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