kernel/sys.c

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/mm.h>
   9 #include <linux/utsname.h>
  10 #include <linux/mman.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/notifier.h>
  13 #include <linux/reboot.h>
  14 #include <linux/prctl.h>
  15 #include <linux/highuid.h>
  16 #include <linux/fs.h>
  17 #include <linux/perf_counter.h>
  18 #include <linux/resource.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kexec.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/capability.h>
  23 #include <linux/device.h>
  24 #include <linux/key.h>
  25 #include <linux/times.h>
  26 #include <linux/posix-timers.h>
  27 #include <linux/security.h>
  28 #include <linux/dcookies.h>
  29 #include <linux/suspend.h>
  30 #include <linux/tty.h>
  31 #include <linux/signal.h>
  32 #include <linux/cn_proc.h>
  33 #include <linux/getcpu.h>
  34 #include <linux/task_io_accounting_ops.h>
  35 #include <linux/seccomp.h>
  36 #include <linux/hardirq.h>
  37 #include <linux/cpu.h>
  38 #include <linux/ptrace.h>
  39
  40 #include <linux/compat.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/rt_lock.h>
  43 #include <linux/kprobes.h>
  44 #include <linux/user_namespace.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/io.h>
  48 #include <asm/unistd.h>
  49
  50 #ifndef SET_UNALIGN_CTL
  51 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  52 #endif
  53 #ifndef GET_UNALIGN_CTL
  54 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  55 #endif
  56 #ifndef SET_FPEMU_CTL
  57 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  58 #endif
  59 #ifndef GET_FPEMU_CTL
  60 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  61 #endif
  62 #ifndef SET_FPEXC_CTL
  63 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  64 #endif
  65 #ifndef GET_FPEXC_CTL
  66 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  67 #endif
  68 #ifndef GET_ENDIAN
  69 # define GET_ENDIAN(a,b)        (-EINVAL)
  70 #endif
  71 #ifndef SET_ENDIAN
  72 # define SET_ENDIAN(a,b)        (-EINVAL)
  73 #endif
  74 #ifndef GET_TSC_CTL
  75 # define GET_TSC_CTL(a)         (-EINVAL)
  76 #endif
  77 #ifndef SET_TSC_CTL
  78 # define SET_TSC_CTL(a)         (-EINVAL)
  79 #endif
  80
  81 /*
  82  * this is where the system-wide overflow UID and GID are defined, for
  83  * architectures that now have 32-bit UID/GID but didn't in the past
  84  */
  85
  86 int overflowuid = DEFAULT_OVERFLOWUID;
  87 int overflowgid = DEFAULT_OVERFLOWGID;
  88
  89 #ifdef CONFIG_UID16
  90 EXPORT_SYMBOL(overflowuid);
  91 EXPORT_SYMBOL(overflowgid);
  92 #endif
  93
  94 /*
  95  * the same as above, but for filesystems which can only store a 16-bit
  96  * UID and GID. as such, this is needed on all architectures
  97  */
  98
  99 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 100 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 101
 102 EXPORT_SYMBOL(fs_overflowuid);
 103 EXPORT_SYMBOL(fs_overflowgid);
 104
 105 /*
 106  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 107  */
 108
 109 int C_A_D = 1;
 110 struct pid *cad_pid;
 111 EXPORT_SYMBOL(cad_pid);
 112
 113 /*
 114  * If set, this is used for preparing the system to power off.
 115  */
 116
 117 void (*pm_power_off_prepare)(void);
 118
 119 /*
 120  * set the priority of a task
 121  * - the caller must hold the RCU read lock
 122  */
 123 static int set_one_prio(struct task_struct *p, int niceval, int error)
 124 {
 125         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 126         int no_nice;
 127
 128         if (pcred->uid  != cred->euid &&
 129             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
 130                 error = -EPERM;
 131                 goto out;
 132         }
 133         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 134                 error = -EACCES;
 135                 goto out;
 136         }
 137         no_nice = security_task_setnice(p, niceval);
 138         if (no_nice) {
 139                 error = no_nice;
 140                 goto out;
 141         }
 142         if (error == -ESRCH)
 143                 error = 0;
 144         set_user_nice(p, niceval);
 145 out:
 146         return error;
 147 }
 148
 149 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 150 {
 151         struct task_struct *g, *p;
 152         struct user_struct *user;
 153         const struct cred *cred = current_cred();
 154         int error = -EINVAL;
 155         struct pid *pgrp;
 156
 157         if (which > PRIO_USER || which < PRIO_PROCESS)
 158                 goto out;
 159
 160         /* normalize: avoid signed division (rounding problems) */
 161         error = -ESRCH;
 162         if (niceval < -20)
 163                 niceval = -20;
 164         if (niceval > 19)
 165                 niceval = 19;
 166
 167         read_lock(&tasklist_lock);
 168         switch (which) {
 169                 case PRIO_PROCESS:
 170                         if (who)
 171                                 p = find_task_by_vpid(who);
 172                         else
 173                                 p = current;
 174                         if (p)
 175                                 error = set_one_prio(p, niceval, error);
 176                         break;
 177                 case PRIO_PGRP:
 178                         if (who)
 179                                 pgrp = find_vpid(who);
 180                         else
 181                                 pgrp = task_pgrp(current);
 182                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 183                                 error = set_one_prio(p, niceval, error);
 184                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 185                         break;
 186                 case PRIO_USER:
 187                         user = (struct user_struct *) cred->user;
 188                         if (!who)
 189                                 who = cred->uid;
 190                         else if ((who != cred->uid) &&
 191                                  !(user = find_user(who)))
 192                                 goto out_unlock;        /* No processes for this user */
 193
 194                         do_each_thread(g, p)
 195                                 if (__task_cred(p)->uid == who)
 196                                         error = set_one_prio(p, niceval, error);
 197                         while_each_thread(g, p);
 198                         if (who != cred->uid)
 199                                 free_uid(user);         /* For find_user() */
 200                         break;
 201         }
 202 out_unlock:
 203         read_unlock(&tasklist_lock);
 204 out:
 205         return error;
 206 }
 207
 208 /*
 209  * Ugh. To avoid negative return values, "getpriority()" will
 210  * not return the normal nice-value, but a negated value that
 211  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 212  * to stay compatible.
 213  */
 214 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 215 {
 216         struct task_struct *g, *p;
 217         struct user_struct *user;
 218         const struct cred *cred = current_cred();
 219         long niceval, retval = -ESRCH;
 220         struct pid *pgrp;
 221
 222         if (which > PRIO_USER || which < PRIO_PROCESS)
 223                 return -EINVAL;
 224
 225         read_lock(&tasklist_lock);
 226         switch (which) {
 227                 case PRIO_PROCESS:
 228                         if (who)
 229                                 p = find_task_by_vpid(who);
 230                         else
 231                                 p = current;
 232                         if (p) {
 233                                 niceval = 20 - task_nice(p);
 234                                 if (niceval > retval)
 235                                         retval = niceval;
 236                         }
 237                         break;
 238                 case PRIO_PGRP:
 239                         if (who)
 240                                 pgrp = find_vpid(who);
 241                         else
 242                                 pgrp = task_pgrp(current);
 243                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 244                                 niceval = 20 - task_nice(p);
 245                                 if (niceval > retval)
 246                                         retval = niceval;
 247                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 248                         break;
 249                 case PRIO_USER:
 250                         user = (struct user_struct *) cred->user;
 251                         if (!who)
 252                                 who = cred->uid;
 253                         else if ((who != cred->uid) &&
 254                                  !(user = find_user(who)))
 255                                 goto out_unlock;        /* No processes for this user */
 256
 257                         do_each_thread(g, p)
 258                                 if (__task_cred(p)->uid == who) {
 259                                         niceval = 20 - task_nice(p);
 260                                         if (niceval > retval)
 261                                                 retval = niceval;
 262                                 }
 263                         while_each_thread(g, p);
 264                         if (who != cred->uid)
 265                                 free_uid(user);         /* for find_user() */
 266                         break;
 267         }
 268 out_unlock:
 269         read_unlock(&tasklist_lock);
 270
 271         return retval;
 272 }
 273
 274 /**
 275  *      emergency_restart - reboot the system
 276  *
 277  *      Without shutting down any hardware or taking any locks
 278  *      reboot the system.  This is called when we know we are in
 279  *      trouble so this is our best effort to reboot.  This is
 280  *      safe to call in interrupt context.
 281  */
 282 void emergency_restart(void)
 283 {
 284         /*
 285          * Call the notifier chain if we are not in an
 286          * atomic context:
 287          */
 288 #ifdef CONFIG_PREEMPT
 289         if (!in_atomic() && !irqs_disabled())
 290                 blocking_notifier_call_chain(&reboot_notifier_list,
 291                                              SYS_RESTART, NULL);
 292 #endif
 293         machine_emergency_restart();
 294 }
 295 EXPORT_SYMBOL_GPL(emergency_restart);
 296
 297 void kernel_restart_prepare(char *cmd)
 298 {
 299         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 300         system_state = SYSTEM_RESTART;
 301         device_shutdown();
 302         sysdev_shutdown();
 303 }
 304
 305 /**
 306  *      kernel_restart - reboot the system
 307  *      @cmd: pointer to buffer containing command to execute for restart
 308  *              or %NULL
 309  *
 310  *      Shutdown everything and perform a clean reboot.
 311  *      This is not safe to call in interrupt context.
 312  */
 313 void kernel_restart(char *cmd)
 314 {
 315         kernel_restart_prepare(cmd);
 316         if (!cmd)
 317                 printk(KERN_EMERG "Restarting system.\n");
 318         else
 319                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 320         machine_restart(cmd);
 321 }
 322 EXPORT_SYMBOL_GPL(kernel_restart);
 323
 324 static void kernel_shutdown_prepare(enum system_states state)
 325 {
 326         blocking_notifier_call_chain(&reboot_notifier_list,
 327                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 328         system_state = state;
 329         device_shutdown();
 330 }
 331 /**
 332  *      kernel_halt - halt the system
 333  *
 334  *      Shutdown everything and perform a clean system halt.
 335  */
 336 void kernel_halt(void)
 337 {
 338         kernel_shutdown_prepare(SYSTEM_HALT);
 339         sysdev_shutdown();
 340         printk(KERN_EMERG "System halted.\n");
 341         machine_halt();
 342 }
 343
 344 EXPORT_SYMBOL_GPL(kernel_halt);
 345
 346 /**
 347  *      kernel_power_off - power_off the system
 348  *
 349  *      Shutdown everything and perform a clean system power_off.
 350  */
 351 void kernel_power_off(void)
 352 {
 353         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 354         if (pm_power_off_prepare)
 355                 pm_power_off_prepare();
 356         disable_nonboot_cpus();
 357         sysdev_shutdown();
 358         printk(KERN_EMERG "Power down.\n");
 359         machine_power_off();
 360 }
 361 EXPORT_SYMBOL_GPL(kernel_power_off);
 362 /*
 363  * Reboot system call: for obvious reasons only root may call it,
 364  * and even root needs to set up some magic numbers in the registers
 365  * so that some mistake won't make this reboot the whole machine.
 366  * You can also set the meaning of the ctrl-alt-del-key here.
 367  *
 368  * reboot doesn't sync: do that yourself before calling this.
 369  */
 370 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 371                 void __user *, arg)
 372 {
 373         char buffer[256];
 374
 375         /* We only trust the superuser with rebooting the system. */
 376         if (!capable(CAP_SYS_BOOT))
 377                 return -EPERM;
 378
 379         /* For safety, we require "magic" arguments. */
 380         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 381             (magic2 != LINUX_REBOOT_MAGIC2 &&
 382                         magic2 != LINUX_REBOOT_MAGIC2A &&
 383                         magic2 != LINUX_REBOOT_MAGIC2B &&
 384                         magic2 != LINUX_REBOOT_MAGIC2C))
 385                 return -EINVAL;
 386
 387         /* Instead of trying to make the power_off code look like
 388          * halt when pm_power_off is not set do it the easy way.
 389          */
 390         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 391                 cmd = LINUX_REBOOT_CMD_HALT;
 392
 393         lock_kernel();
 394         switch (cmd) {
 395         case LINUX_REBOOT_CMD_RESTART:
 396                 kernel_restart(NULL);
 397                 break;
 398
 399         case LINUX_REBOOT_CMD_CAD_ON:
 400                 C_A_D = 1;
 401                 break;
 402
 403         case LINUX_REBOOT_CMD_CAD_OFF:
 404                 C_A_D = 0;
 405                 break;
 406
 407         case LINUX_REBOOT_CMD_HALT:
 408                 kernel_halt();
 409                 unlock_kernel();
 410                 do_exit(0);
 411                 break;
 412
 413         case LINUX_REBOOT_CMD_POWER_OFF:
 414                 kernel_power_off();
 415                 unlock_kernel();
 416                 do_exit(0);
 417                 break;
 418
 419         case LINUX_REBOOT_CMD_RESTART2:
 420                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 421                         unlock_kernel();
 422                         return -EFAULT;
 423                 }
 424                 buffer[sizeof(buffer) - 1] = '\0';
 425
 426                 kernel_restart(buffer);
 427                 break;
 428
 429 #ifdef CONFIG_KEXEC
 430         case LINUX_REBOOT_CMD_KEXEC:
 431                 {
 432                         int ret;
 433                         ret = kernel_kexec();
 434                         unlock_kernel();
 435                         return ret;
 436                 }
 437 #endif
 438
 439 #ifdef CONFIG_HIBERNATION
 440         case LINUX_REBOOT_CMD_SW_SUSPEND:
 441                 {
 442                         int ret = hibernate();
 443                         unlock_kernel();
 444                         return ret;
 445                 }
 446 #endif
 447
 448         default:
 449                 unlock_kernel();
 450                 return -EINVAL;
 451         }
 452         unlock_kernel();
 453         return 0;
 454 }
 455
 456 static void deferred_cad(struct work_struct *dummy)
 457 {
 458         kernel_restart(NULL);
 459 }
 460
 461 /*
 462  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 463  * As it's called within an interrupt, it may NOT sync: the only choice
 464  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 465  */
 466 void ctrl_alt_del(void)
 467 {
 468         static DECLARE_WORK(cad_work, deferred_cad);
 469
 470         if (C_A_D)
 471                 schedule_work(&cad_work);
 472         else
 473                 kill_cad_pid(SIGINT, 1);
 474 }
 475
 476 /*
 477  * Unprivileged users may change the real gid to the effective gid
 478  * or vice versa.  (BSD-style)
 479  *
 480  * If you set the real gid at all, or set the effective gid to a value not
 481  * equal to the real gid, then the saved gid is set to the new effective gid.
 482  *
 483  * This makes it possible for a setgid program to completely drop its
 484  * privileges, which is often a useful assertion to make when you are doing
 485  * a security audit over a program.
 486  *
 487  * The general idea is that a program which uses just setregid() will be
 488  * 100% compatible with BSD.  A program which uses just setgid() will be
 489  * 100% compatible with POSIX with saved IDs.
 490  *
 491  * SMP: There are not races, the GIDs are checked only by filesystem
 492  *      operations (as far as semantic preservation is concerned).
 493  */
 494 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 495 {
 496         const struct cred *old;
 497         struct cred *new;
 498         int retval;
 499
 500         new = prepare_creds();
 501         if (!new)
 502                 return -ENOMEM;
 503         old = current_cred();
 504
 505         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 506         if (retval)
 507                 goto error;
 508
 509         retval = -EPERM;
 510         if (rgid != (gid_t) -1) {
 511                 if (old->gid == rgid ||
 512                     old->egid == rgid ||
 513                     capable(CAP_SETGID))
 514                         new->gid = rgid;
 515                 else
 516                         goto error;
 517         }
 518         if (egid != (gid_t) -1) {
 519                 if (old->gid == egid ||
 520                     old->egid == egid ||
 521                     old->sgid == egid ||
 522                     capable(CAP_SETGID))
 523                         new->egid = egid;
 524                 else
 525                         goto error;
 526         }
 527
 528         if (rgid != (gid_t) -1 ||
 529             (egid != (gid_t) -1 && egid != old->gid))
 530                 new->sgid = new->egid;
 531         new->fsgid = new->egid;
 532
 533         return commit_creds(new);
 534
 535 error:
 536         abort_creds(new);
 537         return retval;
 538 }
 539
 540 /*
 541  * setgid() is implemented like SysV w/ SAVED_IDS
 542  *
 543  * SMP: Same implicit races as above.
 544  */
 545 SYSCALL_DEFINE1(setgid, gid_t, gid)
 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 = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 557         if (retval)
 558                 goto error;
 559
 560         retval = -EPERM;
 561         if (capable(CAP_SETGID))
 562                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 563         else if (gid == old->gid || gid == old->sgid)
 564                 new->egid = new->fsgid = gid;
 565         else
 566                 goto error;
 567
 568         return commit_creds(new);
 569
 570 error:
 571         abort_creds(new);
 572         return retval;
 573 }
 574
 575 /*
 576  * change the user struct in a credentials set to match the new UID
 577  */
 578 static int set_user(struct cred *new)
 579 {
 580         struct user_struct *new_user;
 581
 582         new_user = alloc_uid(current_user_ns(), new->uid);
 583         if (!new_user)
 584                 return -EAGAIN;
 585
 586         if (!task_can_switch_user(new_user, current)) {
 587                 free_uid(new_user);
 588                 return -EINVAL;
 589         }
 590
 591         if (atomic_read(&new_user->processes) >=
 592                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 593                         new_user != INIT_USER) {
 594                 free_uid(new_user);
 595                 return -EAGAIN;
 596         }
 597
 598         free_uid(new->user);
 599         new->user = new_user;
 600         return 0;
 601 }
 602
 603 /*
 604  * Unprivileged users may change the real uid to the effective uid
 605  * or vice versa.  (BSD-style)
 606  *
 607  * If you set the real uid at all, or set the effective uid to a value not
 608  * equal to the real uid, then the saved uid is set to the new effective uid.
 609  *
 610  * This makes it possible for a setuid program to completely drop its
 611  * privileges, which is often a useful assertion to make when you are doing
 612  * a security audit over a program.
 613  *
 614  * The general idea is that a program which uses just setreuid() will be
 615  * 100% compatible with BSD.  A program which uses just setuid() will be
 616  * 100% compatible with POSIX with saved IDs.
 617  */
 618 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 619 {
 620         const struct cred *old;
 621         struct cred *new;
 622         int retval;
 623
 624         new = prepare_creds();
 625         if (!new)
 626                 return -ENOMEM;
 627         old = current_cred();
 628
 629         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 630         if (retval)
 631                 goto error;
 632
 633         retval = -EPERM;
 634         if (ruid != (uid_t) -1) {
 635                 new->uid = ruid;
 636                 if (old->uid != ruid &&
 637                     old->euid != ruid &&
 638                     !capable(CAP_SETUID))
 639                         goto error;
 640         }
 641
 642         if (euid != (uid_t) -1) {
 643                 new->euid = euid;
 644                 if (old->uid != euid &&
 645                     old->euid != euid &&
 646                     old->suid != euid &&
 647                     !capable(CAP_SETUID))
 648                         goto error;
 649         }
 650
 651         if (new->uid != old->uid) {
 652                 retval = set_user(new);
 653                 if (retval < 0)
 654                         goto error;
 655         }
 656         if (ruid != (uid_t) -1 ||
 657             (euid != (uid_t) -1 && euid != old->uid))
 658                 new->suid = new->euid;
 659         new->fsuid = new->euid;
 660
 661         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 662         if (retval < 0)
 663                 goto error;
 664
 665         return commit_creds(new);
 666
 667 error:
 668         abort_creds(new);
 669         return retval;
 670 }
 671
 672 /*
 673  * setuid() is implemented like SysV with SAVED_IDS
 674  *
 675  * Note that SAVED_ID's is deficient in that a setuid root program
 676  * like sendmail, for example, cannot set its uid to be a normal
 677  * user and then switch back, because if you're root, setuid() sets
 678  * the saved uid too.  If you don't like this, blame the bright people
 679  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 680  * will allow a root program to temporarily drop privileges and be able to
 681  * regain them by swapping the real and effective uid.
 682  */
 683 SYSCALL_DEFINE1(setuid, uid_t, uid)
 684 {
 685         const struct cred *old;
 686         struct cred *new;
 687         int retval;
 688
 689         new = prepare_creds();
 690         if (!new)
 691                 return -ENOMEM;
 692         old = current_cred();
 693
 694         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 695         if (retval)
 696                 goto error;
 697
 698         retval = -EPERM;
 699         if (capable(CAP_SETUID)) {
 700                 new->suid = new->uid = uid;
 701                 if (uid != old->uid) {
 702                         retval = set_user(new);
 703                         if (retval < 0)
 704                                 goto error;
 705                 }
 706         } else if (uid != old->uid && uid != new->suid) {
 707                 goto error;
 708         }
 709
 710         new->fsuid = new->euid = uid;
 711
 712         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 713         if (retval < 0)
 714                 goto error;
 715
 716         return commit_creds(new);
 717
 718 error:
 719         abort_creds(new);
 720         return retval;
 721 }
 722
 723
 724 /*
 725  * This function implements a generic ability to update ruid, euid,
 726  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 727  */
 728 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 729 {
 730         const struct cred *old;
 731         struct cred *new;
 732         int retval;
 733
 734         new = prepare_creds();
 735         if (!new)
 736                 return -ENOMEM;
 737
 738         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 739         if (retval)
 740                 goto error;
 741         old = current_cred();
 742
 743         retval = -EPERM;
 744         if (!capable(CAP_SETUID)) {
 745                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 746                     ruid != old->euid  && ruid != old->suid)
 747                         goto error;
 748                 if (euid != (uid_t) -1 && euid != old->uid &&
 749                     euid != old->euid  && euid != old->suid)
 750                         goto error;
 751                 if (suid != (uid_t) -1 && suid != old->uid &&
 752                     suid != old->euid  && suid != old->suid)
 753                         goto error;
 754         }
 755
 756         if (ruid != (uid_t) -1) {
 757                 new->uid = ruid;
 758                 if (ruid != old->uid) {
 759                         retval = set_user(new);
 760                         if (retval < 0)
 761                                 goto error;
 762                 }
 763         }
 764         if (euid != (uid_t) -1)
 765                 new->euid = euid;
 766         if (suid != (uid_t) -1)
 767                 new->suid = suid;
 768         new->fsuid = new->euid;
 769
 770         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 771         if (retval < 0)
 772                 goto error;
 773
 774         return commit_creds(new);
 775
 776 error:
 777         abort_creds(new);
 778         return retval;
 779 }
 780
 781 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 782 {
 783         const struct cred *cred = current_cred();
 784         int retval;
 785
 786         if (!(retval   = put_user(cred->uid,  ruid)) &&
 787             !(retval   = put_user(cred->euid, euid)))
 788                 retval = put_user(cred->suid, suid);
 789
 790         return retval;
 791 }
 792
 793 /*
 794  * Same as above, but for rgid, egid, sgid.
 795  */
 796 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 797 {
 798         const struct cred *old;
 799         struct cred *new;
 800         int retval;
 801
 802         new = prepare_creds();
 803         if (!new)
 804                 return -ENOMEM;
 805         old = current_cred();
 806
 807         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 808         if (retval)
 809                 goto error;
 810
 811         retval = -EPERM;
 812         if (!capable(CAP_SETGID)) {
 813                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 814                     rgid != old->egid  && rgid != old->sgid)
 815                         goto error;
 816                 if (egid != (gid_t) -1 && egid != old->gid &&
 817                     egid != old->egid  && egid != old->sgid)
 818                         goto error;
 819                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 820                     sgid != old->egid  && sgid != old->sgid)
 821                         goto error;
 822         }
 823
 824         if (rgid != (gid_t) -1)
 825                 new->gid = rgid;
 826         if (egid != (gid_t) -1)
 827                 new->egid = egid;
 828         if (sgid != (gid_t) -1)
 829                 new->sgid = sgid;
 830         new->fsgid = new->egid;
 831
 832         return commit_creds(new);
 833
 834 error:
 835         abort_creds(new);
 836         return retval;
 837 }
 838
 839 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 840 {
 841         const struct cred *cred = current_cred();
 842         int retval;
 843
 844         if (!(retval   = put_user(cred->gid,  rgid)) &&
 845             !(retval   = put_user(cred->egid, egid)))
 846                 retval = put_user(cred->sgid, sgid);
 847
 848         return retval;
 849 }
 850
 851
 852 /*
 853  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 854  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 855  * whatever uid it wants to). It normally shadows "euid", except when
 856  * explicitly set by setfsuid() or for access..
 857  */
 858 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 859 {
 860         const struct cred *old;
 861         struct cred *new;
 862         uid_t old_fsuid;
 863
 864         new = prepare_creds();
 865         if (!new)
 866                 return current_fsuid();
 867         old = current_cred();
 868         old_fsuid = old->fsuid;
 869
 870         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
 871                 goto error;
 872
 873         if (uid == old->uid  || uid == old->euid  ||
 874             uid == old->suid || uid == old->fsuid ||
 875             capable(CAP_SETUID)) {
 876                 if (uid != old_fsuid) {
 877                         new->fsuid = uid;
 878                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 879                                 goto change_okay;
 880                 }
 881         }
 882
 883 error:
 884         abort_creds(new);
 885         return old_fsuid;
 886
 887 change_okay:
 888         commit_creds(new);
 889         return old_fsuid;
 890 }
 891
 892 /*
 893  * Samma på svenska..
 894  */
 895 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 896 {
 897         const struct cred *old;
 898         struct cred *new;
 899         gid_t old_fsgid;
 900
 901         new = prepare_creds();
 902         if (!new)
 903                 return current_fsgid();
 904         old = current_cred();
 905         old_fsgid = old->fsgid;
 906
 907         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 908                 goto error;
 909
 910         if (gid == old->gid  || gid == old->egid  ||
 911             gid == old->sgid || gid == old->fsgid ||
 912             capable(CAP_SETGID)) {
 913                 if (gid != old_fsgid) {
 914                         new->fsgid = gid;
 915                         goto change_okay;
 916                 }
 917         }
 918
 919 error:
 920         abort_creds(new);
 921         return old_fsgid;
 922
 923 change_okay:
 924         commit_creds(new);
 925         return old_fsgid;
 926 }
 927
 928 void do_sys_times(struct tms *tms)
 929 {
 930         struct task_cputime cputime;
 931         cputime_t cutime, cstime;
 932
 933         thread_group_cputime(current, &cputime);
 934         spin_lock_irq(&current->sighand->siglock);
 935         cutime = current->signal->cutime;
 936         cstime = current->signal->cstime;
 937         spin_unlock_irq(&current->sighand->siglock);
 938         tms->tms_utime = cputime_to_clock_t(cputime.utime);
 939         tms->tms_stime = cputime_to_clock_t(cputime.stime);
 940         tms->tms_cutime = cputime_to_clock_t(cutime);
 941         tms->tms_cstime = cputime_to_clock_t(cstime);
 942 }
 943
 944 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 945 {
 946         if (tbuf) {
 947                 struct tms tmp;
 948
 949                 do_sys_times(&tmp);
 950                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 951                         return -EFAULT;
 952         }
 953         force_successful_syscall_return();
 954         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 955 }
 956
 957 /*
 958  * This needs some heavy checking ...
 959  * I just haven't the stomach for it. I also don't fully
 960  * understand sessions/pgrp etc. Let somebody who does explain it.
 961  *
 962  * OK, I think I have the protection semantics right.... this is really
 963  * only important on a multi-user system anyway, to make sure one user
 964  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 965  *
 966  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 967  * LBT 04.03.94
 968  */
 969 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 970 {
 971         struct task_struct *p;
 972         struct task_struct *group_leader = current->group_leader;
 973         struct pid *pgrp;
 974         int err;
 975
 976         if (!pid)
 977                 pid = task_pid_vnr(group_leader);
 978         if (!pgid)
 979                 pgid = pid;
 980         if (pgid < 0)
 981                 return -EINVAL;
 982
 983         /* From this point forward we keep holding onto the tasklist lock
 984          * so that our parent does not change from under us. -DaveM
 985          */
 986         write_lock_irq(&tasklist_lock);
 987
 988         err = -ESRCH;
 989         p = find_task_by_vpid(pid);
 990         if (!p)
 991                 goto out;
 992
 993         err = -EINVAL;
 994         if (!thread_group_leader(p))
 995                 goto out;
 996
 997         if (same_thread_group(p->real_parent, group_leader)) {
 998                 err = -EPERM;
 999                 if (task_session(p) != task_session(group_leader))
1000                         goto out;
1001                 err = -EACCES;
1002                 if (p->did_exec)
1003                         goto out;
1004         } else {
1005                 err = -ESRCH;
1006                 if (p != group_leader)
1007                         goto out;
1008         }
1009
1010         err = -EPERM;
1011         if (p->signal->leader)
1012                 goto out;
1013
1014         pgrp = task_pid(p);
1015         if (pgid != pid) {
1016                 struct task_struct *g;
1017
1018                 pgrp = find_vpid(pgid);
1019                 g = pid_task(pgrp, PIDTYPE_PGID);
1020                 if (!g || task_session(g) != task_session(group_leader))
1021                         goto out;
1022         }
1023
1024         err = security_task_setpgid(p, pgid);
1025         if (err)
1026                 goto out;
1027
1028         if (task_pgrp(p) != pgrp) {
1029                 change_pid(p, PIDTYPE_PGID, pgrp);
1030                 set_task_pgrp(p, pid_nr(pgrp));
1031         }
1032
1033         err = 0;
1034 out:
1035         /* All paths lead to here, thus we are safe. -DaveM */
1036         write_unlock_irq(&tasklist_lock);
1037         return err;
1038 }
1039
1040 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1041 {
1042         struct task_struct *p;
1043         struct pid *grp;
1044         int retval;
1045
1046         rcu_read_lock();
1047         if (!pid)
1048                 grp = task_pgrp(current);
1049         else {
1050                 retval = -ESRCH;
1051                 p = find_task_by_vpid(pid);
1052                 if (!p)
1053                         goto out;
1054                 grp = task_pgrp(p);
1055                 if (!grp)
1056                         goto out;
1057
1058                 retval = security_task_getpgid(p);
1059                 if (retval)
1060                         goto out;
1061         }
1062         retval = pid_vnr(grp);
1063 out:
1064         rcu_read_unlock();
1065         return retval;
1066 }
1067
1068 #ifdef __ARCH_WANT_SYS_GETPGRP
1069
1070 SYSCALL_DEFINE0(getpgrp)
1071 {
1072         return sys_getpgid(0);
1073 }
1074
1075 #endif
1076
1077 SYSCALL_DEFINE1(getsid, pid_t, pid)
1078 {
1079         struct task_struct *p;
1080         struct pid *sid;
1081         int retval;
1082
1083         rcu_read_lock();
1084         if (!pid)
1085                 sid = task_session(current);
1086         else {
1087                 retval = -ESRCH;
1088                 p = find_task_by_vpid(pid);
1089                 if (!p)
1090                         goto out;
1091                 sid = task_session(p);
1092                 if (!sid)
1093                         goto out;
1094
1095                 retval = security_task_getsid(p);
1096                 if (retval)
1097                         goto out;
1098         }
1099         retval = pid_vnr(sid);
1100 out:
1101         rcu_read_unlock();
1102         return retval;
1103 }
1104
1105 SYSCALL_DEFINE0(setsid)
1106 {
1107         struct task_struct *group_leader = current->group_leader;
1108         struct pid *sid = task_pid(group_leader);
1109         pid_t session = pid_vnr(sid);
1110         int err = -EPERM;
1111
1112         write_lock_irq(&tasklist_lock);
1113         /* Fail if I am already a session leader */
1114         if (group_leader->signal->leader)
1115                 goto out;
1116
1117         /* Fail if a process group id already exists that equals the
1118          * proposed session id.
1119          */
1120         if (pid_task(sid, PIDTYPE_PGID))
1121                 goto out;
1122
1123         group_leader->signal->leader = 1;
1124         __set_special_pids(sid);
1125
1126         proc_clear_tty(group_leader);
1127
1128         err = session;
1129 out:
1130         write_unlock_irq(&tasklist_lock);
1131         return err;
1132 }
1133
1134 /*
1135  * Supplementary group IDs
1136  */
1137
1138 /* init to 2 - one for init_task, one to ensure it is never freed */
1139 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1140
1141 struct group_info *groups_alloc(int gidsetsize)
1142 {
1143         struct group_info *group_info;
1144         int nblocks;
1145         int i;
1146
1147         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1148         /* Make sure we always allocate at least one indirect block pointer */
1149         nblocks = nblocks ? : 1;
1150         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1151         if (!group_info)
1152                 return NULL;
1153         group_info->ngroups = gidsetsize;
1154         group_info->nblocks = nblocks;
1155         atomic_set(&group_info->usage, 1);
1156
1157         if (gidsetsize <= NGROUPS_SMALL)
1158                 group_info->blocks[0] = group_info->small_block;
1159         else {
1160                 for (i = 0; i < nblocks; i++) {
1161                         gid_t *b;
1162                         b = (void *)__get_free_page(GFP_USER);
1163                         if (!b)
1164                                 goto out_undo_partial_alloc;
1165                         group_info->blocks[i] = b;
1166                 }
1167         }
1168         return group_info;
1169
1170 out_undo_partial_alloc:
1171         while (--i >= 0) {
1172                 free_page((unsigned long)group_info->blocks[i]);
1173         }
1174         kfree(group_info);
1175         return NULL;
1176 }
1177
1178 EXPORT_SYMBOL(groups_alloc);
1179
1180 void groups_free(struct group_info *group_info)
1181 {
1182         if (group_info->blocks[0] != group_info->small_block) {
1183                 int i;
1184                 for (i = 0; i < group_info->nblocks; i++)
1185                         free_page((unsigned long)group_info->blocks[i]);
1186         }
1187         kfree(group_info);
1188 }
1189
1190 EXPORT_SYMBOL(groups_free);
1191
1192 /* export the group_info to a user-space array */
1193 static int groups_to_user(gid_t __user *grouplist,
1194                           const struct group_info *group_info)
1195 {
1196         int i;
1197         unsigned int count = group_info->ngroups;
1198
1199         for (i = 0; i < group_info->nblocks; i++) {
1200                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1201                 unsigned int len = cp_count * sizeof(*grouplist);
1202
1203                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1204                         return -EFAULT;
1205
1206                 grouplist += NGROUPS_PER_BLOCK;
1207                 count -= cp_count;
1208         }
1209         return 0;
1210 }
1211
1212 /* fill a group_info from a user-space array - it must be allocated already */
1213 static int groups_from_user(struct group_info *group_info,
1214     gid_t __user *grouplist)
1215 {
1216         int i;
1217         unsigned int count = group_info->ngroups;
1218
1219         for (i = 0; i < group_info->nblocks; i++) {
1220                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1221                 unsigned int len = cp_count * sizeof(*grouplist);
1222
1223                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1224                         return -EFAULT;
1225
1226                 grouplist += NGROUPS_PER_BLOCK;
1227                 count -= cp_count;
1228         }
1229         return 0;
1230 }
1231
1232 /* a simple Shell sort */
1233 static void groups_sort(struct group_info *group_info)
1234 {
1235         int base, max, stride;
1236         int gidsetsize = group_info->ngroups;
1237
1238         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1239                 ; /* nothing */
1240         stride /= 3;
1241
1242         while (stride) {
1243                 max = gidsetsize - stride;
1244                 for (base = 0; base < max; base++) {
1245                         int left = base;
1246                         int right = left + stride;
1247                         gid_t tmp = GROUP_AT(group_info, right);
1248
1249                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1250                                 GROUP_AT(group_info, right) =
1251                                     GROUP_AT(group_info, left);
1252                                 right = left;
1253                                 left -= stride;
1254                         }
1255                         GROUP_AT(group_info, right) = tmp;
1256                 }
1257                 stride /= 3;
1258         }
1259 }
1260
1261 /* a simple bsearch */
1262 int groups_search(const struct group_info *group_info, gid_t grp)
1263 {
1264         unsigned int left, right;
1265
1266         if (!group_info)
1267                 return 0;
1268
1269         left = 0;
1270         right = group_info->ngroups;
1271         while (left < right) {
1272                 unsigned int mid = (left+right)/2;
1273                 int cmp = grp - GROUP_AT(group_info, mid);
1274                 if (cmp > 0)
1275                         left = mid + 1;
1276                 else if (cmp < 0)
1277                         right = mid;
1278                 else
1279                         return 1;
1280         }
1281         return 0;
1282 }
1283
1284 /**
1285  * set_groups - Change a group subscription in a set of credentials
1286  * @new: The newly prepared set of credentials to alter
1287  * @group_info: The group list to install
1288  *
1289  * Validate a group subscription and, if valid, insert it into a set
1290  * of credentials.
1291  */
1292 int set_groups(struct cred *new, struct group_info *group_info)
1293 {
1294         int retval;
1295
1296         retval = security_task_setgroups(group_info);
1297         if (retval)
1298                 return retval;
1299
1300         put_group_info(new->group_info);
1301         groups_sort(group_info);
1302         get_group_info(group_info);
1303         new->group_info = group_info;
1304         return 0;
1305 }
1306
1307 EXPORT_SYMBOL(set_groups);
1308
1309 /**
1310  * set_current_groups - Change current's group subscription
1311  * @group_info: The group list to impose
1312  *
1313  * Validate a group subscription and, if valid, impose it upon current's task
1314  * security record.
1315  */
1316 int set_current_groups(struct group_info *group_info)
1317 {
1318         struct cred *new;
1319         int ret;
1320
1321         new = prepare_creds();
1322         if (!new)
1323                 return -ENOMEM;
1324
1325         ret = set_groups(new, group_info);
1326         if (ret < 0) {
1327                 abort_creds(new);
1328                 return ret;
1329         }
1330
1331         return commit_creds(new);
1332 }
1333
1334 EXPORT_SYMBOL(set_current_groups);
1335
1336 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1337 {
1338         const struct cred *cred = current_cred();
1339         int i;
1340
1341         if (gidsetsize < 0)
1342                 return -EINVAL;
1343
1344         /* no need to grab task_lock here; it cannot change */
1345         i = cred->group_info->ngroups;
1346         if (gidsetsize) {
1347                 if (i > gidsetsize) {
1348                         i = -EINVAL;
1349                         goto out;
1350                 }
1351                 if (groups_to_user(grouplist, cred->group_info)) {
1352                         i = -EFAULT;
1353                         goto out;
1354                 }
1355         }
1356 out:
1357         return i;
1358 }
1359
1360 /*
1361  *      SMP: Our groups are copy-on-write. We can set them safely
1362  *      without another task interfering.
1363  */
1364
1365 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1366 {
1367         struct group_info *group_info;
1368         int retval;
1369
1370         if (!capable(CAP_SETGID))
1371                 return -EPERM;
1372         if ((unsigned)gidsetsize > NGROUPS_MAX)
1373                 return -EINVAL;
1374
1375         group_info = groups_alloc(gidsetsize);
1376         if (!group_info)
1377                 return -ENOMEM;
1378         retval = groups_from_user(group_info, grouplist);
1379         if (retval) {
1380                 put_group_info(group_info);
1381                 return retval;
1382         }
1383
1384         retval = set_current_groups(group_info);
1385         put_group_info(group_info);
1386
1387         return retval;
1388 }
1389
1390 /*
1391  * Check whether we're fsgid/egid or in the supplemental group..
1392  */
1393 int in_group_p(gid_t grp)
1394 {
1395         const struct cred *cred = current_cred();
1396         int retval = 1;
1397
1398         if (grp != cred->fsgid)
1399                 retval = groups_search(cred->group_info, grp);
1400         return retval;
1401 }
1402
1403 EXPORT_SYMBOL(in_group_p);
1404
1405 int in_egroup_p(gid_t grp)
1406 {
1407         const struct cred *cred = current_cred();
1408         int retval = 1;
1409
1410         if (grp != cred->egid)
1411                 retval = groups_search(cred->group_info, grp);
1412         return retval;
1413 }
1414
1415 EXPORT_SYMBOL(in_egroup_p);
1416
1417 DECLARE_RWSEM(uts_sem);
1418
1419 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1420 {
1421         int errno = 0;
1422
1423         down_read(&uts_sem);
1424         if (copy_to_user(name, utsname(), sizeof *name))
1425                 errno = -EFAULT;
1426         up_read(&uts_sem);
1427         return errno;
1428 }
1429
1430 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1431 {
1432         int errno;
1433         char tmp[__NEW_UTS_LEN];
1434
1435         if (!capable(CAP_SYS_ADMIN))
1436                 return -EPERM;
1437         if (len < 0 || len > __NEW_UTS_LEN)
1438                 return -EINVAL;
1439         down_write(&uts_sem);
1440         errno = -EFAULT;
1441         if (!copy_from_user(tmp, name, len)) {
1442                 struct new_utsname *u = utsname();
1443
1444                 memcpy(u->nodename, tmp, len);
1445                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1446                 errno = 0;
1447         }
1448         up_write(&uts_sem);
1449         return errno;
1450 }
1451
1452 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1453
1454 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1455 {
1456         int i, errno;
1457         struct new_utsname *u;
1458
1459         if (len < 0)
1460                 return -EINVAL;
1461         down_read(&uts_sem);
1462         u = utsname();
1463         i = 1 + strlen(u->nodename);
1464         if (i > len)
1465                 i = len;
1466         errno = 0;
1467         if (copy_to_user(name, u->nodename, i))
1468                 errno = -EFAULT;
1469         up_read(&uts_sem);
1470         return errno;
1471 }
1472
1473 #endif
1474
1475 /*
1476  * Only setdomainname; getdomainname can be implemented by calling
1477  * uname()
1478  */
1479 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1480 {
1481         int errno;
1482         char tmp[__NEW_UTS_LEN];
1483
1484         if (!capable(CAP_SYS_ADMIN))
1485                 return -EPERM;
1486         if (len < 0 || len > __NEW_UTS_LEN)
1487                 return -EINVAL;
1488
1489         down_write(&uts_sem);
1490         errno = -EFAULT;
1491         if (!copy_from_user(tmp, name, len)) {
1492                 struct new_utsname *u = utsname();
1493
1494                 memcpy(u->domainname, tmp, len);
1495                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1496                 errno = 0;
1497         }
1498         up_write(&uts_sem);
1499         return errno;
1500 }
1501
1502 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1503 {
1504         if (resource >= RLIM_NLIMITS)
1505                 return -EINVAL;
1506         else {
1507                 struct rlimit value;
1508                 task_lock(current->group_leader);
1509                 value = current->signal->rlim[resource];
1510                 task_unlock(current->group_leader);
1511                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1512         }
1513 }
1514
1515 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1516
1517 /*
1518  *      Back compatibility for getrlimit. Needed for some apps.
1519  */
1520
1521 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1522                 struct rlimit __user *, rlim)
1523 {
1524         struct rlimit x;
1525         if (resource >= RLIM_NLIMITS)
1526                 return -EINVAL;
1527
1528         task_lock(current->group_leader);
1529         x = current->signal->rlim[resource];
1530         task_unlock(current->group_leader);
1531         if (x.rlim_cur > 0x7FFFFFFF)
1532                 x.rlim_cur = 0x7FFFFFFF;
1533         if (x.rlim_max > 0x7FFFFFFF)
1534                 x.rlim_max = 0x7FFFFFFF;
1535         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1536 }
1537
1538 #endif
1539
1540 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1541 {
1542         struct rlimit new_rlim, *old_rlim;
1543         int retval;
1544
1545         if (resource >= RLIM_NLIMITS)
1546                 return -EINVAL;
1547         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1548                 return -EFAULT;
1549         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1550                 return -EINVAL;
1551         old_rlim = current->signal->rlim + resource;
1552         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1553             !capable(CAP_SYS_RESOURCE))
1554                 return -EPERM;
1555         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1556                 return -EPERM;
1557
1558         retval = security_task_setrlimit(resource, &new_rlim);
1559         if (retval)
1560                 return retval;
1561
1562         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1563                 /*
1564                  * The caller is asking for an immediate RLIMIT_CPU
1565                  * expiry.  But we use the zero value to mean "it was
1566                  * never set".  So let's cheat and make it one second
1567                  * instead
1568                  */
1569                 new_rlim.rlim_cur = 1;
1570         }
1571
1572         task_lock(current->group_leader);
1573         *old_rlim = new_rlim;
1574         task_unlock(current->group_leader);
1575
1576         if (resource != RLIMIT_CPU)
1577                 goto out;
1578
1579         /*
1580          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1581          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1582          * very long-standing error, and fixing it now risks breakage of
1583          * applications, so we live with it
1584          */
1585         if (new_rlim.rlim_cur == RLIM_INFINITY)
1586                 goto out;
1587
1588         update_rlimit_cpu(new_rlim.rlim_cur);
1589 out:
1590         return 0;
1591 }
1592
1593 /*
1594  * It would make sense to put struct rusage in the task_struct,
1595  * except that would make the task_struct be *really big*.  After
1596  * task_struct gets moved into malloc'ed memory, it would
1597  * make sense to do this.  It will make moving the rest of the information
1598  * a lot simpler!  (Which we're not doing right now because we're not
1599  * measuring them yet).
1600  *
1601  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1602  * races with threads incrementing their own counters.  But since word
1603  * reads are atomic, we either get new values or old values and we don't
1604  * care which for the sums.  We always take the siglock to protect reading
1605  * the c* fields from p->signal from races with exit.c updating those
1606  * fields when reaping, so a sample either gets all the additions of a
1607  * given child after it's reaped, or none so this sample is before reaping.
1608  *
1609  * Locking:
1610  * We need to take the siglock for CHILDEREN, SELF and BOTH
1611  * for  the cases current multithreaded, non-current single threaded
1612  * non-current multithreaded.  Thread traversal is now safe with
1613  * the siglock held.
1614  * Strictly speaking, we donot need to take the siglock if we are current and
1615  * single threaded,  as no one else can take our signal_struct away, no one
1616  * else can  reap the  children to update signal->c* counters, and no one else
1617  * can race with the signal-> fields. If we do not take any lock, the
1618  * signal-> fields could be read out of order while another thread was just
1619  * exiting. So we should  place a read memory barrier when we avoid the lock.
1620  * On the writer side,  write memory barrier is implied in  __exit_signal
1621  * as __exit_signal releases  the siglock spinlock after updating the signal->
1622  * fields. But we don't do this yet to keep things simple.
1623  *
1624  */
1625
1626 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1627 {
1628         r->ru_nvcsw += t->nvcsw;
1629         r->ru_nivcsw += t->nivcsw;
1630         r->ru_minflt += t->min_flt;
1631         r->ru_majflt += t->maj_flt;
1632         r->ru_inblock += task_io_get_inblock(t);
1633         r->ru_oublock += task_io_get_oublock(t);
1634 }
1635
1636 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1637 {
1638         struct task_struct *t;
1639         unsigned long flags;
1640         cputime_t utime, stime;
1641         struct task_cputime cputime;
1642
1643         memset((char *) r, 0, sizeof *r);
1644         utime = stime = cputime_zero;
1645
1646         if (who == RUSAGE_THREAD) {
1647                 utime = task_utime(current);
1648                 stime = task_stime(current);
1649                 accumulate_thread_rusage(p, r);
1650                 goto out;
1651         }
1652
1653         if (!lock_task_sighand(p, &flags))
1654                 return;
1655
1656         switch (who) {
1657                 case RUSAGE_BOTH:
1658                 case RUSAGE_CHILDREN:
1659                         utime = p->signal->cutime;
1660                         stime = p->signal->cstime;
1661                         r->ru_nvcsw = p->signal->cnvcsw;
1662                         r->ru_nivcsw = p->signal->cnivcsw;
1663                         r->ru_minflt = p->signal->cmin_flt;
1664                         r->ru_majflt = p->signal->cmaj_flt;
1665                         r->ru_inblock = p->signal->cinblock;
1666                         r->ru_oublock = p->signal->coublock;
1667
1668                         if (who == RUSAGE_CHILDREN)
1669                                 break;
1670
1671                 case RUSAGE_SELF:
1672                         thread_group_cputime(p, &cputime);
1673                         utime = cputime_add(utime, cputime.utime);
1674                         stime = cputime_add(stime, cputime.stime);
1675                         r->ru_nvcsw += p->signal->nvcsw;
1676                         r->ru_nivcsw += p->signal->nivcsw;
1677                         r->ru_minflt += p->signal->min_flt;
1678                         r->ru_majflt += p->signal->maj_flt;
1679                         r->ru_inblock += p->signal->inblock;
1680                         r->ru_oublock += p->signal->oublock;
1681                         t = p;
1682                         do {
1683                                 accumulate_thread_rusage(t, r);
1684                                 t = next_thread(t);
1685                         } while (t != p);
1686                         break;
1687
1688                 default:
1689                         BUG();
1690         }
1691         unlock_task_sighand(p, &flags);
1692
1693 out:
1694         cputime_to_timeval(utime, &r->ru_utime);
1695         cputime_to_timeval(stime, &r->ru_stime);
1696 }
1697
1698 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1699 {
1700         struct rusage r;
1701         k_getrusage(p, who, &r);
1702         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1703 }
1704
1705 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1706 {
1707         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1708             who != RUSAGE_THREAD)
1709                 return -EINVAL;
1710         return getrusage(current, who, ru);
1711 }
1712
1713 SYSCALL_DEFINE1(umask, int, mask)
1714 {
1715         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1716         return mask;
1717 }
1718
1719 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1720                 unsigned long, arg4, unsigned long, arg5)
1721 {
1722         struct task_struct *me = current;
1723         unsigned char comm[sizeof(me->comm)];
1724         long error;
1725
1726         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1727         if (error != -ENOSYS)
1728                 return error;
1729
1730         error = 0;
1731         switch (option) {
1732                 case PR_SET_PDEATHSIG:
1733                         if (!valid_signal(arg2)) {
1734                                 error = -EINVAL;
1735                                 break;
1736                         }
1737                         me->pdeath_signal = arg2;
1738                         error = 0;
1739                         break;
1740                 case PR_GET_PDEATHSIG:
1741                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1742                         break;
1743                 case PR_GET_DUMPABLE:
1744                         error = get_dumpable(me->mm);
1745                         break;
1746                 case PR_SET_DUMPABLE:
1747                         if (arg2 < 0 || arg2 > 1) {
1748                                 error = -EINVAL;
1749                                 break;
1750                         }
1751                         set_dumpable(me->mm, arg2);
1752                         error = 0;
1753                         break;
1754
1755                 case PR_SET_UNALIGN:
1756                         error = SET_UNALIGN_CTL(me, arg2);
1757                         break;
1758                 case PR_GET_UNALIGN:
1759                         error = GET_UNALIGN_CTL(me, arg2);
1760                         break;
1761                 case PR_SET_FPEMU:
1762                         error = SET_FPEMU_CTL(me, arg2);
1763                         break;
1764                 case PR_GET_FPEMU:
1765                         error = GET_FPEMU_CTL(me, arg2);
1766                         break;
1767                 case PR_SET_FPEXC:
1768                         error = SET_FPEXC_CTL(me, arg2);
1769                         break;
1770                 case PR_GET_FPEXC:
1771                         error = GET_FPEXC_CTL(me, arg2);
1772                         break;
1773                 case PR_GET_TIMING:
1774                         error = PR_TIMING_STATISTICAL;
1775                         break;
1776                 case PR_SET_TIMING:
1777                         if (arg2 != PR_TIMING_STATISTICAL)
1778                                 error = -EINVAL;
1779                         else
1780                                 error = 0;
1781                         break;
1782
1783                 case PR_SET_NAME:
1784                         comm[sizeof(me->comm)-1] = 0;
1785                         if (strncpy_from_user(comm, (char __user *)arg2,
1786                                               sizeof(me->comm) - 1) < 0)
1787                                 return -EFAULT;
1788                         set_task_comm(me, comm);
1789                         return 0;
1790                 case PR_GET_NAME:
1791                         get_task_comm(comm, me);
1792                         if (copy_to_user((char __user *)arg2, comm,
1793                                          sizeof(comm)))
1794                                 return -EFAULT;
1795                         return 0;
1796                 case PR_GET_ENDIAN:
1797                         error = GET_ENDIAN(me, arg2);
1798                         break;
1799                 case PR_SET_ENDIAN:
1800                         error = SET_ENDIAN(me, arg2);
1801                         break;
1802
1803                 case PR_GET_SECCOMP:
1804                         error = prctl_get_seccomp();
1805                         break;
1806                 case PR_SET_SECCOMP:
1807                         error = prctl_set_seccomp(arg2);
1808                         break;
1809                 case PR_GET_TSC:
1810                         error = GET_TSC_CTL(arg2);
1811                         break;
1812                 case PR_SET_TSC:
1813                         error = SET_TSC_CTL(arg2);
1814                         break;
1815                 case PR_TASK_PERF_COUNTERS_DISABLE:
1816                         error = perf_counter_task_disable();
1817                         break;
1818                 case PR_TASK_PERF_COUNTERS_ENABLE:
1819                         error = perf_counter_task_enable();
1820                         break;
1821                 case PR_GET_TIMERSLACK:
1822                         error = current->timer_slack_ns;
1823                         break;
1824                 case PR_SET_TIMERSLACK:
1825                         if (arg2 <= 0)
1826                                 current->timer_slack_ns =
1827                                         current->default_timer_slack_ns;
1828                         else
1829                                 current->timer_slack_ns = arg2;
1830                         error = 0;
1831                         break;
1832                 default:
1833                         error = -EINVAL;
1834                         break;
1835         }
1836         return error;
1837 }
1838
1839 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1840                 struct getcpu_cache __user *, unused)
1841 {
1842         int err = 0;
1843         int cpu = raw_smp_processor_id();
1844         if (cpup)
1845                 err |= put_user(cpu, cpup);
1846         if (nodep)
1847                 err |= put_user(cpu_to_node(cpu), nodep);
1848         return err ? -EFAULT : 0;
1849 }
1850
1851 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1852
1853 static void argv_cleanup(char **argv, char **envp)
1854 {
1855         argv_free(argv);
1856 }
1857
1858 /**
1859  * orderly_poweroff - Trigger an orderly system poweroff
1860  * @force: force poweroff if command execution fails
1861  *
1862  * This may be called from any context to trigger a system shutdown.
1863  * If the orderly shutdown fails, it will force an immediate shutdown.
1864  */
1865 int orderly_poweroff(bool force)
1866 {
1867         int argc;
1868         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1869         static char *envp[] = {
1870                 "HOME=/",
1871                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1872                 NULL
1873         };
1874         int ret = -ENOMEM;
1875         struct subprocess_info *info;
1876
1877         if (argv == NULL) {
1878                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1879                        __func__, poweroff_cmd);
1880                 goto out;
1881         }
1882
1883         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1884         if (info == NULL) {
1885                 argv_free(argv);
1886                 goto out;
1887         }
1888
1889         call_usermodehelper_setcleanup(info, argv_cleanup);
1890
1891         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1892
1893   out:
1894         if (ret && force) {
1895                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1896                        "forcing the issue\n");
1897
1898                 /* I guess this should try to kick off some daemon to
1899                    sync and poweroff asap.  Or not even bother syncing
1900                    if we're doing an emergency shutdown? */
1901                 emergency_sync();
1902                 kernel_power_off();
1903         }
1904
1905         return ret;
1906 }
1907 EXPORT_SYMBOL_GPL(orderly_poweroff);