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