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         /*
 625          * We don't fail in case of NPROC limit excess here because too many
 626          * poorly written programs don't check set*uid() return code, assuming
 627          * it never fails if called by root.  We may still enforce NPROC limit
 628          * for programs doing set*uid()+execve() by harmlessly deferring the
 629          * failure to the execve() stage.
 630          */
 631         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 632                         new_user != INIT_USER)
 633                 current->flags |= PF_NPROC_EXCEEDED;
 634         else
 635                 current->flags &= ~PF_NPROC_EXCEEDED;
 636
 637         free_uid(new->user);
 638         new->user = new_user;
 639         return 0;
 640 }
 641
 642 /*
 643  * Unprivileged users may change the real uid to the effective uid
 644  * or vice versa.  (BSD-style)
 645  *
 646  * If you set the real uid at all, or set the effective uid to a value not
 647  * equal to the real uid, then the saved uid is set to the new effective uid.
 648  *
 649  * This makes it possible for a setuid program to completely drop its
 650  * privileges, which is often a useful assertion to make when you are doing
 651  * a security audit over a program.
 652  *
 653  * The general idea is that a program which uses just setreuid() will be
 654  * 100% compatible with BSD.  A program which uses just setuid() will be
 655  * 100% compatible with POSIX with saved IDs.
 656  */
 657 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 658 {
 659         const struct cred *old;
 660         struct cred *new;
 661         int retval;
 662
 663         new = prepare_creds();
 664         if (!new)
 665                 return -ENOMEM;
 666         old = current_cred();
 667
 668         retval = -EPERM;
 669         if (ruid != (uid_t) -1) {
 670                 new->uid = ruid;
 671                 if (old->uid != ruid &&
 672                     old->euid != ruid &&
 673                     !nsown_capable(CAP_SETUID))
 674                         goto error;
 675         }
 676
 677         if (euid != (uid_t) -1) {
 678                 new->euid = euid;
 679                 if (old->uid != euid &&
 680                     old->euid != euid &&
 681                     old->suid != euid &&
 682                     !nsown_capable(CAP_SETUID))
 683                         goto error;
 684         }
 685
 686         if (new->uid != old->uid) {
 687                 retval = set_user(new);
 688                 if (retval < 0)
 689                         goto error;
 690         }
 691         if (ruid != (uid_t) -1 ||
 692             (euid != (uid_t) -1 && euid != old->uid))
 693                 new->suid = new->euid;
 694         new->fsuid = new->euid;
 695
 696         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 697         if (retval < 0)
 698                 goto error;
 699
 700         return commit_creds(new);
 701
 702 error:
 703         abort_creds(new);
 704         return retval;
 705 }
 706
 707 /*
 708  * setuid() is implemented like SysV with SAVED_IDS
 709  *
 710  * Note that SAVED_ID's is deficient in that a setuid root program
 711  * like sendmail, for example, cannot set its uid to be a normal
 712  * user and then switch back, because if you're root, setuid() sets
 713  * the saved uid too.  If you don't like this, blame the bright people
 714  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 715  * will allow a root program to temporarily drop privileges and be able to
 716  * regain them by swapping the real and effective uid.
 717  */
 718 SYSCALL_DEFINE1(setuid, uid_t, uid)
 719 {
 720         const struct cred *old;
 721         struct cred *new;
 722         int retval;
 723
 724         new = prepare_creds();
 725         if (!new)
 726                 return -ENOMEM;
 727         old = current_cred();
 728
 729         retval = -EPERM;
 730         if (nsown_capable(CAP_SETUID)) {
 731                 new->suid = new->uid = uid;
 732                 if (uid != old->uid) {
 733                         retval = set_user(new);
 734                         if (retval < 0)
 735                                 goto error;
 736                 }
 737         } else if (uid != old->uid && uid != new->suid) {
 738                 goto error;
 739         }
 740
 741         new->fsuid = new->euid = uid;
 742
 743         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 744         if (retval < 0)
 745                 goto error;
 746
 747         return commit_creds(new);
 748
 749 error:
 750         abort_creds(new);
 751         return retval;
 752 }
 753
 754
 755 /*
 756  * This function implements a generic ability to update ruid, euid,
 757  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 758  */
 759 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 760 {
 761         const struct cred *old;
 762         struct cred *new;
 763         int retval;
 764
 765         new = prepare_creds();
 766         if (!new)
 767                 return -ENOMEM;
 768
 769         old = current_cred();
 770
 771         retval = -EPERM;
 772         if (!nsown_capable(CAP_SETUID)) {
 773                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 774                     ruid != old->euid  && ruid != old->suid)
 775                         goto error;
 776                 if (euid != (uid_t) -1 && euid != old->uid &&
 777                     euid != old->euid  && euid != old->suid)
 778                         goto error;
 779                 if (suid != (uid_t) -1 && suid != old->uid &&
 780                     suid != old->euid  && suid != old->suid)
 781                         goto error;
 782         }
 783
 784         if (ruid != (uid_t) -1) {
 785                 new->uid = ruid;
 786                 if (ruid != old->uid) {
 787                         retval = set_user(new);
 788                         if (retval < 0)
 789                                 goto error;
 790                 }
 791         }
 792         if (euid != (uid_t) -1)
 793                 new->euid = euid;
 794         if (suid != (uid_t) -1)
 795                 new->suid = suid;
 796         new->fsuid = new->euid;
 797
 798         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 799         if (retval < 0)
 800                 goto error;
 801
 802         return commit_creds(new);
 803
 804 error:
 805         abort_creds(new);
 806         return retval;
 807 }
 808
 809 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 810 {
 811         const struct cred *cred = current_cred();
 812         int retval;
 813
 814         if (!(retval   = put_user(cred->uid,  ruid)) &&
 815             !(retval   = put_user(cred->euid, euid)))
 816                 retval = put_user(cred->suid, suid);
 817
 818         return retval;
 819 }
 820
 821 /*
 822  * Same as above, but for rgid, egid, sgid.
 823  */
 824 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 825 {
 826         const struct cred *old;
 827         struct cred *new;
 828         int retval;
 829
 830         new = prepare_creds();
 831         if (!new)
 832                 return -ENOMEM;
 833         old = current_cred();
 834
 835         retval = -EPERM;
 836         if (!nsown_capable(CAP_SETGID)) {
 837                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 838                     rgid != old->egid  && rgid != old->sgid)
 839                         goto error;
 840                 if (egid != (gid_t) -1 && egid != old->gid &&
 841                     egid != old->egid  && egid != old->sgid)
 842                         goto error;
 843                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 844                     sgid != old->egid  && sgid != old->sgid)
 845                         goto error;
 846         }
 847
 848         if (rgid != (gid_t) -1)
 849                 new->gid = rgid;
 850         if (egid != (gid_t) -1)
 851                 new->egid = egid;
 852         if (sgid != (gid_t) -1)
 853                 new->sgid = sgid;
 854         new->fsgid = new->egid;
 855
 856         return commit_creds(new);
 857
 858 error:
 859         abort_creds(new);
 860         return retval;
 861 }
 862
 863 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 864 {
 865         const struct cred *cred = current_cred();
 866         int retval;
 867
 868         if (!(retval   = put_user(cred->gid,  rgid)) &&
 869             !(retval   = put_user(cred->egid, egid)))
 870                 retval = put_user(cred->sgid, sgid);
 871
 872         return retval;
 873 }
 874
 875
 876 /*
 877  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 878  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 879  * whatever uid it wants to). It normally shadows "euid", except when
 880  * explicitly set by setfsuid() or for access..
 881  */
 882 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 883 {
 884         const struct cred *old;
 885         struct cred *new;
 886         uid_t old_fsuid;
 887
 888         new = prepare_creds();
 889         if (!new)
 890                 return current_fsuid();
 891         old = current_cred();
 892         old_fsuid = old->fsuid;
 893
 894         if (uid == old->uid  || uid == old->euid  ||
 895             uid == old->suid || uid == old->fsuid ||
 896             nsown_capable(CAP_SETUID)) {
 897                 if (uid != old_fsuid) {
 898                         new->fsuid = uid;
 899                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 900                                 goto change_okay;
 901                 }
 902         }
 903
 904         abort_creds(new);
 905         return old_fsuid;
 906
 907 change_okay:
 908         commit_creds(new);
 909         return old_fsuid;
 910 }
 911
 912 /*
 913  * Samma på svenska..
 914  */
 915 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 916 {
 917         const struct cred *old;
 918         struct cred *new;
 919         gid_t old_fsgid;
 920
 921         new = prepare_creds();
 922         if (!new)
 923                 return current_fsgid();
 924         old = current_cred();
 925         old_fsgid = old->fsgid;
 926
 927         if (gid == old->gid  || gid == old->egid  ||
 928             gid == old->sgid || gid == old->fsgid ||
 929             nsown_capable(CAP_SETGID)) {
 930                 if (gid != old_fsgid) {
 931                         new->fsgid = gid;
 932                         goto change_okay;
 933                 }
 934         }
 935
 936         abort_creds(new);
 937         return old_fsgid;
 938
 939 change_okay:
 940         commit_creds(new);
 941         return old_fsgid;
 942 }
 943
 944 void do_sys_times(struct tms *tms)
 945 {
 946         cputime_t tgutime, tgstime, cutime, cstime;
 947
 948         spin_lock_irq(&current->sighand->siglock);
 949         thread_group_times(current, &tgutime, &tgstime);
 950         cutime = current->signal->cutime;
 951         cstime = current->signal->cstime;
 952         spin_unlock_irq(&current->sighand->siglock);
 953         tms->tms_utime = cputime_to_clock_t(tgutime);
 954         tms->tms_stime = cputime_to_clock_t(tgstime);
 955         tms->tms_cutime = cputime_to_clock_t(cutime);
 956         tms->tms_cstime = cputime_to_clock_t(cstime);
 957 }
 958
 959 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 960 {
 961         if (tbuf) {
 962                 struct tms tmp;
 963
 964                 do_sys_times(&tmp);
 965                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 966                         return -EFAULT;
 967         }
 968         force_successful_syscall_return();
 969         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 970 }
 971
 972 /*
 973  * This needs some heavy checking ...
 974  * I just haven't the stomach for it. I also don't fully
 975  * understand sessions/pgrp etc. Let somebody who does explain it.
 976  *
 977  * OK, I think I have the protection semantics right.... this is really
 978  * only important on a multi-user system anyway, to make sure one user
 979  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 980  *
 981  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 982  * LBT 04.03.94
 983  */
 984 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 985 {
 986         struct task_struct *p;
 987         struct task_struct *group_leader = current->group_leader;
 988         struct pid *pgrp;
 989         int err;
 990
 991         if (!pid)
 992                 pid = task_pid_vnr(group_leader);
 993         if (!pgid)
 994                 pgid = pid;
 995         if (pgid < 0)
 996                 return -EINVAL;
 997         rcu_read_lock();
 998
 999         /* From this point forward we keep holding onto the tasklist lock
1000          * so that our parent does not change from under us. -DaveM
1001          */
1002         write_lock_irq(&tasklist_lock);
1003
1004         err = -ESRCH;
1005         p = find_task_by_vpid(pid);
1006         if (!p)
1007                 goto out;
1008
1009         err = -EINVAL;
1010         if (!thread_group_leader(p))
1011                 goto out;
1012
1013         if (same_thread_group(p->real_parent, group_leader)) {
1014                 err = -EPERM;
1015                 if (task_session(p) != task_session(group_leader))
1016                         goto out;
1017                 err = -EACCES;
1018                 if (p->did_exec)
1019                         goto out;
1020         } else {
1021                 err = -ESRCH;
1022                 if (p != group_leader)
1023                         goto out;
1024         }
1025
1026         err = -EPERM;
1027         if (p->signal->leader)
1028                 goto out;
1029
1030         pgrp = task_pid(p);
1031         if (pgid != pid) {
1032                 struct task_struct *g;
1033
1034                 pgrp = find_vpid(pgid);
1035                 g = pid_task(pgrp, PIDTYPE_PGID);
1036                 if (!g || task_session(g) != task_session(group_leader))
1037                         goto out;
1038         }
1039
1040         err = security_task_setpgid(p, pgid);
1041         if (err)
1042                 goto out;
1043
1044         if (task_pgrp(p) != pgrp)
1045                 change_pid(p, PIDTYPE_PGID, pgrp);
1046
1047         err = 0;
1048 out:
1049         /* All paths lead to here, thus we are safe. -DaveM */
1050         write_unlock_irq(&tasklist_lock);
1051         rcu_read_unlock();
1052         return err;
1053 }
1054
1055 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1056 {
1057         struct task_struct *p;
1058         struct pid *grp;
1059         int retval;
1060
1061         rcu_read_lock();
1062         if (!pid)
1063                 grp = task_pgrp(current);
1064         else {
1065                 retval = -ESRCH;
1066                 p = find_task_by_vpid(pid);
1067                 if (!p)
1068                         goto out;
1069                 grp = task_pgrp(p);
1070                 if (!grp)
1071                         goto out;
1072
1073                 retval = security_task_getpgid(p);
1074                 if (retval)
1075                         goto out;
1076         }
1077         retval = pid_vnr(grp);
1078 out:
1079         rcu_read_unlock();
1080         return retval;
1081 }
1082
1083 #ifdef __ARCH_WANT_SYS_GETPGRP
1084
1085 SYSCALL_DEFINE0(getpgrp)
1086 {
1087         return sys_getpgid(0);
1088 }
1089
1090 #endif
1091
1092 SYSCALL_DEFINE1(getsid, pid_t, pid)
1093 {
1094         struct task_struct *p;
1095         struct pid *sid;
1096         int retval;
1097
1098         rcu_read_lock();
1099         if (!pid)
1100                 sid = task_session(current);
1101         else {
1102                 retval = -ESRCH;
1103                 p = find_task_by_vpid(pid);
1104                 if (!p)
1105                         goto out;
1106                 sid = task_session(p);
1107                 if (!sid)
1108                         goto out;
1109
1110                 retval = security_task_getsid(p);
1111                 if (retval)
1112                         goto out;
1113         }
1114         retval = pid_vnr(sid);
1115 out:
1116         rcu_read_unlock();
1117         return retval;
1118 }
1119
1120 SYSCALL_DEFINE0(setsid)
1121 {
1122         struct task_struct *group_leader = current->group_leader;
1123         struct pid *sid = task_pid(group_leader);
1124         pid_t session = pid_vnr(sid);
1125         int err = -EPERM;
1126
1127         write_lock_irq(&tasklist_lock);
1128         /* Fail if I am already a session leader */
1129         if (group_leader->signal->leader)
1130                 goto out;
1131
1132         /* Fail if a process group id already exists that equals the
1133          * proposed session id.
1134          */
1135         if (pid_task(sid, PIDTYPE_PGID))
1136                 goto out;
1137
1138         group_leader->signal->leader = 1;
1139         __set_special_pids(sid);
1140
1141         proc_clear_tty(group_leader);
1142
1143         err = session;
1144 out:
1145         write_unlock_irq(&tasklist_lock);
1146         if (err > 0) {
1147                 proc_sid_connector(group_leader);
1148                 sched_autogroup_create_attach(group_leader);
1149         }
1150         return err;
1151 }
1152
1153 DECLARE_RWSEM(uts_sem);
1154
1155 #ifdef COMPAT_UTS_MACHINE
1156 #define override_architecture(name) \
1157         (personality(current->personality) == PER_LINUX32 && \
1158          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1159                       sizeof(COMPAT_UTS_MACHINE)))
1160 #else
1161 #define override_architecture(name)     0
1162 #endif
1163
1164 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1165 {
1166         int errno = 0;
1167
1168         down_read(&uts_sem);
1169         if (copy_to_user(name, utsname(), sizeof *name))
1170                 errno = -EFAULT;
1171         up_read(&uts_sem);
1172
1173         if (!errno && override_architecture(name))
1174                 errno = -EFAULT;
1175         return errno;
1176 }
1177
1178 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1179 /*
1180  * Old cruft
1181  */
1182 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1183 {
1184         int error = 0;
1185
1186         if (!name)
1187                 return -EFAULT;
1188
1189         down_read(&uts_sem);
1190         if (copy_to_user(name, utsname(), sizeof(*name)))
1191                 error = -EFAULT;
1192         up_read(&uts_sem);
1193
1194         if (!error && override_architecture(name))
1195                 error = -EFAULT;
1196         return error;
1197 }
1198
1199 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1200 {
1201         int error;
1202
1203         if (!name)
1204                 return -EFAULT;
1205         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1206                 return -EFAULT;
1207
1208         down_read(&uts_sem);
1209         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1210                                __OLD_UTS_LEN);
1211         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1212         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1213                                 __OLD_UTS_LEN);
1214         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1215         error |= __copy_to_user(&name->release, &utsname()->release,
1216                                 __OLD_UTS_LEN);
1217         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1218         error |= __copy_to_user(&name->version, &utsname()->version,
1219                                 __OLD_UTS_LEN);
1220         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1221         error |= __copy_to_user(&name->machine, &utsname()->machine,
1222                                 __OLD_UTS_LEN);
1223         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1224         up_read(&uts_sem);
1225
1226         if (!error && override_architecture(name))
1227                 error = -EFAULT;
1228         return error ? -EFAULT : 0;
1229 }
1230 #endif
1231
1232 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1233 {
1234         int errno;
1235         char tmp[__NEW_UTS_LEN];
1236
1237         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1238                 return -EPERM;
1239
1240         if (len < 0 || len > __NEW_UTS_LEN)
1241                 return -EINVAL;
1242         down_write(&uts_sem);
1243         errno = -EFAULT;
1244         if (!copy_from_user(tmp, name, len)) {
1245                 struct new_utsname *u = utsname();
1246
1247                 memcpy(u->nodename, tmp, len);
1248                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1249                 errno = 0;
1250         }
1251         up_write(&uts_sem);
1252         return errno;
1253 }
1254
1255 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1256
1257 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1258 {
1259         int i, errno;
1260         struct new_utsname *u;
1261
1262         if (len < 0)
1263                 return -EINVAL;
1264         down_read(&uts_sem);
1265         u = utsname();
1266         i = 1 + strlen(u->nodename);
1267         if (i > len)
1268                 i = len;
1269         errno = 0;
1270         if (copy_to_user(name, u->nodename, i))
1271                 errno = -EFAULT;
1272         up_read(&uts_sem);
1273         return errno;
1274 }
1275
1276 #endif
1277
1278 /*
1279  * Only setdomainname; getdomainname can be implemented by calling
1280  * uname()
1281  */
1282 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1283 {
1284         int errno;
1285         char tmp[__NEW_UTS_LEN];
1286
1287         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1288                 return -EPERM;
1289         if (len < 0 || len > __NEW_UTS_LEN)
1290                 return -EINVAL;
1291
1292         down_write(&uts_sem);
1293         errno = -EFAULT;
1294         if (!copy_from_user(tmp, name, len)) {
1295                 struct new_utsname *u = utsname();
1296
1297                 memcpy(u->domainname, tmp, len);
1298                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1299                 errno = 0;
1300         }
1301         up_write(&uts_sem);
1302         return errno;
1303 }
1304
1305 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1306 {
1307         struct rlimit value;
1308         int ret;
1309
1310         ret = do_prlimit(current, resource, NULL, &value);
1311         if (!ret)
1312                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1313
1314         return ret;
1315 }
1316
1317 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1318
1319 /*
1320  *      Back compatibility for getrlimit. Needed for some apps.
1321  */
1322
1323 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1324                 struct rlimit __user *, rlim)
1325 {
1326         struct rlimit x;
1327         if (resource >= RLIM_NLIMITS)
1328                 return -EINVAL;
1329
1330         task_lock(current->group_leader);
1331         x = current->signal->rlim[resource];
1332         task_unlock(current->group_leader);
1333         if (x.rlim_cur > 0x7FFFFFFF)
1334                 x.rlim_cur = 0x7FFFFFFF;
1335         if (x.rlim_max > 0x7FFFFFFF)
1336                 x.rlim_max = 0x7FFFFFFF;
1337         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1338 }
1339
1340 #endif
1341
1342 static inline bool rlim64_is_infinity(__u64 rlim64)
1343 {
1344 #if BITS_PER_LONG < 64
1345         return rlim64 >= ULONG_MAX;
1346 #else
1347         return rlim64 == RLIM64_INFINITY;
1348 #endif
1349 }
1350
1351 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1352 {
1353         if (rlim->rlim_cur == RLIM_INFINITY)
1354                 rlim64->rlim_cur = RLIM64_INFINITY;
1355         else
1356                 rlim64->rlim_cur = rlim->rlim_cur;
1357         if (rlim->rlim_max == RLIM_INFINITY)
1358                 rlim64->rlim_max = RLIM64_INFINITY;
1359         else
1360                 rlim64->rlim_max = rlim->rlim_max;
1361 }
1362
1363 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1364 {
1365         if (rlim64_is_infinity(rlim64->rlim_cur))
1366                 rlim->rlim_cur = RLIM_INFINITY;
1367         else
1368                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1369         if (rlim64_is_infinity(rlim64->rlim_max))
1370                 rlim->rlim_max = RLIM_INFINITY;
1371         else
1372                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1373 }
1374
1375 /* make sure you are allowed to change @tsk limits before calling this */
1376 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1377                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1378 {
1379         struct rlimit *rlim;
1380         int retval = 0;
1381
1382         if (resource >= RLIM_NLIMITS)
1383                 return -EINVAL;
1384         if (new_rlim) {
1385                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1386                         return -EINVAL;
1387                 if (resource == RLIMIT_NOFILE &&
1388                                 new_rlim->rlim_max > sysctl_nr_open)
1389                         return -EPERM;
1390         }
1391
1392         /* protect tsk->signal and tsk->sighand from disappearing */
1393         read_lock(&tasklist_lock);
1394         if (!tsk->sighand) {
1395                 retval = -ESRCH;
1396                 goto out;
1397         }
1398
1399         rlim = tsk->signal->rlim + resource;
1400         task_lock(tsk->group_leader);
1401         if (new_rlim) {
1402                 /* Keep the capable check against init_user_ns until
1403                    cgroups can contain all limits */
1404                 if (new_rlim->rlim_max > rlim->rlim_max &&
1405                                 !capable(CAP_SYS_RESOURCE))
1406                         retval = -EPERM;
1407                 if (!retval)
1408                         retval = security_task_setrlimit(tsk->group_leader,
1409                                         resource, new_rlim);
1410                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1411                         /*
1412                          * The caller is asking for an immediate RLIMIT_CPU
1413                          * expiry.  But we use the zero value to mean "it was
1414                          * never set".  So let's cheat and make it one second
1415                          * instead
1416                          */
1417                         new_rlim->rlim_cur = 1;
1418                 }
1419         }
1420         if (!retval) {
1421                 if (old_rlim)
1422                         *old_rlim = *rlim;
1423                 if (new_rlim)
1424                         *rlim = *new_rlim;
1425         }
1426         task_unlock(tsk->group_leader);
1427
1428         /*
1429          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1430          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1431          * very long-standing error, and fixing it now risks breakage of
1432          * applications, so we live with it
1433          */
1434          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1435                          new_rlim->rlim_cur != RLIM_INFINITY)
1436                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1437 out:
1438         read_unlock(&tasklist_lock);
1439         return retval;
1440 }
1441
1442 /* rcu lock must be held */
1443 static int check_prlimit_permission(struct task_struct *task)
1444 {
1445         const struct cred *cred = current_cred(), *tcred;
1446
1447         if (current == task)
1448                 return 0;
1449
1450         tcred = __task_cred(task);
1451         if (cred->user->user_ns == tcred->user->user_ns &&
1452             (cred->uid == tcred->euid &&
1453              cred->uid == tcred->suid &&
1454              cred->uid == tcred->uid  &&
1455              cred->gid == tcred->egid &&
1456              cred->gid == tcred->sgid &&
1457              cred->gid == tcred->gid))
1458                 return 0;
1459         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1460                 return 0;
1461
1462         return -EPERM;
1463 }
1464
1465 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1466                 const struct rlimit64 __user *, new_rlim,
1467                 struct rlimit64 __user *, old_rlim)
1468 {
1469         struct rlimit64 old64, new64;
1470         struct rlimit old, new;
1471         struct task_struct *tsk;
1472         int ret;
1473
1474         if (new_rlim) {
1475                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1476                         return -EFAULT;
1477                 rlim64_to_rlim(&new64, &new);
1478         }
1479
1480         rcu_read_lock();
1481         tsk = pid ? find_task_by_vpid(pid) : current;
1482         if (!tsk) {
1483                 rcu_read_unlock();
1484                 return -ESRCH;
1485         }
1486         ret = check_prlimit_permission(tsk);
1487         if (ret) {
1488                 rcu_read_unlock();
1489                 return ret;
1490         }
1491         get_task_struct(tsk);
1492         rcu_read_unlock();
1493
1494         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1495                         old_rlim ? &old : NULL);
1496
1497         if (!ret && old_rlim) {
1498                 rlim_to_rlim64(&old, &old64);
1499                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1500                         ret = -EFAULT;
1501         }
1502
1503         put_task_struct(tsk);
1504         return ret;
1505 }
1506
1507 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1508 {
1509         struct rlimit new_rlim;
1510
1511         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1512                 return -EFAULT;
1513         return do_prlimit(current, resource, &new_rlim, NULL);
1514 }
1515
1516 /*
1517  * It would make sense to put struct rusage in the task_struct,
1518  * except that would make the task_struct be *really big*.  After
1519  * task_struct gets moved into malloc'ed memory, it would
1520  * make sense to do this.  It will make moving the rest of the information
1521  * a lot simpler!  (Which we're not doing right now because we're not
1522  * measuring them yet).
1523  *
1524  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1525  * races with threads incrementing their own counters.  But since word
1526  * reads are atomic, we either get new values or old values and we don't
1527  * care which for the sums.  We always take the siglock to protect reading
1528  * the c* fields from p->signal from races with exit.c updating those
1529  * fields when reaping, so a sample either gets all the additions of a
1530  * given child after it's reaped, or none so this sample is before reaping.
1531  *
1532  * Locking:
1533  * We need to take the siglock for CHILDEREN, SELF and BOTH
1534  * for  the cases current multithreaded, non-current single threaded
1535  * non-current multithreaded.  Thread traversal is now safe with
1536  * the siglock held.
1537  * Strictly speaking, we donot need to take the siglock if we are current and
1538  * single threaded,  as no one else can take our signal_struct away, no one
1539  * else can  reap the  children to update signal->c* counters, and no one else
1540  * can race with the signal-> fields. If we do not take any lock, the
1541  * signal-> fields could be read out of order while another thread was just
1542  * exiting. So we should  place a read memory barrier when we avoid the lock.
1543  * On the writer side,  write memory barrier is implied in  __exit_signal
1544  * as __exit_signal releases  the siglock spinlock after updating the signal->
1545  * fields. But we don't do this yet to keep things simple.
1546  *
1547  */
1548
1549 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1550 {
1551         r->ru_nvcsw += t->nvcsw;
1552         r->ru_nivcsw += t->nivcsw;
1553         r->ru_minflt += t->min_flt;
1554         r->ru_majflt += t->maj_flt;
1555         r->ru_inblock += task_io_get_inblock(t);
1556         r->ru_oublock += task_io_get_oublock(t);
1557 }
1558
1559 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1560 {
1561         struct task_struct *t;
1562         unsigned long flags;
1563         cputime_t tgutime, tgstime, utime, stime;
1564         unsigned long maxrss = 0;
1565
1566         memset((char *) r, 0, sizeof *r);
1567         utime = stime = cputime_zero;
1568
1569         if (who == RUSAGE_THREAD) {
1570                 task_times(current, &utime, &stime);
1571                 accumulate_thread_rusage(p, r);
1572                 maxrss = p->signal->maxrss;
1573                 goto out;
1574         }
1575
1576         if (!lock_task_sighand(p, &flags))
1577                 return;
1578
1579         switch (who) {
1580                 case RUSAGE_BOTH:
1581                 case RUSAGE_CHILDREN:
1582                         utime = p->signal->cutime;
1583                         stime = p->signal->cstime;
1584                         r->ru_nvcsw = p->signal->cnvcsw;
1585                         r->ru_nivcsw = p->signal->cnivcsw;
1586                         r->ru_minflt = p->signal->cmin_flt;
1587                         r->ru_majflt = p->signal->cmaj_flt;
1588                         r->ru_inblock = p->signal->cinblock;
1589                         r->ru_oublock = p->signal->coublock;
1590                         maxrss = p->signal->cmaxrss;
1591
1592                         if (who == RUSAGE_CHILDREN)
1593                                 break;
1594
1595                 case RUSAGE_SELF:
1596                         thread_group_times(p, &tgutime, &tgstime);
1597                         utime = cputime_add(utime, tgutime);
1598                         stime = cputime_add(stime, tgstime);
1599                         r->ru_nvcsw += p->signal->nvcsw;
1600                         r->ru_nivcsw += p->signal->nivcsw;
1601                         r->ru_minflt += p->signal->min_flt;
1602                         r->ru_majflt += p->signal->maj_flt;
1603                         r->ru_inblock += p->signal->inblock;
1604                         r->ru_oublock += p->signal->oublock;
1605                         if (maxrss < p->signal->maxrss)
1606                                 maxrss = p->signal->maxrss;
1607                         t = p;
1608                         do {
1609                                 accumulate_thread_rusage(t, r);
1610                                 t = next_thread(t);
1611                         } while (t != p);
1612                         break;
1613
1614                 default:
1615                         BUG();
1616         }
1617         unlock_task_sighand(p, &flags);
1618
1619 out:
1620         cputime_to_timeval(utime, &r->ru_utime);
1621         cputime_to_timeval(stime, &r->ru_stime);
1622
1623         if (who != RUSAGE_CHILDREN) {
1624                 struct mm_struct *mm = get_task_mm(p);
1625                 if (mm) {
1626                         setmax_mm_hiwater_rss(&maxrss, mm);
1627                         mmput(mm);
1628                 }
1629         }
1630         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1631 }
1632
1633 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1634 {
1635         struct rusage r;
1636         k_getrusage(p, who, &r);
1637         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1638 }
1639
1640 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1641 {
1642         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1643             who != RUSAGE_THREAD)
1644                 return -EINVAL;
1645         return getrusage(current, who, ru);
1646 }
1647
1648 SYSCALL_DEFINE1(umask, int, mask)
1649 {
1650         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1651         return mask;
1652 }
1653
1654 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1655                 unsigned long, arg4, unsigned long, arg5)
1656 {
1657         struct task_struct *me = current;
1658         unsigned char comm[sizeof(me->comm)];
1659         long error;
1660
1661         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1662         if (error != -ENOSYS)
1663                 return error;
1664
1665         error = 0;
1666         switch (option) {
1667                 case PR_SET_PDEATHSIG:
1668                         if (!valid_signal(arg2)) {
1669                                 error = -EINVAL;
1670                                 break;
1671                         }
1672                         me->pdeath_signal = arg2;
1673                         error = 0;
1674                         break;
1675                 case PR_GET_PDEATHSIG:
1676                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1677                         break;
1678                 case PR_GET_DUMPABLE:
1679                         error = get_dumpable(me->mm);
1680                         break;
1681                 case PR_SET_DUMPABLE:
1682                         if (arg2 < 0 || arg2 > 1) {
1683                                 error = -EINVAL;
1684                                 break;
1685                         }
1686                         set_dumpable(me->mm, arg2);
1687                         error = 0;
1688                         break;
1689
1690                 case PR_SET_UNALIGN:
1691                         error = SET_UNALIGN_CTL(me, arg2);
1692                         break;
1693                 case PR_GET_UNALIGN:
1694                         error = GET_UNALIGN_CTL(me, arg2);
1695                         break;
1696                 case PR_SET_FPEMU:
1697                         error = SET_FPEMU_CTL(me, arg2);
1698                         break;
1699                 case PR_GET_FPEMU:
1700                         error = GET_FPEMU_CTL(me, arg2);
1701                         break;
1702                 case PR_SET_FPEXC:
1703                         error = SET_FPEXC_CTL(me, arg2);
1704                         break;
1705                 case PR_GET_FPEXC:
1706                         error = GET_FPEXC_CTL(me, arg2);
1707                         break;
1708                 case PR_GET_TIMING:
1709                         error = PR_TIMING_STATISTICAL;
1710                         break;
1711                 case PR_SET_TIMING:
1712                         if (arg2 != PR_TIMING_STATISTICAL)
1713                                 error = -EINVAL;
1714                         else
1715                                 error = 0;
1716                         break;
1717
1718                 case PR_SET_NAME:
1719                         comm[sizeof(me->comm)-1] = 0;
1720                         if (strncpy_from_user(comm, (char __user *)arg2,
1721                                               sizeof(me->comm) - 1) < 0)
1722                                 return -EFAULT;
1723                         set_task_comm(me, comm);
1724                         return 0;
1725                 case PR_GET_NAME:
1726                         get_task_comm(comm, me);
1727                         if (copy_to_user((char __user *)arg2, comm,
1728                                          sizeof(comm)))
1729                                 return -EFAULT;
1730                         return 0;
1731                 case PR_GET_ENDIAN:
1732                         error = GET_ENDIAN(me, arg2);
1733                         break;
1734                 case PR_SET_ENDIAN:
1735                         error = SET_ENDIAN(me, arg2);
1736                         break;
1737
1738                 case PR_GET_SECCOMP:
1739                         error = prctl_get_seccomp();
1740                         break;
1741                 case PR_SET_SECCOMP:
1742                         error = prctl_set_seccomp(arg2);
1743                         break;
1744                 case PR_GET_TSC:
1745                         error = GET_TSC_CTL(arg2);
1746                         break;
1747                 case PR_SET_TSC:
1748                         error = SET_TSC_CTL(arg2);
1749                         break;
1750                 case PR_TASK_PERF_EVENTS_DISABLE:
1751                         error = perf_event_task_disable();
1752                         break;
1753                 case PR_TASK_PERF_EVENTS_ENABLE:
1754                         error = perf_event_task_enable();
1755                         break;
1756                 case PR_GET_TIMERSLACK:
1757                         error = current->timer_slack_ns;
1758                         break;
1759                 case PR_SET_TIMERSLACK:
1760                         if (arg2 <= 0)
1761                                 current->timer_slack_ns =
1762                                         current->default_timer_slack_ns;
1763                         else
1764                                 current->timer_slack_ns = arg2;
1765                         error = 0;
1766                         break;
1767                 case PR_MCE_KILL:
1768                         if (arg4 | arg5)
1769                                 return -EINVAL;
1770                         switch (arg2) {
1771                         case PR_MCE_KILL_CLEAR:
1772                                 if (arg3 != 0)
1773                                         return -EINVAL;
1774                                 current->flags &= ~PF_MCE_PROCESS;
1775                                 break;
1776                         case PR_MCE_KILL_SET:
1777                                 current->flags |= PF_MCE_PROCESS;
1778                                 if (arg3 == PR_MCE_KILL_EARLY)
1779                                         current->flags |= PF_MCE_EARLY;
1780                                 else if (arg3 == PR_MCE_KILL_LATE)
1781                                         current->flags &= ~PF_MCE_EARLY;
1782                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1783                                         current->flags &=
1784                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1785                                 else
1786                                         return -EINVAL;
1787                                 break;
1788                         default:
1789                                 return -EINVAL;
1790                         }
1791                         error = 0;
1792                         break;
1793                 case PR_MCE_KILL_GET:
1794                         if (arg2 | arg3 | arg4 | arg5)
1795                                 return -EINVAL;
1796                         if (current->flags & PF_MCE_PROCESS)
1797                                 error = (current->flags & PF_MCE_EARLY) ?
1798                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1799                         else
1800                                 error = PR_MCE_KILL_DEFAULT;
1801                         break;
1802                 default:
1803                         error = -EINVAL;
1804                         break;
1805         }
1806         return error;
1807 }
1808
1809 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1810                 struct getcpu_cache __user *, unused)
1811 {
1812         int err = 0;
1813         int cpu = raw_smp_processor_id();
1814         if (cpup)
1815                 err |= put_user(cpu, cpup);
1816         if (nodep)
1817                 err |= put_user(cpu_to_node(cpu), nodep);
1818         return err ? -EFAULT : 0;
1819 }
1820
1821 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1822
1823 static void argv_cleanup(struct subprocess_info *info)
1824 {
1825         argv_free(info->argv);
1826 }
1827
1828 /**
1829  * orderly_poweroff - Trigger an orderly system poweroff
1830  * @force: force poweroff if command execution fails
1831  *
1832  * This may be called from any context to trigger a system shutdown.
1833  * If the orderly shutdown fails, it will force an immediate shutdown.
1834  */
1835 int orderly_poweroff(bool force)
1836 {
1837         int argc;
1838         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1839         static char *envp[] = {
1840                 "HOME=/",
1841                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1842                 NULL
1843         };
1844         int ret = -ENOMEM;
1845         struct subprocess_info *info;
1846
1847         if (argv == NULL) {
1848                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1849                        __func__, poweroff_cmd);
1850                 goto out;
1851         }
1852
1853         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1854         if (info == NULL) {
1855                 argv_free(argv);
1856                 goto out;
1857         }
1858
1859         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1860
1861         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1862
1863   out:
1864         if (ret && force) {
1865                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1866                        "forcing the issue\n");
1867
1868                 /* I guess this should try to kick off some daemon to
1869                    sync and poweroff asap.  Or not even bother syncing
1870                    if we're doing an emergency shutdown? */
1871                 emergency_sync();
1872                 kernel_power_off();
1873         }
1874
1875         return ret;
1876 }
1877 EXPORT_SYMBOL_GPL(orderly_poweroff);