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 <asm/uaccess.h>
  47 #include <asm/io.h>
  48 #include <asm/unistd.h>
  49
  50 #ifndef SET_UNALIGN_CTL
  51 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  52 #endif
  53 #ifndef GET_UNALIGN_CTL
  54 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  55 #endif
  56 #ifndef SET_FPEMU_CTL
  57 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  58 #endif
  59 #ifndef GET_FPEMU_CTL
  60 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  61 #endif
  62 #ifndef SET_FPEXC_CTL
  63 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  64 #endif
  65 #ifndef GET_FPEXC_CTL
  66 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  67 #endif
  68 #ifndef GET_ENDIAN
  69 # define GET_ENDIAN(a,b)        (-EINVAL)
  70 #endif
  71 #ifndef SET_ENDIAN
  72 # define SET_ENDIAN(a,b)        (-EINVAL)
  73 #endif
  74 #ifndef GET_TSC_CTL
  75 # define GET_TSC_CTL(a)         (-EINVAL)
  76 #endif
  77 #ifndef SET_TSC_CTL
  78 # define SET_TSC_CTL(a)         (-EINVAL)
  79 #endif
  80
  81 /*
  82  * this is where the system-wide overflow UID and GID are defined, for
  83  * architectures that now have 32-bit UID/GID but didn't in the past
  84  */
  85
  86 int overflowuid = DEFAULT_OVERFLOWUID;
  87 int overflowgid = DEFAULT_OVERFLOWGID;
  88
  89 #ifdef CONFIG_UID16
  90 EXPORT_SYMBOL(overflowuid);
  91 EXPORT_SYMBOL(overflowgid);
  92 #endif
  93
  94 /*
  95  * the same as above, but for filesystems which can only store a 16-bit
  96  * UID and GID. as such, this is needed on all architectures
  97  */
  98
  99 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 100 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 101
 102 EXPORT_SYMBOL(fs_overflowuid);
 103 EXPORT_SYMBOL(fs_overflowgid);
 104
 105 /*
 106  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 107  */
 108
 109 int C_A_D = 1;
 110 struct pid *cad_pid;
 111 EXPORT_SYMBOL(cad_pid);
 112
 113 /*
 114  * If set, this is used for preparing the system to power off.
 115  */
 116
 117 void (*pm_power_off_prepare)(void);
 118
 119 /*
 120  * set the priority of a task
 121  * - the caller must hold the RCU read lock
 122  */
 123 static int set_one_prio(struct task_struct *p, int niceval, int error)
 124 {
 125         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 126         int no_nice;
 127
 128         if (pcred->uid  != cred->euid &&
 129             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
 130                 error = -EPERM;
 131                 goto out;
 132         }
 133         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 134                 error = -EACCES;
 135                 goto out;
 136         }
 137         no_nice = security_task_setnice(p, niceval);
 138         if (no_nice) {
 139                 error = no_nice;
 140                 goto out;
 141         }
 142         if (error == -ESRCH)
 143                 error = 0;
 144         set_user_nice(p, niceval);
 145 out:
 146         return error;
 147 }
 148
 149 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 150 {
 151         struct task_struct *g, *p;
 152         struct user_struct *user;
 153         const struct cred *cred = current_cred();
 154         int error = -EINVAL;
 155         struct pid *pgrp;
 156
 157         if (which > PRIO_USER || which < PRIO_PROCESS)
 158                 goto out;
 159
 160         /* normalize: avoid signed division (rounding problems) */
 161         error = -ESRCH;
 162         if (niceval < -20)
 163                 niceval = -20;
 164         if (niceval > 19)
 165                 niceval = 19;
 166
 167         rcu_read_lock();
 168         read_lock(&tasklist_lock);
 169         switch (which) {
 170                 case PRIO_PROCESS:
 171                         if (who)
 172                                 p = find_task_by_vpid(who);
 173                         else
 174                                 p = current;
 175                         if (p)
 176                                 error = set_one_prio(p, niceval, error);
 177                         break;
 178                 case PRIO_PGRP:
 179                         if (who)
 180                                 pgrp = find_vpid(who);
 181                         else
 182                                 pgrp = task_pgrp(current);
 183                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 184                                 error = set_one_prio(p, niceval, error);
 185                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 186                         break;
 187                 case PRIO_USER:
 188                         user = (struct user_struct *) cred->user;
 189                         if (!who)
 190                                 who = cred->uid;
 191                         else if ((who != cred->uid) &&
 192                                  !(user = find_user(who)))
 193                                 goto out_unlock;        /* No processes for this user */
 194
 195                         do_each_thread(g, p) {
 196                                 if (__task_cred(p)->uid == who)
 197                                         error = set_one_prio(p, niceval, error);
 198                         } while_each_thread(g, p);
 199                         if (who != cred->uid)
 200                                 free_uid(user);         /* For find_user() */
 201                         break;
 202         }
 203 out_unlock:
 204         read_unlock(&tasklist_lock);
 205         rcu_read_unlock();
 206 out:
 207         return error;
 208 }
 209
 210 /*
 211  * Ugh. To avoid negative return values, "getpriority()" will
 212  * not return the normal nice-value, but a negated value that
 213  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 214  * to stay compatible.
 215  */
 216 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 217 {
 218         struct task_struct *g, *p;
 219         struct user_struct *user;
 220         const struct cred *cred = current_cred();
 221         long niceval, retval = -ESRCH;
 222         struct pid *pgrp;
 223
 224         if (which > PRIO_USER || which < PRIO_PROCESS)
 225                 return -EINVAL;
 226
 227         rcu_read_lock();
 228         read_lock(&tasklist_lock);
 229         switch (which) {
 230                 case PRIO_PROCESS:
 231                         if (who)
 232                                 p = find_task_by_vpid(who);
 233                         else
 234                                 p = current;
 235                         if (p) {
 236                                 niceval = 20 - task_nice(p);
 237                                 if (niceval > retval)
 238                                         retval = niceval;
 239                         }
 240                         break;
 241                 case PRIO_PGRP:
 242                         if (who)
 243                                 pgrp = find_vpid(who);
 244                         else
 245                                 pgrp = task_pgrp(current);
 246                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 247                                 niceval = 20 - task_nice(p);
 248                                 if (niceval > retval)
 249                                         retval = niceval;
 250                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 251                         break;
 252                 case PRIO_USER:
 253                         user = (struct user_struct *) cred->user;
 254                         if (!who)
 255                                 who = cred->uid;
 256                         else if ((who != cred->uid) &&
 257                                  !(user = find_user(who)))
 258                                 goto out_unlock;        /* No processes for this user */
 259
 260                         do_each_thread(g, p) {
 261                                 if (__task_cred(p)->uid == who) {
 262                                         niceval = 20 - task_nice(p);
 263                                         if (niceval > retval)
 264                                                 retval = niceval;
 265                                 }
 266                         } while_each_thread(g, p);
 267                         if (who != cred->uid)
 268                                 free_uid(user);         /* for find_user() */
 269                         break;
 270         }
 271 out_unlock:
 272         read_unlock(&tasklist_lock);
 273         rcu_read_unlock();
 274
 275         return retval;
 276 }
 277
 278 /**
 279  *      emergency_restart - reboot the system
 280  *
 281  *      Without shutting down any hardware or taking any locks
 282  *      reboot the system.  This is called when we know we are in
 283  *      trouble so this is our best effort to reboot.  This is
 284  *      safe to call in interrupt context.
 285  */
 286 void emergency_restart(void)
 287 {
 288         machine_emergency_restart();
 289 }
 290 EXPORT_SYMBOL_GPL(emergency_restart);
 291
 292 void kernel_restart_prepare(char *cmd)
 293 {
 294         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 295         system_state = SYSTEM_RESTART;
 296         device_shutdown();
 297         sysdev_shutdown();
 298 }
 299
 300 /**
 301  *      kernel_restart - reboot the system
 302  *      @cmd: pointer to buffer containing command to execute for restart
 303  *              or %NULL
 304  *
 305  *      Shutdown everything and perform a clean reboot.
 306  *      This is not safe to call in interrupt context.
 307  */
 308 void kernel_restart(char *cmd)
 309 {
 310         kernel_restart_prepare(cmd);
 311         if (!cmd)
 312                 printk(KERN_EMERG "Restarting system.\n");
 313         else
 314                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 315         machine_restart(cmd);
 316 }
 317 EXPORT_SYMBOL_GPL(kernel_restart);
 318
 319 static void kernel_shutdown_prepare(enum system_states state)
 320 {
 321         blocking_notifier_call_chain(&reboot_notifier_list,
 322                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 323         system_state = state;
 324         device_shutdown();
 325 }
 326 /**
 327  *      kernel_halt - halt the system
 328  *
 329  *      Shutdown everything and perform a clean system halt.
 330  */
 331 void kernel_halt(void)
 332 {
 333         kernel_shutdown_prepare(SYSTEM_HALT);
 334         sysdev_shutdown();
 335         printk(KERN_EMERG "System halted.\n");
 336         machine_halt();
 337 }
 338
 339 EXPORT_SYMBOL_GPL(kernel_halt);
 340
 341 /**
 342  *      kernel_power_off - power_off the system
 343  *
 344  *      Shutdown everything and perform a clean system power_off.
 345  */
 346 void kernel_power_off(void)
 347 {
 348         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 349         if (pm_power_off_prepare)
 350                 pm_power_off_prepare();
 351         disable_nonboot_cpus();
 352         sysdev_shutdown();
 353         printk(KERN_EMERG "Power down.\n");
 354         machine_power_off();
 355 }
 356 EXPORT_SYMBOL_GPL(kernel_power_off);
 357
 358 static DEFINE_MUTEX(reboot_mutex);
 359
 360 /*
 361  * Reboot system call: for obvious reasons only root may call it,
 362  * and even root needs to set up some magic numbers in the registers
 363  * so that some mistake won't make this reboot the whole machine.
 364  * You can also set the meaning of the ctrl-alt-del-key here.
 365  *
 366  * reboot doesn't sync: do that yourself before calling this.
 367  */
 368 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 369                 void __user *, arg)
 370 {
 371         char buffer[256];
 372         int ret = 0;
 373
 374         /* We only trust the superuser with rebooting the system. */
 375         if (!capable(CAP_SYS_BOOT))
 376                 return -EPERM;
 377
 378         /* For safety, we require "magic" arguments. */
 379         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 380             (magic2 != LINUX_REBOOT_MAGIC2 &&
 381                         magic2 != LINUX_REBOOT_MAGIC2A &&
 382                         magic2 != LINUX_REBOOT_MAGIC2B &&
 383                         magic2 != LINUX_REBOOT_MAGIC2C))
 384                 return -EINVAL;
 385
 386         /* Instead of trying to make the power_off code look like
 387          * halt when pm_power_off is not set do it the easy way.
 388          */
 389         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 390                 cmd = LINUX_REBOOT_CMD_HALT;
 391
 392         mutex_lock(&reboot_mutex);
 393         switch (cmd) {
 394         case LINUX_REBOOT_CMD_RESTART:
 395                 kernel_restart(NULL);
 396                 break;
 397
 398         case LINUX_REBOOT_CMD_CAD_ON:
 399                 C_A_D = 1;
 400                 break;
 401
 402         case LINUX_REBOOT_CMD_CAD_OFF:
 403                 C_A_D = 0;
 404                 break;
 405
 406         case LINUX_REBOOT_CMD_HALT:
 407                 kernel_halt();
 408                 do_exit(0);
 409                 panic("cannot halt");
 410
 411         case LINUX_REBOOT_CMD_POWER_OFF:
 412                 kernel_power_off();
 413                 do_exit(0);
 414                 break;
 415
 416         case LINUX_REBOOT_CMD_RESTART2:
 417                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 418                         ret = -EFAULT;
 419                         break;
 420                 }
 421                 buffer[sizeof(buffer) - 1] = '\0';
 422
 423                 kernel_restart(buffer);
 424                 break;
 425
 426 #ifdef CONFIG_KEXEC
 427         case LINUX_REBOOT_CMD_KEXEC:
 428                 ret = kernel_kexec();
 429                 break;
 430 #endif
 431
 432 #ifdef CONFIG_HIBERNATION
 433         case LINUX_REBOOT_CMD_SW_SUSPEND:
 434                 ret = hibernate();
 435                 break;
 436 #endif
 437
 438         default:
 439                 ret = -EINVAL;
 440                 break;
 441         }
 442         mutex_unlock(&reboot_mutex);
 443         return ret;
 444 }
 445
 446 static void deferred_cad(struct work_struct *dummy)
 447 {
 448         kernel_restart(NULL);
 449 }
 450
 451 /*
 452  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 453  * As it's called within an interrupt, it may NOT sync: the only choice
 454  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 455  */
 456 void ctrl_alt_del(void)
 457 {
 458         static DECLARE_WORK(cad_work, deferred_cad);
 459
 460         if (C_A_D)
 461                 schedule_work(&cad_work);
 462         else
 463                 kill_cad_pid(SIGINT, 1);
 464 }
 465
 466 /*
 467  * Unprivileged users may change the real gid to the effective gid
 468  * or vice versa.  (BSD-style)
 469  *
 470  * If you set the real gid at all, or set the effective gid to a value not
 471  * equal to the real gid, then the saved gid is set to the new effective gid.
 472  *
 473  * This makes it possible for a setgid program to completely drop its
 474  * privileges, which is often a useful assertion to make when you are doing
 475  * a security audit over a program.
 476  *
 477  * The general idea is that a program which uses just setregid() will be
 478  * 100% compatible with BSD.  A program which uses just setgid() will be
 479  * 100% compatible with POSIX with saved IDs.
 480  *
 481  * SMP: There are not races, the GIDs are checked only by filesystem
 482  *      operations (as far as semantic preservation is concerned).
 483  */
 484 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 485 {
 486         const struct cred *old;
 487         struct cred *new;
 488         int retval;
 489
 490         new = prepare_creds();
 491         if (!new)
 492                 return -ENOMEM;
 493         old = current_cred();
 494
 495         retval = -EPERM;
 496         if (rgid != (gid_t) -1) {
 497                 if (old->gid == rgid ||
 498                     old->egid == rgid ||
 499                     capable(CAP_SETGID))
 500                         new->gid = rgid;
 501                 else
 502                         goto error;
 503         }
 504         if (egid != (gid_t) -1) {
 505                 if (old->gid == egid ||
 506                     old->egid == egid ||
 507                     old->sgid == egid ||
 508                     capable(CAP_SETGID))
 509                         new->egid = egid;
 510                 else
 511                         goto error;
 512         }
 513
 514         if (rgid != (gid_t) -1 ||
 515             (egid != (gid_t) -1 && egid != old->gid))
 516                 new->sgid = new->egid;
 517         new->fsgid = new->egid;
 518
 519         return commit_creds(new);
 520
 521 error:
 522         abort_creds(new);
 523         return retval;
 524 }
 525
 526 /*
 527  * setgid() is implemented like SysV w/ SAVED_IDS
 528  *
 529  * SMP: Same implicit races as above.
 530  */
 531 SYSCALL_DEFINE1(setgid, gid_t, gid)
 532 {
 533         const struct cred *old;
 534         struct cred *new;
 535         int retval;
 536
 537         new = prepare_creds();
 538         if (!new)
 539                 return -ENOMEM;
 540         old = current_cred();
 541
 542         retval = -EPERM;
 543         if (capable(CAP_SETGID))
 544                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 545         else if (gid == old->gid || gid == old->sgid)
 546                 new->egid = new->fsgid = gid;
 547         else
 548                 goto error;
 549
 550         return commit_creds(new);
 551
 552 error:
 553         abort_creds(new);
 554         return retval;
 555 }
 556
 557 /*
 558  * change the user struct in a credentials set to match the new UID
 559  */
 560 static int set_user(struct cred *new)
 561 {
 562         struct user_struct *new_user;
 563
 564         new_user = alloc_uid(current_user_ns(), new->uid);
 565         if (!new_user)
 566                 return -EAGAIN;
 567
 568         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 569                         new_user != INIT_USER) {
 570                 free_uid(new_user);
 571                 return -EAGAIN;
 572         }
 573
 574         free_uid(new->user);
 575         new->user = new_user;
 576         return 0;
 577 }
 578
 579 /*
 580  * Unprivileged users may change the real uid to the effective uid
 581  * or vice versa.  (BSD-style)
 582  *
 583  * If you set the real uid at all, or set the effective uid to a value not
 584  * equal to the real uid, then the saved uid is set to the new effective uid.
 585  *
 586  * This makes it possible for a setuid program to completely drop its
 587  * privileges, which is often a useful assertion to make when you are doing
 588  * a security audit over a program.
 589  *
 590  * The general idea is that a program which uses just setreuid() will be
 591  * 100% compatible with BSD.  A program which uses just setuid() will be
 592  * 100% compatible with POSIX with saved IDs.
 593  */
 594 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 595 {
 596         const struct cred *old;
 597         struct cred *new;
 598         int retval;
 599
 600         new = prepare_creds();
 601         if (!new)
 602                 return -ENOMEM;
 603         old = current_cred();
 604
 605         retval = -EPERM;
 606         if (ruid != (uid_t) -1) {
 607                 new->uid = ruid;
 608                 if (old->uid != ruid &&
 609                     old->euid != ruid &&
 610                     !capable(CAP_SETUID))
 611                         goto error;
 612         }
 613
 614         if (euid != (uid_t) -1) {
 615                 new->euid = euid;
 616                 if (old->uid != euid &&
 617                     old->euid != euid &&
 618                     old->suid != euid &&
 619                     !capable(CAP_SETUID))
 620                         goto error;
 621         }
 622
 623         if (new->uid != old->uid) {
 624                 retval = set_user(new);
 625                 if (retval < 0)
 626                         goto error;
 627         }
 628         if (ruid != (uid_t) -1 ||
 629             (euid != (uid_t) -1 && euid != old->uid))
 630                 new->suid = new->euid;
 631         new->fsuid = new->euid;
 632
 633         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 634         if (retval < 0)
 635                 goto error;
 636
 637         return commit_creds(new);
 638
 639 error:
 640         abort_creds(new);
 641         return retval;
 642 }
 643
 644 /*
 645  * setuid() is implemented like SysV with SAVED_IDS
 646  *
 647  * Note that SAVED_ID's is deficient in that a setuid root program
 648  * like sendmail, for example, cannot set its uid to be a normal
 649  * user and then switch back, because if you're root, setuid() sets
 650  * the saved uid too.  If you don't like this, blame the bright people
 651  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 652  * will allow a root program to temporarily drop privileges and be able to
 653  * regain them by swapping the real and effective uid.
 654  */
 655 SYSCALL_DEFINE1(setuid, uid_t, uid)
 656 {
 657         const struct cred *old;
 658         struct cred *new;
 659         int retval;
 660
 661         new = prepare_creds();
 662         if (!new)
 663                 return -ENOMEM;
 664         old = current_cred();
 665
 666         retval = -EPERM;
 667         if (capable(CAP_SETUID)) {
 668                 new->suid = new->uid = uid;
 669                 if (uid != old->uid) {
 670                         retval = set_user(new);
 671                         if (retval < 0)
 672                                 goto error;
 673                 }
 674         } else if (uid != old->uid && uid != new->suid) {
 675                 goto error;
 676         }
 677
 678         new->fsuid = new->euid = uid;
 679
 680         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 681         if (retval < 0)
 682                 goto error;
 683
 684         return commit_creds(new);
 685
 686 error:
 687         abort_creds(new);
 688         return retval;
 689 }
 690
 691
 692 /*
 693  * This function implements a generic ability to update ruid, euid,
 694  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 695  */
 696 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 697 {
 698         const struct cred *old;
 699         struct cred *new;
 700         int retval;
 701
 702         new = prepare_creds();
 703         if (!new)
 704                 return -ENOMEM;
 705
 706         old = current_cred();
 707
 708         retval = -EPERM;
 709         if (!capable(CAP_SETUID)) {
 710                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 711                     ruid != old->euid  && ruid != old->suid)
 712                         goto error;
 713                 if (euid != (uid_t) -1 && euid != old->uid &&
 714                     euid != old->euid  && euid != old->suid)
 715                         goto error;
 716                 if (suid != (uid_t) -1 && suid != old->uid &&
 717                     suid != old->euid  && suid != old->suid)
 718                         goto error;
 719         }
 720
 721         if (ruid != (uid_t) -1) {
 722                 new->uid = ruid;
 723                 if (ruid != old->uid) {
 724                         retval = set_user(new);
 725                         if (retval < 0)
 726                                 goto error;
 727                 }
 728         }
 729         if (euid != (uid_t) -1)
 730                 new->euid = euid;
 731         if (suid != (uid_t) -1)
 732                 new->suid = suid;
 733         new->fsuid = new->euid;
 734
 735         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 736         if (retval < 0)
 737                 goto error;
 738
 739         return commit_creds(new);
 740
 741 error:
 742         abort_creds(new);
 743         return retval;
 744 }
 745
 746 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 747 {
 748         const struct cred *cred = current_cred();
 749         int retval;
 750
 751         if (!(retval   = put_user(cred->uid,  ruid)) &&
 752             !(retval   = put_user(cred->euid, euid)))
 753                 retval = put_user(cred->suid, suid);
 754
 755         return retval;
 756 }
 757
 758 /*
 759  * Same as above, but for rgid, egid, sgid.
 760  */
 761 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 762 {
 763         const struct cred *old;
 764         struct cred *new;
 765         int retval;
 766
 767         new = prepare_creds();
 768         if (!new)
 769                 return -ENOMEM;
 770         old = current_cred();
 771
 772         retval = -EPERM;
 773         if (!capable(CAP_SETGID)) {
 774                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 775                     rgid != old->egid  && rgid != old->sgid)
 776                         goto error;
 777                 if (egid != (gid_t) -1 && egid != old->gid &&
 778                     egid != old->egid  && egid != old->sgid)
 779                         goto error;
 780                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 781                     sgid != old->egid  && sgid != old->sgid)
 782                         goto error;
 783         }
 784
 785         if (rgid != (gid_t) -1)
 786                 new->gid = rgid;
 787         if (egid != (gid_t) -1)
 788                 new->egid = egid;
 789         if (sgid != (gid_t) -1)
 790                 new->sgid = sgid;
 791         new->fsgid = new->egid;
 792
 793         return commit_creds(new);
 794
 795 error:
 796         abort_creds(new);
 797         return retval;
 798 }
 799
 800 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 801 {
 802         const struct cred *cred = current_cred();
 803         int retval;
 804
 805         if (!(retval   = put_user(cred->gid,  rgid)) &&
 806             !(retval   = put_user(cred->egid, egid)))
 807                 retval = put_user(cred->sgid, sgid);
 808
 809         return retval;
 810 }
 811
 812
 813 /*
 814  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 815  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 816  * whatever uid it wants to). It normally shadows "euid", except when
 817  * explicitly set by setfsuid() or for access..
 818  */
 819 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 820 {
 821         const struct cred *old;
 822         struct cred *new;
 823         uid_t old_fsuid;
 824
 825         new = prepare_creds();
 826         if (!new)
 827                 return current_fsuid();
 828         old = current_cred();
 829         old_fsuid = old->fsuid;
 830
 831         if (uid == old->uid  || uid == old->euid  ||
 832             uid == old->suid || uid == old->fsuid ||
 833             capable(CAP_SETUID)) {
 834                 if (uid != old_fsuid) {
 835                         new->fsuid = uid;
 836                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 837                                 goto change_okay;
 838                 }
 839         }
 840
 841         abort_creds(new);
 842         return old_fsuid;
 843
 844 change_okay:
 845         commit_creds(new);
 846         return old_fsuid;
 847 }
 848
 849 /*
 850  * Samma på svenska..
 851  */
 852 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 853 {
 854         const struct cred *old;
 855         struct cred *new;
 856         gid_t old_fsgid;
 857
 858         new = prepare_creds();
 859         if (!new)
 860                 return current_fsgid();
 861         old = current_cred();
 862         old_fsgid = old->fsgid;
 863
 864         if (gid == old->gid  || gid == old->egid  ||
 865             gid == old->sgid || gid == old->fsgid ||
 866             capable(CAP_SETGID)) {
 867                 if (gid != old_fsgid) {
 868                         new->fsgid = gid;
 869                         goto change_okay;
 870                 }
 871         }
 872
 873         abort_creds(new);
 874         return old_fsgid;
 875
 876 change_okay:
 877         commit_creds(new);
 878         return old_fsgid;
 879 }
 880
 881 void do_sys_times(struct tms *tms)
 882 {
 883         cputime_t tgutime, tgstime, cutime, cstime;
 884
 885         spin_lock_irq(&current->sighand->siglock);
 886         thread_group_times(current, &tgutime, &tgstime);
 887         cutime = current->signal->cutime;
 888         cstime = current->signal->cstime;
 889         spin_unlock_irq(&current->sighand->siglock);
 890         tms->tms_utime = cputime_to_clock_t(tgutime);
 891         tms->tms_stime = cputime_to_clock_t(tgstime);
 892         tms->tms_cutime = cputime_to_clock_t(cutime);
 893         tms->tms_cstime = cputime_to_clock_t(cstime);
 894 }
 895
 896 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 897 {
 898         if (tbuf) {
 899                 struct tms tmp;
 900
 901                 do_sys_times(&tmp);
 902                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 903                         return -EFAULT;
 904         }
 905         force_successful_syscall_return();
 906         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 907 }
 908
 909 /*
 910  * This needs some heavy checking ...
 911  * I just haven't the stomach for it. I also don't fully
 912  * understand sessions/pgrp etc. Let somebody who does explain it.
 913  *
 914  * OK, I think I have the protection semantics right.... this is really
 915  * only important on a multi-user system anyway, to make sure one user
 916  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 917  *
 918  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 919  * LBT 04.03.94
 920  */
 921 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 922 {
 923         struct task_struct *p;
 924         struct task_struct *group_leader = current->group_leader;
 925         struct pid *pgrp;
 926         int err;
 927
 928         if (!pid)
 929                 pid = task_pid_vnr(group_leader);
 930         if (!pgid)
 931                 pgid = pid;
 932         if (pgid < 0)
 933                 return -EINVAL;
 934         rcu_read_lock();
 935
 936         /* From this point forward we keep holding onto the tasklist lock
 937          * so that our parent does not change from under us. -DaveM
 938          */
 939         write_lock_irq(&tasklist_lock);
 940
 941         err = -ESRCH;
 942         p = find_task_by_vpid(pid);
 943         if (!p)
 944                 goto out;
 945
 946         err = -EINVAL;
 947         if (!thread_group_leader(p))
 948                 goto out;
 949
 950         if (same_thread_group(p->real_parent, group_leader)) {
 951                 err = -EPERM;
 952                 if (task_session(p) != task_session(group_leader))
 953                         goto out;
 954                 err = -EACCES;
 955                 if (p->did_exec)
 956                         goto out;
 957         } else {
 958                 err = -ESRCH;
 959                 if (p != group_leader)
 960                         goto out;
 961         }
 962
 963         err = -EPERM;
 964         if (p->signal->leader)
 965                 goto out;
 966
 967         pgrp = task_pid(p);
 968         if (pgid != pid) {
 969                 struct task_struct *g;
 970
 971                 pgrp = find_vpid(pgid);
 972                 g = pid_task(pgrp, PIDTYPE_PGID);
 973                 if (!g || task_session(g) != task_session(group_leader))
 974                         goto out;
 975         }
 976
 977         err = security_task_setpgid(p, pgid);
 978         if (err)
 979                 goto out;
 980
 981         if (task_pgrp(p) != pgrp)
 982                 change_pid(p, PIDTYPE_PGID, pgrp);
 983
 984         err = 0;
 985 out:
 986         /* All paths lead to here, thus we are safe. -DaveM */
 987         write_unlock_irq(&tasklist_lock);
 988         rcu_read_unlock();
 989         return err;
 990 }
 991
 992 SYSCALL_DEFINE1(getpgid, pid_t, pid)
 993 {
 994         struct task_struct *p;
 995         struct pid *grp;
 996         int retval;
 997
 998         rcu_read_lock();
 999         if (!pid)
1000                 grp = task_pgrp(current);
1001         else {
1002                 retval = -ESRCH;
1003                 p = find_task_by_vpid(pid);
1004                 if (!p)
1005                         goto out;
1006                 grp = task_pgrp(p);
1007                 if (!grp)
1008                         goto out;
1009
1010                 retval = security_task_getpgid(p);
1011                 if (retval)
1012                         goto out;
1013         }
1014         retval = pid_vnr(grp);
1015 out:
1016         rcu_read_unlock();
1017         return retval;
1018 }
1019
1020 #ifdef __ARCH_WANT_SYS_GETPGRP
1021
1022 SYSCALL_DEFINE0(getpgrp)
1023 {
1024         return sys_getpgid(0);
1025 }
1026
1027 #endif
1028
1029 SYSCALL_DEFINE1(getsid, pid_t, pid)
1030 {
1031         struct task_struct *p;
1032         struct pid *sid;
1033         int retval;
1034
1035         rcu_read_lock();
1036         if (!pid)
1037                 sid = task_session(current);
1038         else {
1039                 retval = -ESRCH;
1040                 p = find_task_by_vpid(pid);
1041                 if (!p)
1042                         goto out;
1043                 sid = task_session(p);
1044                 if (!sid)
1045                         goto out;
1046
1047                 retval = security_task_getsid(p);
1048                 if (retval)
1049                         goto out;
1050         }
1051         retval = pid_vnr(sid);
1052 out:
1053         rcu_read_unlock();
1054         return retval;
1055 }
1056
1057 SYSCALL_DEFINE0(setsid)
1058 {
1059         struct task_struct *group_leader = current->group_leader;
1060         struct pid *sid = task_pid(group_leader);
1061         pid_t session = pid_vnr(sid);
1062         int err = -EPERM;
1063
1064         write_lock_irq(&tasklist_lock);
1065         /* Fail if I am already a session leader */
1066         if (group_leader->signal->leader)
1067                 goto out;
1068
1069         /* Fail if a process group id already exists that equals the
1070          * proposed session id.
1071          */
1072         if (pid_task(sid, PIDTYPE_PGID))
1073                 goto out;
1074
1075         group_leader->signal->leader = 1;
1076         __set_special_pids(sid);
1077
1078         proc_clear_tty(group_leader);
1079
1080         err = session;
1081 out:
1082         write_unlock_irq(&tasklist_lock);
1083         if (err > 0)
1084                 proc_sid_connector(group_leader);
1085         return err;
1086 }
1087
1088 DECLARE_RWSEM(uts_sem);
1089
1090 #ifdef COMPAT_UTS_MACHINE
1091 #define override_architecture(name) \
1092         (personality(current->personality) == PER_LINUX32 && \
1093          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1094                       sizeof(COMPAT_UTS_MACHINE)))
1095 #else
1096 #define override_architecture(name)     0
1097 #endif
1098
1099 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1100 {
1101         int errno = 0;
1102
1103         down_read(&uts_sem);
1104         if (copy_to_user(name, utsname(), sizeof *name))
1105                 errno = -EFAULT;
1106         up_read(&uts_sem);
1107
1108         if (!errno && override_architecture(name))
1109                 errno = -EFAULT;
1110         return errno;
1111 }
1112
1113 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1114 /*
1115  * Old cruft
1116  */
1117 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1118 {
1119         int error = 0;
1120
1121         if (!name)
1122                 return -EFAULT;
1123
1124         down_read(&uts_sem);
1125         if (copy_to_user(name, utsname(), sizeof(*name)))
1126                 error = -EFAULT;
1127         up_read(&uts_sem);
1128
1129         if (!error && override_architecture(name))
1130                 error = -EFAULT;
1131         return error;
1132 }
1133
1134 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1135 {
1136         int error;
1137
1138         if (!name)
1139                 return -EFAULT;
1140         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1141                 return -EFAULT;
1142
1143         down_read(&uts_sem);
1144         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1145                                __OLD_UTS_LEN);
1146         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1147         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1148                                 __OLD_UTS_LEN);
1149         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1150         error |= __copy_to_user(&name->release, &utsname()->release,
1151                                 __OLD_UTS_LEN);
1152         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1153         error |= __copy_to_user(&name->version, &utsname()->version,
1154                                 __OLD_UTS_LEN);
1155         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1156         error |= __copy_to_user(&name->machine, &utsname()->machine,
1157                                 __OLD_UTS_LEN);
1158         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1159         up_read(&uts_sem);
1160
1161         if (!error && override_architecture(name))
1162                 error = -EFAULT;
1163         return error ? -EFAULT : 0;
1164 }
1165 #endif
1166
1167 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1168 {
1169         int errno;
1170         char tmp[__NEW_UTS_LEN];
1171
1172         if (!capable(CAP_SYS_ADMIN))
1173                 return -EPERM;
1174         if (len < 0 || len > __NEW_UTS_LEN)
1175                 return -EINVAL;
1176         down_write(&uts_sem);
1177         errno = -EFAULT;
1178         if (!copy_from_user(tmp, name, len)) {
1179                 struct new_utsname *u = utsname();
1180
1181                 memcpy(u->nodename, tmp, len);
1182                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1183                 errno = 0;
1184         }
1185         up_write(&uts_sem);
1186         return errno;
1187 }
1188
1189 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1190
1191 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1192 {
1193         int i, errno;
1194         struct new_utsname *u;
1195
1196         if (len < 0)
1197                 return -EINVAL;
1198         down_read(&uts_sem);
1199         u = utsname();
1200         i = 1 + strlen(u->nodename);
1201         if (i > len)
1202                 i = len;
1203         errno = 0;
1204         if (copy_to_user(name, u->nodename, i))
1205                 errno = -EFAULT;
1206         up_read(&uts_sem);
1207         return errno;
1208 }
1209
1210 #endif
1211
1212 /*
1213  * Only setdomainname; getdomainname can be implemented by calling
1214  * uname()
1215  */
1216 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1217 {
1218         int errno;
1219         char tmp[__NEW_UTS_LEN];
1220
1221         if (!capable(CAP_SYS_ADMIN))
1222                 return -EPERM;
1223         if (len < 0 || len > __NEW_UTS_LEN)
1224                 return -EINVAL;
1225
1226         down_write(&uts_sem);
1227         errno = -EFAULT;
1228         if (!copy_from_user(tmp, name, len)) {
1229                 struct new_utsname *u = utsname();
1230
1231                 memcpy(u->domainname, tmp, len);
1232                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1233                 errno = 0;
1234         }
1235         up_write(&uts_sem);
1236         return errno;
1237 }
1238
1239 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1240 {
1241         struct rlimit value;
1242         int ret;
1243
1244         ret = do_prlimit(current, resource, NULL, &value);
1245         if (!ret)
1246                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1247
1248         return ret;
1249 }
1250
1251 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1252
1253 /*
1254  *      Back compatibility for getrlimit. Needed for some apps.
1255  */
1256
1257 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1258                 struct rlimit __user *, rlim)
1259 {
1260         struct rlimit x;
1261         if (resource >= RLIM_NLIMITS)
1262                 return -EINVAL;
1263
1264         task_lock(current->group_leader);
1265         x = current->signal->rlim[resource];
1266         task_unlock(current->group_leader);
1267         if (x.rlim_cur > 0x7FFFFFFF)
1268                 x.rlim_cur = 0x7FFFFFFF;
1269         if (x.rlim_max > 0x7FFFFFFF)
1270                 x.rlim_max = 0x7FFFFFFF;
1271         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1272 }
1273
1274 #endif
1275
1276 static inline bool rlim64_is_infinity(__u64 rlim64)
1277 {
1278 #if BITS_PER_LONG < 64
1279         return rlim64 >= ULONG_MAX;
1280 #else
1281         return rlim64 == RLIM64_INFINITY;
1282 #endif
1283 }
1284
1285 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1286 {
1287         if (rlim->rlim_cur == RLIM_INFINITY)
1288                 rlim64->rlim_cur = RLIM64_INFINITY;
1289         else
1290                 rlim64->rlim_cur = rlim->rlim_cur;
1291         if (rlim->rlim_max == RLIM_INFINITY)
1292                 rlim64->rlim_max = RLIM64_INFINITY;
1293         else
1294                 rlim64->rlim_max = rlim->rlim_max;
1295 }
1296
1297 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1298 {
1299         if (rlim64_is_infinity(rlim64->rlim_cur))
1300                 rlim->rlim_cur = RLIM_INFINITY;
1301         else
1302                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1303         if (rlim64_is_infinity(rlim64->rlim_max))
1304                 rlim->rlim_max = RLIM_INFINITY;
1305         else
1306                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1307 }
1308
1309 /* make sure you are allowed to change @tsk limits before calling this */
1310 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1311                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1312 {
1313         struct rlimit *rlim;
1314         int retval = 0;
1315
1316         if (resource >= RLIM_NLIMITS)
1317                 return -EINVAL;
1318         if (new_rlim) {
1319                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1320                         return -EINVAL;
1321                 if (resource == RLIMIT_NOFILE &&
1322                                 new_rlim->rlim_max > sysctl_nr_open)
1323                         return -EPERM;
1324         }
1325
1326         /* protect tsk->signal and tsk->sighand from disappearing */
1327         read_lock(&tasklist_lock);
1328         if (!tsk->sighand) {
1329                 retval = -ESRCH;
1330                 goto out;
1331         }
1332
1333         rlim = tsk->signal->rlim + resource;
1334         task_lock(tsk->group_leader);
1335         if (new_rlim) {
1336                 if (new_rlim->rlim_max > rlim->rlim_max &&
1337                                 !capable(CAP_SYS_RESOURCE))
1338                         retval = -EPERM;
1339                 if (!retval)
1340                         retval = security_task_setrlimit(tsk->group_leader,
1341                                         resource, new_rlim);
1342                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1343                         /*
1344                          * The caller is asking for an immediate RLIMIT_CPU
1345                          * expiry.  But we use the zero value to mean "it was
1346                          * never set".  So let's cheat and make it one second
1347                          * instead
1348                          */
1349                         new_rlim->rlim_cur = 1;
1350                 }
1351         }
1352         if (!retval) {
1353                 if (old_rlim)
1354                         *old_rlim = *rlim;
1355                 if (new_rlim)
1356                         *rlim = *new_rlim;
1357         }
1358         task_unlock(tsk->group_leader);
1359
1360         /*
1361          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1362          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1363          * very long-standing error, and fixing it now risks breakage of
1364          * applications, so we live with it
1365          */
1366          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1367                          new_rlim->rlim_cur != RLIM_INFINITY)
1368                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1369 out:
1370         read_unlock(&tasklist_lock);
1371         return retval;
1372 }
1373
1374 /* rcu lock must be held */
1375 static int check_prlimit_permission(struct task_struct *task)
1376 {
1377         const struct cred *cred = current_cred(), *tcred;
1378
1379         tcred = __task_cred(task);
1380         if ((cred->uid != tcred->euid ||
1381              cred->uid != tcred->suid ||
1382              cred->uid != tcred->uid  ||
1383              cred->gid != tcred->egid ||
1384              cred->gid != tcred->sgid ||
1385              cred->gid != tcred->gid) &&
1386              !capable(CAP_SYS_RESOURCE)) {
1387                 return -EPERM;
1388         }
1389
1390         return 0;
1391 }
1392
1393 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1394                 const struct rlimit64 __user *, new_rlim,
1395                 struct rlimit64 __user *, old_rlim)
1396 {
1397         struct rlimit64 old64, new64;
1398         struct rlimit old, new;
1399         struct task_struct *tsk;
1400         int ret;
1401
1402         if (new_rlim) {
1403                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1404                         return -EFAULT;
1405                 rlim64_to_rlim(&new64, &new);
1406         }
1407
1408         rcu_read_lock();
1409         tsk = pid ? find_task_by_vpid(pid) : current;
1410         if (!tsk) {
1411                 rcu_read_unlock();
1412                 return -ESRCH;
1413         }
1414         ret = check_prlimit_permission(tsk);
1415         if (ret) {
1416                 rcu_read_unlock();
1417                 return ret;
1418         }
1419         get_task_struct(tsk);
1420         rcu_read_unlock();
1421
1422         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1423                         old_rlim ? &old : NULL);
1424
1425         if (!ret && old_rlim) {
1426                 rlim_to_rlim64(&old, &old64);
1427                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1428                         ret = -EFAULT;
1429         }
1430
1431         put_task_struct(tsk);
1432         return ret;
1433 }
1434
1435 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1436 {
1437         struct rlimit new_rlim;
1438
1439         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1440                 return -EFAULT;
1441         return do_prlimit(current, resource, &new_rlim, NULL);
1442 }
1443
1444 /*
1445  * It would make sense to put struct rusage in the task_struct,
1446  * except that would make the task_struct be *really big*.  After
1447  * task_struct gets moved into malloc'ed memory, it would
1448  * make sense to do this.  It will make moving the rest of the information
1449  * a lot simpler!  (Which we're not doing right now because we're not
1450  * measuring them yet).
1451  *
1452  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1453  * races with threads incrementing their own counters.  But since word
1454  * reads are atomic, we either get new values or old values and we don't
1455  * care which for the sums.  We always take the siglock to protect reading
1456  * the c* fields from p->signal from races with exit.c updating those
1457  * fields when reaping, so a sample either gets all the additions of a
1458  * given child after it's reaped, or none so this sample is before reaping.
1459  *
1460  * Locking:
1461  * We need to take the siglock for CHILDEREN, SELF and BOTH
1462  * for  the cases current multithreaded, non-current single threaded
1463  * non-current multithreaded.  Thread traversal is now safe with
1464  * the siglock held.
1465  * Strictly speaking, we donot need to take the siglock if we are current and
1466  * single threaded,  as no one else can take our signal_struct away, no one
1467  * else can  reap the  children to update signal->c* counters, and no one else
1468  * can race with the signal-> fields. If we do not take any lock, the
1469  * signal-> fields could be read out of order while another thread was just
1470  * exiting. So we should  place a read memory barrier when we avoid the lock.
1471  * On the writer side,  write memory barrier is implied in  __exit_signal
1472  * as __exit_signal releases  the siglock spinlock after updating the signal->
1473  * fields. But we don't do this yet to keep things simple.
1474  *
1475  */
1476
1477 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1478 {
1479         r->ru_nvcsw += t->nvcsw;
1480         r->ru_nivcsw += t->nivcsw;
1481         r->ru_minflt += t->min_flt;
1482         r->ru_majflt += t->maj_flt;
1483         r->ru_inblock += task_io_get_inblock(t);
1484         r->ru_oublock += task_io_get_oublock(t);
1485 }
1486
1487 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1488 {
1489         struct task_struct *t;
1490         unsigned long flags;
1491         cputime_t tgutime, tgstime, utime, stime;
1492         unsigned long maxrss = 0;
1493
1494         memset((char *) r, 0, sizeof *r);
1495         utime = stime = cputime_zero;
1496
1497         if (who == RUSAGE_THREAD) {
1498                 task_times(current, &utime, &stime);
1499                 accumulate_thread_rusage(p, r);
1500                 maxrss = p->signal->maxrss;
1501                 goto out;
1502         }
1503
1504         if (!lock_task_sighand(p, &flags))
1505                 return;
1506
1507         switch (who) {
1508                 case RUSAGE_BOTH:
1509                 case RUSAGE_CHILDREN:
1510                         utime = p->signal->cutime;
1511                         stime = p->signal->cstime;
1512                         r->ru_nvcsw = p->signal->cnvcsw;
1513                         r->ru_nivcsw = p->signal->cnivcsw;
1514                         r->ru_minflt = p->signal->cmin_flt;
1515                         r->ru_majflt = p->signal->cmaj_flt;
1516                         r->ru_inblock = p->signal->cinblock;
1517                         r->ru_oublock = p->signal->coublock;
1518                         maxrss = p->signal->cmaxrss;
1519
1520                         if (who == RUSAGE_CHILDREN)
1521                                 break;
1522
1523                 case RUSAGE_SELF:
1524                         thread_group_times(p, &tgutime, &tgstime);
1525                         utime = cputime_add(utime, tgutime);
1526                         stime = cputime_add(stime, tgstime);
1527                         r->ru_nvcsw += p->signal->nvcsw;
1528                         r->ru_nivcsw += p->signal->nivcsw;
1529                         r->ru_minflt += p->signal->min_flt;
1530                         r->ru_majflt += p->signal->maj_flt;
1531                         r->ru_inblock += p->signal->inblock;
1532                         r->ru_oublock += p->signal->oublock;
1533                         if (maxrss < p->signal->maxrss)
1534                                 maxrss = p->signal->maxrss;
1535                         t = p;
1536                         do {
1537                                 accumulate_thread_rusage(t, r);
1538                                 t = next_thread(t);
1539                         } while (t != p);
1540                         break;
1541
1542                 default:
1543                         BUG();
1544         }
1545         unlock_task_sighand(p, &flags);
1546
1547 out:
1548         cputime_to_timeval(utime, &r->ru_utime);
1549         cputime_to_timeval(stime, &r->ru_stime);
1550
1551         if (who != RUSAGE_CHILDREN) {
1552                 struct mm_struct *mm = get_task_mm(p);
1553                 if (mm) {
1554                         setmax_mm_hiwater_rss(&maxrss, mm);
1555                         mmput(mm);
1556                 }
1557         }
1558         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1559 }
1560
1561 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1562 {
1563         struct rusage r;
1564         k_getrusage(p, who, &r);
1565         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1566 }
1567
1568 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1569 {
1570         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1571             who != RUSAGE_THREAD)
1572                 return -EINVAL;
1573         return getrusage(current, who, ru);
1574 }
1575
1576 SYSCALL_DEFINE1(umask, int, mask)
1577 {
1578         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1579         return mask;
1580 }
1581
1582 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1583                 unsigned long, arg4, unsigned long, arg5)
1584 {
1585         struct task_struct *me = current;
1586         unsigned char comm[sizeof(me->comm)];
1587         long error;
1588
1589         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1590         if (error != -ENOSYS)
1591                 return error;
1592
1593         error = 0;
1594         switch (option) {
1595                 case PR_SET_PDEATHSIG:
1596                         if (!valid_signal(arg2)) {
1597                                 error = -EINVAL;
1598                                 break;
1599                         }
1600                         me->pdeath_signal = arg2;
1601                         error = 0;
1602                         break;
1603                 case PR_GET_PDEATHSIG:
1604                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1605                         break;
1606                 case PR_GET_DUMPABLE:
1607                         error = get_dumpable(me->mm);
1608                         break;
1609                 case PR_SET_DUMPABLE:
1610                         if (arg2 < 0 || arg2 > 1) {
1611                                 error = -EINVAL;
1612                                 break;
1613                         }
1614                         set_dumpable(me->mm, arg2);
1615                         error = 0;
1616                         break;
1617
1618                 case PR_SET_UNALIGN:
1619                         error = SET_UNALIGN_CTL(me, arg2);
1620                         break;
1621                 case PR_GET_UNALIGN:
1622                         error = GET_UNALIGN_CTL(me, arg2);
1623                         break;
1624                 case PR_SET_FPEMU:
1625                         error = SET_FPEMU_CTL(me, arg2);
1626                         break;
1627                 case PR_GET_FPEMU:
1628                         error = GET_FPEMU_CTL(me, arg2);
1629                         break;
1630                 case PR_SET_FPEXC:
1631                         error = SET_FPEXC_CTL(me, arg2);
1632                         break;
1633                 case PR_GET_FPEXC:
1634                         error = GET_FPEXC_CTL(me, arg2);
1635                         break;
1636                 case PR_GET_TIMING:
1637                         error = PR_TIMING_STATISTICAL;
1638                         break;
1639                 case PR_SET_TIMING:
1640                         if (arg2 != PR_TIMING_STATISTICAL)
1641                                 error = -EINVAL;
1642                         else
1643                                 error = 0;
1644                         break;
1645
1646                 case PR_SET_NAME:
1647                         comm[sizeof(me->comm)-1] = 0;
1648                         if (strncpy_from_user(comm, (char __user *)arg2,
1649                                               sizeof(me->comm) - 1) < 0)
1650                                 return -EFAULT;
1651                         set_task_comm(me, comm);
1652                         return 0;
1653                 case PR_GET_NAME:
1654                         get_task_comm(comm, me);
1655                         if (copy_to_user((char __user *)arg2, comm,
1656                                          sizeof(comm)))
1657                                 return -EFAULT;
1658                         return 0;
1659                 case PR_GET_ENDIAN:
1660                         error = GET_ENDIAN(me, arg2);
1661                         break;
1662                 case PR_SET_ENDIAN:
1663                         error = SET_ENDIAN(me, arg2);
1664                         break;
1665
1666                 case PR_GET_SECCOMP:
1667                         error = prctl_get_seccomp();
1668                         break;
1669                 case PR_SET_SECCOMP:
1670                         error = prctl_set_seccomp(arg2);
1671                         break;
1672                 case PR_GET_TSC:
1673                         error = GET_TSC_CTL(arg2);
1674                         break;
1675                 case PR_SET_TSC:
1676                         error = SET_TSC_CTL(arg2);
1677                         break;
1678                 case PR_TASK_PERF_EVENTS_DISABLE:
1679                         error = perf_event_task_disable();
1680                         break;
1681                 case PR_TASK_PERF_EVENTS_ENABLE:
1682                         error = perf_event_task_enable();
1683                         break;
1684                 case PR_GET_TIMERSLACK:
1685                         error = current->timer_slack_ns;
1686                         break;
1687                 case PR_SET_TIMERSLACK:
1688                         if (arg2 <= 0)
1689                                 current->timer_slack_ns =
1690                                         current->default_timer_slack_ns;
1691                         else
1692                                 current->timer_slack_ns = arg2;
1693                         error = 0;
1694                         break;
1695                 case PR_MCE_KILL:
1696                         if (arg4 | arg5)
1697                                 return -EINVAL;
1698                         switch (arg2) {
1699                         case PR_MCE_KILL_CLEAR:
1700                                 if (arg3 != 0)
1701                                         return -EINVAL;
1702                                 current->flags &= ~PF_MCE_PROCESS;
1703                                 break;
1704                         case PR_MCE_KILL_SET:
1705                                 current->flags |= PF_MCE_PROCESS;
1706                                 if (arg3 == PR_MCE_KILL_EARLY)
1707                                         current->flags |= PF_MCE_EARLY;
1708                                 else if (arg3 == PR_MCE_KILL_LATE)
1709                                         current->flags &= ~PF_MCE_EARLY;
1710                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1711                                         current->flags &=
1712                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1713                                 else
1714                                         return -EINVAL;
1715                                 break;
1716                         default:
1717                                 return -EINVAL;
1718                         }
1719                         error = 0;
1720                         break;
1721                 case PR_MCE_KILL_GET:
1722                         if (arg2 | arg3 | arg4 | arg5)
1723                                 return -EINVAL;
1724                         if (current->flags & PF_MCE_PROCESS)
1725                                 error = (current->flags & PF_MCE_EARLY) ?
1726                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1727                         else
1728                                 error = PR_MCE_KILL_DEFAULT;
1729                         break;
1730                 default:
1731                         error = -EINVAL;
1732                         break;
1733         }
1734         return error;
1735 }
1736
1737 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1738                 struct getcpu_cache __user *, unused)
1739 {
1740         int err = 0;
1741         int cpu = raw_smp_processor_id();
1742         if (cpup)
1743                 err |= put_user(cpu, cpup);
1744         if (nodep)
1745                 err |= put_user(cpu_to_node(cpu), nodep);
1746         return err ? -EFAULT : 0;
1747 }
1748
1749 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1750
1751 static void argv_cleanup(struct subprocess_info *info)
1752 {
1753         argv_free(info->argv);
1754 }
1755
1756 /**
1757  * orderly_poweroff - Trigger an orderly system poweroff
1758  * @force: force poweroff if command execution fails
1759  *
1760  * This may be called from any context to trigger a system shutdown.
1761  * If the orderly shutdown fails, it will force an immediate shutdown.
1762  */
1763 int orderly_poweroff(bool force)
1764 {
1765         int argc;
1766         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1767         static char *envp[] = {
1768                 "HOME=/",
1769                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1770                 NULL
1771         };
1772         int ret = -ENOMEM;
1773         struct subprocess_info *info;
1774
1775         if (argv == NULL) {
1776                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1777                        __func__, poweroff_cmd);
1778                 goto out;
1779         }
1780
1781         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1782         if (info == NULL) {
1783                 argv_free(argv);
1784                 goto out;
1785         }
1786
1787         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1788
1789         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1790
1791   out:
1792         if (ret && force) {
1793                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1794                        "forcing the issue\n");
1795
1796                 /* I guess this should try to kick off some daemon to
1797                    sync and poweroff asap.  Or not even bother syncing
1798                    if we're doing an emergency shutdown? */
1799                 emergency_sync();
1800                 kernel_power_off();
1801         }
1802
1803         return ret;
1804 }
1805 EXPORT_SYMBOL_GPL(orderly_poweroff);