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/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/ptrace.h>
  37 #include <linux/fs_struct.h>
  38
  39 #include <linux/compat.h>
  40 #include <linux/syscalls.h>
  41 #include <linux/kprobes.h>
  42 #include <linux/user_namespace.h>
  43
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46 #include <asm/unistd.h>
  47
  48 #ifndef SET_UNALIGN_CTL
  49 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  50 #endif
  51 #ifndef GET_UNALIGN_CTL
  52 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  53 #endif
  54 #ifndef SET_FPEMU_CTL
  55 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  56 #endif
  57 #ifndef GET_FPEMU_CTL
  58 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  59 #endif
  60 #ifndef SET_FPEXC_CTL
  61 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  62 #endif
  63 #ifndef GET_FPEXC_CTL
  64 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  65 #endif
  66 #ifndef GET_ENDIAN
  67 # define GET_ENDIAN(a,b)        (-EINVAL)
  68 #endif
  69 #ifndef SET_ENDIAN
  70 # define SET_ENDIAN(a,b)        (-EINVAL)
  71 #endif
  72 #ifndef GET_TSC_CTL
  73 # define GET_TSC_CTL(a)         (-EINVAL)
  74 #endif
  75 #ifndef SET_TSC_CTL
  76 # define SET_TSC_CTL(a)         (-EINVAL)
  77 #endif
  78
  79 /*
  80  * this is where the system-wide overflow UID and GID are defined, for
  81  * architectures that now have 32-bit UID/GID but didn't in the past
  82  */
  83
  84 int overflowuid = DEFAULT_OVERFLOWUID;
  85 int overflowgid = DEFAULT_OVERFLOWGID;
  86
  87 #ifdef CONFIG_UID16
  88 EXPORT_SYMBOL(overflowuid);
  89 EXPORT_SYMBOL(overflowgid);
  90 #endif
  91
  92 /*
  93  * the same as above, but for filesystems which can only store a 16-bit
  94  * UID and GID. as such, this is needed on all architectures
  95  */
  96
  97 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  98 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  99
 100 EXPORT_SYMBOL(fs_overflowuid);
 101 EXPORT_SYMBOL(fs_overflowgid);
 102
 103 /*
 104  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 105  */
 106
 107 int C_A_D = 1;
 108 struct pid *cad_pid;
 109 EXPORT_SYMBOL(cad_pid);
 110
 111 /*
 112  * If set, this is used for preparing the system to power off.
 113  */
 114
 115 void (*pm_power_off_prepare)(void);
 116
 117 /*
 118  * set the priority of a task
 119  * - the caller must hold the RCU read lock
 120  */
 121 static int set_one_prio(struct task_struct *p, int niceval, int error)
 122 {
 123         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 124         int no_nice;
 125
 126         if (pcred->uid  != cred->euid &&
 127             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
 128                 error = -EPERM;
 129                 goto out;
 130         }
 131         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 132                 error = -EACCES;
 133                 goto out;
 134         }
 135         no_nice = security_task_setnice(p, niceval);
 136         if (no_nice) {
 137                 error = no_nice;
 138                 goto out;
 139         }
 140         if (error == -ESRCH)
 141                 error = 0;
 142         set_user_nice(p, niceval);
 143 out:
 144         return error;
 145 }
 146
 147 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 148 {
 149         struct task_struct *g, *p;
 150         struct user_struct *user;
 151         const struct cred *cred = current_cred();
 152         int error = -EINVAL;
 153         struct pid *pgrp;
 154
 155         if (which > PRIO_USER || which < PRIO_PROCESS)
 156                 goto out;
 157
 158         /* normalize: avoid signed division (rounding problems) */
 159         error = -ESRCH;
 160         if (niceval < -20)
 161                 niceval = -20;
 162         if (niceval > 19)
 163                 niceval = 19;
 164
 165         read_lock(&tasklist_lock);
 166         switch (which) {
 167                 case PRIO_PROCESS:
 168                         if (who)
 169                                 p = find_task_by_vpid(who);
 170                         else
 171                                 p = current;
 172                         if (p)
 173                                 error = set_one_prio(p, niceval, error);
 174                         break;
 175                 case PRIO_PGRP:
 176                         if (who)
 177                                 pgrp = find_vpid(who);
 178                         else
 179                                 pgrp = task_pgrp(current);
 180                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 181                                 error = set_one_prio(p, niceval, error);
 182                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 183                         break;
 184                 case PRIO_USER:
 185                         user = (struct user_struct *) cred->user;
 186                         if (!who)
 187                                 who = cred->uid;
 188                         else if ((who != cred->uid) &&
 189                                  !(user = find_user(who)))
 190                                 goto out_unlock;        /* No processes for this user */
 191
 192                         do_each_thread(g, p)
 193                                 if (__task_cred(p)->uid == who)
 194                                         error = set_one_prio(p, niceval, error);
 195                         while_each_thread(g, p);
 196                         if (who != cred->uid)
 197                                 free_uid(user);         /* For find_user() */
 198                         break;
 199         }
 200 out_unlock:
 201         read_unlock(&tasklist_lock);
 202 out:
 203         return error;
 204 }
 205
 206 /*
 207  * Ugh. To avoid negative return values, "getpriority()" will
 208  * not return the normal nice-value, but a negated value that
 209  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 210  * to stay compatible.
 211  */
 212 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 213 {
 214         struct task_struct *g, *p;
 215         struct user_struct *user;
 216         const struct cred *cred = current_cred();
 217         long niceval, retval = -ESRCH;
 218         struct pid *pgrp;
 219
 220         if (which > PRIO_USER || which < PRIO_PROCESS)
 221                 return -EINVAL;
 222
 223         read_lock(&tasklist_lock);
 224         switch (which) {
 225                 case PRIO_PROCESS:
 226                         if (who)
 227                                 p = find_task_by_vpid(who);
 228                         else
 229                                 p = current;
 230                         if (p) {
 231                                 niceval = 20 - task_nice(p);
 232                                 if (niceval > retval)
 233                                         retval = niceval;
 234                         }
 235                         break;
 236                 case PRIO_PGRP:
 237                         if (who)
 238                                 pgrp = find_vpid(who);
 239                         else
 240                                 pgrp = task_pgrp(current);
 241                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 242                                 niceval = 20 - task_nice(p);
 243                                 if (niceval > retval)
 244                                         retval = niceval;
 245                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 246                         break;
 247                 case PRIO_USER:
 248                         user = (struct user_struct *) cred->user;
 249                         if (!who)
 250                                 who = cred->uid;
 251                         else if ((who != cred->uid) &&
 252                                  !(user = find_user(who)))
 253                                 goto out_unlock;        /* No processes for this user */
 254
 255                         do_each_thread(g, p)
 256                                 if (__task_cred(p)->uid == who) {
 257                                         niceval = 20 - task_nice(p);
 258                                         if (niceval > retval)
 259                                                 retval = niceval;
 260                                 }
 261                         while_each_thread(g, p);
 262                         if (who != cred->uid)
 263                                 free_uid(user);         /* for find_user() */
 264                         break;
 265         }
 266 out_unlock:
 267         read_unlock(&tasklist_lock);
 268
 269         return retval;
 270 }
 271
 272 /**
 273  *      emergency_restart - reboot the system
 274  *
 275  *      Without shutting down any hardware or taking any locks
 276  *      reboot the system.  This is called when we know we are in
 277  *      trouble so this is our best effort to reboot.  This is
 278  *      safe to call in interrupt context.
 279  */
 280 void emergency_restart(void)
 281 {
 282         machine_emergency_restart();
 283 }
 284 EXPORT_SYMBOL_GPL(emergency_restart);
 285
 286 void kernel_restart_prepare(char *cmd)
 287 {
 288         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 289         system_state = SYSTEM_RESTART;
 290         device_shutdown();
 291         sysdev_shutdown();
 292 }
 293
 294 /**
 295  *      kernel_restart - reboot the system
 296  *      @cmd: pointer to buffer containing command to execute for restart
 297  *              or %NULL
 298  *
 299  *      Shutdown everything and perform a clean reboot.
 300  *      This is not safe to call in interrupt context.
 301  */
 302 void kernel_restart(char *cmd)
 303 {
 304         kernel_restart_prepare(cmd);
 305         if (!cmd)
 306                 printk(KERN_EMERG "Restarting system.\n");
 307         else
 308                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 309         machine_restart(cmd);
 310 }
 311 EXPORT_SYMBOL_GPL(kernel_restart);
 312
 313 static void kernel_shutdown_prepare(enum system_states state)
 314 {
 315         blocking_notifier_call_chain(&reboot_notifier_list,
 316                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 317         system_state = state;
 318         device_shutdown();
 319 }
 320 /**
 321  *      kernel_halt - halt the system
 322  *
 323  *      Shutdown everything and perform a clean system halt.
 324  */
 325 void kernel_halt(void)
 326 {
 327         kernel_shutdown_prepare(SYSTEM_HALT);
 328         sysdev_shutdown();
 329         printk(KERN_EMERG "System halted.\n");
 330         machine_halt();
 331 }
 332
 333 EXPORT_SYMBOL_GPL(kernel_halt);
 334
 335 /**
 336  *      kernel_power_off - power_off the system
 337  *
 338  *      Shutdown everything and perform a clean system power_off.
 339  */
 340 void kernel_power_off(void)
 341 {
 342         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 343         if (pm_power_off_prepare)
 344                 pm_power_off_prepare();
 345         disable_nonboot_cpus();
 346         sysdev_shutdown();
 347         printk(KERN_EMERG "Power down.\n");
 348         machine_power_off();
 349 }
 350 EXPORT_SYMBOL_GPL(kernel_power_off);
 351 /*
 352  * Reboot system call: for obvious reasons only root may call it,
 353  * and even root needs to set up some magic numbers in the registers
 354  * so that some mistake won't make this reboot the whole machine.
 355  * You can also set the meaning of the ctrl-alt-del-key here.
 356  *
 357  * reboot doesn't sync: do that yourself before calling this.
 358  */
 359 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 360                 void __user *, arg)
 361 {
 362         char buffer[256];
 363         int ret = 0;
 364
 365         /* We only trust the superuser with rebooting the system. */
 366         if (!capable(CAP_SYS_BOOT))
 367                 return -EPERM;
 368
 369         /* For safety, we require "magic" arguments. */
 370         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 371             (magic2 != LINUX_REBOOT_MAGIC2 &&
 372                         magic2 != LINUX_REBOOT_MAGIC2A &&
 373                         magic2 != LINUX_REBOOT_MAGIC2B &&
 374                         magic2 != LINUX_REBOOT_MAGIC2C))
 375                 return -EINVAL;
 376
 377         /* Instead of trying to make the power_off code look like
 378          * halt when pm_power_off is not set do it the easy way.
 379          */
 380         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 381                 cmd = LINUX_REBOOT_CMD_HALT;
 382
 383         lock_kernel();
 384         switch (cmd) {
 385         case LINUX_REBOOT_CMD_RESTART:
 386                 kernel_restart(NULL);
 387                 break;
 388
 389         case LINUX_REBOOT_CMD_CAD_ON:
 390                 C_A_D = 1;
 391                 break;
 392
 393         case LINUX_REBOOT_CMD_CAD_OFF:
 394                 C_A_D = 0;
 395                 break;
 396
 397         case LINUX_REBOOT_CMD_HALT:
 398                 kernel_halt();
 399                 unlock_kernel();
 400                 do_exit(0);
 401                 panic("cannot halt");
 402
 403         case LINUX_REBOOT_CMD_POWER_OFF:
 404                 kernel_power_off();
 405                 unlock_kernel();
 406                 do_exit(0);
 407                 break;
 408
 409         case LINUX_REBOOT_CMD_RESTART2:
 410                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 411                         unlock_kernel();
 412                         return -EFAULT;
 413                 }
 414                 buffer[sizeof(buffer) - 1] = '\0';
 415
 416                 kernel_restart(buffer);
 417                 break;
 418
 419 #ifdef CONFIG_KEXEC
 420         case LINUX_REBOOT_CMD_KEXEC:
 421                 ret = kernel_kexec();
 422                 break;
 423 #endif
 424
 425 #ifdef CONFIG_HIBERNATION
 426         case LINUX_REBOOT_CMD_SW_SUSPEND:
 427                 ret = hibernate();
 428                 break;
 429 #endif
 430
 431         default:
 432                 ret = -EINVAL;
 433                 break;
 434         }
 435         unlock_kernel();
 436         return ret;
 437 }
 438
 439 static void deferred_cad(struct work_struct *dummy)
 440 {
 441         kernel_restart(NULL);
 442 }
 443
 444 /*
 445  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 446  * As it's called within an interrupt, it may NOT sync: the only choice
 447  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 448  */
 449 void ctrl_alt_del(void)
 450 {
 451         static DECLARE_WORK(cad_work, deferred_cad);
 452
 453         if (C_A_D)
 454                 schedule_work(&cad_work);
 455         else
 456                 kill_cad_pid(SIGINT, 1);
 457 }
 458
 459 /*
 460  * Unprivileged users may change the real gid to the effective gid
 461  * or vice versa.  (BSD-style)
 462  *
 463  * If you set the real gid at all, or set the effective gid to a value not
 464  * equal to the real gid, then the saved gid is set to the new effective gid.
 465  *
 466  * This makes it possible for a setgid program to completely drop its
 467  * privileges, which is often a useful assertion to make when you are doing
 468  * a security audit over a program.
 469  *
 470  * The general idea is that a program which uses just setregid() will be
 471  * 100% compatible with BSD.  A program which uses just setgid() will be
 472  * 100% compatible with POSIX with saved IDs.
 473  *
 474  * SMP: There are not races, the GIDs are checked only by filesystem
 475  *      operations (as far as semantic preservation is concerned).
 476  */
 477 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 478 {
 479         const struct cred *old;
 480         struct cred *new;
 481         int retval;
 482
 483         new = prepare_creds();
 484         if (!new)
 485                 return -ENOMEM;
 486         old = current_cred();
 487
 488         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 489         if (retval)
 490                 goto error;
 491
 492         retval = -EPERM;
 493         if (rgid != (gid_t) -1) {
 494                 if (old->gid == rgid ||
 495                     old->egid == rgid ||
 496                     capable(CAP_SETGID))
 497                         new->gid = rgid;
 498                 else
 499                         goto error;
 500         }
 501         if (egid != (gid_t) -1) {
 502                 if (old->gid == egid ||
 503                     old->egid == egid ||
 504                     old->sgid == egid ||
 505                     capable(CAP_SETGID))
 506                         new->egid = egid;
 507                 else
 508                         goto error;
 509         }
 510
 511         if (rgid != (gid_t) -1 ||
 512             (egid != (gid_t) -1 && egid != old->gid))
 513                 new->sgid = new->egid;
 514         new->fsgid = new->egid;
 515
 516         return commit_creds(new);
 517
 518 error:
 519         abort_creds(new);
 520         return retval;
 521 }
 522
 523 /*
 524  * setgid() is implemented like SysV w/ SAVED_IDS
 525  *
 526  * SMP: Same implicit races as above.
 527  */
 528 SYSCALL_DEFINE1(setgid, gid_t, gid)
 529 {
 530         const struct cred *old;
 531         struct cred *new;
 532         int retval;
 533
 534         new = prepare_creds();
 535         if (!new)
 536                 return -ENOMEM;
 537         old = current_cred();
 538
 539         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 540         if (retval)
 541                 goto error;
 542
 543         retval = -EPERM;
 544         if (capable(CAP_SETGID))
 545                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 546         else if (gid == old->gid || gid == old->sgid)
 547                 new->egid = new->fsgid = gid;
 548         else
 549                 goto error;
 550
 551         return commit_creds(new);
 552
 553 error:
 554         abort_creds(new);
 555         return retval;
 556 }
 557
 558 /*
 559  * change the user struct in a credentials set to match the new UID
 560  */
 561 static int set_user(struct cred *new)
 562 {
 563         struct user_struct *new_user;
 564
 565         new_user = alloc_uid(current_user_ns(), new->uid);
 566         if (!new_user)
 567                 return -EAGAIN;
 568
 569         if (!task_can_switch_user(new_user, current)) {
 570                 free_uid(new_user);
 571                 return -EINVAL;
 572         }
 573
 574         if (atomic_read(&new_user->processes) >=
 575                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 576                         new_user != INIT_USER) {
 577                 free_uid(new_user);
 578                 return -EAGAIN;
 579         }
 580
 581         free_uid(new->user);
 582         new->user = new_user;
 583         return 0;
 584 }
 585
 586 /*
 587  * Unprivileged users may change the real uid to the effective uid
 588  * or vice versa.  (BSD-style)
 589  *
 590  * If you set the real uid at all, or set the effective uid to a value not
 591  * equal to the real uid, then the saved uid is set to the new effective uid.
 592  *
 593  * This makes it possible for a setuid program to completely drop its
 594  * privileges, which is often a useful assertion to make when you are doing
 595  * a security audit over a program.
 596  *
 597  * The general idea is that a program which uses just setreuid() will be
 598  * 100% compatible with BSD.  A program which uses just setuid() will be
 599  * 100% compatible with POSIX with saved IDs.
 600  */
 601 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 602 {
 603         const struct cred *old;
 604         struct cred *new;
 605         int retval;
 606
 607         new = prepare_creds();
 608         if (!new)
 609                 return -ENOMEM;
 610         old = current_cred();
 611
 612         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 613         if (retval)
 614                 goto error;
 615
 616         retval = -EPERM;
 617         if (ruid != (uid_t) -1) {
 618                 new->uid = ruid;
 619                 if (old->uid != ruid &&
 620                     old->euid != ruid &&
 621                     !capable(CAP_SETUID))
 622                         goto error;
 623         }
 624
 625         if (euid != (uid_t) -1) {
 626                 new->euid = euid;
 627                 if (old->uid != euid &&
 628                     old->euid != euid &&
 629                     old->suid != euid &&
 630                     !capable(CAP_SETUID))
 631                         goto error;
 632         }
 633
 634         if (new->uid != old->uid) {
 635                 retval = set_user(new);
 636                 if (retval < 0)
 637                         goto error;
 638         }
 639         if (ruid != (uid_t) -1 ||
 640             (euid != (uid_t) -1 && euid != old->uid))
 641                 new->suid = new->euid;
 642         new->fsuid = new->euid;
 643
 644         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 645         if (retval < 0)
 646                 goto error;
 647
 648         return commit_creds(new);
 649
 650 error:
 651         abort_creds(new);
 652         return retval;
 653 }
 654
 655 /*
 656  * setuid() is implemented like SysV with SAVED_IDS
 657  *
 658  * Note that SAVED_ID's is deficient in that a setuid root program
 659  * like sendmail, for example, cannot set its uid to be a normal
 660  * user and then switch back, because if you're root, setuid() sets
 661  * the saved uid too.  If you don't like this, blame the bright people
 662  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 663  * will allow a root program to temporarily drop privileges and be able to
 664  * regain them by swapping the real and effective uid.
 665  */
 666 SYSCALL_DEFINE1(setuid, uid_t, uid)
 667 {
 668         const struct cred *old;
 669         struct cred *new;
 670         int retval;
 671
 672         new = prepare_creds();
 673         if (!new)
 674                 return -ENOMEM;
 675         old = current_cred();
 676
 677         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 678         if (retval)
 679                 goto error;
 680
 681         retval = -EPERM;
 682         if (capable(CAP_SETUID)) {
 683                 new->suid = new->uid = uid;
 684                 if (uid != old->uid) {
 685                         retval = set_user(new);
 686                         if (retval < 0)
 687                                 goto error;
 688                 }
 689         } else if (uid != old->uid && uid != new->suid) {
 690                 goto error;
 691         }
 692
 693         new->fsuid = new->euid = uid;
 694
 695         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 696         if (retval < 0)
 697                 goto error;
 698
 699         return commit_creds(new);
 700
 701 error:
 702         abort_creds(new);
 703         return retval;
 704 }
 705
 706
 707 /*
 708  * This function implements a generic ability to update ruid, euid,
 709  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 710  */
 711 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 712 {
 713         const struct cred *old;
 714         struct cred *new;
 715         int retval;
 716
 717         new = prepare_creds();
 718         if (!new)
 719                 return -ENOMEM;
 720
 721         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 722         if (retval)
 723                 goto error;
 724         old = current_cred();
 725
 726         retval = -EPERM;
 727         if (!capable(CAP_SETUID)) {
 728                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 729                     ruid != old->euid  && ruid != old->suid)
 730                         goto error;
 731                 if (euid != (uid_t) -1 && euid != old->uid &&
 732                     euid != old->euid  && euid != old->suid)
 733                         goto error;
 734                 if (suid != (uid_t) -1 && suid != old->uid &&
 735                     suid != old->euid  && suid != old->suid)
 736                         goto error;
 737         }
 738
 739         if (ruid != (uid_t) -1) {
 740                 new->uid = ruid;
 741                 if (ruid != old->uid) {
 742                         retval = set_user(new);
 743                         if (retval < 0)
 744                                 goto error;
 745                 }
 746         }
 747         if (euid != (uid_t) -1)
 748                 new->euid = euid;
 749         if (suid != (uid_t) -1)
 750                 new->suid = suid;
 751         new->fsuid = new->euid;
 752
 753         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 754         if (retval < 0)
 755                 goto error;
 756
 757         return commit_creds(new);
 758
 759 error:
 760         abort_creds(new);
 761         return retval;
 762 }
 763
 764 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 765 {
 766         const struct cred *cred = current_cred();
 767         int retval;
 768
 769         if (!(retval   = put_user(cred->uid,  ruid)) &&
 770             !(retval   = put_user(cred->euid, euid)))
 771                 retval = put_user(cred->suid, suid);
 772
 773         return retval;
 774 }
 775
 776 /*
 777  * Same as above, but for rgid, egid, sgid.
 778  */
 779 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 780 {
 781         const struct cred *old;
 782         struct cred *new;
 783         int retval;
 784
 785         new = prepare_creds();
 786         if (!new)
 787                 return -ENOMEM;
 788         old = current_cred();
 789
 790         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 791         if (retval)
 792                 goto error;
 793
 794         retval = -EPERM;
 795         if (!capable(CAP_SETGID)) {
 796                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 797                     rgid != old->egid  && rgid != old->sgid)
 798                         goto error;
 799                 if (egid != (gid_t) -1 && egid != old->gid &&
 800                     egid != old->egid  && egid != old->sgid)
 801                         goto error;
 802                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 803                     sgid != old->egid  && sgid != old->sgid)
 804                         goto error;
 805         }
 806
 807         if (rgid != (gid_t) -1)
 808                 new->gid = rgid;
 809         if (egid != (gid_t) -1)
 810                 new->egid = egid;
 811         if (sgid != (gid_t) -1)
 812                 new->sgid = sgid;
 813         new->fsgid = new->egid;
 814
 815         return commit_creds(new);
 816
 817 error:
 818         abort_creds(new);
 819         return retval;
 820 }
 821
 822 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 823 {
 824         const struct cred *cred = current_cred();
 825         int retval;
 826
 827         if (!(retval   = put_user(cred->gid,  rgid)) &&
 828             !(retval   = put_user(cred->egid, egid)))
 829                 retval = put_user(cred->sgid, sgid);
 830
 831         return retval;
 832 }
 833
 834
 835 /*
 836  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 837  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 838  * whatever uid it wants to). It normally shadows "euid", except when
 839  * explicitly set by setfsuid() or for access..
 840  */
 841 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 842 {
 843         const struct cred *old;
 844         struct cred *new;
 845         uid_t old_fsuid;
 846
 847         new = prepare_creds();
 848         if (!new)
 849                 return current_fsuid();
 850         old = current_cred();
 851         old_fsuid = old->fsuid;
 852
 853         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
 854                 goto error;
 855
 856         if (uid == old->uid  || uid == old->euid  ||
 857             uid == old->suid || uid == old->fsuid ||
 858             capable(CAP_SETUID)) {
 859                 if (uid != old_fsuid) {
 860                         new->fsuid = uid;
 861                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 862                                 goto change_okay;
 863                 }
 864         }
 865
 866 error:
 867         abort_creds(new);
 868         return old_fsuid;
 869
 870 change_okay:
 871         commit_creds(new);
 872         return old_fsuid;
 873 }
 874
 875 /*
 876  * Samma på svenska..
 877  */
 878 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 879 {
 880         const struct cred *old;
 881         struct cred *new;
 882         gid_t old_fsgid;
 883
 884         new = prepare_creds();
 885         if (!new)
 886                 return current_fsgid();
 887         old = current_cred();
 888         old_fsgid = old->fsgid;
 889
 890         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 891                 goto error;
 892
 893         if (gid == old->gid  || gid == old->egid  ||
 894             gid == old->sgid || gid == old->fsgid ||
 895             capable(CAP_SETGID)) {
 896                 if (gid != old_fsgid) {
 897                         new->fsgid = gid;
 898                         goto change_okay;
 899                 }
 900         }
 901
 902 error:
 903         abort_creds(new);
 904         return old_fsgid;
 905
 906 change_okay:
 907         commit_creds(new);
 908         return old_fsgid;
 909 }
 910
 911 void do_sys_times(struct tms *tms)
 912 {
 913         struct task_cputime cputime;
 914         cputime_t cutime, cstime;
 915
 916         thread_group_cputime(current, &cputime);
 917         spin_lock_irq(&current->sighand->siglock);
 918         cutime = current->signal->cutime;
 919         cstime = current->signal->cstime;
 920         spin_unlock_irq(&current->sighand->siglock);
 921         tms->tms_utime = cputime_to_clock_t(cputime.utime);
 922         tms->tms_stime = cputime_to_clock_t(cputime.stime);
 923         tms->tms_cutime = cputime_to_clock_t(cutime);
 924         tms->tms_cstime = cputime_to_clock_t(cstime);
 925 }
 926
 927 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 928 {
 929         if (tbuf) {
 930                 struct tms tmp;
 931
 932                 do_sys_times(&tmp);
 933                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 934                         return -EFAULT;
 935         }
 936         force_successful_syscall_return();
 937         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 938 }
 939
 940 /*
 941  * This needs some heavy checking ...
 942  * I just haven't the stomach for it. I also don't fully
 943  * understand sessions/pgrp etc. Let somebody who does explain it.
 944  *
 945  * OK, I think I have the protection semantics right.... this is really
 946  * only important on a multi-user system anyway, to make sure one user
 947  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 948  *
 949  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 950  * LBT 04.03.94
 951  */
 952 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 953 {
 954         struct task_struct *p;
 955         struct task_struct *group_leader = current->group_leader;
 956         struct pid *pgrp;
 957         int err;
 958
 959         if (!pid)
 960                 pid = task_pid_vnr(group_leader);
 961         if (!pgid)
 962                 pgid = pid;
 963         if (pgid < 0)
 964                 return -EINVAL;
 965
 966         /* From this point forward we keep holding onto the tasklist lock
 967          * so that our parent does not change from under us. -DaveM
 968          */
 969         write_lock_irq(&tasklist_lock);
 970
 971         err = -ESRCH;
 972         p = find_task_by_vpid(pid);
 973         if (!p)
 974                 goto out;
 975
 976         err = -EINVAL;
 977         if (!thread_group_leader(p))
 978                 goto out;
 979
 980         if (same_thread_group(p->real_parent, group_leader)) {
 981                 err = -EPERM;
 982                 if (task_session(p) != task_session(group_leader))
 983                         goto out;
 984                 err = -EACCES;
 985                 if (p->did_exec)
 986                         goto out;
 987         } else {
 988                 err = -ESRCH;
 989                 if (p != group_leader)
 990                         goto out;
 991         }
 992
 993         err = -EPERM;
 994         if (p->signal->leader)
 995                 goto out;
 996
 997         pgrp = task_pid(p);
 998         if (pgid != pid) {
 999                 struct task_struct *g;
1000
1001                 pgrp = find_vpid(pgid);
1002                 g = pid_task(pgrp, PIDTYPE_PGID);
1003                 if (!g || task_session(g) != task_session(group_leader))
1004                         goto out;
1005         }
1006
1007         err = security_task_setpgid(p, pgid);
1008         if (err)
1009                 goto out;
1010
1011         if (task_pgrp(p) != pgrp)
1012                 change_pid(p, PIDTYPE_PGID, pgrp);
1013
1014         err = 0;
1015 out:
1016         /* All paths lead to here, thus we are safe. -DaveM */
1017         write_unlock_irq(&tasklist_lock);
1018         return err;
1019 }
1020
1021 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1022 {
1023         struct task_struct *p;
1024         struct pid *grp;
1025         int retval;
1026
1027         rcu_read_lock();
1028         if (!pid)
1029                 grp = task_pgrp(current);
1030         else {
1031                 retval = -ESRCH;
1032                 p = find_task_by_vpid(pid);
1033                 if (!p)
1034                         goto out;
1035                 grp = task_pgrp(p);
1036                 if (!grp)
1037                         goto out;
1038
1039                 retval = security_task_getpgid(p);
1040                 if (retval)
1041                         goto out;
1042         }
1043         retval = pid_vnr(grp);
1044 out:
1045         rcu_read_unlock();
1046         return retval;
1047 }
1048
1049 #ifdef __ARCH_WANT_SYS_GETPGRP
1050
1051 SYSCALL_DEFINE0(getpgrp)
1052 {
1053         return sys_getpgid(0);
1054 }
1055
1056 #endif
1057
1058 SYSCALL_DEFINE1(getsid, pid_t, pid)
1059 {
1060         struct task_struct *p;
1061         struct pid *sid;
1062         int retval;
1063
1064         rcu_read_lock();
1065         if (!pid)
1066                 sid = task_session(current);
1067         else {
1068                 retval = -ESRCH;
1069                 p = find_task_by_vpid(pid);
1070                 if (!p)
1071                         goto out;
1072                 sid = task_session(p);
1073                 if (!sid)
1074                         goto out;
1075
1076                 retval = security_task_getsid(p);
1077                 if (retval)
1078                         goto out;
1079         }
1080         retval = pid_vnr(sid);
1081 out:
1082         rcu_read_unlock();
1083         return retval;
1084 }
1085
1086 SYSCALL_DEFINE0(setsid)
1087 {
1088         struct task_struct *group_leader = current->group_leader;
1089         struct pid *sid = task_pid(group_leader);
1090         pid_t session = pid_vnr(sid);
1091         int err = -EPERM;
1092
1093         write_lock_irq(&tasklist_lock);
1094         /* Fail if I am already a session leader */
1095         if (group_leader->signal->leader)
1096                 goto out;
1097
1098         /* Fail if a process group id already exists that equals the
1099          * proposed session id.
1100          */
1101         if (pid_task(sid, PIDTYPE_PGID))
1102                 goto out;
1103
1104         group_leader->signal->leader = 1;
1105         __set_special_pids(sid);
1106
1107         proc_clear_tty(group_leader);
1108
1109         err = session;
1110 out:
1111         write_unlock_irq(&tasklist_lock);
1112         return err;
1113 }
1114
1115 /*
1116  * Supplementary group IDs
1117  */
1118
1119 /* init to 2 - one for init_task, one to ensure it is never freed */
1120 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1121
1122 struct group_info *groups_alloc(int gidsetsize)
1123 {
1124         struct group_info *group_info;
1125         int nblocks;
1126         int i;
1127
1128         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1129         /* Make sure we always allocate at least one indirect block pointer */
1130         nblocks = nblocks ? : 1;
1131         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1132         if (!group_info)
1133                 return NULL;
1134         group_info->ngroups = gidsetsize;
1135         group_info->nblocks = nblocks;
1136         atomic_set(&group_info->usage, 1);
1137
1138         if (gidsetsize <= NGROUPS_SMALL)
1139                 group_info->blocks[0] = group_info->small_block;
1140         else {
1141                 for (i = 0; i < nblocks; i++) {
1142                         gid_t *b;
1143                         b = (void *)__get_free_page(GFP_USER);
1144                         if (!b)
1145                                 goto out_undo_partial_alloc;
1146                         group_info->blocks[i] = b;
1147                 }
1148         }
1149         return group_info;
1150
1151 out_undo_partial_alloc:
1152         while (--i >= 0) {
1153                 free_page((unsigned long)group_info->blocks[i]);
1154         }
1155         kfree(group_info);
1156         return NULL;
1157 }
1158
1159 EXPORT_SYMBOL(groups_alloc);
1160
1161 void groups_free(struct group_info *group_info)
1162 {
1163         if (group_info->blocks[0] != group_info->small_block) {
1164                 int i;
1165                 for (i = 0; i < group_info->nblocks; i++)
1166                         free_page((unsigned long)group_info->blocks[i]);
1167         }
1168         kfree(group_info);
1169 }
1170
1171 EXPORT_SYMBOL(groups_free);
1172
1173 /* export the group_info to a user-space array */
1174 static int groups_to_user(gid_t __user *grouplist,
1175                           const struct group_info *group_info)
1176 {
1177         int i;
1178         unsigned int count = group_info->ngroups;
1179
1180         for (i = 0; i < group_info->nblocks; i++) {
1181                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1182                 unsigned int len = cp_count * sizeof(*grouplist);
1183
1184                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1185                         return -EFAULT;
1186
1187                 grouplist += NGROUPS_PER_BLOCK;
1188                 count -= cp_count;
1189         }
1190         return 0;
1191 }
1192
1193 /* fill a group_info from a user-space array - it must be allocated already */
1194 static int groups_from_user(struct group_info *group_info,
1195     gid_t __user *grouplist)
1196 {
1197         int i;
1198         unsigned int count = group_info->ngroups;
1199
1200         for (i = 0; i < group_info->nblocks; i++) {
1201                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1202                 unsigned int len = cp_count * sizeof(*grouplist);
1203
1204                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1205                         return -EFAULT;
1206
1207                 grouplist += NGROUPS_PER_BLOCK;
1208                 count -= cp_count;
1209         }
1210         return 0;
1211 }
1212
1213 /* a simple Shell sort */
1214 static void groups_sort(struct group_info *group_info)
1215 {
1216         int base, max, stride;
1217         int gidsetsize = group_info->ngroups;
1218
1219         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1220                 ; /* nothing */
1221         stride /= 3;
1222
1223         while (stride) {
1224                 max = gidsetsize - stride;
1225                 for (base = 0; base < max; base++) {
1226                         int left = base;
1227                         int right = left + stride;
1228                         gid_t tmp = GROUP_AT(group_info, right);
1229
1230                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1231                                 GROUP_AT(group_info, right) =
1232                                     GROUP_AT(group_info, left);
1233                                 right = left;
1234                                 left -= stride;
1235                         }
1236                         GROUP_AT(group_info, right) = tmp;
1237                 }
1238                 stride /= 3;
1239         }
1240 }
1241
1242 /* a simple bsearch */
1243 int groups_search(const struct group_info *group_info, gid_t grp)
1244 {
1245         unsigned int left, right;
1246
1247         if (!group_info)
1248                 return 0;
1249
1250         left = 0;
1251         right = group_info->ngroups;
1252         while (left < right) {
1253                 unsigned int mid = (left+right)/2;
1254                 int cmp = grp - GROUP_AT(group_info, mid);
1255                 if (cmp > 0)
1256                         left = mid + 1;
1257                 else if (cmp < 0)
1258                         right = mid;
1259                 else
1260                         return 1;
1261         }
1262         return 0;
1263 }
1264
1265 /**
1266  * set_groups - Change a group subscription in a set of credentials
1267  * @new: The newly prepared set of credentials to alter
1268  * @group_info: The group list to install
1269  *
1270  * Validate a group subscription and, if valid, insert it into a set
1271  * of credentials.
1272  */
1273 int set_groups(struct cred *new, struct group_info *group_info)
1274 {
1275         int retval;
1276
1277         retval = security_task_setgroups(group_info);
1278         if (retval)
1279                 return retval;
1280
1281         put_group_info(new->group_info);
1282         groups_sort(group_info);
1283         get_group_info(group_info);
1284         new->group_info = group_info;
1285         return 0;
1286 }
1287
1288 EXPORT_SYMBOL(set_groups);
1289
1290 /**
1291  * set_current_groups - Change current's group subscription
1292  * @group_info: The group list to impose
1293  *
1294  * Validate a group subscription and, if valid, impose it upon current's task
1295  * security record.
1296  */
1297 int set_current_groups(struct group_info *group_info)
1298 {
1299         struct cred *new;
1300         int ret;
1301
1302         new = prepare_creds();
1303         if (!new)
1304                 return -ENOMEM;
1305
1306         ret = set_groups(new, group_info);
1307         if (ret < 0) {
1308                 abort_creds(new);
1309                 return ret;
1310         }
1311
1312         return commit_creds(new);
1313 }
1314
1315 EXPORT_SYMBOL(set_current_groups);
1316
1317 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1318 {
1319         const struct cred *cred = current_cred();
1320         int i;
1321
1322         if (gidsetsize < 0)
1323                 return -EINVAL;
1324
1325         /* no need to grab task_lock here; it cannot change */
1326         i = cred->group_info->ngroups;
1327         if (gidsetsize) {
1328                 if (i > gidsetsize) {
1329                         i = -EINVAL;
1330                         goto out;
1331                 }
1332                 if (groups_to_user(grouplist, cred->group_info)) {
1333                         i = -EFAULT;
1334                         goto out;
1335                 }
1336         }
1337 out:
1338         return i;
1339 }
1340
1341 /*
1342  *      SMP: Our groups are copy-on-write. We can set them safely
1343  *      without another task interfering.
1344  */
1345
1346 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1347 {
1348         struct group_info *group_info;
1349         int retval;
1350
1351         if (!capable(CAP_SETGID))
1352                 return -EPERM;
1353         if ((unsigned)gidsetsize > NGROUPS_MAX)
1354                 return -EINVAL;
1355
1356         group_info = groups_alloc(gidsetsize);
1357         if (!group_info)
1358                 return -ENOMEM;
1359         retval = groups_from_user(group_info, grouplist);
1360         if (retval) {
1361                 put_group_info(group_info);
1362                 return retval;
1363         }
1364
1365         retval = set_current_groups(group_info);
1366         put_group_info(group_info);
1367
1368         return retval;
1369 }
1370
1371 /*
1372  * Check whether we're fsgid/egid or in the supplemental group..
1373  */
1374 int in_group_p(gid_t grp)
1375 {
1376         const struct cred *cred = current_cred();
1377         int retval = 1;
1378
1379         if (grp != cred->fsgid)
1380                 retval = groups_search(cred->group_info, grp);
1381         return retval;
1382 }
1383
1384 EXPORT_SYMBOL(in_group_p);
1385
1386 int in_egroup_p(gid_t grp)
1387 {
1388         const struct cred *cred = current_cred();
1389         int retval = 1;
1390
1391         if (grp != cred->egid)
1392                 retval = groups_search(cred->group_info, grp);
1393         return retval;
1394 }
1395
1396 EXPORT_SYMBOL(in_egroup_p);
1397
1398 DECLARE_RWSEM(uts_sem);
1399
1400 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1401 {
1402         int errno = 0;
1403
1404         down_read(&uts_sem);
1405         if (copy_to_user(name, utsname(), sizeof *name))
1406                 errno = -EFAULT;
1407         up_read(&uts_sem);
1408         return errno;
1409 }
1410
1411 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1412 {
1413         int errno;
1414         char tmp[__NEW_UTS_LEN];
1415
1416         if (!capable(CAP_SYS_ADMIN))
1417                 return -EPERM;
1418         if (len < 0 || len > __NEW_UTS_LEN)
1419                 return -EINVAL;
1420         down_write(&uts_sem);
1421         errno = -EFAULT;
1422         if (!copy_from_user(tmp, name, len)) {
1423                 struct new_utsname *u = utsname();
1424
1425                 memcpy(u->nodename, tmp, len);
1426                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1427                 errno = 0;
1428         }
1429         up_write(&uts_sem);
1430         return errno;
1431 }
1432
1433 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1434
1435 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1436 {
1437         int i, errno;
1438         struct new_utsname *u;
1439
1440         if (len < 0)
1441                 return -EINVAL;
1442         down_read(&uts_sem);
1443         u = utsname();
1444         i = 1 + strlen(u->nodename);
1445         if (i > len)
1446                 i = len;
1447         errno = 0;
1448         if (copy_to_user(name, u->nodename, i))
1449                 errno = -EFAULT;
1450         up_read(&uts_sem);
1451         return errno;
1452 }
1453
1454 #endif
1455
1456 /*
1457  * Only setdomainname; getdomainname can be implemented by calling
1458  * uname()
1459  */
1460 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1461 {
1462         int errno;
1463         char tmp[__NEW_UTS_LEN];
1464
1465         if (!capable(CAP_SYS_ADMIN))
1466                 return -EPERM;
1467         if (len < 0 || len > __NEW_UTS_LEN)
1468                 return -EINVAL;
1469
1470         down_write(&uts_sem);
1471         errno = -EFAULT;
1472         if (!copy_from_user(tmp, name, len)) {
1473                 struct new_utsname *u = utsname();
1474
1475                 memcpy(u->domainname, tmp, len);
1476                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1477                 errno = 0;
1478         }
1479         up_write(&uts_sem);
1480         return errno;
1481 }
1482
1483 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1484 {
1485         if (resource >= RLIM_NLIMITS)
1486                 return -EINVAL;
1487         else {
1488                 struct rlimit value;
1489                 task_lock(current->group_leader);
1490                 value = current->signal->rlim[resource];
1491                 task_unlock(current->group_leader);
1492                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1493         }
1494 }
1495
1496 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1497
1498 /*
1499  *      Back compatibility for getrlimit. Needed for some apps.
1500  */
1501
1502 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1503                 struct rlimit __user *, rlim)
1504 {
1505         struct rlimit x;
1506         if (resource >= RLIM_NLIMITS)
1507                 return -EINVAL;
1508
1509         task_lock(current->group_leader);
1510         x = current->signal->rlim[resource];
1511         task_unlock(current->group_leader);
1512         if (x.rlim_cur > 0x7FFFFFFF)
1513                 x.rlim_cur = 0x7FFFFFFF;
1514         if (x.rlim_max > 0x7FFFFFFF)
1515                 x.rlim_max = 0x7FFFFFFF;
1516         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1517 }
1518
1519 #endif
1520
1521 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1522 {
1523         struct rlimit new_rlim, *old_rlim;
1524         int retval;
1525
1526         if (resource >= RLIM_NLIMITS)
1527                 return -EINVAL;
1528         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1529                 return -EFAULT;
1530         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1531                 return -EINVAL;
1532         old_rlim = current->signal->rlim + resource;
1533         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1534             !capable(CAP_SYS_RESOURCE))
1535                 return -EPERM;
1536         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1537                 return -EPERM;
1538
1539         retval = security_task_setrlimit(resource, &new_rlim);
1540         if (retval)
1541                 return retval;
1542
1543         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1544                 /*
1545                  * The caller is asking for an immediate RLIMIT_CPU
1546                  * expiry.  But we use the zero value to mean "it was
1547                  * never set".  So let's cheat and make it one second
1548                  * instead
1549                  */
1550                 new_rlim.rlim_cur = 1;
1551         }
1552
1553         task_lock(current->group_leader);
1554         *old_rlim = new_rlim;
1555         task_unlock(current->group_leader);
1556
1557         if (resource != RLIMIT_CPU)
1558                 goto out;
1559
1560         /*
1561          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1562          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1563          * very long-standing error, and fixing it now risks breakage of
1564          * applications, so we live with it
1565          */
1566         if (new_rlim.rlim_cur == RLIM_INFINITY)
1567                 goto out;
1568
1569         update_rlimit_cpu(new_rlim.rlim_cur);
1570 out:
1571         return 0;
1572 }
1573
1574 /*
1575  * It would make sense to put struct rusage in the task_struct,
1576  * except that would make the task_struct be *really big*.  After
1577  * task_struct gets moved into malloc'ed memory, it would
1578  * make sense to do this.  It will make moving the rest of the information
1579  * a lot simpler!  (Which we're not doing right now because we're not
1580  * measuring them yet).
1581  *
1582  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1583  * races with threads incrementing their own counters.  But since word
1584  * reads are atomic, we either get new values or old values and we don't
1585  * care which for the sums.  We always take the siglock to protect reading
1586  * the c* fields from p->signal from races with exit.c updating those
1587  * fields when reaping, so a sample either gets all the additions of a
1588  * given child after it's reaped, or none so this sample is before reaping.
1589  *
1590  * Locking:
1591  * We need to take the siglock for CHILDEREN, SELF and BOTH
1592  * for  the cases current multithreaded, non-current single threaded
1593  * non-current multithreaded.  Thread traversal is now safe with
1594  * the siglock held.
1595  * Strictly speaking, we donot need to take the siglock if we are current and
1596  * single threaded,  as no one else can take our signal_struct away, no one
1597  * else can  reap the  children to update signal->c* counters, and no one else
1598  * can race with the signal-> fields. If we do not take any lock, the
1599  * signal-> fields could be read out of order while another thread was just
1600  * exiting. So we should  place a read memory barrier when we avoid the lock.
1601  * On the writer side,  write memory barrier is implied in  __exit_signal
1602  * as __exit_signal releases  the siglock spinlock after updating the signal->
1603  * fields. But we don't do this yet to keep things simple.
1604  *
1605  */
1606
1607 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1608 {
1609         r->ru_nvcsw += t->nvcsw;
1610         r->ru_nivcsw += t->nivcsw;
1611         r->ru_minflt += t->min_flt;
1612         r->ru_majflt += t->maj_flt;
1613         r->ru_inblock += task_io_get_inblock(t);
1614         r->ru_oublock += task_io_get_oublock(t);
1615 }
1616
1617 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1618 {
1619         struct task_struct *t;
1620         unsigned long flags;
1621         cputime_t utime, stime;
1622         struct task_cputime cputime;
1623
1624         memset((char *) r, 0, sizeof *r);
1625         utime = stime = cputime_zero;
1626
1627         if (who == RUSAGE_THREAD) {
1628                 utime = task_utime(current);
1629                 stime = task_stime(current);
1630                 accumulate_thread_rusage(p, r);
1631                 goto out;
1632         }
1633
1634         if (!lock_task_sighand(p, &flags))
1635                 return;
1636
1637         switch (who) {
1638                 case RUSAGE_BOTH:
1639                 case RUSAGE_CHILDREN:
1640                         utime = p->signal->cutime;
1641                         stime = p->signal->cstime;
1642                         r->ru_nvcsw = p->signal->cnvcsw;
1643                         r->ru_nivcsw = p->signal->cnivcsw;
1644                         r->ru_minflt = p->signal->cmin_flt;
1645                         r->ru_majflt = p->signal->cmaj_flt;
1646                         r->ru_inblock = p->signal->cinblock;
1647                         r->ru_oublock = p->signal->coublock;
1648
1649                         if (who == RUSAGE_CHILDREN)
1650                                 break;
1651
1652                 case RUSAGE_SELF:
1653                         thread_group_cputime(p, &cputime);
1654                         utime = cputime_add(utime, cputime.utime);
1655                         stime = cputime_add(stime, cputime.stime);
1656                         r->ru_nvcsw += p->signal->nvcsw;
1657                         r->ru_nivcsw += p->signal->nivcsw;
1658                         r->ru_minflt += p->signal->min_flt;
1659                         r->ru_majflt += p->signal->maj_flt;
1660                         r->ru_inblock += p->signal->inblock;
1661                         r->ru_oublock += p->signal->oublock;
1662                         t = p;
1663                         do {
1664                                 accumulate_thread_rusage(t, r);
1665                                 t = next_thread(t);
1666                         } while (t != p);
1667                         break;
1668
1669                 default:
1670                         BUG();
1671         }
1672         unlock_task_sighand(p, &flags);
1673
1674 out:
1675         cputime_to_timeval(utime, &r->ru_utime);
1676         cputime_to_timeval(stime, &r->ru_stime);
1677 }
1678
1679 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1680 {
1681         struct rusage r;
1682         k_getrusage(p, who, &r);
1683         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1684 }
1685
1686 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1687 {
1688         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1689             who != RUSAGE_THREAD)
1690                 return -EINVAL;
1691         return getrusage(current, who, ru);
1692 }
1693
1694 SYSCALL_DEFINE1(umask, int, mask)
1695 {
1696         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1697         return mask;
1698 }
1699
1700 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1701                 unsigned long, arg4, unsigned long, arg5)
1702 {
1703         struct task_struct *me = current;
1704         unsigned char comm[sizeof(me->comm)];
1705         long error;
1706
1707         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1708         if (error != -ENOSYS)
1709                 return error;
1710
1711         error = 0;
1712         switch (option) {
1713                 case PR_SET_PDEATHSIG:
1714                         if (!valid_signal(arg2)) {
1715                                 error = -EINVAL;
1716                                 break;
1717                         }
1718                         me->pdeath_signal = arg2;
1719                         error = 0;
1720                         break;
1721                 case PR_GET_PDEATHSIG:
1722                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1723                         break;
1724                 case PR_GET_DUMPABLE:
1725                         error = get_dumpable(me->mm);
1726                         break;
1727                 case PR_SET_DUMPABLE:
1728                         if (arg2 < 0 || arg2 > 1) {
1729                                 error = -EINVAL;
1730                                 break;
1731                         }
1732                         set_dumpable(me->mm, arg2);
1733                         error = 0;
1734                         break;
1735
1736                 case PR_SET_UNALIGN:
1737                         error = SET_UNALIGN_CTL(me, arg2);
1738                         break;
1739                 case PR_GET_UNALIGN:
1740                         error = GET_UNALIGN_CTL(me, arg2);
1741                         break;
1742                 case PR_SET_FPEMU:
1743                         error = SET_FPEMU_CTL(me, arg2);
1744                         break;
1745                 case PR_GET_FPEMU:
1746                         error = GET_FPEMU_CTL(me, arg2);
1747                         break;
1748                 case PR_SET_FPEXC:
1749                         error = SET_FPEXC_CTL(me, arg2);
1750                         break;
1751                 case PR_GET_FPEXC:
1752                         error = GET_FPEXC_CTL(me, arg2);
1753                         break;
1754                 case PR_GET_TIMING:
1755                         error = PR_TIMING_STATISTICAL;
1756                         break;
1757                 case PR_SET_TIMING:
1758                         if (arg2 != PR_TIMING_STATISTICAL)
1759                                 error = -EINVAL;
1760                         else
1761                                 error = 0;
1762                         break;
1763
1764                 case PR_SET_NAME:
1765                         comm[sizeof(me->comm)-1] = 0;
1766                         if (strncpy_from_user(comm, (char __user *)arg2,
1767                                               sizeof(me->comm) - 1) < 0)
1768                                 return -EFAULT;
1769                         set_task_comm(me, comm);
1770                         return 0;
1771                 case PR_GET_NAME:
1772                         get_task_comm(comm, me);
1773                         if (copy_to_user((char __user *)arg2, comm,
1774                                          sizeof(comm)))
1775                                 return -EFAULT;
1776                         return 0;
1777                 case PR_GET_ENDIAN:
1778                         error = GET_ENDIAN(me, arg2);
1779                         break;
1780                 case PR_SET_ENDIAN:
1781                         error = SET_ENDIAN(me, arg2);
1782                         break;
1783
1784                 case PR_GET_SECCOMP:
1785                         error = prctl_get_seccomp();
1786                         break;
1787                 case PR_SET_SECCOMP:
1788                         error = prctl_set_seccomp(arg2);
1789                         break;
1790                 case PR_GET_TSC:
1791                         error = GET_TSC_CTL(arg2);
1792                         break;
1793                 case PR_SET_TSC:
1794                         error = SET_TSC_CTL(arg2);
1795                         break;
1796                 case PR_GET_TIMERSLACK:
1797                         error = current->timer_slack_ns;
1798                         break;
1799                 case PR_SET_TIMERSLACK:
1800                         if (arg2 <= 0)
1801                                 current->timer_slack_ns =
1802                                         current->default_timer_slack_ns;
1803                         else
1804                                 current->timer_slack_ns = arg2;
1805                         error = 0;
1806                         break;
1807                 default:
1808                         error = -EINVAL;
1809                         break;
1810         }
1811         return error;
1812 }
1813
1814 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1815                 struct getcpu_cache __user *, unused)
1816 {
1817         int err = 0;
1818         int cpu = raw_smp_processor_id();
1819         if (cpup)
1820                 err |= put_user(cpu, cpup);
1821         if (nodep)
1822                 err |= put_user(cpu_to_node(cpu), nodep);
1823         return err ? -EFAULT : 0;
1824 }
1825
1826 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1827
1828 static void argv_cleanup(char **argv, char **envp)
1829 {
1830         argv_free(argv);
1831 }
1832
1833 /**
1834  * orderly_poweroff - Trigger an orderly system poweroff
1835  * @force: force poweroff if command execution fails
1836  *
1837  * This may be called from any context to trigger a system shutdown.
1838  * If the orderly shutdown fails, it will force an immediate shutdown.
1839  */
1840 int orderly_poweroff(bool force)
1841 {
1842         int argc;
1843         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1844         static char *envp[] = {
1845                 "HOME=/",
1846                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1847                 NULL
1848         };
1849         int ret = -ENOMEM;
1850         struct subprocess_info *info;
1851
1852         if (argv == NULL) {
1853                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1854                        __func__, poweroff_cmd);
1855                 goto out;
1856         }
1857
1858         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1859         if (info == NULL) {
1860                 argv_free(argv);
1861                 goto out;
1862         }
1863
1864         call_usermodehelper_setcleanup(info, argv_cleanup);
1865
1866         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1867
1868   out:
1869         if (ret && force) {
1870                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1871                        "forcing the issue\n");
1872
1873                 /* I guess this should try to kick off some daemon to
1874                    sync and poweroff asap.  Or not even bother syncing
1875                    if we're doing an emergency shutdown? */
1876                 emergency_sync();
1877                 kernel_power_off();
1878         }
1879
1880         return ret;
1881 }
1882 EXPORT_SYMBOL_GPL(orderly_poweroff);