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 #include <linux/syscore_ops.h>
  41
  42 #include <linux/compat.h>
  43 #include <linux/syscalls.h>
  44 #include <linux/kprobes.h>
  45 #include <linux/user_namespace.h>
  46
  47 #include <linux/kmsg_dump.h>
  48
  49 #include <asm/uaccess.h>
  50 #include <asm/io.h>
  51 #include <asm/unistd.h>
  52
  53 #ifndef SET_UNALIGN_CTL
  54 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  55 #endif
  56 #ifndef GET_UNALIGN_CTL
  57 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  58 #endif
  59 #ifndef SET_FPEMU_CTL
  60 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  61 #endif
  62 #ifndef GET_FPEMU_CTL
  63 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  64 #endif
  65 #ifndef SET_FPEXC_CTL
  66 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  67 #endif
  68 #ifndef GET_FPEXC_CTL
  69 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  70 #endif
  71 #ifndef GET_ENDIAN
  72 # define GET_ENDIAN(a,b)        (-EINVAL)
  73 #endif
  74 #ifndef SET_ENDIAN
  75 # define SET_ENDIAN(a,b)        (-EINVAL)
  76 #endif
  77 #ifndef GET_TSC_CTL
  78 # define GET_TSC_CTL(a)         (-EINVAL)
  79 #endif
  80 #ifndef SET_TSC_CTL
  81 # define SET_TSC_CTL(a)         (-EINVAL)
  82 #endif
  83
  84 /*
  85  * this is where the system-wide overflow UID and GID are defined, for
  86  * architectures that now have 32-bit UID/GID but didn't in the past
  87  */
  88
  89 int overflowuid = DEFAULT_OVERFLOWUID;
  90 int overflowgid = DEFAULT_OVERFLOWGID;
  91
  92 #ifdef CONFIG_UID16
  93 EXPORT_SYMBOL(overflowuid);
  94 EXPORT_SYMBOL(overflowgid);
  95 #endif
  96
  97 /*
  98  * the same as above, but for filesystems which can only store a 16-bit
  99  * UID and GID. as such, this is needed on all architectures
 100  */
 101
 102 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 103 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 104
 105 EXPORT_SYMBOL(fs_overflowuid);
 106 EXPORT_SYMBOL(fs_overflowgid);
 107
 108 /*
 109  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 110  */
 111
 112 int C_A_D = 1;
 113 struct pid *cad_pid;
 114 EXPORT_SYMBOL(cad_pid);
 115
 116 /*
 117  * If set, this is used for preparing the system to power off.
 118  */
 119
 120 void (*pm_power_off_prepare)(void);
 121
 122 /*
 123  * Returns true if current's euid is same as p's uid or euid,
 124  * or has CAP_SYS_NICE to p's user_ns.
 125  *
 126  * Called with rcu_read_lock, creds are safe
 127  */
 128 static bool set_one_prio_perm(struct task_struct *p)
 129 {
 130         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 131
 132         if (pcred->user->user_ns == cred->user->user_ns &&
 133             (pcred->uid  == cred->euid ||
 134              pcred->euid == cred->euid))
 135                 return true;
 136         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 137                 return true;
 138         return false;
 139 }
 140
 141 /*
 142  * set the priority of a task
 143  * - the caller must hold the RCU read lock
 144  */
 145 static int set_one_prio(struct task_struct *p, int niceval, int error)
 146 {
 147         int no_nice;
 148
 149         if (!set_one_prio_perm(p)) {
 150                 error = -EPERM;
 151                 goto out;
 152         }
 153         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 154                 error = -EACCES;
 155                 goto out;
 156         }
 157         no_nice = security_task_setnice(p, niceval);
 158         if (no_nice) {
 159                 error = no_nice;
 160                 goto out;
 161         }
 162         if (error == -ESRCH)
 163                 error = 0;
 164         set_user_nice(p, niceval);
 165 out:
 166         return error;
 167 }
 168
 169 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 170 {
 171         struct task_struct *g, *p;
 172         struct user_struct *user;
 173         const struct cred *cred = current_cred();
 174         int error = -EINVAL;
 175         struct pid *pgrp;
 176
 177         if (which > PRIO_USER || which < PRIO_PROCESS)
 178                 goto out;
 179
 180         /* normalize: avoid signed division (rounding problems) */
 181         error = -ESRCH;
 182         if (niceval < -20)
 183                 niceval = -20;
 184         if (niceval > 19)
 185                 niceval = 19;
 186
 187         rcu_read_lock();
 188         read_lock(&tasklist_lock);
 189         switch (which) {
 190                 case PRIO_PROCESS:
 191                         if (who)
 192                                 p = find_task_by_vpid(who);
 193                         else
 194                                 p = current;
 195                         if (p)
 196                                 error = set_one_prio(p, niceval, error);
 197                         break;
 198                 case PRIO_PGRP:
 199                         if (who)
 200                                 pgrp = find_vpid(who);
 201                         else
 202                                 pgrp = task_pgrp(current);
 203                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 204                                 error = set_one_prio(p, niceval, error);
 205                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 206                         break;
 207                 case PRIO_USER:
 208                         user = (struct user_struct *) cred->user;
 209                         if (!who)
 210                                 who = cred->uid;
 211                         else if ((who != cred->uid) &&
 212                                  !(user = find_user(who)))
 213                                 goto out_unlock;        /* No processes for this user */
 214
 215                         do_each_thread(g, p) {
 216                                 if (__task_cred(p)->uid == who)
 217                                         error = set_one_prio(p, niceval, error);
 218                         } while_each_thread(g, p);
 219                         if (who != cred->uid)
 220                                 free_uid(user);         /* For find_user() */
 221                         break;
 222         }
 223 out_unlock:
 224         read_unlock(&tasklist_lock);
 225         rcu_read_unlock();
 226 out:
 227         return error;
 228 }
 229
 230 /*
 231  * Ugh. To avoid negative return values, "getpriority()" will
 232  * not return the normal nice-value, but a negated value that
 233  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 234  * to stay compatible.
 235  */
 236 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 237 {
 238         struct task_struct *g, *p;
 239         struct user_struct *user;
 240         const struct cred *cred = current_cred();
 241         long niceval, retval = -ESRCH;
 242         struct pid *pgrp;
 243
 244         if (which > PRIO_USER || which < PRIO_PROCESS)
 245                 return -EINVAL;
 246
 247         rcu_read_lock();
 248         read_lock(&tasklist_lock);
 249         switch (which) {
 250                 case PRIO_PROCESS:
 251                         if (who)
 252                                 p = find_task_by_vpid(who);
 253                         else
 254                                 p = current;
 255                         if (p) {
 256                                 niceval = 20 - task_nice(p);
 257                                 if (niceval > retval)
 258                                         retval = niceval;
 259                         }
 260                         break;
 261                 case PRIO_PGRP:
 262                         if (who)
 263                                 pgrp = find_vpid(who);
 264                         else
 265                                 pgrp = task_pgrp(current);
 266                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 267                                 niceval = 20 - task_nice(p);
 268                                 if (niceval > retval)
 269                                         retval = niceval;
 270                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 271                         break;
 272                 case PRIO_USER:
 273                         user = (struct user_struct *) cred->user;
 274                         if (!who)
 275                                 who = cred->uid;
 276                         else if ((who != cred->uid) &&
 277                                  !(user = find_user(who)))
 278                                 goto out_unlock;        /* No processes for this user */
 279
 280                         do_each_thread(g, p) {
 281                                 if (__task_cred(p)->uid == who) {
 282                                         niceval = 20 - task_nice(p);
 283                                         if (niceval > retval)
 284                                                 retval = niceval;
 285                                 }
 286                         } while_each_thread(g, p);
 287                         if (who != cred->uid)
 288                                 free_uid(user);         /* for find_user() */
 289                         break;
 290         }
 291 out_unlock:
 292         read_unlock(&tasklist_lock);
 293         rcu_read_unlock();
 294
 295         return retval;
 296 }
 297
 298 /**
 299  *      emergency_restart - reboot the system
 300  *
 301  *      Without shutting down any hardware or taking any locks
 302  *      reboot the system.  This is called when we know we are in
 303  *      trouble so this is our best effort to reboot.  This is
 304  *      safe to call in interrupt context.
 305  */
 306 void emergency_restart(void)
 307 {
 308         kmsg_dump(KMSG_DUMP_EMERG);
 309         machine_emergency_restart();
 310 }
 311 EXPORT_SYMBOL_GPL(emergency_restart);
 312
 313 void kernel_restart_prepare(char *cmd)
 314 {
 315         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 316         system_state = SYSTEM_RESTART;
 317         device_shutdown();
 318         syscore_shutdown();
 319 }
 320
 321 /**
 322  *      kernel_restart - reboot the system
 323  *      @cmd: pointer to buffer containing command to execute for restart
 324  *              or %NULL
 325  *
 326  *      Shutdown everything and perform a clean reboot.
 327  *      This is not safe to call in interrupt context.
 328  */
 329 void kernel_restart(char *cmd)
 330 {
 331         kernel_restart_prepare(cmd);
 332         if (!cmd)
 333                 printk(KERN_EMERG "Restarting system.\n");
 334         else
 335                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 336         kmsg_dump(KMSG_DUMP_RESTART);
 337         machine_restart(cmd);
 338 }
 339 EXPORT_SYMBOL_GPL(kernel_restart);
 340
 341 static void kernel_shutdown_prepare(enum system_states state)
 342 {
 343         blocking_notifier_call_chain(&reboot_notifier_list,
 344                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 345         system_state = state;
 346         device_shutdown();
 347 }
 348 /**
 349  *      kernel_halt - halt the system
 350  *
 351  *      Shutdown everything and perform a clean system halt.
 352  */
 353 void kernel_halt(void)
 354 {
 355         kernel_shutdown_prepare(SYSTEM_HALT);
 356         syscore_shutdown();
 357         printk(KERN_EMERG "System halted.\n");
 358         kmsg_dump(KMSG_DUMP_HALT);
 359         machine_halt();
 360 }
 361
 362 EXPORT_SYMBOL_GPL(kernel_halt);
 363
 364 /**
 365  *      kernel_power_off - power_off the system
 366  *
 367  *      Shutdown everything and perform a clean system power_off.
 368  */
 369 void kernel_power_off(void)
 370 {
 371         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 372         if (pm_power_off_prepare)
 373                 pm_power_off_prepare();
 374         disable_nonboot_cpus();
 375         syscore_shutdown();
 376         printk(KERN_EMERG "Power down.\n");
 377         kmsg_dump(KMSG_DUMP_POWEROFF);
 378         machine_power_off();
 379 }
 380 EXPORT_SYMBOL_GPL(kernel_power_off);
 381
 382 static DEFINE_MUTEX(reboot_mutex);
 383
 384 /*
 385  * Reboot system call: for obvious reasons only root may call it,
 386  * and even root needs to set up some magic numbers in the registers
 387  * so that some mistake won't make this reboot the whole machine.
 388  * You can also set the meaning of the ctrl-alt-del-key here.
 389  *
 390  * reboot doesn't sync: do that yourself before calling this.
 391  */
 392 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 393                 void __user *, arg)
 394 {
 395         char buffer[256];
 396         int ret = 0;
 397
 398         /* We only trust the superuser with rebooting the system. */
 399         if (!capable(CAP_SYS_BOOT))
 400                 return -EPERM;
 401
 402         /* For safety, we require "magic" arguments. */
 403         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 404             (magic2 != LINUX_REBOOT_MAGIC2 &&
 405                         magic2 != LINUX_REBOOT_MAGIC2A &&
 406                         magic2 != LINUX_REBOOT_MAGIC2B &&
 407                         magic2 != LINUX_REBOOT_MAGIC2C))
 408                 return -EINVAL;
 409
 410         /* Instead of trying to make the power_off code look like
 411          * halt when pm_power_off is not set do it the easy way.
 412          */
 413         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 414                 cmd = LINUX_REBOOT_CMD_HALT;
 415
 416         mutex_lock(&reboot_mutex);
 417         switch (cmd) {
 418         case LINUX_REBOOT_CMD_RESTART:
 419                 kernel_restart(NULL);
 420                 break;
 421
 422         case LINUX_REBOOT_CMD_CAD_ON:
 423                 C_A_D = 1;
 424                 break;
 425
 426         case LINUX_REBOOT_CMD_CAD_OFF:
 427                 C_A_D = 0;
 428                 break;
 429
 430         case LINUX_REBOOT_CMD_HALT:
 431                 kernel_halt();
 432                 do_exit(0);
 433                 panic("cannot halt");
 434
 435         case LINUX_REBOOT_CMD_POWER_OFF:
 436                 kernel_power_off();
 437                 do_exit(0);
 438                 break;
 439
 440         case LINUX_REBOOT_CMD_RESTART2:
 441                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 442                         ret = -EFAULT;
 443                         break;
 444                 }
 445                 buffer[sizeof(buffer) - 1] = '\0';
 446
 447                 kernel_restart(buffer);
 448                 break;
 449
 450 #ifdef CONFIG_KEXEC
 451         case LINUX_REBOOT_CMD_KEXEC:
 452                 ret = kernel_kexec();
 453                 break;
 454 #endif
 455
 456 #ifdef CONFIG_HIBERNATION
 457         case LINUX_REBOOT_CMD_SW_SUSPEND:
 458                 ret = hibernate();
 459                 break;
 460 #endif
 461
 462         default:
 463                 ret = -EINVAL;
 464                 break;
 465         }
 466         mutex_unlock(&reboot_mutex);
 467         return ret;
 468 }
 469
 470 static void deferred_cad(struct work_struct *dummy)
 471 {
 472         kernel_restart(NULL);
 473 }
 474
 475 /*
 476  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 477  * As it's called within an interrupt, it may NOT sync: the only choice
 478  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 479  */
 480 void ctrl_alt_del(void)
 481 {
 482         static DECLARE_WORK(cad_work, deferred_cad);
 483
 484         if (C_A_D)
 485                 schedule_work(&cad_work);
 486         else
 487                 kill_cad_pid(SIGINT, 1);
 488 }
 489
 490 /*
 491  * Unprivileged users may change the real gid to the effective gid
 492  * or vice versa.  (BSD-style)
 493  *
 494  * If you set the real gid at all, or set the effective gid to a value not
 495  * equal to the real gid, then the saved gid is set to the new effective gid.
 496  *
 497  * This makes it possible for a setgid program to completely drop its
 498  * privileges, which is often a useful assertion to make when you are doing
 499  * a security audit over a program.
 500  *
 501  * The general idea is that a program which uses just setregid() will be
 502  * 100% compatible with BSD.  A program which uses just setgid() will be
 503  * 100% compatible with POSIX with saved IDs.
 504  *
 505  * SMP: There are not races, the GIDs are checked only by filesystem
 506  *      operations (as far as semantic preservation is concerned).
 507  */
 508 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 509 {
 510         const struct cred *old;
 511         struct cred *new;
 512         int retval;
 513
 514         new = prepare_creds();
 515         if (!new)
 516                 return -ENOMEM;
 517         old = current_cred();
 518
 519         retval = -EPERM;
 520         if (rgid != (gid_t) -1) {
 521                 if (old->gid == rgid ||
 522                     old->egid == rgid ||
 523                     nsown_capable(CAP_SETGID))
 524                         new->gid = rgid;
 525                 else
 526                         goto error;
 527         }
 528         if (egid != (gid_t) -1) {
 529                 if (old->gid == egid ||
 530                     old->egid == egid ||
 531                     old->sgid == egid ||
 532                     nsown_capable(CAP_SETGID))
 533                         new->egid = egid;
 534                 else
 535                         goto error;
 536         }
 537
 538         if (rgid != (gid_t) -1 ||
 539             (egid != (gid_t) -1 && egid != old->gid))
 540                 new->sgid = new->egid;
 541         new->fsgid = new->egid;
 542
 543         return commit_creds(new);
 544
 545 error:
 546         abort_creds(new);
 547         return retval;
 548 }
 549
 550 /*
 551  * setgid() is implemented like SysV w/ SAVED_IDS
 552  *
 553  * SMP: Same implicit races as above.
 554  */
 555 SYSCALL_DEFINE1(setgid, gid_t, gid)
 556 {
 557         const struct cred *old;
 558         struct cred *new;
 559         int retval;
 560
 561         new = prepare_creds();
 562         if (!new)
 563                 return -ENOMEM;
 564         old = current_cred();
 565
 566         retval = -EPERM;
 567         if (nsown_capable(CAP_SETGID))
 568                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 569         else if (gid == old->gid || gid == old->sgid)
 570                 new->egid = new->fsgid = gid;
 571         else
 572                 goto error;
 573
 574         return commit_creds(new);
 575
 576 error:
 577         abort_creds(new);
 578         return retval;
 579 }
 580
 581 /*
 582  * change the user struct in a credentials set to match the new UID
 583  */
 584 static int set_user(struct cred *new)
 585 {
 586         struct user_struct *new_user;
 587
 588         new_user = alloc_uid(current_user_ns(), new->uid);
 589         if (!new_user)
 590                 return -EAGAIN;
 591
 592         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 593                         new_user != INIT_USER) {
 594                 free_uid(new_user);
 595                 return -EAGAIN;
 596         }
 597
 598         free_uid(new->user);
 599         new->user = new_user;
 600         return 0;
 601 }
 602
 603 /*
 604  * Unprivileged users may change the real uid to the effective uid
 605  * or vice versa.  (BSD-style)
 606  *
 607  * If you set the real uid at all, or set the effective uid to a value not
 608  * equal to the real uid, then the saved uid is set to the new effective uid.
 609  *
 610  * This makes it possible for a setuid program to completely drop its
 611  * privileges, which is often a useful assertion to make when you are doing
 612  * a security audit over a program.
 613  *
 614  * The general idea is that a program which uses just setreuid() will be
 615  * 100% compatible with BSD.  A program which uses just setuid() will be
 616  * 100% compatible with POSIX with saved IDs.
 617  */
 618 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 619 {
 620         const struct cred *old;
 621         struct cred *new;
 622         int retval;
 623
 624         new = prepare_creds();
 625         if (!new)
 626                 return -ENOMEM;
 627         old = current_cred();
 628
 629         retval = -EPERM;
 630         if (ruid != (uid_t) -1) {
 631                 new->uid = ruid;
 632                 if (old->uid != ruid &&
 633                     old->euid != ruid &&
 634                     !nsown_capable(CAP_SETUID))
 635                         goto error;
 636         }
 637
 638         if (euid != (uid_t) -1) {
 639                 new->euid = euid;
 640                 if (old->uid != euid &&
 641                     old->euid != euid &&
 642                     old->suid != euid &&
 643                     !nsown_capable(CAP_SETUID))
 644                         goto error;
 645         }
 646
 647         if (new->uid != old->uid) {
 648                 retval = set_user(new);
 649                 if (retval < 0)
 650                         goto error;
 651         }
 652         if (ruid != (uid_t) -1 ||
 653             (euid != (uid_t) -1 && euid != old->uid))
 654                 new->suid = new->euid;
 655         new->fsuid = new->euid;
 656
 657         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 658         if (retval < 0)
 659                 goto error;
 660
 661         return commit_creds(new);
 662
 663 error:
 664         abort_creds(new);
 665         return retval;
 666 }
 667
 668 /*
 669  * setuid() is implemented like SysV with SAVED_IDS
 670  *
 671  * Note that SAVED_ID's is deficient in that a setuid root program
 672  * like sendmail, for example, cannot set its uid to be a normal
 673  * user and then switch back, because if you're root, setuid() sets
 674  * the saved uid too.  If you don't like this, blame the bright people
 675  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 676  * will allow a root program to temporarily drop privileges and be able to
 677  * regain them by swapping the real and effective uid.
 678  */
 679 SYSCALL_DEFINE1(setuid, uid_t, uid)
 680 {
 681         const struct cred *old;
 682         struct cred *new;
 683         int retval;
 684
 685         new = prepare_creds();
 686         if (!new)
 687                 return -ENOMEM;
 688         old = current_cred();
 689
 690         retval = -EPERM;
 691         if (nsown_capable(CAP_SETUID)) {
 692                 new->suid = new->uid = uid;
 693                 if (uid != old->uid) {
 694                         retval = set_user(new);
 695                         if (retval < 0)
 696                                 goto error;
 697                 }
 698         } else if (uid != old->uid && uid != new->suid) {
 699                 goto error;
 700         }
 701
 702         new->fsuid = new->euid = uid;
 703
 704         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 705         if (retval < 0)
 706                 goto error;
 707
 708         return commit_creds(new);
 709
 710 error:
 711         abort_creds(new);
 712         return retval;
 713 }
 714
 715
 716 /*
 717  * This function implements a generic ability to update ruid, euid,
 718  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 719  */
 720 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 721 {
 722         const struct cred *old;
 723         struct cred *new;
 724         int retval;
 725
 726         new = prepare_creds();
 727         if (!new)
 728                 return -ENOMEM;
 729
 730         old = current_cred();
 731
 732         retval = -EPERM;
 733         if (!nsown_capable(CAP_SETUID)) {
 734                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 735                     ruid != old->euid  && ruid != old->suid)
 736                         goto error;
 737                 if (euid != (uid_t) -1 && euid != old->uid &&
 738                     euid != old->euid  && euid != old->suid)
 739                         goto error;
 740                 if (suid != (uid_t) -1 && suid != old->uid &&
 741                     suid != old->euid  && suid != old->suid)
 742                         goto error;
 743         }
 744
 745         if (ruid != (uid_t) -1) {
 746                 new->uid = ruid;
 747                 if (ruid != old->uid) {
 748                         retval = set_user(new);
 749                         if (retval < 0)
 750                                 goto error;
 751                 }
 752         }
 753         if (euid != (uid_t) -1)
 754                 new->euid = euid;
 755         if (suid != (uid_t) -1)
 756                 new->suid = suid;
 757         new->fsuid = new->euid;
 758
 759         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 760         if (retval < 0)
 761                 goto error;
 762
 763         return commit_creds(new);
 764
 765 error:
 766         abort_creds(new);
 767         return retval;
 768 }
 769
 770 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 771 {
 772         const struct cred *cred = current_cred();
 773         int retval;
 774
 775         if (!(retval   = put_user(cred->uid,  ruid)) &&
 776             !(retval   = put_user(cred->euid, euid)))
 777                 retval = put_user(cred->suid, suid);
 778
 779         return retval;
 780 }
 781
 782 /*
 783  * Same as above, but for rgid, egid, sgid.
 784  */
 785 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 786 {
 787         const struct cred *old;
 788         struct cred *new;
 789         int retval;
 790
 791         new = prepare_creds();
 792         if (!new)
 793                 return -ENOMEM;
 794         old = current_cred();
 795
 796         retval = -EPERM;
 797         if (!nsown_capable(CAP_SETGID)) {
 798                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 799                     rgid != old->egid  && rgid != old->sgid)
 800                         goto error;
 801                 if (egid != (gid_t) -1 && egid != old->gid &&
 802                     egid != old->egid  && egid != old->sgid)
 803                         goto error;
 804                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 805                     sgid != old->egid  && sgid != old->sgid)
 806                         goto error;
 807         }
 808
 809         if (rgid != (gid_t) -1)
 810                 new->gid = rgid;
 811         if (egid != (gid_t) -1)
 812                 new->egid = egid;
 813         if (sgid != (gid_t) -1)
 814                 new->sgid = sgid;
 815         new->fsgid = new->egid;
 816
 817         return commit_creds(new);
 818
 819 error:
 820         abort_creds(new);
 821         return retval;
 822 }
 823
 824 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 825 {
 826         const struct cred *cred = current_cred();
 827         int retval;
 828
 829         if (!(retval   = put_user(cred->gid,  rgid)) &&
 830             !(retval   = put_user(cred->egid, egid)))
 831                 retval = put_user(cred->sgid, sgid);
 832
 833         return retval;
 834 }
 835
 836
 837 /*
 838  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 839  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 840  * whatever uid it wants to). It normally shadows "euid", except when
 841  * explicitly set by setfsuid() or for access..
 842  */
 843 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 844 {
 845         const struct cred *old;
 846         struct cred *new;
 847         uid_t old_fsuid;
 848
 849         new = prepare_creds();
 850         if (!new)
 851                 return current_fsuid();
 852         old = current_cred();
 853         old_fsuid = old->fsuid;
 854
 855         if (uid == old->uid  || uid == old->euid  ||
 856             uid == old->suid || uid == old->fsuid ||
 857             nsown_capable(CAP_SETUID)) {
 858                 if (uid != old_fsuid) {
 859                         new->fsuid = uid;
 860                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 861                                 goto change_okay;
 862                 }
 863         }
 864
 865         abort_creds(new);
 866         return old_fsuid;
 867
 868 change_okay:
 869         commit_creds(new);
 870         return old_fsuid;
 871 }
 872
 873 /*
 874  * Samma på svenska..
 875  */
 876 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 877 {
 878         const struct cred *old;
 879         struct cred *new;
 880         gid_t old_fsgid;
 881
 882         new = prepare_creds();
 883         if (!new)
 884                 return current_fsgid();
 885         old = current_cred();
 886         old_fsgid = old->fsgid;
 887
 888         if (gid == old->gid  || gid == old->egid  ||
 889             gid == old->sgid || gid == old->fsgid ||
 890             nsown_capable(CAP_SETGID)) {
 891                 if (gid != old_fsgid) {
 892                         new->fsgid = gid;
 893                         goto change_okay;
 894                 }
 895         }
 896
 897         abort_creds(new);
 898         return old_fsgid;
 899
 900 change_okay:
 901         commit_creds(new);
 902         return old_fsgid;
 903 }
 904
 905 void do_sys_times(struct tms *tms)
 906 {
 907         cputime_t tgutime, tgstime, cutime, cstime;
 908
 909         spin_lock_irq(&current->sighand->siglock);
 910         thread_group_times(current, &tgutime, &tgstime);
 911         cutime = current->signal->cutime;
 912         cstime = current->signal->cstime;
 913         spin_unlock_irq(&current->sighand->siglock);
 914         tms->tms_utime = cputime_to_clock_t(tgutime);
 915         tms->tms_stime = cputime_to_clock_t(tgstime);
 916         tms->tms_cutime = cputime_to_clock_t(cutime);
 917         tms->tms_cstime = cputime_to_clock_t(cstime);
 918 }
 919
 920 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 921 {
 922         if (tbuf) {
 923                 struct tms tmp;
 924
 925                 do_sys_times(&tmp);
 926                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 927                         return -EFAULT;
 928         }
 929         force_successful_syscall_return();
 930         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 931 }
 932
 933 /*
 934  * This needs some heavy checking ...
 935  * I just haven't the stomach for it. I also don't fully
 936  * understand sessions/pgrp etc. Let somebody who does explain it.
 937  *
 938  * OK, I think I have the protection semantics right.... this is really
 939  * only important on a multi-user system anyway, to make sure one user
 940  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 941  *
 942  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 943  * LBT 04.03.94
 944  */
 945 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 946 {
 947         struct task_struct *p;
 948         struct task_struct *group_leader = current->group_leader;
 949         struct pid *pgrp;
 950         int err;
 951
 952         if (!pid)
 953                 pid = task_pid_vnr(group_leader);
 954         if (!pgid)
 955                 pgid = pid;
 956         if (pgid < 0)
 957                 return -EINVAL;
 958         rcu_read_lock();
 959
 960         /* From this point forward we keep holding onto the tasklist lock
 961          * so that our parent does not change from under us. -DaveM
 962          */
 963         write_lock_irq(&tasklist_lock);
 964
 965         err = -ESRCH;
 966         p = find_task_by_vpid(pid);
 967         if (!p)
 968                 goto out;
 969
 970         err = -EINVAL;
 971         if (!thread_group_leader(p))
 972                 goto out;
 973
 974         if (same_thread_group(p->real_parent, group_leader)) {
 975                 err = -EPERM;
 976                 if (task_session(p) != task_session(group_leader))
 977                         goto out;
 978                 err = -EACCES;
 979                 if (p->did_exec)
 980                         goto out;
 981         } else {
 982                 err = -ESRCH;
 983                 if (p != group_leader)
 984                         goto out;
 985         }
 986
 987         err = -EPERM;
 988         if (p->signal->leader)
 989                 goto out;
 990
 991         pgrp = task_pid(p);
 992         if (pgid != pid) {
 993                 struct task_struct *g;
 994
 995                 pgrp = find_vpid(pgid);
 996                 g = pid_task(pgrp, PIDTYPE_PGID);
 997                 if (!g || task_session(g) != task_session(group_leader))
 998                         goto out;
 999         }
1000
1001         err = security_task_setpgid(p, pgid);
1002         if (err)
1003                 goto out;
1004
1005         if (task_pgrp(p) != pgrp)
1006                 change_pid(p, PIDTYPE_PGID, pgrp);
1007
1008         err = 0;
1009 out:
1010         /* All paths lead to here, thus we are safe. -DaveM */
1011         write_unlock_irq(&tasklist_lock);
1012         rcu_read_unlock();
1013         return err;
1014 }
1015
1016 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1017 {
1018         struct task_struct *p;
1019         struct pid *grp;
1020         int retval;
1021
1022         rcu_read_lock();
1023         if (!pid)
1024                 grp = task_pgrp(current);
1025         else {
1026                 retval = -ESRCH;
1027                 p = find_task_by_vpid(pid);
1028                 if (!p)
1029                         goto out;
1030                 grp = task_pgrp(p);
1031                 if (!grp)
1032                         goto out;
1033
1034                 retval = security_task_getpgid(p);
1035                 if (retval)
1036                         goto out;
1037         }
1038         retval = pid_vnr(grp);
1039 out:
1040         rcu_read_unlock();
1041         return retval;
1042 }
1043
1044 #ifdef __ARCH_WANT_SYS_GETPGRP
1045
1046 SYSCALL_DEFINE0(getpgrp)
1047 {
1048         return sys_getpgid(0);
1049 }
1050
1051 #endif
1052
1053 SYSCALL_DEFINE1(getsid, pid_t, pid)
1054 {
1055         struct task_struct *p;
1056         struct pid *sid;
1057         int retval;
1058
1059         rcu_read_lock();
1060         if (!pid)
1061                 sid = task_session(current);
1062         else {
1063                 retval = -ESRCH;
1064                 p = find_task_by_vpid(pid);
1065                 if (!p)
1066                         goto out;
1067                 sid = task_session(p);
1068                 if (!sid)
1069                         goto out;
1070
1071                 retval = security_task_getsid(p);
1072                 if (retval)
1073                         goto out;
1074         }
1075         retval = pid_vnr(sid);
1076 out:
1077         rcu_read_unlock();
1078         return retval;
1079 }
1080
1081 SYSCALL_DEFINE0(setsid)
1082 {
1083         struct task_struct *group_leader = current->group_leader;
1084         struct pid *sid = task_pid(group_leader);
1085         pid_t session = pid_vnr(sid);
1086         int err = -EPERM;
1087
1088         write_lock_irq(&tasklist_lock);
1089         /* Fail if I am already a session leader */
1090         if (group_leader->signal->leader)
1091                 goto out;
1092
1093         /* Fail if a process group id already exists that equals the
1094          * proposed session id.
1095          */
1096         if (pid_task(sid, PIDTYPE_PGID))
1097                 goto out;
1098
1099         group_leader->signal->leader = 1;
1100         __set_special_pids(sid);
1101
1102         proc_clear_tty(group_leader);
1103
1104         err = session;
1105 out:
1106         write_unlock_irq(&tasklist_lock);
1107         if (err > 0) {
1108                 proc_sid_connector(group_leader);
1109                 sched_autogroup_create_attach(group_leader);
1110         }
1111         return err;
1112 }
1113
1114 DECLARE_RWSEM(uts_sem);
1115
1116 #ifdef COMPAT_UTS_MACHINE
1117 #define override_architecture(name) \
1118         (personality(current->personality) == PER_LINUX32 && \
1119          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1120                       sizeof(COMPAT_UTS_MACHINE)))
1121 #else
1122 #define override_architecture(name)     0
1123 #endif
1124
1125 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1126 {
1127         int errno = 0;
1128
1129         down_read(&uts_sem);
1130         if (copy_to_user(name, utsname(), sizeof *name))
1131                 errno = -EFAULT;
1132         up_read(&uts_sem);
1133
1134         if (!errno && override_architecture(name))
1135                 errno = -EFAULT;
1136         return errno;
1137 }
1138
1139 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1140 /*
1141  * Old cruft
1142  */
1143 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1144 {
1145         int error = 0;
1146
1147         if (!name)
1148                 return -EFAULT;
1149
1150         down_read(&uts_sem);
1151         if (copy_to_user(name, utsname(), sizeof(*name)))
1152                 error = -EFAULT;
1153         up_read(&uts_sem);
1154
1155         if (!error && override_architecture(name))
1156                 error = -EFAULT;
1157         return error;
1158 }
1159
1160 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1161 {
1162         int error;
1163
1164         if (!name)
1165                 return -EFAULT;
1166         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1167                 return -EFAULT;
1168
1169         down_read(&uts_sem);
1170         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1171                                __OLD_UTS_LEN);
1172         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1173         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1174                                 __OLD_UTS_LEN);
1175         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1176         error |= __copy_to_user(&name->release, &utsname()->release,
1177                                 __OLD_UTS_LEN);
1178         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1179         error |= __copy_to_user(&name->version, &utsname()->version,
1180                                 __OLD_UTS_LEN);
1181         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1182         error |= __copy_to_user(&name->machine, &utsname()->machine,
1183                                 __OLD_UTS_LEN);
1184         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1185         up_read(&uts_sem);
1186
1187         if (!error && override_architecture(name))
1188                 error = -EFAULT;
1189         return error ? -EFAULT : 0;
1190 }
1191 #endif
1192
1193 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1194 {
1195         int errno;
1196         char tmp[__NEW_UTS_LEN];
1197
1198         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1199                 return -EPERM;
1200
1201         if (len < 0 || len > __NEW_UTS_LEN)
1202                 return -EINVAL;
1203         down_write(&uts_sem);
1204         errno = -EFAULT;
1205         if (!copy_from_user(tmp, name, len)) {
1206                 struct new_utsname *u = utsname();
1207
1208                 memcpy(u->nodename, tmp, len);
1209                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1210                 errno = 0;
1211         }
1212         up_write(&uts_sem);
1213         return errno;
1214 }
1215
1216 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1217
1218 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1219 {
1220         int i, errno;
1221         struct new_utsname *u;
1222
1223         if (len < 0)
1224                 return -EINVAL;
1225         down_read(&uts_sem);
1226         u = utsname();
1227         i = 1 + strlen(u->nodename);
1228         if (i > len)
1229                 i = len;
1230         errno = 0;
1231         if (copy_to_user(name, u->nodename, i))
1232                 errno = -EFAULT;
1233         up_read(&uts_sem);
1234         return errno;
1235 }
1236
1237 #endif
1238
1239 /*
1240  * Only setdomainname; getdomainname can be implemented by calling
1241  * uname()
1242  */
1243 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1244 {
1245         int errno;
1246         char tmp[__NEW_UTS_LEN];
1247
1248         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1249                 return -EPERM;
1250         if (len < 0 || len > __NEW_UTS_LEN)
1251                 return -EINVAL;
1252
1253         down_write(&uts_sem);
1254         errno = -EFAULT;
1255         if (!copy_from_user(tmp, name, len)) {
1256                 struct new_utsname *u = utsname();
1257
1258                 memcpy(u->domainname, tmp, len);
1259                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1260                 errno = 0;
1261         }
1262         up_write(&uts_sem);
1263         return errno;
1264 }
1265
1266 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1267 {
1268         struct rlimit value;
1269         int ret;
1270
1271         ret = do_prlimit(current, resource, NULL, &value);
1272         if (!ret)
1273                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1274
1275         return ret;
1276 }
1277
1278 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1279
1280 /*
1281  *      Back compatibility for getrlimit. Needed for some apps.
1282  */
1283
1284 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1285                 struct rlimit __user *, rlim)
1286 {
1287         struct rlimit x;
1288         if (resource >= RLIM_NLIMITS)
1289                 return -EINVAL;
1290
1291         task_lock(current->group_leader);
1292         x = current->signal->rlim[resource];
1293         task_unlock(current->group_leader);
1294         if (x.rlim_cur > 0x7FFFFFFF)
1295                 x.rlim_cur = 0x7FFFFFFF;
1296         if (x.rlim_max > 0x7FFFFFFF)
1297                 x.rlim_max = 0x7FFFFFFF;
1298         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1299 }
1300
1301 #endif
1302
1303 static inline bool rlim64_is_infinity(__u64 rlim64)
1304 {
1305 #if BITS_PER_LONG < 64
1306         return rlim64 >= ULONG_MAX;
1307 #else
1308         return rlim64 == RLIM64_INFINITY;
1309 #endif
1310 }
1311
1312 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1313 {
1314         if (rlim->rlim_cur == RLIM_INFINITY)
1315                 rlim64->rlim_cur = RLIM64_INFINITY;
1316         else
1317                 rlim64->rlim_cur = rlim->rlim_cur;
1318         if (rlim->rlim_max == RLIM_INFINITY)
1319                 rlim64->rlim_max = RLIM64_INFINITY;
1320         else
1321                 rlim64->rlim_max = rlim->rlim_max;
1322 }
1323
1324 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1325 {
1326         if (rlim64_is_infinity(rlim64->rlim_cur))
1327                 rlim->rlim_cur = RLIM_INFINITY;
1328         else
1329                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1330         if (rlim64_is_infinity(rlim64->rlim_max))
1331                 rlim->rlim_max = RLIM_INFINITY;
1332         else
1333                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1334 }
1335
1336 /* make sure you are allowed to change @tsk limits before calling this */
1337 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1338                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1339 {
1340         struct rlimit *rlim;
1341         int retval = 0;
1342
1343         if (resource >= RLIM_NLIMITS)
1344                 return -EINVAL;
1345         if (new_rlim) {
1346                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1347                         return -EINVAL;
1348                 if (resource == RLIMIT_NOFILE &&
1349                                 new_rlim->rlim_max > sysctl_nr_open)
1350                         return -EPERM;
1351         }
1352
1353         /* protect tsk->signal and tsk->sighand from disappearing */
1354         read_lock(&tasklist_lock);
1355         if (!tsk->sighand) {
1356                 retval = -ESRCH;
1357                 goto out;
1358         }
1359
1360         rlim = tsk->signal->rlim + resource;
1361         task_lock(tsk->group_leader);
1362         if (new_rlim) {
1363                 /* Keep the capable check against init_user_ns until
1364                    cgroups can contain all limits */
1365                 if (new_rlim->rlim_max > rlim->rlim_max &&
1366                                 !capable(CAP_SYS_RESOURCE))
1367                         retval = -EPERM;
1368                 if (!retval)
1369                         retval = security_task_setrlimit(tsk->group_leader,
1370                                         resource, new_rlim);
1371                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1372                         /*
1373                          * The caller is asking for an immediate RLIMIT_CPU
1374                          * expiry.  But we use the zero value to mean "it was
1375                          * never set".  So let's cheat and make it one second
1376                          * instead
1377                          */
1378                         new_rlim->rlim_cur = 1;
1379                 }
1380         }
1381         if (!retval) {
1382                 if (old_rlim)
1383                         *old_rlim = *rlim;
1384                 if (new_rlim)
1385                         *rlim = *new_rlim;
1386         }
1387         task_unlock(tsk->group_leader);
1388
1389         /*
1390          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1391          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1392          * very long-standing error, and fixing it now risks breakage of
1393          * applications, so we live with it
1394          */
1395          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1396                          new_rlim->rlim_cur != RLIM_INFINITY)
1397                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1398 out:
1399         read_unlock(&tasklist_lock);
1400         return retval;
1401 }
1402
1403 /* rcu lock must be held */
1404 static int check_prlimit_permission(struct task_struct *task)
1405 {
1406         const struct cred *cred = current_cred(), *tcred;
1407
1408         if (current == task)
1409                 return 0;
1410
1411         tcred = __task_cred(task);
1412         if (cred->user->user_ns == tcred->user->user_ns &&
1413             (cred->uid == tcred->euid &&
1414              cred->uid == tcred->suid &&
1415              cred->uid == tcred->uid  &&
1416              cred->gid == tcred->egid &&
1417              cred->gid == tcred->sgid &&
1418              cred->gid == tcred->gid))
1419                 return 0;
1420         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1421                 return 0;
1422
1423         return -EPERM;
1424 }
1425
1426 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1427                 const struct rlimit64 __user *, new_rlim,
1428                 struct rlimit64 __user *, old_rlim)
1429 {
1430         struct rlimit64 old64, new64;
1431         struct rlimit old, new;
1432         struct task_struct *tsk;
1433         int ret;
1434
1435         if (new_rlim) {
1436                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1437                         return -EFAULT;
1438                 rlim64_to_rlim(&new64, &new);
1439         }
1440
1441         rcu_read_lock();
1442         tsk = pid ? find_task_by_vpid(pid) : current;
1443         if (!tsk) {
1444                 rcu_read_unlock();
1445                 return -ESRCH;
1446         }
1447         ret = check_prlimit_permission(tsk);
1448         if (ret) {
1449                 rcu_read_unlock();
1450                 return ret;
1451         }
1452         get_task_struct(tsk);
1453         rcu_read_unlock();
1454
1455         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1456                         old_rlim ? &old : NULL);
1457
1458         if (!ret && old_rlim) {
1459                 rlim_to_rlim64(&old, &old64);
1460                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1461                         ret = -EFAULT;
1462         }
1463
1464         put_task_struct(tsk);
1465         return ret;
1466 }
1467
1468 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1469 {
1470         struct rlimit new_rlim;
1471
1472         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1473                 return -EFAULT;
1474         return do_prlimit(current, resource, &new_rlim, NULL);
1475 }
1476
1477 /*
1478  * It would make sense to put struct rusage in the task_struct,
1479  * except that would make the task_struct be *really big*.  After
1480  * task_struct gets moved into malloc'ed memory, it would
1481  * make sense to do this.  It will make moving the rest of the information
1482  * a lot simpler!  (Which we're not doing right now because we're not
1483  * measuring them yet).
1484  *
1485  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1486  * races with threads incrementing their own counters.  But since word
1487  * reads are atomic, we either get new values or old values and we don't
1488  * care which for the sums.  We always take the siglock to protect reading
1489  * the c* fields from p->signal from races with exit.c updating those
1490  * fields when reaping, so a sample either gets all the additions of a
1491  * given child after it's reaped, or none so this sample is before reaping.
1492  *
1493  * Locking:
1494  * We need to take the siglock for CHILDEREN, SELF and BOTH
1495  * for  the cases current multithreaded, non-current single threaded
1496  * non-current multithreaded.  Thread traversal is now safe with
1497  * the siglock held.
1498  * Strictly speaking, we donot need to take the siglock if we are current and
1499  * single threaded,  as no one else can take our signal_struct away, no one
1500  * else can  reap the  children to update signal->c* counters, and no one else
1501  * can race with the signal-> fields. If we do not take any lock, the
1502  * signal-> fields could be read out of order while another thread was just
1503  * exiting. So we should  place a read memory barrier when we avoid the lock.
1504  * On the writer side,  write memory barrier is implied in  __exit_signal
1505  * as __exit_signal releases  the siglock spinlock after updating the signal->
1506  * fields. But we don't do this yet to keep things simple.
1507  *
1508  */
1509
1510 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1511 {
1512         r->ru_nvcsw += t->nvcsw;
1513         r->ru_nivcsw += t->nivcsw;
1514         r->ru_minflt += t->min_flt;
1515         r->ru_majflt += t->maj_flt;
1516         r->ru_inblock += task_io_get_inblock(t);
1517         r->ru_oublock += task_io_get_oublock(t);
1518 }
1519
1520 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1521 {
1522         struct task_struct *t;
1523         unsigned long flags;
1524         cputime_t tgutime, tgstime, utime, stime;
1525         unsigned long maxrss = 0;
1526
1527         memset((char *) r, 0, sizeof *r);
1528         utime = stime = cputime_zero;
1529
1530         if (who == RUSAGE_THREAD) {
1531                 task_times(current, &utime, &stime);
1532                 accumulate_thread_rusage(p, r);
1533                 maxrss = p->signal->maxrss;
1534                 goto out;
1535         }
1536
1537         if (!lock_task_sighand(p, &flags))
1538                 return;
1539
1540         switch (who) {
1541                 case RUSAGE_BOTH:
1542                 case RUSAGE_CHILDREN:
1543                         utime = p->signal->cutime;
1544                         stime = p->signal->cstime;
1545                         r->ru_nvcsw = p->signal->cnvcsw;
1546                         r->ru_nivcsw = p->signal->cnivcsw;
1547                         r->ru_minflt = p->signal->cmin_flt;
1548                         r->ru_majflt = p->signal->cmaj_flt;
1549                         r->ru_inblock = p->signal->cinblock;
1550                         r->ru_oublock = p->signal->coublock;
1551                         maxrss = p->signal->cmaxrss;
1552
1553                         if (who == RUSAGE_CHILDREN)
1554                                 break;
1555
1556                 case RUSAGE_SELF:
1557                         thread_group_times(p, &tgutime, &tgstime);
1558                         utime = cputime_add(utime, tgutime);
1559                         stime = cputime_add(stime, tgstime);
1560                         r->ru_nvcsw += p->signal->nvcsw;
1561                         r->ru_nivcsw += p->signal->nivcsw;
1562                         r->ru_minflt += p->signal->min_flt;
1563                         r->ru_majflt += p->signal->maj_flt;
1564                         r->ru_inblock += p->signal->inblock;
1565                         r->ru_oublock += p->signal->oublock;
1566                         if (maxrss < p->signal->maxrss)
1567                                 maxrss = p->signal->maxrss;
1568                         t = p;
1569                         do {
1570                                 accumulate_thread_rusage(t, r);
1571                                 t = next_thread(t);
1572                         } while (t != p);
1573                         break;
1574
1575                 default:
1576                         BUG();
1577         }
1578         unlock_task_sighand(p, &flags);
1579
1580 out:
1581         cputime_to_timeval(utime, &r->ru_utime);
1582         cputime_to_timeval(stime, &r->ru_stime);
1583
1584         if (who != RUSAGE_CHILDREN) {
1585                 struct mm_struct *mm = get_task_mm(p);
1586                 if (mm) {
1587                         setmax_mm_hiwater_rss(&maxrss, mm);
1588                         mmput(mm);
1589                 }
1590         }
1591         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1592 }
1593
1594 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1595 {
1596         struct rusage r;
1597         k_getrusage(p, who, &r);
1598         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1599 }
1600
1601 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1602 {
1603         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1604             who != RUSAGE_THREAD)
1605                 return -EINVAL;
1606         return getrusage(current, who, ru);
1607 }
1608
1609 SYSCALL_DEFINE1(umask, int, mask)
1610 {
1611         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1612         return mask;
1613 }
1614
1615 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1616                 unsigned long, arg4, unsigned long, arg5)
1617 {
1618         struct task_struct *me = current;
1619         unsigned char comm[sizeof(me->comm)];
1620         long error;
1621
1622         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1623         if (error != -ENOSYS)
1624                 return error;
1625
1626         error = 0;
1627         switch (option) {
1628                 case PR_SET_PDEATHSIG:
1629                         if (!valid_signal(arg2)) {
1630                                 error = -EINVAL;
1631                                 break;
1632                         }
1633                         me->pdeath_signal = arg2;
1634                         error = 0;
1635                         break;
1636                 case PR_GET_PDEATHSIG:
1637                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1638                         break;
1639                 case PR_GET_DUMPABLE:
1640                         error = get_dumpable(me->mm);
1641                         break;
1642                 case PR_SET_DUMPABLE:
1643                         if (arg2 < 0 || arg2 > 1) {
1644                                 error = -EINVAL;
1645                                 break;
1646                         }
1647                         set_dumpable(me->mm, arg2);
1648                         error = 0;
1649                         break;
1650
1651                 case PR_SET_UNALIGN:
1652                         error = SET_UNALIGN_CTL(me, arg2);
1653                         break;
1654                 case PR_GET_UNALIGN:
1655                         error = GET_UNALIGN_CTL(me, arg2);
1656                         break;
1657                 case PR_SET_FPEMU:
1658                         error = SET_FPEMU_CTL(me, arg2);
1659                         break;
1660                 case PR_GET_FPEMU:
1661                         error = GET_FPEMU_CTL(me, arg2);
1662                         break;
1663                 case PR_SET_FPEXC:
1664                         error = SET_FPEXC_CTL(me, arg2);
1665                         break;
1666                 case PR_GET_FPEXC:
1667                         error = GET_FPEXC_CTL(me, arg2);
1668                         break;
1669                 case PR_GET_TIMING:
1670                         error = PR_TIMING_STATISTICAL;
1671                         break;
1672                 case PR_SET_TIMING:
1673                         if (arg2 != PR_TIMING_STATISTICAL)
1674                                 error = -EINVAL;
1675                         else
1676                                 error = 0;
1677                         break;
1678
1679                 case PR_SET_NAME:
1680                         comm[sizeof(me->comm)-1] = 0;
1681                         if (strncpy_from_user(comm, (char __user *)arg2,
1682                                               sizeof(me->comm) - 1) < 0)
1683                                 return -EFAULT;
1684                         set_task_comm(me, comm);
1685                         return 0;
1686                 case PR_GET_NAME:
1687                         get_task_comm(comm, me);
1688                         if (copy_to_user((char __user *)arg2, comm,
1689                                          sizeof(comm)))
1690                                 return -EFAULT;
1691                         return 0;
1692                 case PR_GET_ENDIAN:
1693                         error = GET_ENDIAN(me, arg2);
1694                         break;
1695                 case PR_SET_ENDIAN:
1696                         error = SET_ENDIAN(me, arg2);
1697                         break;
1698
1699                 case PR_GET_SECCOMP:
1700                         error = prctl_get_seccomp();
1701                         break;
1702                 case PR_SET_SECCOMP:
1703                         error = prctl_set_seccomp(arg2);
1704                         break;
1705                 case PR_GET_TSC:
1706                         error = GET_TSC_CTL(arg2);
1707                         break;
1708                 case PR_SET_TSC:
1709                         error = SET_TSC_CTL(arg2);
1710                         break;
1711                 case PR_TASK_PERF_EVENTS_DISABLE:
1712                         error = perf_event_task_disable();
1713                         break;
1714                 case PR_TASK_PERF_EVENTS_ENABLE:
1715                         error = perf_event_task_enable();
1716                         break;
1717                 case PR_GET_TIMERSLACK:
1718                         error = current->timer_slack_ns;
1719                         break;
1720                 case PR_SET_TIMERSLACK:
1721                         if (arg2 <= 0)
1722                                 current->timer_slack_ns =
1723                                         current->default_timer_slack_ns;
1724                         else
1725                                 current->timer_slack_ns = arg2;
1726                         error = 0;
1727                         break;
1728                 case PR_MCE_KILL:
1729                         if (arg4 | arg5)
1730                                 return -EINVAL;
1731                         switch (arg2) {
1732                         case PR_MCE_KILL_CLEAR:
1733                                 if (arg3 != 0)
1734                                         return -EINVAL;
1735                                 current->flags &= ~PF_MCE_PROCESS;
1736                                 break;
1737                         case PR_MCE_KILL_SET:
1738                                 current->flags |= PF_MCE_PROCESS;
1739                                 if (arg3 == PR_MCE_KILL_EARLY)
1740                                         current->flags |= PF_MCE_EARLY;
1741                                 else if (arg3 == PR_MCE_KILL_LATE)
1742                                         current->flags &= ~PF_MCE_EARLY;
1743                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1744                                         current->flags &=
1745                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1746                                 else
1747                                         return -EINVAL;
1748                                 break;
1749                         default:
1750                                 return -EINVAL;
1751                         }
1752                         error = 0;
1753                         break;
1754                 case PR_MCE_KILL_GET:
1755                         if (arg2 | arg3 | arg4 | arg5)
1756                                 return -EINVAL;
1757                         if (current->flags & PF_MCE_PROCESS)
1758                                 error = (current->flags & PF_MCE_EARLY) ?
1759                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1760                         else
1761                                 error = PR_MCE_KILL_DEFAULT;
1762                         break;
1763                 default:
1764                         error = -EINVAL;
1765                         break;
1766         }
1767         return error;
1768 }
1769
1770 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1771                 struct getcpu_cache __user *, unused)
1772 {
1773         int err = 0;
1774         int cpu = raw_smp_processor_id();
1775         if (cpup)
1776                 err |= put_user(cpu, cpup);
1777         if (nodep)
1778                 err |= put_user(cpu_to_node(cpu), nodep);
1779         return err ? -EFAULT : 0;
1780 }
1781
1782 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1783
1784 static void argv_cleanup(struct subprocess_info *info)
1785 {
1786         argv_free(info->argv);
1787 }
1788
1789 /**
1790  * orderly_poweroff - Trigger an orderly system poweroff
1791  * @force: force poweroff if command execution fails
1792  *
1793  * This may be called from any context to trigger a system shutdown.
1794  * If the orderly shutdown fails, it will force an immediate shutdown.
1795  */
1796 int orderly_poweroff(bool force)
1797 {
1798         int argc;
1799         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1800         static char *envp[] = {
1801                 "HOME=/",
1802                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1803                 NULL
1804         };
1805         int ret = -ENOMEM;
1806         struct subprocess_info *info;
1807
1808         if (argv == NULL) {
1809                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1810                        __func__, poweroff_cmd);
1811                 goto out;
1812         }
1813
1814         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1815         if (info == NULL) {
1816                 argv_free(argv);
1817                 goto out;
1818         }
1819
1820         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1821
1822         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1823
1824   out:
1825         if (ret && force) {
1826                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1827                        "forcing the issue\n");
1828
1829                 /* I guess this should try to kick off some daemon to
1830                    sync and poweroff asap.  Or not even bother syncing
1831                    if we're doing an emergency shutdown? */
1832                 emergency_sync();
1833                 kernel_power_off();
1834         }
1835
1836         return ret;
1837 }
1838 EXPORT_SYMBOL_GPL(orderly_poweroff);