bsd/dev/i386/systemcalls.c

   1 /*
   2  * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 #include <kern/task.h>
  29 #include <kern/thread.h>
  30 #include <kern/assert.h>
  31 #include <kern/clock.h>
  32 #include <kern/locks.h>
  33 #include <kern/sched_prim.h>
  34 #include <kern/debug.h>
  35 #include <mach/machine/thread_status.h>
  36 #include <mach/thread_act.h>
  37 #include <mach/branch_predicates.h>
  38
  39 #include <sys/kernel.h>
  40 #include <sys/vm.h>
  41 #include <sys/proc_internal.h>
  42 #include <sys/syscall.h>
  43 #include <sys/systm.h>
  44 #include <sys/user.h>
  45 #include <sys/errno.h>
  46 #include <sys/kdebug.h>
  47 #include <sys/sysent.h>
  48 #include <sys/sysproto.h>
  49 #include <sys/kauth.h>
  50 #include <sys/systm.h>
  51
  52 #include <security/audit/audit.h>
  53
  54 #include <i386/seg.h>
  55 #include <i386/machine_routines.h>
  56 #include <mach/i386/syscall_sw.h>
  57
  58 #include <machine/pal_routines.h>
  59
  60 #if CONFIG_DTRACE
  61 extern int32_t dtrace_systrace_syscall(struct proc *, void *, int *);
  62 extern void dtrace_systrace_syscall_return(unsigned short, int, int *);
  63 #endif
  64
  65 extern void unix_syscall(x86_saved_state_t *);
  66 extern void unix_syscall64(x86_saved_state_t *);
  67 extern void *find_user_regs(thread_t);
  68
  69 /* dynamically generated at build time based on syscalls.master */
  70 extern const char *syscallnames[];
  71
  72 #define code_is_kdebug_trace(code) (((code) == SYS_kdebug_trace) || ((code) == SYS_kdebug_trace64))
  73
  74 /*
  75  * Function:    unix_syscall
  76  *
  77  * Inputs:      regs    - pointer to i386 save area
  78  *
  79  * Outputs:     none
  80  */
  81 void
  82 unix_syscall(x86_saved_state_t *state)
  83 {
  84         thread_t                thread;
  85         void                    *vt;
  86         unsigned int            code;
  87         struct sysent           *callp;
  88
  89         int                     error;
  90         vm_offset_t             params;
  91         struct proc             *p;
  92         struct uthread          *uthread;
  93         x86_saved_state32_t     *regs;
  94         boolean_t               is_vfork;
  95
  96         assert(is_saved_state32(state));
  97         regs = saved_state32(state);
  98 #if DEBUG
  99         if (regs->eax == 0x800)
 100                 thread_exception_return();
 101 #endif
 102         thread = current_thread();
 103         uthread = get_bsdthread_info(thread);
 104
 105         /* Get the approriate proc; may be different from task's for vfork() */
 106         is_vfork = uthread->uu_flag & UT_VFORK;
 107         if (__improbable(is_vfork != 0))
 108                 p = current_proc();
 109         else
 110                 p = (struct proc *)get_bsdtask_info(current_task());
 111
 112         /* Verify that we are not being called from a task without a proc */
 113         if (__improbable(p == NULL)) {
 114                 regs->eax = EPERM;
 115                 regs->efl |= EFL_CF;
 116                 task_terminate_internal(current_task());
 117                 thread_exception_return();
 118                 /* NOTREACHED */
 119         }
 120
 121         code = regs->eax & I386_SYSCALL_NUMBER_MASK;
 122         DEBUG_KPRINT_SYSCALL_UNIX("unix_syscall: code=%d(%s) eip=%u\n",
 123                                                           code, syscallnames[code >= NUM_SYSENT ? 63 : code], (uint32_t)regs->eip);
 124         params = (vm_offset_t) (regs->uesp + sizeof (int));
 125
 126         regs->efl &= ~(EFL_CF);
 127
 128         callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 129
 130         if (__improbable(callp == sysent)) {
 131                 code = fuword(params);
 132                 params += sizeof(int);
 133                 callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 134         }
 135
 136         vt = (void *)uthread->uu_arg;
 137
 138         if (callp->sy_arg_bytes != 0) {
 139 #if CONFIG_REQUIRES_U32_MUNGING
 140                 sy_munge_t      *mungerp;
 141 #else
 142 #error U32 syscalls on x86_64 kernel requires munging
 143 #endif
 144                 uint32_t         nargs;
 145
 146                 assert((unsigned) callp->sy_arg_bytes <= sizeof (uthread->uu_arg));
 147                 nargs = callp->sy_arg_bytes;
 148                 error = copyin((user_addr_t) params, (char *) vt, nargs);
 149                 if (error) {
 150                         regs->eax = error;
 151                         regs->efl |= EFL_CF;
 152                         thread_exception_return();
 153                         /* NOTREACHED */
 154                 }
 155
 156                 if (__probable(!code_is_kdebug_trace(code))) {
 157                         int *ip = (int *)vt;
 158
 159                         KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 160                                 BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
 161                                 *ip, *(ip+1), *(ip+2), *(ip+3), 0);
 162                 }
 163
 164 #if CONFIG_REQUIRES_U32_MUNGING
 165                 mungerp = callp->sy_arg_munge32;
 166
 167                 if (mungerp != NULL)
 168                         (*mungerp)(vt);
 169 #endif
 170         } else
 171                 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 172                         BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
 173                         0, 0, 0, 0, 0);
 174
 175         /*
 176          * Delayed binding of thread credential to process credential, if we
 177          * are not running with an explicitly set thread credential.
 178          */
 179         kauth_cred_uthread_update(uthread, p);
 180
 181         uthread->uu_rval[0] = 0;
 182         uthread->uu_rval[1] = 0;
 183         uthread->uu_flag |= UT_NOTCANCELPT;
 184         uthread->syscall_code = code;
 185
 186 #ifdef JOE_DEBUG
 187         uthread->uu_iocount = 0;
 188         uthread->uu_vpindex = 0;
 189 #endif
 190
 191         AUDIT_SYSCALL_ENTER(code, p, uthread);
 192         error = (*(callp->sy_call))((void *) p, (void *) vt, &(uthread->uu_rval[0]));
 193         AUDIT_SYSCALL_EXIT(code, p, uthread, error);
 194
 195 #ifdef JOE_DEBUG
 196         if (uthread->uu_iocount)
 197                 printf("system call returned with uu_iocount != 0\n");
 198 #endif
 199 #if CONFIG_DTRACE
 200         uthread->t_dtrace_errno = error;
 201 #endif /* CONFIG_DTRACE */
 202
 203         if (__improbable(error == ERESTART)) {
 204                 /*
 205                  * Move the user's pc back to repeat the syscall:
 206                  * 5 bytes for a sysenter, or 2 for an int 8x.
 207                  * The SYSENTER_TF_CS covers single-stepping over a sysenter
 208                  * - see debug trap handler in idt.s/idt64.s
 209                  */
 210
 211                 pal_syscall_restart(thread, state);
 212         }
 213         else if (error != EJUSTRETURN) {
 214                 if (__improbable(error)) {
 215                     regs->eax = error;
 216                     regs->efl |= EFL_CF;        /* carry bit */
 217                 } else { /* (not error) */
 218                         /*
 219                          * We split retval across two registers, in case the
 220                          * syscall had a 64-bit return value, in which case
 221                          * eax/edx matches the function call ABI.
 222                          */
 223                     regs->eax = uthread->uu_rval[0];
 224                     regs->edx = uthread->uu_rval[1];
 225                 }
 226         }
 227
 228         DEBUG_KPRINT_SYSCALL_UNIX(
 229                 "unix_syscall: error=%d retval=(%u,%u)\n",
 230                 error, regs->eax, regs->edx);
 231
 232         uthread->uu_flag &= ~UT_NOTCANCELPT;
 233
 234         if (__improbable(uthread->uu_lowpri_window)) {
 235                 /*
 236                  * task is marked as a low priority I/O type
 237                  * and the I/O we issued while in this system call
 238                  * collided with normal I/O operations... we'll
 239                  * delay in order to mitigate the impact of this
 240                  * task on the normal operation of the system
 241                  */
 242                 throttle_lowpri_io(1);
 243         }
 244         if (__probable(!code_is_kdebug_trace(code)))
 245                 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 246                         BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
 247                         error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
 248
 249         if (__improbable(!is_vfork && callp->sy_call == (sy_call_t *)execve && !error)) {
 250                 pal_execve_return(thread);
 251         }
 252
 253         thread_exception_return();
 254         /* NOTREACHED */
 255 }
 256
 257
 258 void
 259 unix_syscall64(x86_saved_state_t *state)
 260 {
 261         thread_t        thread;
 262         void                    *vt;
 263         unsigned int    code;
 264         struct sysent   *callp;
 265         int             args_in_regs;
 266         boolean_t       args_start_at_rdi;
 267         int             error;
 268         struct proc     *p;
 269         struct uthread  *uthread;
 270         x86_saved_state64_t *regs;
 271
 272         assert(is_saved_state64(state));
 273         regs = saved_state64(state);
 274 #if     DEBUG
 275         if (regs->rax == 0x2000800)
 276                 thread_exception_return();
 277 #endif
 278         thread = current_thread();
 279         uthread = get_bsdthread_info(thread);
 280
 281         /* Get the approriate proc; may be different from task's for vfork() */
 282         if (__probable(!(uthread->uu_flag & UT_VFORK)))
 283                 p = (struct proc *)get_bsdtask_info(current_task());
 284         else
 285                 p = current_proc();
 286
 287         /* Verify that we are not being called from a task without a proc */
 288         if (__improbable(p == NULL)) {
 289                 regs->rax = EPERM;
 290                 regs->isf.rflags |= EFL_CF;
 291                 task_terminate_internal(current_task());
 292                 thread_exception_return();
 293                 /* NOTREACHED */
 294         }
 295
 296         code = regs->rax & SYSCALL_NUMBER_MASK;
 297         DEBUG_KPRINT_SYSCALL_UNIX(
 298                 "unix_syscall64: code=%d(%s) rip=%llx\n",
 299                 code, syscallnames[code >= NUM_SYSENT ? 63 : code], regs->isf.rip);
 300         callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 301
 302         vt = (void *)uthread->uu_arg;
 303
 304         if (__improbable(callp == sysent)) {
 305                 /*
 306                  * indirect system call... system call number
 307                  * passed as 'arg0'
 308                  */
 309                 code = regs->rdi;
 310                 callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 311                 args_start_at_rdi = FALSE;
 312                 args_in_regs = 5;
 313         } else {
 314                 args_start_at_rdi = TRUE;
 315                 args_in_regs = 6;
 316         }
 317
 318         if (callp->sy_narg != 0) {
 319                 assert(callp->sy_narg <= 8); /* size of uu_arg */
 320
 321                 args_in_regs = MIN(args_in_regs, callp->sy_narg);
 322                 memcpy(vt, args_start_at_rdi ? &regs->rdi : &regs->rsi, args_in_regs * sizeof(syscall_arg_t));
 323
 324
 325                 if (!code_is_kdebug_trace(code)) {
 326                         uint64_t *ip = (uint64_t *)vt;
 327
 328                         KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 329                                 BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
 330                                 (int)(*ip), (int)(*(ip+1)), (int)(*(ip+2)), (int)(*(ip+3)), 0);
 331                 }
 332
 333                 if (__improbable(callp->sy_narg > args_in_regs)) {
 334                         int copyin_count;
 335
 336                         copyin_count = (callp->sy_narg - args_in_regs) * sizeof(syscall_arg_t);
 337
 338                         error = copyin((user_addr_t)(regs->isf.rsp + sizeof(user_addr_t)), (char *)&uthread->uu_arg[args_in_regs], copyin_count);
 339                         if (error) {
 340                                 regs->rax = error;
 341                                 regs->isf.rflags |= EFL_CF;
 342                                 thread_exception_return();
 343                                 /* NOTREACHED */
 344                         }
 345                 }
 346         } else
 347                 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 348                         BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
 349                         0, 0, 0, 0, 0);
 350
 351         /*
 352          * Delayed binding of thread credential to process credential, if we
 353          * are not running with an explicitly set thread credential.
 354          */
 355         kauth_cred_uthread_update(uthread, p);
 356
 357         uthread->uu_rval[0] = 0;
 358         uthread->uu_rval[1] = 0;
 359         uthread->uu_flag |= UT_NOTCANCELPT;
 360         uthread->syscall_code = code;
 361
 362 #ifdef JOE_DEBUG
 363         uthread->uu_iocount = 0;
 364         uthread->uu_vpindex = 0;
 365 #endif
 366
 367         AUDIT_SYSCALL_ENTER(code, p, uthread);
 368         error = (*(callp->sy_call))((void *) p, vt, &(uthread->uu_rval[0]));
 369         AUDIT_SYSCALL_EXIT(code, p, uthread, error);
 370
 371 #ifdef JOE_DEBUG
 372         if (uthread->uu_iocount)
 373                printf("system call returned with uu_iocount != 0\n");
 374 #endif
 375
 376 #if CONFIG_DTRACE
 377         uthread->t_dtrace_errno = error;
 378 #endif /* CONFIG_DTRACE */
 379
 380         if (__improbable(error == ERESTART)) {
 381                 /*
 382                  * all system calls come through via the syscall instruction
 383                  * in 64 bit mode... its 2 bytes in length
 384                  * move the user's pc back to repeat the syscall:
 385                  */
 386                 pal_syscall_restart( thread, state );
 387         }
 388         else if (error != EJUSTRETURN) {
 389                 if (__improbable(error)) {
 390                         regs->rax = error;
 391                         regs->isf.rflags |= EFL_CF;     /* carry bit */
 392                 } else { /* (not error) */
 393
 394                         switch (callp->sy_return_type) {
 395                         case _SYSCALL_RET_INT_T:
 396                                 regs->rax = uthread->uu_rval[0];
 397                                 regs->rdx = uthread->uu_rval[1];
 398                                 break;
 399                         case _SYSCALL_RET_UINT_T:
 400                                 regs->rax = ((u_int)uthread->uu_rval[0]);
 401                                 regs->rdx = ((u_int)uthread->uu_rval[1]);
 402                                 break;
 403                         case _SYSCALL_RET_OFF_T:
 404                         case _SYSCALL_RET_ADDR_T:
 405                         case _SYSCALL_RET_SIZE_T:
 406                         case _SYSCALL_RET_SSIZE_T:
 407                         case _SYSCALL_RET_UINT64_T:
 408                                 regs->rax = *((uint64_t *)(&uthread->uu_rval[0]));
 409                                 regs->rdx = 0;
 410                                 break;
 411                         case _SYSCALL_RET_NONE:
 412                                 break;
 413                         default:
 414                                 panic("unix_syscall: unknown return type");
 415                                 break;
 416                         }
 417                         regs->isf.rflags &= ~EFL_CF;
 418                 }
 419         }
 420
 421         DEBUG_KPRINT_SYSCALL_UNIX(
 422                 "unix_syscall64: error=%d retval=(%llu,%llu)\n",
 423                 error, regs->rax, regs->rdx);
 424
 425         uthread->uu_flag &= ~UT_NOTCANCELPT;
 426
 427         if (__improbable(uthread->uu_lowpri_window)) {
 428                 /*
 429                  * task is marked as a low priority I/O type
 430                  * and the I/O we issued while in this system call
 431                  * collided with normal I/O operations... we'll
 432                  * delay in order to mitigate the impact of this
 433                  * task on the normal operation of the system
 434                  */
 435                 throttle_lowpri_io(1);
 436         }
 437         if (__probable(!code_is_kdebug_trace(code)))
 438                 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 439                         BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
 440                         error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
 441
 442         thread_exception_return();
 443         /* NOTREACHED */
 444 }
 445
 446
 447 void
 448 unix_syscall_return(int error)
 449 {
 450         thread_t                thread;
 451         struct uthread          *uthread;
 452         struct proc *p;
 453         unsigned int code;
 454         struct sysent *callp;
 455
 456         thread = current_thread();
 457         uthread = get_bsdthread_info(thread);
 458
 459         pal_register_cache_state(thread, DIRTY);
 460
 461         p = current_proc();
 462
 463         if (proc_is64bit(p)) {
 464                 x86_saved_state64_t *regs;
 465
 466                 regs = saved_state64(find_user_regs(thread));
 467
 468                 code = uthread->syscall_code;
 469                 callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 470
 471 #if CONFIG_DTRACE
 472                 if (callp->sy_call == dtrace_systrace_syscall)
 473                         dtrace_systrace_syscall_return( code, error, uthread->uu_rval );
 474 #endif /* CONFIG_DTRACE */
 475                 AUDIT_SYSCALL_EXIT(code, p, uthread, error);
 476
 477                 if (error == ERESTART) {
 478                         /*
 479                          * repeat the syscall
 480                          */
 481                         pal_syscall_restart( thread, find_user_regs(thread) );
 482                 }
 483                 else if (error != EJUSTRETURN) {
 484                         if (error) {
 485                                 regs->rax = error;
 486                                 regs->isf.rflags |= EFL_CF;     /* carry bit */
 487                         } else { /* (not error) */
 488
 489                                 switch (callp->sy_return_type) {
 490                                 case _SYSCALL_RET_INT_T:
 491                                         regs->rax = uthread->uu_rval[0];
 492                                         regs->rdx = uthread->uu_rval[1];
 493                                         break;
 494                                 case _SYSCALL_RET_UINT_T:
 495                                         regs->rax = ((u_int)uthread->uu_rval[0]);
 496                                         regs->rdx = ((u_int)uthread->uu_rval[1]);
 497                                         break;
 498                                 case _SYSCALL_RET_OFF_T:
 499                                 case _SYSCALL_RET_ADDR_T:
 500                                 case _SYSCALL_RET_SIZE_T:
 501                                 case _SYSCALL_RET_SSIZE_T:
 502                                 case _SYSCALL_RET_UINT64_T:
 503                                         regs->rax = *((uint64_t *)(&uthread->uu_rval[0]));
 504                                         regs->rdx = 0;
 505                                         break;
 506                                 case _SYSCALL_RET_NONE:
 507                                         break;
 508                                 default:
 509                                         panic("unix_syscall: unknown return type");
 510                                         break;
 511                                 }
 512                                 regs->isf.rflags &= ~EFL_CF;
 513                         }
 514                 }
 515                 DEBUG_KPRINT_SYSCALL_UNIX(
 516                         "unix_syscall_return: error=%d retval=(%llu,%llu)\n",
 517                         error, regs->rax, regs->rdx);
 518         } else {
 519                 x86_saved_state32_t     *regs;
 520
 521                 regs = saved_state32(find_user_regs(thread));
 522
 523                 regs->efl &= ~(EFL_CF);
 524
 525                 code = uthread->syscall_code;
 526                 callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
 527
 528 #if CONFIG_DTRACE
 529                 if (callp->sy_call == dtrace_systrace_syscall)
 530                         dtrace_systrace_syscall_return( code, error, uthread->uu_rval );
 531 #endif /* CONFIG_DTRACE */
 532                 AUDIT_SYSCALL_EXIT(code, p, uthread, error);
 533
 534                 if (error == ERESTART) {
 535                         pal_syscall_restart( thread, find_user_regs(thread) );
 536                 }
 537                 else if (error != EJUSTRETURN) {
 538                         if (error) {
 539                                 regs->eax = error;
 540                                 regs->efl |= EFL_CF;    /* carry bit */
 541                         } else { /* (not error) */
 542                                 regs->eax = uthread->uu_rval[0];
 543                                 regs->edx = uthread->uu_rval[1];
 544                         }
 545                 }
 546                 DEBUG_KPRINT_SYSCALL_UNIX(
 547                         "unix_syscall_return: error=%d retval=(%u,%u)\n",
 548                         error, regs->eax, regs->edx);
 549         }
 550
 551
 552         uthread->uu_flag &= ~UT_NOTCANCELPT;
 553
 554         if (uthread->uu_lowpri_window) {
 555                 /*
 556                  * task is marked as a low priority I/O type
 557                  * and the I/O we issued while in this system call
 558                  * collided with normal I/O operations... we'll
 559                  * delay in order to mitigate the impact of this
 560                  * task on the normal operation of the system
 561                  */
 562                 throttle_lowpri_io(1);
 563         }
 564         if (!code_is_kdebug_trace(code))
 565                 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 566                         BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
 567                         error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
 568
 569         thread_exception_return();
 570         /* NOTREACHED */
 571 }
 572