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/personality.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/fs_struct.h>
  39 #include <linux/gfp.h>
  40
  41 #include <linux/compat.h>
  42 #include <linux/syscalls.h>
  43 #include <linux/kprobes.h>
  44 #include <linux/user_namespace.h>
  45
  46 #include <linux/kmsg_dump.h>
  47
  48 #include <asm/uaccess.h>
  49 #include <asm/io.h>
  50 #include <asm/unistd.h>
  51
  52 #ifndef SET_UNALIGN_CTL
  53 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  54 #endif
  55 #ifndef GET_UNALIGN_CTL
  56 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  57 #endif
  58 #ifndef SET_FPEMU_CTL
  59 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  60 #endif
  61 #ifndef GET_FPEMU_CTL
  62 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  63 #endif
  64 #ifndef SET_FPEXC_CTL
  65 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  66 #endif
  67 #ifndef GET_FPEXC_CTL
  68 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  69 #endif
  70 #ifndef GET_ENDIAN
  71 # define GET_ENDIAN(a,b)        (-EINVAL)
  72 #endif
  73 #ifndef SET_ENDIAN
  74 # define SET_ENDIAN(a,b)        (-EINVAL)
  75 #endif
  76 #ifndef GET_TSC_CTL
  77 # define GET_TSC_CTL(a)         (-EINVAL)
  78 #endif
  79 #ifndef SET_TSC_CTL
  80 # define SET_TSC_CTL(a)         (-EINVAL)
  81 #endif
  82
  83 /*
  84  * this is where the system-wide overflow UID and GID are defined, for
  85  * architectures that now have 32-bit UID/GID but didn't in the past
  86  */
  87
  88 int overflowuid = DEFAULT_OVERFLOWUID;
  89 int overflowgid = DEFAULT_OVERFLOWGID;
  90
  91 #ifdef CONFIG_UID16
  92 EXPORT_SYMBOL(overflowuid);
  93 EXPORT_SYMBOL(overflowgid);
  94 #endif
  95
  96 /*
  97  * the same as above, but for filesystems which can only store a 16-bit
  98  * UID and GID. as such, this is needed on all architectures
  99  */
 100
 101 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 102 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 103
 104 EXPORT_SYMBOL(fs_overflowuid);
 105 EXPORT_SYMBOL(fs_overflowgid);
 106
 107 /*
 108  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 109  */
 110
 111 int C_A_D = 1;
 112 struct pid *cad_pid;
 113 EXPORT_SYMBOL(cad_pid);
 114
 115 /*
 116  * If set, this is used for preparing the system to power off.
 117  */
 118
 119 void (*pm_power_off_prepare)(void);
 120
 121 /*
 122  * set the priority of a task
 123  * - the caller must hold the RCU read lock
 124  */
 125 static int set_one_prio(struct task_struct *p, int niceval, int error)
 126 {
 127         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 128         int no_nice;
 129
 130         if (pcred->uid  != cred->euid &&
 131             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
 132                 error = -EPERM;
 133                 goto out;
 134         }
 135         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 136                 error = -EACCES;
 137                 goto out;
 138         }
 139         no_nice = security_task_setnice(p, niceval);
 140         if (no_nice) {
 141                 error = no_nice;
 142                 goto out;
 143         }
 144         if (error == -ESRCH)
 145                 error = 0;
 146         set_user_nice(p, niceval);
 147 out:
 148         return error;
 149 }
 150
 151 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 152 {
 153         struct task_struct *g, *p;
 154         struct user_struct *user;
 155         const struct cred *cred = current_cred();
 156         int error = -EINVAL;
 157         struct pid *pgrp;
 158
 159         if (which > PRIO_USER || which < PRIO_PROCESS)
 160                 goto out;
 161
 162         /* normalize: avoid signed division (rounding problems) */
 163         error = -ESRCH;
 164         if (niceval < -20)
 165                 niceval = -20;
 166         if (niceval > 19)
 167                 niceval = 19;
 168
 169         rcu_read_lock();
 170         read_lock(&tasklist_lock);
 171         switch (which) {
 172                 case PRIO_PROCESS:
 173                         if (who)
 174                                 p = find_task_by_vpid(who);
 175                         else
 176                                 p = current;
 177                         if (p)
 178                                 error = set_one_prio(p, niceval, error);
 179                         break;
 180                 case PRIO_PGRP:
 181                         if (who)
 182                                 pgrp = find_vpid(who);
 183                         else
 184                                 pgrp = task_pgrp(current);
 185                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 186                                 error = set_one_prio(p, niceval, error);
 187                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 188                         break;
 189                 case PRIO_USER:
 190                         user = (struct user_struct *) cred->user;
 191                         if (!who)
 192                                 who = cred->uid;
 193                         else if ((who != cred->uid) &&
 194                                  !(user = find_user(who)))
 195                                 goto out_unlock;        /* No processes for this user */
 196
 197                         do_each_thread(g, p) {
 198                                 if (__task_cred(p)->uid == who)
 199                                         error = set_one_prio(p, niceval, error);
 200                         } while_each_thread(g, p);
 201                         if (who != cred->uid)
 202                                 free_uid(user);         /* For find_user() */
 203                         break;
 204         }
 205 out_unlock:
 206         read_unlock(&tasklist_lock);
 207         rcu_read_unlock();
 208 out:
 209         return error;
 210 }
 211
 212 /*
 213  * Ugh. To avoid negative return values, "getpriority()" will
 214  * not return the normal nice-value, but a negated value that
 215  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 216  * to stay compatible.
 217  */
 218 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 219 {
 220         struct task_struct *g, *p;
 221         struct user_struct *user;
 222         const struct cred *cred = current_cred();
 223         long niceval, retval = -ESRCH;
 224         struct pid *pgrp;
 225
 226         if (which > PRIO_USER || which < PRIO_PROCESS)
 227                 return -EINVAL;
 228
 229         rcu_read_lock();
 230         read_lock(&tasklist_lock);
 231         switch (which) {
 232                 case PRIO_PROCESS:
 233                         if (who)
 234                                 p = find_task_by_vpid(who);
 235                         else
 236                                 p = current;
 237                         if (p) {
 238                                 niceval = 20 - task_nice(p);
 239                                 if (niceval > retval)
 240                                         retval = niceval;
 241                         }
 242                         break;
 243                 case PRIO_PGRP:
 244                         if (who)
 245                                 pgrp = find_vpid(who);
 246                         else
 247                                 pgrp = task_pgrp(current);
 248                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 249                                 niceval = 20 - task_nice(p);
 250                                 if (niceval > retval)
 251                                         retval = niceval;
 252                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 253                         break;
 254                 case PRIO_USER:
 255                         user = (struct user_struct *) cred->user;
 256                         if (!who)
 257                                 who = cred->uid;
 258                         else if ((who != cred->uid) &&
 259                                  !(user = find_user(who)))
 260                                 goto out_unlock;        /* No processes for this user */
 261
 262                         do_each_thread(g, p) {
 263                                 if (__task_cred(p)->uid == who) {
 264                                         niceval = 20 - task_nice(p);
 265                                         if (niceval > retval)
 266                                                 retval = niceval;
 267                                 }
 268                         } while_each_thread(g, p);
 269                         if (who != cred->uid)
 270                                 free_uid(user);         /* for find_user() */
 271                         break;
 272         }
 273 out_unlock:
 274         read_unlock(&tasklist_lock);
 275         rcu_read_unlock();
 276
 277         return retval;
 278 }
 279
 280 /**
 281  *      emergency_restart - reboot the system
 282  *
 283  *      Without shutting down any hardware or taking any locks
 284  *      reboot the system.  This is called when we know we are in
 285  *      trouble so this is our best effort to reboot.  This is
 286  *      safe to call in interrupt context.
 287  */
 288 void emergency_restart(void)
 289 {
 290         kmsg_dump(KMSG_DUMP_EMERG);
 291         machine_emergency_restart();
 292 }
 293 EXPORT_SYMBOL_GPL(emergency_restart);
 294
 295 void kernel_restart_prepare(char *cmd)
 296 {
 297         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 298         system_state = SYSTEM_RESTART;
 299         device_shutdown();
 300         sysdev_shutdown();
 301 }
 302
 303 /**
 304  *      kernel_restart - reboot the system
 305  *      @cmd: pointer to buffer containing command to execute for restart
 306  *              or %NULL
 307  *
 308  *      Shutdown everything and perform a clean reboot.
 309  *      This is not safe to call in interrupt context.
 310  */
 311 void kernel_restart(char *cmd)
 312 {
 313         kernel_restart_prepare(cmd);
 314         if (!cmd)
 315                 printk(KERN_EMERG "Restarting system.\n");
 316         else
 317                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 318         kmsg_dump(KMSG_DUMP_RESTART);
 319         machine_restart(cmd);
 320 }
 321 EXPORT_SYMBOL_GPL(kernel_restart);
 322
 323 static void kernel_shutdown_prepare(enum system_states state)
 324 {
 325         blocking_notifier_call_chain(&reboot_notifier_list,
 326                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 327         system_state = state;
 328         device_shutdown();
 329 }
 330 /**
 331  *      kernel_halt - halt the system
 332  *
 333  *      Shutdown everything and perform a clean system halt.
 334  */
 335 void kernel_halt(void)
 336 {
 337         kernel_shutdown_prepare(SYSTEM_HALT);
 338         sysdev_shutdown();
 339         printk(KERN_EMERG "System halted.\n");
 340         kmsg_dump(KMSG_DUMP_HALT);
 341         machine_halt();
 342 }
 343
 344 EXPORT_SYMBOL_GPL(kernel_halt);
 345
 346 /**
 347  *      kernel_power_off - power_off the system
 348  *
 349  *      Shutdown everything and perform a clean system power_off.
 350  */
 351 void kernel_power_off(void)
 352 {
 353         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 354         if (pm_power_off_prepare)
 355                 pm_power_off_prepare();
 356         disable_nonboot_cpus();
 357         sysdev_shutdown();
 358         printk(KERN_EMERG "Power down.\n");
 359         kmsg_dump(KMSG_DUMP_POWEROFF);
 360         machine_power_off();
 361 }
 362 EXPORT_SYMBOL_GPL(kernel_power_off);
 363
 364 static DEFINE_MUTEX(reboot_mutex);
 365
 366 /*
 367  * Reboot system call: for obvious reasons only root may call it,
 368  * and even root needs to set up some magic numbers in the registers
 369  * so that some mistake won't make this reboot the whole machine.
 370  * You can also set the meaning of the ctrl-alt-del-key here.
 371  *
 372  * reboot doesn't sync: do that yourself before calling this.
 373  */
 374 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 375                 void __user *, arg)
 376 {
 377         char buffer[256];
 378         int ret = 0;
 379
 380         /* We only trust the superuser with rebooting the system. */
 381         if (!capable(CAP_SYS_BOOT))
 382                 return -EPERM;
 383
 384         /* For safety, we require "magic" arguments. */
 385         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 386             (magic2 != LINUX_REBOOT_MAGIC2 &&
 387                         magic2 != LINUX_REBOOT_MAGIC2A &&
 388                         magic2 != LINUX_REBOOT_MAGIC2B &&
 389                         magic2 != LINUX_REBOOT_MAGIC2C))
 390                 return -EINVAL;
 391
 392         /* Instead of trying to make the power_off code look like
 393          * halt when pm_power_off is not set do it the easy way.
 394          */
 395         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 396                 cmd = LINUX_REBOOT_CMD_HALT;
 397
 398         mutex_lock(&reboot_mutex);
 399         switch (cmd) {
 400         case LINUX_REBOOT_CMD_RESTART:
 401                 kernel_restart(NULL);
 402                 break;
 403
 404         case LINUX_REBOOT_CMD_CAD_ON:
 405                 C_A_D = 1;
 406                 break;
 407
 408         case LINUX_REBOOT_CMD_CAD_OFF:
 409                 C_A_D = 0;
 410                 break;
 411
 412         case LINUX_REBOOT_CMD_HALT:
 413                 kernel_halt();
 414                 do_exit(0);
 415                 panic("cannot halt");
 416
 417         case LINUX_REBOOT_CMD_POWER_OFF:
 418                 kernel_power_off();
 419                 do_exit(0);
 420                 break;
 421
 422         case LINUX_REBOOT_CMD_RESTART2:
 423                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 424                         ret = -EFAULT;
 425                         break;
 426                 }
 427                 buffer[sizeof(buffer) - 1] = '\0';
 428
 429                 kernel_restart(buffer);
 430                 break;
 431
 432 #ifdef CONFIG_KEXEC
 433         case LINUX_REBOOT_CMD_KEXEC:
 434                 ret = kernel_kexec();
 435                 break;
 436 #endif
 437
 438 #ifdef CONFIG_HIBERNATION
 439         case LINUX_REBOOT_CMD_SW_SUSPEND:
 440                 ret = hibernate();
 441                 break;
 442 #endif
 443
 444         default:
 445                 ret = -EINVAL;
 446                 break;
 447         }
 448         mutex_unlock(&reboot_mutex);
 449         return ret;
 450 }
 451
 452 static void deferred_cad(struct work_struct *dummy)
 453 {
 454         kernel_restart(NULL);
 455 }
 456
 457 /*
 458  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 459  * As it's called within an interrupt, it may NOT sync: the only choice
 460  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 461  */
 462 void ctrl_alt_del(void)
 463 {
 464         static DECLARE_WORK(cad_work, deferred_cad);
 465
 466         if (C_A_D)
 467                 schedule_work(&cad_work);
 468         else
 469                 kill_cad_pid(SIGINT, 1);
 470 }
 471
 472 /*
 473  * Unprivileged users may change the real gid to the effective gid
 474  * or vice versa.  (BSD-style)
 475  *
 476  * If you set the real gid at all, or set the effective gid to a value not
 477  * equal to the real gid, then the saved gid is set to the new effective gid.
 478  *
 479  * This makes it possible for a setgid program to completely drop its
 480  * privileges, which is often a useful assertion to make when you are doing
 481  * a security audit over a program.
 482  *
 483  * The general idea is that a program which uses just setregid() will be
 484  * 100% compatible with BSD.  A program which uses just setgid() will be
 485  * 100% compatible with POSIX with saved IDs.
 486  *
 487  * SMP: There are not races, the GIDs are checked only by filesystem
 488  *      operations (as far as semantic preservation is concerned).
 489  */
 490 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 491 {
 492         const struct cred *old;
 493         struct cred *new;
 494         int retval;
 495
 496         new = prepare_creds();
 497         if (!new)
 498                 return -ENOMEM;
 499         old = current_cred();
 500
 501         retval = -EPERM;
 502         if (rgid != (gid_t) -1) {
 503                 if (old->gid == rgid ||
 504                     old->egid == rgid ||
 505                     capable(CAP_SETGID))
 506                         new->gid = rgid;
 507                 else
 508                         goto error;
 509         }
 510         if (egid != (gid_t) -1) {
 511                 if (old->gid == egid ||
 512                     old->egid == egid ||
 513                     old->sgid == egid ||
 514                     capable(CAP_SETGID))
 515                         new->egid = egid;
 516                 else
 517                         goto error;
 518         }
 519
 520         if (rgid != (gid_t) -1 ||
 521             (egid != (gid_t) -1 && egid != old->gid))
 522                 new->sgid = new->egid;
 523         new->fsgid = new->egid;
 524
 525         return commit_creds(new);
 526
 527 error:
 528         abort_creds(new);
 529         return retval;
 530 }
 531
 532 /*
 533  * setgid() is implemented like SysV w/ SAVED_IDS
 534  *
 535  * SMP: Same implicit races as above.
 536  */
 537 SYSCALL_DEFINE1(setgid, gid_t, gid)
 538 {
 539         const struct cred *old;
 540         struct cred *new;
 541         int retval;
 542
 543         new = prepare_creds();
 544         if (!new)
 545                 return -ENOMEM;
 546         old = current_cred();
 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 = -EPERM;
 612         if (ruid != (uid_t) -1) {
 613                 new->uid = ruid;
 614                 if (old->uid != ruid &&
 615                     old->euid != ruid &&
 616                     !capable(CAP_SETUID))
 617                         goto error;
 618         }
 619
 620         if (euid != (uid_t) -1) {
 621                 new->euid = euid;
 622                 if (old->uid != euid &&
 623                     old->euid != euid &&
 624                     old->suid != euid &&
 625                     !capable(CAP_SETUID))
 626                         goto error;
 627         }
 628
 629         if (new->uid != old->uid) {
 630                 retval = set_user(new);
 631                 if (retval < 0)
 632                         goto error;
 633         }
 634         if (ruid != (uid_t) -1 ||
 635             (euid != (uid_t) -1 && euid != old->uid))
 636                 new->suid = new->euid;
 637         new->fsuid = new->euid;
 638
 639         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 640         if (retval < 0)
 641                 goto error;
 642
 643         return commit_creds(new);
 644
 645 error:
 646         abort_creds(new);
 647         return retval;
 648 }
 649
 650 /*
 651  * setuid() is implemented like SysV with SAVED_IDS
 652  *
 653  * Note that SAVED_ID's is deficient in that a setuid root program
 654  * like sendmail, for example, cannot set its uid to be a normal
 655  * user and then switch back, because if you're root, setuid() sets
 656  * the saved uid too.  If you don't like this, blame the bright people
 657  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 658  * will allow a root program to temporarily drop privileges and be able to
 659  * regain them by swapping the real and effective uid.
 660  */
 661 SYSCALL_DEFINE1(setuid, uid_t, uid)
 662 {
 663         const struct cred *old;
 664         struct cred *new;
 665         int retval;
 666
 667         new = prepare_creds();
 668         if (!new)
 669                 return -ENOMEM;
 670         old = current_cred();
 671
 672         retval = -EPERM;
 673         if (capable(CAP_SETUID)) {
 674                 new->suid = new->uid = uid;
 675                 if (uid != old->uid) {
 676                         retval = set_user(new);
 677                         if (retval < 0)
 678                                 goto error;
 679                 }
 680         } else if (uid != old->uid && uid != new->suid) {
 681                 goto error;
 682         }
 683
 684         new->fsuid = new->euid = uid;
 685
 686         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 687         if (retval < 0)
 688                 goto error;
 689
 690         return commit_creds(new);
 691
 692 error:
 693         abort_creds(new);
 694         return retval;
 695 }
 696
 697
 698 /*
 699  * This function implements a generic ability to update ruid, euid,
 700  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 701  */
 702 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 703 {
 704         const struct cred *old;
 705         struct cred *new;
 706         int retval;
 707
 708         new = prepare_creds();
 709         if (!new)
 710                 return -ENOMEM;
 711
 712         old = current_cred();
 713
 714         retval = -EPERM;
 715         if (!capable(CAP_SETUID)) {
 716                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 717                     ruid != old->euid  && ruid != old->suid)
 718                         goto error;
 719                 if (euid != (uid_t) -1 && euid != old->uid &&
 720                     euid != old->euid  && euid != old->suid)
 721                         goto error;
 722                 if (suid != (uid_t) -1 && suid != old->uid &&
 723                     suid != old->euid  && suid != old->suid)
 724                         goto error;
 725         }
 726
 727         if (ruid != (uid_t) -1) {
 728                 new->uid = ruid;
 729                 if (ruid != old->uid) {
 730                         retval = set_user(new);
 731                         if (retval < 0)
 732                                 goto error;
 733                 }
 734         }
 735         if (euid != (uid_t) -1)
 736                 new->euid = euid;
 737         if (suid != (uid_t) -1)
 738                 new->suid = suid;
 739         new->fsuid = new->euid;
 740
 741         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 742         if (retval < 0)
 743                 goto error;
 744
 745         return commit_creds(new);
 746
 747 error:
 748         abort_creds(new);
 749         return retval;
 750 }
 751
 752 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 753 {
 754         const struct cred *cred = current_cred();
 755         int retval;
 756
 757         if (!(retval   = put_user(cred->uid,  ruid)) &&
 758             !(retval   = put_user(cred->euid, euid)))
 759                 retval = put_user(cred->suid, suid);
 760
 761         return retval;
 762 }
 763
 764 /*
 765  * Same as above, but for rgid, egid, sgid.
 766  */
 767 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 768 {
 769         const struct cred *old;
 770         struct cred *new;
 771         int retval;
 772
 773         new = prepare_creds();
 774         if (!new)
 775                 return -ENOMEM;
 776         old = current_cred();
 777
 778         retval = -EPERM;
 779         if (!capable(CAP_SETGID)) {
 780                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 781                     rgid != old->egid  && rgid != old->sgid)
 782                         goto error;
 783                 if (egid != (gid_t) -1 && egid != old->gid &&
 784                     egid != old->egid  && egid != old->sgid)
 785                         goto error;
 786                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 787                     sgid != old->egid  && sgid != old->sgid)
 788                         goto error;
 789         }
 790
 791         if (rgid != (gid_t) -1)
 792                 new->gid = rgid;
 793         if (egid != (gid_t) -1)
 794                 new->egid = egid;
 795         if (sgid != (gid_t) -1)
 796                 new->sgid = sgid;
 797         new->fsgid = new->egid;
 798
 799         return commit_creds(new);
 800
 801 error:
 802         abort_creds(new);
 803         return retval;
 804 }
 805
 806 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 807 {
 808         const struct cred *cred = current_cred();
 809         int retval;
 810
 811         if (!(retval   = put_user(cred->gid,  rgid)) &&
 812             !(retval   = put_user(cred->egid, egid)))
 813                 retval = put_user(cred->sgid, sgid);
 814
 815         return retval;
 816 }
 817
 818
 819 /*
 820  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 821  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 822  * whatever uid it wants to). It normally shadows "euid", except when
 823  * explicitly set by setfsuid() or for access..
 824  */
 825 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 826 {
 827         const struct cred *old;
 828         struct cred *new;
 829         uid_t old_fsuid;
 830
 831         new = prepare_creds();
 832         if (!new)
 833                 return current_fsuid();
 834         old = current_cred();
 835         old_fsuid = old->fsuid;
 836
 837         if (uid == old->uid  || uid == old->euid  ||
 838             uid == old->suid || uid == old->fsuid ||
 839             capable(CAP_SETUID)) {
 840                 if (uid != old_fsuid) {
 841                         new->fsuid = uid;
 842                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 843                                 goto change_okay;
 844                 }
 845         }
 846
 847         abort_creds(new);
 848         return old_fsuid;
 849
 850 change_okay:
 851         commit_creds(new);
 852         return old_fsuid;
 853 }
 854
 855 /*
 856  * Samma på svenska..
 857  */
 858 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 859 {
 860         const struct cred *old;
 861         struct cred *new;
 862         gid_t old_fsgid;
 863
 864         new = prepare_creds();
 865         if (!new)
 866                 return current_fsgid();
 867         old = current_cred();
 868         old_fsgid = old->fsgid;
 869
 870         if (gid == old->gid  || gid == old->egid  ||
 871             gid == old->sgid || gid == old->fsgid ||
 872             capable(CAP_SETGID)) {
 873                 if (gid != old_fsgid) {
 874                         new->fsgid = gid;
 875                         goto change_okay;
 876                 }
 877         }
 878
 879         abort_creds(new);
 880         return old_fsgid;
 881
 882 change_okay:
 883         commit_creds(new);
 884         return old_fsgid;
 885 }
 886
 887 void do_sys_times(struct tms *tms)
 888 {
 889         cputime_t tgutime, tgstime, cutime, cstime;
 890
 891         spin_lock_irq(&current->sighand->siglock);
 892         thread_group_times(current, &tgutime, &tgstime);
 893         cutime = current->signal->cutime;
 894         cstime = current->signal->cstime;
 895         spin_unlock_irq(&current->sighand->siglock);
 896         tms->tms_utime = cputime_to_clock_t(tgutime);
 897         tms->tms_stime = cputime_to_clock_t(tgstime);
 898         tms->tms_cutime = cputime_to_clock_t(cutime);
 899         tms->tms_cstime = cputime_to_clock_t(cstime);
 900 }
 901
 902 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 903 {
 904         if (tbuf) {
 905                 struct tms tmp;
 906
 907                 do_sys_times(&tmp);
 908                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 909                         return -EFAULT;
 910         }
 911         force_successful_syscall_return();
 912         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 913 }
 914
 915 /*
 916  * This needs some heavy checking ...
 917  * I just haven't the stomach for it. I also don't fully
 918  * understand sessions/pgrp etc. Let somebody who does explain it.
 919  *
 920  * OK, I think I have the protection semantics right.... this is really
 921  * only important on a multi-user system anyway, to make sure one user
 922  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 923  *
 924  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 925  * LBT 04.03.94
 926  */
 927 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 928 {
 929         struct task_struct *p;
 930         struct task_struct *group_leader = current->group_leader;
 931         struct pid *pgrp;
 932         int err;
 933
 934         if (!pid)
 935                 pid = task_pid_vnr(group_leader);
 936         if (!pgid)
 937                 pgid = pid;
 938         if (pgid < 0)
 939                 return -EINVAL;
 940         rcu_read_lock();
 941
 942         /* From this point forward we keep holding onto the tasklist lock
 943          * so that our parent does not change from under us. -DaveM
 944          */
 945         write_lock_irq(&tasklist_lock);
 946
 947         err = -ESRCH;
 948         p = find_task_by_vpid(pid);
 949         if (!p)
 950                 goto out;
 951
 952         err = -EINVAL;
 953         if (!thread_group_leader(p))
 954                 goto out;
 955
 956         if (same_thread_group(p->real_parent, group_leader)) {
 957                 err = -EPERM;
 958                 if (task_session(p) != task_session(group_leader))
 959                         goto out;
 960                 err = -EACCES;
 961                 if (p->did_exec)
 962                         goto out;
 963         } else {
 964                 err = -ESRCH;
 965                 if (p != group_leader)
 966                         goto out;
 967         }
 968
 969         err = -EPERM;
 970         if (p->signal->leader)
 971                 goto out;
 972
 973         pgrp = task_pid(p);
 974         if (pgid != pid) {
 975                 struct task_struct *g;
 976
 977                 pgrp = find_vpid(pgid);
 978                 g = pid_task(pgrp, PIDTYPE_PGID);
 979                 if (!g || task_session(g) != task_session(group_leader))
 980                         goto out;
 981         }
 982
 983         err = security_task_setpgid(p, pgid);
 984         if (err)
 985                 goto out;
 986
 987         if (task_pgrp(p) != pgrp)
 988                 change_pid(p, PIDTYPE_PGID, pgrp);
 989
 990         err = 0;
 991 out:
 992         /* All paths lead to here, thus we are safe. -DaveM */
 993         write_unlock_irq(&tasklist_lock);
 994         rcu_read_unlock();
 995         return err;
 996 }
 997
 998 SYSCALL_DEFINE1(getpgid, pid_t, pid)
 999 {
1000         struct task_struct *p;
1001         struct pid *grp;
1002         int retval;
1003
1004         rcu_read_lock();
1005         if (!pid)
1006                 grp = task_pgrp(current);
1007         else {
1008                 retval = -ESRCH;
1009                 p = find_task_by_vpid(pid);
1010                 if (!p)
1011                         goto out;
1012                 grp = task_pgrp(p);
1013                 if (!grp)
1014                         goto out;
1015
1016                 retval = security_task_getpgid(p);
1017                 if (retval)
1018                         goto out;
1019         }
1020         retval = pid_vnr(grp);
1021 out:
1022         rcu_read_unlock();
1023         return retval;
1024 }
1025
1026 #ifdef __ARCH_WANT_SYS_GETPGRP
1027
1028 SYSCALL_DEFINE0(getpgrp)
1029 {
1030         return sys_getpgid(0);
1031 }
1032
1033 #endif
1034
1035 SYSCALL_DEFINE1(getsid, pid_t, pid)
1036 {
1037         struct task_struct *p;
1038         struct pid *sid;
1039         int retval;
1040
1041         rcu_read_lock();
1042         if (!pid)
1043                 sid = task_session(current);
1044         else {
1045                 retval = -ESRCH;
1046                 p = find_task_by_vpid(pid);
1047                 if (!p)
1048                         goto out;
1049                 sid = task_session(p);
1050                 if (!sid)
1051                         goto out;
1052
1053                 retval = security_task_getsid(p);
1054                 if (retval)
1055                         goto out;
1056         }
1057         retval = pid_vnr(sid);
1058 out:
1059         rcu_read_unlock();
1060         return retval;
1061 }
1062
1063 SYSCALL_DEFINE0(setsid)
1064 {
1065         struct task_struct *group_leader = current->group_leader;
1066         struct pid *sid = task_pid(group_leader);
1067         pid_t session = pid_vnr(sid);
1068         int err = -EPERM;
1069
1070         write_lock_irq(&tasklist_lock);
1071         /* Fail if I am already a session leader */
1072         if (group_leader->signal->leader)
1073                 goto out;
1074
1075         /* Fail if a process group id already exists that equals the
1076          * proposed session id.
1077          */
1078         if (pid_task(sid, PIDTYPE_PGID))
1079                 goto out;
1080
1081         group_leader->signal->leader = 1;
1082         __set_special_pids(sid);
1083
1084         proc_clear_tty(group_leader);
1085
1086         err = session;
1087 out:
1088         write_unlock_irq(&tasklist_lock);
1089         if (err > 0) {
1090                 proc_sid_connector(group_leader);
1091                 sched_autogroup_create_attach(group_leader);
1092         }
1093         return err;
1094 }
1095
1096 DECLARE_RWSEM(uts_sem);
1097
1098 #ifdef COMPAT_UTS_MACHINE
1099 #define override_architecture(name) \
1100         (personality(current->personality) == PER_LINUX32 && \
1101          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1102                       sizeof(COMPAT_UTS_MACHINE)))
1103 #else
1104 #define override_architecture(name)     0
1105 #endif
1106
1107 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1108 {
1109         int errno = 0;
1110
1111         down_read(&uts_sem);
1112         if (copy_to_user(name, utsname(), sizeof *name))
1113                 errno = -EFAULT;
1114         up_read(&uts_sem);
1115
1116         if (!errno && override_architecture(name))
1117                 errno = -EFAULT;
1118         return errno;
1119 }
1120
1121 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1122 /*
1123  * Old cruft
1124  */
1125 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1126 {
1127         int error = 0;
1128
1129         if (!name)
1130                 return -EFAULT;
1131
1132         down_read(&uts_sem);
1133         if (copy_to_user(name, utsname(), sizeof(*name)))
1134                 error = -EFAULT;
1135         up_read(&uts_sem);
1136
1137         if (!error && override_architecture(name))
1138                 error = -EFAULT;
1139         return error;
1140 }
1141
1142 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1143 {
1144         int error;
1145
1146         if (!name)
1147                 return -EFAULT;
1148         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1149                 return -EFAULT;
1150
1151         down_read(&uts_sem);
1152         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1153                                __OLD_UTS_LEN);
1154         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1155         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1156                                 __OLD_UTS_LEN);
1157         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1158         error |= __copy_to_user(&name->release, &utsname()->release,
1159                                 __OLD_UTS_LEN);
1160         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1161         error |= __copy_to_user(&name->version, &utsname()->version,
1162                                 __OLD_UTS_LEN);
1163         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1164         error |= __copy_to_user(&name->machine, &utsname()->machine,
1165                                 __OLD_UTS_LEN);
1166         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1167         up_read(&uts_sem);
1168
1169         if (!error && override_architecture(name))
1170                 error = -EFAULT;
1171         return error ? -EFAULT : 0;
1172 }
1173 #endif
1174
1175 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1176 {
1177         int errno;
1178         char tmp[__NEW_UTS_LEN];
1179
1180         if (!capable(CAP_SYS_ADMIN))
1181                 return -EPERM;
1182         if (len < 0 || len > __NEW_UTS_LEN)
1183                 return -EINVAL;
1184         down_write(&uts_sem);
1185         errno = -EFAULT;
1186         if (!copy_from_user(tmp, name, len)) {
1187                 struct new_utsname *u = utsname();
1188
1189                 memcpy(u->nodename, tmp, len);
1190                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1191                 errno = 0;
1192         }
1193         up_write(&uts_sem);
1194         return errno;
1195 }
1196
1197 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1198
1199 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1200 {
1201         int i, errno;
1202         struct new_utsname *u;
1203
1204         if (len < 0)
1205                 return -EINVAL;
1206         down_read(&uts_sem);
1207         u = utsname();
1208         i = 1 + strlen(u->nodename);
1209         if (i > len)
1210                 i = len;
1211         errno = 0;
1212         if (copy_to_user(name, u->nodename, i))
1213                 errno = -EFAULT;
1214         up_read(&uts_sem);
1215         return errno;
1216 }
1217
1218 #endif
1219
1220 /*
1221  * Only setdomainname; getdomainname can be implemented by calling
1222  * uname()
1223  */
1224 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1225 {
1226         int errno;
1227         char tmp[__NEW_UTS_LEN];
1228
1229         if (!capable(CAP_SYS_ADMIN))
1230                 return -EPERM;
1231         if (len < 0 || len > __NEW_UTS_LEN)
1232                 return -EINVAL;
1233
1234         down_write(&uts_sem);
1235         errno = -EFAULT;
1236         if (!copy_from_user(tmp, name, len)) {
1237                 struct new_utsname *u = utsname();
1238
1239                 memcpy(u->domainname, tmp, len);
1240                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1241                 errno = 0;
1242         }
1243         up_write(&uts_sem);
1244         return errno;
1245 }
1246
1247 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1248 {
1249         struct rlimit value;
1250         int ret;
1251
1252         ret = do_prlimit(current, resource, NULL, &value);
1253         if (!ret)
1254                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1255
1256         return ret;
1257 }
1258
1259 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1260
1261 /*
1262  *      Back compatibility for getrlimit. Needed for some apps.
1263  */
1264
1265 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1266                 struct rlimit __user *, rlim)
1267 {
1268         struct rlimit x;
1269         if (resource >= RLIM_NLIMITS)
1270                 return -EINVAL;
1271
1272         task_lock(current->group_leader);
1273         x = current->signal->rlim[resource];
1274         task_unlock(current->group_leader);
1275         if (x.rlim_cur > 0x7FFFFFFF)
1276                 x.rlim_cur = 0x7FFFFFFF;
1277         if (x.rlim_max > 0x7FFFFFFF)
1278                 x.rlim_max = 0x7FFFFFFF;
1279         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1280 }
1281
1282 #endif
1283
1284 static inline bool rlim64_is_infinity(__u64 rlim64)
1285 {
1286 #if BITS_PER_LONG < 64
1287         return rlim64 >= ULONG_MAX;
1288 #else
1289         return rlim64 == RLIM64_INFINITY;
1290 #endif
1291 }
1292
1293 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1294 {
1295         if (rlim->rlim_cur == RLIM_INFINITY)
1296                 rlim64->rlim_cur = RLIM64_INFINITY;
1297         else
1298                 rlim64->rlim_cur = rlim->rlim_cur;
1299         if (rlim->rlim_max == RLIM_INFINITY)
1300                 rlim64->rlim_max = RLIM64_INFINITY;
1301         else
1302                 rlim64->rlim_max = rlim->rlim_max;
1303 }
1304
1305 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1306 {
1307         if (rlim64_is_infinity(rlim64->rlim_cur))
1308                 rlim->rlim_cur = RLIM_INFINITY;
1309         else
1310                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1311         if (rlim64_is_infinity(rlim64->rlim_max))
1312                 rlim->rlim_max = RLIM_INFINITY;
1313         else
1314                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1315 }
1316
1317 /* make sure you are allowed to change @tsk limits before calling this */
1318 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1319                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1320 {
1321         struct rlimit *rlim;
1322         int retval = 0;
1323
1324         if (resource >= RLIM_NLIMITS)
1325                 return -EINVAL;
1326         if (new_rlim) {
1327                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1328                         return -EINVAL;
1329                 if (resource == RLIMIT_NOFILE &&
1330                                 new_rlim->rlim_max > sysctl_nr_open)
1331                         return -EPERM;
1332         }
1333
1334         /* protect tsk->signal and tsk->sighand from disappearing */
1335         read_lock(&tasklist_lock);
1336         if (!tsk->sighand) {
1337                 retval = -ESRCH;
1338                 goto out;
1339         }
1340
1341         rlim = tsk->signal->rlim + resource;
1342         task_lock(tsk->group_leader);
1343         if (new_rlim) {
1344                 if (new_rlim->rlim_max > rlim->rlim_max &&
1345                                 !capable(CAP_SYS_RESOURCE))
1346                         retval = -EPERM;
1347                 if (!retval)
1348                         retval = security_task_setrlimit(tsk->group_leader,
1349                                         resource, new_rlim);
1350                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1351                         /*
1352                          * The caller is asking for an immediate RLIMIT_CPU
1353                          * expiry.  But we use the zero value to mean "it was
1354                          * never set".  So let's cheat and make it one second
1355                          * instead
1356                          */
1357                         new_rlim->rlim_cur = 1;
1358                 }
1359         }
1360         if (!retval) {
1361                 if (old_rlim)
1362                         *old_rlim = *rlim;
1363                 if (new_rlim)
1364                         *rlim = *new_rlim;
1365         }
1366         task_unlock(tsk->group_leader);
1367
1368         /*
1369          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1370          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1371          * very long-standing error, and fixing it now risks breakage of
1372          * applications, so we live with it
1373          */
1374          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1375                          new_rlim->rlim_cur != RLIM_INFINITY)
1376                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1377 out:
1378         read_unlock(&tasklist_lock);
1379         return retval;
1380 }
1381
1382 /* rcu lock must be held */
1383 static int check_prlimit_permission(struct task_struct *task)
1384 {
1385         const struct cred *cred = current_cred(), *tcred;
1386
1387         tcred = __task_cred(task);
1388         if (current != task &&
1389             (cred->uid != tcred->euid ||
1390              cred->uid != tcred->suid ||
1391              cred->uid != tcred->uid  ||
1392              cred->gid != tcred->egid ||
1393              cred->gid != tcred->sgid ||
1394              cred->gid != tcred->gid) &&
1395              !capable(CAP_SYS_RESOURCE)) {
1396                 return -EPERM;
1397         }
1398
1399         return 0;
1400 }
1401
1402 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1403                 const struct rlimit64 __user *, new_rlim,
1404                 struct rlimit64 __user *, old_rlim)
1405 {
1406         struct rlimit64 old64, new64;
1407         struct rlimit old, new;
1408         struct task_struct *tsk;
1409         int ret;
1410
1411         if (new_rlim) {
1412                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1413                         return -EFAULT;
1414                 rlim64_to_rlim(&new64, &new);
1415         }
1416
1417         rcu_read_lock();
1418         tsk = pid ? find_task_by_vpid(pid) : current;
1419         if (!tsk) {
1420                 rcu_read_unlock();
1421                 return -ESRCH;
1422         }
1423         ret = check_prlimit_permission(tsk);
1424         if (ret) {
1425                 rcu_read_unlock();
1426                 return ret;
1427         }
1428         get_task_struct(tsk);
1429         rcu_read_unlock();
1430
1431         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1432                         old_rlim ? &old : NULL);
1433
1434         if (!ret && old_rlim) {
1435                 rlim_to_rlim64(&old, &old64);
1436                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1437                         ret = -EFAULT;
1438         }
1439
1440         put_task_struct(tsk);
1441         return ret;
1442 }
1443
1444 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1445 {
1446         struct rlimit new_rlim;
1447
1448         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1449                 return -EFAULT;
1450         return do_prlimit(current, resource, &new_rlim, NULL);
1451 }
1452
1453 /*
1454  * It would make sense to put struct rusage in the task_struct,
1455  * except that would make the task_struct be *really big*.  After
1456  * task_struct gets moved into malloc'ed memory, it would
1457  * make sense to do this.  It will make moving the rest of the information
1458  * a lot simpler!  (Which we're not doing right now because we're not
1459  * measuring them yet).
1460  *
1461  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1462  * races with threads incrementing their own counters.  But since word
1463  * reads are atomic, we either get new values or old values and we don't
1464  * care which for the sums.  We always take the siglock to protect reading
1465  * the c* fields from p->signal from races with exit.c updating those
1466  * fields when reaping, so a sample either gets all the additions of a
1467  * given child after it's reaped, or none so this sample is before reaping.
1468  *
1469  * Locking:
1470  * We need to take the siglock for CHILDEREN, SELF and BOTH
1471  * for  the cases current multithreaded, non-current single threaded
1472  * non-current multithreaded.  Thread traversal is now safe with
1473  * the siglock held.
1474  * Strictly speaking, we donot need to take the siglock if we are current and
1475  * single threaded,  as no one else can take our signal_struct away, no one
1476  * else can  reap the  children to update signal->c* counters, and no one else
1477  * can race with the signal-> fields. If we do not take any lock, the
1478  * signal-> fields could be read out of order while another thread was just
1479  * exiting. So we should  place a read memory barrier when we avoid the lock.
1480  * On the writer side,  write memory barrier is implied in  __exit_signal
1481  * as __exit_signal releases  the siglock spinlock after updating the signal->
1482  * fields. But we don't do this yet to keep things simple.
1483  *
1484  */
1485
1486 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1487 {
1488         r->ru_nvcsw += t->nvcsw;
1489         r->ru_nivcsw += t->nivcsw;
1490         r->ru_minflt += t->min_flt;
1491         r->ru_majflt += t->maj_flt;
1492         r->ru_inblock += task_io_get_inblock(t);
1493         r->ru_oublock += task_io_get_oublock(t);
1494 }
1495
1496 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1497 {
1498         struct task_struct *t;
1499         unsigned long flags;
1500         cputime_t tgutime, tgstime, utime, stime;
1501         unsigned long maxrss = 0;
1502
1503         memset((char *) r, 0, sizeof *r);
1504         utime = stime = cputime_zero;
1505
1506         if (who == RUSAGE_THREAD) {
1507                 task_times(current, &utime, &stime);
1508                 accumulate_thread_rusage(p, r);
1509                 maxrss = p->signal->maxrss;
1510                 goto out;
1511         }
1512
1513         if (!lock_task_sighand(p, &flags))
1514                 return;
1515
1516         switch (who) {
1517                 case RUSAGE_BOTH:
1518                 case RUSAGE_CHILDREN:
1519                         utime = p->signal->cutime;
1520                         stime = p->signal->cstime;
1521                         r->ru_nvcsw = p->signal->cnvcsw;
1522                         r->ru_nivcsw = p->signal->cnivcsw;
1523                         r->ru_minflt = p->signal->cmin_flt;
1524                         r->ru_majflt = p->signal->cmaj_flt;
1525                         r->ru_inblock = p->signal->cinblock;
1526                         r->ru_oublock = p->signal->coublock;
1527                         maxrss = p->signal->cmaxrss;
1528
1529                         if (who == RUSAGE_CHILDREN)
1530                                 break;
1531
1532                 case RUSAGE_SELF:
1533                         thread_group_times(p, &tgutime, &tgstime);
1534                         utime = cputime_add(utime, tgutime);
1535                         stime = cputime_add(stime, tgstime);
1536                         r->ru_nvcsw += p->signal->nvcsw;
1537                         r->ru_nivcsw += p->signal->nivcsw;
1538                         r->ru_minflt += p->signal->min_flt;
1539                         r->ru_majflt += p->signal->maj_flt;
1540                         r->ru_inblock += p->signal->inblock;
1541                         r->ru_oublock += p->signal->oublock;
1542                         if (maxrss < p->signal->maxrss)
1543                                 maxrss = p->signal->maxrss;
1544                         t = p;
1545                         do {
1546                                 accumulate_thread_rusage(t, r);
1547                                 t = next_thread(t);
1548                         } while (t != p);
1549                         break;
1550
1551                 default:
1552                         BUG();
1553         }
1554         unlock_task_sighand(p, &flags);
1555
1556 out:
1557         cputime_to_timeval(utime, &r->ru_utime);
1558         cputime_to_timeval(stime, &r->ru_stime);
1559
1560         if (who != RUSAGE_CHILDREN) {
1561                 struct mm_struct *mm = get_task_mm(p);
1562                 if (mm) {
1563                         setmax_mm_hiwater_rss(&maxrss, mm);
1564                         mmput(mm);
1565                 }
1566         }
1567         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1568 }
1569
1570 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1571 {
1572         struct rusage r;
1573         k_getrusage(p, who, &r);
1574         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1575 }
1576
1577 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1578 {
1579         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1580             who != RUSAGE_THREAD)
1581                 return -EINVAL;
1582         return getrusage(current, who, ru);
1583 }
1584
1585 SYSCALL_DEFINE1(umask, int, mask)
1586 {
1587         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1588         return mask;
1589 }
1590
1591 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1592                 unsigned long, arg4, unsigned long, arg5)
1593 {
1594         struct task_struct *me = current;
1595         unsigned char comm[sizeof(me->comm)];
1596         long error;
1597
1598         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1599         if (error != -ENOSYS)
1600                 return error;
1601
1602         error = 0;
1603         switch (option) {
1604                 case PR_SET_PDEATHSIG:
1605                         if (!valid_signal(arg2)) {
1606                                 error = -EINVAL;
1607                                 break;
1608                         }
1609                         me->pdeath_signal = arg2;
1610                         error = 0;
1611                         break;
1612                 case PR_GET_PDEATHSIG:
1613                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1614                         break;
1615                 case PR_GET_DUMPABLE:
1616                         error = get_dumpable(me->mm);
1617                         break;
1618                 case PR_SET_DUMPABLE:
1619                         if (arg2 < 0 || arg2 > 1) {
1620                                 error = -EINVAL;
1621                                 break;
1622                         }
1623                         set_dumpable(me->mm, arg2);
1624                         error = 0;
1625                         break;
1626
1627                 case PR_SET_UNALIGN:
1628                         error = SET_UNALIGN_CTL(me, arg2);
1629                         break;
1630                 case PR_GET_UNALIGN:
1631                         error = GET_UNALIGN_CTL(me, arg2);
1632                         break;
1633                 case PR_SET_FPEMU:
1634                         error = SET_FPEMU_CTL(me, arg2);
1635                         break;
1636                 case PR_GET_FPEMU:
1637                         error = GET_FPEMU_CTL(me, arg2);
1638                         break;
1639                 case PR_SET_FPEXC:
1640                         error = SET_FPEXC_CTL(me, arg2);
1641                         break;
1642                 case PR_GET_FPEXC:
1643                         error = GET_FPEXC_CTL(me, arg2);
1644                         break;
1645                 case PR_GET_TIMING:
1646                         error = PR_TIMING_STATISTICAL;
1647                         break;
1648                 case PR_SET_TIMING:
1649                         if (arg2 != PR_TIMING_STATISTICAL)
1650                                 error = -EINVAL;
1651                         else
1652                                 error = 0;
1653                         break;
1654
1655                 case PR_SET_NAME:
1656                         comm[sizeof(me->comm)-1] = 0;
1657                         if (strncpy_from_user(comm, (char __user *)arg2,
1658                                               sizeof(me->comm) - 1) < 0)
1659                                 return -EFAULT;
1660                         set_task_comm(me, comm);
1661                         return 0;
1662                 case PR_GET_NAME:
1663                         get_task_comm(comm, me);
1664                         if (copy_to_user((char __user *)arg2, comm,
1665                                          sizeof(comm)))
1666                                 return -EFAULT;
1667                         return 0;
1668                 case PR_GET_ENDIAN:
1669                         error = GET_ENDIAN(me, arg2);
1670                         break;
1671                 case PR_SET_ENDIAN:
1672                         error = SET_ENDIAN(me, arg2);
1673                         break;
1674
1675                 case PR_GET_SECCOMP:
1676                         error = prctl_get_seccomp();
1677                         break;
1678                 case PR_SET_SECCOMP:
1679                         error = prctl_set_seccomp(arg2);
1680                         break;
1681                 case PR_GET_TSC:
1682                         error = GET_TSC_CTL(arg2);
1683                         break;
1684                 case PR_SET_TSC:
1685                         error = SET_TSC_CTL(arg2);
1686                         break;
1687                 case PR_TASK_PERF_EVENTS_DISABLE:
1688                         error = perf_event_task_disable();
1689                         break;
1690                 case PR_TASK_PERF_EVENTS_ENABLE:
1691                         error = perf_event_task_enable();
1692                         break;
1693                 case PR_GET_TIMERSLACK:
1694                         error = current->timer_slack_ns;
1695                         break;
1696                 case PR_SET_TIMERSLACK:
1697                         if (arg2 <= 0)
1698                                 current->timer_slack_ns =
1699                                         current->default_timer_slack_ns;
1700                         else
1701                                 current->timer_slack_ns = arg2;
1702                         error = 0;
1703                         break;
1704                 case PR_MCE_KILL:
1705                         if (arg4 | arg5)
1706                                 return -EINVAL;
1707                         switch (arg2) {
1708                         case PR_MCE_KILL_CLEAR:
1709                                 if (arg3 != 0)
1710                                         return -EINVAL;
1711                                 current->flags &= ~PF_MCE_PROCESS;
1712                                 break;
1713                         case PR_MCE_KILL_SET:
1714                                 current->flags |= PF_MCE_PROCESS;
1715                                 if (arg3 == PR_MCE_KILL_EARLY)
1716                                         current->flags |= PF_MCE_EARLY;
1717                                 else if (arg3 == PR_MCE_KILL_LATE)
1718                                         current->flags &= ~PF_MCE_EARLY;
1719                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1720                                         current->flags &=
1721                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1722                                 else
1723                                         return -EINVAL;
1724                                 break;
1725                         default:
1726                                 return -EINVAL;
1727                         }
1728                         error = 0;
1729                         break;
1730                 case PR_MCE_KILL_GET:
1731                         if (arg2 | arg3 | arg4 | arg5)
1732                                 return -EINVAL;
1733                         if (current->flags & PF_MCE_PROCESS)
1734                                 error = (current->flags & PF_MCE_EARLY) ?
1735                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1736                         else
1737                                 error = PR_MCE_KILL_DEFAULT;
1738                         break;
1739                 default:
1740                         error = -EINVAL;
1741                         break;
1742         }
1743         return error;
1744 }
1745
1746 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1747                 struct getcpu_cache __user *, unused)
1748 {
1749         int err = 0;
1750         int cpu = raw_smp_processor_id();
1751         if (cpup)
1752                 err |= put_user(cpu, cpup);
1753         if (nodep)
1754                 err |= put_user(cpu_to_node(cpu), nodep);
1755         return err ? -EFAULT : 0;
1756 }
1757
1758 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1759
1760 static void argv_cleanup(struct subprocess_info *info)
1761 {
1762         argv_free(info->argv);
1763 }
1764
1765 /**
1766  * orderly_poweroff - Trigger an orderly system poweroff
1767  * @force: force poweroff if command execution fails
1768  *
1769  * This may be called from any context to trigger a system shutdown.
1770  * If the orderly shutdown fails, it will force an immediate shutdown.
1771  */
1772 int orderly_poweroff(bool force)
1773 {
1774         int argc;
1775         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1776         static char *envp[] = {
1777                 "HOME=/",
1778                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1779                 NULL
1780         };
1781         int ret = -ENOMEM;
1782         struct subprocess_info *info;
1783
1784         if (argv == NULL) {
1785                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1786                        __func__, poweroff_cmd);
1787                 goto out;
1788         }
1789
1790         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1791         if (info == NULL) {
1792                 argv_free(argv);
1793                 goto out;
1794         }
1795
1796         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1797
1798         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1799
1800   out:
1801         if (ret && force) {
1802                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1803                        "forcing the issue\n");
1804
1805                 /* I guess this should try to kick off some daemon to
1806                    sync and poweroff asap.  Or not even bother syncing
1807                    if we're doing an emergency shutdown? */
1808                 emergency_sync();
1809                 kernel_power_off();
1810         }
1811
1812         return ret;
1813 }
1814 EXPORT_SYMBOL_GPL(orderly_poweroff);