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 #include <linux/version.h>
  42 #include <linux/ctype.h>
  43
  44 #include <linux/compat.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/kprobes.h>
  47 #include <linux/user_namespace.h>
  48
  49 #include <linux/kmsg_dump.h>
  50 /* Move somewhere else to avoid recompiling? */
  51 #include <generated/utsrelease.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/io.h>
  55 #include <asm/unistd.h>
  56 /***************
  57 *        DEBUG
  58 ****************/
  59 #define RESTART_DEBUG
  60 #ifdef RESTART_DEBUG
  61 #define restart_dbg(format, arg...) \
  62         printk("RESTART_DEBUG : " format "\n" , ## arg)
  63 #else
  64 #define restart_dbg(format, arg...) do {} while (0)
  65 #endif
  66
  67
  68
  69 #ifndef SET_UNALIGN_CTL
  70 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  71 #endif
  72 #ifndef GET_UNALIGN_CTL
  73 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  74 #endif
  75 #ifndef SET_FPEMU_CTL
  76 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  77 #endif
  78 #ifndef GET_FPEMU_CTL
  79 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  80 #endif
  81 #ifndef SET_FPEXC_CTL
  82 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  83 #endif
  84 #ifndef GET_FPEXC_CTL
  85 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  86 #endif
  87 #ifndef GET_ENDIAN
  88 # define GET_ENDIAN(a,b)        (-EINVAL)
  89 #endif
  90 #ifndef SET_ENDIAN
  91 # define SET_ENDIAN(a,b)        (-EINVAL)
  92 #endif
  93 #ifndef GET_TSC_CTL
  94 # define GET_TSC_CTL(a)         (-EINVAL)
  95 #endif
  96 #ifndef SET_TSC_CTL
  97 # define SET_TSC_CTL(a)         (-EINVAL)
  98 #endif
  99
 100 /*
 101  * this is where the system-wide overflow UID and GID are defined, for
 102  * architectures that now have 32-bit UID/GID but didn't in the past
 103  */
 104
 105 int overflowuid = DEFAULT_OVERFLOWUID;
 106 int overflowgid = DEFAULT_OVERFLOWGID;
 107
 108 #ifdef CONFIG_UID16
 109 EXPORT_SYMBOL(overflowuid);
 110 EXPORT_SYMBOL(overflowgid);
 111 #endif
 112
 113 /*
 114  * the same as above, but for filesystems which can only store a 16-bit
 115  * UID and GID. as such, this is needed on all architectures
 116  */
 117
 118 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 119 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 120
 121 EXPORT_SYMBOL(fs_overflowuid);
 122 EXPORT_SYMBOL(fs_overflowgid);
 123
 124 /*
 125  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 126  */
 127
 128 int C_A_D = 1;
 129 struct pid *cad_pid;
 130 EXPORT_SYMBOL(cad_pid);
 131
 132 /*
 133  * If set, this is used for preparing the system to power off.
 134  */
 135
 136 void (*pm_power_off_prepare)(void);
 137
 138 /*
 139  * Returns true if current's euid is same as p's uid or euid,
 140  * or has CAP_SYS_NICE to p's user_ns.
 141  *
 142  * Called with rcu_read_lock, creds are safe
 143  */
 144 static bool set_one_prio_perm(struct task_struct *p)
 145 {
 146         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 147
 148         if (pcred->user->user_ns == cred->user->user_ns &&
 149             (pcred->uid  == cred->euid ||
 150              pcred->euid == cred->euid))
 151                 return true;
 152         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 153                 return true;
 154         return false;
 155 }
 156
 157 /*
 158  * set the priority of a task
 159  * - the caller must hold the RCU read lock
 160  */
 161 static int set_one_prio(struct task_struct *p, int niceval, int error)
 162 {
 163         int no_nice;
 164
 165         if (!set_one_prio_perm(p)) {
 166                 error = -EPERM;
 167                 goto out;
 168         }
 169         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 170                 error = -EACCES;
 171                 goto out;
 172         }
 173         no_nice = security_task_setnice(p, niceval);
 174         if (no_nice) {
 175                 error = no_nice;
 176                 goto out;
 177         }
 178         if (error == -ESRCH)
 179                 error = 0;
 180         set_user_nice(p, niceval);
 181 out:
 182         return error;
 183 }
 184
 185 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 186 {
 187         struct task_struct *g, *p;
 188         struct user_struct *user;
 189         const struct cred *cred = current_cred();
 190         int error = -EINVAL;
 191         struct pid *pgrp;
 192
 193         if (which > PRIO_USER || which < PRIO_PROCESS)
 194                 goto out;
 195
 196         /* normalize: avoid signed division (rounding problems) */
 197         error = -ESRCH;
 198         if (niceval < -20)
 199                 niceval = -20;
 200         if (niceval > 19)
 201                 niceval = 19;
 202
 203         rcu_read_lock();
 204         read_lock(&tasklist_lock);
 205         switch (which) {
 206                 case PRIO_PROCESS:
 207                         if (who)
 208                                 p = find_task_by_vpid(who);
 209                         else
 210                                 p = current;
 211                         if (p)
 212                                 error = set_one_prio(p, niceval, error);
 213                         break;
 214                 case PRIO_PGRP:
 215                         if (who)
 216                                 pgrp = find_vpid(who);
 217                         else
 218                                 pgrp = task_pgrp(current);
 219                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 220                                 error = set_one_prio(p, niceval, error);
 221                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 222                         break;
 223                 case PRIO_USER:
 224                         user = (struct user_struct *) cred->user;
 225                         if (!who)
 226                                 who = cred->uid;
 227                         else if ((who != cred->uid) &&
 228                                  !(user = find_user(who)))
 229                                 goto out_unlock;        /* No processes for this user */
 230
 231                         do_each_thread(g, p) {
 232                                 if (__task_cred(p)->uid == who)
 233                                         error = set_one_prio(p, niceval, error);
 234                         } while_each_thread(g, p);
 235                         if (who != cred->uid)
 236                                 free_uid(user);         /* For find_user() */
 237                         break;
 238         }
 239 out_unlock:
 240         read_unlock(&tasklist_lock);
 241         rcu_read_unlock();
 242 out:
 243         return error;
 244 }
 245
 246 /*
 247  * Ugh. To avoid negative return values, "getpriority()" will
 248  * not return the normal nice-value, but a negated value that
 249  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 250  * to stay compatible.
 251  */
 252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 253 {
 254         struct task_struct *g, *p;
 255         struct user_struct *user;
 256         const struct cred *cred = current_cred();
 257         long niceval, retval = -ESRCH;
 258         struct pid *pgrp;
 259
 260         if (which > PRIO_USER || which < PRIO_PROCESS)
 261                 return -EINVAL;
 262
 263         rcu_read_lock();
 264         read_lock(&tasklist_lock);
 265         switch (which) {
 266                 case PRIO_PROCESS:
 267                         if (who)
 268                                 p = find_task_by_vpid(who);
 269                         else
 270                                 p = current;
 271                         if (p) {
 272                                 niceval = 20 - task_nice(p);
 273                                 if (niceval > retval)
 274                                         retval = niceval;
 275                         }
 276                         break;
 277                 case PRIO_PGRP:
 278                         if (who)
 279                                 pgrp = find_vpid(who);
 280                         else
 281                                 pgrp = task_pgrp(current);
 282                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 283                                 niceval = 20 - task_nice(p);
 284                                 if (niceval > retval)
 285                                         retval = niceval;
 286                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 287                         break;
 288                 case PRIO_USER:
 289                         user = (struct user_struct *) cred->user;
 290                         if (!who)
 291                                 who = cred->uid;
 292                         else if ((who != cred->uid) &&
 293                                  !(user = find_user(who)))
 294                                 goto out_unlock;        /* No processes for this user */
 295
 296                         do_each_thread(g, p) {
 297                                 if (__task_cred(p)->uid == who) {
 298                                         niceval = 20 - task_nice(p);
 299                                         if (niceval > retval)
 300                                                 retval = niceval;
 301                                 }
 302                         } while_each_thread(g, p);
 303                         if (who != cred->uid)
 304                                 free_uid(user);         /* for find_user() */
 305                         break;
 306         }
 307 out_unlock:
 308         read_unlock(&tasklist_lock);
 309         rcu_read_unlock();
 310
 311         return retval;
 312 }
 313
 314 /**
 315  *      emergency_restart - reboot the system
 316  *
 317  *      Without shutting down any hardware or taking any locks
 318  *      reboot the system.  This is called when we know we are in
 319  *      trouble so this is our best effort to reboot.  This is
 320  *      safe to call in interrupt context.
 321  */
 322 void emergency_restart(void)
 323 {
 324         kmsg_dump(KMSG_DUMP_EMERG);
 325         machine_emergency_restart();
 326 }
 327 EXPORT_SYMBOL_GPL(emergency_restart);
 328
 329 void kernel_restart_prepare(char *cmd)
 330 {
 331         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 332         system_state = SYSTEM_RESTART;
 333         usermodehelper_disable();
 334         device_shutdown();
 335         syscore_shutdown();
 336 }
 337
 338 /**
 339  *      kernel_restart - reboot the system
 340  *      @cmd: pointer to buffer containing command to execute for restart
 341  *              or %NULL
 342  *
 343  *      Shutdown everything and perform a clean reboot.
 344  *      This is not safe to call in interrupt context.
 345  */
 346 void kernel_restart(char *cmd)
 347 {
 348         /*
 349         *  debug trace
 350         */
 351         restart_dbg("%s->%d->cmd=%s",__FUNCTION__,__LINE__,cmd);
 352
 353         kernel_restart_prepare(cmd);
 354         disable_nonboot_cpus();
 355         if (!cmd)
 356                 printk(KERN_EMERG "Restarting system.\n");
 357         else
 358                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 359         kmsg_dump(KMSG_DUMP_RESTART);
 360         machine_restart(cmd);
 361 }
 362 EXPORT_SYMBOL_GPL(kernel_restart);
 363
 364 static void kernel_shutdown_prepare(enum system_states state)
 365 {
 366         blocking_notifier_call_chain(&reboot_notifier_list,
 367                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 368         system_state = state;
 369         usermodehelper_disable();
 370         device_shutdown();
 371 }
 372 /**
 373  *      kernel_halt - halt the system
 374  *
 375  *      Shutdown everything and perform a clean system halt.
 376  */
 377 void kernel_halt(void)
 378 {
 379         kernel_shutdown_prepare(SYSTEM_HALT);
 380         syscore_shutdown();
 381         printk(KERN_EMERG "System halted.\n");
 382         kmsg_dump(KMSG_DUMP_HALT);
 383         machine_halt();
 384 }
 385
 386 EXPORT_SYMBOL_GPL(kernel_halt);
 387
 388 /**
 389  *      kernel_power_off - power_off the system
 390  *
 391  *      Shutdown everything and perform a clean system power_off.
 392  */
 393 void kernel_power_off(void)
 394 {
 395         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 396         if (pm_power_off_prepare)
 397                 pm_power_off_prepare();
 398         disable_nonboot_cpus();
 399         syscore_shutdown();
 400         printk(KERN_EMERG "Power down.\n");
 401         kmsg_dump(KMSG_DUMP_POWEROFF);
 402         machine_power_off();
 403 }
 404 EXPORT_SYMBOL_GPL(kernel_power_off);
 405
 406 static DEFINE_MUTEX(reboot_mutex);
 407
 408 /*
 409  * Reboot system call: for obvious reasons only root may call it,
 410  * and even root needs to set up some magic numbers in the registers
 411  * so that some mistake won't make this reboot the whole machine.
 412  * You can also set the meaning of the ctrl-alt-del-key here.
 413  *
 414  * reboot doesn't sync: do that yourself before calling this.
 415  */
 416 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 417                 void __user *, arg)
 418 {
 419         char buffer[256];
 420         int ret = 0;
 421
 422         /* We only trust the superuser with rebooting the system. */
 423         if (!capable(CAP_SYS_BOOT))
 424                 return -EPERM;
 425
 426         /* For safety, we require "magic" arguments. */
 427         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 428             (magic2 != LINUX_REBOOT_MAGIC2 &&
 429                         magic2 != LINUX_REBOOT_MAGIC2A &&
 430                         magic2 != LINUX_REBOOT_MAGIC2B &&
 431                         magic2 != LINUX_REBOOT_MAGIC2C))
 432                 return -EINVAL;
 433
 434         /* Instead of trying to make the power_off code look like
 435          * halt when pm_power_off is not set do it the easy way.
 436          */
 437         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 438                 cmd = LINUX_REBOOT_CMD_HALT;
 439
 440         mutex_lock(&reboot_mutex);
 441         switch (cmd) {
 442         case LINUX_REBOOT_CMD_RESTART:
 443                 /*
 444                 *  debug trace
 445                 */
 446                 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
 447
 448                 kernel_restart(NULL);
 449                 break;
 450
 451         case LINUX_REBOOT_CMD_CAD_ON:
 452                 C_A_D = 1;
 453                 break;
 454
 455         case LINUX_REBOOT_CMD_CAD_OFF:
 456                 C_A_D = 0;
 457                 break;
 458
 459         case LINUX_REBOOT_CMD_HALT:
 460                 kernel_halt();
 461                 do_exit(0);
 462                 panic("cannot halt");
 463
 464         case LINUX_REBOOT_CMD_POWER_OFF:
 465                 /*
 466                 *  debug trace
 467                 */
 468                 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
 469
 470                 kernel_power_off();
 471                 do_exit(0);
 472                 break;
 473
 474         case LINUX_REBOOT_CMD_RESTART2:
 475                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 476                         ret = -EFAULT;
 477                         break;
 478                 }
 479                 buffer[sizeof(buffer) - 1] = '\0';
 480                 /*
 481                 *  debug trace
 482                 */
 483                 restart_dbg("%s->%d->cmd=%x args=%s",__FUNCTION__,__LINE__,cmd,buffer);
 484
 485                 kernel_restart(buffer);
 486                 break;
 487
 488 #ifdef CONFIG_KEXEC
 489         case LINUX_REBOOT_CMD_KEXEC:
 490                 ret = kernel_kexec();
 491                 break;
 492 #endif
 493
 494 #ifdef CONFIG_HIBERNATION
 495         case LINUX_REBOOT_CMD_SW_SUSPEND:
 496                 ret = hibernate();
 497                 break;
 498 #endif
 499
 500         default:
 501                 ret = -EINVAL;
 502                 break;
 503         }
 504         mutex_unlock(&reboot_mutex);
 505         return ret;
 506 }
 507
 508 static void deferred_cad(struct work_struct *dummy)
 509 {
 510         kernel_restart(NULL);
 511 }
 512
 513 /*
 514  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 515  * As it's called within an interrupt, it may NOT sync: the only choice
 516  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 517  */
 518 void ctrl_alt_del(void)
 519 {
 520         static DECLARE_WORK(cad_work, deferred_cad);
 521
 522         if (C_A_D)
 523                 schedule_work(&cad_work);
 524         else
 525                 kill_cad_pid(SIGINT, 1);
 526 }
 527
 528 /*
 529  * Unprivileged users may change the real gid to the effective gid
 530  * or vice versa.  (BSD-style)
 531  *
 532  * If you set the real gid at all, or set the effective gid to a value not
 533  * equal to the real gid, then the saved gid is set to the new effective gid.
 534  *
 535  * This makes it possible for a setgid program to completely drop its
 536  * privileges, which is often a useful assertion to make when you are doing
 537  * a security audit over a program.
 538  *
 539  * The general idea is that a program which uses just setregid() will be
 540  * 100% compatible with BSD.  A program which uses just setgid() will be
 541  * 100% compatible with POSIX with saved IDs.
 542  *
 543  * SMP: There are not races, the GIDs are checked only by filesystem
 544  *      operations (as far as semantic preservation is concerned).
 545  */
 546 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 547 {
 548         const struct cred *old;
 549         struct cred *new;
 550         int retval;
 551
 552         new = prepare_creds();
 553         if (!new)
 554                 return -ENOMEM;
 555         old = current_cred();
 556
 557         retval = -EPERM;
 558         if (rgid != (gid_t) -1) {
 559                 if (old->gid == rgid ||
 560                     old->egid == rgid ||
 561                     nsown_capable(CAP_SETGID))
 562                         new->gid = rgid;
 563                 else
 564                         goto error;
 565         }
 566         if (egid != (gid_t) -1) {
 567                 if (old->gid == egid ||
 568                     old->egid == egid ||
 569                     old->sgid == egid ||
 570                     nsown_capable(CAP_SETGID))
 571                         new->egid = egid;
 572                 else
 573                         goto error;
 574         }
 575
 576         if (rgid != (gid_t) -1 ||
 577             (egid != (gid_t) -1 && egid != old->gid))
 578                 new->sgid = new->egid;
 579         new->fsgid = new->egid;
 580
 581         return commit_creds(new);
 582
 583 error:
 584         abort_creds(new);
 585         return retval;
 586 }
 587
 588 /*
 589  * setgid() is implemented like SysV w/ SAVED_IDS
 590  *
 591  * SMP: Same implicit races as above.
 592  */
 593 SYSCALL_DEFINE1(setgid, gid_t, gid)
 594 {
 595         const struct cred *old;
 596         struct cred *new;
 597         int retval;
 598
 599         new = prepare_creds();
 600         if (!new)
 601                 return -ENOMEM;
 602         old = current_cred();
 603
 604         retval = -EPERM;
 605         if (nsown_capable(CAP_SETGID))
 606                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 607         else if (gid == old->gid || gid == old->sgid)
 608                 new->egid = new->fsgid = gid;
 609         else
 610                 goto error;
 611
 612         return commit_creds(new);
 613
 614 error:
 615         abort_creds(new);
 616         return retval;
 617 }
 618
 619 /*
 620  * change the user struct in a credentials set to match the new UID
 621  */
 622 static int set_user(struct cred *new)
 623 {
 624         struct user_struct *new_user;
 625
 626         new_user = alloc_uid(current_user_ns(), new->uid);
 627         if (!new_user)
 628                 return -EAGAIN;
 629
 630         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 631                         new_user != INIT_USER) {
 632                 free_uid(new_user);
 633                 return -EAGAIN;
 634         }
 635
 636         free_uid(new->user);
 637         new->user = new_user;
 638         return 0;
 639 }
 640
 641 /*
 642  * Unprivileged users may change the real uid to the effective uid
 643  * or vice versa.  (BSD-style)
 644  *
 645  * If you set the real uid at all, or set the effective uid to a value not
 646  * equal to the real uid, then the saved uid is set to the new effective uid.
 647  *
 648  * This makes it possible for a setuid program to completely drop its
 649  * privileges, which is often a useful assertion to make when you are doing
 650  * a security audit over a program.
 651  *
 652  * The general idea is that a program which uses just setreuid() will be
 653  * 100% compatible with BSD.  A program which uses just setuid() will be
 654  * 100% compatible with POSIX with saved IDs.
 655  */
 656 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 657 {
 658         const struct cred *old;
 659         struct cred *new;
 660         int retval;
 661
 662         new = prepare_creds();
 663         if (!new)
 664                 return -ENOMEM;
 665         old = current_cred();
 666
 667         retval = -EPERM;
 668         if (ruid != (uid_t) -1) {
 669                 new->uid = ruid;
 670                 if (old->uid != ruid &&
 671                     old->euid != ruid &&
 672                     !nsown_capable(CAP_SETUID))
 673                         goto error;
 674         }
 675
 676         if (euid != (uid_t) -1) {
 677                 new->euid = euid;
 678                 if (old->uid != euid &&
 679                     old->euid != euid &&
 680                     old->suid != euid &&
 681                     !nsown_capable(CAP_SETUID))
 682                         goto error;
 683         }
 684
 685         if (new->uid != old->uid) {
 686                 retval = set_user(new);
 687                 if (retval < 0)
 688                         goto error;
 689         }
 690         if (ruid != (uid_t) -1 ||
 691             (euid != (uid_t) -1 && euid != old->uid))
 692                 new->suid = new->euid;
 693         new->fsuid = new->euid;
 694
 695         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 696         if (retval < 0)
 697                 goto error;
 698
 699         return commit_creds(new);
 700
 701 error:
 702         abort_creds(new);
 703         return retval;
 704 }
 705
 706 /*
 707  * setuid() is implemented like SysV with SAVED_IDS
 708  *
 709  * Note that SAVED_ID's is deficient in that a setuid root program
 710  * like sendmail, for example, cannot set its uid to be a normal
 711  * user and then switch back, because if you're root, setuid() sets
 712  * the saved uid too.  If you don't like this, blame the bright people
 713  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 714  * will allow a root program to temporarily drop privileges and be able to
 715  * regain them by swapping the real and effective uid.
 716  */
 717 SYSCALL_DEFINE1(setuid, uid_t, uid)
 718 {
 719         const struct cred *old;
 720         struct cred *new;
 721         int retval;
 722
 723         new = prepare_creds();
 724         if (!new)
 725                 return -ENOMEM;
 726         old = current_cred();
 727
 728         retval = -EPERM;
 729         if (nsown_capable(CAP_SETUID)) {
 730                 new->suid = new->uid = uid;
 731                 if (uid != old->uid) {
 732                         retval = set_user(new);
 733                         if (retval < 0)
 734                                 goto error;
 735                 }
 736         } else if (uid != old->uid && uid != new->suid) {
 737                 goto error;
 738         }
 739
 740         new->fsuid = new->euid = uid;
 741
 742         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 743         if (retval < 0)
 744                 goto error;
 745
 746         return commit_creds(new);
 747
 748 error:
 749         abort_creds(new);
 750         return retval;
 751 }
 752
 753
 754 /*
 755  * This function implements a generic ability to update ruid, euid,
 756  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 757  */
 758 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 759 {
 760         const struct cred *old;
 761         struct cred *new;
 762         int retval;
 763
 764         new = prepare_creds();
 765         if (!new)
 766                 return -ENOMEM;
 767
 768         old = current_cred();
 769
 770         retval = -EPERM;
 771         if (!nsown_capable(CAP_SETUID)) {
 772                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 773                     ruid != old->euid  && ruid != old->suid)
 774                         goto error;
 775                 if (euid != (uid_t) -1 && euid != old->uid &&
 776                     euid != old->euid  && euid != old->suid)
 777                         goto error;
 778                 if (suid != (uid_t) -1 && suid != old->uid &&
 779                     suid != old->euid  && suid != old->suid)
 780                         goto error;
 781         }
 782
 783         if (ruid != (uid_t) -1) {
 784                 new->uid = ruid;
 785                 if (ruid != old->uid) {
 786                         retval = set_user(new);
 787                         if (retval < 0)
 788                                 goto error;
 789                 }
 790         }
 791         if (euid != (uid_t) -1)
 792                 new->euid = euid;
 793         if (suid != (uid_t) -1)
 794                 new->suid = suid;
 795         new->fsuid = new->euid;
 796
 797         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 798         if (retval < 0)
 799                 goto error;
 800
 801         return commit_creds(new);
 802
 803 error:
 804         abort_creds(new);
 805         return retval;
 806 }
 807
 808 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 809 {
 810         const struct cred *cred = current_cred();
 811         int retval;
 812
 813         if (!(retval   = put_user(cred->uid,  ruid)) &&
 814             !(retval   = put_user(cred->euid, euid)))
 815                 retval = put_user(cred->suid, suid);
 816
 817         return retval;
 818 }
 819
 820 /*
 821  * Same as above, but for rgid, egid, sgid.
 822  */
 823 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 824 {
 825         const struct cred *old;
 826         struct cred *new;
 827         int retval;
 828
 829         new = prepare_creds();
 830         if (!new)
 831                 return -ENOMEM;
 832         old = current_cred();
 833
 834         retval = -EPERM;
 835         if (!nsown_capable(CAP_SETGID)) {
 836                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 837                     rgid != old->egid  && rgid != old->sgid)
 838                         goto error;
 839                 if (egid != (gid_t) -1 && egid != old->gid &&
 840                     egid != old->egid  && egid != old->sgid)
 841                         goto error;
 842                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 843                     sgid != old->egid  && sgid != old->sgid)
 844                         goto error;
 845         }
 846
 847         if (rgid != (gid_t) -1)
 848                 new->gid = rgid;
 849         if (egid != (gid_t) -1)
 850                 new->egid = egid;
 851         if (sgid != (gid_t) -1)
 852                 new->sgid = sgid;
 853         new->fsgid = new->egid;
 854
 855         return commit_creds(new);
 856
 857 error:
 858         abort_creds(new);
 859         return retval;
 860 }
 861
 862 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 863 {
 864         const struct cred *cred = current_cred();
 865         int retval;
 866
 867         if (!(retval   = put_user(cred->gid,  rgid)) &&
 868             !(retval   = put_user(cred->egid, egid)))
 869                 retval = put_user(cred->sgid, sgid);
 870
 871         return retval;
 872 }
 873
 874
 875 /*
 876  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 877  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 878  * whatever uid it wants to). It normally shadows "euid", except when
 879  * explicitly set by setfsuid() or for access..
 880  */
 881 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 882 {
 883         const struct cred *old;
 884         struct cred *new;
 885         uid_t old_fsuid;
 886
 887         new = prepare_creds();
 888         if (!new)
 889                 return current_fsuid();
 890         old = current_cred();
 891         old_fsuid = old->fsuid;
 892
 893         if (uid == old->uid  || uid == old->euid  ||
 894             uid == old->suid || uid == old->fsuid ||
 895             nsown_capable(CAP_SETUID)) {
 896                 if (uid != old_fsuid) {
 897                         new->fsuid = uid;
 898                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 899                                 goto change_okay;
 900                 }
 901         }
 902
 903         abort_creds(new);
 904         return old_fsuid;
 905
 906 change_okay:
 907         commit_creds(new);
 908         return old_fsuid;
 909 }
 910
 911 /*
 912  * Samma på svenska..
 913  */
 914 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 915 {
 916         const struct cred *old;
 917         struct cred *new;
 918         gid_t old_fsgid;
 919
 920         new = prepare_creds();
 921         if (!new)
 922                 return current_fsgid();
 923         old = current_cred();
 924         old_fsgid = old->fsgid;
 925
 926         if (gid == old->gid  || gid == old->egid  ||
 927             gid == old->sgid || gid == old->fsgid ||
 928             nsown_capable(CAP_SETGID)) {
 929                 if (gid != old_fsgid) {
 930                         new->fsgid = gid;
 931                         goto change_okay;
 932                 }
 933         }
 934
 935         abort_creds(new);
 936         return old_fsgid;
 937
 938 change_okay:
 939         commit_creds(new);
 940         return old_fsgid;
 941 }
 942
 943 void do_sys_times(struct tms *tms)
 944 {
 945         cputime_t tgutime, tgstime, cutime, cstime;
 946
 947         spin_lock_irq(&current->sighand->siglock);
 948         thread_group_times(current, &tgutime, &tgstime);
 949         cutime = current->signal->cutime;
 950         cstime = current->signal->cstime;
 951         spin_unlock_irq(&current->sighand->siglock);
 952         tms->tms_utime = cputime_to_clock_t(tgutime);
 953         tms->tms_stime = cputime_to_clock_t(tgstime);
 954         tms->tms_cutime = cputime_to_clock_t(cutime);
 955         tms->tms_cstime = cputime_to_clock_t(cstime);
 956 }
 957
 958 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 959 {
 960         if (tbuf) {
 961                 struct tms tmp;
 962
 963                 do_sys_times(&tmp);
 964                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 965                         return -EFAULT;
 966         }
 967         force_successful_syscall_return();
 968         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 969 }
 970
 971 /*
 972  * This needs some heavy checking ...
 973  * I just haven't the stomach for it. I also don't fully
 974  * understand sessions/pgrp etc. Let somebody who does explain it.
 975  *
 976  * OK, I think I have the protection semantics right.... this is really
 977  * only important on a multi-user system anyway, to make sure one user
 978  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 979  *
 980  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 981  * LBT 04.03.94
 982  */
 983 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 984 {
 985         struct task_struct *p;
 986         struct task_struct *group_leader = current->group_leader;
 987         struct pid *pgrp;
 988         int err;
 989
 990         if (!pid)
 991                 pid = task_pid_vnr(group_leader);
 992         if (!pgid)
 993                 pgid = pid;
 994         if (pgid < 0)
 995                 return -EINVAL;
 996         rcu_read_lock();
 997
 998         /* From this point forward we keep holding onto the tasklist lock
 999          * so that our parent does not change from under us. -DaveM
1000          */
1001         write_lock_irq(&tasklist_lock);
1002
1003         err = -ESRCH;
1004         p = find_task_by_vpid(pid);
1005         if (!p)
1006                 goto out;
1007
1008         err = -EINVAL;
1009         if (!thread_group_leader(p))
1010                 goto out;
1011
1012         if (same_thread_group(p->real_parent, group_leader)) {
1013                 err = -EPERM;
1014                 if (task_session(p) != task_session(group_leader))
1015                         goto out;
1016                 err = -EACCES;
1017                 if (p->did_exec)
1018                         goto out;
1019         } else {
1020                 err = -ESRCH;
1021                 if (p != group_leader)
1022                         goto out;
1023         }
1024
1025         err = -EPERM;
1026         if (p->signal->leader)
1027                 goto out;
1028
1029         pgrp = task_pid(p);
1030         if (pgid != pid) {
1031                 struct task_struct *g;
1032
1033                 pgrp = find_vpid(pgid);
1034                 g = pid_task(pgrp, PIDTYPE_PGID);
1035                 if (!g || task_session(g) != task_session(group_leader))
1036                         goto out;
1037         }
1038
1039         err = security_task_setpgid(p, pgid);
1040         if (err)
1041                 goto out;
1042
1043         if (task_pgrp(p) != pgrp)
1044                 change_pid(p, PIDTYPE_PGID, pgrp);
1045
1046         err = 0;
1047 out:
1048         /* All paths lead to here, thus we are safe. -DaveM */
1049         write_unlock_irq(&tasklist_lock);
1050         rcu_read_unlock();
1051         return err;
1052 }
1053
1054 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1055 {
1056         struct task_struct *p;
1057         struct pid *grp;
1058         int retval;
1059
1060         rcu_read_lock();
1061         if (!pid)
1062                 grp = task_pgrp(current);
1063         else {
1064                 retval = -ESRCH;
1065                 p = find_task_by_vpid(pid);
1066                 if (!p)
1067                         goto out;
1068                 grp = task_pgrp(p);
1069                 if (!grp)
1070                         goto out;
1071
1072                 retval = security_task_getpgid(p);
1073                 if (retval)
1074                         goto out;
1075         }
1076         retval = pid_vnr(grp);
1077 out:
1078         rcu_read_unlock();
1079         return retval;
1080 }
1081
1082 #ifdef __ARCH_WANT_SYS_GETPGRP
1083
1084 SYSCALL_DEFINE0(getpgrp)
1085 {
1086         return sys_getpgid(0);
1087 }
1088
1089 #endif
1090
1091 SYSCALL_DEFINE1(getsid, pid_t, pid)
1092 {
1093         struct task_struct *p;
1094         struct pid *sid;
1095         int retval;
1096
1097         rcu_read_lock();
1098         if (!pid)
1099                 sid = task_session(current);
1100         else {
1101                 retval = -ESRCH;
1102                 p = find_task_by_vpid(pid);
1103                 if (!p)
1104                         goto out;
1105                 sid = task_session(p);
1106                 if (!sid)
1107                         goto out;
1108
1109                 retval = security_task_getsid(p);
1110                 if (retval)
1111                         goto out;
1112         }
1113         retval = pid_vnr(sid);
1114 out:
1115         rcu_read_unlock();
1116         return retval;
1117 }
1118
1119 SYSCALL_DEFINE0(setsid)
1120 {
1121         struct task_struct *group_leader = current->group_leader;
1122         struct pid *sid = task_pid(group_leader);
1123         pid_t session = pid_vnr(sid);
1124         int err = -EPERM;
1125
1126         write_lock_irq(&tasklist_lock);
1127         /* Fail if I am already a session leader */
1128         if (group_leader->signal->leader)
1129                 goto out;
1130
1131         /* Fail if a process group id already exists that equals the
1132          * proposed session id.
1133          */
1134         if (pid_task(sid, PIDTYPE_PGID))
1135                 goto out;
1136
1137         group_leader->signal->leader = 1;
1138         __set_special_pids(sid);
1139
1140         proc_clear_tty(group_leader);
1141
1142         err = session;
1143 out:
1144         write_unlock_irq(&tasklist_lock);
1145         if (err > 0) {
1146                 proc_sid_connector(group_leader);
1147                 sched_autogroup_create_attach(group_leader);
1148         }
1149         return err;
1150 }
1151
1152 DECLARE_RWSEM(uts_sem);
1153
1154 #ifdef COMPAT_UTS_MACHINE
1155 #define override_architecture(name) \
1156         (personality(current->personality) == PER_LINUX32 && \
1157          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1158                       sizeof(COMPAT_UTS_MACHINE)))
1159 #else
1160 #define override_architecture(name)     0
1161 #endif
1162
1163 /*
1164  * Work around broken programs that cannot handle "Linux 3.0".
1165  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1166  */
1167 static int override_release(char __user *release, size_t len)
1168 {
1169         int ret = 0;
1170
1171         if (current->personality & UNAME26) {
1172                 const char *rest = UTS_RELEASE;
1173                 char buf[65] = { 0 };
1174                 int ndots = 0;
1175                 unsigned v;
1176                 size_t copy;
1177
1178                 while (*rest) {
1179                         if (*rest == '.' && ++ndots >= 3)
1180                                 break;
1181                         if (!isdigit(*rest) && *rest != '.')
1182                                 break;
1183                         rest++;
1184                 }
1185                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1186                 copy = clamp_t(size_t, len, 1, sizeof(buf));
1187                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1188                 ret = copy_to_user(release, buf, copy + 1);
1189         }
1190         return ret;
1191 }
1192
1193 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1194 {
1195         int errno = 0;
1196
1197         down_read(&uts_sem);
1198         if (copy_to_user(name, utsname(), sizeof *name))
1199                 errno = -EFAULT;
1200         up_read(&uts_sem);
1201
1202         if (!errno && override_release(name->release, sizeof(name->release)))
1203                 errno = -EFAULT;
1204         if (!errno && override_architecture(name))
1205                 errno = -EFAULT;
1206         return errno;
1207 }
1208
1209 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1210 /*
1211  * Old cruft
1212  */
1213 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1214 {
1215         int error = 0;
1216
1217         if (!name)
1218                 return -EFAULT;
1219
1220         down_read(&uts_sem);
1221         if (copy_to_user(name, utsname(), sizeof(*name)))
1222                 error = -EFAULT;
1223         up_read(&uts_sem);
1224
1225         if (!error && override_release(name->release, sizeof(name->release)))
1226                 error = -EFAULT;
1227         if (!error && override_architecture(name))
1228                 error = -EFAULT;
1229         return error;
1230 }
1231
1232 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1233 {
1234         int error;
1235
1236         if (!name)
1237                 return -EFAULT;
1238         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1239                 return -EFAULT;
1240
1241         down_read(&uts_sem);
1242         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1243                                __OLD_UTS_LEN);
1244         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1245         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1246                                 __OLD_UTS_LEN);
1247         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1248         error |= __copy_to_user(&name->release, &utsname()->release,
1249                                 __OLD_UTS_LEN);
1250         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1251         error |= __copy_to_user(&name->version, &utsname()->version,
1252                                 __OLD_UTS_LEN);
1253         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1254         error |= __copy_to_user(&name->machine, &utsname()->machine,
1255                                 __OLD_UTS_LEN);
1256         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1257         up_read(&uts_sem);
1258
1259         if (!error && override_architecture(name))
1260                 error = -EFAULT;
1261         if (!error && override_release(name->release, sizeof(name->release)))
1262                 error = -EFAULT;
1263         return error ? -EFAULT : 0;
1264 }
1265 #endif
1266
1267 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1268 {
1269         int errno;
1270         char tmp[__NEW_UTS_LEN];
1271
1272         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1273                 return -EPERM;
1274
1275         if (len < 0 || len > __NEW_UTS_LEN)
1276                 return -EINVAL;
1277         down_write(&uts_sem);
1278         errno = -EFAULT;
1279         if (!copy_from_user(tmp, name, len)) {
1280                 struct new_utsname *u = utsname();
1281
1282                 memcpy(u->nodename, tmp, len);
1283                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1284                 errno = 0;
1285         }
1286         up_write(&uts_sem);
1287         return errno;
1288 }
1289
1290 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1291
1292 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1293 {
1294         int i, errno;
1295         struct new_utsname *u;
1296
1297         if (len < 0)
1298                 return -EINVAL;
1299         down_read(&uts_sem);
1300         u = utsname();
1301         i = 1 + strlen(u->nodename);
1302         if (i > len)
1303                 i = len;
1304         errno = 0;
1305         if (copy_to_user(name, u->nodename, i))
1306                 errno = -EFAULT;
1307         up_read(&uts_sem);
1308         return errno;
1309 }
1310
1311 #endif
1312
1313 /*
1314  * Only setdomainname; getdomainname can be implemented by calling
1315  * uname()
1316  */
1317 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1318 {
1319         int errno;
1320         char tmp[__NEW_UTS_LEN];
1321
1322         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1323                 return -EPERM;
1324         if (len < 0 || len > __NEW_UTS_LEN)
1325                 return -EINVAL;
1326
1327         down_write(&uts_sem);
1328         errno = -EFAULT;
1329         if (!copy_from_user(tmp, name, len)) {
1330                 struct new_utsname *u = utsname();
1331
1332                 memcpy(u->domainname, tmp, len);
1333                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1334                 errno = 0;
1335         }
1336         up_write(&uts_sem);
1337         return errno;
1338 }
1339
1340 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1341 {
1342         struct rlimit value;
1343         int ret;
1344
1345         ret = do_prlimit(current, resource, NULL, &value);
1346         if (!ret)
1347                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1348
1349         return ret;
1350 }
1351
1352 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1353
1354 /*
1355  *      Back compatibility for getrlimit. Needed for some apps.
1356  */
1357
1358 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1359                 struct rlimit __user *, rlim)
1360 {
1361         struct rlimit x;
1362         if (resource >= RLIM_NLIMITS)
1363                 return -EINVAL;
1364
1365         task_lock(current->group_leader);
1366         x = current->signal->rlim[resource];
1367         task_unlock(current->group_leader);
1368         if (x.rlim_cur > 0x7FFFFFFF)
1369                 x.rlim_cur = 0x7FFFFFFF;
1370         if (x.rlim_max > 0x7FFFFFFF)
1371                 x.rlim_max = 0x7FFFFFFF;
1372         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1373 }
1374
1375 #endif
1376
1377 static inline bool rlim64_is_infinity(__u64 rlim64)
1378 {
1379 #if BITS_PER_LONG < 64
1380         return rlim64 >= ULONG_MAX;
1381 #else
1382         return rlim64 == RLIM64_INFINITY;
1383 #endif
1384 }
1385
1386 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1387 {
1388         if (rlim->rlim_cur == RLIM_INFINITY)
1389                 rlim64->rlim_cur = RLIM64_INFINITY;
1390         else
1391                 rlim64->rlim_cur = rlim->rlim_cur;
1392         if (rlim->rlim_max == RLIM_INFINITY)
1393                 rlim64->rlim_max = RLIM64_INFINITY;
1394         else
1395                 rlim64->rlim_max = rlim->rlim_max;
1396 }
1397
1398 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1399 {
1400         if (rlim64_is_infinity(rlim64->rlim_cur))
1401                 rlim->rlim_cur = RLIM_INFINITY;
1402         else
1403                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1404         if (rlim64_is_infinity(rlim64->rlim_max))
1405                 rlim->rlim_max = RLIM_INFINITY;
1406         else
1407                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1408 }
1409
1410 /* make sure you are allowed to change @tsk limits before calling this */
1411 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1412                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1413 {
1414         struct rlimit *rlim;
1415         int retval = 0;
1416
1417         if (resource >= RLIM_NLIMITS)
1418                 return -EINVAL;
1419         if (new_rlim) {
1420                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1421                         return -EINVAL;
1422                 if (resource == RLIMIT_NOFILE &&
1423                                 new_rlim->rlim_max > sysctl_nr_open)
1424                         return -EPERM;
1425         }
1426
1427         /* protect tsk->signal and tsk->sighand from disappearing */
1428         read_lock(&tasklist_lock);
1429         if (!tsk->sighand) {
1430                 retval = -ESRCH;
1431                 goto out;
1432         }
1433
1434         rlim = tsk->signal->rlim + resource;
1435         task_lock(tsk->group_leader);
1436         if (new_rlim) {
1437                 /* Keep the capable check against init_user_ns until
1438                    cgroups can contain all limits */
1439                 if (new_rlim->rlim_max > rlim->rlim_max &&
1440                                 !capable(CAP_SYS_RESOURCE))
1441                         retval = -EPERM;
1442                 if (!retval)
1443                         retval = security_task_setrlimit(tsk->group_leader,
1444                                         resource, new_rlim);
1445                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1446                         /*
1447                          * The caller is asking for an immediate RLIMIT_CPU
1448                          * expiry.  But we use the zero value to mean "it was
1449                          * never set".  So let's cheat and make it one second
1450                          * instead
1451                          */
1452                         new_rlim->rlim_cur = 1;
1453                 }
1454         }
1455         if (!retval) {
1456                 if (old_rlim)
1457                         *old_rlim = *rlim;
1458                 if (new_rlim)
1459                         *rlim = *new_rlim;
1460         }
1461         task_unlock(tsk->group_leader);
1462
1463         /*
1464          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1465          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1466          * very long-standing error, and fixing it now risks breakage of
1467          * applications, so we live with it
1468          */
1469          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1470                          new_rlim->rlim_cur != RLIM_INFINITY)
1471                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1472 out:
1473         read_unlock(&tasklist_lock);
1474         return retval;
1475 }
1476
1477 /* rcu lock must be held */
1478 static int check_prlimit_permission(struct task_struct *task)
1479 {
1480         const struct cred *cred = current_cred(), *tcred;
1481
1482         if (current == task)
1483                 return 0;
1484
1485         tcred = __task_cred(task);
1486         if (cred->user->user_ns == tcred->user->user_ns &&
1487             (cred->uid == tcred->euid &&
1488              cred->uid == tcred->suid &&
1489              cred->uid == tcred->uid  &&
1490              cred->gid == tcred->egid &&
1491              cred->gid == tcred->sgid &&
1492              cred->gid == tcred->gid))
1493                 return 0;
1494         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1495                 return 0;
1496
1497         return -EPERM;
1498 }
1499
1500 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1501                 const struct rlimit64 __user *, new_rlim,
1502                 struct rlimit64 __user *, old_rlim)
1503 {
1504         struct rlimit64 old64, new64;
1505         struct rlimit old, new;
1506         struct task_struct *tsk;
1507         int ret;
1508
1509         if (new_rlim) {
1510                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1511                         return -EFAULT;
1512                 rlim64_to_rlim(&new64, &new);
1513         }
1514
1515         rcu_read_lock();
1516         tsk = pid ? find_task_by_vpid(pid) : current;
1517         if (!tsk) {
1518                 rcu_read_unlock();
1519                 return -ESRCH;
1520         }
1521         ret = check_prlimit_permission(tsk);
1522         if (ret) {
1523                 rcu_read_unlock();
1524                 return ret;
1525         }
1526         get_task_struct(tsk);
1527         rcu_read_unlock();
1528
1529         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1530                         old_rlim ? &old : NULL);
1531
1532         if (!ret && old_rlim) {
1533                 rlim_to_rlim64(&old, &old64);
1534                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1535                         ret = -EFAULT;
1536         }
1537
1538         put_task_struct(tsk);
1539         return ret;
1540 }
1541
1542 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1543 {
1544         struct rlimit new_rlim;
1545
1546         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1547                 return -EFAULT;
1548         return do_prlimit(current, resource, &new_rlim, NULL);
1549 }
1550
1551 /*
1552  * It would make sense to put struct rusage in the task_struct,
1553  * except that would make the task_struct be *really big*.  After
1554  * task_struct gets moved into malloc'ed memory, it would
1555  * make sense to do this.  It will make moving the rest of the information
1556  * a lot simpler!  (Which we're not doing right now because we're not
1557  * measuring them yet).
1558  *
1559  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1560  * races with threads incrementing their own counters.  But since word
1561  * reads are atomic, we either get new values or old values and we don't
1562  * care which for the sums.  We always take the siglock to protect reading
1563  * the c* fields from p->signal from races with exit.c updating those
1564  * fields when reaping, so a sample either gets all the additions of a
1565  * given child after it's reaped, or none so this sample is before reaping.
1566  *
1567  * Locking:
1568  * We need to take the siglock for CHILDEREN, SELF and BOTH
1569  * for  the cases current multithreaded, non-current single threaded
1570  * non-current multithreaded.  Thread traversal is now safe with
1571  * the siglock held.
1572  * Strictly speaking, we donot need to take the siglock if we are current and
1573  * single threaded,  as no one else can take our signal_struct away, no one
1574  * else can  reap the  children to update signal->c* counters, and no one else
1575  * can race with the signal-> fields. If we do not take any lock, the
1576  * signal-> fields could be read out of order while another thread was just
1577  * exiting. So we should  place a read memory barrier when we avoid the lock.
1578  * On the writer side,  write memory barrier is implied in  __exit_signal
1579  * as __exit_signal releases  the siglock spinlock after updating the signal->
1580  * fields. But we don't do this yet to keep things simple.
1581  *
1582  */
1583
1584 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1585 {
1586         r->ru_nvcsw += t->nvcsw;
1587         r->ru_nivcsw += t->nivcsw;
1588         r->ru_minflt += t->min_flt;
1589         r->ru_majflt += t->maj_flt;
1590         r->ru_inblock += task_io_get_inblock(t);
1591         r->ru_oublock += task_io_get_oublock(t);
1592 }
1593
1594 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1595 {
1596         struct task_struct *t;
1597         unsigned long flags;
1598         cputime_t tgutime, tgstime, utime, stime;
1599         unsigned long maxrss = 0;
1600
1601         memset((char *) r, 0, sizeof *r);
1602         utime = stime = cputime_zero;
1603
1604         if (who == RUSAGE_THREAD) {
1605                 task_times(current, &utime, &stime);
1606                 accumulate_thread_rusage(p, r);
1607                 maxrss = p->signal->maxrss;
1608                 goto out;
1609         }
1610
1611         if (!lock_task_sighand(p, &flags))
1612                 return;
1613
1614         switch (who) {
1615                 case RUSAGE_BOTH:
1616                 case RUSAGE_CHILDREN:
1617                         utime = p->signal->cutime;
1618                         stime = p->signal->cstime;
1619                         r->ru_nvcsw = p->signal->cnvcsw;
1620                         r->ru_nivcsw = p->signal->cnivcsw;
1621                         r->ru_minflt = p->signal->cmin_flt;
1622                         r->ru_majflt = p->signal->cmaj_flt;
1623                         r->ru_inblock = p->signal->cinblock;
1624                         r->ru_oublock = p->signal->coublock;
1625                         maxrss = p->signal->cmaxrss;
1626
1627                         if (who == RUSAGE_CHILDREN)
1628                                 break;
1629
1630                 case RUSAGE_SELF:
1631                         thread_group_times(p, &tgutime, &tgstime);
1632                         utime = cputime_add(utime, tgutime);
1633                         stime = cputime_add(stime, tgstime);
1634                         r->ru_nvcsw += p->signal->nvcsw;
1635                         r->ru_nivcsw += p->signal->nivcsw;
1636                         r->ru_minflt += p->signal->min_flt;
1637                         r->ru_majflt += p->signal->maj_flt;
1638                         r->ru_inblock += p->signal->inblock;
1639                         r->ru_oublock += p->signal->oublock;
1640                         if (maxrss < p->signal->maxrss)
1641                                 maxrss = p->signal->maxrss;
1642                         t = p;
1643                         do {
1644                                 accumulate_thread_rusage(t, r);
1645                                 t = next_thread(t);
1646                         } while (t != p);
1647                         break;
1648
1649                 default:
1650                         BUG();
1651         }
1652         unlock_task_sighand(p, &flags);
1653
1654 out:
1655         cputime_to_timeval(utime, &r->ru_utime);
1656         cputime_to_timeval(stime, &r->ru_stime);
1657
1658         if (who != RUSAGE_CHILDREN) {
1659                 struct mm_struct *mm = get_task_mm(p);
1660                 if (mm) {
1661                         setmax_mm_hiwater_rss(&maxrss, mm);
1662                         mmput(mm);
1663                 }
1664         }
1665         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1666 }
1667
1668 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1669 {
1670         struct rusage r;
1671         k_getrusage(p, who, &r);
1672         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1673 }
1674
1675 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1676 {
1677         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1678             who != RUSAGE_THREAD)
1679                 return -EINVAL;
1680         return getrusage(current, who, ru);
1681 }
1682
1683 SYSCALL_DEFINE1(umask, int, mask)
1684 {
1685         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1686         return mask;
1687 }
1688
1689 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1690                 unsigned long, arg4, unsigned long, arg5)
1691 {
1692         struct task_struct *me = current;
1693         unsigned char comm[sizeof(me->comm)];
1694         long error;
1695
1696         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1697         if (error != -ENOSYS)
1698                 return error;
1699
1700         error = 0;
1701         switch (option) {
1702                 case PR_SET_PDEATHSIG:
1703                         if (!valid_signal(arg2)) {
1704                                 error = -EINVAL;
1705                                 break;
1706                         }
1707                         me->pdeath_signal = arg2;
1708                         error = 0;
1709                         break;
1710                 case PR_GET_PDEATHSIG:
1711                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1712                         break;
1713                 case PR_GET_DUMPABLE:
1714                         error = get_dumpable(me->mm);
1715                         break;
1716                 case PR_SET_DUMPABLE:
1717                         if (arg2 < 0 || arg2 > 1) {
1718                                 error = -EINVAL;
1719                                 break;
1720                         }
1721                         set_dumpable(me->mm, arg2);
1722                         error = 0;
1723                         break;
1724
1725                 case PR_SET_UNALIGN:
1726                         error = SET_UNALIGN_CTL(me, arg2);
1727                         break;
1728                 case PR_GET_UNALIGN:
1729                         error = GET_UNALIGN_CTL(me, arg2);
1730                         break;
1731                 case PR_SET_FPEMU:
1732                         error = SET_FPEMU_CTL(me, arg2);
1733                         break;
1734                 case PR_GET_FPEMU:
1735                         error = GET_FPEMU_CTL(me, arg2);
1736                         break;
1737                 case PR_SET_FPEXC:
1738                         error = SET_FPEXC_CTL(me, arg2);
1739                         break;
1740                 case PR_GET_FPEXC:
1741                         error = GET_FPEXC_CTL(me, arg2);
1742                         break;
1743                 case PR_GET_TIMING:
1744                         error = PR_TIMING_STATISTICAL;
1745                         break;
1746                 case PR_SET_TIMING:
1747                         if (arg2 != PR_TIMING_STATISTICAL)
1748                                 error = -EINVAL;
1749                         else
1750                                 error = 0;
1751                         break;
1752
1753                 case PR_SET_NAME:
1754                         comm[sizeof(me->comm)-1] = 0;
1755                         if (strncpy_from_user(comm, (char __user *)arg2,
1756                                               sizeof(me->comm) - 1) < 0)
1757                                 return -EFAULT;
1758                         set_task_comm(me, comm);
1759                         return 0;
1760                 case PR_GET_NAME:
1761                         get_task_comm(comm, me);
1762                         if (copy_to_user((char __user *)arg2, comm,
1763                                          sizeof(comm)))
1764                                 return -EFAULT;
1765                         return 0;
1766                 case PR_GET_ENDIAN:
1767                         error = GET_ENDIAN(me, arg2);
1768                         break;
1769                 case PR_SET_ENDIAN:
1770                         error = SET_ENDIAN(me, arg2);
1771                         break;
1772
1773                 case PR_GET_SECCOMP:
1774                         error = prctl_get_seccomp();
1775                         break;
1776                 case PR_SET_SECCOMP:
1777                         error = prctl_set_seccomp(arg2);
1778                         break;
1779                 case PR_GET_TSC:
1780                         error = GET_TSC_CTL(arg2);
1781                         break;
1782                 case PR_SET_TSC:
1783                         error = SET_TSC_CTL(arg2);
1784                         break;
1785                 case PR_TASK_PERF_EVENTS_DISABLE:
1786                         error = perf_event_task_disable();
1787                         break;
1788                 case PR_TASK_PERF_EVENTS_ENABLE:
1789                         error = perf_event_task_enable();
1790                         break;
1791                 case PR_GET_TIMERSLACK:
1792                         error = current->timer_slack_ns;
1793                         break;
1794                 case PR_SET_TIMERSLACK:
1795                         if (arg2 <= 0)
1796                                 current->timer_slack_ns =
1797                                         current->default_timer_slack_ns;
1798                         else
1799                                 current->timer_slack_ns = arg2;
1800                         error = 0;
1801                         break;
1802                 case PR_MCE_KILL:
1803                         if (arg4 | arg5)
1804                                 return -EINVAL;
1805                         switch (arg2) {
1806                         case PR_MCE_KILL_CLEAR:
1807                                 if (arg3 != 0)
1808                                         return -EINVAL;
1809                                 current->flags &= ~PF_MCE_PROCESS;
1810                                 break;
1811                         case PR_MCE_KILL_SET:
1812                                 current->flags |= PF_MCE_PROCESS;
1813                                 if (arg3 == PR_MCE_KILL_EARLY)
1814                                         current->flags |= PF_MCE_EARLY;
1815                                 else if (arg3 == PR_MCE_KILL_LATE)
1816                                         current->flags &= ~PF_MCE_EARLY;
1817                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1818                                         current->flags &=
1819                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1820                                 else
1821                                         return -EINVAL;
1822                                 break;
1823                         default:
1824                                 return -EINVAL;
1825                         }
1826                         error = 0;
1827                         break;
1828                 case PR_MCE_KILL_GET:
1829                         if (arg2 | arg3 | arg4 | arg5)
1830                                 return -EINVAL;
1831                         if (current->flags & PF_MCE_PROCESS)
1832                                 error = (current->flags & PF_MCE_EARLY) ?
1833                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1834                         else
1835                                 error = PR_MCE_KILL_DEFAULT;
1836                         break;
1837                 default:
1838                         error = -EINVAL;
1839                         break;
1840         }
1841         return error;
1842 }
1843
1844 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1845                 struct getcpu_cache __user *, unused)
1846 {
1847         int err = 0;
1848         int cpu = raw_smp_processor_id();
1849         if (cpup)
1850                 err |= put_user(cpu, cpup);
1851         if (nodep)
1852                 err |= put_user(cpu_to_node(cpu), nodep);
1853         return err ? -EFAULT : 0;
1854 }
1855
1856 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1857
1858 static void argv_cleanup(struct subprocess_info *info)
1859 {
1860         argv_free(info->argv);
1861 }
1862
1863 /**
1864  * orderly_poweroff - Trigger an orderly system poweroff
1865  * @force: force poweroff if command execution fails
1866  *
1867  * This may be called from any context to trigger a system shutdown.
1868  * If the orderly shutdown fails, it will force an immediate shutdown.
1869  */
1870 int orderly_poweroff(bool force)
1871 {
1872         int argc;
1873         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1874         static char *envp[] = {
1875                 "HOME=/",
1876                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1877                 NULL
1878         };
1879         int ret = -ENOMEM;
1880         struct subprocess_info *info;
1881
1882         if (argv == NULL) {
1883                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1884                        __func__, poweroff_cmd);
1885                 goto out;
1886         }
1887
1888         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1889         if (info == NULL) {
1890                 argv_free(argv);
1891                 goto out;
1892         }
1893
1894         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1895
1896         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1897
1898   out:
1899         if (ret && force) {
1900                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1901                        "forcing the issue\n");
1902
1903                 /* I guess this should try to kick off some daemon to
1904                    sync and poweroff asap.  Or not even bother syncing
1905                    if we're doing an emergency shutdown? */
1906                 emergency_sync();
1907                 kernel_power_off();
1908         }
1909
1910         return ret;
1911 }
1912 EXPORT_SYMBOL_GPL(orderly_poweroff);