kernel/sys.c

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/export.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/kmod.h>
  16 #include <linux/perf_event.h>
  17 #include <linux/resource.h>
  18 #include <linux/kernel.h>
  19 #include <linux/kexec.h>
  20 #include <linux/workqueue.h>
  21 #include <linux/capability.h>
  22 #include <linux/device.h>
  23 #include <linux/key.h>
  24 #include <linux/times.h>
  25 #include <linux/posix-timers.h>
  26 #include <linux/security.h>
  27 #include <linux/dcookies.h>
  28 #include <linux/suspend.h>
  29 #include <linux/tty.h>
  30 #include <linux/signal.h>
  31 #include <linux/cn_proc.h>
  32 #include <linux/getcpu.h>
  33 #include <linux/task_io_accounting_ops.h>
  34 #include <linux/seccomp.h>
  35 #include <linux/cpu.h>
  36 #include <linux/personality.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/fs_struct.h>
  39 #include <linux/gfp.h>
  40 #include <linux/syscore_ops.h>
  41 #include <linux/version.h>
  42 #include <linux/ctype.h>
  43
  44 #include <linux/compat.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/kprobes.h>
  47 #include <linux/user_namespace.h>
  48
  49 #include <linux/kmsg_dump.h>
  50 /* Move somewhere else to avoid recompiling? */
  51 #include <generated/utsrelease.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/io.h>
  55 #include <asm/unistd.h>
  56
  57 #ifndef SET_UNALIGN_CTL
  58 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  59 #endif
  60 #ifndef GET_UNALIGN_CTL
  61 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  62 #endif
  63 #ifndef SET_FPEMU_CTL
  64 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  65 #endif
  66 #ifndef GET_FPEMU_CTL
  67 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  68 #endif
  69 #ifndef SET_FPEXC_CTL
  70 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  71 #endif
  72 #ifndef GET_FPEXC_CTL
  73 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  74 #endif
  75 #ifndef GET_ENDIAN
  76 # define GET_ENDIAN(a,b)        (-EINVAL)
  77 #endif
  78 #ifndef SET_ENDIAN
  79 # define SET_ENDIAN(a,b)        (-EINVAL)
  80 #endif
  81 #ifndef GET_TSC_CTL
  82 # define GET_TSC_CTL(a)         (-EINVAL)
  83 #endif
  84 #ifndef SET_TSC_CTL
  85 # define SET_TSC_CTL(a)         (-EINVAL)
  86 #endif
  87
  88 /*
  89  * this is where the system-wide overflow UID and GID are defined, for
  90  * architectures that now have 32-bit UID/GID but didn't in the past
  91  */
  92
  93 int overflowuid = DEFAULT_OVERFLOWUID;
  94 int overflowgid = DEFAULT_OVERFLOWGID;
  95
  96 #ifdef CONFIG_UID16
  97 EXPORT_SYMBOL(overflowuid);
  98 EXPORT_SYMBOL(overflowgid);
  99 #endif
 100
 101 /*
 102  * the same as above, but for filesystems which can only store a 16-bit
 103  * UID and GID. as such, this is needed on all architectures
 104  */
 105
 106 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 107 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 108
 109 EXPORT_SYMBOL(fs_overflowuid);
 110 EXPORT_SYMBOL(fs_overflowgid);
 111
 112 /*
 113  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 114  */
 115
 116 int C_A_D = 1;
 117 struct pid *cad_pid;
 118 EXPORT_SYMBOL(cad_pid);
 119
 120 /*
 121  * If set, this is used for preparing the system to power off.
 122  */
 123
 124 void (*pm_power_off_prepare)(void);
 125
 126 /*
 127  * Returns true if current's euid is same as p's uid or euid,
 128  * or has CAP_SYS_NICE to p's user_ns.
 129  *
 130  * Called with rcu_read_lock, creds are safe
 131  */
 132 static bool set_one_prio_perm(struct task_struct *p)
 133 {
 134         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 135
 136         if (pcred->user->user_ns == cred->user->user_ns &&
 137             (pcred->uid  == cred->euid ||
 138              pcred->euid == cred->euid))
 139                 return true;
 140         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 141                 return true;
 142         return false;
 143 }
 144
 145 /*
 146  * set the priority of a task
 147  * - the caller must hold the RCU read lock
 148  */
 149 static int set_one_prio(struct task_struct *p, int niceval, int error)
 150 {
 151         int no_nice;
 152
 153         if (!set_one_prio_perm(p)) {
 154                 error = -EPERM;
 155                 goto out;
 156         }
 157         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 158                 error = -EACCES;
 159                 goto out;
 160         }
 161         no_nice = security_task_setnice(p, niceval);
 162         if (no_nice) {
 163                 error = no_nice;
 164                 goto out;
 165         }
 166         if (error == -ESRCH)
 167                 error = 0;
 168         set_user_nice(p, niceval);
 169 out:
 170         return error;
 171 }
 172
 173 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 174 {
 175         struct task_struct *g, *p;
 176         struct user_struct *user;
 177         const struct cred *cred = current_cred();
 178         int error = -EINVAL;
 179         struct pid *pgrp;
 180
 181         if (which > PRIO_USER || which < PRIO_PROCESS)
 182                 goto out;
 183
 184         /* normalize: avoid signed division (rounding problems) */
 185         error = -ESRCH;
 186         if (niceval < -20)
 187                 niceval = -20;
 188         if (niceval > 19)
 189                 niceval = 19;
 190
 191         rcu_read_lock();
 192         read_lock(&tasklist_lock);
 193         switch (which) {
 194                 case PRIO_PROCESS:
 195                         if (who)
 196                                 p = find_task_by_vpid(who);
 197                         else
 198                                 p = current;
 199                         if (p)
 200                                 error = set_one_prio(p, niceval, error);
 201                         break;
 202                 case PRIO_PGRP:
 203                         if (who)
 204                                 pgrp = find_vpid(who);
 205                         else
 206                                 pgrp = task_pgrp(current);
 207                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 208                                 error = set_one_prio(p, niceval, error);
 209                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 210                         break;
 211                 case PRIO_USER:
 212                         user = (struct user_struct *) cred->user;
 213                         if (!who)
 214                                 who = cred->uid;
 215                         else if ((who != cred->uid) &&
 216                                  !(user = find_user(who)))
 217                                 goto out_unlock;        /* No processes for this user */
 218
 219                         do_each_thread(g, p) {
 220                                 if (__task_cred(p)->uid == who)
 221                                         error = set_one_prio(p, niceval, error);
 222                         } while_each_thread(g, p);
 223                         if (who != cred->uid)
 224                                 free_uid(user);         /* For find_user() */
 225                         break;
 226         }
 227 out_unlock:
 228         read_unlock(&tasklist_lock);
 229         rcu_read_unlock();
 230 out:
 231         return error;
 232 }
 233
 234 /*
 235  * Ugh. To avoid negative return values, "getpriority()" will
 236  * not return the normal nice-value, but a negated value that
 237  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 238  * to stay compatible.
 239  */
 240 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 241 {
 242         struct task_struct *g, *p;
 243         struct user_struct *user;
 244         const struct cred *cred = current_cred();
 245         long niceval, retval = -ESRCH;
 246         struct pid *pgrp;
 247
 248         if (which > PRIO_USER || which < PRIO_PROCESS)
 249                 return -EINVAL;
 250
 251         rcu_read_lock();
 252         read_lock(&tasklist_lock);
 253         switch (which) {
 254                 case PRIO_PROCESS:
 255                         if (who)
 256                                 p = find_task_by_vpid(who);
 257                         else
 258                                 p = current;
 259                         if (p) {
 260                                 niceval = 20 - task_nice(p);
 261                                 if (niceval > retval)
 262                                         retval = niceval;
 263                         }
 264                         break;
 265                 case PRIO_PGRP:
 266                         if (who)
 267                                 pgrp = find_vpid(who);
 268                         else
 269                                 pgrp = task_pgrp(current);
 270                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 271                                 niceval = 20 - task_nice(p);
 272                                 if (niceval > retval)
 273                                         retval = niceval;
 274                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 275                         break;
 276                 case PRIO_USER:
 277                         user = (struct user_struct *) cred->user;
 278                         if (!who)
 279                                 who = cred->uid;
 280                         else if ((who != cred->uid) &&
 281                                  !(user = find_user(who)))
 282                                 goto out_unlock;        /* No processes for this user */
 283
 284                         do_each_thread(g, p) {
 285                                 if (__task_cred(p)->uid == who) {
 286                                         niceval = 20 - task_nice(p);
 287                                         if (niceval > retval)
 288                                                 retval = niceval;
 289                                 }
 290                         } while_each_thread(g, p);
 291                         if (who != cred->uid)
 292                                 free_uid(user);         /* for find_user() */
 293                         break;
 294         }
 295 out_unlock:
 296         read_unlock(&tasklist_lock);
 297         rcu_read_unlock();
 298
 299         return retval;
 300 }
 301
 302 /**
 303  *      emergency_restart - reboot the system
 304  *
 305  *      Without shutting down any hardware or taking any locks
 306  *      reboot the system.  This is called when we know we are in
 307  *      trouble so this is our best effort to reboot.  This is
 308  *      safe to call in interrupt context.
 309  */
 310 void emergency_restart(void)
 311 {
 312         kmsg_dump(KMSG_DUMP_EMERG);
 313         machine_emergency_restart();
 314 }
 315 EXPORT_SYMBOL_GPL(emergency_restart);
 316
 317 void kernel_restart_prepare(char *cmd)
 318 {
 319         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 320         system_state = SYSTEM_RESTART;
 321         usermodehelper_disable();
 322         device_shutdown();
 323         syscore_shutdown();
 324 }
 325
 326 /**
 327  *      register_reboot_notifier - Register function to be called at reboot time
 328  *      @nb: Info about notifier function to be called
 329  *
 330  *      Registers a function with the list of functions
 331  *      to be called at reboot time.
 332  *
 333  *      Currently always returns zero, as blocking_notifier_chain_register()
 334  *      always returns zero.
 335  */
 336 int register_reboot_notifier(struct notifier_block *nb)
 337 {
 338         return blocking_notifier_chain_register(&reboot_notifier_list, nb);
 339 }
 340 EXPORT_SYMBOL(register_reboot_notifier);
 341
 342 /**
 343  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 344  *      @nb: Hook to be unregistered
 345  *
 346  *      Unregisters a previously registered reboot
 347  *      notifier function.
 348  *
 349  *      Returns zero on success, or %-ENOENT on failure.
 350  */
 351 int unregister_reboot_notifier(struct notifier_block *nb)
 352 {
 353         return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
 354 }
 355 EXPORT_SYMBOL(unregister_reboot_notifier);
 356
 357 /**
 358  *      kernel_restart - reboot the system
 359  *      @cmd: pointer to buffer containing command to execute for restart
 360  *              or %NULL
 361  *
 362  *      Shutdown everything and perform a clean reboot.
 363  *      This is not safe to call in interrupt context.
 364  */
 365 void kernel_restart(char *cmd)
 366 {
 367         kernel_restart_prepare(cmd);
 368         disable_nonboot_cpus();
 369         if (!cmd)
 370                 printk(KERN_EMERG "Restarting system.\n");
 371         else
 372                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 373         kmsg_dump(KMSG_DUMP_RESTART);
 374         machine_restart(cmd);
 375 }
 376 EXPORT_SYMBOL_GPL(kernel_restart);
 377
 378 static void kernel_shutdown_prepare(enum system_states state)
 379 {
 380         blocking_notifier_call_chain(&reboot_notifier_list,
 381                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 382         system_state = state;
 383         usermodehelper_disable();
 384         device_shutdown();
 385 }
 386 /**
 387  *      kernel_halt - halt the system
 388  *
 389  *      Shutdown everything and perform a clean system halt.
 390  */
 391 void kernel_halt(void)
 392 {
 393         kernel_shutdown_prepare(SYSTEM_HALT);
 394         syscore_shutdown();
 395         printk(KERN_EMERG "System halted.\n");
 396         kmsg_dump(KMSG_DUMP_HALT);
 397         machine_halt();
 398 }
 399
 400 EXPORT_SYMBOL_GPL(kernel_halt);
 401
 402 /**
 403  *      kernel_power_off - power_off the system
 404  *
 405  *      Shutdown everything and perform a clean system power_off.
 406  */
 407 void kernel_power_off(void)
 408 {
 409         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 410         if (pm_power_off_prepare)
 411                 pm_power_off_prepare();
 412         disable_nonboot_cpus();
 413         syscore_shutdown();
 414         printk(KERN_EMERG "Power down.\n");
 415         kmsg_dump(KMSG_DUMP_POWEROFF);
 416         machine_power_off();
 417 }
 418 EXPORT_SYMBOL_GPL(kernel_power_off);
 419
 420 static DEFINE_MUTEX(reboot_mutex);
 421
 422 /*
 423  * Reboot system call: for obvious reasons only root may call it,
 424  * and even root needs to set up some magic numbers in the registers
 425  * so that some mistake won't make this reboot the whole machine.
 426  * You can also set the meaning of the ctrl-alt-del-key here.
 427  *
 428  * reboot doesn't sync: do that yourself before calling this.
 429  */
 430 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 431                 void __user *, arg)
 432 {
 433         char buffer[256];
 434         int ret = 0;
 435
 436         /* We only trust the superuser with rebooting the system. */
 437         if (!capable(CAP_SYS_BOOT))
 438                 return -EPERM;
 439
 440         /* For safety, we require "magic" arguments. */
 441         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 442             (magic2 != LINUX_REBOOT_MAGIC2 &&
 443                         magic2 != LINUX_REBOOT_MAGIC2A &&
 444                         magic2 != LINUX_REBOOT_MAGIC2B &&
 445                         magic2 != LINUX_REBOOT_MAGIC2C))
 446                 return -EINVAL;
 447
 448         /* Instead of trying to make the power_off code look like
 449          * halt when pm_power_off is not set do it the easy way.
 450          */
 451         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 452                 cmd = LINUX_REBOOT_CMD_HALT;
 453
 454         mutex_lock(&reboot_mutex);
 455         switch (cmd) {
 456         case LINUX_REBOOT_CMD_RESTART:
 457                 kernel_restart(NULL);
 458                 break;
 459
 460         case LINUX_REBOOT_CMD_CAD_ON:
 461                 C_A_D = 1;
 462                 break;
 463
 464         case LINUX_REBOOT_CMD_CAD_OFF:
 465                 C_A_D = 0;
 466                 break;
 467
 468         case LINUX_REBOOT_CMD_HALT:
 469                 kernel_halt();
 470                 do_exit(0);
 471                 panic("cannot halt");
 472
 473         case LINUX_REBOOT_CMD_POWER_OFF:
 474                 kernel_power_off();
 475                 do_exit(0);
 476                 break;
 477
 478         case LINUX_REBOOT_CMD_RESTART2:
 479                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 480                         ret = -EFAULT;
 481                         break;
 482                 }
 483                 buffer[sizeof(buffer) - 1] = '\0';
 484
 485                 kernel_restart(buffer);
 486                 break;
 487
 488 #ifdef CONFIG_KEXEC
 489         case LINUX_REBOOT_CMD_KEXEC:
 490                 ret = kernel_kexec();
 491                 break;
 492 #endif
 493
 494 #ifdef CONFIG_HIBERNATION
 495         case LINUX_REBOOT_CMD_SW_SUSPEND:
 496                 ret = hibernate();
 497                 break;
 498 #endif
 499
 500         default:
 501                 ret = -EINVAL;
 502                 break;
 503         }
 504         mutex_unlock(&reboot_mutex);
 505         return ret;
 506 }
 507
 508 static void deferred_cad(struct work_struct *dummy)
 509 {
 510         kernel_restart(NULL);
 511 }
 512
 513 /*
 514  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 515  * As it's called within an interrupt, it may NOT sync: the only choice
 516  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 517  */
 518 void ctrl_alt_del(void)
 519 {
 520         static DECLARE_WORK(cad_work, deferred_cad);
 521
 522         if (C_A_D)
 523                 schedule_work(&cad_work);
 524         else
 525                 kill_cad_pid(SIGINT, 1);
 526 }
 527
 528 /*
 529  * Unprivileged users may change the real gid to the effective gid
 530  * or vice versa.  (BSD-style)
 531  *
 532  * If you set the real gid at all, or set the effective gid to a value not
 533  * equal to the real gid, then the saved gid is set to the new effective gid.
 534  *
 535  * This makes it possible for a setgid program to completely drop its
 536  * privileges, which is often a useful assertion to make when you are doing
 537  * a security audit over a program.
 538  *
 539  * The general idea is that a program which uses just setregid() will be
 540  * 100% compatible with BSD.  A program which uses just setgid() will be
 541  * 100% compatible with POSIX with saved IDs.
 542  *
 543  * SMP: There are not races, the GIDs are checked only by filesystem
 544  *      operations (as far as semantic preservation is concerned).
 545  */
 546 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 547 {
 548         const struct cred *old;
 549         struct cred *new;
 550         int retval;
 551
 552         new = prepare_creds();
 553         if (!new)
 554                 return -ENOMEM;
 555         old = current_cred();
 556
 557         retval = -EPERM;
 558         if (rgid != (gid_t) -1) {
 559                 if (old->gid == rgid ||
 560                     old->egid == rgid ||
 561                     nsown_capable(CAP_SETGID))
 562                         new->gid = rgid;
 563                 else
 564                         goto error;
 565         }
 566         if (egid != (gid_t) -1) {
 567                 if (old->gid == egid ||
 568                     old->egid == egid ||
 569                     old->sgid == egid ||
 570                     nsown_capable(CAP_SETGID))
 571                         new->egid = egid;
 572                 else
 573                         goto error;
 574         }
 575
 576         if (rgid != (gid_t) -1 ||
 577             (egid != (gid_t) -1 && egid != old->gid))
 578                 new->sgid = new->egid;
 579         new->fsgid = new->egid;
 580
 581         return commit_creds(new);
 582
 583 error:
 584         abort_creds(new);
 585         return retval;
 586 }
 587
 588 /*
 589  * setgid() is implemented like SysV w/ SAVED_IDS
 590  *
 591  * SMP: Same implicit races as above.
 592  */
 593 SYSCALL_DEFINE1(setgid, gid_t, gid)
 594 {
 595         const struct cred *old;
 596         struct cred *new;
 597         int retval;
 598
 599         new = prepare_creds();
 600         if (!new)
 601                 return -ENOMEM;
 602         old = current_cred();
 603
 604         retval = -EPERM;
 605         if (nsown_capable(CAP_SETGID))
 606                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 607         else if (gid == old->gid || gid == old->sgid)
 608                 new->egid = new->fsgid = gid;
 609         else
 610                 goto error;
 611
 612         return commit_creds(new);
 613
 614 error:
 615         abort_creds(new);
 616         return retval;
 617 }
 618
 619 /*
 620  * change the user struct in a credentials set to match the new UID
 621  */
 622 static int set_user(struct cred *new)
 623 {
 624         struct user_struct *new_user;
 625
 626         new_user = alloc_uid(current_user_ns(), new->uid);
 627         if (!new_user)
 628                 return -EAGAIN;
 629
 630         /*
 631          * We don't fail in case of NPROC limit excess here because too many
 632          * poorly written programs don't check set*uid() return code, assuming
 633          * it never fails if called by root.  We may still enforce NPROC limit
 634          * for programs doing set*uid()+execve() by harmlessly deferring the
 635          * failure to the execve() stage.
 636          */
 637         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 638                         new_user != INIT_USER)
 639                 current->flags |= PF_NPROC_EXCEEDED;
 640         else
 641                 current->flags &= ~PF_NPROC_EXCEEDED;
 642
 643         free_uid(new->user);
 644         new->user = new_user;
 645         return 0;
 646 }
 647
 648 /*
 649  * Unprivileged users may change the real uid to the effective uid
 650  * or vice versa.  (BSD-style)
 651  *
 652  * If you set the real uid at all, or set the effective uid to a value not
 653  * equal to the real uid, then the saved uid is set to the new effective uid.
 654  *
 655  * This makes it possible for a setuid program to completely drop its
 656  * privileges, which is often a useful assertion to make when you are doing
 657  * a security audit over a program.
 658  *
 659  * The general idea is that a program which uses just setreuid() will be
 660  * 100% compatible with BSD.  A program which uses just setuid() will be
 661  * 100% compatible with POSIX with saved IDs.
 662  */
 663 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 664 {
 665         const struct cred *old;
 666         struct cred *new;
 667         int retval;
 668
 669         new = prepare_creds();
 670         if (!new)
 671                 return -ENOMEM;
 672         old = current_cred();
 673
 674         retval = -EPERM;
 675         if (ruid != (uid_t) -1) {
 676                 new->uid = ruid;
 677                 if (old->uid != ruid &&
 678                     old->euid != ruid &&
 679                     !nsown_capable(CAP_SETUID))
 680                         goto error;
 681         }
 682
 683         if (euid != (uid_t) -1) {
 684                 new->euid = euid;
 685                 if (old->uid != euid &&
 686                     old->euid != euid &&
 687                     old->suid != euid &&
 688                     !nsown_capable(CAP_SETUID))
 689                         goto error;
 690         }
 691
 692         if (new->uid != old->uid) {
 693                 retval = set_user(new);
 694                 if (retval < 0)
 695                         goto error;
 696         }
 697         if (ruid != (uid_t) -1 ||
 698             (euid != (uid_t) -1 && euid != old->uid))
 699                 new->suid = new->euid;
 700         new->fsuid = new->euid;
 701
 702         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 703         if (retval < 0)
 704                 goto error;
 705
 706         return commit_creds(new);
 707
 708 error:
 709         abort_creds(new);
 710         return retval;
 711 }
 712
 713 /*
 714  * setuid() is implemented like SysV with SAVED_IDS
 715  *
 716  * Note that SAVED_ID's is deficient in that a setuid root program
 717  * like sendmail, for example, cannot set its uid to be a normal
 718  * user and then switch back, because if you're root, setuid() sets
 719  * the saved uid too.  If you don't like this, blame the bright people
 720  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 721  * will allow a root program to temporarily drop privileges and be able to
 722  * regain them by swapping the real and effective uid.
 723  */
 724 SYSCALL_DEFINE1(setuid, uid_t, uid)
 725 {
 726         const struct cred *old;
 727         struct cred *new;
 728         int retval;
 729
 730         new = prepare_creds();
 731         if (!new)
 732                 return -ENOMEM;
 733         old = current_cred();
 734
 735         retval = -EPERM;
 736         if (nsown_capable(CAP_SETUID)) {
 737                 new->suid = new->uid = uid;
 738                 if (uid != old->uid) {
 739                         retval = set_user(new);
 740                         if (retval < 0)
 741                                 goto error;
 742                 }
 743         } else if (uid != old->uid && uid != new->suid) {
 744                 goto error;
 745         }
 746
 747         new->fsuid = new->euid = uid;
 748
 749         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 750         if (retval < 0)
 751                 goto error;
 752
 753         return commit_creds(new);
 754
 755 error:
 756         abort_creds(new);
 757         return retval;
 758 }
 759
 760
 761 /*
 762  * This function implements a generic ability to update ruid, euid,
 763  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 764  */
 765 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 766 {
 767         const struct cred *old;
 768         struct cred *new;
 769         int retval;
 770
 771         new = prepare_creds();
 772         if (!new)
 773                 return -ENOMEM;
 774
 775         old = current_cred();
 776
 777         retval = -EPERM;
 778         if (!nsown_capable(CAP_SETUID)) {
 779                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 780                     ruid != old->euid  && ruid != old->suid)
 781                         goto error;
 782                 if (euid != (uid_t) -1 && euid != old->uid &&
 783                     euid != old->euid  && euid != old->suid)
 784                         goto error;
 785                 if (suid != (uid_t) -1 && suid != old->uid &&
 786                     suid != old->euid  && suid != old->suid)
 787                         goto error;
 788         }
 789
 790         if (ruid != (uid_t) -1) {
 791                 new->uid = ruid;
 792                 if (ruid != old->uid) {
 793                         retval = set_user(new);
 794                         if (retval < 0)
 795                                 goto error;
 796                 }
 797         }
 798         if (euid != (uid_t) -1)
 799                 new->euid = euid;
 800         if (suid != (uid_t) -1)
 801                 new->suid = suid;
 802         new->fsuid = new->euid;
 803
 804         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 805         if (retval < 0)
 806                 goto error;
 807
 808         return commit_creds(new);
 809
 810 error:
 811         abort_creds(new);
 812         return retval;
 813 }
 814
 815 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 816 {
 817         const struct cred *cred = current_cred();
 818         int retval;
 819
 820         if (!(retval   = put_user(cred->uid,  ruid)) &&
 821             !(retval   = put_user(cred->euid, euid)))
 822                 retval = put_user(cred->suid, suid);
 823
 824         return retval;
 825 }
 826
 827 /*
 828  * Same as above, but for rgid, egid, sgid.
 829  */
 830 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 831 {
 832         const struct cred *old;
 833         struct cred *new;
 834         int retval;
 835
 836         new = prepare_creds();
 837         if (!new)
 838                 return -ENOMEM;
 839         old = current_cred();
 840
 841         retval = -EPERM;
 842         if (!nsown_capable(CAP_SETGID)) {
 843                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 844                     rgid != old->egid  && rgid != old->sgid)
 845                         goto error;
 846                 if (egid != (gid_t) -1 && egid != old->gid &&
 847                     egid != old->egid  && egid != old->sgid)
 848                         goto error;
 849                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 850                     sgid != old->egid  && sgid != old->sgid)
 851                         goto error;
 852         }
 853
 854         if (rgid != (gid_t) -1)
 855                 new->gid = rgid;
 856         if (egid != (gid_t) -1)
 857                 new->egid = egid;
 858         if (sgid != (gid_t) -1)
 859                 new->sgid = sgid;
 860         new->fsgid = new->egid;
 861
 862         return commit_creds(new);
 863
 864 error:
 865         abort_creds(new);
 866         return retval;
 867 }
 868
 869 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 870 {
 871         const struct cred *cred = current_cred();
 872         int retval;
 873
 874         if (!(retval   = put_user(cred->gid,  rgid)) &&
 875             !(retval   = put_user(cred->egid, egid)))
 876                 retval = put_user(cred->sgid, sgid);
 877
 878         return retval;
 879 }
 880
 881
 882 /*
 883  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 884  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 885  * whatever uid it wants to). It normally shadows "euid", except when
 886  * explicitly set by setfsuid() or for access..
 887  */
 888 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 889 {
 890         const struct cred *old;
 891         struct cred *new;
 892         uid_t old_fsuid;
 893
 894         new = prepare_creds();
 895         if (!new)
 896                 return current_fsuid();
 897         old = current_cred();
 898         old_fsuid = old->fsuid;
 899
 900         if (uid == old->uid  || uid == old->euid  ||
 901             uid == old->suid || uid == old->fsuid ||
 902             nsown_capable(CAP_SETUID)) {
 903                 if (uid != old_fsuid) {
 904                         new->fsuid = uid;
 905                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 906                                 goto change_okay;
 907                 }
 908         }
 909
 910         abort_creds(new);
 911         return old_fsuid;
 912
 913 change_okay:
 914         commit_creds(new);
 915         return old_fsuid;
 916 }
 917
 918 /*
 919  * Samma på svenska..
 920  */
 921 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 922 {
 923         const struct cred *old;
 924         struct cred *new;
 925         gid_t old_fsgid;
 926
 927         new = prepare_creds();
 928         if (!new)
 929                 return current_fsgid();
 930         old = current_cred();
 931         old_fsgid = old->fsgid;
 932
 933         if (gid == old->gid  || gid == old->egid  ||
 934             gid == old->sgid || gid == old->fsgid ||
 935             nsown_capable(CAP_SETGID)) {
 936                 if (gid != old_fsgid) {
 937                         new->fsgid = gid;
 938                         goto change_okay;
 939                 }
 940         }
 941
 942         abort_creds(new);
 943         return old_fsgid;
 944
 945 change_okay:
 946         commit_creds(new);
 947         return old_fsgid;
 948 }
 949
 950 void do_sys_times(struct tms *tms)
 951 {
 952         cputime_t tgutime, tgstime, cutime, cstime;
 953
 954         spin_lock_irq(&current->sighand->siglock);
 955         thread_group_times(current, &tgutime, &tgstime);
 956         cutime = current->signal->cutime;
 957         cstime = current->signal->cstime;
 958         spin_unlock_irq(&current->sighand->siglock);
 959         tms->tms_utime = cputime_to_clock_t(tgutime);
 960         tms->tms_stime = cputime_to_clock_t(tgstime);
 961         tms->tms_cutime = cputime_to_clock_t(cutime);
 962         tms->tms_cstime = cputime_to_clock_t(cstime);
 963 }
 964
 965 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 966 {
 967         if (tbuf) {
 968                 struct tms tmp;
 969
 970                 do_sys_times(&tmp);
 971                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 972                         return -EFAULT;
 973         }
 974         force_successful_syscall_return();
 975         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 976 }
 977
 978 /*
 979  * This needs some heavy checking ...
 980  * I just haven't the stomach for it. I also don't fully
 981  * understand sessions/pgrp etc. Let somebody who does explain it.
 982  *
 983  * OK, I think I have the protection semantics right.... this is really
 984  * only important on a multi-user system anyway, to make sure one user
 985  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 986  *
 987  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 988  * LBT 04.03.94
 989  */
 990 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 991 {
 992         struct task_struct *p;
 993         struct task_struct *group_leader = current->group_leader;
 994         struct pid *pgrp;
 995         int err;
 996
 997         if (!pid)
 998                 pid = task_pid_vnr(group_leader);
 999         if (!pgid)
1000                 pgid = pid;
1001         if (pgid < 0)
1002                 return -EINVAL;
1003         rcu_read_lock();
1004
1005         /* From this point forward we keep holding onto the tasklist lock
1006          * so that our parent does not change from under us. -DaveM
1007          */
1008         write_lock_irq(&tasklist_lock);
1009
1010         err = -ESRCH;
1011         p = find_task_by_vpid(pid);
1012         if (!p)
1013                 goto out;
1014
1015         err = -EINVAL;
1016         if (!thread_group_leader(p))
1017                 goto out;
1018
1019         if (same_thread_group(p->real_parent, group_leader)) {
1020                 err = -EPERM;
1021                 if (task_session(p) != task_session(group_leader))
1022                         goto out;
1023                 err = -EACCES;
1024                 if (p->did_exec)
1025                         goto out;
1026         } else {
1027                 err = -ESRCH;
1028                 if (p != group_leader)
1029                         goto out;
1030         }
1031
1032         err = -EPERM;
1033         if (p->signal->leader)
1034                 goto out;
1035
1036         pgrp = task_pid(p);
1037         if (pgid != pid) {
1038                 struct task_struct *g;
1039
1040                 pgrp = find_vpid(pgid);
1041                 g = pid_task(pgrp, PIDTYPE_PGID);
1042                 if (!g || task_session(g) != task_session(group_leader))
1043                         goto out;
1044         }
1045
1046         err = security_task_setpgid(p, pgid);
1047         if (err)
1048                 goto out;
1049
1050         if (task_pgrp(p) != pgrp)
1051                 change_pid(p, PIDTYPE_PGID, pgrp);
1052
1053         err = 0;
1054 out:
1055         /* All paths lead to here, thus we are safe. -DaveM */
1056         write_unlock_irq(&tasklist_lock);
1057         rcu_read_unlock();
1058         return err;
1059 }
1060
1061 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1062 {
1063         struct task_struct *p;
1064         struct pid *grp;
1065         int retval;
1066
1067         rcu_read_lock();
1068         if (!pid)
1069                 grp = task_pgrp(current);
1070         else {
1071                 retval = -ESRCH;
1072                 p = find_task_by_vpid(pid);
1073                 if (!p)
1074                         goto out;
1075                 grp = task_pgrp(p);
1076                 if (!grp)
1077                         goto out;
1078
1079                 retval = security_task_getpgid(p);
1080                 if (retval)
1081                         goto out;
1082         }
1083         retval = pid_vnr(grp);
1084 out:
1085         rcu_read_unlock();
1086         return retval;
1087 }
1088
1089 #ifdef __ARCH_WANT_SYS_GETPGRP
1090
1091 SYSCALL_DEFINE0(getpgrp)
1092 {
1093         return sys_getpgid(0);
1094 }
1095
1096 #endif
1097
1098 SYSCALL_DEFINE1(getsid, pid_t, pid)
1099 {
1100         struct task_struct *p;
1101         struct pid *sid;
1102         int retval;
1103
1104         rcu_read_lock();
1105         if (!pid)
1106                 sid = task_session(current);
1107         else {
1108                 retval = -ESRCH;
1109                 p = find_task_by_vpid(pid);
1110                 if (!p)
1111                         goto out;
1112                 sid = task_session(p);
1113                 if (!sid)
1114                         goto out;
1115
1116                 retval = security_task_getsid(p);
1117                 if (retval)
1118                         goto out;
1119         }
1120         retval = pid_vnr(sid);
1121 out:
1122         rcu_read_unlock();
1123         return retval;
1124 }
1125
1126 SYSCALL_DEFINE0(setsid)
1127 {
1128         struct task_struct *group_leader = current->group_leader;
1129         struct pid *sid = task_pid(group_leader);
1130         pid_t session = pid_vnr(sid);
1131         int err = -EPERM;
1132
1133         write_lock_irq(&tasklist_lock);
1134         /* Fail if I am already a session leader */
1135         if (group_leader->signal->leader)
1136                 goto out;
1137
1138         /* Fail if a process group id already exists that equals the
1139          * proposed session id.
1140          */
1141         if (pid_task(sid, PIDTYPE_PGID))
1142                 goto out;
1143
1144         group_leader->signal->leader = 1;
1145         __set_special_pids(sid);
1146
1147         proc_clear_tty(group_leader);
1148
1149         err = session;
1150 out:
1151         write_unlock_irq(&tasklist_lock);
1152         if (err > 0) {
1153                 proc_sid_connector(group_leader);
1154                 sched_autogroup_create_attach(group_leader);
1155         }
1156         return err;
1157 }
1158
1159 DECLARE_RWSEM(uts_sem);
1160
1161 #ifdef COMPAT_UTS_MACHINE
1162 #define override_architecture(name) \
1163         (personality(current->personality) == PER_LINUX32 && \
1164          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1165                       sizeof(COMPAT_UTS_MACHINE)))
1166 #else
1167 #define override_architecture(name)     0
1168 #endif
1169
1170 /*
1171  * Work around broken programs that cannot handle "Linux 3.0".
1172  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1173  */
1174 static int override_release(char __user *release, size_t len)
1175 {
1176         int ret = 0;
1177
1178         if (current->personality & UNAME26) {
1179                 const char *rest = UTS_RELEASE;
1180                 char buf[65] = { 0 };
1181                 int ndots = 0;
1182                 unsigned v;
1183                 size_t copy;
1184
1185                 while (*rest) {
1186                         if (*rest == '.' && ++ndots >= 3)
1187                                 break;
1188                         if (!isdigit(*rest) && *rest != '.')
1189                                 break;
1190                         rest++;
1191                 }
1192                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1193                 copy = min(sizeof(buf), max_t(size_t, 1, len));
1194                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1195                 ret = copy_to_user(release, buf, copy + 1);
1196         }
1197         return ret;
1198 }
1199
1200 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1201 {
1202         int errno = 0;
1203
1204         down_read(&uts_sem);
1205         if (copy_to_user(name, utsname(), sizeof *name))
1206                 errno = -EFAULT;
1207         up_read(&uts_sem);
1208
1209         if (!errno && override_release(name->release, sizeof(name->release)))
1210                 errno = -EFAULT;
1211         if (!errno && override_architecture(name))
1212                 errno = -EFAULT;
1213         return errno;
1214 }
1215
1216 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1217 /*
1218  * Old cruft
1219  */
1220 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1221 {
1222         int error = 0;
1223
1224         if (!name)
1225                 return -EFAULT;
1226
1227         down_read(&uts_sem);
1228         if (copy_to_user(name, utsname(), sizeof(*name)))
1229                 error = -EFAULT;
1230         up_read(&uts_sem);
1231
1232         if (!error && override_release(name->release, sizeof(name->release)))
1233                 error = -EFAULT;
1234         if (!error && override_architecture(name))
1235                 error = -EFAULT;
1236         return error;
1237 }
1238
1239 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1240 {
1241         int error;
1242
1243         if (!name)
1244                 return -EFAULT;
1245         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1246                 return -EFAULT;
1247
1248         down_read(&uts_sem);
1249         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1250                                __OLD_UTS_LEN);
1251         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1252         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1253                                 __OLD_UTS_LEN);
1254         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1255         error |= __copy_to_user(&name->release, &utsname()->release,
1256                                 __OLD_UTS_LEN);
1257         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1258         error |= __copy_to_user(&name->version, &utsname()->version,
1259                                 __OLD_UTS_LEN);
1260         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1261         error |= __copy_to_user(&name->machine, &utsname()->machine,
1262                                 __OLD_UTS_LEN);
1263         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1264         up_read(&uts_sem);
1265
1266         if (!error && override_architecture(name))
1267                 error = -EFAULT;
1268         if (!error && override_release(name->release, sizeof(name->release)))
1269                 error = -EFAULT;
1270         return error ? -EFAULT : 0;
1271 }
1272 #endif
1273
1274 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1275 {
1276         int errno;
1277         char tmp[__NEW_UTS_LEN];
1278
1279         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1280                 return -EPERM;
1281
1282         if (len < 0 || len > __NEW_UTS_LEN)
1283                 return -EINVAL;
1284         down_write(&uts_sem);
1285         errno = -EFAULT;
1286         if (!copy_from_user(tmp, name, len)) {
1287                 struct new_utsname *u = utsname();
1288
1289                 memcpy(u->nodename, tmp, len);
1290                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1291                 errno = 0;
1292         }
1293         uts_proc_notify(UTS_PROC_HOSTNAME);
1294         up_write(&uts_sem);
1295         return errno;
1296 }
1297
1298 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1299
1300 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1301 {
1302         int i, errno;
1303         struct new_utsname *u;
1304
1305         if (len < 0)
1306                 return -EINVAL;
1307         down_read(&uts_sem);
1308         u = utsname();
1309         i = 1 + strlen(u->nodename);
1310         if (i > len)
1311                 i = len;
1312         errno = 0;
1313         if (copy_to_user(name, u->nodename, i))
1314                 errno = -EFAULT;
1315         up_read(&uts_sem);
1316         return errno;
1317 }
1318
1319 #endif
1320
1321 /*
1322  * Only setdomainname; getdomainname can be implemented by calling
1323  * uname()
1324  */
1325 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1326 {
1327         int errno;
1328         char tmp[__NEW_UTS_LEN];
1329
1330         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1331                 return -EPERM;
1332         if (len < 0 || len > __NEW_UTS_LEN)
1333                 return -EINVAL;
1334
1335         down_write(&uts_sem);
1336         errno = -EFAULT;
1337         if (!copy_from_user(tmp, name, len)) {
1338                 struct new_utsname *u = utsname();
1339
1340                 memcpy(u->domainname, tmp, len);
1341                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1342                 errno = 0;
1343         }
1344         uts_proc_notify(UTS_PROC_DOMAINNAME);
1345         up_write(&uts_sem);
1346         return errno;
1347 }
1348
1349 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1350 {
1351         struct rlimit value;
1352         int ret;
1353
1354         ret = do_prlimit(current, resource, NULL, &value);
1355         if (!ret)
1356                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1357
1358         return ret;
1359 }
1360
1361 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1362
1363 /*
1364  *      Back compatibility for getrlimit. Needed for some apps.
1365  */
1366
1367 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1368                 struct rlimit __user *, rlim)
1369 {
1370         struct rlimit x;
1371         if (resource >= RLIM_NLIMITS)
1372                 return -EINVAL;
1373
1374         task_lock(current->group_leader);
1375         x = current->signal->rlim[resource];
1376         task_unlock(current->group_leader);
1377         if (x.rlim_cur > 0x7FFFFFFF)
1378                 x.rlim_cur = 0x7FFFFFFF;
1379         if (x.rlim_max > 0x7FFFFFFF)
1380                 x.rlim_max = 0x7FFFFFFF;
1381         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1382 }
1383
1384 #endif
1385
1386 static inline bool rlim64_is_infinity(__u64 rlim64)
1387 {
1388 #if BITS_PER_LONG < 64
1389         return rlim64 >= ULONG_MAX;
1390 #else
1391         return rlim64 == RLIM64_INFINITY;
1392 #endif
1393 }
1394
1395 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1396 {
1397         if (rlim->rlim_cur == RLIM_INFINITY)
1398                 rlim64->rlim_cur = RLIM64_INFINITY;
1399         else
1400                 rlim64->rlim_cur = rlim->rlim_cur;
1401         if (rlim->rlim_max == RLIM_INFINITY)
1402                 rlim64->rlim_max = RLIM64_INFINITY;
1403         else
1404                 rlim64->rlim_max = rlim->rlim_max;
1405 }
1406
1407 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1408 {
1409         if (rlim64_is_infinity(rlim64->rlim_cur))
1410                 rlim->rlim_cur = RLIM_INFINITY;
1411         else
1412                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1413         if (rlim64_is_infinity(rlim64->rlim_max))
1414                 rlim->rlim_max = RLIM_INFINITY;
1415         else
1416                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1417 }
1418
1419 /* make sure you are allowed to change @tsk limits before calling this */
1420 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1421                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1422 {
1423         struct rlimit *rlim;
1424         int retval = 0;
1425
1426         if (resource >= RLIM_NLIMITS)
1427                 return -EINVAL;
1428         if (new_rlim) {
1429                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1430                         return -EINVAL;
1431                 if (resource == RLIMIT_NOFILE &&
1432                                 new_rlim->rlim_max > sysctl_nr_open)
1433                         return -EPERM;
1434         }
1435
1436         /* protect tsk->signal and tsk->sighand from disappearing */
1437         read_lock(&tasklist_lock);
1438         if (!tsk->sighand) {
1439                 retval = -ESRCH;
1440                 goto out;
1441         }
1442
1443         rlim = tsk->signal->rlim + resource;
1444         task_lock(tsk->group_leader);
1445         if (new_rlim) {
1446                 /* Keep the capable check against init_user_ns until
1447                    cgroups can contain all limits */
1448                 if (new_rlim->rlim_max > rlim->rlim_max &&
1449                                 !capable(CAP_SYS_RESOURCE))
1450                         retval = -EPERM;
1451                 if (!retval)
1452                         retval = security_task_setrlimit(tsk->group_leader,
1453                                         resource, new_rlim);
1454                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1455                         /*
1456                          * The caller is asking for an immediate RLIMIT_CPU
1457                          * expiry.  But we use the zero value to mean "it was
1458                          * never set".  So let's cheat and make it one second
1459                          * instead
1460                          */
1461                         new_rlim->rlim_cur = 1;
1462                 }
1463         }
1464         if (!retval) {
1465                 if (old_rlim)
1466                         *old_rlim = *rlim;
1467                 if (new_rlim)
1468                         *rlim = *new_rlim;
1469         }
1470         task_unlock(tsk->group_leader);
1471
1472         /*
1473          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1474          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1475          * very long-standing error, and fixing it now risks breakage of
1476          * applications, so we live with it
1477          */
1478          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1479                          new_rlim->rlim_cur != RLIM_INFINITY)
1480                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1481 out:
1482         read_unlock(&tasklist_lock);
1483         return retval;
1484 }
1485
1486 /* rcu lock must be held */
1487 static int check_prlimit_permission(struct task_struct *task)
1488 {
1489         const struct cred *cred = current_cred(), *tcred;
1490
1491         if (current == task)
1492                 return 0;
1493
1494         tcred = __task_cred(task);
1495         if (cred->user->user_ns == tcred->user->user_ns &&
1496             (cred->uid == tcred->euid &&
1497              cred->uid == tcred->suid &&
1498              cred->uid == tcred->uid  &&
1499              cred->gid == tcred->egid &&
1500              cred->gid == tcred->sgid &&
1501              cred->gid == tcred->gid))
1502                 return 0;
1503         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1504                 return 0;
1505
1506         return -EPERM;
1507 }
1508
1509 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1510                 const struct rlimit64 __user *, new_rlim,
1511                 struct rlimit64 __user *, old_rlim)
1512 {
1513         struct rlimit64 old64, new64;
1514         struct rlimit old, new;
1515         struct task_struct *tsk;
1516         int ret;
1517
1518         if (new_rlim) {
1519                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1520                         return -EFAULT;
1521                 rlim64_to_rlim(&new64, &new);
1522         }
1523
1524         rcu_read_lock();
1525         tsk = pid ? find_task_by_vpid(pid) : current;
1526         if (!tsk) {
1527                 rcu_read_unlock();
1528                 return -ESRCH;
1529         }
1530         ret = check_prlimit_permission(tsk);
1531         if (ret) {
1532                 rcu_read_unlock();
1533                 return ret;
1534         }
1535         get_task_struct(tsk);
1536         rcu_read_unlock();
1537
1538         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1539                         old_rlim ? &old : NULL);
1540
1541         if (!ret && old_rlim) {
1542                 rlim_to_rlim64(&old, &old64);
1543                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1544                         ret = -EFAULT;
1545         }
1546
1547         put_task_struct(tsk);
1548         return ret;
1549 }
1550
1551 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1552 {
1553         struct rlimit new_rlim;
1554
1555         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1556                 return -EFAULT;
1557         return do_prlimit(current, resource, &new_rlim, NULL);
1558 }
1559
1560 /*
1561  * It would make sense to put struct rusage in the task_struct,
1562  * except that would make the task_struct be *really big*.  After
1563  * task_struct gets moved into malloc'ed memory, it would
1564  * make sense to do this.  It will make moving the rest of the information
1565  * a lot simpler!  (Which we're not doing right now because we're not
1566  * measuring them yet).
1567  *
1568  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1569  * races with threads incrementing their own counters.  But since word
1570  * reads are atomic, we either get new values or old values and we don't
1571  * care which for the sums.  We always take the siglock to protect reading
1572  * the c* fields from p->signal from races with exit.c updating those
1573  * fields when reaping, so a sample either gets all the additions of a
1574  * given child after it's reaped, or none so this sample is before reaping.
1575  *
1576  * Locking:
1577  * We need to take the siglock for CHILDEREN, SELF and BOTH
1578  * for  the cases current multithreaded, non-current single threaded
1579  * non-current multithreaded.  Thread traversal is now safe with
1580  * the siglock held.
1581  * Strictly speaking, we donot need to take the siglock if we are current and
1582  * single threaded,  as no one else can take our signal_struct away, no one
1583  * else can  reap the  children to update signal->c* counters, and no one else
1584  * can race with the signal-> fields. If we do not take any lock, the
1585  * signal-> fields could be read out of order while another thread was just
1586  * exiting. So we should  place a read memory barrier when we avoid the lock.
1587  * On the writer side,  write memory barrier is implied in  __exit_signal
1588  * as __exit_signal releases  the siglock spinlock after updating the signal->
1589  * fields. But we don't do this yet to keep things simple.
1590  *
1591  */
1592
1593 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1594 {
1595         r->ru_nvcsw += t->nvcsw;
1596         r->ru_nivcsw += t->nivcsw;
1597         r->ru_minflt += t->min_flt;
1598         r->ru_majflt += t->maj_flt;
1599         r->ru_inblock += task_io_get_inblock(t);
1600         r->ru_oublock += task_io_get_oublock(t);
1601 }
1602
1603 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1604 {
1605         struct task_struct *t;
1606         unsigned long flags;
1607         cputime_t tgutime, tgstime, utime, stime;
1608         unsigned long maxrss = 0;
1609
1610         memset((char *) r, 0, sizeof *r);
1611         utime = stime = cputime_zero;
1612
1613         if (who == RUSAGE_THREAD) {
1614                 task_times(current, &utime, &stime);
1615                 accumulate_thread_rusage(p, r);
1616                 maxrss = p->signal->maxrss;
1617                 goto out;
1618         }
1619
1620         if (!lock_task_sighand(p, &flags))
1621                 return;
1622
1623         switch (who) {
1624                 case RUSAGE_BOTH:
1625                 case RUSAGE_CHILDREN:
1626                         utime = p->signal->cutime;
1627                         stime = p->signal->cstime;
1628                         r->ru_nvcsw = p->signal->cnvcsw;
1629                         r->ru_nivcsw = p->signal->cnivcsw;
1630                         r->ru_minflt = p->signal->cmin_flt;
1631                         r->ru_majflt = p->signal->cmaj_flt;
1632                         r->ru_inblock = p->signal->cinblock;
1633                         r->ru_oublock = p->signal->coublock;
1634                         maxrss = p->signal->cmaxrss;
1635
1636                         if (who == RUSAGE_CHILDREN)
1637                                 break;
1638
1639                 case RUSAGE_SELF:
1640                         thread_group_times(p, &tgutime, &tgstime);
1641                         utime = cputime_add(utime, tgutime);
1642                         stime = cputime_add(stime, tgstime);
1643                         r->ru_nvcsw += p->signal->nvcsw;
1644                         r->ru_nivcsw += p->signal->nivcsw;
1645                         r->ru_minflt += p->signal->min_flt;
1646                         r->ru_majflt += p->signal->maj_flt;
1647                         r->ru_inblock += p->signal->inblock;
1648                         r->ru_oublock += p->signal->oublock;
1649                         if (maxrss < p->signal->maxrss)
1650                                 maxrss = p->signal->maxrss;
1651                         t = p;
1652                         do {
1653                                 accumulate_thread_rusage(t, r);
1654                                 t = next_thread(t);
1655                         } while (t != p);
1656                         break;
1657
1658                 default:
1659                         BUG();
1660         }
1661         unlock_task_sighand(p, &flags);
1662
1663 out:
1664         cputime_to_timeval(utime, &r->ru_utime);
1665         cputime_to_timeval(stime, &r->ru_stime);
1666
1667         if (who != RUSAGE_CHILDREN) {
1668                 struct mm_struct *mm = get_task_mm(p);
1669                 if (mm) {
1670                         setmax_mm_hiwater_rss(&maxrss, mm);
1671                         mmput(mm);
1672                 }
1673         }
1674         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1675 }
1676
1677 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1678 {
1679         struct rusage r;
1680         k_getrusage(p, who, &r);
1681         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1682 }
1683
1684 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1685 {
1686         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1687             who != RUSAGE_THREAD)
1688                 return -EINVAL;
1689         return getrusage(current, who, ru);
1690 }
1691
1692 SYSCALL_DEFINE1(umask, int, mask)
1693 {
1694         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1695         return mask;
1696 }
1697
1698 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1699                 unsigned long, arg4, unsigned long, arg5)
1700 {
1701         struct task_struct *me = current;
1702         unsigned char comm[sizeof(me->comm)];
1703         long error;
1704
1705         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1706         if (error != -ENOSYS)
1707                 return error;
1708
1709         error = 0;
1710         switch (option) {
1711                 case PR_SET_PDEATHSIG:
1712                         if (!valid_signal(arg2)) {
1713                                 error = -EINVAL;
1714                                 break;
1715                         }
1716                         me->pdeath_signal = arg2;
1717                         error = 0;
1718                         break;
1719                 case PR_GET_PDEATHSIG:
1720                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1721                         break;
1722                 case PR_GET_DUMPABLE:
1723                         error = get_dumpable(me->mm);
1724                         break;
1725                 case PR_SET_DUMPABLE:
1726                         if (arg2 < 0 || arg2 > 1) {
1727                                 error = -EINVAL;
1728                                 break;
1729                         }
1730                         set_dumpable(me->mm, arg2);
1731                         error = 0;
1732                         break;
1733
1734                 case PR_SET_UNALIGN:
1735                         error = SET_UNALIGN_CTL(me, arg2);
1736                         break;
1737                 case PR_GET_UNALIGN:
1738                         error = GET_UNALIGN_CTL(me, arg2);
1739                         break;
1740                 case PR_SET_FPEMU:
1741                         error = SET_FPEMU_CTL(me, arg2);
1742                         break;
1743                 case PR_GET_FPEMU:
1744                         error = GET_FPEMU_CTL(me, arg2);
1745                         break;
1746                 case PR_SET_FPEXC:
1747                         error = SET_FPEXC_CTL(me, arg2);
1748                         break;
1749                 case PR_GET_FPEXC:
1750                         error = GET_FPEXC_CTL(me, arg2);
1751                         break;
1752                 case PR_GET_TIMING:
1753                         error = PR_TIMING_STATISTICAL;
1754                         break;
1755                 case PR_SET_TIMING:
1756                         if (arg2 != PR_TIMING_STATISTICAL)
1757                                 error = -EINVAL;
1758                         else
1759                                 error = 0;
1760                         break;
1761
1762                 case PR_SET_NAME:
1763                         comm[sizeof(me->comm)-1] = 0;
1764                         if (strncpy_from_user(comm, (char __user *)arg2,
1765                                               sizeof(me->comm) - 1) < 0)
1766                                 return -EFAULT;
1767                         set_task_comm(me, comm);
1768                         proc_comm_connector(me);
1769                         return 0;
1770                 case PR_GET_NAME:
1771                         get_task_comm(comm, me);
1772                         if (copy_to_user((char __user *)arg2, comm,
1773                                          sizeof(comm)))
1774                                 return -EFAULT;
1775                         return 0;
1776                 case PR_GET_ENDIAN:
1777                         error = GET_ENDIAN(me, arg2);
1778                         break;
1779                 case PR_SET_ENDIAN:
1780                         error = SET_ENDIAN(me, arg2);
1781                         break;
1782
1783                 case PR_GET_SECCOMP:
1784                         error = prctl_get_seccomp();
1785                         break;
1786                 case PR_SET_SECCOMP:
1787                         error = prctl_set_seccomp(arg2);
1788                         break;
1789                 case PR_GET_TSC:
1790                         error = GET_TSC_CTL(arg2);
1791                         break;
1792                 case PR_SET_TSC:
1793                         error = SET_TSC_CTL(arg2);
1794                         break;
1795                 case PR_TASK_PERF_EVENTS_DISABLE:
1796                         error = perf_event_task_disable();
1797                         break;
1798                 case PR_TASK_PERF_EVENTS_ENABLE:
1799                         error = perf_event_task_enable();
1800                         break;
1801                 case PR_GET_TIMERSLACK:
1802                         error = current->timer_slack_ns;
1803                         break;
1804                 case PR_SET_TIMERSLACK:
1805                         if (arg2 <= 0)
1806                                 current->timer_slack_ns =
1807                                         current->default_timer_slack_ns;
1808                         else
1809                                 current->timer_slack_ns = arg2;
1810                         error = 0;
1811                         break;
1812                 case PR_MCE_KILL:
1813                         if (arg4 | arg5)
1814                                 return -EINVAL;
1815                         switch (arg2) {
1816                         case PR_MCE_KILL_CLEAR:
1817                                 if (arg3 != 0)
1818                                         return -EINVAL;
1819                                 current->flags &= ~PF_MCE_PROCESS;
1820                                 break;
1821                         case PR_MCE_KILL_SET:
1822                                 current->flags |= PF_MCE_PROCESS;
1823                                 if (arg3 == PR_MCE_KILL_EARLY)
1824                                         current->flags |= PF_MCE_EARLY;
1825                                 else if (arg3 == PR_MCE_KILL_LATE)
1826                                         current->flags &= ~PF_MCE_EARLY;
1827                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1828                                         current->flags &=
1829                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1830                                 else
1831                                         return -EINVAL;
1832                                 break;
1833                         default:
1834                                 return -EINVAL;
1835                         }
1836                         error = 0;
1837                         break;
1838                 case PR_MCE_KILL_GET:
1839                         if (arg2 | arg3 | arg4 | arg5)
1840                                 return -EINVAL;
1841                         if (current->flags & PF_MCE_PROCESS)
1842                                 error = (current->flags & PF_MCE_EARLY) ?
1843                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1844                         else
1845                                 error = PR_MCE_KILL_DEFAULT;
1846                         break;
1847                 default:
1848                         error = -EINVAL;
1849                         break;
1850         }
1851         return error;
1852 }
1853
1854 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1855                 struct getcpu_cache __user *, unused)
1856 {
1857         int err = 0;
1858         int cpu = raw_smp_processor_id();
1859         if (cpup)
1860                 err |= put_user(cpu, cpup);
1861         if (nodep)
1862                 err |= put_user(cpu_to_node(cpu), nodep);
1863         return err ? -EFAULT : 0;
1864 }
1865
1866 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1867
1868 static void argv_cleanup(struct subprocess_info *info)
1869 {
1870         argv_free(info->argv);
1871 }
1872
1873 /**
1874  * orderly_poweroff - Trigger an orderly system poweroff
1875  * @force: force poweroff if command execution fails
1876  *
1877  * This may be called from any context to trigger a system shutdown.
1878  * If the orderly shutdown fails, it will force an immediate shutdown.
1879  */
1880 int orderly_poweroff(bool force)
1881 {
1882         int argc;
1883         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1884         static char *envp[] = {
1885                 "HOME=/",
1886                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1887                 NULL
1888         };
1889         int ret = -ENOMEM;
1890         struct subprocess_info *info;
1891
1892         if (argv == NULL) {
1893                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1894                        __func__, poweroff_cmd);
1895                 goto out;
1896         }
1897
1898         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1899         if (info == NULL) {
1900                 argv_free(argv);
1901                 goto out;
1902         }
1903
1904         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1905
1906         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1907
1908   out:
1909         if (ret && force) {
1910                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1911                        "forcing the issue\n");
1912
1913                 /* I guess this should try to kick off some daemon to
1914                    sync and poweroff asap.  Or not even bother syncing
1915                    if we're doing an emergency shutdown? */
1916                 emergency_sync();
1917                 kernel_power_off();
1918         }
1919
1920         return ret;
1921 }
1922 EXPORT_SYMBOL_GPL(orderly_poweroff);