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
  22 #include <asm/uaccess.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/system.h>
  25 #include <asm/processor.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 /* determines which flags the user has access to. */
  39 /* 1 = access 0 = no access */
  40 #define FLAG_MASK 0x44dd5UL
  41
  42 /* set's the trap flag. */
  43 #define TRAP_FLAG 0x100UL
  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 static inline struct pt_regs *get_child_regs(struct task_struct *task)
  67 {
  68         struct pt_regs *regs = (void *)task->thread.rsp0;
  69         return regs - 1;
  70 }
  71
  72 /*
  73  * this routine will put a word on the processes privileged stack.
  74  * the offset is how far from the base addr as stored in the TSS.
  75  * this routine assumes that all the privileged stacks are in our
  76  * data space.
  77  */
  78 static inline long put_stack_long(struct task_struct *task, int offset,
  79         unsigned long data)
  80 {
  81         unsigned char * stack;
  82
  83         stack = (unsigned char *) task->thread.rsp0;
  84         stack += offset;
  85         *(unsigned long *) stack = data;
  86         return 0;
  87 }
  88
  89 #define LDT_SEGMENT 4
  90
  91 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  92 {
  93         unsigned long addr, seg;
  94
  95         addr = regs->rip;
  96         seg = regs->cs & 0xffff;
  97
  98         /*
  99          * We'll assume that the code segments in the GDT
 100          * are all zero-based. That is largely true: the
 101          * TLS segments are used for data, and the PNPBIOS
 102          * and APM bios ones we just ignore here.
 103          */
 104         if (seg & LDT_SEGMENT) {
 105                 u32 *desc;
 106                 unsigned long base;
 107
 108                 down(&child->mm->context.sem);
 109                 desc = child->mm->context.ldt + (seg & ~7);
 110                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 111
 112                 /* 16-bit code segment? */
 113                 if (!((desc[1] >> 22) & 1))
 114                         addr &= 0xffff;
 115                 addr += base;
 116                 up(&child->mm->context.sem);
 117         }
 118         return addr;
 119 }
 120
 121 static int is_at_popf(struct task_struct *child, struct pt_regs *regs)
 122 {
 123         int i, copied;
 124         unsigned char opcode[16];
 125         unsigned long addr = convert_rip_to_linear(child, regs);
 126
 127         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 128         for (i = 0; i < copied; i++) {
 129                 switch (opcode[i]) {
 130                 /* popf */
 131                 case 0x9d:
 132                         return 1;
 133
 134                         /* CHECKME: 64 65 */
 135
 136                 /* opcode and address size prefixes */
 137                 case 0x66: case 0x67:
 138                         continue;
 139                 /* irrelevant prefixes (segment overrides and repeats) */
 140                 case 0x26: case 0x2e:
 141                 case 0x36: case 0x3e:
 142                 case 0x64: case 0x65:
 143                 case 0xf0: case 0xf2: case 0xf3:
 144                         continue;
 145
 146                 /* REX prefixes */
 147                 case 0x40 ... 0x4f:
 148                         continue;
 149
 150                         /* CHECKME: f0, f2, f3 */
 151
 152                 /*
 153                  * pushf: NOTE! We should probably not let
 154                  * the user see the TF bit being set. But
 155                  * it's more pain than it's worth to avoid
 156                  * it, and a debugger could emulate this
 157                  * all in user space if it _really_ cares.
 158                  */
 159                 case 0x9c:
 160                 default:
 161                         return 0;
 162                 }
 163         }
 164         return 0;
 165 }
 166
 167 static void set_singlestep(struct task_struct *child)
 168 {
 169         struct pt_regs *regs = get_child_regs(child);
 170
 171         /*
 172          * Always set TIF_SINGLESTEP - this guarantees that
 173          * we single-step system calls etc..  This will also
 174          * cause us to set TF when returning to user mode.
 175          */
 176         set_tsk_thread_flag(child, TIF_SINGLESTEP);
 177
 178         /*
 179          * If TF was already set, don't do anything else
 180          */
 181         if (regs->eflags & TRAP_FLAG)
 182                 return;
 183
 184         /* Set TF on the kernel stack.. */
 185         regs->eflags |= TRAP_FLAG;
 186
 187         /*
 188          * ..but if TF is changed by the instruction we will trace,
 189          * don't mark it as being "us" that set it, so that we
 190          * won't clear it by hand later.
 191          *
 192          * AK: this is not enough, LAHF and IRET can change TF in user space too.
 193          */
 194         if (is_at_popf(child, regs))
 195                 return;
 196
 197         child->ptrace |= PT_DTRACE;
 198 }
 199
 200 static void clear_singlestep(struct task_struct *child)
 201 {
 202         /* Always clear TIF_SINGLESTEP... */
 203         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 204
 205         /* But touch TF only if it was set by us.. */
 206         if (child->ptrace & PT_DTRACE) {
 207                 struct pt_regs *regs = get_child_regs(child);
 208                 regs->eflags &= ~TRAP_FLAG;
 209                 child->ptrace &= ~PT_DTRACE;
 210         }
 211 }
 212
 213 /*
 214  * Called by kernel/ptrace.c when detaching..
 215  *
 216  * Make sure the single step bit is not set.
 217  */
 218 void ptrace_disable(struct task_struct *child)
 219 {
 220         clear_singlestep(child);
 221 }
 222
 223 static int putreg(struct task_struct *child,
 224         unsigned long regno, unsigned long value)
 225 {
 226         unsigned long tmp;
 227
 228         /* Some code in the 64bit emulation may not be 64bit clean.
 229            Don't take any chances. */
 230         if (test_tsk_thread_flag(child, TIF_IA32))
 231                 value &= 0xffffffff;
 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 >> 48) == 0 || (value >> 48) == 0xffff))
 260                                 return -EIO;
 261                         child->thread.fs = value;
 262                         return 0;
 263                 case offsetof(struct user_regs_struct,gs_base):
 264                         if (!((value >> 48) == 0 || (value >> 48) == 0xffff))
 265                                 return -EIO;
 266                         child->thread.gs = value;
 267                         return 0;
 268                 case offsetof(struct user_regs_struct, eflags):
 269                         value &= FLAG_MASK;
 270                         tmp = get_stack_long(child, EFL_OFFSET);
 271                         tmp &= ~FLAG_MASK;
 272                         value |= tmp;
 273                         break;
 274                 case offsetof(struct user_regs_struct,cs):
 275                         if ((value & 3) != 3)
 276                                 return -EIO;
 277                         value &= 0xffff;
 278                         break;
 279         }
 280         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 281         return 0;
 282 }
 283
 284 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 285 {
 286         unsigned long val;
 287         switch (regno) {
 288                 case offsetof(struct user_regs_struct, fs):
 289                         return child->thread.fsindex;
 290                 case offsetof(struct user_regs_struct, gs):
 291                         return child->thread.gsindex;
 292                 case offsetof(struct user_regs_struct, ds):
 293                         return child->thread.ds;
 294                 case offsetof(struct user_regs_struct, es):
 295                         return child->thread.es;
 296                 case offsetof(struct user_regs_struct, fs_base):
 297                         return child->thread.fs;
 298                 case offsetof(struct user_regs_struct, gs_base):
 299                         return child->thread.gs;
 300                 default:
 301                         regno = regno - sizeof(struct pt_regs);
 302                         val = get_stack_long(child, regno);
 303                         if (test_tsk_thread_flag(child, TIF_IA32))
 304                                 val &= 0xffffffff;
 305                         return val;
 306         }
 307
 308 }
 309
 310 asmlinkage long sys_ptrace(long request, long pid, unsigned long addr, long data)
 311 {
 312         struct task_struct *child;
 313         long i, ret;
 314         unsigned ui;
 315
 316         /* This lock_kernel fixes a subtle race with suid exec */
 317         lock_kernel();
 318         ret = -EPERM;
 319         if (request == PTRACE_TRACEME) {
 320                 /* are we already being traced? */
 321                 if (current->ptrace & PT_PTRACED)
 322                         goto out;
 323                 ret = security_ptrace(current->parent, current);
 324                 if (ret)
 325                         goto out;
 326                 /* set the ptrace bit in the process flags. */
 327                 current->ptrace |= PT_PTRACED;
 328                 ret = 0;
 329                 goto out;
 330         }
 331         ret = -ESRCH;
 332         read_lock(&tasklist_lock);
 333         child = find_task_by_pid(pid);
 334         if (child)
 335                 get_task_struct(child);
 336         read_unlock(&tasklist_lock);
 337         if (!child)
 338                 goto out;
 339
 340         ret = -EPERM;
 341         if (pid == 1)           /* you may not mess with init */
 342                 goto out_tsk;
 343
 344         if (request == PTRACE_ATTACH) {
 345                 ret = ptrace_attach(child);
 346                 goto out_tsk;
 347         }
 348         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 349         if (ret < 0)
 350                 goto out_tsk;
 351
 352         switch (request) {
 353         /* when I and D space are separate, these will need to be fixed. */
 354         case PTRACE_PEEKTEXT: /* read word at location addr. */
 355         case PTRACE_PEEKDATA: {
 356                 unsigned long tmp;
 357                 int copied;
 358
 359                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 360                 ret = -EIO;
 361                 if (copied != sizeof(tmp))
 362                         break;
 363                 ret = put_user(tmp,(unsigned long __user *) data);
 364                 break;
 365         }
 366
 367         /* read the word at location addr in the USER area. */
 368         case PTRACE_PEEKUSR: {
 369                 unsigned long tmp;
 370
 371                 ret = -EIO;
 372                 if ((addr & 7) ||
 373                     addr > sizeof(struct user) - 7)
 374                         break;
 375
 376                 switch (addr) {
 377                 case 0 ... sizeof(struct user_regs_struct):
 378                         tmp = getreg(child, addr);
 379                         break;
 380                 case offsetof(struct user, u_debugreg[0]):
 381                         tmp = child->thread.debugreg0;
 382                         break;
 383                 case offsetof(struct user, u_debugreg[1]):
 384                         tmp = child->thread.debugreg1;
 385                         break;
 386                 case offsetof(struct user, u_debugreg[2]):
 387                         tmp = child->thread.debugreg2;
 388                         break;
 389                 case offsetof(struct user, u_debugreg[3]):
 390                         tmp = child->thread.debugreg3;
 391                         break;
 392                 case offsetof(struct user, u_debugreg[6]):
 393                         tmp = child->thread.debugreg6;
 394                         break;
 395                 case offsetof(struct user, u_debugreg[7]):
 396                         tmp = child->thread.debugreg7;
 397                         break;
 398                 default:
 399                         tmp = 0;
 400                         break;
 401                 }
 402                 ret = put_user(tmp,(unsigned long __user *) data);
 403                 break;
 404         }
 405
 406         /* when I and D space are separate, this will have to be fixed. */
 407         case PTRACE_POKETEXT: /* write the word at location addr. */
 408         case PTRACE_POKEDATA:
 409                 ret = 0;
 410                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 411                         break;
 412                 ret = -EIO;
 413                 break;
 414
 415         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 416                 ret = -EIO;
 417                 if ((addr & 7) ||
 418                     addr > sizeof(struct user) - 7)
 419                         break;
 420
 421                 switch (addr) {
 422                 case 0 ... sizeof(struct user_regs_struct):
 423                         ret = putreg(child, addr, data);
 424                         break;
 425                 /* Disallows to set a breakpoint into the vsyscall */
 426                 case offsetof(struct user, u_debugreg[0]):
 427                         if (data >= TASK_SIZE-7) break;
 428                         child->thread.debugreg0 = data;
 429                         ret = 0;
 430                         break;
 431                 case offsetof(struct user, u_debugreg[1]):
 432                         if (data >= TASK_SIZE-7) break;
 433                         child->thread.debugreg1 = data;
 434                         ret = 0;
 435                         break;
 436                 case offsetof(struct user, u_debugreg[2]):
 437                         if (data >= TASK_SIZE-7) break;
 438                         child->thread.debugreg2 = data;
 439                         ret = 0;
 440                         break;
 441                 case offsetof(struct user, u_debugreg[3]):
 442                         if (data >= TASK_SIZE-7) break;
 443                         child->thread.debugreg3 = data;
 444                         ret = 0;
 445                         break;
 446                 case offsetof(struct user, u_debugreg[6]):
 447                                   if (data >> 32)
 448                                 break;
 449                         child->thread.debugreg6 = data;
 450                         ret = 0;
 451                         break;
 452                 case offsetof(struct user, u_debugreg[7]):
 453                         /* See arch/i386/kernel/ptrace.c for an explanation of
 454                          * this awkward check.*/
 455                                   data &= ~DR_CONTROL_RESERVED;
 456                                   for(i=0; i<4; i++)
 457                                           if ((0x5454 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 458                                         break;
 459                         if (i == 4) {
 460                                 child->thread.debugreg7 = data;
 461                           ret = 0;
 462                   }
 463                   break;
 464                 }
 465                 break;
 466         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 467         case PTRACE_CONT:    /* restart after signal. */
 468
 469                 ret = -EIO;
 470                 if ((unsigned long) data > _NSIG)
 471                         break;
 472                 if (request == PTRACE_SYSCALL)
 473                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 474                 else
 475                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 476                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 477                 child->exit_code = data;
 478                 /* make sure the single step bit is not set. */
 479                 clear_singlestep(child);
 480                 wake_up_process(child);
 481                 ret = 0;
 482                 break;
 483
 484 #ifdef CONFIG_IA32_EMULATION
 485                 /* This makes only sense with 32bit programs. Allow a
 486                    64bit debugger to fully examine them too. Better
 487                    don't use it against 64bit processes, use
 488                    PTRACE_ARCH_PRCTL instead. */
 489         case PTRACE_SET_THREAD_AREA: {
 490                 struct user_desc __user *p;
 491                 int old;
 492                 p = (struct user_desc __user *)data;
 493                 get_user(old,  &p->entry_number);
 494                 put_user(addr, &p->entry_number);
 495                 ret = do_set_thread_area(&child->thread, p);
 496                 put_user(old,  &p->entry_number);
 497                 break;
 498         case PTRACE_GET_THREAD_AREA:
 499                 p = (struct user_desc __user *)data;
 500                 get_user(old,  &p->entry_number);
 501                 put_user(addr, &p->entry_number);
 502                 ret = do_get_thread_area(&child->thread, p);
 503                 put_user(old,  &p->entry_number);
 504                 break;
 505         }
 506 #endif
 507                 /* normal 64bit interface to access TLS data.
 508                    Works just like arch_prctl, except that the arguments
 509                    are reversed. */
 510         case PTRACE_ARCH_PRCTL:
 511                 ret = do_arch_prctl(child, data, addr);
 512                 break;
 513
 514 /*
 515  * make the child exit.  Best I can do is send it a sigkill.
 516  * perhaps it should be put in the status that it wants to
 517  * exit.
 518  */
 519         case PTRACE_KILL:
 520                 ret = 0;
 521                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 522                         break;
 523                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 524                 child->exit_code = SIGKILL;
 525                 /* make sure the single step bit is not set. */
 526                 clear_singlestep(child);
 527                 wake_up_process(child);
 528                 break;
 529
 530         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 531                 ret = -EIO;
 532                 if ((unsigned long) data > _NSIG)
 533                         break;
 534                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 535                 set_singlestep(child);
 536                 child->exit_code = data;
 537                 /* give it a chance to run. */
 538                 wake_up_process(child);
 539                 ret = 0;
 540                 break;
 541
 542         case PTRACE_DETACH:
 543                 /* detach a process that was attached. */
 544                 ret = ptrace_detach(child, data);
 545                 break;
 546
 547         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 548                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 549                                sizeof(struct user_regs_struct))) {
 550                         ret = -EIO;
 551                         break;
 552                 }
 553                 ret = 0;
 554                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 555                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 556                         data += sizeof(long);
 557                 }
 558                 break;
 559         }
 560
 561         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 562                 unsigned long tmp;
 563                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 564                                sizeof(struct user_regs_struct))) {
 565                         ret = -EIO;
 566                         break;
 567                 }
 568                 ret = 0;
 569                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 570                         ret |= __get_user(tmp, (unsigned long __user *) data);
 571                         putreg(child, ui, tmp);
 572                         data += sizeof(long);
 573                 }
 574                 break;
 575         }
 576
 577         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 578                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 579                                sizeof(struct user_i387_struct))) {
 580                         ret = -EIO;
 581                         break;
 582                 }
 583                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 584                 break;
 585         }
 586
 587         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 588                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 589                                sizeof(struct user_i387_struct))) {
 590                         ret = -EIO;
 591                         break;
 592                 }
 593                 set_stopped_child_used_math(child);
 594                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 595                 break;
 596         }
 597
 598         default:
 599                 ret = ptrace_request(child, request, addr, data);
 600                 break;
 601         }
 602 out_tsk:
 603         put_task_struct(child);
 604 out:
 605         unlock_kernel();
 606         return ret;
 607 }
 608
 609 static void syscall_trace(struct pt_regs *regs)
 610 {
 611
 612 #if 0
 613         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 614                current->comm,
 615                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 616                current_thread_info()->flags, current->ptrace);
 617 #endif
 618
 619         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 620                                 ? 0x80 : 0));
 621         /*
 622          * this isn't the same as continuing with a signal, but it will do
 623          * for normal use.  strace only continues with a signal if the
 624          * stopping signal is not SIGTRAP.  -brl
 625          */
 626         if (current->exit_code) {
 627                 send_sig(current->exit_code, current, 1);
 628                 current->exit_code = 0;
 629         }
 630 }
 631
 632 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 633 {
 634         /* do the secure computing check first */
 635         secure_computing(regs->orig_rax);
 636
 637         if (unlikely(current->audit_context))
 638                 audit_syscall_entry(current, regs->orig_rax,
 639                                     regs->rdi, regs->rsi,
 640                                     regs->rdx, regs->r10);
 641
 642         if (test_thread_flag(TIF_SYSCALL_TRACE)
 643             && (current->ptrace & PT_PTRACED))
 644                 syscall_trace(regs);
 645 }
 646
 647 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 648 {
 649         if (unlikely(current->audit_context))
 650                 audit_syscall_exit(current, regs->rax);
 651
 652         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 653              || test_thread_flag(TIF_SINGLESTEP))
 654             && (current->ptrace & PT_PTRACED))
 655                 syscall_trace(regs);
 656 }