
 620                 syscall_trace(regs);
 621 }