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 /*
  46  * eflags and offset of eflags on child stack..
  47  */
  48 #define EFLAGS offsetof(struct pt_regs, eflags)
  49 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  50
  51 /*
  52  * this routine will get a word off of the processes privileged stack.
  53  * the offset is how far from the base addr as stored in the TSS.
  54  * this routine assumes that all the privileged stacks are in our
  55  * data space.
  56  */
  57 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  58 {
  59         unsigned char *stack;
  60
  61         stack = (unsigned char *)task->thread.rsp0;
  62         stack += offset;
  63         return (*((unsigned long *)stack));
  64 }
  65
  66 /*
  67  * this routine will put a word on the processes privileged stack.
  68  * the offset is how far from the base addr as stored in the TSS.
  69  * this routine assumes that all the privileged stacks are in our
  70  * data space.
  71  */
  72 static inline long put_stack_long(struct task_struct *task, int offset,
  73         unsigned long data)
  74 {
  75         unsigned char * stack;
  76
  77         stack = (unsigned char *) task->thread.rsp0;
  78         stack += offset;
  79         *(unsigned long *) stack = data;
  80         return 0;
  81 }
  82
  83 #define LDT_SEGMENT 4
  84
  85 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  86 {
  87         unsigned long addr, seg;
  88
  89         addr = regs->rip;
  90         seg = regs->cs & 0xffff;
  91
  92         /*
  93          * We'll assume that the code segments in the GDT
  94          * are all zero-based. That is largely true: the
  95          * TLS segments are used for data, and the PNPBIOS
  96          * and APM bios ones we just ignore here.
  97          */
  98         if (seg & LDT_SEGMENT) {
  99                 u32 *desc;
 100                 unsigned long base;
 101
 102                 seg &= ~7UL;
 103
 104                 mutex_lock(&child->mm->context.lock);
 105                 if (unlikely((seg >> 3) >= child->mm->context.size))
 106                         addr = -1L; /* bogus selector, access would fault */
 107                 else {
 108                         desc = child->mm->context.ldt + seg;
 109                         base = ((desc[0] >> 16) |
 110                                 ((desc[1] & 0xff) << 16) |
 111                                 (desc[1] & 0xff000000));
 112
 113                         /* 16-bit code segment? */
 114                         if (!((desc[1] >> 22) & 1))
 115                                 addr &= 0xffff;
 116                         addr += base;
 117                 }
 118                 mutex_unlock(&child->mm->context.lock);
 119         }
 120
 121         return addr;
 122 }
 123
 124 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 125 {
 126         int i, copied;
 127         unsigned char opcode[15];
 128         unsigned long addr = convert_rip_to_linear(child, regs);
 129
 130         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 131         for (i = 0; i < copied; i++) {
 132                 switch (opcode[i]) {
 133                 /* popf and iret */
 134                 case 0x9d: case 0xcf:
 135                         return 1;
 136
 137                         /* CHECKME: 64 65 */
 138
 139                 /* opcode and address size prefixes */
 140                 case 0x66: case 0x67:
 141                         continue;
 142                 /* irrelevant prefixes (segment overrides and repeats) */
 143                 case 0x26: case 0x2e:
 144                 case 0x36: case 0x3e:
 145                 case 0x64: case 0x65:
 146                 case 0xf2: case 0xf3:
 147                         continue;
 148
 149                 case 0x40 ... 0x4f:
 150                         if (regs->cs != __USER_CS)
 151                                 /* 32-bit mode: register increment */
 152                                 return 0;
 153                         /* 64-bit mode: REX prefix */
 154                         continue;
 155
 156                         /* CHECKME: f2, f3 */
 157
 158                 /*
 159                  * pushf: NOTE! We should probably not let
 160                  * the user see the TF bit being set. But
 161                  * it's more pain than it's worth to avoid
 162                  * it, and a debugger could emulate this
 163                  * all in user space if it _really_ cares.
 164                  */
 165                 case 0x9c:
 166                 default:
 167                         return 0;
 168                 }
 169         }
 170         return 0;
 171 }
 172
 173 static void set_singlestep(struct task_struct *child)
 174 {
 175         struct pt_regs *regs = task_pt_regs(child);
 176
 177         /*
 178          * Always set TIF_SINGLESTEP - this guarantees that
 179          * we single-step system calls etc..  This will also
 180          * cause us to set TF when returning to user mode.
 181          */
 182         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 183
 184         /*
 185          * If TF was already set, don't do anything else
 186          */
 187         if (regs->eflags & X86_EFLAGS_TF)
 188                 return;
 189
 190         /* Set TF on the kernel stack.. */
 191         regs->eflags |= X86_EFLAGS_TF;
 192
 193         /*
 194          * ..but if TF is changed by the instruction we will trace,
 195          * don't mark it as being "us" that set it, so that we
 196          * won't clear it by hand later.
 197          */
 198         if (is_setting_trap_flag(child, regs))
 199                 return;
 200
 201         child->ptrace |= PT_DTRACE;
 202 }
 203
 204 static void clear_singlestep(struct task_struct *child)
 205 {
 206         /* Always clear TIF_SINGLESTEP... */
 207         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 208
 209         /* But touch TF only if it was set by us.. */
 210         if (child->ptrace & PT_DTRACE) {
 211                 struct pt_regs *regs = task_pt_regs(child);
 212                 regs->eflags &= ~X86_EFLAGS_TF;
 213                 child->ptrace &= ~PT_DTRACE;
 214         }
 215 }
 216
 217 /*
 218  * Called by kernel/ptrace.c when detaching..
 219  *
 220  * Make sure the single step bit is not set.
 221  */
 222 void ptrace_disable(struct task_struct *child)
 223 {
 224         clear_singlestep(child);
 225 }
 226
 227 static int putreg(struct task_struct *child,
 228         unsigned long regno, unsigned long value)
 229 {
 230         unsigned long tmp;
 231
 232         switch (regno) {
 233                 case offsetof(struct user_regs_struct,fs):
 234                         if (value && (value & 3) != 3)
 235                                 return -EIO;
 236                         child->thread.fsindex = value & 0xffff;
 237                         return 0;
 238                 case offsetof(struct user_regs_struct,gs):
 239                         if (value && (value & 3) != 3)
 240                                 return -EIO;
 241                         child->thread.gsindex = value & 0xffff;
 242                         return 0;
 243                 case offsetof(struct user_regs_struct,ds):
 244                         if (value && (value & 3) != 3)
 245                                 return -EIO;
 246                         child->thread.ds = value & 0xffff;
 247                         return 0;
 248                 case offsetof(struct user_regs_struct,es):
 249                         if (value && (value & 3) != 3)
 250                                 return -EIO;
 251                         child->thread.es = value & 0xffff;
 252                         return 0;
 253                 case offsetof(struct user_regs_struct,ss):
 254                         if ((value & 3) != 3)
 255                                 return -EIO;
 256                         value &= 0xffff;
 257                         return 0;
 258                 case offsetof(struct user_regs_struct,fs_base):
 259                         if (value >= TASK_SIZE_OF(child))
 260                                 return -EIO;
 261                         /*
 262                          * When changing the segment base, use do_arch_prctl
 263                          * to set either thread.fs or thread.fsindex and the
 264                          * corresponding GDT slot.
 265                          */
 266                         if (child->thread.fs != value)
 267                                 return do_arch_prctl(child, ARCH_SET_FS, value);
 268                         return 0;
 269                 case offsetof(struct user_regs_struct,gs_base):
 270                         /*
 271                          * Exactly the same here as the %fs handling above.
 272                          */
 273                         if (value >= TASK_SIZE_OF(child))
 274                                 return -EIO;
 275                         if (child->thread.gs != value)
 276                                 return do_arch_prctl(child, ARCH_SET_GS, value);
 277                         return 0;
 278                 case offsetof(struct user_regs_struct, eflags):
 279                         value &= FLAG_MASK;
 280                         tmp = get_stack_long(child, EFL_OFFSET);
 281                         tmp &= ~FLAG_MASK;
 282                         value |= tmp;
 283                         break;
 284                 case offsetof(struct user_regs_struct,cs):
 285                         if ((value & 3) != 3)
 286                                 return -EIO;
 287                         value &= 0xffff;
 288                         break;
 289         }
 290         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 291         return 0;
 292 }
 293
 294 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 295 {
 296         unsigned long val;
 297         switch (regno) {
 298                 case offsetof(struct user_regs_struct, fs):
 299                         return child->thread.fsindex;
 300                 case offsetof(struct user_regs_struct, gs):
 301                         return child->thread.gsindex;
 302                 case offsetof(struct user_regs_struct, ds):
 303                         return child->thread.ds;
 304                 case offsetof(struct user_regs_struct, es):
 305                         return child->thread.es;
 306                 case offsetof(struct user_regs_struct, fs_base):
 307                         /*
 308                          * do_arch_prctl may have used a GDT slot instead of
 309                          * the MSR.  To userland, it appears the same either
 310                          * way, except the %fs segment selector might not be 0.
 311                          */
 312                         if (child->thread.fs != 0)
 313                                 return child->thread.fs;
 314                         if (child->thread.fsindex != FS_TLS_SEL)
 315                                 return 0;
 316                         return get_desc_base(&child->thread.tls_array[FS_TLS]);
 317                 case offsetof(struct user_regs_struct, gs_base):
 318                         /*
 319                          * Exactly the same here as the %fs handling above.
 320                          */
 321                         if (child->thread.gs != 0)
 322                                 return child->thread.gs;
 323                         if (child->thread.gsindex != GS_TLS_SEL)
 324                                 return 0;
 325                         return get_desc_base(&child->thread.tls_array[GS_TLS]);
 326                 default:
 327                         regno = regno - sizeof(struct pt_regs);
 328                         val = get_stack_long(child, regno);
 329                         if (test_tsk_thread_flag(child, TIF_IA32))
 330                                 val &= 0xffffffff;
 331                         return val;
 332         }
 333
 334 }
 335
 336 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 337 {
 338         long i, ret;
 339         unsigned ui;
 340
 341         switch (request) {
 342         /* when I and D space are separate, these will need to be fixed. */
 343         case PTRACE_PEEKTEXT: /* read word at location addr. */
 344         case PTRACE_PEEKDATA:
 345                 ret = generic_ptrace_peekdata(child, addr, data);
 346                 break;
 347
 348         /* read the word at location addr in the USER area. */
 349         case PTRACE_PEEKUSR: {
 350                 unsigned long tmp;
 351
 352                 ret = -EIO;
 353                 if ((addr & 7) ||
 354                     addr > sizeof(struct user) - 7)
 355                         break;
 356
 357                 switch (addr) {
 358                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 359                         tmp = getreg(child, addr);
 360                         break;
 361                 case offsetof(struct user, u_debugreg[0]):
 362                         tmp = child->thread.debugreg0;
 363                         break;
 364                 case offsetof(struct user, u_debugreg[1]):
 365                         tmp = child->thread.debugreg1;
 366                         break;
 367                 case offsetof(struct user, u_debugreg[2]):
 368                         tmp = child->thread.debugreg2;
 369                         break;
 370                 case offsetof(struct user, u_debugreg[3]):
 371                         tmp = child->thread.debugreg3;
 372                         break;
 373                 case offsetof(struct user, u_debugreg[6]):
 374                         tmp = child->thread.debugreg6;
 375                         break;
 376                 case offsetof(struct user, u_debugreg[7]):
 377                         tmp = child->thread.debugreg7;
 378                         break;
 379                 default:
 380                         tmp = 0;
 381                         break;
 382                 }
 383                 ret = put_user(tmp,(unsigned long __user *) data);
 384                 break;
 385         }
 386
 387         /* when I and D space are separate, this will have to be fixed. */
 388         case PTRACE_POKETEXT: /* write the word at location addr. */
 389         case PTRACE_POKEDATA:
 390                 ret = generic_ptrace_pokedata(child, addr, data);
 391                 break;
 392
 393         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 394         {
 395                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 396                 ret = -EIO;
 397                 if ((addr & 7) ||
 398                     addr > sizeof(struct user) - 7)
 399                         break;
 400
 401                 switch (addr) {
 402                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 403                         ret = putreg(child, addr, data);
 404                         break;
 405                 /* Disallows to set a breakpoint into the vsyscall */
 406                 case offsetof(struct user, u_debugreg[0]):
 407                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 408                         child->thread.debugreg0 = data;
 409                         ret = 0;
 410                         break;
 411                 case offsetof(struct user, u_debugreg[1]):
 412                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 413                         child->thread.debugreg1 = data;
 414                         ret = 0;
 415                         break;
 416                 case offsetof(struct user, u_debugreg[2]):
 417                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 418                         child->thread.debugreg2 = data;
 419                         ret = 0;
 420                         break;
 421                 case offsetof(struct user, u_debugreg[3]):
 422                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 423                         child->thread.debugreg3 = data;
 424                         ret = 0;
 425                         break;
 426                 case offsetof(struct user, u_debugreg[6]):
 427                                   if (data >> 32)
 428                                 break;
 429                         child->thread.debugreg6 = data;
 430                         ret = 0;
 431                         break;
 432                 case offsetof(struct user, u_debugreg[7]):
 433                         /* See arch/i386/kernel/ptrace.c for an explanation of
 434                          * this awkward check.*/
 435                         data &= ~DR_CONTROL_RESERVED;
 436                         for(i=0; i<4; i++)
 437                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 438                                         break;
 439                         if (i == 4) {
 440                           child->thread.debugreg7 = data;
 441                           if (data)
 442                                 set_tsk_thread_flag(child, TIF_DEBUG);
 443                           else
 444                                 clear_tsk_thread_flag(child, TIF_DEBUG);
 445                           ret = 0;
 446                         }
 447                   break;
 448                 }
 449                 break;
 450         }
 451         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 452         case PTRACE_CONT:    /* restart after signal. */
 453
 454                 ret = -EIO;
 455                 if (!valid_signal(data))
 456                         break;
 457                 if (request == PTRACE_SYSCALL)
 458                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 459                 else
 460                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 461                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 462                 child->exit_code = data;
 463                 /* make sure the single step bit is not set. */
 464                 clear_singlestep(child);
 465                 wake_up_process(child);
 466                 ret = 0;
 467                 break;
 468
 469 #ifdef CONFIG_IA32_EMULATION
 470                 /* This makes only sense with 32bit programs. Allow a
 471                    64bit debugger to fully examine them too. Better
 472                    don't use it against 64bit processes, use
 473                    PTRACE_ARCH_PRCTL instead. */
 474         case PTRACE_GET_THREAD_AREA:
 475                 if (addr < 0)
 476                         return -EIO;
 477                 ret = do_get_thread_area(child, addr,
 478                                          (struct user_desc __user *) data);
 479
 480                 break;
 481         case PTRACE_SET_THREAD_AREA:
 482                 if (addr < 0)
 483                         return -EIO;
 484                 ret = do_set_thread_area(child, addr,
 485                                          (struct user_desc __user *) data, 0);
 486                 break;
 487 #endif
 488                 /* normal 64bit interface to access TLS data.
 489                    Works just like arch_prctl, except that the arguments
 490                    are reversed. */
 491         case PTRACE_ARCH_PRCTL:
 492                 ret = do_arch_prctl(child, data, addr);
 493                 break;
 494
 495 /*
 496  * make the child exit.  Best I can do is send it a sigkill.
 497  * perhaps it should be put in the status that it wants to
 498  * exit.
 499  */
 500         case PTRACE_KILL:
 501                 ret = 0;
 502                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 503                         break;
 504                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 505                 child->exit_code = SIGKILL;
 506                 /* make sure the single step bit is not set. */
 507                 clear_singlestep(child);
 508                 wake_up_process(child);
 509                 break;
 510
 511         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 512                 ret = -EIO;
 513                 if (!valid_signal(data))
 514                         break;
 515                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 516                 set_singlestep(child);
 517                 child->exit_code = data;
 518                 /* give it a chance to run. */
 519                 wake_up_process(child);
 520                 ret = 0;
 521                 break;
 522
 523         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 524                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 525                                sizeof(struct user_regs_struct))) {
 526                         ret = -EIO;
 527                         break;
 528                 }
 529                 ret = 0;
 530                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 531                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 532                         data += sizeof(long);
 533                 }
 534                 break;
 535         }
 536
 537         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 538                 unsigned long tmp;
 539                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 540                                sizeof(struct user_regs_struct))) {
 541                         ret = -EIO;
 542                         break;
 543                 }
 544                 ret = 0;
 545                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 546                         ret = __get_user(tmp, (unsigned long __user *) data);
 547                         if (ret)
 548                                 break;
 549                         ret = putreg(child, ui, tmp);
 550                         if (ret)
 551                                 break;
 552                         data += sizeof(long);
 553                 }
 554                 break;
 555         }
 556
 557         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 558                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 559                                sizeof(struct user_i387_struct))) {
 560                         ret = -EIO;
 561                         break;
 562                 }
 563                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 564                 break;
 565         }
 566
 567         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 568                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 569                                sizeof(struct user_i387_struct))) {
 570                         ret = -EIO;
 571                         break;
 572                 }
 573                 set_stopped_child_used_math(child);
 574                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 575                 break;
 576         }
 577
 578         default:
 579                 ret = ptrace_request(child, request, addr, data);
 580                 break;
 581         }
 582         return ret;
 583 }
 584
 585 static void syscall_trace(struct pt_regs *regs)
 586 {
 587
 588 #if 0
 589         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 590                current->comm,
 591                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 592                current_thread_info()->flags, current->ptrace);
 593 #endif
 594
 595         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 596                                 ? 0x80 : 0));
 597         /*
 598          * this isn't the same as continuing with a signal, but it will do
 599          * for normal use.  strace only continues with a signal if the
 600          * stopping signal is not SIGTRAP.  -brl
 601          */
 602         if (current->exit_code) {
 603                 send_sig(current->exit_code, current, 1);
 604                 current->exit_code = 0;
 605         }
 606 }
 607
 608 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 609 {
 610         /* do the secure computing check first */
 611         secure_computing(regs->orig_rax);
 612
 613         if (test_thread_flag(TIF_SYSCALL_TRACE)
 614             && (current->ptrace & PT_PTRACED))
 615                 syscall_trace(regs);
 616
 617         if (unlikely(current->audit_context)) {
 618                 if (test_thread_flag(TIF_IA32)) {
 619                         audit_syscall_entry(AUDIT_ARCH_I386,
 620                                             regs->orig_rax,
 621                                             regs->rbx, regs->rcx,
 622                                             regs->rdx, regs->rsi);
 623                 } else {
 624                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 625                                             regs->orig_rax,
 626                                             regs->rdi, regs->rsi,
 627                                             regs->rdx, regs->r10);
 628                 }
 629         }
 630 }
 631
 632 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 633 {
 634         if (unlikely(current->audit_context))
 635                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 636
 637         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 638              || test_thread_flag(TIF_SINGLESTEP))
 639             && (current->ptrace & PT_PTRACED))
 640                 syscall_trace(regs);
 641 }