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 /* determines which flags the user has access to. */
  40 /* 1 = access 0 = no access */
  41 #define FLAG_MASK 0x44dd5UL
  42
  43 /* set's the trap flag. */
  44 #define TRAP_FLAG 0x100UL
  45
  46 /*
  47  * eflags and offset of eflags on child stack..
  48  */
  49 #define EFLAGS offsetof(struct pt_regs, eflags)
  50 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  51
  52 /*
  53  * this routine will get a word off of the processes privileged stack.
  54  * the offset is how far from the base addr as stored in the TSS.
  55  * this routine assumes that all the privileged stacks are in our
  56  * data space.
  57  */
  58 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  59 {
  60         unsigned char *stack;
  61
  62         stack = (unsigned char *)task->thread.rsp0;
  63         stack += offset;
  64         return (*((unsigned long *)stack));
  65 }
  66
  67 static inline struct pt_regs *get_child_regs(struct task_struct *task)
  68 {
  69         struct pt_regs *regs = (void *)task->thread.rsp0;
  70         return regs - 1;
  71 }
  72
  73 /*
  74  * this routine will put a word on the processes privileged stack.
  75  * the offset is how far from the base addr as stored in the TSS.
  76  * this routine assumes that all the privileged stacks are in our
  77  * data space.
  78  */
  79 static inline long put_stack_long(struct task_struct *task, int offset,
  80         unsigned long data)
  81 {
  82         unsigned char * stack;
  83
  84         stack = (unsigned char *) task->thread.rsp0;
  85         stack += offset;
  86         *(unsigned long *) stack = data;
  87         return 0;
  88 }
  89
  90 #define LDT_SEGMENT 4
  91
  92 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  93 {
  94         unsigned long addr, seg;
  95
  96         addr = regs->rip;
  97         seg = regs->cs & 0xffff;
  98
  99         /*
 100          * We'll assume that the code segments in the GDT
 101          * are all zero-based. That is largely true: the
 102          * TLS segments are used for data, and the PNPBIOS
 103          * and APM bios ones we just ignore here.
 104          */
 105         if (seg & LDT_SEGMENT) {
 106                 u32 *desc;
 107                 unsigned long base;
 108
 109                 down(&child->mm->context.sem);
 110                 desc = child->mm->context.ldt + (seg & ~7);
 111                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 112
 113                 /* 16-bit code segment? */
 114                 if (!((desc[1] >> 22) & 1))
 115                         addr &= 0xffff;
 116                 addr += base;
 117                 up(&child->mm->context.sem);
 118         }
 119         return addr;
 120 }
 121
 122 static int is_at_popf(struct task_struct *child, struct pt_regs *regs)
 123 {
 124         int i, copied;
 125         unsigned char opcode[16];
 126         unsigned long addr = convert_rip_to_linear(child, regs);
 127
 128         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 129         for (i = 0; i < copied; i++) {
 130                 switch (opcode[i]) {
 131                 /* popf */
 132                 case 0x9d:
 133                         return 1;
 134
 135                         /* CHECKME: 64 65 */
 136
 137                 /* opcode and address size prefixes */
 138                 case 0x66: case 0x67:
 139                         continue;
 140                 /* irrelevant prefixes (segment overrides and repeats) */
 141                 case 0x26: case 0x2e:
 142                 case 0x36: case 0x3e:
 143                 case 0x64: case 0x65:
 144                 case 0xf0: case 0xf2: case 0xf3:
 145                         continue;
 146
 147                 /* REX prefixes */
 148                 case 0x40 ... 0x4f:
 149                         continue;
 150
 151                         /* CHECKME: f0, f2, f3 */
 152
 153                 /*
 154                  * pushf: NOTE! We should probably not let
 155                  * the user see the TF bit being set. But
 156                  * it's more pain than it's worth to avoid
 157                  * it, and a debugger could emulate this
 158                  * all in user space if it _really_ cares.
 159                  */
 160                 case 0x9c:
 161                 default:
 162                         return 0;
 163                 }
 164         }
 165         return 0;
 166 }
 167
 168 static void set_singlestep(struct task_struct *child)
 169 {
 170         struct pt_regs *regs = get_child_regs(child);
 171
 172         /*
 173          * Always set TIF_SINGLESTEP - this guarantees that
 174          * we single-step system calls etc..  This will also
 175          * cause us to set TF when returning to user mode.
 176          */
 177         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 178
 179         /*
 180          * If TF was already set, don't do anything else
 181          */
 182         if (regs->eflags & TRAP_FLAG)
 183                 return;
 184
 185         /* Set TF on the kernel stack.. */
 186         regs->eflags |= TRAP_FLAG;
 187
 188         /*
 189          * ..but if TF is changed by the instruction we will trace,
 190          * don't mark it as being "us" that set it, so that we
 191          * won't clear it by hand later.
 192          *
 193          * AK: this is not enough, LAHF and IRET can change TF in user space too.
 194          */
 195         if (is_at_popf(child, regs))
 196                 return;
 197
 198         child->ptrace |= PT_DTRACE;
 199 }
 200
 201 static void clear_singlestep(struct task_struct *child)
 202 {
 203         /* Always clear TIF_SINGLESTEP... */
 204         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 205
 206         /* But touch TF only if it was set by us.. */
 207         if (child->ptrace & PT_DTRACE) {
 208                 struct pt_regs *regs = get_child_regs(child);
 209                 regs->eflags &= ~TRAP_FLAG;
 210                 child->ptrace &= ~PT_DTRACE;
 211         }
 212 }
 213
 214 /*
 215  * Called by kernel/ptrace.c when detaching..
 216  *
 217  * Make sure the single step bit is not set.
 218  */
 219 void ptrace_disable(struct task_struct *child)
 220 {
 221         clear_singlestep(child);
 222 }
 223
 224 static int putreg(struct task_struct *child,
 225         unsigned long regno, unsigned long value)
 226 {
 227         unsigned long tmp;
 228
 229         /* Some code in the 64bit emulation may not be 64bit clean.
 230            Don't take any chances. */
 231         if (test_tsk_thread_flag(child, TIF_IA32))
 232                 value &= 0xffffffff;
 233         switch (regno) {
 234                 case offsetof(struct user_regs_struct,fs):
 235                         if (value && (value & 3) != 3)
 236                                 return -EIO;
 237                         child->thread.fsindex = value & 0xffff;
 238                         return 0;
 239                 case offsetof(struct user_regs_struct,gs):
 240                         if (value && (value & 3) != 3)
 241                                 return -EIO;
 242                         child->thread.gsindex = value & 0xffff;
 243                         return 0;
 244                 case offsetof(struct user_regs_struct,ds):
 245                         if (value && (value & 3) != 3)
 246                                 return -EIO;
 247                         child->thread.ds = value & 0xffff;
 248                         return 0;
 249                 case offsetof(struct user_regs_struct,es):
 250                         if (value && (value & 3) != 3)
 251                                 return -EIO;
 252                         child->thread.es = value & 0xffff;
 253                         return 0;
 254                 case offsetof(struct user_regs_struct,ss):
 255                         if ((value & 3) != 3)
 256                                 return -EIO;
 257                         value &= 0xffff;
 258                         return 0;
 259                 case offsetof(struct user_regs_struct,fs_base):
 260                         if (value >= TASK_SIZE)
 261                                 return -EIO;
 262                         child->thread.fs = value;
 263                         return 0;
 264                 case offsetof(struct user_regs_struct,gs_base):
 265                         if (value >= TASK_SIZE)
 266                                 return -EIO;
 267                         child->thread.gs = value;
 268                         return 0;
 269                 case offsetof(struct user_regs_struct, eflags):
 270                         value &= FLAG_MASK;
 271                         tmp = get_stack_long(child, EFL_OFFSET);
 272                         tmp &= ~FLAG_MASK;
 273                         value |= tmp;
 274                         break;
 275                 case offsetof(struct user_regs_struct,cs):
 276                         if ((value & 3) != 3)
 277                                 return -EIO;
 278                         value &= 0xffff;
 279                         break;
 280                 case offsetof(struct user_regs_struct, rip):
 281                         /* Check if the new RIP address is canonical */
 282                         if (value >= TASK_SIZE)
 283                                 return -EIO;
 284                         break;
 285         }
 286         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 287         return 0;
 288 }
 289
 290 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 291 {
 292         unsigned long val;
 293         switch (regno) {
 294                 case offsetof(struct user_regs_struct, fs):
 295                         return child->thread.fsindex;
 296                 case offsetof(struct user_regs_struct, gs):
 297                         return child->thread.gsindex;
 298                 case offsetof(struct user_regs_struct, ds):
 299                         return child->thread.ds;
 300                 case offsetof(struct user_regs_struct, es):
 301                         return child->thread.es;
 302                 case offsetof(struct user_regs_struct, fs_base):
 303                         return child->thread.fs;
 304                 case offsetof(struct user_regs_struct, gs_base):
 305                         return child->thread.gs;
 306                 default:
 307                         regno = regno - sizeof(struct pt_regs);
 308                         val = get_stack_long(child, regno);
 309                         if (test_tsk_thread_flag(child, TIF_IA32))
 310                                 val &= 0xffffffff;
 311                         return val;
 312         }
 313
 314 }
 315
 316 asmlinkage long sys_ptrace(long request, long pid, unsigned long addr, long data)
 317 {
 318         struct task_struct *child;
 319         long i, ret;
 320         unsigned ui;
 321
 322         /* This lock_kernel fixes a subtle race with suid exec */
 323         lock_kernel();
 324         ret = -EPERM;
 325         if (request == PTRACE_TRACEME) {
 326                 /* are we already being traced? */
 327                 if (current->ptrace & PT_PTRACED)
 328                         goto out;
 329                 ret = security_ptrace(current->parent, current);
 330                 if (ret)
 331                         goto out;
 332                 /* set the ptrace bit in the process flags. */
 333                 current->ptrace |= PT_PTRACED;
 334                 ret = 0;
 335                 goto out;
 336         }
 337         ret = -ESRCH;
 338         read_lock(&tasklist_lock);
 339         child = find_task_by_pid(pid);
 340         if (child)
 341                 get_task_struct(child);
 342         read_unlock(&tasklist_lock);
 343         if (!child)
 344                 goto out;
 345
 346         ret = -EPERM;
 347         if (pid == 1)           /* you may not mess with init */
 348                 goto out_tsk;
 349
 350         if (request == PTRACE_ATTACH) {
 351                 ret = ptrace_attach(child);
 352                 goto out_tsk;
 353         }
 354         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 355         if (ret < 0)
 356                 goto out_tsk;
 357
 358         switch (request) {
 359         /* when I and D space are separate, these will need to be fixed. */
 360         case PTRACE_PEEKTEXT: /* read word at location addr. */
 361         case PTRACE_PEEKDATA: {
 362                 unsigned long tmp;
 363                 int copied;
 364
 365                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 366                 ret = -EIO;
 367                 if (copied != sizeof(tmp))
 368                         break;
 369                 ret = put_user(tmp,(unsigned long __user *) data);
 370                 break;
 371         }
 372
 373         /* read the word at location addr in the USER area. */
 374         case PTRACE_PEEKUSR: {
 375                 unsigned long tmp;
 376
 377                 ret = -EIO;
 378                 if ((addr & 7) ||
 379                     addr > sizeof(struct user) - 7)
 380                         break;
 381
 382                 switch (addr) {
 383                 case 0 ... sizeof(struct user_regs_struct):
 384                         tmp = getreg(child, addr);
 385                         break;
 386                 case offsetof(struct user, u_debugreg[0]):
 387                         tmp = child->thread.debugreg0;
 388                         break;
 389                 case offsetof(struct user, u_debugreg[1]):
 390                         tmp = child->thread.debugreg1;
 391                         break;
 392                 case offsetof(struct user, u_debugreg[2]):
 393                         tmp = child->thread.debugreg2;
 394                         break;
 395                 case offsetof(struct user, u_debugreg[3]):
 396                         tmp = child->thread.debugreg3;
 397                         break;
 398                 case offsetof(struct user, u_debugreg[6]):
 399                         tmp = child->thread.debugreg6;
 400                         break;
 401                 case offsetof(struct user, u_debugreg[7]):
 402                         tmp = child->thread.debugreg7;
 403                         break;
 404                 default:
 405                         tmp = 0;
 406                         break;
 407                 }
 408                 ret = put_user(tmp,(unsigned long __user *) data);
 409                 break;
 410         }
 411
 412         /* when I and D space are separate, this will have to be fixed. */
 413         case PTRACE_POKETEXT: /* write the word at location addr. */
 414         case PTRACE_POKEDATA:
 415                 ret = 0;
 416                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 417                         break;
 418                 ret = -EIO;
 419                 break;
 420
 421         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 422                 ret = -EIO;
 423                 if ((addr & 7) ||
 424                     addr > sizeof(struct user) - 7)
 425                         break;
 426
 427                 switch (addr) {
 428                 case 0 ... sizeof(struct user_regs_struct):
 429                         ret = putreg(child, addr, data);
 430                         break;
 431                 /* Disallows to set a breakpoint into the vsyscall */
 432                 case offsetof(struct user, u_debugreg[0]):
 433                         if (data >= TASK_SIZE-7) break;
 434                         child->thread.debugreg0 = data;
 435                         ret = 0;
 436                         break;
 437                 case offsetof(struct user, u_debugreg[1]):
 438                         if (data >= TASK_SIZE-7) break;
 439                         child->thread.debugreg1 = data;
 440                         ret = 0;
 441                         break;
 442                 case offsetof(struct user, u_debugreg[2]):
 443                         if (data >= TASK_SIZE-7) break;
 444                         child->thread.debugreg2 = data;
 445                         ret = 0;
 446                         break;
 447                 case offsetof(struct user, u_debugreg[3]):
 448                         if (data >= TASK_SIZE-7) break;
 449                         child->thread.debugreg3 = data;
 450                         ret = 0;
 451                         break;
 452                 case offsetof(struct user, u_debugreg[6]):
 453                                   if (data >> 32)
 454                                 break;
 455                         child->thread.debugreg6 = data;
 456                         ret = 0;
 457                         break;
 458                 case offsetof(struct user, u_debugreg[7]):
 459                         /* See arch/i386/kernel/ptrace.c for an explanation of
 460                          * this awkward check.*/
 461                                   data &= ~DR_CONTROL_RESERVED;
 462                                   for(i=0; i<4; i++)
 463                                           if ((0x5454 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 464                                         break;
 465                         if (i == 4) {
 466                                 child->thread.debugreg7 = data;
 467                           ret = 0;
 468                   }
 469                   break;
 470                 }
 471                 break;
 472         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 473         case PTRACE_CONT:    /* restart after signal. */
 474
 475                 ret = -EIO;
 476                 if (!valid_signal(data))
 477                         break;
 478                 if (request == PTRACE_SYSCALL)
 479                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 480                 else
 481                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 482                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 483                 child->exit_code = data;
 484                 /* make sure the single step bit is not set. */
 485                 clear_singlestep(child);
 486                 wake_up_process(child);
 487                 ret = 0;
 488                 break;
 489
 490 #ifdef CONFIG_IA32_EMULATION
 491                 /* This makes only sense with 32bit programs. Allow a
 492                    64bit debugger to fully examine them too. Better
 493                    don't use it against 64bit processes, use
 494                    PTRACE_ARCH_PRCTL instead. */
 495         case PTRACE_SET_THREAD_AREA: {
 496                 struct user_desc __user *p;
 497                 int old;
 498                 p = (struct user_desc __user *)data;
 499                 get_user(old,  &p->entry_number);
 500                 put_user(addr, &p->entry_number);
 501                 ret = do_set_thread_area(&child->thread, p);
 502                 put_user(old,  &p->entry_number);
 503                 break;
 504         case PTRACE_GET_THREAD_AREA:
 505                 p = (struct user_desc __user *)data;
 506                 get_user(old,  &p->entry_number);
 507                 put_user(addr, &p->entry_number);
 508                 ret = do_get_thread_area(&child->thread, p);
 509                 put_user(old,  &p->entry_number);
 510                 break;
 511         }
 512 #endif
 513                 /* normal 64bit interface to access TLS data.
 514                    Works just like arch_prctl, except that the arguments
 515                    are reversed. */
 516         case PTRACE_ARCH_PRCTL:
 517                 ret = do_arch_prctl(child, data, addr);
 518                 break;
 519
 520 /*
 521  * make the child exit.  Best I can do is send it a sigkill.
 522  * perhaps it should be put in the status that it wants to
 523  * exit.
 524  */
 525         case PTRACE_KILL:
 526                 ret = 0;
 527                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 528                         break;
 529                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 530                 child->exit_code = SIGKILL;
 531                 /* make sure the single step bit is not set. */
 532                 clear_singlestep(child);
 533                 wake_up_process(child);
 534                 break;
 535
 536         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 537                 ret = -EIO;
 538                 if (!valid_signal(data))
 539                         break;
 540                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 541                 set_singlestep(child);
 542                 child->exit_code = data;
 543                 /* give it a chance to run. */
 544                 wake_up_process(child);
 545                 ret = 0;
 546                 break;
 547
 548         case PTRACE_DETACH:
 549                 /* detach a process that was attached. */
 550                 ret = ptrace_detach(child, data);
 551                 break;
 552
 553         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 554                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 555                                sizeof(struct user_regs_struct))) {
 556                         ret = -EIO;
 557                         break;
 558                 }
 559                 ret = 0;
 560                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 561                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 562                         data += sizeof(long);
 563                 }
 564                 break;
 565         }
 566
 567         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 568                 unsigned long tmp;
 569                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 570                                sizeof(struct user_regs_struct))) {
 571                         ret = -EIO;
 572                         break;
 573                 }
 574                 ret = 0;
 575                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 576                         ret |= __get_user(tmp, (unsigned long __user *) data);
 577                         putreg(child, ui, tmp);
 578                         data += sizeof(long);
 579                 }
 580                 break;
 581         }
 582
 583         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 584                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 585                                sizeof(struct user_i387_struct))) {
 586                         ret = -EIO;
 587                         break;
 588                 }
 589                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 590                 break;
 591         }
 592
 593         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 594                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 595                                sizeof(struct user_i387_struct))) {
 596                         ret = -EIO;
 597                         break;
 598                 }
 599                 set_stopped_child_used_math(child);
 600                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 601                 break;
 602         }
 603
 604         default:
 605                 ret = ptrace_request(child, request, addr, data);
 606                 break;
 607         }
 608 out_tsk:
 609         put_task_struct(child);
 610 out:
 611         unlock_kernel();
 612         return ret;
 613 }
 614
 615 static void syscall_trace(struct pt_regs *regs)
 616 {
 617
 618 #if 0
 619         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 620                current->comm,
 621                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 622                current_thread_info()->flags, current->ptrace);
 623 #endif
 624
 625         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 626                                 ? 0x80 : 0));
 627         /*
 628          * this isn't the same as continuing with a signal, but it will do
 629          * for normal use.  strace only continues with a signal if the
 630          * stopping signal is not SIGTRAP.  -brl
 631          */
 632         if (current->exit_code) {
 633                 send_sig(current->exit_code, current, 1);
 634                 current->exit_code = 0;
 635         }
 636 }
 637
 638 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 639 {
 640         /* do the secure computing check first */
 641         secure_computing(regs->orig_rax);
 642
 643         if (test_thread_flag(TIF_SYSCALL_TRACE)
 644             && (current->ptrace & PT_PTRACED))
 645                 syscall_trace(regs);
 646
 647         if (unlikely(current->audit_context)) {
 648                 if (test_thread_flag(TIF_IA32)) {
 649                         audit_syscall_entry(current, AUDIT_ARCH_I386,
 650                                             regs->orig_rax,
 651                                             regs->rbx, regs->rcx,
 652                                             regs->rdx, regs->rsi);
 653                 } else {
 654                         audit_syscall_entry(current, AUDIT_ARCH_X86_64,
 655                                             regs->orig_rax,
 656                                             regs->rdi, regs->rsi,
 657                                             regs->rdx, regs->r10);
 658                 }
 659         }
 660 }
 661
 662 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 663 {
 664         if (unlikely(current->audit_context))
 665                 audit_syscall_exit(current, AUDITSC_RESULT(regs->rax), regs->rax);
 666
 667         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 668              || test_thread_flag(TIF_SINGLESTEP))
 669             && (current->ptrace & PT_PTRACED))
 670                 syscall_trace(regs);
 671 }