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