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