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/prctl.h>
  26 #include <asm/i387.h>
  27 #include <asm/debugreg.h>
  28 #include <asm/ldt.h>
  29 #include <asm/desc.h>
  30 #include <asm/proto.h>
  31 #include <asm/ia32.h>
  32
  33 /*
  34  * does not yet catch signals sent when the child dies.
  35  * in exit.c or in signal.c.
  36  */
  37
  38 /*
  39  * Determines which flags the user has access to [1 = access, 0 = no access].
  40  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
  41  * Also masks reserved bits (63-22, 15, 5, 3, 1).
  42  */
  43 #define FLAG_MASK 0x54dd5UL
  44
  45 /* set's the trap flag. */
  46 #define TRAP_FLAG 0x100UL
  47
  48 /*
  49  * eflags and offset of eflags on child stack..
  50  */
  51 #define EFLAGS offsetof(struct pt_regs, eflags)
  52 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  53
  54 /*
  55  * this routine will get a word off of the processes privileged stack.
  56  * the offset is how far from the base addr as stored in the TSS.
  57  * this routine assumes that all the privileged stacks are in our
  58  * data space.
  59  */
  60 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  61 {
  62         unsigned char *stack;
  63
  64         stack = (unsigned char *)task->thread.rsp0;
  65         stack += offset;
  66         return (*((unsigned long *)stack));
  67 }
  68
  69 /*
  70  * this routine will put a word on the processes privileged stack.
  71  * the offset is how far from the base addr as stored in the TSS.
  72  * this routine assumes that all the privileged stacks are in our
  73  * data space.
  74  */
  75 static inline long put_stack_long(struct task_struct *task, int offset,
  76         unsigned long data)
  77 {
  78         unsigned char * stack;
  79
  80         stack = (unsigned char *) task->thread.rsp0;
  81         stack += offset;
  82         *(unsigned long *) stack = data;
  83         return 0;
  84 }
  85
  86 #define LDT_SEGMENT 4
  87
  88 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  89 {
  90         unsigned long addr, seg;
  91
  92         addr = regs->rip;
  93         seg = regs->cs & 0xffff;
  94
  95         /*
  96          * We'll assume that the code segments in the GDT
  97          * are all zero-based. That is largely true: the
  98          * TLS segments are used for data, and the PNPBIOS
  99          * and APM bios ones we just ignore here.
 100          */
 101         if (seg & LDT_SEGMENT) {
 102                 u32 *desc;
 103                 unsigned long base;
 104
 105                 seg &= ~7UL;
 106
 107                 mutex_lock(&child->mm->context.lock);
 108                 if (unlikely((seg >> 3) >= child->mm->context.size))
 109                         addr = -1L; /* bogus selector, access would fault */
 110                 else {
 111                         desc = child->mm->context.ldt + seg;
 112                         base = ((desc[0] >> 16) |
 113                                 ((desc[1] & 0xff) << 16) |
 114                                 (desc[1] & 0xff000000));
 115
 116                         /* 16-bit code segment? */
 117                         if (!((desc[1] >> 22) & 1))
 118                                 addr &= 0xffff;
 119                         addr += base;
 120                 }
 121                 mutex_unlock(&child->mm->context.lock);
 122         }
 123
 124         return addr;
 125 }
 126
 127 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 128 {
 129         int i, copied;
 130         unsigned char opcode[15];
 131         unsigned long addr = convert_rip_to_linear(child, regs);
 132
 133         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 134         for (i = 0; i < copied; i++) {
 135                 switch (opcode[i]) {
 136                 /* popf and iret */
 137                 case 0x9d: case 0xcf:
 138                         return 1;
 139
 140                         /* CHECKME: 64 65 */
 141
 142                 /* opcode and address size prefixes */
 143                 case 0x66: case 0x67:
 144                         continue;
 145                 /* irrelevant prefixes (segment overrides and repeats) */
 146                 case 0x26: case 0x2e:
 147                 case 0x36: case 0x3e:
 148                 case 0x64: case 0x65:
 149                 case 0xf2: case 0xf3:
 150                         continue;
 151
 152                 case 0x40 ... 0x4f:
 153                         if (regs->cs != __USER_CS)
 154                                 /* 32-bit mode: register increment */
 155                                 return 0;
 156                         /* 64-bit mode: REX prefix */
 157                         continue;
 158
 159                         /* CHECKME: f2, f3 */
 160
 161                 /*
 162                  * pushf: NOTE! We should probably not let
 163                  * the user see the TF bit being set. But
 164                  * it's more pain than it's worth to avoid
 165                  * it, and a debugger could emulate this
 166                  * all in user space if it _really_ cares.
 167                  */
 168                 case 0x9c:
 169                 default:
 170                         return 0;
 171                 }
 172         }
 173         return 0;
 174 }
 175
 176 static void set_singlestep(struct task_struct *child)
 177 {
 178         struct pt_regs *regs = task_pt_regs(child);
 179
 180         /*
 181          * Always set TIF_SINGLESTEP - this guarantees that
 182          * we single-step system calls etc..  This will also
 183          * cause us to set TF when returning to user mode.
 184          */
 185         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 186
 187         /*
 188          * If TF was already set, don't do anything else
 189          */
 190         if (regs->eflags & TRAP_FLAG)
 191                 return;
 192
 193         /* Set TF on the kernel stack.. */
 194         regs->eflags |= TRAP_FLAG;
 195
 196         /*
 197          * ..but if TF is changed by the instruction we will trace,
 198          * don't mark it as being "us" that set it, so that we
 199          * won't clear it by hand later.
 200          */
 201         if (is_setting_trap_flag(child, regs))
 202                 return;
 203
 204         child->ptrace |= PT_DTRACE;
 205 }
 206
 207 static void clear_singlestep(struct task_struct *child)
 208 {
 209         /* Always clear TIF_SINGLESTEP... */
 210         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 211
 212         /* But touch TF only if it was set by us.. */
 213         if (child->ptrace & PT_DTRACE) {
 214                 struct pt_regs *regs = task_pt_regs(child);
 215                 regs->eflags &= ~TRAP_FLAG;
 216                 child->ptrace &= ~PT_DTRACE;
 217         }
 218 }
 219
 220 /*
 221  * Called by kernel/ptrace.c when detaching..
 222  *
 223  * Make sure the single step bit is not set.
 224  */
 225 void ptrace_disable(struct task_struct *child)
 226 {
 227         clear_singlestep(child);
 228 }
 229
 230 static int putreg(struct task_struct *child,
 231         unsigned long regno, unsigned long value)
 232 {
 233         unsigned long tmp;
 234
 235         switch (regno) {
 236                 case offsetof(struct user_regs_struct,fs):
 237                         if (value && (value & 3) != 3)
 238                                 return -EIO;
 239                         child->thread.fsindex = value & 0xffff;
 240                         return 0;
 241                 case offsetof(struct user_regs_struct,gs):
 242                         if (value && (value & 3) != 3)
 243                                 return -EIO;
 244                         child->thread.gsindex = value & 0xffff;
 245                         return 0;
 246                 case offsetof(struct user_regs_struct,ds):
 247                         if (value && (value & 3) != 3)
 248                                 return -EIO;
 249                         child->thread.ds = value & 0xffff;
 250                         return 0;
 251                 case offsetof(struct user_regs_struct,es):
 252                         if (value && (value & 3) != 3)
 253                                 return -EIO;
 254                         child->thread.es = value & 0xffff;
 255                         return 0;
 256                 case offsetof(struct user_regs_struct,ss):
 257                         if ((value & 3) != 3)
 258                                 return -EIO;
 259                         value &= 0xffff;
 260                         return 0;
 261                 case offsetof(struct user_regs_struct,fs_base):
 262                         if (value >= TASK_SIZE_OF(child))
 263                                 return -EIO;
 264                         /*
 265                          * When changing the segment base, use do_arch_prctl
 266                          * to set either thread.fs or thread.fsindex and the
 267                          * corresponding GDT slot.
 268                          */
 269                         if (child->thread.fs != value)
 270                                 return do_arch_prctl(child, ARCH_SET_FS, value);
 271                         return 0;
 272                 case offsetof(struct user_regs_struct,gs_base):
 273                         /*
 274                          * Exactly the same here as the %fs handling above.
 275                          */
 276                         if (value >= TASK_SIZE_OF(child))
 277                                 return -EIO;
 278                         if (child->thread.gs != value)
 279                                 return do_arch_prctl(child, ARCH_SET_GS, value);
 280                         return 0;
 281                 case offsetof(struct user_regs_struct, eflags):
 282                         value &= FLAG_MASK;
 283                         tmp = get_stack_long(child, EFL_OFFSET);
 284                         tmp &= ~FLAG_MASK;
 285                         value |= tmp;
 286                         break;
 287                 case offsetof(struct user_regs_struct,cs):
 288                         if ((value & 3) != 3)
 289                                 return -EIO;
 290                         value &= 0xffff;
 291                         break;
 292         }
 293         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 294         return 0;
 295 }
 296
 297 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 298 {
 299         unsigned long val;
 300         switch (regno) {
 301                 case offsetof(struct user_regs_struct, fs):
 302                         return child->thread.fsindex;
 303                 case offsetof(struct user_regs_struct, gs):
 304                         return child->thread.gsindex;
 305                 case offsetof(struct user_regs_struct, ds):
 306                         return child->thread.ds;
 307                 case offsetof(struct user_regs_struct, es):
 308                         return child->thread.es;
 309                 case offsetof(struct user_regs_struct, fs_base):
 310                         /*
 311                          * do_arch_prctl may have used a GDT slot instead of
 312                          * the MSR.  To userland, it appears the same either
 313                          * way, except the %fs segment selector might not be 0.
 314                          */
 315                         if (child->thread.fs != 0)
 316                                 return child->thread.fs;
 317                         if (child->thread.fsindex != FS_TLS_SEL)
 318                                 return 0;
 319                         return get_desc_base(&child->thread.tls_array[FS_TLS]);
 320                 case offsetof(struct user_regs_struct, gs_base):
 321                         /*
 322                          * Exactly the same here as the %fs handling above.
 323                          */
 324                         if (child->thread.gs != 0)
 325                                 return child->thread.gs;
 326                         if (child->thread.gsindex != GS_TLS_SEL)
 327                                 return 0;
 328                         return get_desc_base(&child->thread.tls_array[GS_TLS]);
 329                 default:
 330                         regno = regno - sizeof(struct pt_regs);
 331                         val = get_stack_long(child, regno);
 332                         if (test_tsk_thread_flag(child, TIF_IA32))
 333                                 val &= 0xffffffff;
 334                         return val;
 335         }
 336
 337 }
 338
 339 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 340 {
 341         long i, ret;
 342         unsigned ui;
 343
 344         switch (request) {
 345         /* when I and D space are separate, these will need to be fixed. */
 346         case PTRACE_PEEKTEXT: /* read word at location addr. */
 347         case PTRACE_PEEKDATA:
 348                 ret = generic_ptrace_peekdata(child, addr, data);
 349                 break;
 350
 351         /* read the word at location addr in the USER area. */
 352         case PTRACE_PEEKUSR: {
 353                 unsigned long tmp;
 354
 355                 ret = -EIO;
 356                 if ((addr & 7) ||
 357                     addr > sizeof(struct user) - 7)
 358                         break;
 359
 360                 switch (addr) {
 361                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 362                         tmp = getreg(child, addr);
 363                         break;
 364                 case offsetof(struct user, u_debugreg[0]):
 365                         tmp = child->thread.debugreg0;
 366                         break;
 367                 case offsetof(struct user, u_debugreg[1]):
 368                         tmp = child->thread.debugreg1;
 369                         break;
 370                 case offsetof(struct user, u_debugreg[2]):
 371                         tmp = child->thread.debugreg2;
 372                         break;
 373                 case offsetof(struct user, u_debugreg[3]):
 374                         tmp = child->thread.debugreg3;
 375                         break;
 376                 case offsetof(struct user, u_debugreg[6]):
 377                         tmp = child->thread.debugreg6;
 378                         break;
 379                 case offsetof(struct user, u_debugreg[7]):
 380                         tmp = child->thread.debugreg7;
 381                         break;
 382                 default:
 383                         tmp = 0;
 384                         break;
 385                 }
 386                 ret = put_user(tmp,(unsigned long __user *) data);
 387                 break;
 388         }
 389
 390         /* when I and D space are separate, this will have to be fixed. */
 391         case PTRACE_POKETEXT: /* write the word at location addr. */
 392         case PTRACE_POKEDATA:
 393                 ret = generic_ptrace_pokedata(child, addr, data);
 394                 break;
 395
 396         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 397         {
 398                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 399                 ret = -EIO;
 400                 if ((addr & 7) ||
 401                     addr > sizeof(struct user) - 7)
 402                         break;
 403
 404                 switch (addr) {
 405                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 406                         ret = putreg(child, addr, data);
 407                         break;
 408                 /* Disallows to set a breakpoint into the vsyscall */
 409                 case offsetof(struct user, u_debugreg[0]):
 410                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 411                         child->thread.debugreg0 = data;
 412                         ret = 0;
 413                         break;
 414                 case offsetof(struct user, u_debugreg[1]):
 415                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 416                         child->thread.debugreg1 = data;
 417                         ret = 0;
 418                         break;
 419                 case offsetof(struct user, u_debugreg[2]):
 420                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 421                         child->thread.debugreg2 = data;
 422                         ret = 0;
 423                         break;
 424                 case offsetof(struct user, u_debugreg[3]):
 425                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 426                         child->thread.debugreg3 = data;
 427                         ret = 0;
 428                         break;
 429                 case offsetof(struct user, u_debugreg[6]):
 430                                   if (data >> 32)
 431                                 break;
 432                         child->thread.debugreg6 = data;
 433                         ret = 0;
 434                         break;
 435                 case offsetof(struct user, u_debugreg[7]):
 436                         /* See arch/i386/kernel/ptrace.c for an explanation of
 437                          * this awkward check.*/
 438                         data &= ~DR_CONTROL_RESERVED;
 439                         for(i=0; i<4; i++)
 440                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 441                                         break;
 442                         if (i == 4) {
 443                           child->thread.debugreg7 = data;
 444                           if (data)
 445                                 set_tsk_thread_flag(child, TIF_DEBUG);
 446                           else
 447                                 clear_tsk_thread_flag(child, TIF_DEBUG);
 448                           ret = 0;
 449                         }
 450                   break;
 451                 }
 452                 break;
 453         }
 454         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 455         case PTRACE_CONT:    /* restart after signal. */
 456
 457                 ret = -EIO;
 458                 if (!valid_signal(data))
 459                         break;
 460                 if (request == PTRACE_SYSCALL)
 461                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 462                 else
 463                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 464                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 465                 child->exit_code = data;
 466                 /* make sure the single step bit is not set. */
 467                 clear_singlestep(child);
 468                 wake_up_process(child);
 469                 ret = 0;
 470                 break;
 471
 472 #ifdef CONFIG_IA32_EMULATION
 473                 /* This makes only sense with 32bit programs. Allow a
 474                    64bit debugger to fully examine them too. Better
 475                    don't use it against 64bit processes, use
 476                    PTRACE_ARCH_PRCTL instead. */
 477         case PTRACE_GET_THREAD_AREA:
 478                 if (addr < 0)
 479                         return -EIO;
 480                 ret = do_get_thread_area(child, addr,
 481                                          (struct user_desc __user *) data);
 482
 483                 break;
 484         case PTRACE_SET_THREAD_AREA:
 485                 if (addr < 0)
 486                         return -EIO;
 487                 ret = do_set_thread_area(child, addr,
 488                                          (struct user_desc __user *) data, 0);
 489                 break;
 490 #endif
 491                 /* normal 64bit interface to access TLS data.
 492                    Works just like arch_prctl, except that the arguments
 493                    are reversed. */
 494         case PTRACE_ARCH_PRCTL:
 495                 ret = do_arch_prctl(child, data, addr);
 496                 break;
 497
 498 /*
 499  * make the child exit.  Best I can do is send it a sigkill.
 500  * perhaps it should be put in the status that it wants to
 501  * exit.
 502  */
 503         case PTRACE_KILL:
 504                 ret = 0;
 505                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 506                         break;
 507                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 508                 child->exit_code = SIGKILL;
 509                 /* make sure the single step bit is not set. */
 510                 clear_singlestep(child);
 511                 wake_up_process(child);
 512                 break;
 513
 514         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 515                 ret = -EIO;
 516                 if (!valid_signal(data))
 517                         break;
 518                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 519                 set_singlestep(child);
 520                 child->exit_code = data;
 521                 /* give it a chance to run. */
 522                 wake_up_process(child);
 523                 ret = 0;
 524                 break;
 525
 526         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 527                 if (!access_ok(VERIFY_WRITE, (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 |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 535                         data += sizeof(long);
 536                 }
 537                 break;
 538         }
 539
 540         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 541                 unsigned long tmp;
 542                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 543                                sizeof(struct user_regs_struct))) {
 544                         ret = -EIO;
 545                         break;
 546                 }
 547                 ret = 0;
 548                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 549                         ret = __get_user(tmp, (unsigned long __user *) data);
 550                         if (ret)
 551                                 break;
 552                         ret = putreg(child, ui, tmp);
 553                         if (ret)
 554                                 break;
 555                         data += sizeof(long);
 556                 }
 557                 break;
 558         }
 559
 560         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 561                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 562                                sizeof(struct user_i387_struct))) {
 563                         ret = -EIO;
 564                         break;
 565                 }
 566                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 567                 break;
 568         }
 569
 570         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 571                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 572                                sizeof(struct user_i387_struct))) {
 573                         ret = -EIO;
 574                         break;
 575                 }
 576                 set_stopped_child_used_math(child);
 577                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 578                 break;
 579         }
 580
 581         default:
 582                 ret = ptrace_request(child, request, addr, data);
 583                 break;
 584         }
 585         return ret;
 586 }
 587
 588 static void syscall_trace(struct pt_regs *regs)
 589 {
 590
 591 #if 0
 592         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 593                current->comm,
 594                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 595                current_thread_info()->flags, current->ptrace);
 596 #endif
 597
 598         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 599                                 ? 0x80 : 0));
 600         /*
 601          * this isn't the same as continuing with a signal, but it will do
 602          * for normal use.  strace only continues with a signal if the
 603          * stopping signal is not SIGTRAP.  -brl
 604          */
 605         if (current->exit_code) {
 606                 send_sig(current->exit_code, current, 1);
 607                 current->exit_code = 0;
 608         }
 609 }
 610
 611 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 612 {
 613         /* do the secure computing check first */
 614         secure_computing(regs->orig_rax);
 615
 616         if (test_thread_flag(TIF_SYSCALL_TRACE)
 617             && (current->ptrace & PT_PTRACED))
 618                 syscall_trace(regs);
 619
 620         if (unlikely(current->audit_context)) {
 621                 if (test_thread_flag(TIF_IA32)) {
 622                         audit_syscall_entry(AUDIT_ARCH_I386,
 623                                             regs->orig_rax,
 624                                             regs->rbx, regs->rcx,
 625                                             regs->rdx, regs->rsi);
 626                 } else {
 627                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 628                                             regs->orig_rax,
 629                                             regs->rdi, regs->rsi,
 630                                             regs->rdx, regs->r10);
 631                 }
 632         }
 633 }
 634
 635 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 636 {
 637         if (unlikely(current->audit_context))
 638                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 639
 640         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 641              || test_thread_flag(TIF_SINGLESTEP))
 642             && (current->ptrace & PT_PTRACED))
 643                 syscall_trace(regs);
 644 }