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