arch/x86/kernel/ptrace_32.c

   1 /* By Ross Biro 1/23/92 */
   2 /*
   3  * Pentium III FXSR, SSE support
   4  *      Gareth Hughes <gareth@valinux.com>, May 2000
   5  */
   6
   7 #include <linux/kernel.h>
   8 #include <linux/sched.h>
   9 #include <linux/mm.h>
  10 #include <linux/smp.h>
  11 #include <linux/errno.h>
  12 #include <linux/ptrace.h>
  13 #include <linux/user.h>
  14 #include <linux/security.h>
  15 #include <linux/audit.h>
  16 #include <linux/seccomp.h>
  17 #include <linux/signal.h>
  18
  19 #include <asm/uaccess.h>
  20 #include <asm/pgtable.h>
  21 #include <asm/system.h>
  22 #include <asm/processor.h>
  23 #include <asm/i387.h>
  24 #include <asm/debugreg.h>
  25 #include <asm/ldt.h>
  26 #include <asm/desc.h>
  27
  28 /*
  29  * does not yet catch signals sent when the child dies.
  30  * in exit.c or in signal.c.
  31  */
  32
  33 /*
  34  * Determines which flags the user has access to [1 = access, 0 = no access].
  35  * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), NT(14), IOPL(12-13), IF(9).
  36  * Also masks reserved bits (31-22, 15, 5, 3, 1).
  37  */
  38 #define FLAG_MASK 0x00050dd5
  39
  40 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  41 {
  42         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  43         if (regno > FS)
  44                 --regno;
  45         return &regs->bx + regno;
  46 }
  47
  48 static int putreg(struct task_struct *child,
  49         unsigned long regno, unsigned long value)
  50 {
  51         struct pt_regs *regs = task_pt_regs(child);
  52         regno >>= 2;
  53         switch (regno) {
  54                 case GS:
  55                         if (value && (value & 3) != 3)
  56                                 return -EIO;
  57                         child->thread.gs = value;
  58                         return 0;
  59                 case DS:
  60                 case ES:
  61                 case FS:
  62                         if (value && (value & 3) != 3)
  63                                 return -EIO;
  64                         value &= 0xffff;
  65                         break;
  66                 case SS:
  67                 case CS:
  68                         if ((value & 3) != 3)
  69                                 return -EIO;
  70                         value &= 0xffff;
  71                         break;
  72                 case EFL:
  73                         value &= FLAG_MASK;
  74                         /*
  75                          * If the user value contains TF, mark that
  76                          * it was not "us" (the debugger) that set it.
  77                          * If not, make sure it stays set if we had.
  78                          */
  79                         if (value & X86_EFLAGS_TF)
  80                                 clear_tsk_thread_flag(child, TIF_FORCED_TF);
  81                         else if (test_tsk_thread_flag(child, TIF_FORCED_TF))
  82                                 value |= X86_EFLAGS_TF;
  83                         value |= regs->flags & ~FLAG_MASK;
  84                         break;
  85         }
  86         *pt_regs_access(regs, regno) = value;
  87         return 0;
  88 }
  89
  90 static unsigned long getreg(struct task_struct *child, unsigned long regno)
  91 {
  92         struct pt_regs *regs = task_pt_regs(child);
  93         unsigned long retval = ~0UL;
  94
  95         regno >>= 2;
  96         switch (regno) {
  97                 case EFL:
  98                         /*
  99                          * If the debugger set TF, hide it from the readout.
 100                          */
 101                         retval = regs->flags;
 102                         if (test_tsk_thread_flag(child, TIF_FORCED_TF))
 103                                 retval &= ~X86_EFLAGS_TF;
 104                         break;
 105                 case GS:
 106                         retval = child->thread.gs;
 107                         break;
 108                 case DS:
 109                 case ES:
 110                 case FS:
 111                 case SS:
 112                 case CS:
 113                         retval = 0xffff;
 114                         /* fall through */
 115                 default:
 116                         retval &= *pt_regs_access(regs, regno);
 117         }
 118         return retval;
 119 }
 120
 121 /*
 122  * This function is trivial and will be inlined by the compiler.
 123  * Having it separates the implementation details of debug
 124  * registers from the interface details of ptrace.
 125  */
 126 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 127 {
 128         return child->thread.debugreg[n];
 129 }
 130
 131 static int ptrace_set_debugreg(struct task_struct *child,
 132                                int n, unsigned long data)
 133 {
 134         if (unlikely(n == 4 || n == 5))
 135                 return -EIO;
 136
 137         if (n < 4 && unlikely(data >= TASK_SIZE - 3))
 138                 return -EIO;
 139
 140         if (n == 7) {
 141                 /*
 142                  * Sanity-check data. Take one half-byte at once with
 143                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 144                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 145                  * 2 and 3 are LENi. Given a list of invalid values,
 146                  * we do mask |= 1 << invalid_value, so that
 147                  * (mask >> check) & 1 is a correct test for invalid
 148                  * values.
 149                  *
 150                  * R/Wi contains the type of the breakpoint /
 151                  * watchpoint, LENi contains the length of the watched
 152                  * data in the watchpoint case.
 153                  *
 154                  * The invalid values are:
 155                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.
 156                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 157                  *   mask |= 0x4444.
 158                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 159                  *   0x1110.
 160                  *
 161                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 162                  *
 163                  * See the Intel Manual "System Programming Guide",
 164                  * 15.2.4
 165                  *
 166                  * Note that LENi == 0x10 is defined on x86_64 in long
 167                  * mode (i.e. even for 32-bit userspace software, but
 168                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 169                  * See the AMD manual no. 24593 (AMD64 System Programming)
 170                  */
 171                 int i;
 172                 data &= ~DR_CONTROL_RESERVED;
 173                 for (i = 0; i < 4; i++)
 174                         if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 175                                 return -EIO;
 176                 if (data)
 177                         set_tsk_thread_flag(child, TIF_DEBUG);
 178                 else
 179                         clear_tsk_thread_flag(child, TIF_DEBUG);
 180         }
 181
 182         child->thread.debugreg[n] = data;
 183
 184         return 0;
 185 }
 186
 187 /*
 188  * Called by kernel/ptrace.c when detaching..
 189  *
 190  * Make sure the single step bit is not set.
 191  */
 192 void ptrace_disable(struct task_struct *child)
 193 {
 194         user_disable_single_step(child);
 195         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 196 }
 197
 198 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 199 {
 200         struct user * dummy = NULL;
 201         int i, ret;
 202         unsigned long __user *datap = (unsigned long __user *)data;
 203
 204         switch (request) {
 205         /* when I and D space are separate, these will need to be fixed. */
 206         case PTRACE_PEEKTEXT: /* read word at location addr. */
 207         case PTRACE_PEEKDATA:
 208                 ret = generic_ptrace_peekdata(child, addr, data);
 209                 break;
 210
 211         /* read the word at location addr in the USER area. */
 212         case PTRACE_PEEKUSR: {
 213                 unsigned long tmp;
 214
 215                 ret = -EIO;
 216                 if ((addr & 3) || addr < 0 ||
 217                     addr > sizeof(struct user) - 3)
 218                         break;
 219
 220                 tmp = 0;  /* Default return condition */
 221                 if(addr < FRAME_SIZE*sizeof(long))
 222                         tmp = getreg(child, addr);
 223                 if(addr >= (long) &dummy->u_debugreg[0] &&
 224                    addr <= (long) &dummy->u_debugreg[7]){
 225                         addr -= (long) &dummy->u_debugreg[0];
 226                         addr = addr >> 2;
 227                         tmp = ptrace_get_debugreg(child, addr);
 228                 }
 229                 ret = put_user(tmp, datap);
 230                 break;
 231         }
 232
 233         /* when I and D space are separate, this will have to be fixed. */
 234         case PTRACE_POKETEXT: /* write the word at location addr. */
 235         case PTRACE_POKEDATA:
 236                 ret = generic_ptrace_pokedata(child, addr, data);
 237                 break;
 238
 239         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 240                 ret = -EIO;
 241                 if ((addr & 3) || addr < 0 ||
 242                     addr > sizeof(struct user) - 3)
 243                         break;
 244
 245                 if (addr < FRAME_SIZE*sizeof(long)) {
 246                         ret = putreg(child, addr, data);
 247                         break;
 248                 }
 249                 /* We need to be very careful here.  We implicitly
 250                    want to modify a portion of the task_struct, and we
 251                    have to be selective about what portions we allow someone
 252                    to modify. */
 253
 254                   ret = -EIO;
 255                   if(addr >= (long) &dummy->u_debugreg[0] &&
 256                      addr <= (long) &dummy->u_debugreg[7]){
 257                           addr -= (long) &dummy->u_debugreg;
 258                           addr = addr >> 2;
 259                           ret = ptrace_set_debugreg(child, addr, data);
 260                   }
 261                   break;
 262
 263         case PTRACE_GETREGS: { /* Get all gp regs from the child. */
 264                 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
 265                         ret = -EIO;
 266                         break;
 267                 }
 268                 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
 269                         __put_user(getreg(child, i), datap);
 270                         datap++;
 271                 }
 272                 ret = 0;
 273                 break;
 274         }
 275
 276         case PTRACE_SETREGS: { /* Set all gp regs in the child. */
 277                 unsigned long tmp;
 278                 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
 279                         ret = -EIO;
 280                         break;
 281                 }
 282                 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
 283                         __get_user(tmp, datap);
 284                         putreg(child, i, tmp);
 285                         datap++;
 286                 }
 287                 ret = 0;
 288                 break;
 289         }
 290
 291         case PTRACE_GETFPREGS: { /* Get the child FPU state. */
 292                 if (!access_ok(VERIFY_WRITE, datap,
 293                                sizeof(struct user_i387_struct))) {
 294                         ret = -EIO;
 295                         break;
 296                 }
 297                 ret = 0;
 298                 if (!tsk_used_math(child))
 299                         init_fpu(child);
 300                 get_fpregs((struct user_i387_struct __user *)data, child);
 301                 break;
 302         }
 303
 304         case PTRACE_SETFPREGS: { /* Set the child FPU state. */
 305                 if (!access_ok(VERIFY_READ, datap,
 306                                sizeof(struct user_i387_struct))) {
 307                         ret = -EIO;
 308                         break;
 309                 }
 310                 set_stopped_child_used_math(child);
 311                 set_fpregs(child, (struct user_i387_struct __user *)data);
 312                 ret = 0;
 313                 break;
 314         }
 315
 316         case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
 317                 if (!access_ok(VERIFY_WRITE, datap,
 318                                sizeof(struct user_fxsr_struct))) {
 319                         ret = -EIO;
 320                         break;
 321                 }
 322                 if (!tsk_used_math(child))
 323                         init_fpu(child);
 324                 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
 325                 break;
 326         }
 327
 328         case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
 329                 if (!access_ok(VERIFY_READ, datap,
 330                                sizeof(struct user_fxsr_struct))) {
 331                         ret = -EIO;
 332                         break;
 333                 }
 334                 set_stopped_child_used_math(child);
 335                 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
 336                 break;
 337         }
 338
 339         case PTRACE_GET_THREAD_AREA:
 340                 if (addr < 0)
 341                         return -EIO;
 342                 ret = do_get_thread_area(child, addr,
 343                                          (struct user_desc __user *) data);
 344                 break;
 345
 346         case PTRACE_SET_THREAD_AREA:
 347                 if (addr < 0)
 348                         return -EIO;
 349                 ret = do_set_thread_area(child, addr,
 350                                          (struct user_desc __user *) data, 0);
 351                 break;
 352
 353         default:
 354                 ret = ptrace_request(child, request, addr, data);
 355                 break;
 356         }
 357
 358         return ret;
 359 }
 360
 361 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
 362 {
 363         struct siginfo info;
 364
 365         tsk->thread.trap_no = 1;
 366         tsk->thread.error_code = error_code;
 367
 368         memset(&info, 0, sizeof(info));
 369         info.si_signo = SIGTRAP;
 370         info.si_code = TRAP_BRKPT;
 371
 372         /* User-mode ip? */
 373         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
 374
 375         /* Send us the fake SIGTRAP */
 376         force_sig_info(SIGTRAP, &info, tsk);
 377 }
 378
 379 /* notification of system call entry/exit
 380  * - triggered by current->work.syscall_trace
 381  */
 382 __attribute__((regparm(3)))
 383 int do_syscall_trace(struct pt_regs *regs, int entryexit)
 384 {
 385         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
 386         /*
 387          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
 388          * interception
 389          */
 390         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
 391         int ret = 0;
 392
 393         /* do the secure computing check first */
 394         if (!entryexit)
 395                 secure_computing(regs->orig_ax);
 396
 397         if (unlikely(current->audit_context)) {
 398                 if (entryexit)
 399                         audit_syscall_exit(AUDITSC_RESULT(regs->ax),
 400                                                 regs->ax);
 401                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
 402                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
 403                  * not used, entry.S will call us only on syscall exit, not
 404                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
 405                  * calling send_sigtrap() on syscall entry.
 406                  *
 407                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
 408                  * is_singlestep is false, despite his name, so we will still do
 409                  * the correct thing.
 410                  */
 411                 else if (is_singlestep)
 412                         goto out;
 413         }
 414
 415         if (!(current->ptrace & PT_PTRACED))
 416                 goto out;
 417
 418         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
 419          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
 420          * here. We have to check this and return */
 421         if (is_sysemu && entryexit)
 422                 return 0;
 423
 424         /* Fake a debug trap */
 425         if (is_singlestep)
 426                 send_sigtrap(current, regs, 0);
 427
 428         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
 429                 goto out;
 430
 431         /* the 0x80 provides a way for the tracing parent to distinguish
 432            between a syscall stop and SIGTRAP delivery */
 433         /* Note that the debugger could change the result of test_thread_flag!*/
 434         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
 435
 436         /*
 437          * this isn't the same as continuing with a signal, but it will do
 438          * for normal use.  strace only continues with a signal if the
 439          * stopping signal is not SIGTRAP.  -brl
 440          */
 441         if (current->exit_code) {
 442                 send_sig(current->exit_code, current, 1);
 443                 current->exit_code = 0;
 444         }
 445         ret = is_sysemu;
 446 out:
 447         if (unlikely(current->audit_context) && !entryexit)
 448                 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
 449                                     regs->bx, regs->cx, regs->dx, regs->si);
 450         if (ret == 0)
 451                 return 0;
 452
 453         regs->orig_ax = -1; /* force skip of syscall restarting */
 454         if (unlikely(current->audit_context))
 455                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
 456         return 1;
 457 }