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/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/user.h>
  17 #include <linux/security.h>
  18 #include <linux/audit.h>
  19 #include <linux/seccomp.h>
  20 #include <linux/signal.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 /*
  39  * Determines which flags the user has access to [1 = access, 0 = no access].
  40  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
  41  * Also masks reserved bits (63-22, 15, 5, 3, 1).
  42  */
  43 #define FLAG_MASK 0x54dd5UL
  44
  45 /* set's the trap flag. */
  46 #define TRAP_FLAG 0x100UL
  47
  48 /*
  49  * eflags and offset of eflags on child stack..
  50  */
  51 #define EFLAGS offsetof(struct pt_regs, eflags)
  52 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
  53
  54 /*
  55  * this routine will get a word off of the processes privileged stack.
  56  * the offset is how far from the base addr as stored in the TSS.
  57  * this routine assumes that all the privileged stacks are in our
  58  * data space.
  59  */
  60 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
  61 {
  62         unsigned char *stack;
  63
  64         stack = (unsigned char *)task->thread.rsp0;
  65         stack += offset;
  66         return (*((unsigned long *)stack));
  67 }
  68
  69 /*
  70  * this routine will put a word on the processes privileged stack.
  71  * the offset is how far from the base addr as stored in the TSS.
  72  * this routine assumes that all the privileged stacks are in our
  73  * data space.
  74  */
  75 static inline long put_stack_long(struct task_struct *task, int offset,
  76         unsigned long data)
  77 {
  78         unsigned char * stack;
  79
  80         stack = (unsigned char *) task->thread.rsp0;
  81         stack += offset;
  82         *(unsigned long *) stack = data;
  83         return 0;
  84 }
  85
  86 #define LDT_SEGMENT 4
  87
  88 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
  89 {
  90         unsigned long addr, seg;
  91
  92         addr = regs->rip;
  93         seg = regs->cs & 0xffff;
  94
  95         /*
  96          * We'll assume that the code segments in the GDT
  97          * are all zero-based. That is largely true: the
  98          * TLS segments are used for data, and the PNPBIOS
  99          * and APM bios ones we just ignore here.
 100          */
 101         if (seg & LDT_SEGMENT) {
 102                 u32 *desc;
 103                 unsigned long base;
 104
 105                 down(&child->mm->context.sem);
 106                 desc = child->mm->context.ldt + (seg & ~7);
 107                 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
 108
 109                 /* 16-bit code segment? */
 110                 if (!((desc[1] >> 22) & 1))
 111                         addr &= 0xffff;
 112                 addr += base;
 113                 up(&child->mm->context.sem);
 114         }
 115         return addr;
 116 }
 117
 118 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 119 {
 120         int i, copied;
 121         unsigned char opcode[15];
 122         unsigned long addr = convert_rip_to_linear(child, regs);
 123
 124         copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 125         for (i = 0; i < copied; i++) {
 126                 switch (opcode[i]) {
 127                 /* popf and iret */
 128                 case 0x9d: case 0xcf:
 129                         return 1;
 130
 131                         /* CHECKME: 64 65 */
 132
 133                 /* opcode and address size prefixes */
 134                 case 0x66: case 0x67:
 135                         continue;
 136                 /* irrelevant prefixes (segment overrides and repeats) */
 137                 case 0x26: case 0x2e:
 138                 case 0x36: case 0x3e:
 139                 case 0x64: case 0x65:
 140                 case 0xf2: case 0xf3:
 141                         continue;
 142
 143                 case 0x40 ... 0x4f:
 144                         if (regs->cs != __USER_CS)
 145                                 /* 32-bit mode: register increment */
 146                                 return 0;
 147                         /* 64-bit mode: REX prefix */
 148                         continue;
 149
 150                         /* CHECKME: 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 = task_pt_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         if (is_setting_trap_flag(child, regs))
 193                 return;
 194
 195         child->ptrace |= PT_DTRACE;
 196 }
 197
 198 static void clear_singlestep(struct task_struct *child)
 199 {
 200         /* Always clear TIF_SINGLESTEP... */
 201         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 202
 203         /* But touch TF only if it was set by us.. */
 204         if (child->ptrace & PT_DTRACE) {
 205                 struct pt_regs *regs = task_pt_regs(child);
 206                 regs->eflags &= ~TRAP_FLAG;
 207                 child->ptrace &= ~PT_DTRACE;
 208         }
 209 }
 210
 211 /*
 212  * Called by kernel/ptrace.c when detaching..
 213  *
 214  * Make sure the single step bit is not set.
 215  */
 216 void ptrace_disable(struct task_struct *child)
 217 {
 218         clear_singlestep(child);
 219 }
 220
 221 static int putreg(struct task_struct *child,
 222         unsigned long regno, unsigned long value)
 223 {
 224         unsigned long tmp;
 225
 226         /* Some code in the 64bit emulation may not be 64bit clean.
 227            Don't take any chances. */
 228         if (test_tsk_thread_flag(child, TIF_IA32))
 229                 value &= 0xffffffff;
 230         switch (regno) {
 231                 case offsetof(struct user_regs_struct,fs):
 232                         if (value && (value & 3) != 3)
 233                                 return -EIO;
 234                         child->thread.fsindex = value & 0xffff;
 235                         return 0;
 236                 case offsetof(struct user_regs_struct,gs):
 237                         if (value && (value & 3) != 3)
 238                                 return -EIO;
 239                         child->thread.gsindex = value & 0xffff;
 240                         return 0;
 241                 case offsetof(struct user_regs_struct,ds):
 242                         if (value && (value & 3) != 3)
 243                                 return -EIO;
 244                         child->thread.ds = value & 0xffff;
 245                         return 0;
 246                 case offsetof(struct user_regs_struct,es):
 247                         if (value && (value & 3) != 3)
 248                                 return -EIO;
 249                         child->thread.es = value & 0xffff;
 250                         return 0;
 251                 case offsetof(struct user_regs_struct,ss):
 252                         if ((value & 3) != 3)
 253                                 return -EIO;
 254                         value &= 0xffff;
 255                         return 0;
 256                 case offsetof(struct user_regs_struct,fs_base):
 257                         if (value >= TASK_SIZE_OF(child))
 258                                 return -EIO;
 259                         child->thread.fs = value;
 260                         return 0;
 261                 case offsetof(struct user_regs_struct,gs_base):
 262                         if (value >= TASK_SIZE_OF(child))
 263                                 return -EIO;
 264                         child->thread.gs = value;
 265                         return 0;
 266                 case offsetof(struct user_regs_struct, eflags):
 267                         value &= FLAG_MASK;
 268                         tmp = get_stack_long(child, EFL_OFFSET);
 269                         tmp &= ~FLAG_MASK;
 270                         value |= tmp;
 271                         break;
 272                 case offsetof(struct user_regs_struct,cs):
 273                         if ((value & 3) != 3)
 274                                 return -EIO;
 275                         value &= 0xffff;
 276                         break;
 277         }
 278         put_stack_long(child, regno - sizeof(struct pt_regs), value);
 279         return 0;
 280 }
 281
 282 static unsigned long getreg(struct task_struct *child, unsigned long regno)
 283 {
 284         unsigned long val;
 285         switch (regno) {
 286                 case offsetof(struct user_regs_struct, fs):
 287                         return child->thread.fsindex;
 288                 case offsetof(struct user_regs_struct, gs):
 289                         return child->thread.gsindex;
 290                 case offsetof(struct user_regs_struct, ds):
 291                         return child->thread.ds;
 292                 case offsetof(struct user_regs_struct, es):
 293                         return child->thread.es;
 294                 case offsetof(struct user_regs_struct, fs_base):
 295                         return child->thread.fs;
 296                 case offsetof(struct user_regs_struct, gs_base):
 297                         return child->thread.gs;
 298                 default:
 299                         regno = regno - sizeof(struct pt_regs);
 300                         val = get_stack_long(child, regno);
 301                         if (test_tsk_thread_flag(child, TIF_IA32))
 302                                 val &= 0xffffffff;
 303                         return val;
 304         }
 305
 306 }
 307
 308 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 309 {
 310         long i, ret;
 311         unsigned ui;
 312
 313         switch (request) {
 314         /* when I and D space are separate, these will need to be fixed. */
 315         case PTRACE_PEEKTEXT: /* read word at location addr. */
 316         case PTRACE_PEEKDATA:
 317                 ret = generic_ptrace_peekdata(child, addr, data);
 318                 break;
 319
 320         /* read the word at location addr in the USER area. */
 321         case PTRACE_PEEKUSR: {
 322                 unsigned long tmp;
 323
 324                 ret = -EIO;
 325                 if ((addr & 7) ||
 326                     addr > sizeof(struct user) - 7)
 327                         break;
 328
 329                 switch (addr) {
 330                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 331                         tmp = getreg(child, addr);
 332                         break;
 333                 case offsetof(struct user, u_debugreg[0]):
 334                         tmp = child->thread.debugreg0;
 335                         break;
 336                 case offsetof(struct user, u_debugreg[1]):
 337                         tmp = child->thread.debugreg1;
 338                         break;
 339                 case offsetof(struct user, u_debugreg[2]):
 340                         tmp = child->thread.debugreg2;
 341                         break;
 342                 case offsetof(struct user, u_debugreg[3]):
 343                         tmp = child->thread.debugreg3;
 344                         break;
 345                 case offsetof(struct user, u_debugreg[6]):
 346                         tmp = child->thread.debugreg6;
 347                         break;
 348                 case offsetof(struct user, u_debugreg[7]):
 349                         tmp = child->thread.debugreg7;
 350                         break;
 351                 default:
 352                         tmp = 0;
 353                         break;
 354                 }
 355                 ret = put_user(tmp,(unsigned long __user *) data);
 356                 break;
 357         }
 358
 359         /* when I and D space are separate, this will have to be fixed. */
 360         case PTRACE_POKETEXT: /* write the word at location addr. */
 361         case PTRACE_POKEDATA:
 362                 ret = generic_ptrace_pokedata(child, addr, data);
 363                 break;
 364
 365         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 366         {
 367                 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
 368                 ret = -EIO;
 369                 if ((addr & 7) ||
 370                     addr > sizeof(struct user) - 7)
 371                         break;
 372
 373                 switch (addr) {
 374                 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
 375                         ret = putreg(child, addr, data);
 376                         break;
 377                 /* Disallows to set a breakpoint into the vsyscall */
 378                 case offsetof(struct user, u_debugreg[0]):
 379                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 380                         child->thread.debugreg0 = data;
 381                         ret = 0;
 382                         break;
 383                 case offsetof(struct user, u_debugreg[1]):
 384                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 385                         child->thread.debugreg1 = data;
 386                         ret = 0;
 387                         break;
 388                 case offsetof(struct user, u_debugreg[2]):
 389                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 390                         child->thread.debugreg2 = data;
 391                         ret = 0;
 392                         break;
 393                 case offsetof(struct user, u_debugreg[3]):
 394                         if (data >= TASK_SIZE_OF(child) - dsize) break;
 395                         child->thread.debugreg3 = data;
 396                         ret = 0;
 397                         break;
 398                 case offsetof(struct user, u_debugreg[6]):
 399                                   if (data >> 32)
 400                                 break;
 401                         child->thread.debugreg6 = data;
 402                         ret = 0;
 403                         break;
 404                 case offsetof(struct user, u_debugreg[7]):
 405                         /* See arch/i386/kernel/ptrace.c for an explanation of
 406                          * this awkward check.*/
 407                         data &= ~DR_CONTROL_RESERVED;
 408                         for(i=0; i<4; i++)
 409                                 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 410                                         break;
 411                         if (i == 4) {
 412                           child->thread.debugreg7 = data;
 413                           if (data)
 414                                 set_tsk_thread_flag(child, TIF_DEBUG);
 415                           else
 416                                 clear_tsk_thread_flag(child, TIF_DEBUG);
 417                           ret = 0;
 418                         }
 419                   break;
 420                 }
 421                 break;
 422         }
 423         case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 424         case PTRACE_CONT:    /* restart after signal. */
 425
 426                 ret = -EIO;
 427                 if (!valid_signal(data))
 428                         break;
 429                 if (request == PTRACE_SYSCALL)
 430                         set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 431                 else
 432                         clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 433                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 434                 child->exit_code = data;
 435                 /* make sure the single step bit is not set. */
 436                 clear_singlestep(child);
 437                 wake_up_process(child);
 438                 ret = 0;
 439                 break;
 440
 441 #ifdef CONFIG_IA32_EMULATION
 442                 /* This makes only sense with 32bit programs. Allow a
 443                    64bit debugger to fully examine them too. Better
 444                    don't use it against 64bit processes, use
 445                    PTRACE_ARCH_PRCTL instead. */
 446         case PTRACE_SET_THREAD_AREA: {
 447                 struct user_desc __user *p;
 448                 int old;
 449                 p = (struct user_desc __user *)data;
 450                 get_user(old,  &p->entry_number);
 451                 put_user(addr, &p->entry_number);
 452                 ret = do_set_thread_area(&child->thread, p);
 453                 put_user(old,  &p->entry_number);
 454                 break;
 455         case PTRACE_GET_THREAD_AREA:
 456                 p = (struct user_desc __user *)data;
 457                 get_user(old,  &p->entry_number);
 458                 put_user(addr, &p->entry_number);
 459                 ret = do_get_thread_area(&child->thread, p);
 460                 put_user(old,  &p->entry_number);
 461                 break;
 462         }
 463 #endif
 464                 /* normal 64bit interface to access TLS data.
 465                    Works just like arch_prctl, except that the arguments
 466                    are reversed. */
 467         case PTRACE_ARCH_PRCTL:
 468                 ret = do_arch_prctl(child, data, addr);
 469                 break;
 470
 471 /*
 472  * make the child exit.  Best I can do is send it a sigkill.
 473  * perhaps it should be put in the status that it wants to
 474  * exit.
 475  */
 476         case PTRACE_KILL:
 477                 ret = 0;
 478                 if (child->exit_state == EXIT_ZOMBIE)   /* already dead */
 479                         break;
 480                 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 481                 child->exit_code = SIGKILL;
 482                 /* make sure the single step bit is not set. */
 483                 clear_singlestep(child);
 484                 wake_up_process(child);
 485                 break;
 486
 487         case PTRACE_SINGLESTEP:    /* set the trap flag. */
 488                 ret = -EIO;
 489                 if (!valid_signal(data))
 490                         break;
 491                 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
 492                 set_singlestep(child);
 493                 child->exit_code = data;
 494                 /* give it a chance to run. */
 495                 wake_up_process(child);
 496                 ret = 0;
 497                 break;
 498
 499         case PTRACE_DETACH:
 500                 /* detach a process that was attached. */
 501                 ret = ptrace_detach(child, data);
 502                 break;
 503
 504         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 505                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 506                                sizeof(struct user_regs_struct))) {
 507                         ret = -EIO;
 508                         break;
 509                 }
 510                 ret = 0;
 511                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 512                         ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
 513                         data += sizeof(long);
 514                 }
 515                 break;
 516         }
 517
 518         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 519                 unsigned long tmp;
 520                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 521                                sizeof(struct user_regs_struct))) {
 522                         ret = -EIO;
 523                         break;
 524                 }
 525                 ret = 0;
 526                 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
 527                         ret = __get_user(tmp, (unsigned long __user *) data);
 528                         if (ret)
 529                                 break;
 530                         ret = putreg(child, ui, tmp);
 531                         if (ret)
 532                                 break;
 533                         data += sizeof(long);
 534                 }
 535                 break;
 536         }
 537
 538         case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
 539                 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
 540                                sizeof(struct user_i387_struct))) {
 541                         ret = -EIO;
 542                         break;
 543                 }
 544                 ret = get_fpregs((struct user_i387_struct __user *)data, child);
 545                 break;
 546         }
 547
 548         case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
 549                 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
 550                                sizeof(struct user_i387_struct))) {
 551                         ret = -EIO;
 552                         break;
 553                 }
 554                 set_stopped_child_used_math(child);
 555                 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
 556                 break;
 557         }
 558
 559         default:
 560                 ret = ptrace_request(child, request, addr, data);
 561                 break;
 562         }
 563         return ret;
 564 }
 565
 566 static void syscall_trace(struct pt_regs *regs)
 567 {
 568
 569 #if 0
 570         printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
 571                current->comm,
 572                regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
 573                current_thread_info()->flags, current->ptrace);
 574 #endif
 575
 576         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
 577                                 ? 0x80 : 0));
 578         /*
 579          * this isn't the same as continuing with a signal, but it will do
 580          * for normal use.  strace only continues with a signal if the
 581          * stopping signal is not SIGTRAP.  -brl
 582          */
 583         if (current->exit_code) {
 584                 send_sig(current->exit_code, current, 1);
 585                 current->exit_code = 0;
 586         }
 587 }
 588
 589 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
 590 {
 591         /* do the secure computing check first */
 592         secure_computing(regs->orig_rax);
 593
 594         if (test_thread_flag(TIF_SYSCALL_TRACE)
 595             && (current->ptrace & PT_PTRACED))
 596                 syscall_trace(regs);
 597
 598         if (unlikely(current->audit_context)) {
 599                 if (test_thread_flag(TIF_IA32)) {
 600                         audit_syscall_entry(AUDIT_ARCH_I386,
 601                                             regs->orig_rax,
 602                                             regs->rbx, regs->rcx,
 603                                             regs->rdx, regs->rsi);
 604                 } else {
 605                         audit_syscall_entry(AUDIT_ARCH_X86_64,
 606                                             regs->orig_rax,
 607                                             regs->rdi, regs->rsi,
 608                                             regs->rdx, regs->r10);
 609                 }
 610         }
 611 }
 612
 613 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
 614 {
 615         if (unlikely(current->audit_context))
 616                 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
 617
 618         if ((test_thread_flag(TIF_SYSCALL_TRACE)
 619              || test_thread_flag(TIF_SINGLESTEP))
 620             && (current->ptrace & PT_PTRACED))
 621                 syscall_trace(regs);
 622 }