SAFE public projects git trees. - safe/jmp/linux-2.6/blob - kernel/sys.c

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/module.h>
   9 #include <linux/mm.h>
  10 #include <linux/utsname.h>
  11 #include <linux/mman.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/notifier.h>
  14 #include <linux/reboot.h>
  15 #include <linux/prctl.h>
  16 #include <linux/init.h>
  17 #include <linux/highuid.h>
  18 #include <linux/fs.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kexec.h>
  21 #include <linux/workqueue.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
  32 #include <linux/compat.h>
  33 #include <linux/syscalls.h>
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/io.h>
  37 #include <asm/unistd.h>
  38
  39 #ifndef SET_UNALIGN_CTL
  40 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  41 #endif
  42 #ifndef GET_UNALIGN_CTL
  43 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  44 #endif
  45 #ifndef SET_FPEMU_CTL
  46 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  47 #endif
  48 #ifndef GET_FPEMU_CTL
  49 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  50 #endif
  51 #ifndef SET_FPEXC_CTL
  52 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  53 #endif
  54 #ifndef GET_FPEXC_CTL
  55 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  56 #endif
  57
  58 /*
  59  * this is where the system-wide overflow UID and GID are defined, for
  60  * architectures that now have 32-bit UID/GID but didn't in the past
  61  */
  62
  63 int overflowuid = DEFAULT_OVERFLOWUID;
  64 int overflowgid = DEFAULT_OVERFLOWGID;
  65
  66 #ifdef CONFIG_UID16
  67 EXPORT_SYMBOL(overflowuid);
  68 EXPORT_SYMBOL(overflowgid);
  69 #endif
  70
  71 /*
  72  * the same as above, but for filesystems which can only store a 16-bit
  73  * UID and GID. as such, this is needed on all architectures
  74  */
  75
  76 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  77 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  78
  79 EXPORT_SYMBOL(fs_overflowuid);
  80 EXPORT_SYMBOL(fs_overflowgid);
  81
  82 /*
  83  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
  84  */
  85
  86 int C_A_D = 1;
  87 int cad_pid = 1;
  88
  89 /*
  90  *      Notifier list for kernel code which wants to be called
  91  *      at shutdown. This is used to stop any idling DMA operations
  92  *      and the like.
  93  */
  94
  95 static struct notifier_block *reboot_notifier_list;
  96 static DEFINE_RWLOCK(notifier_lock);
  97
  98 /**
  99  *      notifier_chain_register - Add notifier to a notifier chain
 100  *      @list: Pointer to root list pointer
 101  *      @n: New entry in notifier chain
 102  *
 103  *      Adds a notifier to a notifier chain.
 104  *
 105  *      Currently always returns zero.
 106  */
 107
 108 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
 109 {
 110         write_lock(&notifier_lock);
 111         while(*list)
 112         {
 113                 if(n->priority > (*list)->priority)
 114                         break;
 115                 list= &((*list)->next);
 116         }
 117         n->next = *list;
 118         *list=n;
 119         write_unlock(&notifier_lock);
 120         return 0;
 121 }
 122
 123 EXPORT_SYMBOL(notifier_chain_register);
 124
 125 /**
 126  *      notifier_chain_unregister - Remove notifier from a notifier chain
 127  *      @nl: Pointer to root list pointer
 128  *      @n: New entry in notifier chain
 129  *
 130  *      Removes a notifier from a notifier chain.
 131  *
 132  *      Returns zero on success, or %-ENOENT on failure.
 133  */
 134
 135 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
 136 {
 137         write_lock(&notifier_lock);
 138         while((*nl)!=NULL)
 139         {
 140                 if((*nl)==n)
 141                 {
 142                         *nl=n->next;
 143                         write_unlock(&notifier_lock);
 144                         return 0;
 145                 }
 146                 nl=&((*nl)->next);
 147         }
 148         write_unlock(&notifier_lock);
 149         return -ENOENT;
 150 }
 151
 152 EXPORT_SYMBOL(notifier_chain_unregister);
 153
 154 /**
 155  *      notifier_call_chain - Call functions in a notifier chain
 156  *      @n: Pointer to root pointer of notifier chain
 157  *      @val: Value passed unmodified to notifier function
 158  *      @v: Pointer passed unmodified to notifier function
 159  *
 160  *      Calls each function in a notifier chain in turn.
 161  *
 162  *      If the return value of the notifier can be and'd
 163  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
 164  *      will return immediately, with the return value of
 165  *      the notifier function which halted execution.
 166  *      Otherwise, the return value is the return value
 167  *      of the last notifier function called.
 168  */
 169
 170 int notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
 171 {
 172         int ret=NOTIFY_DONE;
 173         struct notifier_block *nb = *n;
 174
 175         while(nb)
 176         {
 177                 ret=nb->notifier_call(nb,val,v);
 178                 if(ret&NOTIFY_STOP_MASK)
 179                 {
 180                         return ret;
 181                 }
 182                 nb=nb->next;
 183         }
 184         return ret;
 185 }
 186
 187 EXPORT_SYMBOL(notifier_call_chain);
 188
 189 /**
 190  *      register_reboot_notifier - Register function to be called at reboot time
 191  *      @nb: Info about notifier function to be called
 192  *
 193  *      Registers a function with the list of functions
 194  *      to be called at reboot time.
 195  *
 196  *      Currently always returns zero, as notifier_chain_register
 197  *      always returns zero.
 198  */
 199
 200 int register_reboot_notifier(struct notifier_block * nb)
 201 {
 202         return notifier_chain_register(&reboot_notifier_list, nb);
 203 }
 204
 205 EXPORT_SYMBOL(register_reboot_notifier);
 206
 207 /**
 208  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 209  *      @nb: Hook to be unregistered
 210  *
 211  *      Unregisters a previously registered reboot
 212  *      notifier function.
 213  *
 214  *      Returns zero on success, or %-ENOENT on failure.
 215  */
 216
 217 int unregister_reboot_notifier(struct notifier_block * nb)
 218 {
 219         return notifier_chain_unregister(&reboot_notifier_list, nb);
 220 }
 221
 222 EXPORT_SYMBOL(unregister_reboot_notifier);
 223
 224 static int set_one_prio(struct task_struct *p, int niceval, int error)
 225 {
 226         int no_nice;
 227
 228         if (p->uid != current->euid &&
 229                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
 230                 error = -EPERM;
 231                 goto out;
 232         }
 233         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 234                 error = -EACCES;
 235                 goto out;
 236         }
 237         no_nice = security_task_setnice(p, niceval);
 238         if (no_nice) {
 239                 error = no_nice;
 240                 goto out;
 241         }
 242         if (error == -ESRCH)
 243                 error = 0;
 244         set_user_nice(p, niceval);
 245 out:
 246         return error;
 247 }
 248
 249 asmlinkage long sys_setpriority(int which, int who, int niceval)
 250 {
 251         struct task_struct *g, *p;
 252         struct user_struct *user;
 253         int error = -EINVAL;
 254
 255         if (which > 2 || which < 0)
 256                 goto out;
 257
 258         /* normalize: avoid signed division (rounding problems) */
 259         error = -ESRCH;
 260         if (niceval < -20)
 261                 niceval = -20;
 262         if (niceval > 19)
 263                 niceval = 19;
 264
 265         read_lock(&tasklist_lock);
 266         switch (which) {
 267                 case PRIO_PROCESS:
 268                         if (!who)
 269                                 who = current->pid;
 270                         p = find_task_by_pid(who);
 271                         if (p)
 272                                 error = set_one_prio(p, niceval, error);
 273                         break;
 274                 case PRIO_PGRP:
 275                         if (!who)
 276                                 who = process_group(current);
 277                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 278                                 error = set_one_prio(p, niceval, error);
 279                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 280                         break;
 281                 case PRIO_USER:
 282                         user = current->user;
 283                         if (!who)
 284                                 who = current->uid;
 285                         else
 286                                 if ((who != current->uid) && !(user = find_user(who)))
 287                                         goto out_unlock;        /* No processes for this user */
 288
 289                         do_each_thread(g, p)
 290                                 if (p->uid == who)
 291                                         error = set_one_prio(p, niceval, error);
 292                         while_each_thread(g, p);
 293                         if (who != current->uid)
 294                                 free_uid(user);         /* For find_user() */
 295                         break;
 296         }
 297 out_unlock:
 298         read_unlock(&tasklist_lock);
 299 out:
 300         return error;
 301 }
 302
 303 /*
 304  * Ugh. To avoid negative return values, "getpriority()" will
 305  * not return the normal nice-value, but a negated value that
 306  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 307  * to stay compatible.
 308  */
 309 asmlinkage long sys_getpriority(int which, int who)
 310 {
 311         struct task_struct *g, *p;
 312         struct user_struct *user;
 313         long niceval, retval = -ESRCH;
 314
 315         if (which > 2 || which < 0)
 316                 return -EINVAL;
 317
 318         read_lock(&tasklist_lock);
 319         switch (which) {
 320                 case PRIO_PROCESS:
 321                         if (!who)
 322                                 who = current->pid;
 323                         p = find_task_by_pid(who);
 324                         if (p) {
 325                                 niceval = 20 - task_nice(p);
 326                                 if (niceval > retval)
 327                                         retval = niceval;
 328                         }
 329                         break;
 330                 case PRIO_PGRP:
 331                         if (!who)
 332                                 who = process_group(current);
 333                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 334                                 niceval = 20 - task_nice(p);
 335                                 if (niceval > retval)
 336                                         retval = niceval;
 337                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 338                         break;
 339                 case PRIO_USER:
 340                         user = current->user;
 341                         if (!who)
 342                                 who = current->uid;
 343                         else
 344                                 if ((who != current->uid) && !(user = find_user(who)))
 345                                         goto out_unlock;        /* No processes for this user */
 346
 347                         do_each_thread(g, p)
 348                                 if (p->uid == who) {
 349                                         niceval = 20 - task_nice(p);
 350                                         if (niceval > retval)
 351                                                 retval = niceval;
 352                                 }
 353                         while_each_thread(g, p);
 354                         if (who != current->uid)
 355                                 free_uid(user);         /* for find_user() */
 356                         break;
 357         }
 358 out_unlock:
 359         read_unlock(&tasklist_lock);
 360
 361         return retval;
 362 }
 363
 364 void kernel_restart(char *cmd)
 365 {
 366         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 367         system_state = SYSTEM_RESTART;
 368         device_suspend(PMSG_FREEZE);
 369         device_shutdown();
 370         if (!cmd) {
 371                 printk(KERN_EMERG "Restarting system.\n");
 372         } else {
 373                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 374         }
 375         printk(".\n");
 376         machine_restart(cmd);
 377 }
 378 EXPORT_SYMBOL_GPL(kernel_restart);
 379
 380 void kernel_kexec(void)
 381 {
 382 #ifdef CONFIG_KEXEC
 383         struct kimage *image;
 384         image = xchg(&kexec_image, 0);
 385         if (!image) {
 386                 return;
 387         }
 388         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
 389         system_state = SYSTEM_RESTART;
 390         device_suspend(PMSG_FREEZE);
 391         device_shutdown();
 392         printk(KERN_EMERG "Starting new kernel\n");
 393         machine_shutdown();
 394         machine_kexec(image);
 395 #endif
 396 }
 397 EXPORT_SYMBOL_GPL(kernel_kexec);
 398
 399 void kernel_halt(void)
 400 {
 401         notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
 402         system_state = SYSTEM_HALT;
 403         device_suspend(PMSG_SUSPEND);
 404         device_shutdown();
 405         printk(KERN_EMERG "System halted.\n");
 406         machine_halt();
 407 }
 408 EXPORT_SYMBOL_GPL(kernel_halt);
 409
 410 void kernel_power_off(void)
 411 {
 412         notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
 413         system_state = SYSTEM_POWER_OFF;
 414         device_suspend(PMSG_SUSPEND);
 415         device_shutdown();
 416         printk(KERN_EMERG "Power down.\n");
 417         machine_power_off();
 418 }
 419 EXPORT_SYMBOL_GPL(kernel_power_off);
 420
 421 /*
 422  * Reboot system call: for obvious reasons only root may call it,
 423  * and even root needs to set up some magic numbers in the registers
 424  * so that some mistake won't make this reboot the whole machine.
 425  * You can also set the meaning of the ctrl-alt-del-key here.
 426  *
 427  * reboot doesn't sync: do that yourself before calling this.
 428  */
 429 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 430 {
 431         char buffer[256];
 432
 433         /* We only trust the superuser with rebooting the system. */
 434         if (!capable(CAP_SYS_BOOT))
 435                 return -EPERM;
 436
 437         /* For safety, we require "magic" arguments. */
 438         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 439             (magic2 != LINUX_REBOOT_MAGIC2 &&
 440                         magic2 != LINUX_REBOOT_MAGIC2A &&
 441                         magic2 != LINUX_REBOOT_MAGIC2B &&
 442                         magic2 != LINUX_REBOOT_MAGIC2C))
 443                 return -EINVAL;
 444
 445         lock_kernel();
 446         switch (cmd) {
 447         case LINUX_REBOOT_CMD_RESTART:
 448                 kernel_restart(NULL);
 449                 break;
 450
 451         case LINUX_REBOOT_CMD_CAD_ON:
 452                 C_A_D = 1;
 453                 break;
 454
 455         case LINUX_REBOOT_CMD_CAD_OFF:
 456                 C_A_D = 0;
 457                 break;
 458
 459         case LINUX_REBOOT_CMD_HALT:
 460                 kernel_halt();
 461                 unlock_kernel();
 462                 do_exit(0);
 463                 break;
 464
 465         case LINUX_REBOOT_CMD_POWER_OFF:
 466                 kernel_power_off();
 467                 unlock_kernel();
 468                 do_exit(0);
 469                 break;
 470
 471         case LINUX_REBOOT_CMD_RESTART2:
 472                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 473                         unlock_kernel();
 474                         return -EFAULT;
 475                 }
 476                 buffer[sizeof(buffer) - 1] = '\0';
 477
 478                 kernel_restart(buffer);
 479                 break;
 480
 481         case LINUX_REBOOT_CMD_KEXEC:
 482                 kernel_kexec();
 483                 unlock_kernel();
 484                 return -EINVAL;
 485
 486 #ifdef CONFIG_SOFTWARE_SUSPEND
 487         case LINUX_REBOOT_CMD_SW_SUSPEND:
 488                 {
 489                         int ret = software_suspend();
 490                         unlock_kernel();
 491                         return ret;
 492                 }
 493 #endif
 494
 495         default:
 496                 unlock_kernel();
 497                 return -EINVAL;
 498         }
 499         unlock_kernel();
 500         return 0;
 501 }
 502
 503 static void deferred_cad(void *dummy)
 504 {
 505         kernel_restart(NULL);
 506 }
 507
 508 /*
 509  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 510  * As it's called within an interrupt, it may NOT sync: the only choice
 511  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 512  */
 513 void ctrl_alt_del(void)
 514 {
 515         static DECLARE_WORK(cad_work, deferred_cad, NULL);
 516
 517         if (C_A_D)
 518                 schedule_work(&cad_work);
 519         else
 520                 kill_proc(cad_pid, SIGINT, 1);
 521 }
 522
 523
 524 /*
 525  * Unprivileged users may change the real gid to the effective gid
 526  * or vice versa.  (BSD-style)
 527  *
 528  * If you set the real gid at all, or set the effective gid to a value not
 529  * equal to the real gid, then the saved gid is set to the new effective gid.
 530  *
 531  * This makes it possible for a setgid program to completely drop its
 532  * privileges, which is often a useful assertion to make when you are doing
 533  * a security audit over a program.
 534  *
 535  * The general idea is that a program which uses just setregid() will be
 536  * 100% compatible with BSD.  A program which uses just setgid() will be
 537  * 100% compatible with POSIX with saved IDs.
 538  *
 539  * SMP: There are not races, the GIDs are checked only by filesystem
 540  *      operations (as far as semantic preservation is concerned).
 541  */
 542 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 543 {
 544         int old_rgid = current->gid;
 545         int old_egid = current->egid;
 546         int new_rgid = old_rgid;
 547         int new_egid = old_egid;
 548         int retval;
 549
 550         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 551         if (retval)
 552                 return retval;
 553
 554         if (rgid != (gid_t) -1) {
 555                 if ((old_rgid == rgid) ||
 556                     (current->egid==rgid) ||
 557                     capable(CAP_SETGID))
 558                         new_rgid = rgid;
 559                 else
 560                         return -EPERM;
 561         }
 562         if (egid != (gid_t) -1) {
 563                 if ((old_rgid == egid) ||
 564                     (current->egid == egid) ||
 565                     (current->sgid == egid) ||
 566                     capable(CAP_SETGID))
 567                         new_egid = egid;
 568                 else {
 569                         return -EPERM;
 570                 }
 571         }
 572         if (new_egid != old_egid)
 573         {
 574                 current->mm->dumpable = suid_dumpable;
 575                 smp_wmb();
 576         }
 577         if (rgid != (gid_t) -1 ||
 578             (egid != (gid_t) -1 && egid != old_rgid))
 579                 current->sgid = new_egid;
 580         current->fsgid = new_egid;
 581         current->egid = new_egid;
 582         current->gid = new_rgid;
 583         key_fsgid_changed(current);
 584         return 0;
 585 }
 586
 587 /*
 588  * setgid() is implemented like SysV w/ SAVED_IDS
 589  *
 590  * SMP: Same implicit races as above.
 591  */
 592 asmlinkage long sys_setgid(gid_t gid)
 593 {
 594         int old_egid = current->egid;
 595         int retval;
 596
 597         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 598         if (retval)
 599                 return retval;
 600
 601         if (capable(CAP_SETGID))
 602         {
 603                 if(old_egid != gid)
 604                 {
 605                         current->mm->dumpable = suid_dumpable;
 606                         smp_wmb();
 607                 }
 608                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 609         }
 610         else if ((gid == current->gid) || (gid == current->sgid))
 611         {
 612                 if(old_egid != gid)
 613                 {
 614                         current->mm->dumpable = suid_dumpable;
 615                         smp_wmb();
 616                 }
 617                 current->egid = current->fsgid = gid;
 618         }
 619         else
 620                 return -EPERM;
 621
 622         key_fsgid_changed(current);
 623         return 0;
 624 }
 625
 626 static int set_user(uid_t new_ruid, int dumpclear)
 627 {
 628         struct user_struct *new_user;
 629
 630         new_user = alloc_uid(new_ruid);
 631         if (!new_user)
 632                 return -EAGAIN;
 633
 634         if (atomic_read(&new_user->processes) >=
 635                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 636                         new_user != &root_user) {
 637                 free_uid(new_user);
 638                 return -EAGAIN;
 639         }
 640
 641         switch_uid(new_user);
 642
 643         if(dumpclear)
 644         {
 645                 current->mm->dumpable = suid_dumpable;
 646                 smp_wmb();
 647         }
 648         current->uid = new_ruid;
 649         return 0;
 650 }
 651
 652 /*
 653  * Unprivileged users may change the real uid to the effective uid
 654  * or vice versa.  (BSD-style)
 655  *
 656  * If you set the real uid at all, or set the effective uid to a value not
 657  * equal to the real uid, then the saved uid is set to the new effective uid.
 658  *
 659  * This makes it possible for a setuid program to completely drop its
 660  * privileges, which is often a useful assertion to make when you are doing
 661  * a security audit over a program.
 662  *
 663  * The general idea is that a program which uses just setreuid() will be
 664  * 100% compatible with BSD.  A program which uses just setuid() will be
 665  * 100% compatible with POSIX with saved IDs.
 666  */
 667 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 668 {
 669         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 670         int retval;
 671
 672         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 673         if (retval)
 674                 return retval;
 675
 676         new_ruid = old_ruid = current->uid;
 677         new_euid = old_euid = current->euid;
 678         old_suid = current->suid;
 679
 680         if (ruid != (uid_t) -1) {
 681                 new_ruid = ruid;
 682                 if ((old_ruid != ruid) &&
 683                     (current->euid != ruid) &&
 684                     !capable(CAP_SETUID))
 685                         return -EPERM;
 686         }
 687
 688         if (euid != (uid_t) -1) {
 689                 new_euid = euid;
 690                 if ((old_ruid != euid) &&
 691                     (current->euid != euid) &&
 692                     (current->suid != euid) &&
 693                     !capable(CAP_SETUID))
 694                         return -EPERM;
 695         }
 696
 697         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 698                 return -EAGAIN;
 699
 700         if (new_euid != old_euid)
 701         {
 702                 current->mm->dumpable = suid_dumpable;
 703                 smp_wmb();
 704         }
 705         current->fsuid = current->euid = new_euid;
 706         if (ruid != (uid_t) -1 ||
 707             (euid != (uid_t) -1 && euid != old_ruid))
 708                 current->suid = current->euid;
 709         current->fsuid = current->euid;
 710
 711         key_fsuid_changed(current);
 712
 713         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 714 }
 715
 716
 717
 718 /*
 719  * setuid() is implemented like SysV with SAVED_IDS
 720  *
 721  * Note that SAVED_ID's is deficient in that a setuid root program
 722  * like sendmail, for example, cannot set its uid to be a normal
 723  * user and then switch back, because if you're root, setuid() sets
 724  * the saved uid too.  If you don't like this, blame the bright people
 725  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 726  * will allow a root program to temporarily drop privileges and be able to
 727  * regain them by swapping the real and effective uid.
 728  */
 729 asmlinkage long sys_setuid(uid_t uid)
 730 {
 731         int old_euid = current->euid;
 732         int old_ruid, old_suid, new_ruid, new_suid;
 733         int retval;
 734
 735         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 736         if (retval)
 737                 return retval;
 738
 739         old_ruid = new_ruid = current->uid;
 740         old_suid = current->suid;
 741         new_suid = old_suid;
 742
 743         if (capable(CAP_SETUID)) {
 744                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 745                         return -EAGAIN;
 746                 new_suid = uid;
 747         } else if ((uid != current->uid) && (uid != new_suid))
 748                 return -EPERM;
 749
 750         if (old_euid != uid)
 751         {
 752                 current->mm->dumpable = suid_dumpable;
 753                 smp_wmb();
 754         }
 755         current->fsuid = current->euid = uid;
 756         current->suid = new_suid;
 757
 758         key_fsuid_changed(current);
 759
 760         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 761 }
 762
 763
 764 /*
 765  * This function implements a generic ability to update ruid, euid,
 766  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 767  */
 768 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 769 {
 770         int old_ruid = current->uid;
 771         int old_euid = current->euid;
 772         int old_suid = current->suid;
 773         int retval;
 774
 775         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 776         if (retval)
 777                 return retval;
 778
 779         if (!capable(CAP_SETUID)) {
 780                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 781                     (ruid != current->euid) && (ruid != current->suid))
 782                         return -EPERM;
 783                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 784                     (euid != current->euid) && (euid != current->suid))
 785                         return -EPERM;
 786                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 787                     (suid != current->euid) && (suid != current->suid))
 788                         return -EPERM;
 789         }
 790         if (ruid != (uid_t) -1) {
 791                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 792                         return -EAGAIN;
 793         }
 794         if (euid != (uid_t) -1) {
 795                 if (euid != current->euid)
 796                 {
 797                         current->mm->dumpable = suid_dumpable;
 798                         smp_wmb();
 799                 }
 800                 current->euid = euid;
 801         }
 802         current->fsuid = current->euid;
 803         if (suid != (uid_t) -1)
 804                 current->suid = suid;
 805
 806         key_fsuid_changed(current);
 807
 808         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 809 }
 810
 811 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 812 {
 813         int retval;
 814
 815         if (!(retval = put_user(current->uid, ruid)) &&
 816             !(retval = put_user(current->euid, euid)))
 817                 retval = put_user(current->suid, suid);
 818
 819         return retval;
 820 }
 821
 822 /*
 823  * Same as above, but for rgid, egid, sgid.
 824  */
 825 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 826 {
 827         int retval;
 828
 829         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 830         if (retval)
 831                 return retval;
 832
 833         if (!capable(CAP_SETGID)) {
 834                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 835                     (rgid != current->egid) && (rgid != current->sgid))
 836                         return -EPERM;
 837                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 838                     (egid != current->egid) && (egid != current->sgid))
 839                         return -EPERM;
 840                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 841                     (sgid != current->egid) && (sgid != current->sgid))
 842                         return -EPERM;
 843         }
 844         if (egid != (gid_t) -1) {
 845                 if (egid != current->egid)
 846                 {
 847                         current->mm->dumpable = suid_dumpable;
 848                         smp_wmb();
 849                 }
 850                 current->egid = egid;
 851         }
 852         current->fsgid = current->egid;
 853         if (rgid != (gid_t) -1)
 854                 current->gid = rgid;
 855         if (sgid != (gid_t) -1)
 856                 current->sgid = sgid;
 857
 858         key_fsgid_changed(current);
 859         return 0;
 860 }
 861
 862 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 863 {
 864         int retval;
 865
 866         if (!(retval = put_user(current->gid, rgid)) &&
 867             !(retval = put_user(current->egid, egid)))
 868                 retval = put_user(current->sgid, sgid);
 869
 870         return retval;
 871 }
 872
 873
 874 /*
 875  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 876  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 877  * whatever uid it wants to). It normally shadows "euid", except when
 878  * explicitly set by setfsuid() or for access..
 879  */
 880 asmlinkage long sys_setfsuid(uid_t uid)
 881 {
 882         int old_fsuid;
 883
 884         old_fsuid = current->fsuid;
 885         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 886                 return old_fsuid;
 887
 888         if (uid == current->uid || uid == current->euid ||
 889             uid == current->suid || uid == current->fsuid ||
 890             capable(CAP_SETUID))
 891         {
 892                 if (uid != old_fsuid)
 893                 {
 894                         current->mm->dumpable = suid_dumpable;
 895                         smp_wmb();
 896                 }
 897                 current->fsuid = uid;
 898         }
 899
 900         key_fsuid_changed(current);
 901
 902         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 903
 904         return old_fsuid;
 905 }
 906
 907 /*
 908  * Samma på svenska..
 909  */
 910 asmlinkage long sys_setfsgid(gid_t gid)
 911 {
 912         int old_fsgid;
 913
 914         old_fsgid = current->fsgid;
 915         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 916                 return old_fsgid;
 917
 918         if (gid == current->gid || gid == current->egid ||
 919             gid == current->sgid || gid == current->fsgid ||
 920             capable(CAP_SETGID))
 921         {
 922                 if (gid != old_fsgid)
 923                 {
 924                         current->mm->dumpable = suid_dumpable;
 925                         smp_wmb();
 926                 }
 927                 current->fsgid = gid;
 928                 key_fsgid_changed(current);
 929         }
 930         return old_fsgid;
 931 }
 932
 933 asmlinkage long sys_times(struct tms __user * tbuf)
 934 {
 935         /*
 936          *      In the SMP world we might just be unlucky and have one of
 937          *      the times increment as we use it. Since the value is an
 938          *      atomically safe type this is just fine. Conceptually its
 939          *      as if the syscall took an instant longer to occur.
 940          */
 941         if (tbuf) {
 942                 struct tms tmp;
 943                 cputime_t utime, stime, cutime, cstime;
 944
 945 #ifdef CONFIG_SMP
 946                 if (thread_group_empty(current)) {
 947                         /*
 948                          * Single thread case without the use of any locks.
 949                          *
 950                          * We may race with release_task if two threads are
 951                          * executing. However, release task first adds up the
 952                          * counters (__exit_signal) before  removing the task
 953                          * from the process tasklist (__unhash_process).
 954                          * __exit_signal also acquires and releases the
 955                          * siglock which results in the proper memory ordering
 956                          * so that the list modifications are always visible
 957                          * after the counters have been updated.
 958                          *
 959                          * If the counters have been updated by the second thread
 960                          * but the thread has not yet been removed from the list
 961                          * then the other branch will be executing which will
 962                          * block on tasklist_lock until the exit handling of the
 963                          * other task is finished.
 964                          *
 965                          * This also implies that the sighand->siglock cannot
 966                          * be held by another processor. So we can also
 967                          * skip acquiring that lock.
 968                          */
 969                         utime = cputime_add(current->signal->utime, current->utime);
 970                         stime = cputime_add(current->signal->utime, current->stime);
 971                         cutime = current->signal->cutime;
 972                         cstime = current->signal->cstime;
 973                 } else
 974 #endif
 975                 {
 976
 977                         /* Process with multiple threads */
 978                         struct task_struct *tsk = current;
 979                         struct task_struct *t;
 980
 981                         read_lock(&tasklist_lock);
 982                         utime = tsk->signal->utime;
 983                         stime = tsk->signal->stime;
 984                         t = tsk;
 985                         do {
 986                                 utime = cputime_add(utime, t->utime);
 987                                 stime = cputime_add(stime, t->stime);
 988                                 t = next_thread(t);
 989                         } while (t != tsk);
 990
 991                         /*
 992                          * While we have tasklist_lock read-locked, no dying thread
 993                          * can be updating current->signal->[us]time.  Instead,
 994                          * we got their counts included in the live thread loop.
 995                          * However, another thread can come in right now and
 996                          * do a wait call that updates current->signal->c[us]time.
 997                          * To make sure we always see that pair updated atomically,
 998                          * we take the siglock around fetching them.
 999                          */
1000                         spin_lock_irq(&tsk->sighand->siglock);
1001                         cutime = tsk->signal->cutime;
1002                         cstime = tsk->signal->cstime;
1003                         spin_unlock_irq(&tsk->sighand->siglock);
1004                         read_unlock(&tasklist_lock);
1005                 }
1006                 tmp.tms_utime = cputime_to_clock_t(utime);
1007                 tmp.tms_stime = cputime_to_clock_t(stime);
1008                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1009                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1010                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1011                         return -EFAULT;
1012         }
1013         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1014 }
1015
1016 /*
1017  * This needs some heavy checking ...
1018  * I just haven't the stomach for it. I also don't fully
1019  * understand sessions/pgrp etc. Let somebody who does explain it.
1020  *
1021  * OK, I think I have the protection semantics right.... this is really
1022  * only important on a multi-user system anyway, to make sure one user
1023  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1024  *
1025  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1026  * LBT 04.03.94
1027  */
1028
1029 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1030 {
1031         struct task_struct *p;
1032         int err = -EINVAL;
1033
1034         if (!pid)
1035                 pid = current->pid;
1036         if (!pgid)
1037                 pgid = pid;
1038         if (pgid < 0)
1039                 return -EINVAL;
1040
1041         /* From this point forward we keep holding onto the tasklist lock
1042          * so that our parent does not change from under us. -DaveM
1043          */
1044         write_lock_irq(&tasklist_lock);
1045
1046         err = -ESRCH;
1047         p = find_task_by_pid(pid);
1048         if (!p)
1049                 goto out;
1050
1051         err = -EINVAL;
1052         if (!thread_group_leader(p))
1053                 goto out;
1054
1055         if (p->parent == current || p->real_parent == current) {
1056                 err = -EPERM;
1057                 if (p->signal->session != current->signal->session)
1058                         goto out;
1059                 err = -EACCES;
1060                 if (p->did_exec)
1061                         goto out;
1062         } else {
1063                 err = -ESRCH;
1064                 if (p != current)
1065                         goto out;
1066         }
1067
1068         err = -EPERM;
1069         if (p->signal->leader)
1070                 goto out;
1071
1072         if (pgid != pid) {
1073                 struct task_struct *p;
1074
1075                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1076                         if (p->signal->session == current->signal->session)
1077                                 goto ok_pgid;
1078                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1079                 goto out;
1080         }
1081
1082 ok_pgid:
1083         err = security_task_setpgid(p, pgid);
1084         if (err)
1085                 goto out;
1086
1087         if (process_group(p) != pgid) {
1088                 detach_pid(p, PIDTYPE_PGID);
1089                 p->signal->pgrp = pgid;
1090                 attach_pid(p, PIDTYPE_PGID, pgid);
1091         }
1092
1093         err = 0;
1094 out:
1095         /* All paths lead to here, thus we are safe. -DaveM */
1096         write_unlock_irq(&tasklist_lock);
1097         return err;
1098 }
1099
1100 asmlinkage long sys_getpgid(pid_t pid)
1101 {
1102         if (!pid) {
1103                 return process_group(current);
1104         } else {
1105                 int retval;
1106                 struct task_struct *p;
1107
1108                 read_lock(&tasklist_lock);
1109                 p = find_task_by_pid(pid);
1110
1111                 retval = -ESRCH;
1112                 if (p) {
1113                         retval = security_task_getpgid(p);
1114                         if (!retval)
1115                                 retval = process_group(p);
1116                 }
1117                 read_unlock(&tasklist_lock);
1118                 return retval;
1119         }
1120 }
1121
1122 #ifdef __ARCH_WANT_SYS_GETPGRP
1123
1124 asmlinkage long sys_getpgrp(void)
1125 {
1126         /* SMP - assuming writes are word atomic this is fine */
1127         return process_group(current);
1128 }
1129
1130 #endif
1131
1132 asmlinkage long sys_getsid(pid_t pid)
1133 {
1134         if (!pid) {
1135                 return current->signal->session;
1136         } else {
1137                 int retval;
1138                 struct task_struct *p;
1139
1140                 read_lock(&tasklist_lock);
1141                 p = find_task_by_pid(pid);
1142
1143                 retval = -ESRCH;
1144                 if(p) {
1145                         retval = security_task_getsid(p);
1146                         if (!retval)
1147                                 retval = p->signal->session;
1148                 }
1149                 read_unlock(&tasklist_lock);
1150                 return retval;
1151         }
1152 }
1153
1154 asmlinkage long sys_setsid(void)
1155 {
1156         struct pid *pid;
1157         int err = -EPERM;
1158
1159         if (!thread_group_leader(current))
1160                 return -EINVAL;
1161
1162         down(&tty_sem);
1163         write_lock_irq(&tasklist_lock);
1164
1165         pid = find_pid(PIDTYPE_PGID, current->pid);
1166         if (pid)
1167                 goto out;
1168
1169         current->signal->leader = 1;
1170         __set_special_pids(current->pid, current->pid);
1171         current->signal->tty = NULL;
1172         current->signal->tty_old_pgrp = 0;
1173         err = process_group(current);
1174 out:
1175         write_unlock_irq(&tasklist_lock);
1176         up(&tty_sem);
1177         return err;
1178 }
1179
1180 /*
1181  * Supplementary group IDs
1182  */
1183
1184 /* init to 2 - one for init_task, one to ensure it is never freed */
1185 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1186
1187 struct group_info *groups_alloc(int gidsetsize)
1188 {
1189         struct group_info *group_info;
1190         int nblocks;
1191         int i;
1192
1193         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1194         /* Make sure we always allocate at least one indirect block pointer */
1195         nblocks = nblocks ? : 1;
1196         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1197         if (!group_info)
1198                 return NULL;
1199         group_info->ngroups = gidsetsize;
1200         group_info->nblocks = nblocks;
1201         atomic_set(&group_info->usage, 1);
1202
1203         if (gidsetsize <= NGROUPS_SMALL) {
1204                 group_info->blocks[0] = group_info->small_block;
1205         } else {
1206                 for (i = 0; i < nblocks; i++) {
1207                         gid_t *b;
1208                         b = (void *)__get_free_page(GFP_USER);
1209                         if (!b)
1210                                 goto out_undo_partial_alloc;
1211                         group_info->blocks[i] = b;
1212                 }
1213         }
1214         return group_info;
1215
1216 out_undo_partial_alloc:
1217         while (--i >= 0) {
1218                 free_page((unsigned long)group_info->blocks[i]);
1219         }
1220         kfree(group_info);
1221         return NULL;
1222 }
1223
1224 EXPORT_SYMBOL(groups_alloc);
1225
1226 void groups_free(struct group_info *group_info)
1227 {
1228         if (group_info->blocks[0] != group_info->small_block) {
1229                 int i;
1230                 for (i = 0; i < group_info->nblocks; i++)
1231                         free_page((unsigned long)group_info->blocks[i]);
1232         }
1233         kfree(group_info);
1234 }
1235
1236 EXPORT_SYMBOL(groups_free);
1237
1238 /* export the group_info to a user-space array */
1239 static int groups_to_user(gid_t __user *grouplist,
1240     struct group_info *group_info)
1241 {
1242         int i;
1243         int count = group_info->ngroups;
1244
1245         for (i = 0; i < group_info->nblocks; i++) {
1246                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1247                 int off = i * NGROUPS_PER_BLOCK;
1248                 int len = cp_count * sizeof(*grouplist);
1249
1250                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1251                         return -EFAULT;
1252
1253                 count -= cp_count;
1254         }
1255         return 0;
1256 }
1257
1258 /* fill a group_info from a user-space array - it must be allocated already */
1259 static int groups_from_user(struct group_info *group_info,
1260     gid_t __user *grouplist)
1261  {
1262         int i;
1263         int count = group_info->ngroups;
1264
1265         for (i = 0; i < group_info->nblocks; i++) {
1266                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1267                 int off = i * NGROUPS_PER_BLOCK;
1268                 int len = cp_count * sizeof(*grouplist);
1269
1270                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1271                         return -EFAULT;
1272
1273                 count -= cp_count;
1274         }
1275         return 0;
1276 }
1277
1278 /* a simple Shell sort */
1279 static void groups_sort(struct group_info *group_info)
1280 {
1281         int base, max, stride;
1282         int gidsetsize = group_info->ngroups;
1283
1284         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1285                 ; /* nothing */
1286         stride /= 3;
1287
1288         while (stride) {
1289                 max = gidsetsize - stride;
1290                 for (base = 0; base < max; base++) {
1291                         int left = base;
1292                         int right = left + stride;
1293                         gid_t tmp = GROUP_AT(group_info, right);
1294
1295                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1296                                 GROUP_AT(group_info, right) =
1297                                     GROUP_AT(group_info, left);
1298                                 right = left;
1299                                 left -= stride;
1300                         }
1301                         GROUP_AT(group_info, right) = tmp;
1302                 }
1303                 stride /= 3;
1304         }
1305 }
1306
1307 /* a simple bsearch */
1308 int groups_search(struct group_info *group_info, gid_t grp)
1309 {
1310         int left, right;
1311
1312         if (!group_info)
1313                 return 0;
1314
1315         left = 0;
1316         right = group_info->ngroups;
1317         while (left < right) {
1318                 int mid = (left+right)/2;
1319                 int cmp = grp - GROUP_AT(group_info, mid);
1320                 if (cmp > 0)
1321                         left = mid + 1;
1322                 else if (cmp < 0)
1323                         right = mid;
1324                 else
1325                         return 1;
1326         }
1327         return 0;
1328 }
1329
1330 /* validate and set current->group_info */
1331 int set_current_groups(struct group_info *group_info)
1332 {
1333         int retval;
1334         struct group_info *old_info;
1335
1336         retval = security_task_setgroups(group_info);
1337         if (retval)
1338                 return retval;
1339
1340         groups_sort(group_info);
1341         get_group_info(group_info);
1342
1343         task_lock(current);
1344         old_info = current->group_info;
1345         current->group_info = group_info;
1346         task_unlock(current);
1347
1348         put_group_info(old_info);
1349
1350         return 0;
1351 }
1352
1353 EXPORT_SYMBOL(set_current_groups);
1354
1355 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1356 {
1357         int i = 0;
1358
1359         /*
1360          *      SMP: Nobody else can change our grouplist. Thus we are
1361          *      safe.
1362          */
1363
1364         if (gidsetsize < 0)
1365                 return -EINVAL;
1366
1367         /* no need to grab task_lock here; it cannot change */
1368         get_group_info(current->group_info);
1369         i = current->group_info->ngroups;
1370         if (gidsetsize) {
1371                 if (i > gidsetsize) {
1372                         i = -EINVAL;
1373                         goto out;
1374                 }
1375                 if (groups_to_user(grouplist, current->group_info)) {
1376                         i = -EFAULT;
1377                         goto out;
1378                 }
1379         }
1380 out:
1381         put_group_info(current->group_info);
1382         return i;
1383 }
1384
1385 /*
1386  *      SMP: Our groups are copy-on-write. We can set them safely
1387  *      without another task interfering.
1388  */
1389
1390 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1391 {
1392         struct group_info *group_info;
1393         int retval;
1394
1395         if (!capable(CAP_SETGID))
1396                 return -EPERM;
1397         if ((unsigned)gidsetsize > NGROUPS_MAX)
1398                 return -EINVAL;
1399
1400         group_info = groups_alloc(gidsetsize);
1401         if (!group_info)
1402                 return -ENOMEM;
1403         retval = groups_from_user(group_info, grouplist);
1404         if (retval) {
1405                 put_group_info(group_info);
1406                 return retval;
1407         }
1408
1409         retval = set_current_groups(group_info);
1410         put_group_info(group_info);
1411
1412         return retval;
1413 }
1414
1415 /*
1416  * Check whether we're fsgid/egid or in the supplemental group..
1417  */
1418 int in_group_p(gid_t grp)
1419 {
1420         int retval = 1;
1421         if (grp != current->fsgid) {
1422                 get_group_info(current->group_info);
1423                 retval = groups_search(current->group_info, grp);
1424                 put_group_info(current->group_info);
1425         }
1426         return retval;
1427 }
1428
1429 EXPORT_SYMBOL(in_group_p);
1430
1431 int in_egroup_p(gid_t grp)
1432 {
1433         int retval = 1;
1434         if (grp != current->egid) {
1435                 get_group_info(current->group_info);
1436                 retval = groups_search(current->group_info, grp);
1437                 put_group_info(current->group_info);
1438         }
1439         return retval;
1440 }
1441
1442 EXPORT_SYMBOL(in_egroup_p);
1443
1444 DECLARE_RWSEM(uts_sem);
1445
1446 EXPORT_SYMBOL(uts_sem);
1447
1448 asmlinkage long sys_newuname(struct new_utsname __user * name)
1449 {
1450         int errno = 0;
1451
1452         down_read(&uts_sem);
1453         if (copy_to_user(name,&system_utsname,sizeof *name))
1454                 errno = -EFAULT;
1455         up_read(&uts_sem);
1456         return errno;
1457 }
1458
1459 asmlinkage long sys_sethostname(char __user *name, int len)
1460 {
1461         int errno;
1462         char tmp[__NEW_UTS_LEN];
1463
1464         if (!capable(CAP_SYS_ADMIN))
1465                 return -EPERM;
1466         if (len < 0 || len > __NEW_UTS_LEN)
1467                 return -EINVAL;
1468         down_write(&uts_sem);
1469         errno = -EFAULT;
1470         if (!copy_from_user(tmp, name, len)) {
1471                 memcpy(system_utsname.nodename, tmp, len);
1472                 system_utsname.nodename[len] = 0;
1473                 errno = 0;
1474         }
1475         up_write(&uts_sem);
1476         return errno;
1477 }
1478
1479 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1480
1481 asmlinkage long sys_gethostname(char __user *name, int len)
1482 {
1483         int i, errno;
1484
1485         if (len < 0)
1486                 return -EINVAL;
1487         down_read(&uts_sem);
1488         i = 1 + strlen(system_utsname.nodename);
1489         if (i > len)
1490                 i = len;
1491         errno = 0;
1492         if (copy_to_user(name, system_utsname.nodename, i))
1493                 errno = -EFAULT;
1494         up_read(&uts_sem);
1495         return errno;
1496 }
1497
1498 #endif
1499
1500 /*
1501  * Only setdomainname; getdomainname can be implemented by calling
1502  * uname()
1503  */
1504 asmlinkage long sys_setdomainname(char __user *name, int len)
1505 {
1506         int errno;
1507         char tmp[__NEW_UTS_LEN];
1508
1509         if (!capable(CAP_SYS_ADMIN))
1510                 return -EPERM;
1511         if (len < 0 || len > __NEW_UTS_LEN)
1512                 return -EINVAL;
1513
1514         down_write(&uts_sem);
1515         errno = -EFAULT;
1516         if (!copy_from_user(tmp, name, len)) {
1517                 memcpy(system_utsname.domainname, tmp, len);
1518                 system_utsname.domainname[len] = 0;
1519                 errno = 0;
1520         }
1521         up_write(&uts_sem);
1522         return errno;
1523 }
1524
1525 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1526 {
1527         if (resource >= RLIM_NLIMITS)
1528                 return -EINVAL;
1529         else {
1530                 struct rlimit value;
1531                 task_lock(current->group_leader);
1532                 value = current->signal->rlim[resource];
1533                 task_unlock(current->group_leader);
1534                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1535         }
1536 }
1537
1538 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1539
1540 /*
1541  *      Back compatibility for getrlimit. Needed for some apps.
1542  */
1543
1544 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1545 {
1546         struct rlimit x;
1547         if (resource >= RLIM_NLIMITS)
1548                 return -EINVAL;
1549
1550         task_lock(current->group_leader);
1551         x = current->signal->rlim[resource];
1552         task_unlock(current->group_leader);
1553         if(x.rlim_cur > 0x7FFFFFFF)
1554                 x.rlim_cur = 0x7FFFFFFF;
1555         if(x.rlim_max > 0x7FFFFFFF)
1556                 x.rlim_max = 0x7FFFFFFF;
1557         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1558 }
1559
1560 #endif
1561
1562 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1563 {
1564         struct rlimit new_rlim, *old_rlim;
1565         int retval;
1566
1567         if (resource >= RLIM_NLIMITS)
1568                 return -EINVAL;
1569         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1570                 return -EFAULT;
1571        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1572                return -EINVAL;
1573         old_rlim = current->signal->rlim + resource;
1574         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1575             !capable(CAP_SYS_RESOURCE))
1576                 return -EPERM;
1577         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1578                         return -EPERM;
1579
1580         retval = security_task_setrlimit(resource, &new_rlim);
1581         if (retval)
1582                 return retval;
1583
1584         task_lock(current->group_leader);
1585         *old_rlim = new_rlim;
1586         task_unlock(current->group_leader);
1587
1588         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1589             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1590              new_rlim.rlim_cur <= cputime_to_secs(
1591                      current->signal->it_prof_expires))) {
1592                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1593                 read_lock(&tasklist_lock);
1594                 spin_lock_irq(&current->sighand->siglock);
1595                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1596                                       &cputime, NULL);
1597                 spin_unlock_irq(&current->sighand->siglock);
1598                 read_unlock(&tasklist_lock);
1599         }
1600
1601         return 0;
1602 }
1603
1604 /*
1605  * It would make sense to put struct rusage in the task_struct,
1606  * except that would make the task_struct be *really big*.  After
1607  * task_struct gets moved into malloc'ed memory, it would
1608  * make sense to do this.  It will make moving the rest of the information
1609  * a lot simpler!  (Which we're not doing right now because we're not
1610  * measuring them yet).
1611  *
1612  * This expects to be called with tasklist_lock read-locked or better,
1613  * and the siglock not locked.  It may momentarily take the siglock.
1614  *
1615  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1616  * races with threads incrementing their own counters.  But since word
1617  * reads are atomic, we either get new values or old values and we don't
1618  * care which for the sums.  We always take the siglock to protect reading
1619  * the c* fields from p->signal from races with exit.c updating those
1620  * fields when reaping, so a sample either gets all the additions of a
1621  * given child after it's reaped, or none so this sample is before reaping.
1622  */
1623
1624 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1625 {
1626         struct task_struct *t;
1627         unsigned long flags;
1628         cputime_t utime, stime;
1629
1630         memset((char *) r, 0, sizeof *r);
1631
1632         if (unlikely(!p->signal))
1633                 return;
1634
1635         switch (who) {
1636                 case RUSAGE_CHILDREN:
1637                         spin_lock_irqsave(&p->sighand->siglock, flags);
1638                         utime = p->signal->cutime;
1639                         stime = p->signal->cstime;
1640                         r->ru_nvcsw = p->signal->cnvcsw;
1641                         r->ru_nivcsw = p->signal->cnivcsw;
1642                         r->ru_minflt = p->signal->cmin_flt;
1643                         r->ru_majflt = p->signal->cmaj_flt;
1644                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1645                         cputime_to_timeval(utime, &r->ru_utime);
1646                         cputime_to_timeval(stime, &r->ru_stime);
1647                         break;
1648                 case RUSAGE_SELF:
1649                         spin_lock_irqsave(&p->sighand->siglock, flags);
1650                         utime = stime = cputime_zero;
1651                         goto sum_group;
1652                 case RUSAGE_BOTH:
1653                         spin_lock_irqsave(&p->sighand->siglock, flags);
1654                         utime = p->signal->cutime;
1655                         stime = p->signal->cstime;
1656                         r->ru_nvcsw = p->signal->cnvcsw;
1657                         r->ru_nivcsw = p->signal->cnivcsw;
1658                         r->ru_minflt = p->signal->cmin_flt;
1659                         r->ru_majflt = p->signal->cmaj_flt;
1660                 sum_group:
1661                         utime = cputime_add(utime, p->signal->utime);
1662                         stime = cputime_add(stime, p->signal->stime);
1663                         r->ru_nvcsw += p->signal->nvcsw;
1664                         r->ru_nivcsw += p->signal->nivcsw;
1665                         r->ru_minflt += p->signal->min_flt;
1666                         r->ru_majflt += p->signal->maj_flt;
1667                         t = p;
1668                         do {
1669                                 utime = cputime_add(utime, t->utime);
1670                                 stime = cputime_add(stime, t->stime);
1671                                 r->ru_nvcsw += t->nvcsw;
1672                                 r->ru_nivcsw += t->nivcsw;
1673                                 r->ru_minflt += t->min_flt;
1674                                 r->ru_majflt += t->maj_flt;
1675                                 t = next_thread(t);
1676                         } while (t != p);
1677                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1678                         cputime_to_timeval(utime, &r->ru_utime);
1679                         cputime_to_timeval(stime, &r->ru_stime);
1680                         break;
1681                 default:
1682                         BUG();
1683         }
1684 }
1685
1686 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1687 {
1688         struct rusage r;
1689         read_lock(&tasklist_lock);
1690         k_getrusage(p, who, &r);
1691         read_unlock(&tasklist_lock);
1692         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1693 }
1694
1695 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1696 {
1697         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1698                 return -EINVAL;
1699         return getrusage(current, who, ru);
1700 }
1701
1702 asmlinkage long sys_umask(int mask)
1703 {
1704         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1705         return mask;
1706 }
1707
1708 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1709                           unsigned long arg4, unsigned long arg5)
1710 {
1711         long error;
1712         int sig;
1713
1714         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1715         if (error)
1716                 return error;
1717
1718         switch (option) {
1719                 case PR_SET_PDEATHSIG:
1720                         sig = arg2;
1721                         if (!valid_signal(sig)) {
1722                                 error = -EINVAL;
1723                                 break;
1724                         }
1725                         current->pdeath_signal = sig;
1726                         break;
1727                 case PR_GET_PDEATHSIG:
1728                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1729                         break;
1730                 case PR_GET_DUMPABLE:
1731                         if (current->mm->dumpable)
1732                                 error = 1;
1733                         break;
1734                 case PR_SET_DUMPABLE:
1735                         if (arg2 < 0 || arg2 > 2) {
1736                                 error = -EINVAL;
1737                                 break;
1738                         }
1739                         current->mm->dumpable = arg2;
1740                         break;
1741
1742                 case PR_SET_UNALIGN:
1743                         error = SET_UNALIGN_CTL(current, arg2);
1744                         break;
1745                 case PR_GET_UNALIGN:
1746                         error = GET_UNALIGN_CTL(current, arg2);
1747                         break;
1748                 case PR_SET_FPEMU:
1749                         error = SET_FPEMU_CTL(current, arg2);
1750                         break;
1751                 case PR_GET_FPEMU:
1752                         error = GET_FPEMU_CTL(current, arg2);
1753                         break;
1754                 case PR_SET_FPEXC:
1755                         error = SET_FPEXC_CTL(current, arg2);
1756                         break;
1757                 case PR_GET_FPEXC:
1758                         error = GET_FPEXC_CTL(current, arg2);
1759                         break;
1760                 case PR_GET_TIMING:
1761                         error = PR_TIMING_STATISTICAL;
1762                         break;
1763                 case PR_SET_TIMING:
1764                         if (arg2 == PR_TIMING_STATISTICAL)
1765                                 error = 0;
1766                         else
1767                                 error = -EINVAL;
1768                         break;
1769
1770                 case PR_GET_KEEPCAPS:
1771                         if (current->keep_capabilities)
1772                                 error = 1;
1773                         break;
1774                 case PR_SET_KEEPCAPS:
1775                         if (arg2 != 0 && arg2 != 1) {
1776                                 error = -EINVAL;
1777                                 break;
1778                         }
1779                         current->keep_capabilities = arg2;
1780                         break;
1781                 case PR_SET_NAME: {
1782                         struct task_struct *me = current;
1783                         unsigned char ncomm[sizeof(me->comm)];
1784
1785                         ncomm[sizeof(me->comm)-1] = 0;
1786                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1787                                                 sizeof(me->comm)-1) < 0)
1788                                 return -EFAULT;
1789                         set_task_comm(me, ncomm);
1790                         return 0;
1791                 }
1792                 case PR_GET_NAME: {
1793                         struct task_struct *me = current;
1794                         unsigned char tcomm[sizeof(me->comm)];
1795
1796                         get_task_comm(tcomm, me);
1797                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1798                                 return -EFAULT;
1799                         return 0;
1800                 }
1801                 default:
1802                         error = -EINVAL;
1803                         break;
1804         }
1805         return error;
1806 }