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