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 /*
  40  * Determines which flags the user has access to [1 = access, 0 = no access].
  41  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
  42  * Also masks reserved bits (63-22, 15, 5, 3, 1).
  43  */
  44 #define FLAG_MASK 0x54dd5UL
  45
  46 /* set's the trap flag. */
  47 #define TRAP_FLAG 0x100UL
  48
  49 /*
  50  * eflags and offset of eflags on child stack..
  51  */
  52 #define EFLAGS offsetof(struct pt_regs, eflags)
  53 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  54
  55 /*
  56  * this routine will get a word off of the processes privileged stack.
  57  * the offset is how far from the base addr as stored in the TSS.
  58  * this routine assumes that all the privileged stacks are in our
  59  * data space.
  60  */
  61 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  62 {
  63         unsigned char *stack;
  64
  65         stack = (unsigned char *)task->thread.rsp0;
  66         stack += offset;
  67         return (*((unsigned long *)stack));
  68 }
  69
  70 /*
  71  * this routine will put a word on the processes privileged stack.
  72  * the offset is how far from the base addr as stored in the TSS.
  73  * this routine assumes that all the privileged stacks are in our
  74  * data space.
  75  */
  76 static inline long put_stack_long(struct task_struct *task, int offset,
  77         unsigned long data)
  78 {
  79         unsigned char * stack;
  80
  81         stack = (unsigned char *) task->thread.rsp0;
  82         stack += offset;
  83         *(unsigned long *) stack = data;
  84         return 0;
  85 }
  86
  87 #define LDT_SEGMENT 4
  88
  89 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  90 {
  91         unsigned long addr, seg;
  92
  93         addr = regs->rip;
  94         seg = regs->cs & 0xffff;
  95
  96         /*
  97          * We'll assume that the code segments in the GDT
  98          * are all zero-based. That is largely true: the
  99          * TLS segments are used for data, and the PNPBIOS
 100          * and APM bios ones we just ignore here.
 101          */
 102         if (seg & LDT_SEGMENT) {
 103                 u32 *desc;
 104                 unsigned long base;
 105
 106                 down(&child->mm->context.sem);
 107                 desc = child->mm->context.ldt + (seg & ~7);
 108                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 109
 110                 /* 16-bit code segment? */
 111                 if (!((desc[1] >> 22) & 1))
 112                         addr &= 0xffff;
 113                 addr += base;
 114                 up(&child->mm->context.sem);
 115         }
 116         return addr;
 117 }
 118
 119 static int is_at_popf(struct task_struct *child, struct pt_regs *regs)
 120 {
 121         int i, copied;
 122         unsigned char opcode[16];
 123         unsigned long addr = convert_rip_to_linear(child, regs);
 124
 125         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 126         for (i = 0; i < copied; i++) {
 127                 switch (opcode[i]) {
 128                 /* popf */
 129                 case 0x9d:
 130                         return 1;
 131
 132                         /* CHECKME: 64 65 */
 133
 134                 /* opcode and address size prefixes */
 135                 case 0x66: case 0x67:
 136                         continue;
 137                 /* irrelevant prefixes (segment overrides and repeats) */
 138                 case 0x26: case 0x2e:
 139                 case 0x36: case 0x3e:
 140                 case 0x64: case 0x65:
 141                 case 0xf0: case 0xf2: case 0xf3:
 142                         continue;
 143
 144                 case 0x40 ... 0x4f:
 145                         if (regs->cs != __USER_CS)
 146                                 /* 32-bit mode: register increment */
 147                                 return 0;
 148                         /* 64-bit mode: REX prefix */
 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 = task_pt_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 = task_pt_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_OF(child))
 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_OF(child))
 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         }
 281         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 282         return 0;
 283 }
 284
 285 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 286 {
 287         unsigned long val;
 288         switch (regno) {
 289                 case offsetof(struct user_regs_struct, fs):
 290                         return child->thread.fsindex;
 291                 case offsetof(struct user_regs_struct, gs):
 292                         return child->thread.gsindex;
 293                 case offsetof(struct user_regs_struct, ds):
 294                         return child->thread.ds;
 295                 case offsetof(struct user_regs_struct, es):
 296                         return child->thread.es;
 297                 case offsetof(struct user_regs_struct, fs_base):
 298                         return child->thread.fs;
 299                 case offsetof(struct user_regs_struct, gs_base):
 300                         return child->thread.gs;
 301                 default:
 302                         regno = regno - sizeof(struct pt_regs);
 303                         val = get_stack_long(child, regno);
 304                         if (test_tsk_thread_flag(child, TIF_IA32))
 305                                 val &= 0xffffffff;
 306                         return val;
 307         }
 308
 309 }
 310
 311 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 312 {
 313         long i, ret;
 314         unsigned ui;
 315
 316         switch (request) {
 317         /* when I and D space are separate, these will need to be fixed. */
 318         case PTRACE_PEEKTEXT: /* read word at location addr. */
 319         case PTRACE_PEEKDATA: {
 320                 unsigned long tmp;
 321                 int copied;
 322
 323                 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
 324                 ret = -EIO;
 325                 if (copied != sizeof(tmp))
 326                         break;
 327                 ret = put_user(tmp,(unsigned long __user *) data);
 328                 break;
 329         }
 330
 331         /* read the word at location addr in the USER area. */
 332         case PTRACE_PEEKUSR: {
 333                 unsigned long tmp;
 334
 335                 ret = -EIO;
 336                 if ((addr & 7) ||
 337                     addr > sizeof(struct user) - 7)
 338                         break;
 339
 340                 switch (addr) {
 341                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 342                         tmp = getreg(child, addr);
 343                         break;
 344                 case offsetof(struct user, u_debugreg[0]):
 345                         tmp = child->thread.debugreg0;
 346                         break;
 347                 case offsetof(struct user, u_debugreg[1]):
 348                         tmp = child->thread.debugreg1;
 349                         break;
 350                 case offsetof(struct user, u_debugreg[2]):
 351                         tmp = child->thread.debugreg2;
 352                         break;
 353                 case offsetof(struct user, u_debugreg[3]):
 354                         tmp = child->thread.debugreg3;
 355                         break;
 356                 case offsetof(struct user, u_debugreg[6]):
 357                         tmp = child->thread.debugreg6;
 358                         break;
 359                 case offsetof(struct user, u_debugreg[7]):
 360                         tmp = child->thread.debugreg7;
 361                         break;
 362                 default:
 363                         tmp = 0;
 364                         break;
 365                 }
 366                 ret = put_user(tmp,(unsigned long __user *) data);
 367                 break;
 368         }
 369
 370         /* when I and D space are separate, this will have to be fixed. */
 371         case PTRACE_POKETEXT: /* write the word at location addr. */
 372         case PTRACE_POKEDATA:
 373                 ret = 0;
 374                 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
 375                         break;
 376                 ret = -EIO;
 377                 break;
 378
 379         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 380         {
 381                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 382                 ret = -EIO;
 383                 if ((addr & 7) ||
 384                     addr > sizeof(struct user) - 7)
 385                         break;
 386
 387                 switch (addr) {
 388                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 389                         ret = putreg(child, addr, data);
 390                         break;
 391                 /* Disallows to set a breakpoint into the vsyscall */
 392                 case offsetof(struct user, u_debugreg[0]):
 393                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 394                         child->thread.debugreg0 = data;
 395                         ret = 0;
 396                         break;
 397                 case offsetof(struct user, u_debugreg[1]):
 398                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 399                         child->thread.debugreg1 = data;
 400                         ret = 0;
 401                         break;
 402                 case offsetof(struct user, u_debugreg[2]):
 403                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 404                         child->thread.debugreg2 = data;
 405                         ret = 0;
 406                         break;
 407                 case offsetof(struct user, u_debugreg[3]):
 408                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 409                         child->thread.debugreg3 = data;
 410                         ret = 0;
 411                         break;
 412                 case offsetof(struct user, u_debugreg[6]):
 413                                   if (data >> 32)
 414                                 break;
 415                         child->thread.debugreg6 = data;
 416                         ret = 0;
 417                         break;
 418                 case offsetof(struct user, u_debugreg[7]):
 419                         /* See arch/i386/kernel/ptrace.c for an explanation of
 420                          * this awkward check.*/
 421                         data &= ~DR_CONTROL_RESERVED;
 422                         for(i=0; i<4; i++)
 423                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 424                                         break;
 425                         if (i == 4) {
 426                           child->thread.debugreg7 = data;
 427                           if (data)
 428                                 set_tsk_thread_flag(child, TIF_DEBUG);
 429                           else
 430                                 clear_tsk_thread_flag(child, TIF_DEBUG);
 431                           ret = 0;
 432                         }
 433                   break;
 434                 }
 435                 break;
 436         }
 437         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 438         case PTRACE_CONT:    /* restart after signal. */
 439
 440                 ret = -EIO;
 441                 if (!valid_signal(data))
 442                         break;
 443                 if (request == PTRACE_SYSCALL)
 444                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 445                 else
 446                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 447                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 448                 child->exit_code = data;
 449                 /* make sure the single step bit is not set. */
 450                 clear_singlestep(child);
 451                 wake_up_process(child);
 452                 ret = 0;
 453                 break;
 454
 455 #ifdef CONFIG_IA32_EMULATION
 456                 /* This makes only sense with 32bit programs. Allow a
 457                    64bit debugger to fully examine them too. Better
 458                    don't use it against 64bit processes, use
 459                    PTRACE_ARCH_PRCTL instead. */
 460         case PTRACE_SET_THREAD_AREA: {
 461                 struct user_desc __user *p;
 462                 int old;
 463                 p = (struct user_desc __user *)data;
 464                 get_user(old,  &p->entry_number);
 465                 put_user(addr, &p->entry_number);
 466                 ret = do_set_thread_area(&child->thread, p);
 467                 put_user(old,  &p->entry_number);
 468                 break;
 469         case PTRACE_GET_THREAD_AREA:
 470                 p = (struct user_desc __user *)data;
 471                 get_user(old,  &p->entry_number);
 472                 put_user(addr, &p->entry_number);
 473                 ret = do_get_thread_area(&child->thread, p);
 474                 put_user(old,  &p->entry_number);
 475                 break;
 476         }
 477 #endif
 478                 /* normal 64bit interface to access TLS data.
 479                    Works just like arch_prctl, except that the arguments
 480                    are reversed. */
 481         case PTRACE_ARCH_PRCTL:
 482                 ret = do_arch_prctl(child, data, addr);
 483                 break;
 484
 485 /*
 486  * make the child exit.  Best I can do is send it a sigkill.
 487  * perhaps it should be put in the status that it wants to
 488  * exit.
 489  */
 490         case PTRACE_KILL:
 491                 ret = 0;
 492                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 493                         break;
 494                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 495                 child->exit_code = SIGKILL;
 496                 /* make sure the single step bit is not set. */
 497                 clear_singlestep(child);
 498                 wake_up_process(child);
 499                 break;
 500
 501         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 502                 ret = -EIO;
 503                 if (!valid_signal(data))
 504                         break;
 505                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 506                 set_singlestep(child);
 507                 child->exit_code = data;
 508                 /* give it a chance to run. */
 509                 wake_up_process(child);
 510                 ret = 0;
 511                 break;
 512
 513         case PTRACE_DETACH:
 514                 /* detach a process that was attached. */
 515                 ret = ptrace_detach(child, data);
 516                 break;
 517
 518         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 519                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 520                                sizeof(struct user_regs_struct))) {
 521                         ret = -EIO;
 522                         break;
 523                 }
 524                 ret = 0;
 525                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 526                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 527                         data += sizeof(long);
 528                 }
 529                 break;
 530         }
 531
 532         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 533                 unsigned long tmp;
 534                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 535                                sizeof(struct user_regs_struct))) {
 536                         ret = -EIO;
 537                         break;
 538                 }
 539                 ret = 0;
 540                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 541                         ret |= __get_user(tmp, (unsigned long __user *) data);
 542                         putreg(child, ui, tmp);
 543                         data += sizeof(long);
 544                 }
 545                 break;
 546         }
 547
 548         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 549                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 550                                sizeof(struct user_i387_struct))) {
 551                         ret = -EIO;
 552                         break;
 553                 }
 554                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 555                 break;
 556         }
 557
 558         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 559                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 560                                sizeof(struct user_i387_struct))) {
 561                         ret = -EIO;
 562                         break;
 563                 }
 564                 set_stopped_child_used_math(child);
 565                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 566                 break;
 567         }
 568
 569         default:
 570                 ret = ptrace_request(child, request, addr, data);
 571                 break;
 572         }
 573         return ret;
 574 }
 575
 576 static void syscall_trace(struct pt_regs *regs)
 577 {
 578
 579 #if 0
 580         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 581                current->comm,
 582                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 583                current_thread_info()->flags, current->ptrace);
 584 #endif
 585
 586         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 587                                 ? 0x80 : 0));
 588         /*
 589          * this isn't the same as continuing with a signal, but it will do
 590          * for normal use.  strace only continues with a signal if the
 591          * stopping signal is not SIGTRAP.  -brl
 592          */
 593         if (current->exit_code) {
 594                 send_sig(current->exit_code, current, 1);
 595                 current->exit_code = 0;
 596         }
 597 }
 598
 599 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 600 {
 601         /* do the secure computing check first */
 602         secure_computing(regs->orig_rax);
 603
 604         if (test_thread_flag(TIF_SYSCALL_TRACE)
 605             && (current->ptrace & PT_PTRACED))
 606                 syscall_trace(regs);
 607
 608         if (unlikely(current->audit_context)) {
 609                 if (test_thread_flag(TIF_IA32)) {
 610                         audit_syscall_entry(AUDIT_ARCH_I386,
 611                                             regs->orig_rax,
 612                                             regs->rbx, regs->rcx,
 613                                             regs->rdx, regs->rsi);
 614                 } else {
 615                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 616                                             regs->orig_rax,
 617                                             regs->rdi, regs->rsi,
 618                                             regs->rdx, regs->r10);
 619                 }
 620         }
 621 }
 622
 623 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 624 {
 625         if (unlikely(current->audit_context))
 626                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 627
 628         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 629              || test_thread_flag(TIF_SINGLESTEP))
 630             && (current->ptrace & PT_PTRACED))
 631                 syscall_trace(regs);
 632 }