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 }
 324
 325 /**
 326  *      register_reboot_notifier - Register function to be called at reboot time
 327  *      @nb: Info about notifier function to be called
 328  *
 329  *      Registers a function with the list of functions
 330  *      to be called at reboot time.
 331  *
 332  *      Currently always returns zero, as blocking_notifier_chain_register()
 333  *      always returns zero.
 334  */
 335 int register_reboot_notifier(struct notifier_block *nb)
 336 {
 337         return blocking_notifier_chain_register(&reboot_notifier_list, nb);
 338 }
 339 EXPORT_SYMBOL(register_reboot_notifier);
 340
 341 /**
 342  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 343  *      @nb: Hook to be unregistered
 344  *
 345  *      Unregisters a previously registered reboot
 346  *      notifier function.
 347  *
 348  *      Returns zero on success, or %-ENOENT on failure.
 349  */
 350 int unregister_reboot_notifier(struct notifier_block *nb)
 351 {
 352         return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
 353 }
 354 EXPORT_SYMBOL(unregister_reboot_notifier);
 355
 356 /**
 357  *      kernel_restart - reboot the system
 358  *      @cmd: pointer to buffer containing command to execute for restart
 359  *              or %NULL
 360  *
 361  *      Shutdown everything and perform a clean reboot.
 362  *      This is not safe to call in interrupt context.
 363  */
 364 void kernel_restart(char *cmd)
 365 {
 366         kernel_restart_prepare(cmd);
 367         disable_nonboot_cpus();
 368         syscore_shutdown();
 369         if (!cmd)
 370                 printk(KERN_EMERG "Restarting system.\n");
 371         else
 372                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 373         kmsg_dump(KMSG_DUMP_RESTART);
 374         machine_restart(cmd);
 375 }
 376 EXPORT_SYMBOL_GPL(kernel_restart);
 377
 378 static void kernel_shutdown_prepare(enum system_states state)
 379 {
 380         blocking_notifier_call_chain(&reboot_notifier_list,
 381                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 382         system_state = state;
 383         usermodehelper_disable();
 384         device_shutdown();
 385 }
 386 /**
 387  *      kernel_halt - halt the system
 388  *
 389  *      Shutdown everything and perform a clean system halt.
 390  */
 391 void kernel_halt(void)
 392 {
 393         kernel_shutdown_prepare(SYSTEM_HALT);
 394         disable_nonboot_cpus();
 395         syscore_shutdown();
 396         printk(KERN_EMERG "System halted.\n");
 397         kmsg_dump(KMSG_DUMP_HALT);
 398         machine_halt();
 399 }
 400
 401 EXPORT_SYMBOL_GPL(kernel_halt);
 402
 403 /**
 404  *      kernel_power_off - power_off the system
 405  *
 406  *      Shutdown everything and perform a clean system power_off.
 407  */
 408 void kernel_power_off(void)
 409 {
 410         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 411         if (pm_power_off_prepare)
 412                 pm_power_off_prepare();
 413         disable_nonboot_cpus();
 414         syscore_shutdown();
 415         printk(KERN_EMERG "Power down.\n");
 416         kmsg_dump(KMSG_DUMP_POWEROFF);
 417         machine_power_off();
 418 }
 419 EXPORT_SYMBOL_GPL(kernel_power_off);
 420
 421 static DEFINE_MUTEX(reboot_mutex);
 422
 423 /*
 424  * Reboot system call: for obvious reasons only root may call it,
 425  * and even root needs to set up some magic numbers in the registers
 426  * so that some mistake won't make this reboot the whole machine.
 427  * You can also set the meaning of the ctrl-alt-del-key here.
 428  *
 429  * reboot doesn't sync: do that yourself before calling this.
 430  */
 431 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 432                 void __user *, arg)
 433 {
 434         char buffer[256];
 435         int ret = 0;
 436
 437         /* We only trust the superuser with rebooting the system. */
 438         if (!capable(CAP_SYS_BOOT))
 439                 return -EPERM;
 440
 441         /* For safety, we require "magic" arguments. */
 442         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 443             (magic2 != LINUX_REBOOT_MAGIC2 &&
 444                         magic2 != LINUX_REBOOT_MAGIC2A &&
 445                         magic2 != LINUX_REBOOT_MAGIC2B &&
 446                         magic2 != LINUX_REBOOT_MAGIC2C))
 447                 return -EINVAL;
 448
 449         /* Instead of trying to make the power_off code look like
 450          * halt when pm_power_off is not set do it the easy way.
 451          */
 452         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 453                 cmd = LINUX_REBOOT_CMD_HALT;
 454
 455         mutex_lock(&reboot_mutex);
 456         switch (cmd) {
 457         case LINUX_REBOOT_CMD_RESTART:
 458                 kernel_restart(NULL);
 459                 break;
 460
 461         case LINUX_REBOOT_CMD_CAD_ON:
 462                 C_A_D = 1;
 463                 break;
 464
 465         case LINUX_REBOOT_CMD_CAD_OFF:
 466                 C_A_D = 0;
 467                 break;
 468
 469         case LINUX_REBOOT_CMD_HALT:
 470                 kernel_halt();
 471                 do_exit(0);
 472                 panic("cannot halt");
 473
 474         case LINUX_REBOOT_CMD_POWER_OFF:
 475                 kernel_power_off();
 476                 do_exit(0);
 477                 break;
 478
 479         case LINUX_REBOOT_CMD_RESTART2:
 480                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 481                         ret = -EFAULT;
 482                         break;
 483                 }
 484                 buffer[sizeof(buffer) - 1] = '\0';
 485
 486                 kernel_restart(buffer);
 487                 break;
 488
 489 #ifdef CONFIG_KEXEC
 490         case LINUX_REBOOT_CMD_KEXEC:
 491                 ret = kernel_kexec();
 492                 break;
 493 #endif
 494
 495 #ifdef CONFIG_HIBERNATION
 496         case LINUX_REBOOT_CMD_SW_SUSPEND:
 497                 ret = hibernate();
 498                 break;
 499 #endif
 500
 501         default:
 502                 ret = -EINVAL;
 503                 break;
 504         }
 505         mutex_unlock(&reboot_mutex);
 506         return ret;
 507 }
 508
 509 static void deferred_cad(struct work_struct *dummy)
 510 {
 511         kernel_restart(NULL);
 512 }
 513
 514 /*
 515  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 516  * As it's called within an interrupt, it may NOT sync: the only choice
 517  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 518  */
 519 void ctrl_alt_del(void)
 520 {
 521         static DECLARE_WORK(cad_work, deferred_cad);
 522
 523         if (C_A_D)
 524                 schedule_work(&cad_work);
 525         else
 526                 kill_cad_pid(SIGINT, 1);
 527 }
 528
 529 /*
 530  * Unprivileged users may change the real gid to the effective gid
 531  * or vice versa.  (BSD-style)
 532  *
 533  * If you set the real gid at all, or set the effective gid to a value not
 534  * equal to the real gid, then the saved gid is set to the new effective gid.
 535  *
 536  * This makes it possible for a setgid program to completely drop its
 537  * privileges, which is often a useful assertion to make when you are doing
 538  * a security audit over a program.
 539  *
 540  * The general idea is that a program which uses just setregid() will be
 541  * 100% compatible with BSD.  A program which uses just setgid() will be
 542  * 100% compatible with POSIX with saved IDs.
 543  *
 544  * SMP: There are not races, the GIDs are checked only by filesystem
 545  *      operations (as far as semantic preservation is concerned).
 546  */
 547 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 548 {
 549         const struct cred *old;
 550         struct cred *new;
 551         int retval;
 552
 553         new = prepare_creds();
 554         if (!new)
 555                 return -ENOMEM;
 556         old = current_cred();
 557
 558         retval = -EPERM;
 559         if (rgid != (gid_t) -1) {
 560                 if (old->gid == rgid ||
 561                     old->egid == rgid ||
 562                     nsown_capable(CAP_SETGID))
 563                         new->gid = rgid;
 564                 else
 565                         goto error;
 566         }
 567         if (egid != (gid_t) -1) {
 568                 if (old->gid == egid ||
 569                     old->egid == egid ||
 570                     old->sgid == egid ||
 571                     nsown_capable(CAP_SETGID))
 572                         new->egid = egid;
 573                 else
 574                         goto error;
 575         }
 576
 577         if (rgid != (gid_t) -1 ||
 578             (egid != (gid_t) -1 && egid != old->gid))
 579                 new->sgid = new->egid;
 580         new->fsgid = new->egid;
 581
 582         return commit_creds(new);
 583
 584 error:
 585         abort_creds(new);
 586         return retval;
 587 }
 588
 589 /*
 590  * setgid() is implemented like SysV w/ SAVED_IDS
 591  *
 592  * SMP: Same implicit races as above.
 593  */
 594 SYSCALL_DEFINE1(setgid, gid_t, gid)
 595 {
 596         const struct cred *old;
 597         struct cred *new;
 598         int retval;
 599
 600         new = prepare_creds();
 601         if (!new)
 602                 return -ENOMEM;
 603         old = current_cred();
 604
 605         retval = -EPERM;
 606         if (nsown_capable(CAP_SETGID))
 607                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 608         else if (gid == old->gid || gid == old->sgid)
 609                 new->egid = new->fsgid = gid;
 610         else
 611                 goto error;
 612
 613         return commit_creds(new);
 614
 615 error:
 616         abort_creds(new);
 617         return retval;
 618 }
 619
 620 /*
 621  * change the user struct in a credentials set to match the new UID
 622  */
 623 static int set_user(struct cred *new)
 624 {
 625         struct user_struct *new_user;
 626
 627         new_user = alloc_uid(current_user_ns(), new->uid);
 628         if (!new_user)
 629                 return -EAGAIN;
 630
 631         /*
 632          * We don't fail in case of NPROC limit excess here because too many
 633          * poorly written programs don't check set*uid() return code, assuming
 634          * it never fails if called by root.  We may still enforce NPROC limit
 635          * for programs doing set*uid()+execve() by harmlessly deferring the
 636          * failure to the execve() stage.
 637          */
 638         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 639                         new_user != INIT_USER)
 640                 current->flags |= PF_NPROC_EXCEEDED;
 641         else
 642                 current->flags &= ~PF_NPROC_EXCEEDED;
 643
 644         free_uid(new->user);
 645         new->user = new_user;
 646         return 0;
 647 }
 648
 649 /*
 650  * Unprivileged users may change the real uid to the effective uid
 651  * or vice versa.  (BSD-style)
 652  *
 653  * If you set the real uid at all, or set the effective uid to a value not
 654  * equal to the real uid, then the saved uid is set to the new effective uid.
 655  *
 656  * This makes it possible for a setuid program to completely drop its
 657  * privileges, which is often a useful assertion to make when you are doing
 658  * a security audit over a program.
 659  *
 660  * The general idea is that a program which uses just setreuid() will be
 661  * 100% compatible with BSD.  A program which uses just setuid() will be
 662  * 100% compatible with POSIX with saved IDs.
 663  */
 664 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 665 {
 666         const struct cred *old;
 667         struct cred *new;
 668         int retval;
 669
 670         new = prepare_creds();
 671         if (!new)
 672                 return -ENOMEM;
 673         old = current_cred();
 674
 675         retval = -EPERM;
 676         if (ruid != (uid_t) -1) {
 677                 new->uid = ruid;
 678                 if (old->uid != ruid &&
 679                     old->euid != ruid &&
 680                     !nsown_capable(CAP_SETUID))
 681                         goto error;
 682         }
 683
 684         if (euid != (uid_t) -1) {
 685                 new->euid = euid;
 686                 if (old->uid != euid &&
 687                     old->euid != euid &&
 688                     old->suid != euid &&
 689                     !nsown_capable(CAP_SETUID))
 690                         goto error;
 691         }
 692
 693         if (new->uid != old->uid) {
 694                 retval = set_user(new);
 695                 if (retval < 0)
 696                         goto error;
 697         }
 698         if (ruid != (uid_t) -1 ||
 699             (euid != (uid_t) -1 && euid != old->uid))
 700                 new->suid = new->euid;
 701         new->fsuid = new->euid;
 702
 703         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 704         if (retval < 0)
 705                 goto error;
 706
 707         return commit_creds(new);
 708
 709 error:
 710         abort_creds(new);
 711         return retval;
 712 }
 713
 714 /*
 715  * setuid() is implemented like SysV with SAVED_IDS
 716  *
 717  * Note that SAVED_ID's is deficient in that a setuid root program
 718  * like sendmail, for example, cannot set its uid to be a normal
 719  * user and then switch back, because if you're root, setuid() sets
 720  * the saved uid too.  If you don't like this, blame the bright people
 721  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 722  * will allow a root program to temporarily drop privileges and be able to
 723  * regain them by swapping the real and effective uid.
 724  */
 725 SYSCALL_DEFINE1(setuid, uid_t, uid)
 726 {
 727         const struct cred *old;
 728         struct cred *new;
 729         int retval;
 730
 731         new = prepare_creds();
 732         if (!new)
 733                 return -ENOMEM;
 734         old = current_cred();
 735
 736         retval = -EPERM;
 737         if (nsown_capable(CAP_SETUID)) {
 738                 new->suid = new->uid = uid;
 739                 if (uid != old->uid) {
 740                         retval = set_user(new);
 741                         if (retval < 0)
 742                                 goto error;
 743                 }
 744         } else if (uid != old->uid && uid != new->suid) {
 745                 goto error;
 746         }
 747
 748         new->fsuid = new->euid = uid;
 749
 750         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 751         if (retval < 0)
 752                 goto error;
 753
 754         return commit_creds(new);
 755
 756 error:
 757         abort_creds(new);
 758         return retval;
 759 }
 760
 761
 762 /*
 763  * This function implements a generic ability to update ruid, euid,
 764  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 765  */
 766 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 767 {
 768         const struct cred *old;
 769         struct cred *new;
 770         int retval;
 771
 772         new = prepare_creds();
 773         if (!new)
 774                 return -ENOMEM;
 775
 776         old = current_cred();
 777
 778         retval = -EPERM;
 779         if (!nsown_capable(CAP_SETUID)) {
 780                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 781                     ruid != old->euid  && ruid != old->suid)
 782                         goto error;
 783                 if (euid != (uid_t) -1 && euid != old->uid &&
 784                     euid != old->euid  && euid != old->suid)
 785                         goto error;
 786                 if (suid != (uid_t) -1 && suid != old->uid &&
 787                     suid != old->euid  && suid != old->suid)
 788                         goto error;
 789         }
 790
 791         if (ruid != (uid_t) -1) {
 792                 new->uid = ruid;
 793                 if (ruid != old->uid) {
 794                         retval = set_user(new);
 795                         if (retval < 0)
 796                                 goto error;
 797                 }
 798         }
 799         if (euid != (uid_t) -1)
 800                 new->euid = euid;
 801         if (suid != (uid_t) -1)
 802                 new->suid = suid;
 803         new->fsuid = new->euid;
 804
 805         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 806         if (retval < 0)
 807                 goto error;
 808
 809         return commit_creds(new);
 810
 811 error:
 812         abort_creds(new);
 813         return retval;
 814 }
 815
 816 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 817 {
 818         const struct cred *cred = current_cred();
 819         int retval;
 820
 821         if (!(retval   = put_user(cred->uid,  ruid)) &&
 822             !(retval   = put_user(cred->euid, euid)))
 823                 retval = put_user(cred->suid, suid);
 824
 825         return retval;
 826 }
 827
 828 /*
 829  * Same as above, but for rgid, egid, sgid.
 830  */
 831 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 832 {
 833         const struct cred *old;
 834         struct cred *new;
 835         int retval;
 836
 837         new = prepare_creds();
 838         if (!new)
 839                 return -ENOMEM;
 840         old = current_cred();
 841
 842         retval = -EPERM;
 843         if (!nsown_capable(CAP_SETGID)) {
 844                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 845                     rgid != old->egid  && rgid != old->sgid)
 846                         goto error;
 847                 if (egid != (gid_t) -1 && egid != old->gid &&
 848                     egid != old->egid  && egid != old->sgid)
 849                         goto error;
 850                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 851                     sgid != old->egid  && sgid != old->sgid)
 852                         goto error;
 853         }
 854
 855         if (rgid != (gid_t) -1)
 856                 new->gid = rgid;
 857         if (egid != (gid_t) -1)
 858                 new->egid = egid;
 859         if (sgid != (gid_t) -1)
 860                 new->sgid = sgid;
 861         new->fsgid = new->egid;
 862
 863         return commit_creds(new);
 864
 865 error:
 866         abort_creds(new);
 867         return retval;
 868 }
 869
 870 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 871 {
 872         const struct cred *cred = current_cred();
 873         int retval;
 874
 875         if (!(retval   = put_user(cred->gid,  rgid)) &&
 876             !(retval   = put_user(cred->egid, egid)))
 877                 retval = put_user(cred->sgid, sgid);
 878
 879         return retval;
 880 }
 881
 882
 883 /*
 884  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 885  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 886  * whatever uid it wants to). It normally shadows "euid", except when
 887  * explicitly set by setfsuid() or for access..
 888  */
 889 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 890 {
 891         const struct cred *old;
 892         struct cred *new;
 893         uid_t old_fsuid;
 894
 895         new = prepare_creds();
 896         if (!new)
 897                 return current_fsuid();
 898         old = current_cred();
 899         old_fsuid = old->fsuid;
 900
 901         if (uid == old->uid  || uid == old->euid  ||
 902             uid == old->suid || uid == old->fsuid ||
 903             nsown_capable(CAP_SETUID)) {
 904                 if (uid != old_fsuid) {
 905                         new->fsuid = uid;
 906                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 907                                 goto change_okay;
 908                 }
 909         }
 910
 911         abort_creds(new);
 912         return old_fsuid;
 913
 914 change_okay:
 915         commit_creds(new);
 916         return old_fsuid;
 917 }
 918
 919 /*
 920  * Samma på svenska..
 921  */
 922 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 923 {
 924         const struct cred *old;
 925         struct cred *new;
 926         gid_t old_fsgid;
 927
 928         new = prepare_creds();
 929         if (!new)
 930                 return current_fsgid();
 931         old = current_cred();
 932         old_fsgid = old->fsgid;
 933
 934         if (gid == old->gid  || gid == old->egid  ||
 935             gid == old->sgid || gid == old->fsgid ||
 936             nsown_capable(CAP_SETGID)) {
 937                 if (gid != old_fsgid) {
 938                         new->fsgid = gid;
 939                         goto change_okay;
 940                 }
 941         }
 942
 943         abort_creds(new);
 944         return old_fsgid;
 945
 946 change_okay:
 947         commit_creds(new);
 948         return old_fsgid;
 949 }
 950
 951 void do_sys_times(struct tms *tms)
 952 {
 953         cputime_t tgutime, tgstime, cutime, cstime;
 954
 955         spin_lock_irq(&current->sighand->siglock);
 956         thread_group_times(current, &tgutime, &tgstime);
 957         cutime = current->signal->cutime;
 958         cstime = current->signal->cstime;
 959         spin_unlock_irq(&current->sighand->siglock);
 960         tms->tms_utime = cputime_to_clock_t(tgutime);
 961         tms->tms_stime = cputime_to_clock_t(tgstime);
 962         tms->tms_cutime = cputime_to_clock_t(cutime);
 963         tms->tms_cstime = cputime_to_clock_t(cstime);
 964 }
 965
 966 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 967 {
 968         if (tbuf) {
 969                 struct tms tmp;
 970
 971                 do_sys_times(&tmp);
 972                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 973                         return -EFAULT;
 974         }
 975         force_successful_syscall_return();
 976         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 977 }
 978
 979 /*
 980  * This needs some heavy checking ...
 981  * I just haven't the stomach for it. I also don't fully
 982  * understand sessions/pgrp etc. Let somebody who does explain it.
 983  *
 984  * OK, I think I have the protection semantics right.... this is really
 985  * only important on a multi-user system anyway, to make sure one user
 986  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 987  *
 988  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 989  * LBT 04.03.94
 990  */
 991 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 992 {
 993         struct task_struct *p;
 994         struct task_struct *group_leader = current->group_leader;
 995         struct pid *pgrp;
 996         int err;
 997
 998         if (!pid)
 999                 pid = task_pid_vnr(group_leader);
1000         if (!pgid)
1001                 pgid = pid;
1002         if (pgid < 0)
1003                 return -EINVAL;
1004         rcu_read_lock();
1005
1006         /* From this point forward we keep holding onto the tasklist lock
1007          * so that our parent does not change from under us. -DaveM
1008          */
1009         write_lock_irq(&tasklist_lock);
1010
1011         err = -ESRCH;
1012         p = find_task_by_vpid(pid);
1013         if (!p)
1014                 goto out;
1015
1016         err = -EINVAL;
1017         if (!thread_group_leader(p))
1018                 goto out;
1019
1020         if (same_thread_group(p->real_parent, group_leader)) {
1021                 err = -EPERM;
1022                 if (task_session(p) != task_session(group_leader))
1023                         goto out;
1024                 err = -EACCES;
1025                 if (p->did_exec)
1026                         goto out;
1027         } else {
1028                 err = -ESRCH;
1029                 if (p != group_leader)
1030                         goto out;
1031         }
1032
1033         err = -EPERM;
1034         if (p->signal->leader)
1035                 goto out;
1036
1037         pgrp = task_pid(p);
1038         if (pgid != pid) {
1039                 struct task_struct *g;
1040
1041                 pgrp = find_vpid(pgid);
1042                 g = pid_task(pgrp, PIDTYPE_PGID);
1043                 if (!g || task_session(g) != task_session(group_leader))
1044                         goto out;
1045         }
1046
1047         err = security_task_setpgid(p, pgid);
1048         if (err)
1049                 goto out;
1050
1051         if (task_pgrp(p) != pgrp)
1052                 change_pid(p, PIDTYPE_PGID, pgrp);
1053
1054         err = 0;
1055 out:
1056         /* All paths lead to here, thus we are safe. -DaveM */
1057         write_unlock_irq(&tasklist_lock);
1058         rcu_read_unlock();
1059         return err;
1060 }
1061
1062 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1063 {
1064         struct task_struct *p;
1065         struct pid *grp;
1066         int retval;
1067
1068         rcu_read_lock();
1069         if (!pid)
1070                 grp = task_pgrp(current);
1071         else {
1072                 retval = -ESRCH;
1073                 p = find_task_by_vpid(pid);
1074                 if (!p)
1075                         goto out;
1076                 grp = task_pgrp(p);
1077                 if (!grp)
1078                         goto out;
1079
1080                 retval = security_task_getpgid(p);
1081                 if (retval)
1082                         goto out;
1083         }
1084         retval = pid_vnr(grp);
1085 out:
1086         rcu_read_unlock();
1087         return retval;
1088 }
1089
1090 #ifdef __ARCH_WANT_SYS_GETPGRP
1091
1092 SYSCALL_DEFINE0(getpgrp)
1093 {
1094         return sys_getpgid(0);
1095 }
1096
1097 #endif
1098
1099 SYSCALL_DEFINE1(getsid, pid_t, pid)
1100 {
1101         struct task_struct *p;
1102         struct pid *sid;
1103         int retval;
1104
1105         rcu_read_lock();
1106         if (!pid)
1107                 sid = task_session(current);
1108         else {
1109                 retval = -ESRCH;
1110                 p = find_task_by_vpid(pid);
1111                 if (!p)
1112                         goto out;
1113                 sid = task_session(p);
1114                 if (!sid)
1115                         goto out;
1116
1117                 retval = security_task_getsid(p);
1118                 if (retval)
1119                         goto out;
1120         }
1121         retval = pid_vnr(sid);
1122 out:
1123         rcu_read_unlock();
1124         return retval;
1125 }
1126
1127 SYSCALL_DEFINE0(setsid)
1128 {
1129         struct task_struct *group_leader = current->group_leader;
1130         struct pid *sid = task_pid(group_leader);
1131         pid_t session = pid_vnr(sid);
1132         int err = -EPERM;
1133
1134         write_lock_irq(&tasklist_lock);
1135         /* Fail if I am already a session leader */
1136         if (group_leader->signal->leader)
1137                 goto out;
1138
1139         /* Fail if a process group id already exists that equals the
1140          * proposed session id.
1141          */
1142         if (pid_task(sid, PIDTYPE_PGID))
1143                 goto out;
1144
1145         group_leader->signal->leader = 1;
1146         __set_special_pids(sid);
1147
1148         proc_clear_tty(group_leader);
1149
1150         err = session;
1151 out:
1152         write_unlock_irq(&tasklist_lock);
1153         if (err > 0) {
1154                 proc_sid_connector(group_leader);
1155                 sched_autogroup_create_attach(group_leader);
1156         }
1157         return err;
1158 }
1159
1160 DECLARE_RWSEM(uts_sem);
1161
1162 #ifdef COMPAT_UTS_MACHINE
1163 #define override_architecture(name) \
1164         (personality(current->personality) == PER_LINUX32 && \
1165          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1166                       sizeof(COMPAT_UTS_MACHINE)))
1167 #else
1168 #define override_architecture(name)     0
1169 #endif
1170
1171 /*
1172  * Work around broken programs that cannot handle "Linux 3.0".
1173  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1174  */
1175 static int override_release(char __user *release, size_t len)
1176 {
1177         int ret = 0;
1178
1179         if (current->personality & UNAME26) {
1180                 const char *rest = UTS_RELEASE;
1181                 char buf[65] = { 0 };
1182                 int ndots = 0;
1183                 unsigned v;
1184                 size_t copy;
1185
1186                 while (*rest) {
1187                         if (*rest == '.' && ++ndots >= 3)
1188                                 break;
1189                         if (!isdigit(*rest) && *rest != '.')
1190                                 break;
1191                         rest++;
1192                 }
1193                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1194                 copy = clamp_t(size_t, len, 1, sizeof(buf));
1195                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1196                 ret = copy_to_user(release, buf, copy + 1);
1197         }
1198         return ret;
1199 }
1200
1201 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1202 {
1203         int errno = 0;
1204
1205         down_read(&uts_sem);
1206         if (copy_to_user(name, utsname(), sizeof *name))
1207                 errno = -EFAULT;
1208         up_read(&uts_sem);
1209
1210         if (!errno && override_release(name->release, sizeof(name->release)))
1211                 errno = -EFAULT;
1212         if (!errno && override_architecture(name))
1213                 errno = -EFAULT;
1214         return errno;
1215 }
1216
1217 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1218 /*
1219  * Old cruft
1220  */
1221 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1222 {
1223         int error = 0;
1224
1225         if (!name)
1226                 return -EFAULT;
1227
1228         down_read(&uts_sem);
1229         if (copy_to_user(name, utsname(), sizeof(*name)))
1230                 error = -EFAULT;
1231         up_read(&uts_sem);
1232
1233         if (!error && override_release(name->release, sizeof(name->release)))
1234                 error = -EFAULT;
1235         if (!error && override_architecture(name))
1236                 error = -EFAULT;
1237         return error;
1238 }
1239
1240 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1241 {
1242         int error;
1243
1244         if (!name)
1245                 return -EFAULT;
1246         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1247                 return -EFAULT;
1248
1249         down_read(&uts_sem);
1250         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1251                                __OLD_UTS_LEN);
1252         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1253         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1254                                 __OLD_UTS_LEN);
1255         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1256         error |= __copy_to_user(&name->release, &utsname()->release,
1257                                 __OLD_UTS_LEN);
1258         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1259         error |= __copy_to_user(&name->version, &utsname()->version,
1260                                 __OLD_UTS_LEN);
1261         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1262         error |= __copy_to_user(&name->machine, &utsname()->machine,
1263                                 __OLD_UTS_LEN);
1264         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1265         up_read(&uts_sem);
1266
1267         if (!error && override_architecture(name))
1268                 error = -EFAULT;
1269         if (!error && override_release(name->release, sizeof(name->release)))
1270                 error = -EFAULT;
1271         return error ? -EFAULT : 0;
1272 }
1273 #endif
1274
1275 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1276 {
1277         int errno;
1278         char tmp[__NEW_UTS_LEN];
1279
1280         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1281                 return -EPERM;
1282
1283         if (len < 0 || len > __NEW_UTS_LEN)
1284                 return -EINVAL;
1285         down_write(&uts_sem);
1286         errno = -EFAULT;
1287         if (!copy_from_user(tmp, name, len)) {
1288                 struct new_utsname *u = utsname();
1289
1290                 memcpy(u->nodename, tmp, len);
1291                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1292                 errno = 0;
1293         }
1294         uts_proc_notify(UTS_PROC_HOSTNAME);
1295         up_write(&uts_sem);
1296         return errno;
1297 }
1298
1299 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1300
1301 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1302 {
1303         int i, errno;
1304         struct new_utsname *u;
1305
1306         if (len < 0)
1307                 return -EINVAL;
1308         down_read(&uts_sem);
1309         u = utsname();
1310         i = 1 + strlen(u->nodename);
1311         if (i > len)
1312                 i = len;
1313         errno = 0;
1314         if (copy_to_user(name, u->nodename, i))
1315                 errno = -EFAULT;
1316         up_read(&uts_sem);
1317         return errno;
1318 }
1319
1320 #endif
1321
1322 /*
1323  * Only setdomainname; getdomainname can be implemented by calling
1324  * uname()
1325  */
1326 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1327 {
1328         int errno;
1329         char tmp[__NEW_UTS_LEN];
1330
1331         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1332                 return -EPERM;
1333         if (len < 0 || len > __NEW_UTS_LEN)
1334                 return -EINVAL;
1335
1336         down_write(&uts_sem);
1337         errno = -EFAULT;
1338         if (!copy_from_user(tmp, name, len)) {
1339                 struct new_utsname *u = utsname();
1340
1341                 memcpy(u->domainname, tmp, len);
1342                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1343                 errno = 0;
1344         }
1345         uts_proc_notify(UTS_PROC_DOMAINNAME);
1346         up_write(&uts_sem);
1347         return errno;
1348 }
1349
1350 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1351 {
1352         struct rlimit value;
1353         int ret;
1354
1355         ret = do_prlimit(current, resource, NULL, &value);
1356         if (!ret)
1357                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1358
1359         return ret;
1360 }
1361
1362 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1363
1364 /*
1365  *      Back compatibility for getrlimit. Needed for some apps.
1366  */
1367
1368 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1369                 struct rlimit __user *, rlim)
1370 {
1371         struct rlimit x;
1372         if (resource >= RLIM_NLIMITS)
1373                 return -EINVAL;
1374
1375         task_lock(current->group_leader);
1376         x = current->signal->rlim[resource];
1377         task_unlock(current->group_leader);
1378         if (x.rlim_cur > 0x7FFFFFFF)
1379                 x.rlim_cur = 0x7FFFFFFF;
1380         if (x.rlim_max > 0x7FFFFFFF)
1381                 x.rlim_max = 0x7FFFFFFF;
1382         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1383 }
1384
1385 #endif
1386
1387 static inline bool rlim64_is_infinity(__u64 rlim64)
1388 {
1389 #if BITS_PER_LONG < 64
1390         return rlim64 >= ULONG_MAX;
1391 #else
1392         return rlim64 == RLIM64_INFINITY;
1393 #endif
1394 }
1395
1396 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1397 {
1398         if (rlim->rlim_cur == RLIM_INFINITY)
1399                 rlim64->rlim_cur = RLIM64_INFINITY;
1400         else
1401                 rlim64->rlim_cur = rlim->rlim_cur;
1402         if (rlim->rlim_max == RLIM_INFINITY)
1403                 rlim64->rlim_max = RLIM64_INFINITY;
1404         else
1405                 rlim64->rlim_max = rlim->rlim_max;
1406 }
1407
1408 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1409 {
1410         if (rlim64_is_infinity(rlim64->rlim_cur))
1411                 rlim->rlim_cur = RLIM_INFINITY;
1412         else
1413                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1414         if (rlim64_is_infinity(rlim64->rlim_max))
1415                 rlim->rlim_max = RLIM_INFINITY;
1416         else
1417                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1418 }
1419
1420 /* make sure you are allowed to change @tsk limits before calling this */
1421 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1422                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1423 {
1424         struct rlimit *rlim;
1425         int retval = 0;
1426
1427         if (resource >= RLIM_NLIMITS)
1428                 return -EINVAL;
1429         if (new_rlim) {
1430                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1431                         return -EINVAL;
1432                 if (resource == RLIMIT_NOFILE &&
1433                                 new_rlim->rlim_max > sysctl_nr_open)
1434                         return -EPERM;
1435         }
1436
1437         /* protect tsk->signal and tsk->sighand from disappearing */
1438         read_lock(&tasklist_lock);
1439         if (!tsk->sighand) {
1440                 retval = -ESRCH;
1441                 goto out;
1442         }
1443
1444         rlim = tsk->signal->rlim + resource;
1445         task_lock(tsk->group_leader);
1446         if (new_rlim) {
1447                 /* Keep the capable check against init_user_ns until
1448                    cgroups can contain all limits */
1449                 if (new_rlim->rlim_max > rlim->rlim_max &&
1450                                 !capable(CAP_SYS_RESOURCE))
1451                         retval = -EPERM;
1452                 if (!retval)
1453                         retval = security_task_setrlimit(tsk->group_leader,
1454                                         resource, new_rlim);
1455                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1456                         /*
1457                          * The caller is asking for an immediate RLIMIT_CPU
1458                          * expiry.  But we use the zero value to mean "it was
1459                          * never set".  So let's cheat and make it one second
1460                          * instead
1461                          */
1462                         new_rlim->rlim_cur = 1;
1463                 }
1464         }
1465         if (!retval) {
1466                 if (old_rlim)
1467                         *old_rlim = *rlim;
1468                 if (new_rlim)
1469                         *rlim = *new_rlim;
1470         }
1471         task_unlock(tsk->group_leader);
1472
1473         /*
1474          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1475          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1476          * very long-standing error, and fixing it now risks breakage of
1477          * applications, so we live with it
1478          */
1479          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1480                          new_rlim->rlim_cur != RLIM_INFINITY)
1481                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1482 out:
1483         read_unlock(&tasklist_lock);
1484         return retval;
1485 }
1486
1487 /* rcu lock must be held */
1488 static int check_prlimit_permission(struct task_struct *task)
1489 {
1490         const struct cred *cred = current_cred(), *tcred;
1491
1492         if (current == task)
1493                 return 0;
1494
1495         tcred = __task_cred(task);
1496         if (cred->user->user_ns == tcred->user->user_ns &&
1497             (cred->uid == tcred->euid &&
1498              cred->uid == tcred->suid &&
1499              cred->uid == tcred->uid  &&
1500              cred->gid == tcred->egid &&
1501              cred->gid == tcred->sgid &&
1502              cred->gid == tcred->gid))
1503                 return 0;
1504         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1505                 return 0;
1506
1507         return -EPERM;
1508 }
1509
1510 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1511                 const struct rlimit64 __user *, new_rlim,
1512                 struct rlimit64 __user *, old_rlim)
1513 {
1514         struct rlimit64 old64, new64;
1515         struct rlimit old, new;
1516         struct task_struct *tsk;
1517         int ret;
1518
1519         if (new_rlim) {
1520                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1521                         return -EFAULT;
1522                 rlim64_to_rlim(&new64, &new);
1523         }
1524
1525         rcu_read_lock();
1526         tsk = pid ? find_task_by_vpid(pid) : current;
1527         if (!tsk) {
1528                 rcu_read_unlock();
1529                 return -ESRCH;
1530         }
1531         ret = check_prlimit_permission(tsk);
1532         if (ret) {
1533                 rcu_read_unlock();
1534                 return ret;
1535         }
1536         get_task_struct(tsk);
1537         rcu_read_unlock();
1538
1539         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1540                         old_rlim ? &old : NULL);
1541
1542         if (!ret && old_rlim) {
1543                 rlim_to_rlim64(&old, &old64);
1544                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1545                         ret = -EFAULT;
1546         }
1547
1548         put_task_struct(tsk);
1549         return ret;
1550 }
1551
1552 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1553 {
1554         struct rlimit new_rlim;
1555
1556         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1557                 return -EFAULT;
1558         return do_prlimit(current, resource, &new_rlim, NULL);
1559 }
1560
1561 /*
1562  * It would make sense to put struct rusage in the task_struct,
1563  * except that would make the task_struct be *really big*.  After
1564  * task_struct gets moved into malloc'ed memory, it would
1565  * make sense to do this.  It will make moving the rest of the information
1566  * a lot simpler!  (Which we're not doing right now because we're not
1567  * measuring them yet).
1568  *
1569  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1570  * races with threads incrementing their own counters.  But since word
1571  * reads are atomic, we either get new values or old values and we don't
1572  * care which for the sums.  We always take the siglock to protect reading
1573  * the c* fields from p->signal from races with exit.c updating those
1574  * fields when reaping, so a sample either gets all the additions of a
1575  * given child after it's reaped, or none so this sample is before reaping.
1576  *
1577  * Locking:
1578  * We need to take the siglock for CHILDEREN, SELF and BOTH
1579  * for  the cases current multithreaded, non-current single threaded
1580  * non-current multithreaded.  Thread traversal is now safe with
1581  * the siglock held.
1582  * Strictly speaking, we donot need to take the siglock if we are current and
1583  * single threaded,  as no one else can take our signal_struct away, no one
1584  * else can  reap the  children to update signal->c* counters, and no one else
1585  * can race with the signal-> fields. If we do not take any lock, the
1586  * signal-> fields could be read out of order while another thread was just
1587  * exiting. So we should  place a read memory barrier when we avoid the lock.
1588  * On the writer side,  write memory barrier is implied in  __exit_signal
1589  * as __exit_signal releases  the siglock spinlock after updating the signal->
1590  * fields. But we don't do this yet to keep things simple.
1591  *
1592  */
1593
1594 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1595 {
1596         r->ru_nvcsw += t->nvcsw;
1597         r->ru_nivcsw += t->nivcsw;
1598         r->ru_minflt += t->min_flt;
1599         r->ru_majflt += t->maj_flt;
1600         r->ru_inblock += task_io_get_inblock(t);
1601         r->ru_oublock += task_io_get_oublock(t);
1602 }
1603
1604 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1605 {
1606         struct task_struct *t;
1607         unsigned long flags;
1608         cputime_t tgutime, tgstime, utime, stime;
1609         unsigned long maxrss = 0;
1610
1611         memset((char *) r, 0, sizeof *r);
1612         utime = stime = cputime_zero;
1613
1614         if (who == RUSAGE_THREAD) {
1615                 task_times(current, &utime, &stime);
1616                 accumulate_thread_rusage(p, r);
1617                 maxrss = p->signal->maxrss;
1618                 goto out;
1619         }
1620
1621         if (!lock_task_sighand(p, &flags))
1622                 return;
1623
1624         switch (who) {
1625                 case RUSAGE_BOTH:
1626                 case RUSAGE_CHILDREN:
1627                         utime = p->signal->cutime;
1628                         stime = p->signal->cstime;
1629                         r->ru_nvcsw = p->signal->cnvcsw;
1630                         r->ru_nivcsw = p->signal->cnivcsw;
1631                         r->ru_minflt = p->signal->cmin_flt;
1632                         r->ru_majflt = p->signal->cmaj_flt;
1633                         r->ru_inblock = p->signal->cinblock;
1634                         r->ru_oublock = p->signal->coublock;
1635                         maxrss = p->signal->cmaxrss;
1636
1637                         if (who == RUSAGE_CHILDREN)
1638                                 break;
1639
1640                 case RUSAGE_SELF:
1641                         thread_group_times(p, &tgutime, &tgstime);
1642                         utime = cputime_add(utime, tgutime);
1643                         stime = cputime_add(stime, tgstime);
1644                         r->ru_nvcsw += p->signal->nvcsw;
1645                         r->ru_nivcsw += p->signal->nivcsw;
1646                         r->ru_minflt += p->signal->min_flt;
1647                         r->ru_majflt += p->signal->maj_flt;
1648                         r->ru_inblock += p->signal->inblock;
1649                         r->ru_oublock += p->signal->oublock;
1650                         if (maxrss < p->signal->maxrss)
1651                                 maxrss = p->signal->maxrss;
1652                         t = p;
1653                         do {
1654                                 accumulate_thread_rusage(t, r);
1655                                 t = next_thread(t);
1656                         } while (t != p);
1657                         break;
1658
1659                 default:
1660                         BUG();
1661         }
1662         unlock_task_sighand(p, &flags);
1663
1664 out:
1665         cputime_to_timeval(utime, &r->ru_utime);
1666         cputime_to_timeval(stime, &r->ru_stime);
1667
1668         if (who != RUSAGE_CHILDREN) {
1669                 struct mm_struct *mm = get_task_mm(p);
1670                 if (mm) {
1671                         setmax_mm_hiwater_rss(&maxrss, mm);
1672                         mmput(mm);
1673                 }
1674         }
1675         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1676 }
1677
1678 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1679 {
1680         struct rusage r;
1681         k_getrusage(p, who, &r);
1682         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1683 }
1684
1685 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1686 {
1687         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1688             who != RUSAGE_THREAD)
1689                 return -EINVAL;
1690         return getrusage(current, who, ru);
1691 }
1692
1693 SYSCALL_DEFINE1(umask, int, mask)
1694 {
1695         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1696         return mask;
1697 }
1698
1699 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1700                 unsigned long, arg4, unsigned long, arg5)
1701 {
1702         struct task_struct *me = current;
1703         unsigned char comm[sizeof(me->comm)];
1704         long error;
1705
1706         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1707         if (error != -ENOSYS)
1708                 return error;
1709
1710         error = 0;
1711         switch (option) {
1712                 case PR_SET_PDEATHSIG:
1713                         if (!valid_signal(arg2)) {
1714                                 error = -EINVAL;
1715                                 break;
1716                         }
1717                         me->pdeath_signal = arg2;
1718                         error = 0;
1719                         break;
1720                 case PR_GET_PDEATHSIG:
1721                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1722                         break;
1723                 case PR_GET_DUMPABLE:
1724                         error = get_dumpable(me->mm);
1725                         break;
1726                 case PR_SET_DUMPABLE:
1727                         if (arg2 < 0 || arg2 > 1) {
1728                                 error = -EINVAL;
1729                                 break;
1730                         }
1731                         set_dumpable(me->mm, arg2);
1732                         error = 0;
1733                         break;
1734
1735                 case PR_SET_UNALIGN:
1736                         error = SET_UNALIGN_CTL(me, arg2);
1737                         break;
1738                 case PR_GET_UNALIGN:
1739                         error = GET_UNALIGN_CTL(me, arg2);
1740                         break;
1741                 case PR_SET_FPEMU:
1742                         error = SET_FPEMU_CTL(me, arg2);
1743                         break;
1744                 case PR_GET_FPEMU:
1745                         error = GET_FPEMU_CTL(me, arg2);
1746                         break;
1747                 case PR_SET_FPEXC:
1748                         error = SET_FPEXC_CTL(me, arg2);
1749                         break;
1750                 case PR_GET_FPEXC:
1751                         error = GET_FPEXC_CTL(me, arg2);
1752                         break;
1753                 case PR_GET_TIMING:
1754                         error = PR_TIMING_STATISTICAL;
1755                         break;
1756                 case PR_SET_TIMING:
1757                         if (arg2 != PR_TIMING_STATISTICAL)
1758                                 error = -EINVAL;
1759                         else
1760                                 error = 0;
1761                         break;
1762
1763                 case PR_SET_NAME:
1764                         comm[sizeof(me->comm)-1] = 0;
1765                         if (strncpy_from_user(comm, (char __user *)arg2,
1766                                               sizeof(me->comm) - 1) < 0)
1767                                 return -EFAULT;
1768                         set_task_comm(me, comm);
1769                         proc_comm_connector(me);
1770                         return 0;
1771                 case PR_GET_NAME:
1772                         get_task_comm(comm, me);
1773                         if (copy_to_user((char __user *)arg2, comm,
1774                                          sizeof(comm)))
1775                                 return -EFAULT;
1776                         return 0;
1777                 case PR_GET_ENDIAN:
1778                         error = GET_ENDIAN(me, arg2);
1779                         break;
1780                 case PR_SET_ENDIAN:
1781                         error = SET_ENDIAN(me, arg2);
1782                         break;
1783
1784                 case PR_GET_SECCOMP:
1785                         error = prctl_get_seccomp();
1786                         break;
1787                 case PR_SET_SECCOMP:
1788                         error = prctl_set_seccomp(arg2);
1789                         break;
1790                 case PR_GET_TSC:
1791                         error = GET_TSC_CTL(arg2);
1792                         break;
1793                 case PR_SET_TSC:
1794                         error = SET_TSC_CTL(arg2);
1795                         break;
1796                 case PR_TASK_PERF_EVENTS_DISABLE:
1797                         error = perf_event_task_disable();
1798                         break;
1799                 case PR_TASK_PERF_EVENTS_ENABLE:
1800                         error = perf_event_task_enable();
1801                         break;
1802                 case PR_GET_TIMERSLACK:
1803                         error = current->timer_slack_ns;
1804                         break;
1805                 case PR_SET_TIMERSLACK:
1806                         if (arg2 <= 0)
1807                                 current->timer_slack_ns =
1808                                         current->default_timer_slack_ns;
1809                         else
1810                                 current->timer_slack_ns = arg2;
1811                         error = 0;
1812                         break;
1813                 case PR_MCE_KILL:
1814                         if (arg4 | arg5)
1815                                 return -EINVAL;
1816                         switch (arg2) {
1817                         case PR_MCE_KILL_CLEAR:
1818                                 if (arg3 != 0)
1819                                         return -EINVAL;
1820                                 current->flags &= ~PF_MCE_PROCESS;
1821                                 break;
1822                         case PR_MCE_KILL_SET:
1823                                 current->flags |= PF_MCE_PROCESS;
1824                                 if (arg3 == PR_MCE_KILL_EARLY)
1825                                         current->flags |= PF_MCE_EARLY;
1826                                 else if (arg3 == PR_MCE_KILL_LATE)
1827                                         current->flags &= ~PF_MCE_EARLY;
1828                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1829                                         current->flags &=
1830                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1831                                 else
1832                                         return -EINVAL;
1833                                 break;
1834                         default:
1835                                 return -EINVAL;
1836                         }
1837                         error = 0;
1838                         break;
1839                 case PR_MCE_KILL_GET:
1840                         if (arg2 | arg3 | arg4 | arg5)
1841                                 return -EINVAL;
1842                         if (current->flags & PF_MCE_PROCESS)
1843                                 error = (current->flags & PF_MCE_EARLY) ?
1844                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1845                         else
1846                                 error = PR_MCE_KILL_DEFAULT;
1847                         break;
1848                 default:
1849                         error = -EINVAL;
1850                         break;
1851         }
1852         return error;
1853 }
1854
1855 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1856                 struct getcpu_cache __user *, unused)
1857 {
1858         int err = 0;
1859         int cpu = raw_smp_processor_id();
1860         if (cpup)
1861                 err |= put_user(cpu, cpup);
1862         if (nodep)
1863                 err |= put_user(cpu_to_node(cpu), nodep);
1864         return err ? -EFAULT : 0;
1865 }
1866
1867 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1868
1869 static void argv_cleanup(struct subprocess_info *info)
1870 {
1871         argv_free(info->argv);
1872 }
1873
1874 /**
1875  * orderly_poweroff - Trigger an orderly system poweroff
1876  * @force: force poweroff if command execution fails
1877  *
1878  * This may be called from any context to trigger a system shutdown.
1879  * If the orderly shutdown fails, it will force an immediate shutdown.
1880  */
1881 int orderly_poweroff(bool force)
1882 {
1883         int argc;
1884         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1885         static char *envp[] = {
1886                 "HOME=/",
1887                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1888                 NULL
1889         };
1890         int ret = -ENOMEM;
1891         struct subprocess_info *info;
1892
1893         if (argv == NULL) {
1894                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1895                        __func__, poweroff_cmd);
1896                 goto out;
1897         }
1898
1899         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1900         if (info == NULL) {
1901                 argv_free(argv);
1902                 goto out;
1903         }
1904
1905         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1906
1907         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1908
1909   out:
1910         if (ret && force) {
1911                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1912                        "forcing the issue\n");
1913
1914                 /* I guess this should try to kick off some daemon to
1915                    sync and poweroff asap.  Or not even bother syncing
1916                    if we're doing an emergency shutdown? */
1917                 emergency_sync();
1918                 kernel_power_off();
1919         }
1920
1921         return ret;
1922 }
1923 EXPORT_SYMBOL_GPL(orderly_poweroff);