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         if (!cmd)
 369                 printk(KERN_EMERG "Restarting system.\n");
 370         else
 371                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 372         kmsg_dump(KMSG_DUMP_RESTART);
 373         machine_restart(cmd);
 374 }
 375 EXPORT_SYMBOL_GPL(kernel_restart);
 376
 377 static void kernel_shutdown_prepare(enum system_states state)
 378 {
 379         blocking_notifier_call_chain(&reboot_notifier_list,
 380                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 381         system_state = state;
 382         usermodehelper_disable();
 383         device_shutdown();
 384 }
 385 /**
 386  *      kernel_halt - halt the system
 387  *
 388  *      Shutdown everything and perform a clean system halt.
 389  */
 390 void kernel_halt(void)
 391 {
 392         kernel_shutdown_prepare(SYSTEM_HALT);
 393         syscore_shutdown();
 394         printk(KERN_EMERG "System halted.\n");
 395         kmsg_dump(KMSG_DUMP_HALT);
 396         machine_halt();
 397 }
 398
 399 EXPORT_SYMBOL_GPL(kernel_halt);
 400
 401 /**
 402  *      kernel_power_off - power_off the system
 403  *
 404  *      Shutdown everything and perform a clean system power_off.
 405  */
 406 void kernel_power_off(void)
 407 {
 408         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 409         if (pm_power_off_prepare)
 410                 pm_power_off_prepare();
 411         disable_nonboot_cpus();
 412         syscore_shutdown();
 413         printk(KERN_EMERG "Power down.\n");
 414         kmsg_dump(KMSG_DUMP_POWEROFF);
 415         machine_power_off();
 416 }
 417 EXPORT_SYMBOL_GPL(kernel_power_off);
 418
 419 static DEFINE_MUTEX(reboot_mutex);
 420
 421 /*
 422  * Reboot system call: for obvious reasons only root may call it,
 423  * and even root needs to set up some magic numbers in the registers
 424  * so that some mistake won't make this reboot the whole machine.
 425  * You can also set the meaning of the ctrl-alt-del-key here.
 426  *
 427  * reboot doesn't sync: do that yourself before calling this.
 428  */
 429 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 430                 void __user *, arg)
 431 {
 432         char buffer[256];
 433         int ret = 0;
 434
 435         /* We only trust the superuser with rebooting the system. */
 436         if (!capable(CAP_SYS_BOOT))
 437                 return -EPERM;
 438
 439         /* For safety, we require "magic" arguments. */
 440         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 441             (magic2 != LINUX_REBOOT_MAGIC2 &&
 442                         magic2 != LINUX_REBOOT_MAGIC2A &&
 443                         magic2 != LINUX_REBOOT_MAGIC2B &&
 444                         magic2 != LINUX_REBOOT_MAGIC2C))
 445                 return -EINVAL;
 446
 447         /*
 448          * If pid namespaces are enabled and the current task is in a child
 449          * pid_namespace, the command is handled by reboot_pid_ns() which will
 450          * call do_exit().
 451          */
 452         ret = reboot_pid_ns(task_active_pid_ns(current), cmd);
 453         if (ret)
 454                 return ret;
 455
 456         /* Instead of trying to make the power_off code look like
 457          * halt when pm_power_off is not set do it the easy way.
 458          */
 459         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 460                 cmd = LINUX_REBOOT_CMD_HALT;
 461
 462         mutex_lock(&reboot_mutex);
 463         switch (cmd) {
 464         case LINUX_REBOOT_CMD_RESTART:
 465                 kernel_restart(NULL);
 466                 break;
 467
 468         case LINUX_REBOOT_CMD_CAD_ON:
 469                 C_A_D = 1;
 470                 break;
 471
 472         case LINUX_REBOOT_CMD_CAD_OFF:
 473                 C_A_D = 0;
 474                 break;
 475
 476         case LINUX_REBOOT_CMD_HALT:
 477                 kernel_halt();
 478                 do_exit(0);
 479                 panic("cannot halt");
 480
 481         case LINUX_REBOOT_CMD_POWER_OFF:
 482                 kernel_power_off();
 483                 do_exit(0);
 484                 break;
 485
 486         case LINUX_REBOOT_CMD_RESTART2:
 487                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 488                         ret = -EFAULT;
 489                         break;
 490                 }
 491                 buffer[sizeof(buffer) - 1] = '\0';
 492
 493                 kernel_restart(buffer);
 494                 break;
 495
 496 #ifdef CONFIG_KEXEC
 497         case LINUX_REBOOT_CMD_KEXEC:
 498                 ret = kernel_kexec();
 499                 break;
 500 #endif
 501
 502 #ifdef CONFIG_HIBERNATION
 503         case LINUX_REBOOT_CMD_SW_SUSPEND:
 504                 ret = hibernate();
 505                 break;
 506 #endif
 507
 508         default:
 509                 ret = -EINVAL;
 510                 break;
 511         }
 512         mutex_unlock(&reboot_mutex);
 513         return ret;
 514 }
 515
 516 static void deferred_cad(struct work_struct *dummy)
 517 {
 518         kernel_restart(NULL);
 519 }
 520
 521 /*
 522  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 523  * As it's called within an interrupt, it may NOT sync: the only choice
 524  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 525  */
 526 void ctrl_alt_del(void)
 527 {
 528         static DECLARE_WORK(cad_work, deferred_cad);
 529
 530         if (C_A_D)
 531                 schedule_work(&cad_work);
 532         else
 533                 kill_cad_pid(SIGINT, 1);
 534 }
 535
 536 /*
 537  * Unprivileged users may change the real gid to the effective gid
 538  * or vice versa.  (BSD-style)
 539  *
 540  * If you set the real gid at all, or set the effective gid to a value not
 541  * equal to the real gid, then the saved gid is set to the new effective gid.
 542  *
 543  * This makes it possible for a setgid program to completely drop its
 544  * privileges, which is often a useful assertion to make when you are doing
 545  * a security audit over a program.
 546  *
 547  * The general idea is that a program which uses just setregid() will be
 548  * 100% compatible with BSD.  A program which uses just setgid() will be
 549  * 100% compatible with POSIX with saved IDs.
 550  *
 551  * SMP: There are not races, the GIDs are checked only by filesystem
 552  *      operations (as far as semantic preservation is concerned).
 553  */
 554 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 555 {
 556         const struct cred *old;
 557         struct cred *new;
 558         int retval;
 559
 560         new = prepare_creds();
 561         if (!new)
 562                 return -ENOMEM;
 563         old = current_cred();
 564
 565         retval = -EPERM;
 566         if (rgid != (gid_t) -1) {
 567                 if (old->gid == rgid ||
 568                     old->egid == rgid ||
 569                     nsown_capable(CAP_SETGID))
 570                         new->gid = rgid;
 571                 else
 572                         goto error;
 573         }
 574         if (egid != (gid_t) -1) {
 575                 if (old->gid == egid ||
 576                     old->egid == egid ||
 577                     old->sgid == egid ||
 578                     nsown_capable(CAP_SETGID))
 579                         new->egid = egid;
 580                 else
 581                         goto error;
 582         }
 583
 584         if (rgid != (gid_t) -1 ||
 585             (egid != (gid_t) -1 && egid != old->gid))
 586                 new->sgid = new->egid;
 587         new->fsgid = new->egid;
 588
 589         return commit_creds(new);
 590
 591 error:
 592         abort_creds(new);
 593         return retval;
 594 }
 595
 596 /*
 597  * setgid() is implemented like SysV w/ SAVED_IDS
 598  *
 599  * SMP: Same implicit races as above.
 600  */
 601 SYSCALL_DEFINE1(setgid, gid_t, gid)
 602 {
 603         const struct cred *old;
 604         struct cred *new;
 605         int retval;
 606
 607         new = prepare_creds();
 608         if (!new)
 609                 return -ENOMEM;
 610         old = current_cred();
 611
 612         retval = -EPERM;
 613         if (nsown_capable(CAP_SETGID))
 614                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 615         else if (gid == old->gid || gid == old->sgid)
 616                 new->egid = new->fsgid = gid;
 617         else
 618                 goto error;
 619
 620         return commit_creds(new);
 621
 622 error:
 623         abort_creds(new);
 624         return retval;
 625 }
 626
 627 /*
 628  * change the user struct in a credentials set to match the new UID
 629  */
 630 static int set_user(struct cred *new)
 631 {
 632         struct user_struct *new_user;
 633
 634         new_user = alloc_uid(current_user_ns(), new->uid);
 635         if (!new_user)
 636                 return -EAGAIN;
 637
 638         /*
 639          * We don't fail in case of NPROC limit excess here because too many
 640          * poorly written programs don't check set*uid() return code, assuming
 641          * it never fails if called by root.  We may still enforce NPROC limit
 642          * for programs doing set*uid()+execve() by harmlessly deferring the
 643          * failure to the execve() stage.
 644          */
 645         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 646                         new_user != INIT_USER)
 647                 current->flags |= PF_NPROC_EXCEEDED;
 648         else
 649                 current->flags &= ~PF_NPROC_EXCEEDED;
 650
 651         free_uid(new->user);
 652         new->user = new_user;
 653         return 0;
 654 }
 655
 656 /*
 657  * Unprivileged users may change the real uid to the effective uid
 658  * or vice versa.  (BSD-style)
 659  *
 660  * If you set the real uid at all, or set the effective uid to a value not
 661  * equal to the real uid, then the saved uid is set to the new effective uid.
 662  *
 663  * This makes it possible for a setuid program to completely drop its
 664  * privileges, which is often a useful assertion to make when you are doing
 665  * a security audit over a program.
 666  *
 667  * The general idea is that a program which uses just setreuid() will be
 668  * 100% compatible with BSD.  A program which uses just setuid() will be
 669  * 100% compatible with POSIX with saved IDs.
 670  */
 671 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 672 {
 673         const struct cred *old;
 674         struct cred *new;
 675         int retval;
 676
 677         new = prepare_creds();
 678         if (!new)
 679                 return -ENOMEM;
 680         old = current_cred();
 681
 682         retval = -EPERM;
 683         if (ruid != (uid_t) -1) {
 684                 new->uid = ruid;
 685                 if (old->uid != ruid &&
 686                     old->euid != ruid &&
 687                     !nsown_capable(CAP_SETUID))
 688                         goto error;
 689         }
 690
 691         if (euid != (uid_t) -1) {
 692                 new->euid = euid;
 693                 if (old->uid != euid &&
 694                     old->euid != euid &&
 695                     old->suid != euid &&
 696                     !nsown_capable(CAP_SETUID))
 697                         goto error;
 698         }
 699
 700         if (new->uid != old->uid) {
 701                 retval = set_user(new);
 702                 if (retval < 0)
 703                         goto error;
 704         }
 705         if (ruid != (uid_t) -1 ||
 706             (euid != (uid_t) -1 && euid != old->uid))
 707                 new->suid = new->euid;
 708         new->fsuid = new->euid;
 709
 710         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 711         if (retval < 0)
 712                 goto error;
 713
 714         return commit_creds(new);
 715
 716 error:
 717         abort_creds(new);
 718         return retval;
 719 }
 720
 721 /*
 722  * setuid() is implemented like SysV with SAVED_IDS
 723  *
 724  * Note that SAVED_ID's is deficient in that a setuid root program
 725  * like sendmail, for example, cannot set its uid to be a normal
 726  * user and then switch back, because if you're root, setuid() sets
 727  * the saved uid too.  If you don't like this, blame the bright people
 728  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 729  * will allow a root program to temporarily drop privileges and be able to
 730  * regain them by swapping the real and effective uid.
 731  */
 732 SYSCALL_DEFINE1(setuid, uid_t, uid)
 733 {
 734         const struct cred *old;
 735         struct cred *new;
 736         int retval;
 737
 738         new = prepare_creds();
 739         if (!new)
 740                 return -ENOMEM;
 741         old = current_cred();
 742
 743         retval = -EPERM;
 744         if (nsown_capable(CAP_SETUID)) {
 745                 new->suid = new->uid = uid;
 746                 if (uid != old->uid) {
 747                         retval = set_user(new);
 748                         if (retval < 0)
 749                                 goto error;
 750                 }
 751         } else if (uid != old->uid && uid != new->suid) {
 752                 goto error;
 753         }
 754
 755         new->fsuid = new->euid = uid;
 756
 757         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 758         if (retval < 0)
 759                 goto error;
 760
 761         return commit_creds(new);
 762
 763 error:
 764         abort_creds(new);
 765         return retval;
 766 }
 767
 768
 769 /*
 770  * This function implements a generic ability to update ruid, euid,
 771  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 772  */
 773 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 774 {
 775         const struct cred *old;
 776         struct cred *new;
 777         int retval;
 778
 779         new = prepare_creds();
 780         if (!new)
 781                 return -ENOMEM;
 782
 783         old = current_cred();
 784
 785         retval = -EPERM;
 786         if (!nsown_capable(CAP_SETUID)) {
 787                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 788                     ruid != old->euid  && ruid != old->suid)
 789                         goto error;
 790                 if (euid != (uid_t) -1 && euid != old->uid &&
 791                     euid != old->euid  && euid != old->suid)
 792                         goto error;
 793                 if (suid != (uid_t) -1 && suid != old->uid &&
 794                     suid != old->euid  && suid != old->suid)
 795                         goto error;
 796         }
 797
 798         if (ruid != (uid_t) -1) {
 799                 new->uid = ruid;
 800                 if (ruid != old->uid) {
 801                         retval = set_user(new);
 802                         if (retval < 0)
 803                                 goto error;
 804                 }
 805         }
 806         if (euid != (uid_t) -1)
 807                 new->euid = euid;
 808         if (suid != (uid_t) -1)
 809                 new->suid = suid;
 810         new->fsuid = new->euid;
 811
 812         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 813         if (retval < 0)
 814                 goto error;
 815
 816         return commit_creds(new);
 817
 818 error:
 819         abort_creds(new);
 820         return retval;
 821 }
 822
 823 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 824 {
 825         const struct cred *cred = current_cred();
 826         int retval;
 827
 828         if (!(retval   = put_user(cred->uid,  ruid)) &&
 829             !(retval   = put_user(cred->euid, euid)))
 830                 retval = put_user(cred->suid, suid);
 831
 832         return retval;
 833 }
 834
 835 /*
 836  * Same as above, but for rgid, egid, sgid.
 837  */
 838 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 839 {
 840         const struct cred *old;
 841         struct cred *new;
 842         int retval;
 843
 844         new = prepare_creds();
 845         if (!new)
 846                 return -ENOMEM;
 847         old = current_cred();
 848
 849         retval = -EPERM;
 850         if (!nsown_capable(CAP_SETGID)) {
 851                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 852                     rgid != old->egid  && rgid != old->sgid)
 853                         goto error;
 854                 if (egid != (gid_t) -1 && egid != old->gid &&
 855                     egid != old->egid  && egid != old->sgid)
 856                         goto error;
 857                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 858                     sgid != old->egid  && sgid != old->sgid)
 859                         goto error;
 860         }
 861
 862         if (rgid != (gid_t) -1)
 863                 new->gid = rgid;
 864         if (egid != (gid_t) -1)
 865                 new->egid = egid;
 866         if (sgid != (gid_t) -1)
 867                 new->sgid = sgid;
 868         new->fsgid = new->egid;
 869
 870         return commit_creds(new);
 871
 872 error:
 873         abort_creds(new);
 874         return retval;
 875 }
 876
 877 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 878 {
 879         const struct cred *cred = current_cred();
 880         int retval;
 881
 882         if (!(retval   = put_user(cred->gid,  rgid)) &&
 883             !(retval   = put_user(cred->egid, egid)))
 884                 retval = put_user(cred->sgid, sgid);
 885
 886         return retval;
 887 }
 888
 889
 890 /*
 891  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 892  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 893  * whatever uid it wants to). It normally shadows "euid", except when
 894  * explicitly set by setfsuid() or for access..
 895  */
 896 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 897 {
 898         const struct cred *old;
 899         struct cred *new;
 900         uid_t old_fsuid;
 901
 902         new = prepare_creds();
 903         if (!new)
 904                 return current_fsuid();
 905         old = current_cred();
 906         old_fsuid = old->fsuid;
 907
 908         if (uid == old->uid  || uid == old->euid  ||
 909             uid == old->suid || uid == old->fsuid ||
 910             nsown_capable(CAP_SETUID)) {
 911                 if (uid != old_fsuid) {
 912                         new->fsuid = uid;
 913                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 914                                 goto change_okay;
 915                 }
 916         }
 917
 918         abort_creds(new);
 919         return old_fsuid;
 920
 921 change_okay:
 922         commit_creds(new);
 923         return old_fsuid;
 924 }
 925
 926 /*
 927  * Samma på svenska..
 928  */
 929 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 930 {
 931         const struct cred *old;
 932         struct cred *new;
 933         gid_t old_fsgid;
 934
 935         new = prepare_creds();
 936         if (!new)
 937                 return current_fsgid();
 938         old = current_cred();
 939         old_fsgid = old->fsgid;
 940
 941         if (gid == old->gid  || gid == old->egid  ||
 942             gid == old->sgid || gid == old->fsgid ||
 943             nsown_capable(CAP_SETGID)) {
 944                 if (gid != old_fsgid) {
 945                         new->fsgid = gid;
 946                         goto change_okay;
 947                 }
 948         }
 949
 950         abort_creds(new);
 951         return old_fsgid;
 952
 953 change_okay:
 954         commit_creds(new);
 955         return old_fsgid;
 956 }
 957
 958 void do_sys_times(struct tms *tms)
 959 {
 960         cputime_t tgutime, tgstime, cutime, cstime;
 961
 962         spin_lock_irq(&current->sighand->siglock);
 963         thread_group_times(current, &tgutime, &tgstime);
 964         cutime = current->signal->cutime;
 965         cstime = current->signal->cstime;
 966         spin_unlock_irq(&current->sighand->siglock);
 967         tms->tms_utime = cputime_to_clock_t(tgutime);
 968         tms->tms_stime = cputime_to_clock_t(tgstime);
 969         tms->tms_cutime = cputime_to_clock_t(cutime);
 970         tms->tms_cstime = cputime_to_clock_t(cstime);
 971 }
 972
 973 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 974 {
 975         if (tbuf) {
 976                 struct tms tmp;
 977
 978                 do_sys_times(&tmp);
 979                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 980                         return -EFAULT;
 981         }
 982         force_successful_syscall_return();
 983         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 984 }
 985
 986 /*
 987  * This needs some heavy checking ...
 988  * I just haven't the stomach for it. I also don't fully
 989  * understand sessions/pgrp etc. Let somebody who does explain it.
 990  *
 991  * OK, I think I have the protection semantics right.... this is really
 992  * only important on a multi-user system anyway, to make sure one user
 993  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 994  *
 995  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 996  * LBT 04.03.94
 997  */
 998 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 999 {
1000         struct task_struct *p;
1001         struct task_struct *group_leader = current->group_leader;
1002         struct pid *pgrp;
1003         int err;
1004
1005         if (!pid)
1006                 pid = task_pid_vnr(group_leader);
1007         if (!pgid)
1008                 pgid = pid;
1009         if (pgid < 0)
1010                 return -EINVAL;
1011         rcu_read_lock();
1012
1013         /* From this point forward we keep holding onto the tasklist lock
1014          * so that our parent does not change from under us. -DaveM
1015          */
1016         write_lock_irq(&tasklist_lock);
1017
1018         err = -ESRCH;
1019         p = find_task_by_vpid(pid);
1020         if (!p)
1021                 goto out;
1022
1023         err = -EINVAL;
1024         if (!thread_group_leader(p))
1025                 goto out;
1026
1027         if (same_thread_group(p->real_parent, group_leader)) {
1028                 err = -EPERM;
1029                 if (task_session(p) != task_session(group_leader))
1030                         goto out;
1031                 err = -EACCES;
1032                 if (p->did_exec)
1033                         goto out;
1034         } else {
1035                 err = -ESRCH;
1036                 if (p != group_leader)
1037                         goto out;
1038         }
1039
1040         err = -EPERM;
1041         if (p->signal->leader)
1042                 goto out;
1043
1044         pgrp = task_pid(p);
1045         if (pgid != pid) {
1046                 struct task_struct *g;
1047
1048                 pgrp = find_vpid(pgid);
1049                 g = pid_task(pgrp, PIDTYPE_PGID);
1050                 if (!g || task_session(g) != task_session(group_leader))
1051                         goto out;
1052         }
1053
1054         err = security_task_setpgid(p, pgid);
1055         if (err)
1056                 goto out;
1057
1058         if (task_pgrp(p) != pgrp)
1059                 change_pid(p, PIDTYPE_PGID, pgrp);
1060
1061         err = 0;
1062 out:
1063         /* All paths lead to here, thus we are safe. -DaveM */
1064         write_unlock_irq(&tasklist_lock);
1065         rcu_read_unlock();
1066         return err;
1067 }
1068
1069 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1070 {
1071         struct task_struct *p;
1072         struct pid *grp;
1073         int retval;
1074
1075         rcu_read_lock();
1076         if (!pid)
1077                 grp = task_pgrp(current);
1078         else {
1079                 retval = -ESRCH;
1080                 p = find_task_by_vpid(pid);
1081                 if (!p)
1082                         goto out;
1083                 grp = task_pgrp(p);
1084                 if (!grp)
1085                         goto out;
1086
1087                 retval = security_task_getpgid(p);
1088                 if (retval)
1089                         goto out;
1090         }
1091         retval = pid_vnr(grp);
1092 out:
1093         rcu_read_unlock();
1094         return retval;
1095 }
1096
1097 #ifdef __ARCH_WANT_SYS_GETPGRP
1098
1099 SYSCALL_DEFINE0(getpgrp)
1100 {
1101         return sys_getpgid(0);
1102 }
1103
1104 #endif
1105
1106 SYSCALL_DEFINE1(getsid, pid_t, pid)
1107 {
1108         struct task_struct *p;
1109         struct pid *sid;
1110         int retval;
1111
1112         rcu_read_lock();
1113         if (!pid)
1114                 sid = task_session(current);
1115         else {
1116                 retval = -ESRCH;
1117                 p = find_task_by_vpid(pid);
1118                 if (!p)
1119                         goto out;
1120                 sid = task_session(p);
1121                 if (!sid)
1122                         goto out;
1123
1124                 retval = security_task_getsid(p);
1125                 if (retval)
1126                         goto out;
1127         }
1128         retval = pid_vnr(sid);
1129 out:
1130         rcu_read_unlock();
1131         return retval;
1132 }
1133
1134 SYSCALL_DEFINE0(setsid)
1135 {
1136         struct task_struct *group_leader = current->group_leader;
1137         struct pid *sid = task_pid(group_leader);
1138         pid_t session = pid_vnr(sid);
1139         int err = -EPERM;
1140
1141         write_lock_irq(&tasklist_lock);
1142         /* Fail if I am already a session leader */
1143         if (group_leader->signal->leader)
1144                 goto out;
1145
1146         /* Fail if a process group id already exists that equals the
1147          * proposed session id.
1148          */
1149         if (pid_task(sid, PIDTYPE_PGID))
1150                 goto out;
1151
1152         group_leader->signal->leader = 1;
1153         __set_special_pids(sid);
1154
1155         proc_clear_tty(group_leader);
1156
1157         err = session;
1158 out:
1159         write_unlock_irq(&tasklist_lock);
1160         if (err > 0) {
1161                 proc_sid_connector(group_leader);
1162                 sched_autogroup_create_attach(group_leader);
1163         }
1164         return err;
1165 }
1166
1167 DECLARE_RWSEM(uts_sem);
1168
1169 #ifdef COMPAT_UTS_MACHINE
1170 #define override_architecture(name) \
1171         (personality(current->personality) == PER_LINUX32 && \
1172          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1173                       sizeof(COMPAT_UTS_MACHINE)))
1174 #else
1175 #define override_architecture(name)     0
1176 #endif
1177
1178 /*
1179  * Work around broken programs that cannot handle "Linux 3.0".
1180  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1181  */
1182 static int override_release(char __user *release, size_t len)
1183 {
1184         int ret = 0;
1185         char buf[65] = { 0 };
1186         const char *rest = UTS_RELEASE;
1187
1188         if (current->personality & UNAME26) {
1189                 int ndots = 0;
1190                 unsigned v;
1191
1192                 while (*rest) {
1193                         if (*rest == '.' && ++ndots >= 3)
1194                                 break;
1195                         if (!isdigit(*rest) && *rest != '.')
1196                                 break;
1197                         rest++;
1198                 }
1199                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1200                 snprintf(buf, sizeof(buf), "2.6.%u%s", v, rest);
1201                 if (sizeof(buf) < len)
1202                         len = sizeof(buf);
1203                 ret = copy_to_user(release, buf, len);
1204         }
1205         return ret;
1206 }
1207
1208 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1209 {
1210         int errno = 0;
1211
1212         down_read(&uts_sem);
1213         if (copy_to_user(name, utsname(), sizeof *name))
1214                 errno = -EFAULT;
1215         up_read(&uts_sem);
1216
1217         if (!errno && override_release(name->release, sizeof(name->release)))
1218                 errno = -EFAULT;
1219         if (!errno && override_architecture(name))
1220                 errno = -EFAULT;
1221         return errno;
1222 }
1223
1224 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1225 /*
1226  * Old cruft
1227  */
1228 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1229 {
1230         int error = 0;
1231
1232         if (!name)
1233                 return -EFAULT;
1234
1235         down_read(&uts_sem);
1236         if (copy_to_user(name, utsname(), sizeof(*name)))
1237                 error = -EFAULT;
1238         up_read(&uts_sem);
1239
1240         if (!error && override_release(name->release, sizeof(name->release)))
1241                 error = -EFAULT;
1242         if (!error && override_architecture(name))
1243                 error = -EFAULT;
1244         return error;
1245 }
1246
1247 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1248 {
1249         int error;
1250
1251         if (!name)
1252                 return -EFAULT;
1253         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1254                 return -EFAULT;
1255
1256         down_read(&uts_sem);
1257         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1258                                __OLD_UTS_LEN);
1259         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1260         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1261                                 __OLD_UTS_LEN);
1262         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1263         error |= __copy_to_user(&name->release, &utsname()->release,
1264                                 __OLD_UTS_LEN);
1265         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1266         error |= __copy_to_user(&name->version, &utsname()->version,
1267                                 __OLD_UTS_LEN);
1268         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1269         error |= __copy_to_user(&name->machine, &utsname()->machine,
1270                                 __OLD_UTS_LEN);
1271         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1272         up_read(&uts_sem);
1273
1274         if (!error && override_architecture(name))
1275                 error = -EFAULT;
1276         if (!error && override_release(name->release, sizeof(name->release)))
1277                 error = -EFAULT;
1278         return error ? -EFAULT : 0;
1279 }
1280 #endif
1281
1282 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1283 {
1284         int errno;
1285         char tmp[__NEW_UTS_LEN];
1286
1287         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1288                 return -EPERM;
1289
1290         if (len < 0 || len > __NEW_UTS_LEN)
1291                 return -EINVAL;
1292         down_write(&uts_sem);
1293         errno = -EFAULT;
1294         if (!copy_from_user(tmp, name, len)) {
1295                 struct new_utsname *u = utsname();
1296
1297                 memcpy(u->nodename, tmp, len);
1298                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1299                 errno = 0;
1300         }
1301         uts_proc_notify(UTS_PROC_HOSTNAME);
1302         up_write(&uts_sem);
1303         return errno;
1304 }
1305
1306 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1307
1308 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1309 {
1310         int i, errno;
1311         struct new_utsname *u;
1312
1313         if (len < 0)
1314                 return -EINVAL;
1315         down_read(&uts_sem);
1316         u = utsname();
1317         i = 1 + strlen(u->nodename);
1318         if (i > len)
1319                 i = len;
1320         errno = 0;
1321         if (copy_to_user(name, u->nodename, i))
1322                 errno = -EFAULT;
1323         up_read(&uts_sem);
1324         return errno;
1325 }
1326
1327 #endif
1328
1329 /*
1330  * Only setdomainname; getdomainname can be implemented by calling
1331  * uname()
1332  */
1333 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1334 {
1335         int errno;
1336         char tmp[__NEW_UTS_LEN];
1337
1338         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1339                 return -EPERM;
1340         if (len < 0 || len > __NEW_UTS_LEN)
1341                 return -EINVAL;
1342
1343         down_write(&uts_sem);
1344         errno = -EFAULT;
1345         if (!copy_from_user(tmp, name, len)) {
1346                 struct new_utsname *u = utsname();
1347
1348                 memcpy(u->domainname, tmp, len);
1349                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1350                 errno = 0;
1351         }
1352         uts_proc_notify(UTS_PROC_DOMAINNAME);
1353         up_write(&uts_sem);
1354         return errno;
1355 }
1356
1357 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1358 {
1359         struct rlimit value;
1360         int ret;
1361
1362         ret = do_prlimit(current, resource, NULL, &value);
1363         if (!ret)
1364                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1365
1366         return ret;
1367 }
1368
1369 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1370
1371 /*
1372  *      Back compatibility for getrlimit. Needed for some apps.
1373  */
1374
1375 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1376                 struct rlimit __user *, rlim)
1377 {
1378         struct rlimit x;
1379         if (resource >= RLIM_NLIMITS)
1380                 return -EINVAL;
1381
1382         task_lock(current->group_leader);
1383         x = current->signal->rlim[resource];
1384         task_unlock(current->group_leader);
1385         if (x.rlim_cur > 0x7FFFFFFF)
1386                 x.rlim_cur = 0x7FFFFFFF;
1387         if (x.rlim_max > 0x7FFFFFFF)
1388                 x.rlim_max = 0x7FFFFFFF;
1389         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1390 }
1391
1392 #endif
1393
1394 static inline bool rlim64_is_infinity(__u64 rlim64)
1395 {
1396 #if BITS_PER_LONG < 64
1397         return rlim64 >= ULONG_MAX;
1398 #else
1399         return rlim64 == RLIM64_INFINITY;
1400 #endif
1401 }
1402
1403 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1404 {
1405         if (rlim->rlim_cur == RLIM_INFINITY)
1406                 rlim64->rlim_cur = RLIM64_INFINITY;
1407         else
1408                 rlim64->rlim_cur = rlim->rlim_cur;
1409         if (rlim->rlim_max == RLIM_INFINITY)
1410                 rlim64->rlim_max = RLIM64_INFINITY;
1411         else
1412                 rlim64->rlim_max = rlim->rlim_max;
1413 }
1414
1415 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1416 {
1417         if (rlim64_is_infinity(rlim64->rlim_cur))
1418                 rlim->rlim_cur = RLIM_INFINITY;
1419         else
1420                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1421         if (rlim64_is_infinity(rlim64->rlim_max))
1422                 rlim->rlim_max = RLIM_INFINITY;
1423         else
1424                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1425 }
1426
1427 /* make sure you are allowed to change @tsk limits before calling this */
1428 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1429                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1430 {
1431         struct rlimit *rlim;
1432         int retval = 0;
1433
1434         if (resource >= RLIM_NLIMITS)
1435                 return -EINVAL;
1436         if (new_rlim) {
1437                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1438                         return -EINVAL;
1439                 if (resource == RLIMIT_NOFILE &&
1440                                 new_rlim->rlim_max > sysctl_nr_open)
1441                         return -EPERM;
1442         }
1443
1444         /* protect tsk->signal and tsk->sighand from disappearing */
1445         read_lock(&tasklist_lock);
1446         if (!tsk->sighand) {
1447                 retval = -ESRCH;
1448                 goto out;
1449         }
1450
1451         rlim = tsk->signal->rlim + resource;
1452         task_lock(tsk->group_leader);
1453         if (new_rlim) {
1454                 /* Keep the capable check against init_user_ns until
1455                    cgroups can contain all limits */
1456                 if (new_rlim->rlim_max > rlim->rlim_max &&
1457                                 !capable(CAP_SYS_RESOURCE))
1458                         retval = -EPERM;
1459                 if (!retval)
1460                         retval = security_task_setrlimit(tsk->group_leader,
1461                                         resource, new_rlim);
1462                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1463                         /*
1464                          * The caller is asking for an immediate RLIMIT_CPU
1465                          * expiry.  But we use the zero value to mean "it was
1466                          * never set".  So let's cheat and make it one second
1467                          * instead
1468                          */
1469                         new_rlim->rlim_cur = 1;
1470                 }
1471         }
1472         if (!retval) {
1473                 if (old_rlim)
1474                         *old_rlim = *rlim;
1475                 if (new_rlim)
1476                         *rlim = *new_rlim;
1477         }
1478         task_unlock(tsk->group_leader);
1479
1480         /*
1481          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1482          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1483          * very long-standing error, and fixing it now risks breakage of
1484          * applications, so we live with it
1485          */
1486          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1487                          new_rlim->rlim_cur != RLIM_INFINITY)
1488                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1489 out:
1490         read_unlock(&tasklist_lock);
1491         return retval;
1492 }
1493
1494 /* rcu lock must be held */
1495 static int check_prlimit_permission(struct task_struct *task)
1496 {
1497         const struct cred *cred = current_cred(), *tcred;
1498
1499         if (current == task)
1500                 return 0;
1501
1502         tcred = __task_cred(task);
1503         if (cred->user->user_ns == tcred->user->user_ns &&
1504             (cred->uid == tcred->euid &&
1505              cred->uid == tcred->suid &&
1506              cred->uid == tcred->uid  &&
1507              cred->gid == tcred->egid &&
1508              cred->gid == tcred->sgid &&
1509              cred->gid == tcred->gid))
1510                 return 0;
1511         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1512                 return 0;
1513
1514         return -EPERM;
1515 }
1516
1517 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1518                 const struct rlimit64 __user *, new_rlim,
1519                 struct rlimit64 __user *, old_rlim)
1520 {
1521         struct rlimit64 old64, new64;
1522         struct rlimit old, new;
1523         struct task_struct *tsk;
1524         int ret;
1525
1526         if (new_rlim) {
1527                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1528                         return -EFAULT;
1529                 rlim64_to_rlim(&new64, &new);
1530         }
1531
1532         rcu_read_lock();
1533         tsk = pid ? find_task_by_vpid(pid) : current;
1534         if (!tsk) {
1535                 rcu_read_unlock();
1536                 return -ESRCH;
1537         }
1538         ret = check_prlimit_permission(tsk);
1539         if (ret) {
1540                 rcu_read_unlock();
1541                 return ret;
1542         }
1543         get_task_struct(tsk);
1544         rcu_read_unlock();
1545
1546         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1547                         old_rlim ? &old : NULL);
1548
1549         if (!ret && old_rlim) {
1550                 rlim_to_rlim64(&old, &old64);
1551                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1552                         ret = -EFAULT;
1553         }
1554
1555         put_task_struct(tsk);
1556         return ret;
1557 }
1558
1559 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1560 {
1561         struct rlimit new_rlim;
1562
1563         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1564                 return -EFAULT;
1565         return do_prlimit(current, resource, &new_rlim, NULL);
1566 }
1567
1568 /*
1569  * It would make sense to put struct rusage in the task_struct,
1570  * except that would make the task_struct be *really big*.  After
1571  * task_struct gets moved into malloc'ed memory, it would
1572  * make sense to do this.  It will make moving the rest of the information
1573  * a lot simpler!  (Which we're not doing right now because we're not
1574  * measuring them yet).
1575  *
1576  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1577  * races with threads incrementing their own counters.  But since word
1578  * reads are atomic, we either get new values or old values and we don't
1579  * care which for the sums.  We always take the siglock to protect reading
1580  * the c* fields from p->signal from races with exit.c updating those
1581  * fields when reaping, so a sample either gets all the additions of a
1582  * given child after it's reaped, or none so this sample is before reaping.
1583  *
1584  * Locking:
1585  * We need to take the siglock for CHILDEREN, SELF and BOTH
1586  * for  the cases current multithreaded, non-current single threaded
1587  * non-current multithreaded.  Thread traversal is now safe with
1588  * the siglock held.
1589  * Strictly speaking, we donot need to take the siglock if we are current and
1590  * single threaded,  as no one else can take our signal_struct away, no one
1591  * else can  reap the  children to update signal->c* counters, and no one else
1592  * can race with the signal-> fields. If we do not take any lock, the
1593  * signal-> fields could be read out of order while another thread was just
1594  * exiting. So we should  place a read memory barrier when we avoid the lock.
1595  * On the writer side,  write memory barrier is implied in  __exit_signal
1596  * as __exit_signal releases  the siglock spinlock after updating the signal->
1597  * fields. But we don't do this yet to keep things simple.
1598  *
1599  */
1600
1601 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1602 {
1603         r->ru_nvcsw += t->nvcsw;
1604         r->ru_nivcsw += t->nivcsw;
1605         r->ru_minflt += t->min_flt;
1606         r->ru_majflt += t->maj_flt;
1607         r->ru_inblock += task_io_get_inblock(t);
1608         r->ru_oublock += task_io_get_oublock(t);
1609 }
1610
1611 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1612 {
1613         struct task_struct *t;
1614         unsigned long flags;
1615         cputime_t tgutime, tgstime, utime, stime;
1616         unsigned long maxrss = 0;
1617
1618         memset((char *) r, 0, sizeof *r);
1619         utime = stime = 0;
1620
1621         if (who == RUSAGE_THREAD) {
1622                 task_times(current, &utime, &stime);
1623                 accumulate_thread_rusage(p, r);
1624                 maxrss = p->signal->maxrss;
1625                 goto out;
1626         }
1627
1628         if (!lock_task_sighand(p, &flags))
1629                 return;
1630
1631         switch (who) {
1632                 case RUSAGE_BOTH:
1633                 case RUSAGE_CHILDREN:
1634                         utime = p->signal->cutime;
1635                         stime = p->signal->cstime;
1636                         r->ru_nvcsw = p->signal->cnvcsw;
1637                         r->ru_nivcsw = p->signal->cnivcsw;
1638                         r->ru_minflt = p->signal->cmin_flt;
1639                         r->ru_majflt = p->signal->cmaj_flt;
1640                         r->ru_inblock = p->signal->cinblock;
1641                         r->ru_oublock = p->signal->coublock;
1642                         maxrss = p->signal->cmaxrss;
1643
1644                         if (who == RUSAGE_CHILDREN)
1645                                 break;
1646
1647                 case RUSAGE_SELF:
1648                         thread_group_times(p, &tgutime, &tgstime);
1649                         utime += tgutime;
1650                         stime += tgstime;
1651                         r->ru_nvcsw += p->signal->nvcsw;
1652                         r->ru_nivcsw += p->signal->nivcsw;
1653                         r->ru_minflt += p->signal->min_flt;
1654                         r->ru_majflt += p->signal->maj_flt;
1655                         r->ru_inblock += p->signal->inblock;
1656                         r->ru_oublock += p->signal->oublock;
1657                         if (maxrss < p->signal->maxrss)
1658                                 maxrss = p->signal->maxrss;
1659                         t = p;
1660                         do {
1661                                 accumulate_thread_rusage(t, r);
1662                                 t = next_thread(t);
1663                         } while (t != p);
1664                         break;
1665
1666                 default:
1667                         BUG();
1668         }
1669         unlock_task_sighand(p, &flags);
1670
1671 out:
1672         cputime_to_timeval(utime, &r->ru_utime);
1673         cputime_to_timeval(stime, &r->ru_stime);
1674
1675         if (who != RUSAGE_CHILDREN) {
1676                 struct mm_struct *mm = get_task_mm(p);
1677                 if (mm) {
1678                         setmax_mm_hiwater_rss(&maxrss, mm);
1679                         mmput(mm);
1680                 }
1681         }
1682         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1683 }
1684
1685 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1686 {
1687         struct rusage r;
1688         k_getrusage(p, who, &r);
1689         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1690 }
1691
1692 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1693 {
1694         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1695             who != RUSAGE_THREAD)
1696                 return -EINVAL;
1697         return getrusage(current, who, ru);
1698 }
1699
1700 SYSCALL_DEFINE1(umask, int, mask)
1701 {
1702         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1703         return mask;
1704 }
1705
1706 #ifdef CONFIG_CHECKPOINT_RESTORE
1707 static int prctl_set_mm(int opt, unsigned long addr,
1708                         unsigned long arg4, unsigned long arg5)
1709 {
1710         unsigned long rlim = rlimit(RLIMIT_DATA);
1711         unsigned long vm_req_flags;
1712         unsigned long vm_bad_flags;
1713         struct vm_area_struct *vma;
1714         int error = 0;
1715         struct mm_struct *mm = current->mm;
1716
1717         if (arg4 | arg5)
1718                 return -EINVAL;
1719
1720         if (!capable(CAP_SYS_RESOURCE))
1721                 return -EPERM;
1722
1723         if (addr >= TASK_SIZE)
1724                 return -EINVAL;
1725
1726         down_read(&mm->mmap_sem);
1727         vma = find_vma(mm, addr);
1728
1729         if (opt != PR_SET_MM_START_BRK && opt != PR_SET_MM_BRK) {
1730                 /* It must be existing VMA */
1731                 if (!vma || vma->vm_start > addr)
1732                         goto out;
1733         }
1734
1735         error = -EINVAL;
1736         switch (opt) {
1737         case PR_SET_MM_START_CODE:
1738         case PR_SET_MM_END_CODE:
1739                 vm_req_flags = VM_READ | VM_EXEC;
1740                 vm_bad_flags = VM_WRITE | VM_MAYSHARE;
1741
1742                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
1743                     (vma->vm_flags & vm_bad_flags))
1744                         goto out;
1745
1746                 if (opt == PR_SET_MM_START_CODE)
1747                         mm->start_code = addr;
1748                 else
1749                         mm->end_code = addr;
1750                 break;
1751
1752         case PR_SET_MM_START_DATA:
1753         case PR_SET_MM_END_DATA:
1754                 vm_req_flags = VM_READ | VM_WRITE;
1755                 vm_bad_flags = VM_EXEC | VM_MAYSHARE;
1756
1757                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
1758                     (vma->vm_flags & vm_bad_flags))
1759                         goto out;
1760
1761                 if (opt == PR_SET_MM_START_DATA)
1762                         mm->start_data = addr;
1763                 else
1764                         mm->end_data = addr;
1765                 break;
1766
1767         case PR_SET_MM_START_STACK:
1768
1769 #ifdef CONFIG_STACK_GROWSUP
1770                 vm_req_flags = VM_READ | VM_WRITE | VM_GROWSUP;
1771 #else
1772                 vm_req_flags = VM_READ | VM_WRITE | VM_GROWSDOWN;
1773 #endif
1774                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags)
1775                         goto out;
1776
1777                 mm->start_stack = addr;
1778                 break;
1779
1780         case PR_SET_MM_START_BRK:
1781                 if (addr <= mm->end_data)
1782                         goto out;
1783
1784                 if (rlim < RLIM_INFINITY &&
1785                     (mm->brk - addr) +
1786                     (mm->end_data - mm->start_data) > rlim)
1787                         goto out;
1788
1789                 mm->start_brk = addr;
1790                 break;
1791
1792         case PR_SET_MM_BRK:
1793                 if (addr <= mm->end_data)
1794                         goto out;
1795
1796                 if (rlim < RLIM_INFINITY &&
1797                     (addr - mm->start_brk) +
1798                     (mm->end_data - mm->start_data) > rlim)
1799                         goto out;
1800
1801                 mm->brk = addr;
1802                 break;
1803
1804         default:
1805                 error = -EINVAL;
1806                 goto out;
1807         }
1808
1809         error = 0;
1810
1811 out:
1812         up_read(&mm->mmap_sem);
1813
1814         return error;
1815 }
1816 #else /* CONFIG_CHECKPOINT_RESTORE */
1817 static int prctl_set_mm(int opt, unsigned long addr,
1818                         unsigned long arg4, unsigned long arg5)
1819 {
1820         return -EINVAL;
1821 }
1822 #endif
1823
1824 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1825                 unsigned long, arg4, unsigned long, arg5)
1826 {
1827         struct task_struct *me = current;
1828         unsigned char comm[sizeof(me->comm)];
1829         long error;
1830
1831         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1832         if (error != -ENOSYS)
1833                 return error;
1834
1835         error = 0;
1836         switch (option) {
1837                 case PR_SET_PDEATHSIG:
1838                         if (!valid_signal(arg2)) {
1839                                 error = -EINVAL;
1840                                 break;
1841                         }
1842                         me->pdeath_signal = arg2;
1843                         error = 0;
1844                         break;
1845                 case PR_GET_PDEATHSIG:
1846                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1847                         break;
1848                 case PR_GET_DUMPABLE:
1849                         error = get_dumpable(me->mm);
1850                         break;
1851                 case PR_SET_DUMPABLE:
1852                         if (arg2 < 0 || arg2 > 1) {
1853                                 error = -EINVAL;
1854                                 break;
1855                         }
1856                         set_dumpable(me->mm, arg2);
1857                         error = 0;
1858                         break;
1859
1860                 case PR_SET_UNALIGN:
1861                         error = SET_UNALIGN_CTL(me, arg2);
1862                         break;
1863                 case PR_GET_UNALIGN:
1864                         error = GET_UNALIGN_CTL(me, arg2);
1865                         break;
1866                 case PR_SET_FPEMU:
1867                         error = SET_FPEMU_CTL(me, arg2);
1868                         break;
1869                 case PR_GET_FPEMU:
1870                         error = GET_FPEMU_CTL(me, arg2);
1871                         break;
1872                 case PR_SET_FPEXC:
1873                         error = SET_FPEXC_CTL(me, arg2);
1874                         break;
1875                 case PR_GET_FPEXC:
1876                         error = GET_FPEXC_CTL(me, arg2);
1877                         break;
1878                 case PR_GET_TIMING:
1879                         error = PR_TIMING_STATISTICAL;
1880                         break;
1881                 case PR_SET_TIMING:
1882                         if (arg2 != PR_TIMING_STATISTICAL)
1883                                 error = -EINVAL;
1884                         else
1885                                 error = 0;
1886                         break;
1887
1888                 case PR_SET_NAME:
1889                         comm[sizeof(me->comm)-1] = 0;
1890                         if (strncpy_from_user(comm, (char __user *)arg2,
1891                                               sizeof(me->comm) - 1) < 0)
1892                                 return -EFAULT;
1893                         set_task_comm(me, comm);
1894                         proc_comm_connector(me);
1895                         return 0;
1896                 case PR_GET_NAME:
1897                         get_task_comm(comm, me);
1898                         if (copy_to_user((char __user *)arg2, comm,
1899                                          sizeof(comm)))
1900                                 return -EFAULT;
1901                         return 0;
1902                 case PR_GET_ENDIAN:
1903                         error = GET_ENDIAN(me, arg2);
1904                         break;
1905                 case PR_SET_ENDIAN:
1906                         error = SET_ENDIAN(me, arg2);
1907                         break;
1908
1909                 case PR_GET_SECCOMP:
1910                         error = prctl_get_seccomp();
1911                         break;
1912                 case PR_SET_SECCOMP:
1913                         error = prctl_set_seccomp(arg2, (char __user *)arg3);
1914                         break;
1915                 case PR_GET_TSC:
1916                         error = GET_TSC_CTL(arg2);
1917                         break;
1918                 case PR_SET_TSC:
1919                         error = SET_TSC_CTL(arg2);
1920                         break;
1921                 case PR_TASK_PERF_EVENTS_DISABLE:
1922                         error = perf_event_task_disable();
1923                         break;
1924                 case PR_TASK_PERF_EVENTS_ENABLE:
1925                         error = perf_event_task_enable();
1926                         break;
1927                 case PR_GET_TIMERSLACK:
1928                         error = current->timer_slack_ns;
1929                         break;
1930                 case PR_SET_TIMERSLACK:
1931                         if (arg2 <= 0)
1932                                 current->timer_slack_ns =
1933                                         current->default_timer_slack_ns;
1934                         else
1935                                 current->timer_slack_ns = arg2;
1936                         error = 0;
1937                         break;
1938                 case PR_MCE_KILL:
1939                         if (arg4 | arg5)
1940                                 return -EINVAL;
1941                         switch (arg2) {
1942                         case PR_MCE_KILL_CLEAR:
1943                                 if (arg3 != 0)
1944                                         return -EINVAL;
1945                                 current->flags &= ~PF_MCE_PROCESS;
1946                                 break;
1947                         case PR_MCE_KILL_SET:
1948                                 current->flags |= PF_MCE_PROCESS;
1949                                 if (arg3 == PR_MCE_KILL_EARLY)
1950                                         current->flags |= PF_MCE_EARLY;
1951                                 else if (arg3 == PR_MCE_KILL_LATE)
1952                                         current->flags &= ~PF_MCE_EARLY;
1953                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1954                                         current->flags &=
1955                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1956                                 else
1957                                         return -EINVAL;
1958                                 break;
1959                         default:
1960                                 return -EINVAL;
1961                         }
1962                         error = 0;
1963                         break;
1964                 case PR_MCE_KILL_GET:
1965                         if (arg2 | arg3 | arg4 | arg5)
1966                                 return -EINVAL;
1967                         if (current->flags & PF_MCE_PROCESS)
1968                                 error = (current->flags & PF_MCE_EARLY) ?
1969                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1970                         else
1971                                 error = PR_MCE_KILL_DEFAULT;
1972                         break;
1973                 case PR_SET_MM:
1974                         error = prctl_set_mm(arg2, arg3, arg4, arg5);
1975                         break;
1976                 case PR_SET_CHILD_SUBREAPER:
1977                         me->signal->is_child_subreaper = !!arg2;
1978                         error = 0;
1979                         break;
1980                 case PR_GET_CHILD_SUBREAPER:
1981                         error = put_user(me->signal->is_child_subreaper,
1982                                          (int __user *) arg2);
1983                         break;
1984                 case PR_SET_NO_NEW_PRIVS:
1985                         if (arg2 != 1 || arg3 || arg4 || arg5)
1986                                 return -EINVAL;
1987
1988                         current->no_new_privs = 1;
1989                         break;
1990                 case PR_GET_NO_NEW_PRIVS:
1991                         if (arg2 || arg3 || arg4 || arg5)
1992                                 return -EINVAL;
1993                         return current->no_new_privs ? 1 : 0;
1994                 default:
1995                         error = -EINVAL;
1996                         break;
1997         }
1998         return error;
1999 }
2000
2001 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2002                 struct getcpu_cache __user *, unused)
2003 {
2004         int err = 0;
2005         int cpu = raw_smp_processor_id();
2006         if (cpup)
2007                 err |= put_user(cpu, cpup);
2008         if (nodep)
2009                 err |= put_user(cpu_to_node(cpu), nodep);
2010         return err ? -EFAULT : 0;
2011 }
2012
2013 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2014
2015 static void argv_cleanup(struct subprocess_info *info)
2016 {
2017         argv_free(info->argv);
2018 }
2019
2020 /**
2021  * orderly_poweroff - Trigger an orderly system poweroff
2022  * @force: force poweroff if command execution fails
2023  *
2024  * This may be called from any context to trigger a system shutdown.
2025  * If the orderly shutdown fails, it will force an immediate shutdown.
2026  */
2027 int orderly_poweroff(bool force)
2028 {
2029         int argc;
2030         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
2031         static char *envp[] = {
2032                 "HOME=/",
2033                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2034                 NULL
2035         };
2036         int ret = -ENOMEM;
2037         struct subprocess_info *info;
2038
2039         if (argv == NULL) {
2040                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2041                        __func__, poweroff_cmd);
2042                 goto out;
2043         }
2044
2045         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
2046         if (info == NULL) {
2047                 argv_free(argv);
2048                 goto out;
2049         }
2050
2051         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
2052
2053         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
2054
2055   out:
2056         if (ret && force) {
2057                 printk(KERN_WARNING "Failed to start orderly shutdown: "
2058                        "forcing the issue\n");
2059
2060                 /* I guess this should try to kick off some daemon to
2061                    sync and poweroff asap.  Or not even bother syncing
2062                    if we're doing an emergency shutdown? */
2063                 emergency_sync();
2064                 kernel_power_off();
2065         }
2066
2067         return ret;
2068 }
2069 EXPORT_SYMBOL_GPL(orderly_poweroff);