kernel/sys.c

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/mm.h>
   9 #include <linux/utsname.h>
  10 #include <linux/mman.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/notifier.h>
  13 #include <linux/reboot.h>
  14 #include <linux/prctl.h>
  15 #include <linux/highuid.h>
  16 #include <linux/fs.h>
  17 #include <linux/perf_event.h>
  18 #include <linux/resource.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kexec.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/capability.h>
  23 #include <linux/device.h>
  24 #include <linux/key.h>
  25 #include <linux/times.h>
  26 #include <linux/posix-timers.h>
  27 #include <linux/security.h>
  28 #include <linux/dcookies.h>
  29 #include <linux/suspend.h>
  30 #include <linux/tty.h>
  31 #include <linux/signal.h>
  32 #include <linux/cn_proc.h>
  33 #include <linux/getcpu.h>
  34 #include <linux/task_io_accounting_ops.h>
  35 #include <linux/seccomp.h>
  36 #include <linux/cpu.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/fs_struct.h>
  39
  40 #include <linux/compat.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/kprobes.h>
  43 #include <linux/user_namespace.h>
  44
  45 #include <asm/uaccess.h>
  46 #include <asm/io.h>
  47 #include <asm/unistd.h>
  48
  49 #ifndef SET_UNALIGN_CTL
  50 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  51 #endif
  52 #ifndef GET_UNALIGN_CTL
  53 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  54 #endif
  55 #ifndef SET_FPEMU_CTL
  56 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  57 #endif
  58 #ifndef GET_FPEMU_CTL
  59 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  60 #endif
  61 #ifndef SET_FPEXC_CTL
  62 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  63 #endif
  64 #ifndef GET_FPEXC_CTL
  65 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  66 #endif
  67 #ifndef GET_ENDIAN
  68 # define GET_ENDIAN(a,b)        (-EINVAL)
  69 #endif
  70 #ifndef SET_ENDIAN
  71 # define SET_ENDIAN(a,b)        (-EINVAL)
  72 #endif
  73 #ifndef GET_TSC_CTL
  74 # define GET_TSC_CTL(a)         (-EINVAL)
  75 #endif
  76 #ifndef SET_TSC_CTL
  77 # define SET_TSC_CTL(a)         (-EINVAL)
  78 #endif
  79
  80 /*
  81  * this is where the system-wide overflow UID and GID are defined, for
  82  * architectures that now have 32-bit UID/GID but didn't in the past
  83  */
  84
  85 int overflowuid = DEFAULT_OVERFLOWUID;
  86 int overflowgid = DEFAULT_OVERFLOWGID;
  87
  88 #ifdef CONFIG_UID16
  89 EXPORT_SYMBOL(overflowuid);
  90 EXPORT_SYMBOL(overflowgid);
  91 #endif
  92
  93 /*
  94  * the same as above, but for filesystems which can only store a 16-bit
  95  * UID and GID. as such, this is needed on all architectures
  96  */
  97
  98 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  99 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 100
 101 EXPORT_SYMBOL(fs_overflowuid);
 102 EXPORT_SYMBOL(fs_overflowgid);
 103
 104 /*
 105  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 106  */
 107
 108 int C_A_D = 1;
 109 struct pid *cad_pid;
 110 EXPORT_SYMBOL(cad_pid);
 111
 112 /*
 113  * If set, this is used for preparing the system to power off.
 114  */
 115
 116 void (*pm_power_off_prepare)(void);
 117
 118 /*
 119  * set the priority of a task
 120  * - the caller must hold the RCU read lock
 121  */
 122 static int set_one_prio(struct task_struct *p, int niceval, int error)
 123 {
 124         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 125         int no_nice;
 126
 127         if (pcred->uid  != cred->euid &&
 128             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
 129                 error = -EPERM;
 130                 goto out;
 131         }
 132         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 133                 error = -EACCES;
 134                 goto out;
 135         }
 136         no_nice = security_task_setnice(p, niceval);
 137         if (no_nice) {
 138                 error = no_nice;
 139                 goto out;
 140         }
 141         if (error == -ESRCH)
 142                 error = 0;
 143         set_user_nice(p, niceval);
 144 out:
 145         return error;
 146 }
 147
 148 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 149 {
 150         struct task_struct *g, *p;
 151         struct user_struct *user;
 152         const struct cred *cred = current_cred();
 153         int error = -EINVAL;
 154         struct pid *pgrp;
 155
 156         if (which > PRIO_USER || which < PRIO_PROCESS)
 157                 goto out;
 158
 159         /* normalize: avoid signed division (rounding problems) */
 160         error = -ESRCH;
 161         if (niceval < -20)
 162                 niceval = -20;
 163         if (niceval > 19)
 164                 niceval = 19;
 165
 166         read_lock(&tasklist_lock);
 167         switch (which) {
 168                 case PRIO_PROCESS:
 169                         if (who)
 170                                 p = find_task_by_vpid(who);
 171                         else
 172                                 p = current;
 173                         if (p)
 174                                 error = set_one_prio(p, niceval, error);
 175                         break;
 176                 case PRIO_PGRP:
 177                         if (who)
 178                                 pgrp = find_vpid(who);
 179                         else
 180                                 pgrp = task_pgrp(current);
 181                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 182                                 error = set_one_prio(p, niceval, error);
 183                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 184                         break;
 185                 case PRIO_USER:
 186                         user = (struct user_struct *) cred->user;
 187                         if (!who)
 188                                 who = cred->uid;
 189                         else if ((who != cred->uid) &&
 190                                  !(user = find_user(who)))
 191                                 goto out_unlock;        /* No processes for this user */
 192
 193                         do_each_thread(g, p)
 194                                 if (__task_cred(p)->uid == who)
 195                                         error = set_one_prio(p, niceval, error);
 196                         while_each_thread(g, p);
 197                         if (who != cred->uid)
 198                                 free_uid(user);         /* For find_user() */
 199                         break;
 200         }
 201 out_unlock:
 202         read_unlock(&tasklist_lock);
 203 out:
 204         return error;
 205 }
 206
 207 /*
 208  * Ugh. To avoid negative return values, "getpriority()" will
 209  * not return the normal nice-value, but a negated value that
 210  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 211  * to stay compatible.
 212  */
 213 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 214 {
 215         struct task_struct *g, *p;
 216         struct user_struct *user;
 217         const struct cred *cred = current_cred();
 218         long niceval, retval = -ESRCH;
 219         struct pid *pgrp;
 220
 221         if (which > PRIO_USER || which < PRIO_PROCESS)
 222                 return -EINVAL;
 223
 224         read_lock(&tasklist_lock);
 225         switch (which) {
 226                 case PRIO_PROCESS:
 227                         if (who)
 228                                 p = find_task_by_vpid(who);
 229                         else
 230                                 p = current;
 231                         if (p) {
 232                                 niceval = 20 - task_nice(p);
 233                                 if (niceval > retval)
 234                                         retval = niceval;
 235                         }
 236                         break;
 237                 case PRIO_PGRP:
 238                         if (who)
 239                                 pgrp = find_vpid(who);
 240                         else
 241                                 pgrp = task_pgrp(current);
 242                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 243                                 niceval = 20 - task_nice(p);
 244                                 if (niceval > retval)
 245                                         retval = niceval;
 246                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 247                         break;
 248                 case PRIO_USER:
 249                         user = (struct user_struct *) cred->user;
 250                         if (!who)
 251                                 who = cred->uid;
 252                         else if ((who != cred->uid) &&
 253                                  !(user = find_user(who)))
 254                                 goto out_unlock;        /* No processes for this user */
 255
 256                         do_each_thread(g, p)
 257                                 if (__task_cred(p)->uid == who) {
 258                                         niceval = 20 - task_nice(p);
 259                                         if (niceval > retval)
 260                                                 retval = niceval;
 261                                 }
 262                         while_each_thread(g, p);
 263                         if (who != cred->uid)
 264                                 free_uid(user);         /* for find_user() */
 265                         break;
 266         }
 267 out_unlock:
 268         read_unlock(&tasklist_lock);
 269
 270         return retval;
 271 }
 272
 273 /**
 274  *      emergency_restart - reboot the system
 275  *
 276  *      Without shutting down any hardware or taking any locks
 277  *      reboot the system.  This is called when we know we are in
 278  *      trouble so this is our best effort to reboot.  This is
 279  *      safe to call in interrupt context.
 280  */
 281 void emergency_restart(void)
 282 {
 283         machine_emergency_restart();
 284 }
 285 EXPORT_SYMBOL_GPL(emergency_restart);
 286
 287 void kernel_restart_prepare(char *cmd)
 288 {
 289         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 290         system_state = SYSTEM_RESTART;
 291         device_shutdown();
 292         sysdev_shutdown();
 293 }
 294
 295 /**
 296  *      kernel_restart - reboot the system
 297  *      @cmd: pointer to buffer containing command to execute for restart
 298  *              or %NULL
 299  *
 300  *      Shutdown everything and perform a clean reboot.
 301  *      This is not safe to call in interrupt context.
 302  */
 303 void kernel_restart(char *cmd)
 304 {
 305         kernel_restart_prepare(cmd);
 306         if (!cmd)
 307                 printk(KERN_EMERG "Restarting system.\n");
 308         else
 309                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 310         machine_restart(cmd);
 311 }
 312 EXPORT_SYMBOL_GPL(kernel_restart);
 313
 314 static void kernel_shutdown_prepare(enum system_states state)
 315 {
 316         blocking_notifier_call_chain(&reboot_notifier_list,
 317                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 318         system_state = state;
 319         device_shutdown();
 320 }
 321 /**
 322  *      kernel_halt - halt the system
 323  *
 324  *      Shutdown everything and perform a clean system halt.
 325  */
 326 void kernel_halt(void)
 327 {
 328         kernel_shutdown_prepare(SYSTEM_HALT);
 329         sysdev_shutdown();
 330         printk(KERN_EMERG "System halted.\n");
 331         machine_halt();
 332 }
 333
 334 EXPORT_SYMBOL_GPL(kernel_halt);
 335
 336 /**
 337  *      kernel_power_off - power_off the system
 338  *
 339  *      Shutdown everything and perform a clean system power_off.
 340  */
 341 void kernel_power_off(void)
 342 {
 343         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 344         if (pm_power_off_prepare)
 345                 pm_power_off_prepare();
 346         disable_nonboot_cpus();
 347         sysdev_shutdown();
 348         printk(KERN_EMERG "Power down.\n");
 349         machine_power_off();
 350 }
 351 EXPORT_SYMBOL_GPL(kernel_power_off);
 352 /*
 353  * Reboot system call: for obvious reasons only root may call it,
 354  * and even root needs to set up some magic numbers in the registers
 355  * so that some mistake won't make this reboot the whole machine.
 356  * You can also set the meaning of the ctrl-alt-del-key here.
 357  *
 358  * reboot doesn't sync: do that yourself before calling this.
 359  */
 360 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 361                 void __user *, arg)
 362 {
 363         char buffer[256];
 364         int ret = 0;
 365
 366         /* We only trust the superuser with rebooting the system. */
 367         if (!capable(CAP_SYS_BOOT))
 368                 return -EPERM;
 369
 370         /* For safety, we require "magic" arguments. */
 371         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 372             (magic2 != LINUX_REBOOT_MAGIC2 &&
 373                         magic2 != LINUX_REBOOT_MAGIC2A &&
 374                         magic2 != LINUX_REBOOT_MAGIC2B &&
 375                         magic2 != LINUX_REBOOT_MAGIC2C))
 376                 return -EINVAL;
 377
 378         /* Instead of trying to make the power_off code look like
 379          * halt when pm_power_off is not set do it the easy way.
 380          */
 381         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 382                 cmd = LINUX_REBOOT_CMD_HALT;
 383
 384         lock_kernel();
 385         switch (cmd) {
 386         case LINUX_REBOOT_CMD_RESTART:
 387                 kernel_restart(NULL);
 388                 break;
 389
 390         case LINUX_REBOOT_CMD_CAD_ON:
 391                 C_A_D = 1;
 392                 break;
 393
 394         case LINUX_REBOOT_CMD_CAD_OFF:
 395                 C_A_D = 0;
 396                 break;
 397
 398         case LINUX_REBOOT_CMD_HALT:
 399                 kernel_halt();
 400                 unlock_kernel();
 401                 do_exit(0);
 402                 panic("cannot halt");
 403
 404         case LINUX_REBOOT_CMD_POWER_OFF:
 405                 kernel_power_off();
 406                 unlock_kernel();
 407                 do_exit(0);
 408                 break;
 409
 410         case LINUX_REBOOT_CMD_RESTART2:
 411                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 412                         unlock_kernel();
 413                         return -EFAULT;
 414                 }
 415                 buffer[sizeof(buffer) - 1] = '\0';
 416
 417                 kernel_restart(buffer);
 418                 break;
 419
 420 #ifdef CONFIG_KEXEC
 421         case LINUX_REBOOT_CMD_KEXEC:
 422                 ret = kernel_kexec();
 423                 break;
 424 #endif
 425
 426 #ifdef CONFIG_HIBERNATION
 427         case LINUX_REBOOT_CMD_SW_SUSPEND:
 428                 ret = hibernate();
 429                 break;
 430 #endif
 431
 432         default:
 433                 ret = -EINVAL;
 434                 break;
 435         }
 436         unlock_kernel();
 437         return ret;
 438 }
 439
 440 static void deferred_cad(struct work_struct *dummy)
 441 {
 442         kernel_restart(NULL);
 443 }
 444
 445 /*
 446  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 447  * As it's called within an interrupt, it may NOT sync: the only choice
 448  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 449  */
 450 void ctrl_alt_del(void)
 451 {
 452         static DECLARE_WORK(cad_work, deferred_cad);
 453
 454         if (C_A_D)
 455                 schedule_work(&cad_work);
 456         else
 457                 kill_cad_pid(SIGINT, 1);
 458 }
 459
 460 /*
 461  * Unprivileged users may change the real gid to the effective gid
 462  * or vice versa.  (BSD-style)
 463  *
 464  * If you set the real gid at all, or set the effective gid to a value not
 465  * equal to the real gid, then the saved gid is set to the new effective gid.
 466  *
 467  * This makes it possible for a setgid program to completely drop its
 468  * privileges, which is often a useful assertion to make when you are doing
 469  * a security audit over a program.
 470  *
 471  * The general idea is that a program which uses just setregid() will be
 472  * 100% compatible with BSD.  A program which uses just setgid() will be
 473  * 100% compatible with POSIX with saved IDs.
 474  *
 475  * SMP: There are not races, the GIDs are checked only by filesystem
 476  *      operations (as far as semantic preservation is concerned).
 477  */
 478 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 479 {
 480         const struct cred *old;
 481         struct cred *new;
 482         int retval;
 483
 484         new = prepare_creds();
 485         if (!new)
 486                 return -ENOMEM;
 487         old = current_cred();
 488
 489         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 490         if (retval)
 491                 goto error;
 492
 493         retval = -EPERM;
 494         if (rgid != (gid_t) -1) {
 495                 if (old->gid == rgid ||
 496                     old->egid == rgid ||
 497                     capable(CAP_SETGID))
 498                         new->gid = rgid;
 499                 else
 500                         goto error;
 501         }
 502         if (egid != (gid_t) -1) {
 503                 if (old->gid == egid ||
 504                     old->egid == egid ||
 505                     old->sgid == egid ||
 506                     capable(CAP_SETGID))
 507                         new->egid = egid;
 508                 else
 509                         goto error;
 510         }
 511
 512         if (rgid != (gid_t) -1 ||
 513             (egid != (gid_t) -1 && egid != old->gid))
 514                 new->sgid = new->egid;
 515         new->fsgid = new->egid;
 516
 517         return commit_creds(new);
 518
 519 error:
 520         abort_creds(new);
 521         return retval;
 522 }
 523
 524 /*
 525  * setgid() is implemented like SysV w/ SAVED_IDS
 526  *
 527  * SMP: Same implicit races as above.
 528  */
 529 SYSCALL_DEFINE1(setgid, gid_t, gid)
 530 {
 531         const struct cred *old;
 532         struct cred *new;
 533         int retval;
 534
 535         new = prepare_creds();
 536         if (!new)
 537                 return -ENOMEM;
 538         old = current_cred();
 539
 540         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 541         if (retval)
 542                 goto error;
 543
 544         retval = -EPERM;
 545         if (capable(CAP_SETGID))
 546                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 547         else if (gid == old->gid || gid == old->sgid)
 548                 new->egid = new->fsgid = gid;
 549         else
 550                 goto error;
 551
 552         return commit_creds(new);
 553
 554 error:
 555         abort_creds(new);
 556         return retval;
 557 }
 558
 559 /*
 560  * change the user struct in a credentials set to match the new UID
 561  */
 562 static int set_user(struct cred *new)
 563 {
 564         struct user_struct *new_user;
 565
 566         new_user = alloc_uid(current_user_ns(), new->uid);
 567         if (!new_user)
 568                 return -EAGAIN;
 569
 570         if (atomic_read(&new_user->processes) >=
 571                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 572                         new_user != INIT_USER) {
 573                 free_uid(new_user);
 574                 return -EAGAIN;
 575         }
 576
 577         free_uid(new->user);
 578         new->user = new_user;
 579         return 0;
 580 }
 581
 582 /*
 583  * Unprivileged users may change the real uid to the effective uid
 584  * or vice versa.  (BSD-style)
 585  *
 586  * If you set the real uid at all, or set the effective uid to a value not
 587  * equal to the real uid, then the saved uid is set to the new effective uid.
 588  *
 589  * This makes it possible for a setuid program to completely drop its
 590  * privileges, which is often a useful assertion to make when you are doing
 591  * a security audit over a program.
 592  *
 593  * The general idea is that a program which uses just setreuid() will be
 594  * 100% compatible with BSD.  A program which uses just setuid() will be
 595  * 100% compatible with POSIX with saved IDs.
 596  */
 597 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 598 {
 599         const struct cred *old;
 600         struct cred *new;
 601         int retval;
 602
 603         new = prepare_creds();
 604         if (!new)
 605                 return -ENOMEM;
 606         old = current_cred();
 607
 608         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 609         if (retval)
 610                 goto error;
 611
 612         retval = -EPERM;
 613         if (ruid != (uid_t) -1) {
 614                 new->uid = ruid;
 615                 if (old->uid != ruid &&
 616                     old->euid != ruid &&
 617                     !capable(CAP_SETUID))
 618                         goto error;
 619         }
 620
 621         if (euid != (uid_t) -1) {
 622                 new->euid = euid;
 623                 if (old->uid != euid &&
 624                     old->euid != euid &&
 625                     old->suid != euid &&
 626                     !capable(CAP_SETUID))
 627                         goto error;
 628         }
 629
 630         if (new->uid != old->uid) {
 631                 retval = set_user(new);
 632                 if (retval < 0)
 633                         goto error;
 634         }
 635         if (ruid != (uid_t) -1 ||
 636             (euid != (uid_t) -1 && euid != old->uid))
 637                 new->suid = new->euid;
 638         new->fsuid = new->euid;
 639
 640         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 641         if (retval < 0)
 642                 goto error;
 643
 644         return commit_creds(new);
 645
 646 error:
 647         abort_creds(new);
 648         return retval;
 649 }
 650
 651 /*
 652  * setuid() is implemented like SysV with SAVED_IDS
 653  *
 654  * Note that SAVED_ID's is deficient in that a setuid root program
 655  * like sendmail, for example, cannot set its uid to be a normal
 656  * user and then switch back, because if you're root, setuid() sets
 657  * the saved uid too.  If you don't like this, blame the bright people
 658  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 659  * will allow a root program to temporarily drop privileges and be able to
 660  * regain them by swapping the real and effective uid.
 661  */
 662 SYSCALL_DEFINE1(setuid, uid_t, uid)
 663 {
 664         const struct cred *old;
 665         struct cred *new;
 666         int retval;
 667
 668         new = prepare_creds();
 669         if (!new)
 670                 return -ENOMEM;
 671         old = current_cred();
 672
 673         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 674         if (retval)
 675                 goto error;
 676
 677         retval = -EPERM;
 678         if (capable(CAP_SETUID)) {
 679                 new->suid = new->uid = uid;
 680                 if (uid != old->uid) {
 681                         retval = set_user(new);
 682                         if (retval < 0)
 683                                 goto error;
 684                 }
 685         } else if (uid != old->uid && uid != new->suid) {
 686                 goto error;
 687         }
 688
 689         new->fsuid = new->euid = uid;
 690
 691         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 692         if (retval < 0)
 693                 goto error;
 694
 695         return commit_creds(new);
 696
 697 error:
 698         abort_creds(new);
 699         return retval;
 700 }
 701
 702
 703 /*
 704  * This function implements a generic ability to update ruid, euid,
 705  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 706  */
 707 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 708 {
 709         const struct cred *old;
 710         struct cred *new;
 711         int retval;
 712
 713         new = prepare_creds();
 714         if (!new)
 715                 return -ENOMEM;
 716
 717         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 718         if (retval)
 719                 goto error;
 720         old = current_cred();
 721
 722         retval = -EPERM;
 723         if (!capable(CAP_SETUID)) {
 724                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 725                     ruid != old->euid  && ruid != old->suid)
 726                         goto error;
 727                 if (euid != (uid_t) -1 && euid != old->uid &&
 728                     euid != old->euid  && euid != old->suid)
 729                         goto error;
 730                 if (suid != (uid_t) -1 && suid != old->uid &&
 731                     suid != old->euid  && suid != old->suid)
 732                         goto error;
 733         }
 734
 735         if (ruid != (uid_t) -1) {
 736                 new->uid = ruid;
 737                 if (ruid != old->uid) {
 738                         retval = set_user(new);
 739                         if (retval < 0)
 740                                 goto error;
 741                 }
 742         }
 743         if (euid != (uid_t) -1)
 744                 new->euid = euid;
 745         if (suid != (uid_t) -1)
 746                 new->suid = suid;
 747         new->fsuid = new->euid;
 748
 749         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 750         if (retval < 0)
 751                 goto error;
 752
 753         return commit_creds(new);
 754
 755 error:
 756         abort_creds(new);
 757         return retval;
 758 }
 759
 760 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 761 {
 762         const struct cred *cred = current_cred();
 763         int retval;
 764
 765         if (!(retval   = put_user(cred->uid,  ruid)) &&
 766             !(retval   = put_user(cred->euid, euid)))
 767                 retval = put_user(cred->suid, suid);
 768
 769         return retval;
 770 }
 771
 772 /*
 773  * Same as above, but for rgid, egid, sgid.
 774  */
 775 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 776 {
 777         const struct cred *old;
 778         struct cred *new;
 779         int retval;
 780
 781         new = prepare_creds();
 782         if (!new)
 783                 return -ENOMEM;
 784         old = current_cred();
 785
 786         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 787         if (retval)
 788                 goto error;
 789
 790         retval = -EPERM;
 791         if (!capable(CAP_SETGID)) {
 792                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 793                     rgid != old->egid  && rgid != old->sgid)
 794                         goto error;
 795                 if (egid != (gid_t) -1 && egid != old->gid &&
 796                     egid != old->egid  && egid != old->sgid)
 797                         goto error;
 798                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 799                     sgid != old->egid  && sgid != old->sgid)
 800                         goto error;
 801         }
 802
 803         if (rgid != (gid_t) -1)
 804                 new->gid = rgid;
 805         if (egid != (gid_t) -1)
 806                 new->egid = egid;
 807         if (sgid != (gid_t) -1)
 808                 new->sgid = sgid;
 809         new->fsgid = new->egid;
 810
 811         return commit_creds(new);
 812
 813 error:
 814         abort_creds(new);
 815         return retval;
 816 }
 817
 818 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 819 {
 820         const struct cred *cred = current_cred();
 821         int retval;
 822
 823         if (!(retval   = put_user(cred->gid,  rgid)) &&
 824             !(retval   = put_user(cred->egid, egid)))
 825                 retval = put_user(cred->sgid, sgid);
 826
 827         return retval;
 828 }
 829
 830
 831 /*
 832  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 833  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 834  * whatever uid it wants to). It normally shadows "euid", except when
 835  * explicitly set by setfsuid() or for access..
 836  */
 837 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 838 {
 839         const struct cred *old;
 840         struct cred *new;
 841         uid_t old_fsuid;
 842
 843         new = prepare_creds();
 844         if (!new)
 845                 return current_fsuid();
 846         old = current_cred();
 847         old_fsuid = old->fsuid;
 848
 849         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
 850                 goto error;
 851
 852         if (uid == old->uid  || uid == old->euid  ||
 853             uid == old->suid || uid == old->fsuid ||
 854             capable(CAP_SETUID)) {
 855                 if (uid != old_fsuid) {
 856                         new->fsuid = uid;
 857                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 858                                 goto change_okay;
 859                 }
 860         }
 861
 862 error:
 863         abort_creds(new);
 864         return old_fsuid;
 865
 866 change_okay:
 867         commit_creds(new);
 868         return old_fsuid;
 869 }
 870
 871 /*
 872  * Samma på svenska..
 873  */
 874 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 875 {
 876         const struct cred *old;
 877         struct cred *new;
 878         gid_t old_fsgid;
 879
 880         new = prepare_creds();
 881         if (!new)
 882                 return current_fsgid();
 883         old = current_cred();
 884         old_fsgid = old->fsgid;
 885
 886         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 887                 goto error;
 888
 889         if (gid == old->gid  || gid == old->egid  ||
 890             gid == old->sgid || gid == old->fsgid ||
 891             capable(CAP_SETGID)) {
 892                 if (gid != old_fsgid) {
 893                         new->fsgid = gid;
 894                         goto change_okay;
 895                 }
 896         }
 897
 898 error:
 899         abort_creds(new);
 900         return old_fsgid;
 901
 902 change_okay:
 903         commit_creds(new);
 904         return old_fsgid;
 905 }
 906
 907 void do_sys_times(struct tms *tms)
 908 {
 909         cputime_t tgutime, tgstime, cutime, cstime;
 910
 911         spin_lock_irq(&current->sighand->siglock);
 912         thread_group_times(current, &tgutime, &tgstime);
 913         cutime = current->signal->cutime;
 914         cstime = current->signal->cstime;
 915         spin_unlock_irq(&current->sighand->siglock);
 916         tms->tms_utime = cputime_to_clock_t(tgutime);
 917         tms->tms_stime = cputime_to_clock_t(tgstime);
 918         tms->tms_cutime = cputime_to_clock_t(cutime);
 919         tms->tms_cstime = cputime_to_clock_t(cstime);
 920 }
 921
 922 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 923 {
 924         if (tbuf) {
 925                 struct tms tmp;
 926
 927                 do_sys_times(&tmp);
 928                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 929                         return -EFAULT;
 930         }
 931         force_successful_syscall_return();
 932         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 933 }
 934
 935 /*
 936  * This needs some heavy checking ...
 937  * I just haven't the stomach for it. I also don't fully
 938  * understand sessions/pgrp etc. Let somebody who does explain it.
 939  *
 940  * OK, I think I have the protection semantics right.... this is really
 941  * only important on a multi-user system anyway, to make sure one user
 942  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 943  *
 944  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 945  * LBT 04.03.94
 946  */
 947 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 948 {
 949         struct task_struct *p;
 950         struct task_struct *group_leader = current->group_leader;
 951         struct pid *pgrp;
 952         int err;
 953
 954         if (!pid)
 955                 pid = task_pid_vnr(group_leader);
 956         if (!pgid)
 957                 pgid = pid;
 958         if (pgid < 0)
 959                 return -EINVAL;
 960         rcu_read_lock();
 961
 962         /* From this point forward we keep holding onto the tasklist lock
 963          * so that our parent does not change from under us. -DaveM
 964          */
 965         write_lock_irq(&tasklist_lock);
 966
 967         err = -ESRCH;
 968         p = find_task_by_vpid(pid);
 969         if (!p)
 970                 goto out;
 971
 972         err = -EINVAL;
 973         if (!thread_group_leader(p))
 974                 goto out;
 975
 976         if (same_thread_group(p->real_parent, group_leader)) {
 977                 err = -EPERM;
 978                 if (task_session(p) != task_session(group_leader))
 979                         goto out;
 980                 err = -EACCES;
 981                 if (p->did_exec)
 982                         goto out;
 983         } else {
 984                 err = -ESRCH;
 985                 if (p != group_leader)
 986                         goto out;
 987         }
 988
 989         err = -EPERM;
 990         if (p->signal->leader)
 991                 goto out;
 992
 993         pgrp = task_pid(p);
 994         if (pgid != pid) {
 995                 struct task_struct *g;
 996
 997                 pgrp = find_vpid(pgid);
 998                 g = pid_task(pgrp, PIDTYPE_PGID);
 999                 if (!g || task_session(g) != task_session(group_leader))
1000                         goto out;
1001         }
1002
1003         err = security_task_setpgid(p, pgid);
1004         if (err)
1005                 goto out;
1006
1007         if (task_pgrp(p) != pgrp)
1008                 change_pid(p, PIDTYPE_PGID, pgrp);
1009
1010         err = 0;
1011 out:
1012         /* All paths lead to here, thus we are safe. -DaveM */
1013         write_unlock_irq(&tasklist_lock);
1014         rcu_read_unlock();
1015         return err;
1016 }
1017
1018 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1019 {
1020         struct task_struct *p;
1021         struct pid *grp;
1022         int retval;
1023
1024         rcu_read_lock();
1025         if (!pid)
1026                 grp = task_pgrp(current);
1027         else {
1028                 retval = -ESRCH;
1029                 p = find_task_by_vpid(pid);
1030                 if (!p)
1031                         goto out;
1032                 grp = task_pgrp(p);
1033                 if (!grp)
1034                         goto out;
1035
1036                 retval = security_task_getpgid(p);
1037                 if (retval)
1038                         goto out;
1039         }
1040         retval = pid_vnr(grp);
1041 out:
1042         rcu_read_unlock();
1043         return retval;
1044 }
1045
1046 #ifdef __ARCH_WANT_SYS_GETPGRP
1047
1048 SYSCALL_DEFINE0(getpgrp)
1049 {
1050         return sys_getpgid(0);
1051 }
1052
1053 #endif
1054
1055 SYSCALL_DEFINE1(getsid, pid_t, pid)
1056 {
1057         struct task_struct *p;
1058         struct pid *sid;
1059         int retval;
1060
1061         rcu_read_lock();
1062         if (!pid)
1063                 sid = task_session(current);
1064         else {
1065                 retval = -ESRCH;
1066                 p = find_task_by_vpid(pid);
1067                 if (!p)
1068                         goto out;
1069                 sid = task_session(p);
1070                 if (!sid)
1071                         goto out;
1072
1073                 retval = security_task_getsid(p);
1074                 if (retval)
1075                         goto out;
1076         }
1077         retval = pid_vnr(sid);
1078 out:
1079         rcu_read_unlock();
1080         return retval;
1081 }
1082
1083 SYSCALL_DEFINE0(setsid)
1084 {
1085         struct task_struct *group_leader = current->group_leader;
1086         struct pid *sid = task_pid(group_leader);
1087         pid_t session = pid_vnr(sid);
1088         int err = -EPERM;
1089
1090         write_lock_irq(&tasklist_lock);
1091         /* Fail if I am already a session leader */
1092         if (group_leader->signal->leader)
1093                 goto out;
1094
1095         /* Fail if a process group id already exists that equals the
1096          * proposed session id.
1097          */
1098         if (pid_task(sid, PIDTYPE_PGID))
1099                 goto out;
1100
1101         group_leader->signal->leader = 1;
1102         __set_special_pids(sid);
1103
1104         proc_clear_tty(group_leader);
1105
1106         err = session;
1107 out:
1108         write_unlock_irq(&tasklist_lock);
1109         if (err > 0)
1110                 proc_sid_connector(group_leader);
1111         return err;
1112 }
1113
1114 DECLARE_RWSEM(uts_sem);
1115
1116 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1117 {
1118         int errno = 0;
1119
1120         down_read(&uts_sem);
1121         if (copy_to_user(name, utsname(), sizeof *name))
1122                 errno = -EFAULT;
1123         up_read(&uts_sem);
1124         return errno;
1125 }
1126
1127 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1128 {
1129         int errno;
1130         char tmp[__NEW_UTS_LEN];
1131
1132         if (!capable(CAP_SYS_ADMIN))
1133                 return -EPERM;
1134         if (len < 0 || len > __NEW_UTS_LEN)
1135                 return -EINVAL;
1136         down_write(&uts_sem);
1137         errno = -EFAULT;
1138         if (!copy_from_user(tmp, name, len)) {
1139                 struct new_utsname *u = utsname();
1140
1141                 memcpy(u->nodename, tmp, len);
1142                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1143                 errno = 0;
1144         }
1145         up_write(&uts_sem);
1146         return errno;
1147 }
1148
1149 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1150
1151 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1152 {
1153         int i, errno;
1154         struct new_utsname *u;
1155
1156         if (len < 0)
1157                 return -EINVAL;
1158         down_read(&uts_sem);
1159         u = utsname();
1160         i = 1 + strlen(u->nodename);
1161         if (i > len)
1162                 i = len;
1163         errno = 0;
1164         if (copy_to_user(name, u->nodename, i))
1165                 errno = -EFAULT;
1166         up_read(&uts_sem);
1167         return errno;
1168 }
1169
1170 #endif
1171
1172 /*
1173  * Only setdomainname; getdomainname can be implemented by calling
1174  * uname()
1175  */
1176 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1177 {
1178         int errno;
1179         char tmp[__NEW_UTS_LEN];
1180
1181         if (!capable(CAP_SYS_ADMIN))
1182                 return -EPERM;
1183         if (len < 0 || len > __NEW_UTS_LEN)
1184                 return -EINVAL;
1185
1186         down_write(&uts_sem);
1187         errno = -EFAULT;
1188         if (!copy_from_user(tmp, name, len)) {
1189                 struct new_utsname *u = utsname();
1190
1191                 memcpy(u->domainname, tmp, len);
1192                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1193                 errno = 0;
1194         }
1195         up_write(&uts_sem);
1196         return errno;
1197 }
1198
1199 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1200 {
1201         if (resource >= RLIM_NLIMITS)
1202                 return -EINVAL;
1203         else {
1204                 struct rlimit value;
1205                 task_lock(current->group_leader);
1206                 value = current->signal->rlim[resource];
1207                 task_unlock(current->group_leader);
1208                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1209         }
1210 }
1211
1212 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1213
1214 /*
1215  *      Back compatibility for getrlimit. Needed for some apps.
1216  */
1217
1218 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1219                 struct rlimit __user *, rlim)
1220 {
1221         struct rlimit x;
1222         if (resource >= RLIM_NLIMITS)
1223                 return -EINVAL;
1224
1225         task_lock(current->group_leader);
1226         x = current->signal->rlim[resource];
1227         task_unlock(current->group_leader);
1228         if (x.rlim_cur > 0x7FFFFFFF)
1229                 x.rlim_cur = 0x7FFFFFFF;
1230         if (x.rlim_max > 0x7FFFFFFF)
1231                 x.rlim_max = 0x7FFFFFFF;
1232         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1233 }
1234
1235 #endif
1236
1237 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1238 {
1239         struct rlimit new_rlim, *old_rlim;
1240         int retval;
1241
1242         if (resource >= RLIM_NLIMITS)
1243                 return -EINVAL;
1244         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1245                 return -EFAULT;
1246         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1247                 return -EINVAL;
1248         old_rlim = current->signal->rlim + resource;
1249         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1250             !capable(CAP_SYS_RESOURCE))
1251                 return -EPERM;
1252         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1253                 return -EPERM;
1254
1255         retval = security_task_setrlimit(resource, &new_rlim);
1256         if (retval)
1257                 return retval;
1258
1259         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1260                 /*
1261                  * The caller is asking for an immediate RLIMIT_CPU
1262                  * expiry.  But we use the zero value to mean "it was
1263                  * never set".  So let's cheat and make it one second
1264                  * instead
1265                  */
1266                 new_rlim.rlim_cur = 1;
1267         }
1268
1269         task_lock(current->group_leader);
1270         *old_rlim = new_rlim;
1271         task_unlock(current->group_leader);
1272
1273         if (resource != RLIMIT_CPU)
1274                 goto out;
1275
1276         /*
1277          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1278          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1279          * very long-standing error, and fixing it now risks breakage of
1280          * applications, so we live with it
1281          */
1282         if (new_rlim.rlim_cur == RLIM_INFINITY)
1283                 goto out;
1284
1285         update_rlimit_cpu(new_rlim.rlim_cur);
1286 out:
1287         return 0;
1288 }
1289
1290 /*
1291  * It would make sense to put struct rusage in the task_struct,
1292  * except that would make the task_struct be *really big*.  After
1293  * task_struct gets moved into malloc'ed memory, it would
1294  * make sense to do this.  It will make moving the rest of the information
1295  * a lot simpler!  (Which we're not doing right now because we're not
1296  * measuring them yet).
1297  *
1298  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1299  * races with threads incrementing their own counters.  But since word
1300  * reads are atomic, we either get new values or old values and we don't
1301  * care which for the sums.  We always take the siglock to protect reading
1302  * the c* fields from p->signal from races with exit.c updating those
1303  * fields when reaping, so a sample either gets all the additions of a
1304  * given child after it's reaped, or none so this sample is before reaping.
1305  *
1306  * Locking:
1307  * We need to take the siglock for CHILDEREN, SELF and BOTH
1308  * for  the cases current multithreaded, non-current single threaded
1309  * non-current multithreaded.  Thread traversal is now safe with
1310  * the siglock held.
1311  * Strictly speaking, we donot need to take the siglock if we are current and
1312  * single threaded,  as no one else can take our signal_struct away, no one
1313  * else can  reap the  children to update signal->c* counters, and no one else
1314  * can race with the signal-> fields. If we do not take any lock, the
1315  * signal-> fields could be read out of order while another thread was just
1316  * exiting. So we should  place a read memory barrier when we avoid the lock.
1317  * On the writer side,  write memory barrier is implied in  __exit_signal
1318  * as __exit_signal releases  the siglock spinlock after updating the signal->
1319  * fields. But we don't do this yet to keep things simple.
1320  *
1321  */
1322
1323 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1324 {
1325         r->ru_nvcsw += t->nvcsw;
1326         r->ru_nivcsw += t->nivcsw;
1327         r->ru_minflt += t->min_flt;
1328         r->ru_majflt += t->maj_flt;
1329         r->ru_inblock += task_io_get_inblock(t);
1330         r->ru_oublock += task_io_get_oublock(t);
1331 }
1332
1333 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1334 {
1335         struct task_struct *t;
1336         unsigned long flags;
1337         cputime_t tgutime, tgstime, utime, stime;
1338         unsigned long maxrss = 0;
1339
1340         memset((char *) r, 0, sizeof *r);
1341         utime = stime = cputime_zero;
1342
1343         if (who == RUSAGE_THREAD) {
1344                 utime = task_utime(current);
1345                 stime = task_stime(current);
1346                 accumulate_thread_rusage(p, r);
1347                 maxrss = p->signal->maxrss;
1348                 goto out;
1349         }
1350
1351         if (!lock_task_sighand(p, &flags))
1352                 return;
1353
1354         switch (who) {
1355                 case RUSAGE_BOTH:
1356                 case RUSAGE_CHILDREN:
1357                         utime = p->signal->cutime;
1358                         stime = p->signal->cstime;
1359                         r->ru_nvcsw = p->signal->cnvcsw;
1360                         r->ru_nivcsw = p->signal->cnivcsw;
1361                         r->ru_minflt = p->signal->cmin_flt;
1362                         r->ru_majflt = p->signal->cmaj_flt;
1363                         r->ru_inblock = p->signal->cinblock;
1364                         r->ru_oublock = p->signal->coublock;
1365                         maxrss = p->signal->cmaxrss;
1366
1367                         if (who == RUSAGE_CHILDREN)
1368                                 break;
1369
1370                 case RUSAGE_SELF:
1371                         thread_group_times(p, &tgutime, &tgstime);
1372                         utime = cputime_add(utime, tgutime);
1373                         stime = cputime_add(stime, tgstime);
1374                         r->ru_nvcsw += p->signal->nvcsw;
1375                         r->ru_nivcsw += p->signal->nivcsw;
1376                         r->ru_minflt += p->signal->min_flt;
1377                         r->ru_majflt += p->signal->maj_flt;
1378                         r->ru_inblock += p->signal->inblock;
1379                         r->ru_oublock += p->signal->oublock;
1380                         if (maxrss < p->signal->maxrss)
1381                                 maxrss = p->signal->maxrss;
1382                         t = p;
1383                         do {
1384                                 accumulate_thread_rusage(t, r);
1385                                 t = next_thread(t);
1386                         } while (t != p);
1387                         break;
1388
1389                 default:
1390                         BUG();
1391         }
1392         unlock_task_sighand(p, &flags);
1393
1394 out:
1395         cputime_to_timeval(utime, &r->ru_utime);
1396         cputime_to_timeval(stime, &r->ru_stime);
1397
1398         if (who != RUSAGE_CHILDREN) {
1399                 struct mm_struct *mm = get_task_mm(p);
1400                 if (mm) {
1401                         setmax_mm_hiwater_rss(&maxrss, mm);
1402                         mmput(mm);
1403                 }
1404         }
1405         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1406 }
1407
1408 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1409 {
1410         struct rusage r;
1411         k_getrusage(p, who, &r);
1412         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1413 }
1414
1415 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1416 {
1417         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1418             who != RUSAGE_THREAD)
1419                 return -EINVAL;
1420         return getrusage(current, who, ru);
1421 }
1422
1423 SYSCALL_DEFINE1(umask, int, mask)
1424 {
1425         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1426         return mask;
1427 }
1428
1429 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1430                 unsigned long, arg4, unsigned long, arg5)
1431 {
1432         struct task_struct *me = current;
1433         unsigned char comm[sizeof(me->comm)];
1434         long error;
1435
1436         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1437         if (error != -ENOSYS)
1438                 return error;
1439
1440         error = 0;
1441         switch (option) {
1442                 case PR_SET_PDEATHSIG:
1443                         if (!valid_signal(arg2)) {
1444                                 error = -EINVAL;
1445                                 break;
1446                         }
1447                         me->pdeath_signal = arg2;
1448                         error = 0;
1449                         break;
1450                 case PR_GET_PDEATHSIG:
1451                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1452                         break;
1453                 case PR_GET_DUMPABLE:
1454                         error = get_dumpable(me->mm);
1455                         break;
1456                 case PR_SET_DUMPABLE:
1457                         if (arg2 < 0 || arg2 > 1) {
1458                                 error = -EINVAL;
1459                                 break;
1460                         }
1461                         set_dumpable(me->mm, arg2);
1462                         error = 0;
1463                         break;
1464
1465                 case PR_SET_UNALIGN:
1466                         error = SET_UNALIGN_CTL(me, arg2);
1467                         break;
1468                 case PR_GET_UNALIGN:
1469                         error = GET_UNALIGN_CTL(me, arg2);
1470                         break;
1471                 case PR_SET_FPEMU:
1472                         error = SET_FPEMU_CTL(me, arg2);
1473                         break;
1474                 case PR_GET_FPEMU:
1475                         error = GET_FPEMU_CTL(me, arg2);
1476                         break;
1477                 case PR_SET_FPEXC:
1478                         error = SET_FPEXC_CTL(me, arg2);
1479                         break;
1480                 case PR_GET_FPEXC:
1481                         error = GET_FPEXC_CTL(me, arg2);
1482                         break;
1483                 case PR_GET_TIMING:
1484                         error = PR_TIMING_STATISTICAL;
1485                         break;
1486                 case PR_SET_TIMING:
1487                         if (arg2 != PR_TIMING_STATISTICAL)
1488                                 error = -EINVAL;
1489                         else
1490                                 error = 0;
1491                         break;
1492
1493                 case PR_SET_NAME:
1494                         comm[sizeof(me->comm)-1] = 0;
1495                         if (strncpy_from_user(comm, (char __user *)arg2,
1496                                               sizeof(me->comm) - 1) < 0)
1497                                 return -EFAULT;
1498                         set_task_comm(me, comm);
1499                         return 0;
1500                 case PR_GET_NAME:
1501                         get_task_comm(comm, me);
1502                         if (copy_to_user((char __user *)arg2, comm,
1503                                          sizeof(comm)))
1504                                 return -EFAULT;
1505                         return 0;
1506                 case PR_GET_ENDIAN:
1507                         error = GET_ENDIAN(me, arg2);
1508                         break;
1509                 case PR_SET_ENDIAN:
1510                         error = SET_ENDIAN(me, arg2);
1511                         break;
1512
1513                 case PR_GET_SECCOMP:
1514                         error = prctl_get_seccomp();
1515                         break;
1516                 case PR_SET_SECCOMP:
1517                         error = prctl_set_seccomp(arg2);
1518                         break;
1519                 case PR_GET_TSC:
1520                         error = GET_TSC_CTL(arg2);
1521                         break;
1522                 case PR_SET_TSC:
1523                         error = SET_TSC_CTL(arg2);
1524                         break;
1525                 case PR_TASK_PERF_EVENTS_DISABLE:
1526                         error = perf_event_task_disable();
1527                         break;
1528                 case PR_TASK_PERF_EVENTS_ENABLE:
1529                         error = perf_event_task_enable();
1530                         break;
1531                 case PR_GET_TIMERSLACK:
1532                         error = current->timer_slack_ns;
1533                         break;
1534                 case PR_SET_TIMERSLACK:
1535                         if (arg2 <= 0)
1536                                 current->timer_slack_ns =
1537                                         current->default_timer_slack_ns;
1538                         else
1539                                 current->timer_slack_ns = arg2;
1540                         error = 0;
1541                         break;
1542                 case PR_MCE_KILL:
1543                         if (arg4 | arg5)
1544                                 return -EINVAL;
1545                         switch (arg2) {
1546                         case PR_MCE_KILL_CLEAR:
1547                                 if (arg3 != 0)
1548                                         return -EINVAL;
1549                                 current->flags &= ~PF_MCE_PROCESS;
1550                                 break;
1551                         case PR_MCE_KILL_SET:
1552                                 current->flags |= PF_MCE_PROCESS;
1553                                 if (arg3 == PR_MCE_KILL_EARLY)
1554                                         current->flags |= PF_MCE_EARLY;
1555                                 else if (arg3 == PR_MCE_KILL_LATE)
1556                                         current->flags &= ~PF_MCE_EARLY;
1557                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1558                                         current->flags &=
1559                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1560                                 else
1561                                         return -EINVAL;
1562                                 break;
1563                         default:
1564                                 return -EINVAL;
1565                         }
1566                         error = 0;
1567                         break;
1568                 case PR_MCE_KILL_GET:
1569                         if (arg2 | arg3 | arg4 | arg5)
1570                                 return -EINVAL;
1571                         if (current->flags & PF_MCE_PROCESS)
1572                                 error = (current->flags & PF_MCE_EARLY) ?
1573                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1574                         else
1575                                 error = PR_MCE_KILL_DEFAULT;
1576                         break;
1577                 default:
1578                         error = -EINVAL;
1579                         break;
1580         }
1581         return error;
1582 }
1583
1584 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1585                 struct getcpu_cache __user *, unused)
1586 {
1587         int err = 0;
1588         int cpu = raw_smp_processor_id();
1589         if (cpup)
1590                 err |= put_user(cpu, cpup);
1591         if (nodep)
1592                 err |= put_user(cpu_to_node(cpu), nodep);
1593         return err ? -EFAULT : 0;
1594 }
1595
1596 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1597
1598 static void argv_cleanup(char **argv, char **envp)
1599 {
1600         argv_free(argv);
1601 }
1602
1603 /**
1604  * orderly_poweroff - Trigger an orderly system poweroff
1605  * @force: force poweroff if command execution fails
1606  *
1607  * This may be called from any context to trigger a system shutdown.
1608  * If the orderly shutdown fails, it will force an immediate shutdown.
1609  */
1610 int orderly_poweroff(bool force)
1611 {
1612         int argc;
1613         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1614         static char *envp[] = {
1615                 "HOME=/",
1616                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1617                 NULL
1618         };
1619         int ret = -ENOMEM;
1620         struct subprocess_info *info;
1621
1622         if (argv == NULL) {
1623                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1624                        __func__, poweroff_cmd);
1625                 goto out;
1626         }
1627
1628         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1629         if (info == NULL) {
1630                 argv_free(argv);
1631                 goto out;
1632         }
1633
1634         call_usermodehelper_setcleanup(info, argv_cleanup);
1635
1636         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1637
1638   out:
1639         if (ret && force) {
1640                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1641                        "forcing the issue\n");
1642
1643                 /* I guess this should try to kick off some daemon to
1644                    sync and poweroff asap.  Or not even bother syncing
1645                    if we're doing an emergency shutdown? */
1646                 emergency_sync();
1647                 kernel_power_off();
1648         }
1649
1650         return ret;
1651 }
1652 EXPORT_SYMBOL_GPL(orderly_poweroff);