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 #include <linux/version.h>
  41 #include <linux/ctype.h>
  42
  43 #include <linux/compat.h>
  44 #include <linux/syscalls.h>
  45 #include <linux/kprobes.h>
  46 #include <linux/user_namespace.h>
  47
  48 #include <linux/kmsg_dump.h>
  49 /* Move somewhere else to avoid recompiling? */
  50 #include <generated/utsrelease.h>
  51
  52 #include <asm/uaccess.h>
  53 #include <asm/io.h>
  54 #include <asm/unistd.h>
  55
  56 #ifndef SET_UNALIGN_CTL
  57 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  58 #endif
  59 #ifndef GET_UNALIGN_CTL
  60 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  61 #endif
  62 #ifndef SET_FPEMU_CTL
  63 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  64 #endif
  65 #ifndef GET_FPEMU_CTL
  66 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  67 #endif
  68 #ifndef SET_FPEXC_CTL
  69 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  70 #endif
  71 #ifndef GET_FPEXC_CTL
  72 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  73 #endif
  74 #ifndef GET_ENDIAN
  75 # define GET_ENDIAN(a,b)        (-EINVAL)
  76 #endif
  77 #ifndef SET_ENDIAN
  78 # define SET_ENDIAN(a,b)        (-EINVAL)
  79 #endif
  80 #ifndef GET_TSC_CTL
  81 # define GET_TSC_CTL(a)         (-EINVAL)
  82 #endif
  83 #ifndef SET_TSC_CTL
  84 # define SET_TSC_CTL(a)         (-EINVAL)
  85 #endif
  86
  87 /*
  88  * this is where the system-wide overflow UID and GID are defined, for
  89  * architectures that now have 32-bit UID/GID but didn't in the past
  90  */
  91
  92 int overflowuid = DEFAULT_OVERFLOWUID;
  93 int overflowgid = DEFAULT_OVERFLOWGID;
  94
  95 #ifdef CONFIG_UID16
  96 EXPORT_SYMBOL(overflowuid);
  97 EXPORT_SYMBOL(overflowgid);
  98 #endif
  99
 100 /*
 101  * the same as above, but for filesystems which can only store a 16-bit
 102  * UID and GID. as such, this is needed on all architectures
 103  */
 104
 105 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 106 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 107
 108 EXPORT_SYMBOL(fs_overflowuid);
 109 EXPORT_SYMBOL(fs_overflowgid);
 110
 111 /*
 112  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 113  */
 114
 115 int C_A_D = 1;
 116 struct pid *cad_pid;
 117 EXPORT_SYMBOL(cad_pid);
 118
 119 /*
 120  * If set, this is used for preparing the system to power off.
 121  */
 122
 123 void (*pm_power_off_prepare)(void);
 124
 125 /*
 126  * Returns true if current's euid is same as p's uid or euid,
 127  * or has CAP_SYS_NICE to p's user_ns.
 128  *
 129  * Called with rcu_read_lock, creds are safe
 130  */
 131 static bool set_one_prio_perm(struct task_struct *p)
 132 {
 133         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 134
 135         if (pcred->user->user_ns == cred->user->user_ns &&
 136             (pcred->uid  == cred->euid ||
 137              pcred->euid == cred->euid))
 138                 return true;
 139         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 140                 return true;
 141         return false;
 142 }
 143
 144 /*
 145  * set the priority of a task
 146  * - the caller must hold the RCU read lock
 147  */
 148 static int set_one_prio(struct task_struct *p, int niceval, int error)
 149 {
 150         int no_nice;
 151
 152         if (!set_one_prio_perm(p)) {
 153                 error = -EPERM;
 154                 goto out;
 155         }
 156         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 157                 error = -EACCES;
 158                 goto out;
 159         }
 160         no_nice = security_task_setnice(p, niceval);
 161         if (no_nice) {
 162                 error = no_nice;
 163                 goto out;
 164         }
 165         if (error == -ESRCH)
 166                 error = 0;
 167         set_user_nice(p, niceval);
 168 out:
 169         return error;
 170 }
 171
 172 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 173 {
 174         struct task_struct *g, *p;
 175         struct user_struct *user;
 176         const struct cred *cred = current_cred();
 177         int error = -EINVAL;
 178         struct pid *pgrp;
 179
 180         if (which > PRIO_USER || which < PRIO_PROCESS)
 181                 goto out;
 182
 183         /* normalize: avoid signed division (rounding problems) */
 184         error = -ESRCH;
 185         if (niceval < -20)
 186                 niceval = -20;
 187         if (niceval > 19)
 188                 niceval = 19;
 189
 190         rcu_read_lock();
 191         read_lock(&tasklist_lock);
 192         switch (which) {
 193                 case PRIO_PROCESS:
 194                         if (who)
 195                                 p = find_task_by_vpid(who);
 196                         else
 197                                 p = current;
 198                         if (p)
 199                                 error = set_one_prio(p, niceval, error);
 200                         break;
 201                 case PRIO_PGRP:
 202                         if (who)
 203                                 pgrp = find_vpid(who);
 204                         else
 205                                 pgrp = task_pgrp(current);
 206                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 207                                 error = set_one_prio(p, niceval, error);
 208                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 209                         break;
 210                 case PRIO_USER:
 211                         user = (struct user_struct *) cred->user;
 212                         if (!who)
 213                                 who = cred->uid;
 214                         else if ((who != cred->uid) &&
 215                                  !(user = find_user(who)))
 216                                 goto out_unlock;        /* No processes for this user */
 217
 218                         do_each_thread(g, p) {
 219                                 if (__task_cred(p)->uid == who)
 220                                         error = set_one_prio(p, niceval, error);
 221                         } while_each_thread(g, p);
 222                         if (who != cred->uid)
 223                                 free_uid(user);         /* For find_user() */
 224                         break;
 225         }
 226 out_unlock:
 227         read_unlock(&tasklist_lock);
 228         rcu_read_unlock();
 229 out:
 230         return error;
 231 }
 232
 233 /*
 234  * Ugh. To avoid negative return values, "getpriority()" will
 235  * not return the normal nice-value, but a negated value that
 236  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 237  * to stay compatible.
 238  */
 239 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 240 {
 241         struct task_struct *g, *p;
 242         struct user_struct *user;
 243         const struct cred *cred = current_cred();
 244         long niceval, retval = -ESRCH;
 245         struct pid *pgrp;
 246
 247         if (which > PRIO_USER || which < PRIO_PROCESS)
 248                 return -EINVAL;
 249
 250         rcu_read_lock();
 251         read_lock(&tasklist_lock);
 252         switch (which) {
 253                 case PRIO_PROCESS:
 254                         if (who)
 255                                 p = find_task_by_vpid(who);
 256                         else
 257                                 p = current;
 258                         if (p) {
 259                                 niceval = 20 - task_nice(p);
 260                                 if (niceval > retval)
 261                                         retval = niceval;
 262                         }
 263                         break;
 264                 case PRIO_PGRP:
 265                         if (who)
 266                                 pgrp = find_vpid(who);
 267                         else
 268                                 pgrp = task_pgrp(current);
 269                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 270                                 niceval = 20 - task_nice(p);
 271                                 if (niceval > retval)
 272                                         retval = niceval;
 273                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 274                         break;
 275                 case PRIO_USER:
 276                         user = (struct user_struct *) cred->user;
 277                         if (!who)
 278                                 who = cred->uid;
 279                         else if ((who != cred->uid) &&
 280                                  !(user = find_user(who)))
 281                                 goto out_unlock;        /* No processes for this user */
 282
 283                         do_each_thread(g, p) {
 284                                 if (__task_cred(p)->uid == who) {
 285                                         niceval = 20 - task_nice(p);
 286                                         if (niceval > retval)
 287                                                 retval = niceval;
 288                                 }
 289                         } while_each_thread(g, p);
 290                         if (who != cred->uid)
 291                                 free_uid(user);         /* for find_user() */
 292                         break;
 293         }
 294 out_unlock:
 295         read_unlock(&tasklist_lock);
 296         rcu_read_unlock();
 297
 298         return retval;
 299 }
 300
 301 /**
 302  *      emergency_restart - reboot the system
 303  *
 304  *      Without shutting down any hardware or taking any locks
 305  *      reboot the system.  This is called when we know we are in
 306  *      trouble so this is our best effort to reboot.  This is
 307  *      safe to call in interrupt context.
 308  */
 309 void emergency_restart(void)
 310 {
 311         kmsg_dump(KMSG_DUMP_EMERG);
 312         machine_emergency_restart();
 313 }
 314 EXPORT_SYMBOL_GPL(emergency_restart);
 315
 316 void kernel_restart_prepare(char *cmd)
 317 {
 318         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 319         system_state = SYSTEM_RESTART;
 320         usermodehelper_disable();
 321         device_shutdown();
 322         syscore_shutdown();
 323 }
 324
 325 /**
 326  *      register_reboot_notifier - Register function to be called at reboot time
 327  *      @nb: Info about notifier function to be called
 328  *
 329  *      Registers a function with the list of functions
 330  *      to be called at reboot time.
 331  *
 332  *      Currently always returns zero, as blocking_notifier_chain_register()
 333  *      always returns zero.
 334  */
 335 int register_reboot_notifier(struct notifier_block *nb)
 336 {
 337         return blocking_notifier_chain_register(&reboot_notifier_list, nb);
 338 }
 339 EXPORT_SYMBOL(register_reboot_notifier);
 340
 341 /**
 342  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 343  *      @nb: Hook to be unregistered
 344  *
 345  *      Unregisters a previously registered reboot
 346  *      notifier function.
 347  *
 348  *      Returns zero on success, or %-ENOENT on failure.
 349  */
 350 int unregister_reboot_notifier(struct notifier_block *nb)
 351 {
 352         return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
 353 }
 354 EXPORT_SYMBOL(unregister_reboot_notifier);
 355
 356 /**
 357  *      kernel_restart - reboot the system
 358  *      @cmd: pointer to buffer containing command to execute for restart
 359  *              or %NULL
 360  *
 361  *      Shutdown everything and perform a clean reboot.
 362  *      This is not safe to call in interrupt context.
 363  */
 364 void kernel_restart(char *cmd)
 365 {
 366         kernel_restart_prepare(cmd);
 367         if (!cmd)
 368                 printk(KERN_EMERG "Restarting system.\n");
 369         else
 370                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 371         kmsg_dump(KMSG_DUMP_RESTART);
 372         machine_restart(cmd);
 373 }
 374 EXPORT_SYMBOL_GPL(kernel_restart);
 375
 376 static void kernel_shutdown_prepare(enum system_states state)
 377 {
 378         blocking_notifier_call_chain(&reboot_notifier_list,
 379                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 380         system_state = state;
 381         usermodehelper_disable();
 382         device_shutdown();
 383 }
 384 /**
 385  *      kernel_halt - halt the system
 386  *
 387  *      Shutdown everything and perform a clean system halt.
 388  */
 389 void kernel_halt(void)
 390 {
 391         kernel_shutdown_prepare(SYSTEM_HALT);
 392         syscore_shutdown();
 393         printk(KERN_EMERG "System halted.\n");
 394         kmsg_dump(KMSG_DUMP_HALT);
 395         machine_halt();
 396 }
 397
 398 EXPORT_SYMBOL_GPL(kernel_halt);
 399
 400 /**
 401  *      kernel_power_off - power_off the system
 402  *
 403  *      Shutdown everything and perform a clean system power_off.
 404  */
 405 void kernel_power_off(void)
 406 {
 407         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 408         if (pm_power_off_prepare)
 409                 pm_power_off_prepare();
 410         disable_nonboot_cpus();
 411         syscore_shutdown();
 412         printk(KERN_EMERG "Power down.\n");
 413         kmsg_dump(KMSG_DUMP_POWEROFF);
 414         machine_power_off();
 415 }
 416 EXPORT_SYMBOL_GPL(kernel_power_off);
 417
 418 static DEFINE_MUTEX(reboot_mutex);
 419
 420 /*
 421  * Reboot system call: for obvious reasons only root may call it,
 422  * and even root needs to set up some magic numbers in the registers
 423  * so that some mistake won't make this reboot the whole machine.
 424  * You can also set the meaning of the ctrl-alt-del-key here.
 425  *
 426  * reboot doesn't sync: do that yourself before calling this.
 427  */
 428 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 429                 void __user *, arg)
 430 {
 431         char buffer[256];
 432         int ret = 0;
 433
 434         /* We only trust the superuser with rebooting the system. */
 435         if (!capable(CAP_SYS_BOOT))
 436                 return -EPERM;
 437
 438         /* For safety, we require "magic" arguments. */
 439         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 440             (magic2 != LINUX_REBOOT_MAGIC2 &&
 441                         magic2 != LINUX_REBOOT_MAGIC2A &&
 442                         magic2 != LINUX_REBOOT_MAGIC2B &&
 443                         magic2 != LINUX_REBOOT_MAGIC2C))
 444                 return -EINVAL;
 445
 446         /* Instead of trying to make the power_off code look like
 447          * halt when pm_power_off is not set do it the easy way.
 448          */
 449         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 450                 cmd = LINUX_REBOOT_CMD_HALT;
 451
 452         mutex_lock(&reboot_mutex);
 453         switch (cmd) {
 454         case LINUX_REBOOT_CMD_RESTART:
 455                 kernel_restart(NULL);
 456                 break;
 457
 458         case LINUX_REBOOT_CMD_CAD_ON:
 459                 C_A_D = 1;
 460                 break;
 461
 462         case LINUX_REBOOT_CMD_CAD_OFF:
 463                 C_A_D = 0;
 464                 break;
 465
 466         case LINUX_REBOOT_CMD_HALT:
 467                 kernel_halt();
 468                 do_exit(0);
 469                 panic("cannot halt");
 470
 471         case LINUX_REBOOT_CMD_POWER_OFF:
 472                 kernel_power_off();
 473                 do_exit(0);
 474                 break;
 475
 476         case LINUX_REBOOT_CMD_RESTART2:
 477                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 478                         ret = -EFAULT;
 479                         break;
 480                 }
 481                 buffer[sizeof(buffer) - 1] = '\0';
 482
 483                 kernel_restart(buffer);
 484                 break;
 485
 486 #ifdef CONFIG_KEXEC
 487         case LINUX_REBOOT_CMD_KEXEC:
 488                 ret = kernel_kexec();
 489                 break;
 490 #endif
 491
 492 #ifdef CONFIG_HIBERNATION
 493         case LINUX_REBOOT_CMD_SW_SUSPEND:
 494                 ret = hibernate();
 495                 break;
 496 #endif
 497
 498         default:
 499                 ret = -EINVAL;
 500                 break;
 501         }
 502         mutex_unlock(&reboot_mutex);
 503         return ret;
 504 }
 505
 506 static void deferred_cad(struct work_struct *dummy)
 507 {
 508         kernel_restart(NULL);
 509 }
 510
 511 /*
 512  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 513  * As it's called within an interrupt, it may NOT sync: the only choice
 514  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 515  */
 516 void ctrl_alt_del(void)
 517 {
 518         static DECLARE_WORK(cad_work, deferred_cad);
 519
 520         if (C_A_D)
 521                 schedule_work(&cad_work);
 522         else
 523                 kill_cad_pid(SIGINT, 1);
 524 }
 525
 526 /*
 527  * Unprivileged users may change the real gid to the effective gid
 528  * or vice versa.  (BSD-style)
 529  *
 530  * If you set the real gid at all, or set the effective gid to a value not
 531  * equal to the real gid, then the saved gid is set to the new effective gid.
 532  *
 533  * This makes it possible for a setgid program to completely drop its
 534  * privileges, which is often a useful assertion to make when you are doing
 535  * a security audit over a program.
 536  *
 537  * The general idea is that a program which uses just setregid() will be
 538  * 100% compatible with BSD.  A program which uses just setgid() will be
 539  * 100% compatible with POSIX with saved IDs.
 540  *
 541  * SMP: There are not races, the GIDs are checked only by filesystem
 542  *      operations (as far as semantic preservation is concerned).
 543  */
 544 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 545 {
 546         const struct cred *old;
 547         struct cred *new;
 548         int retval;
 549
 550         new = prepare_creds();
 551         if (!new)
 552                 return -ENOMEM;
 553         old = current_cred();
 554
 555         retval = -EPERM;
 556         if (rgid != (gid_t) -1) {
 557                 if (old->gid == rgid ||
 558                     old->egid == rgid ||
 559                     nsown_capable(CAP_SETGID))
 560                         new->gid = rgid;
 561                 else
 562                         goto error;
 563         }
 564         if (egid != (gid_t) -1) {
 565                 if (old->gid == egid ||
 566                     old->egid == egid ||
 567                     old->sgid == egid ||
 568                     nsown_capable(CAP_SETGID))
 569                         new->egid = egid;
 570                 else
 571                         goto error;
 572         }
 573
 574         if (rgid != (gid_t) -1 ||
 575             (egid != (gid_t) -1 && egid != old->gid))
 576                 new->sgid = new->egid;
 577         new->fsgid = new->egid;
 578
 579         return commit_creds(new);
 580
 581 error:
 582         abort_creds(new);
 583         return retval;
 584 }
 585
 586 /*
 587  * setgid() is implemented like SysV w/ SAVED_IDS
 588  *
 589  * SMP: Same implicit races as above.
 590  */
 591 SYSCALL_DEFINE1(setgid, gid_t, gid)
 592 {
 593         const struct cred *old;
 594         struct cred *new;
 595         int retval;
 596
 597         new = prepare_creds();
 598         if (!new)
 599                 return -ENOMEM;
 600         old = current_cred();
 601
 602         retval = -EPERM;
 603         if (nsown_capable(CAP_SETGID))
 604                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 605         else if (gid == old->gid || gid == old->sgid)
 606                 new->egid = new->fsgid = gid;
 607         else
 608                 goto error;
 609
 610         return commit_creds(new);
 611
 612 error:
 613         abort_creds(new);
 614         return retval;
 615 }
 616
 617 /*
 618  * change the user struct in a credentials set to match the new UID
 619  */
 620 static int set_user(struct cred *new)
 621 {
 622         struct user_struct *new_user;
 623
 624         new_user = alloc_uid(current_user_ns(), new->uid);
 625         if (!new_user)
 626                 return -EAGAIN;
 627
 628         /*
 629          * We don't fail in case of NPROC limit excess here because too many
 630          * poorly written programs don't check set*uid() return code, assuming
 631          * it never fails if called by root.  We may still enforce NPROC limit
 632          * for programs doing set*uid()+execve() by harmlessly deferring the
 633          * failure to the execve() stage.
 634          */
 635         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 636                         new_user != INIT_USER)
 637                 current->flags |= PF_NPROC_EXCEEDED;
 638         else
 639                 current->flags &= ~PF_NPROC_EXCEEDED;
 640
 641         free_uid(new->user);
 642         new->user = new_user;
 643         return 0;
 644 }
 645
 646 /*
 647  * Unprivileged users may change the real uid to the effective uid
 648  * or vice versa.  (BSD-style)
 649  *
 650  * If you set the real uid at all, or set the effective uid to a value not
 651  * equal to the real uid, then the saved uid is set to the new effective uid.
 652  *
 653  * This makes it possible for a setuid program to completely drop its
 654  * privileges, which is often a useful assertion to make when you are doing
 655  * a security audit over a program.
 656  *
 657  * The general idea is that a program which uses just setreuid() will be
 658  * 100% compatible with BSD.  A program which uses just setuid() will be
 659  * 100% compatible with POSIX with saved IDs.
 660  */
 661 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 662 {
 663         const struct cred *old;
 664         struct cred *new;
 665         int retval;
 666
 667         new = prepare_creds();
 668         if (!new)
 669                 return -ENOMEM;
 670         old = current_cred();
 671
 672         retval = -EPERM;
 673         if (ruid != (uid_t) -1) {
 674                 new->uid = ruid;
 675                 if (old->uid != ruid &&
 676                     old->euid != ruid &&
 677                     !nsown_capable(CAP_SETUID))
 678                         goto error;
 679         }
 680
 681         if (euid != (uid_t) -1) {
 682                 new->euid = euid;
 683                 if (old->uid != euid &&
 684                     old->euid != euid &&
 685                     old->suid != euid &&
 686                     !nsown_capable(CAP_SETUID))
 687                         goto error;
 688         }
 689
 690         if (new->uid != old->uid) {
 691                 retval = set_user(new);
 692                 if (retval < 0)
 693                         goto error;
 694         }
 695         if (ruid != (uid_t) -1 ||
 696             (euid != (uid_t) -1 && euid != old->uid))
 697                 new->suid = new->euid;
 698         new->fsuid = new->euid;
 699
 700         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 701         if (retval < 0)
 702                 goto error;
 703
 704         return commit_creds(new);
 705
 706 error:
 707         abort_creds(new);
 708         return retval;
 709 }
 710
 711 /*
 712  * setuid() is implemented like SysV with SAVED_IDS
 713  *
 714  * Note that SAVED_ID's is deficient in that a setuid root program
 715  * like sendmail, for example, cannot set its uid to be a normal
 716  * user and then switch back, because if you're root, setuid() sets
 717  * the saved uid too.  If you don't like this, blame the bright people
 718  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 719  * will allow a root program to temporarily drop privileges and be able to
 720  * regain them by swapping the real and effective uid.
 721  */
 722 SYSCALL_DEFINE1(setuid, uid_t, uid)
 723 {
 724         const struct cred *old;
 725         struct cred *new;
 726         int retval;
 727
 728         new = prepare_creds();
 729         if (!new)
 730                 return -ENOMEM;
 731         old = current_cred();
 732
 733         retval = -EPERM;
 734         if (nsown_capable(CAP_SETUID)) {
 735                 new->suid = new->uid = uid;
 736                 if (uid != old->uid) {
 737                         retval = set_user(new);
 738                         if (retval < 0)
 739                                 goto error;
 740                 }
 741         } else if (uid != old->uid && uid != new->suid) {
 742                 goto error;
 743         }
 744
 745         new->fsuid = new->euid = uid;
 746
 747         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 748         if (retval < 0)
 749                 goto error;
 750
 751         return commit_creds(new);
 752
 753 error:
 754         abort_creds(new);
 755         return retval;
 756 }
 757
 758
 759 /*
 760  * This function implements a generic ability to update ruid, euid,
 761  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 762  */
 763 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 764 {
 765         const struct cred *old;
 766         struct cred *new;
 767         int retval;
 768
 769         new = prepare_creds();
 770         if (!new)
 771                 return -ENOMEM;
 772
 773         old = current_cred();
 774
 775         retval = -EPERM;
 776         if (!nsown_capable(CAP_SETUID)) {
 777                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 778                     ruid != old->euid  && ruid != old->suid)
 779                         goto error;
 780                 if (euid != (uid_t) -1 && euid != old->uid &&
 781                     euid != old->euid  && euid != old->suid)
 782                         goto error;
 783                 if (suid != (uid_t) -1 && suid != old->uid &&
 784                     suid != old->euid  && suid != old->suid)
 785                         goto error;
 786         }
 787
 788         if (ruid != (uid_t) -1) {
 789                 new->uid = ruid;
 790                 if (ruid != old->uid) {
 791                         retval = set_user(new);
 792                         if (retval < 0)
 793                                 goto error;
 794                 }
 795         }
 796         if (euid != (uid_t) -1)
 797                 new->euid = euid;
 798         if (suid != (uid_t) -1)
 799                 new->suid = suid;
 800         new->fsuid = new->euid;
 801
 802         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 803         if (retval < 0)
 804                 goto error;
 805
 806         return commit_creds(new);
 807
 808 error:
 809         abort_creds(new);
 810         return retval;
 811 }
 812
 813 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 814 {
 815         const struct cred *cred = current_cred();
 816         int retval;
 817
 818         if (!(retval   = put_user(cred->uid,  ruid)) &&
 819             !(retval   = put_user(cred->euid, euid)))
 820                 retval = put_user(cred->suid, suid);
 821
 822         return retval;
 823 }
 824
 825 /*
 826  * Same as above, but for rgid, egid, sgid.
 827  */
 828 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 829 {
 830         const struct cred *old;
 831         struct cred *new;
 832         int retval;
 833
 834         new = prepare_creds();
 835         if (!new)
 836                 return -ENOMEM;
 837         old = current_cred();
 838
 839         retval = -EPERM;
 840         if (!nsown_capable(CAP_SETGID)) {
 841                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 842                     rgid != old->egid  && rgid != old->sgid)
 843                         goto error;
 844                 if (egid != (gid_t) -1 && egid != old->gid &&
 845                     egid != old->egid  && egid != old->sgid)
 846                         goto error;
 847                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 848                     sgid != old->egid  && sgid != old->sgid)
 849                         goto error;
 850         }
 851
 852         if (rgid != (gid_t) -1)
 853                 new->gid = rgid;
 854         if (egid != (gid_t) -1)
 855                 new->egid = egid;
 856         if (sgid != (gid_t) -1)
 857                 new->sgid = sgid;
 858         new->fsgid = new->egid;
 859
 860         return commit_creds(new);
 861
 862 error:
 863         abort_creds(new);
 864         return retval;
 865 }
 866
 867 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 868 {
 869         const struct cred *cred = current_cred();
 870         int retval;
 871
 872         if (!(retval   = put_user(cred->gid,  rgid)) &&
 873             !(retval   = put_user(cred->egid, egid)))
 874                 retval = put_user(cred->sgid, sgid);
 875
 876         return retval;
 877 }
 878
 879
 880 /*
 881  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 882  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 883  * whatever uid it wants to). It normally shadows "euid", except when
 884  * explicitly set by setfsuid() or for access..
 885  */
 886 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 887 {
 888         const struct cred *old;
 889         struct cred *new;
 890         uid_t old_fsuid;
 891
 892         new = prepare_creds();
 893         if (!new)
 894                 return current_fsuid();
 895         old = current_cred();
 896         old_fsuid = old->fsuid;
 897
 898         if (uid == old->uid  || uid == old->euid  ||
 899             uid == old->suid || uid == old->fsuid ||
 900             nsown_capable(CAP_SETUID)) {
 901                 if (uid != old_fsuid) {
 902                         new->fsuid = uid;
 903                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 904                                 goto change_okay;
 905                 }
 906         }
 907
 908         abort_creds(new);
 909         return old_fsuid;
 910
 911 change_okay:
 912         commit_creds(new);
 913         return old_fsuid;
 914 }
 915
 916 /*
 917  * Samma på svenska..
 918  */
 919 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 920 {
 921         const struct cred *old;
 922         struct cred *new;
 923         gid_t old_fsgid;
 924
 925         new = prepare_creds();
 926         if (!new)
 927                 return current_fsgid();
 928         old = current_cred();
 929         old_fsgid = old->fsgid;
 930
 931         if (gid == old->gid  || gid == old->egid  ||
 932             gid == old->sgid || gid == old->fsgid ||
 933             nsown_capable(CAP_SETGID)) {
 934                 if (gid != old_fsgid) {
 935                         new->fsgid = gid;
 936                         goto change_okay;
 937                 }
 938         }
 939
 940         abort_creds(new);
 941         return old_fsgid;
 942
 943 change_okay:
 944         commit_creds(new);
 945         return old_fsgid;
 946 }
 947
 948 void do_sys_times(struct tms *tms)
 949 {
 950         cputime_t tgutime, tgstime, cutime, cstime;
 951
 952         spin_lock_irq(&current->sighand->siglock);
 953         thread_group_times(current, &tgutime, &tgstime);
 954         cutime = current->signal->cutime;
 955         cstime = current->signal->cstime;
 956         spin_unlock_irq(&current->sighand->siglock);
 957         tms->tms_utime = cputime_to_clock_t(tgutime);
 958         tms->tms_stime = cputime_to_clock_t(tgstime);
 959         tms->tms_cutime = cputime_to_clock_t(cutime);
 960         tms->tms_cstime = cputime_to_clock_t(cstime);
 961 }
 962
 963 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 964 {
 965         if (tbuf) {
 966                 struct tms tmp;
 967
 968                 do_sys_times(&tmp);
 969                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 970                         return -EFAULT;
 971         }
 972         force_successful_syscall_return();
 973         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 974 }
 975
 976 /*
 977  * This needs some heavy checking ...
 978  * I just haven't the stomach for it. I also don't fully
 979  * understand sessions/pgrp etc. Let somebody who does explain it.
 980  *
 981  * OK, I think I have the protection semantics right.... this is really
 982  * only important on a multi-user system anyway, to make sure one user
 983  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 984  *
 985  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 986  * LBT 04.03.94
 987  */
 988 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 989 {
 990         struct task_struct *p;
 991         struct task_struct *group_leader = current->group_leader;
 992         struct pid *pgrp;
 993         int err;
 994
 995         if (!pid)
 996                 pid = task_pid_vnr(group_leader);
 997         if (!pgid)
 998                 pgid = pid;
 999         if (pgid < 0)
1000                 return -EINVAL;
1001         rcu_read_lock();
1002
1003         /* From this point forward we keep holding onto the tasklist lock
1004          * so that our parent does not change from under us. -DaveM
1005          */
1006         write_lock_irq(&tasklist_lock);
1007
1008         err = -ESRCH;
1009         p = find_task_by_vpid(pid);
1010         if (!p)
1011                 goto out;
1012
1013         err = -EINVAL;
1014         if (!thread_group_leader(p))
1015                 goto out;
1016
1017         if (same_thread_group(p->real_parent, group_leader)) {
1018                 err = -EPERM;
1019                 if (task_session(p) != task_session(group_leader))
1020                         goto out;
1021                 err = -EACCES;
1022                 if (p->did_exec)
1023                         goto out;
1024         } else {
1025                 err = -ESRCH;
1026                 if (p != group_leader)
1027                         goto out;
1028         }
1029
1030         err = -EPERM;
1031         if (p->signal->leader)
1032                 goto out;
1033
1034         pgrp = task_pid(p);
1035         if (pgid != pid) {
1036                 struct task_struct *g;
1037
1038                 pgrp = find_vpid(pgid);
1039                 g = pid_task(pgrp, PIDTYPE_PGID);
1040                 if (!g || task_session(g) != task_session(group_leader))
1041                         goto out;
1042         }
1043
1044         err = security_task_setpgid(p, pgid);
1045         if (err)
1046                 goto out;
1047
1048         if (task_pgrp(p) != pgrp)
1049                 change_pid(p, PIDTYPE_PGID, pgrp);
1050
1051         err = 0;
1052 out:
1053         /* All paths lead to here, thus we are safe. -DaveM */
1054         write_unlock_irq(&tasklist_lock);
1055         rcu_read_unlock();
1056         return err;
1057 }
1058
1059 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1060 {
1061         struct task_struct *p;
1062         struct pid *grp;
1063         int retval;
1064
1065         rcu_read_lock();
1066         if (!pid)
1067                 grp = task_pgrp(current);
1068         else {
1069                 retval = -ESRCH;
1070                 p = find_task_by_vpid(pid);
1071                 if (!p)
1072                         goto out;
1073                 grp = task_pgrp(p);
1074                 if (!grp)
1075                         goto out;
1076
1077                 retval = security_task_getpgid(p);
1078                 if (retval)
1079                         goto out;
1080         }
1081         retval = pid_vnr(grp);
1082 out:
1083         rcu_read_unlock();
1084         return retval;
1085 }
1086
1087 #ifdef __ARCH_WANT_SYS_GETPGRP
1088
1089 SYSCALL_DEFINE0(getpgrp)
1090 {
1091         return sys_getpgid(0);
1092 }
1093
1094 #endif
1095
1096 SYSCALL_DEFINE1(getsid, pid_t, pid)
1097 {
1098         struct task_struct *p;
1099         struct pid *sid;
1100         int retval;
1101
1102         rcu_read_lock();
1103         if (!pid)
1104                 sid = task_session(current);
1105         else {
1106                 retval = -ESRCH;
1107                 p = find_task_by_vpid(pid);
1108                 if (!p)
1109                         goto out;
1110                 sid = task_session(p);
1111                 if (!sid)
1112                         goto out;
1113
1114                 retval = security_task_getsid(p);
1115                 if (retval)
1116                         goto out;
1117         }
1118         retval = pid_vnr(sid);
1119 out:
1120         rcu_read_unlock();
1121         return retval;
1122 }
1123
1124 SYSCALL_DEFINE0(setsid)
1125 {
1126         struct task_struct *group_leader = current->group_leader;
1127         struct pid *sid = task_pid(group_leader);
1128         pid_t session = pid_vnr(sid);
1129         int err = -EPERM;
1130
1131         write_lock_irq(&tasklist_lock);
1132         /* Fail if I am already a session leader */
1133         if (group_leader->signal->leader)
1134                 goto out;
1135
1136         /* Fail if a process group id already exists that equals the
1137          * proposed session id.
1138          */
1139         if (pid_task(sid, PIDTYPE_PGID))
1140                 goto out;
1141
1142         group_leader->signal->leader = 1;
1143         __set_special_pids(sid);
1144
1145         proc_clear_tty(group_leader);
1146
1147         err = session;
1148 out:
1149         write_unlock_irq(&tasklist_lock);
1150         if (err > 0) {
1151                 proc_sid_connector(group_leader);
1152                 sched_autogroup_create_attach(group_leader);
1153         }
1154         return err;
1155 }
1156
1157 DECLARE_RWSEM(uts_sem);
1158
1159 #ifdef COMPAT_UTS_MACHINE
1160 #define override_architecture(name) \
1161         (personality(current->personality) == PER_LINUX32 && \
1162          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1163                       sizeof(COMPAT_UTS_MACHINE)))
1164 #else
1165 #define override_architecture(name)     0
1166 #endif
1167
1168 /*
1169  * Work around broken programs that cannot handle "Linux 3.0".
1170  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1171  */
1172 static int override_release(char __user *release, int len)
1173 {
1174         int ret = 0;
1175         char buf[len];
1176
1177         if (current->personality & UNAME26) {
1178                 char *rest = UTS_RELEASE;
1179                 int ndots = 0;
1180                 unsigned v;
1181
1182                 while (*rest) {
1183                         if (*rest == '.' && ++ndots >= 3)
1184                                 break;
1185                         if (!isdigit(*rest) && *rest != '.')
1186                                 break;
1187                         rest++;
1188                 }
1189                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1190                 snprintf(buf, len, "2.6.%u%s", v, rest);
1191                 ret = copy_to_user(release, buf, len);
1192         }
1193         return ret;
1194 }
1195
1196 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1197 {
1198         int errno = 0;
1199
1200         down_read(&uts_sem);
1201         if (copy_to_user(name, utsname(), sizeof *name))
1202                 errno = -EFAULT;
1203         up_read(&uts_sem);
1204
1205         if (!errno && override_release(name->release, sizeof(name->release)))
1206                 errno = -EFAULT;
1207         if (!errno && override_architecture(name))
1208                 errno = -EFAULT;
1209         return errno;
1210 }
1211
1212 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1213 /*
1214  * Old cruft
1215  */
1216 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1217 {
1218         int error = 0;
1219
1220         if (!name)
1221                 return -EFAULT;
1222
1223         down_read(&uts_sem);
1224         if (copy_to_user(name, utsname(), sizeof(*name)))
1225                 error = -EFAULT;
1226         up_read(&uts_sem);
1227
1228         if (!error && override_release(name->release, sizeof(name->release)))
1229                 error = -EFAULT;
1230         if (!error && override_architecture(name))
1231                 error = -EFAULT;
1232         return error;
1233 }
1234
1235 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1236 {
1237         int error;
1238
1239         if (!name)
1240                 return -EFAULT;
1241         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1242                 return -EFAULT;
1243
1244         down_read(&uts_sem);
1245         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1246                                __OLD_UTS_LEN);
1247         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1248         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1249                                 __OLD_UTS_LEN);
1250         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1251         error |= __copy_to_user(&name->release, &utsname()->release,
1252                                 __OLD_UTS_LEN);
1253         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1254         error |= __copy_to_user(&name->version, &utsname()->version,
1255                                 __OLD_UTS_LEN);
1256         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1257         error |= __copy_to_user(&name->machine, &utsname()->machine,
1258                                 __OLD_UTS_LEN);
1259         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1260         up_read(&uts_sem);
1261
1262         if (!error && override_architecture(name))
1263                 error = -EFAULT;
1264         if (!error && override_release(name->release, sizeof(name->release)))
1265                 error = -EFAULT;
1266         return error ? -EFAULT : 0;
1267 }
1268 #endif
1269
1270 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1271 {
1272         int errno;
1273         char tmp[__NEW_UTS_LEN];
1274
1275         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1276                 return -EPERM;
1277
1278         if (len < 0 || len > __NEW_UTS_LEN)
1279                 return -EINVAL;
1280         down_write(&uts_sem);
1281         errno = -EFAULT;
1282         if (!copy_from_user(tmp, name, len)) {
1283                 struct new_utsname *u = utsname();
1284
1285                 memcpy(u->nodename, tmp, len);
1286                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1287                 errno = 0;
1288         }
1289         up_write(&uts_sem);
1290         return errno;
1291 }
1292
1293 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1294
1295 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1296 {
1297         int i, errno;
1298         struct new_utsname *u;
1299
1300         if (len < 0)
1301                 return -EINVAL;
1302         down_read(&uts_sem);
1303         u = utsname();
1304         i = 1 + strlen(u->nodename);
1305         if (i > len)
1306                 i = len;
1307         errno = 0;
1308         if (copy_to_user(name, u->nodename, i))
1309                 errno = -EFAULT;
1310         up_read(&uts_sem);
1311         return errno;
1312 }
1313
1314 #endif
1315
1316 /*
1317  * Only setdomainname; getdomainname can be implemented by calling
1318  * uname()
1319  */
1320 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1321 {
1322         int errno;
1323         char tmp[__NEW_UTS_LEN];
1324
1325         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1326                 return -EPERM;
1327         if (len < 0 || len > __NEW_UTS_LEN)
1328                 return -EINVAL;
1329
1330         down_write(&uts_sem);
1331         errno = -EFAULT;
1332         if (!copy_from_user(tmp, name, len)) {
1333                 struct new_utsname *u = utsname();
1334
1335                 memcpy(u->domainname, tmp, len);
1336                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1337                 errno = 0;
1338         }
1339         up_write(&uts_sem);
1340         return errno;
1341 }
1342
1343 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1344 {
1345         struct rlimit value;
1346         int ret;
1347
1348         ret = do_prlimit(current, resource, NULL, &value);
1349         if (!ret)
1350                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1351
1352         return ret;
1353 }
1354
1355 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1356
1357 /*
1358  *      Back compatibility for getrlimit. Needed for some apps.
1359  */
1360
1361 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1362                 struct rlimit __user *, rlim)
1363 {
1364         struct rlimit x;
1365         if (resource >= RLIM_NLIMITS)
1366                 return -EINVAL;
1367
1368         task_lock(current->group_leader);
1369         x = current->signal->rlim[resource];
1370         task_unlock(current->group_leader);
1371         if (x.rlim_cur > 0x7FFFFFFF)
1372                 x.rlim_cur = 0x7FFFFFFF;
1373         if (x.rlim_max > 0x7FFFFFFF)
1374                 x.rlim_max = 0x7FFFFFFF;
1375         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1376 }
1377
1378 #endif
1379
1380 static inline bool rlim64_is_infinity(__u64 rlim64)
1381 {
1382 #if BITS_PER_LONG < 64
1383         return rlim64 >= ULONG_MAX;
1384 #else
1385         return rlim64 == RLIM64_INFINITY;
1386 #endif
1387 }
1388
1389 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1390 {
1391         if (rlim->rlim_cur == RLIM_INFINITY)
1392                 rlim64->rlim_cur = RLIM64_INFINITY;
1393         else
1394                 rlim64->rlim_cur = rlim->rlim_cur;
1395         if (rlim->rlim_max == RLIM_INFINITY)
1396                 rlim64->rlim_max = RLIM64_INFINITY;
1397         else
1398                 rlim64->rlim_max = rlim->rlim_max;
1399 }
1400
1401 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1402 {
1403         if (rlim64_is_infinity(rlim64->rlim_cur))
1404                 rlim->rlim_cur = RLIM_INFINITY;
1405         else
1406                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1407         if (rlim64_is_infinity(rlim64->rlim_max))
1408                 rlim->rlim_max = RLIM_INFINITY;
1409         else
1410                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1411 }
1412
1413 /* make sure you are allowed to change @tsk limits before calling this */
1414 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1415                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1416 {
1417         struct rlimit *rlim;
1418         int retval = 0;
1419
1420         if (resource >= RLIM_NLIMITS)
1421                 return -EINVAL;
1422         if (new_rlim) {
1423                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1424                         return -EINVAL;
1425                 if (resource == RLIMIT_NOFILE &&
1426                                 new_rlim->rlim_max > sysctl_nr_open)
1427                         return -EPERM;
1428         }
1429
1430         /* protect tsk->signal and tsk->sighand from disappearing */
1431         read_lock(&tasklist_lock);
1432         if (!tsk->sighand) {
1433                 retval = -ESRCH;
1434                 goto out;
1435         }
1436
1437         rlim = tsk->signal->rlim + resource;
1438         task_lock(tsk->group_leader);
1439         if (new_rlim) {
1440                 /* Keep the capable check against init_user_ns until
1441                    cgroups can contain all limits */
1442                 if (new_rlim->rlim_max > rlim->rlim_max &&
1443                                 !capable(CAP_SYS_RESOURCE))
1444                         retval = -EPERM;
1445                 if (!retval)
1446                         retval = security_task_setrlimit(tsk->group_leader,
1447                                         resource, new_rlim);
1448                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1449                         /*
1450                          * The caller is asking for an immediate RLIMIT_CPU
1451                          * expiry.  But we use the zero value to mean "it was
1452                          * never set".  So let's cheat and make it one second
1453                          * instead
1454                          */
1455                         new_rlim->rlim_cur = 1;
1456                 }
1457         }
1458         if (!retval) {
1459                 if (old_rlim)
1460                         *old_rlim = *rlim;
1461                 if (new_rlim)
1462                         *rlim = *new_rlim;
1463         }
1464         task_unlock(tsk->group_leader);
1465
1466         /*
1467          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1468          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1469          * very long-standing error, and fixing it now risks breakage of
1470          * applications, so we live with it
1471          */
1472          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1473                          new_rlim->rlim_cur != RLIM_INFINITY)
1474                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1475 out:
1476         read_unlock(&tasklist_lock);
1477         return retval;
1478 }
1479
1480 /* rcu lock must be held */
1481 static int check_prlimit_permission(struct task_struct *task)
1482 {
1483         const struct cred *cred = current_cred(), *tcred;
1484
1485         if (current == task)
1486                 return 0;
1487
1488         tcred = __task_cred(task);
1489         if (cred->user->user_ns == tcred->user->user_ns &&
1490             (cred->uid == tcred->euid &&
1491              cred->uid == tcred->suid &&
1492              cred->uid == tcred->uid  &&
1493              cred->gid == tcred->egid &&
1494              cred->gid == tcred->sgid &&
1495              cred->gid == tcred->gid))
1496                 return 0;
1497         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1498                 return 0;
1499
1500         return -EPERM;
1501 }
1502
1503 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1504                 const struct rlimit64 __user *, new_rlim,
1505                 struct rlimit64 __user *, old_rlim)
1506 {
1507         struct rlimit64 old64, new64;
1508         struct rlimit old, new;
1509         struct task_struct *tsk;
1510         int ret;
1511
1512         if (new_rlim) {
1513                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1514                         return -EFAULT;
1515                 rlim64_to_rlim(&new64, &new);
1516         }
1517
1518         rcu_read_lock();
1519         tsk = pid ? find_task_by_vpid(pid) : current;
1520         if (!tsk) {
1521                 rcu_read_unlock();
1522                 return -ESRCH;
1523         }
1524         ret = check_prlimit_permission(tsk);
1525         if (ret) {
1526                 rcu_read_unlock();
1527                 return ret;
1528         }
1529         get_task_struct(tsk);
1530         rcu_read_unlock();
1531
1532         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1533                         old_rlim ? &old : NULL);
1534
1535         if (!ret && old_rlim) {
1536                 rlim_to_rlim64(&old, &old64);
1537                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1538                         ret = -EFAULT;
1539         }
1540
1541         put_task_struct(tsk);
1542         return ret;
1543 }
1544
1545 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1546 {
1547         struct rlimit new_rlim;
1548
1549         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1550                 return -EFAULT;
1551         return do_prlimit(current, resource, &new_rlim, NULL);
1552 }
1553
1554 /*
1555  * It would make sense to put struct rusage in the task_struct,
1556  * except that would make the task_struct be *really big*.  After
1557  * task_struct gets moved into malloc'ed memory, it would
1558  * make sense to do this.  It will make moving the rest of the information
1559  * a lot simpler!  (Which we're not doing right now because we're not
1560  * measuring them yet).
1561  *
1562  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1563  * races with threads incrementing their own counters.  But since word
1564  * reads are atomic, we either get new values or old values and we don't
1565  * care which for the sums.  We always take the siglock to protect reading
1566  * the c* fields from p->signal from races with exit.c updating those
1567  * fields when reaping, so a sample either gets all the additions of a
1568  * given child after it's reaped, or none so this sample is before reaping.
1569  *
1570  * Locking:
1571  * We need to take the siglock for CHILDEREN, SELF and BOTH
1572  * for  the cases current multithreaded, non-current single threaded
1573  * non-current multithreaded.  Thread traversal is now safe with
1574  * the siglock held.
1575  * Strictly speaking, we donot need to take the siglock if we are current and
1576  * single threaded,  as no one else can take our signal_struct away, no one
1577  * else can  reap the  children to update signal->c* counters, and no one else
1578  * can race with the signal-> fields. If we do not take any lock, the
1579  * signal-> fields could be read out of order while another thread was just
1580  * exiting. So we should  place a read memory barrier when we avoid the lock.
1581  * On the writer side,  write memory barrier is implied in  __exit_signal
1582  * as __exit_signal releases  the siglock spinlock after updating the signal->
1583  * fields. But we don't do this yet to keep things simple.
1584  *
1585  */
1586
1587 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1588 {
1589         r->ru_nvcsw += t->nvcsw;
1590         r->ru_nivcsw += t->nivcsw;
1591         r->ru_minflt += t->min_flt;
1592         r->ru_majflt += t->maj_flt;
1593         r->ru_inblock += task_io_get_inblock(t);
1594         r->ru_oublock += task_io_get_oublock(t);
1595 }
1596
1597 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1598 {
1599         struct task_struct *t;
1600         unsigned long flags;
1601         cputime_t tgutime, tgstime, utime, stime;
1602         unsigned long maxrss = 0;
1603
1604         memset((char *) r, 0, sizeof *r);
1605         utime = stime = cputime_zero;
1606
1607         if (who == RUSAGE_THREAD) {
1608                 task_times(current, &utime, &stime);
1609                 accumulate_thread_rusage(p, r);
1610                 maxrss = p->signal->maxrss;
1611                 goto out;
1612         }
1613
1614         if (!lock_task_sighand(p, &flags))
1615                 return;
1616
1617         switch (who) {
1618                 case RUSAGE_BOTH:
1619                 case RUSAGE_CHILDREN:
1620                         utime = p->signal->cutime;
1621                         stime = p->signal->cstime;
1622                         r->ru_nvcsw = p->signal->cnvcsw;
1623                         r->ru_nivcsw = p->signal->cnivcsw;
1624                         r->ru_minflt = p->signal->cmin_flt;
1625                         r->ru_majflt = p->signal->cmaj_flt;
1626                         r->ru_inblock = p->signal->cinblock;
1627                         r->ru_oublock = p->signal->coublock;
1628                         maxrss = p->signal->cmaxrss;
1629
1630                         if (who == RUSAGE_CHILDREN)
1631                                 break;
1632
1633                 case RUSAGE_SELF:
1634                         thread_group_times(p, &tgutime, &tgstime);
1635                         utime = cputime_add(utime, tgutime);
1636                         stime = cputime_add(stime, tgstime);
1637                         r->ru_nvcsw += p->signal->nvcsw;
1638                         r->ru_nivcsw += p->signal->nivcsw;
1639                         r->ru_minflt += p->signal->min_flt;
1640                         r->ru_majflt += p->signal->maj_flt;
1641                         r->ru_inblock += p->signal->inblock;
1642                         r->ru_oublock += p->signal->oublock;
1643                         if (maxrss < p->signal->maxrss)
1644                                 maxrss = p->signal->maxrss;
1645                         t = p;
1646                         do {
1647                                 accumulate_thread_rusage(t, r);
1648                                 t = next_thread(t);
1649                         } while (t != p);
1650                         break;
1651
1652                 default:
1653                         BUG();
1654         }
1655         unlock_task_sighand(p, &flags);
1656
1657 out:
1658         cputime_to_timeval(utime, &r->ru_utime);
1659         cputime_to_timeval(stime, &r->ru_stime);
1660
1661         if (who != RUSAGE_CHILDREN) {
1662                 struct mm_struct *mm = get_task_mm(p);
1663                 if (mm) {
1664                         setmax_mm_hiwater_rss(&maxrss, mm);
1665                         mmput(mm);
1666                 }
1667         }
1668         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1669 }
1670
1671 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1672 {
1673         struct rusage r;
1674         k_getrusage(p, who, &r);
1675         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1676 }
1677
1678 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1679 {
1680         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1681             who != RUSAGE_THREAD)
1682                 return -EINVAL;
1683         return getrusage(current, who, ru);
1684 }
1685
1686 SYSCALL_DEFINE1(umask, int, mask)
1687 {
1688         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1689         return mask;
1690 }
1691
1692 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1693                 unsigned long, arg4, unsigned long, arg5)
1694 {
1695         struct task_struct *me = current;
1696         unsigned char comm[sizeof(me->comm)];
1697         long error;
1698
1699         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1700         if (error != -ENOSYS)
1701                 return error;
1702
1703         error = 0;
1704         switch (option) {
1705                 case PR_SET_PDEATHSIG:
1706                         if (!valid_signal(arg2)) {
1707                                 error = -EINVAL;
1708                                 break;
1709                         }
1710                         me->pdeath_signal = arg2;
1711                         error = 0;
1712                         break;
1713                 case PR_GET_PDEATHSIG:
1714                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1715                         break;
1716                 case PR_GET_DUMPABLE:
1717                         error = get_dumpable(me->mm);
1718                         break;
1719                 case PR_SET_DUMPABLE:
1720                         if (arg2 < 0 || arg2 > 1) {
1721                                 error = -EINVAL;
1722                                 break;
1723                         }
1724                         set_dumpable(me->mm, arg2);
1725                         error = 0;
1726                         break;
1727
1728                 case PR_SET_UNALIGN:
1729                         error = SET_UNALIGN_CTL(me, arg2);
1730                         break;
1731                 case PR_GET_UNALIGN:
1732                         error = GET_UNALIGN_CTL(me, arg2);
1733                         break;
1734                 case PR_SET_FPEMU:
1735                         error = SET_FPEMU_CTL(me, arg2);
1736                         break;
1737                 case PR_GET_FPEMU:
1738                         error = GET_FPEMU_CTL(me, arg2);
1739                         break;
1740                 case PR_SET_FPEXC:
1741                         error = SET_FPEXC_CTL(me, arg2);
1742                         break;
1743                 case PR_GET_FPEXC:
1744                         error = GET_FPEXC_CTL(me, arg2);
1745                         break;
1746                 case PR_GET_TIMING:
1747                         error = PR_TIMING_STATISTICAL;
1748                         break;
1749                 case PR_SET_TIMING:
1750                         if (arg2 != PR_TIMING_STATISTICAL)
1751                                 error = -EINVAL;
1752                         else
1753                                 error = 0;
1754                         break;
1755
1756                 case PR_SET_NAME:
1757                         comm[sizeof(me->comm)-1] = 0;
1758                         if (strncpy_from_user(comm, (char __user *)arg2,
1759                                               sizeof(me->comm) - 1) < 0)
1760                                 return -EFAULT;
1761                         set_task_comm(me, comm);
1762                         return 0;
1763                 case PR_GET_NAME:
1764                         get_task_comm(comm, me);
1765                         if (copy_to_user((char __user *)arg2, comm,
1766                                          sizeof(comm)))
1767                                 return -EFAULT;
1768                         return 0;
1769                 case PR_GET_ENDIAN:
1770                         error = GET_ENDIAN(me, arg2);
1771                         break;
1772                 case PR_SET_ENDIAN:
1773                         error = SET_ENDIAN(me, arg2);
1774                         break;
1775
1776                 case PR_GET_SECCOMP:
1777                         error = prctl_get_seccomp();
1778                         break;
1779                 case PR_SET_SECCOMP:
1780                         error = prctl_set_seccomp(arg2);
1781                         break;
1782                 case PR_GET_TSC:
1783                         error = GET_TSC_CTL(arg2);
1784                         break;
1785                 case PR_SET_TSC:
1786                         error = SET_TSC_CTL(arg2);
1787                         break;
1788                 case PR_TASK_PERF_EVENTS_DISABLE:
1789                         error = perf_event_task_disable();
1790                         break;
1791                 case PR_TASK_PERF_EVENTS_ENABLE:
1792                         error = perf_event_task_enable();
1793                         break;
1794                 case PR_GET_TIMERSLACK:
1795                         error = current->timer_slack_ns;
1796                         break;
1797                 case PR_SET_TIMERSLACK:
1798                         if (arg2 <= 0)
1799                                 current->timer_slack_ns =
1800                                         current->default_timer_slack_ns;
1801                         else
1802                                 current->timer_slack_ns = arg2;
1803                         error = 0;
1804                         break;
1805                 case PR_MCE_KILL:
1806                         if (arg4 | arg5)
1807                                 return -EINVAL;
1808                         switch (arg2) {
1809                         case PR_MCE_KILL_CLEAR:
1810                                 if (arg3 != 0)
1811                                         return -EINVAL;
1812                                 current->flags &= ~PF_MCE_PROCESS;
1813                                 break;
1814                         case PR_MCE_KILL_SET:
1815                                 current->flags |= PF_MCE_PROCESS;
1816                                 if (arg3 == PR_MCE_KILL_EARLY)
1817                                         current->flags |= PF_MCE_EARLY;
1818                                 else if (arg3 == PR_MCE_KILL_LATE)
1819                                         current->flags &= ~PF_MCE_EARLY;
1820                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1821                                         current->flags &=
1822                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1823                                 else
1824                                         return -EINVAL;
1825                                 break;
1826                         default:
1827                                 return -EINVAL;
1828                         }
1829                         error = 0;
1830                         break;
1831                 case PR_MCE_KILL_GET:
1832                         if (arg2 | arg3 | arg4 | arg5)
1833                                 return -EINVAL;
1834                         if (current->flags & PF_MCE_PROCESS)
1835                                 error = (current->flags & PF_MCE_EARLY) ?
1836                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1837                         else
1838                                 error = PR_MCE_KILL_DEFAULT;
1839                         break;
1840                 default:
1841                         error = -EINVAL;
1842                         break;
1843         }
1844         return error;
1845 }
1846
1847 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1848                 struct getcpu_cache __user *, unused)
1849 {
1850         int err = 0;
1851         int cpu = raw_smp_processor_id();
1852         if (cpup)
1853                 err |= put_user(cpu, cpup);
1854         if (nodep)
1855                 err |= put_user(cpu_to_node(cpu), nodep);
1856         return err ? -EFAULT : 0;
1857 }
1858
1859 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1860
1861 static void argv_cleanup(struct subprocess_info *info)
1862 {
1863         argv_free(info->argv);
1864 }
1865
1866 /**
1867  * orderly_poweroff - Trigger an orderly system poweroff
1868  * @force: force poweroff if command execution fails
1869  *
1870  * This may be called from any context to trigger a system shutdown.
1871  * If the orderly shutdown fails, it will force an immediate shutdown.
1872  */
1873 int orderly_poweroff(bool force)
1874 {
1875         int argc;
1876         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1877         static char *envp[] = {
1878                 "HOME=/",
1879                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1880                 NULL
1881         };
1882         int ret = -ENOMEM;
1883         struct subprocess_info *info;
1884
1885         if (argv == NULL) {
1886                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1887                        __func__, poweroff_cmd);
1888                 goto out;
1889         }
1890
1891         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1892         if (info == NULL) {
1893                 argv_free(argv);
1894                 goto out;
1895         }
1896
1897         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1898
1899         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1900
1901   out:
1902         if (ret && force) {
1903                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1904                        "forcing the issue\n");
1905
1906                 /* I guess this should try to kick off some daemon to
1907                    sync and poweroff asap.  Or not even bother syncing
1908                    if we're doing an emergency shutdown? */
1909                 emergency_sync();
1910                 kernel_power_off();
1911         }
1912
1913         return ret;
1914 }
1915 EXPORT_SYMBOL_GPL(orderly_poweroff);