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