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 emergency_restart(void)
 365 {
 366         machine_emergency_restart();
 367 }
 368 EXPORT_SYMBOL_GPL(emergency_restart);
 369
 370 void kernel_restart(char *cmd)
 371 {
 372         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 373         system_state = SYSTEM_RESTART;
 374         device_shutdown();
 375         if (!cmd) {
 376                 printk(KERN_EMERG "Restarting system.\n");
 377         } else {
 378                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 379         }
 380         printk(".\n");
 381         machine_restart(cmd);
 382 }
 383 EXPORT_SYMBOL_GPL(kernel_restart);
 384
 385 void kernel_kexec(void)
 386 {
 387 #ifdef CONFIG_KEXEC
 388         struct kimage *image;
 389         image = xchg(&kexec_image, 0);
 390         if (!image) {
 391                 return;
 392         }
 393         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
 394         system_state = SYSTEM_RESTART;
 395         device_shutdown();
 396         printk(KERN_EMERG "Starting new kernel\n");
 397         machine_shutdown();
 398         machine_kexec(image);
 399 #endif
 400 }
 401 EXPORT_SYMBOL_GPL(kernel_kexec);
 402
 403 void kernel_halt(void)
 404 {
 405         notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
 406         system_state = SYSTEM_HALT;
 407         device_suspend(PMSG_SUSPEND);
 408         device_shutdown();
 409         printk(KERN_EMERG "System halted.\n");
 410         machine_halt();
 411 }
 412 EXPORT_SYMBOL_GPL(kernel_halt);
 413
 414 void kernel_power_off(void)
 415 {
 416         notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
 417         system_state = SYSTEM_POWER_OFF;
 418         device_suspend(PMSG_SUSPEND);
 419         device_shutdown();
 420         printk(KERN_EMERG "Power down.\n");
 421         machine_power_off();
 422 }
 423 EXPORT_SYMBOL_GPL(kernel_power_off);
 424
 425 /*
 426  * Reboot system call: for obvious reasons only root may call it,
 427  * and even root needs to set up some magic numbers in the registers
 428  * so that some mistake won't make this reboot the whole machine.
 429  * You can also set the meaning of the ctrl-alt-del-key here.
 430  *
 431  * reboot doesn't sync: do that yourself before calling this.
 432  */
 433 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 434 {
 435         char buffer[256];
 436
 437         /* We only trust the superuser with rebooting the system. */
 438         if (!capable(CAP_SYS_BOOT))
 439                 return -EPERM;
 440
 441         /* For safety, we require "magic" arguments. */
 442         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 443             (magic2 != LINUX_REBOOT_MAGIC2 &&
 444                         magic2 != LINUX_REBOOT_MAGIC2A &&
 445                         magic2 != LINUX_REBOOT_MAGIC2B &&
 446                         magic2 != LINUX_REBOOT_MAGIC2C))
 447                 return -EINVAL;
 448
 449         lock_kernel();
 450         switch (cmd) {
 451         case LINUX_REBOOT_CMD_RESTART:
 452                 kernel_restart(NULL);
 453                 break;
 454
 455         case LINUX_REBOOT_CMD_CAD_ON:
 456                 C_A_D = 1;
 457                 break;
 458
 459         case LINUX_REBOOT_CMD_CAD_OFF:
 460                 C_A_D = 0;
 461                 break;
 462
 463         case LINUX_REBOOT_CMD_HALT:
 464                 kernel_halt();
 465                 unlock_kernel();
 466                 do_exit(0);
 467                 break;
 468
 469         case LINUX_REBOOT_CMD_POWER_OFF:
 470                 kernel_power_off();
 471                 unlock_kernel();
 472                 do_exit(0);
 473                 break;
 474
 475         case LINUX_REBOOT_CMD_RESTART2:
 476                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 477                         unlock_kernel();
 478                         return -EFAULT;
 479                 }
 480                 buffer[sizeof(buffer) - 1] = '\0';
 481
 482                 kernel_restart(buffer);
 483                 break;
 484
 485         case LINUX_REBOOT_CMD_KEXEC:
 486                 kernel_kexec();
 487                 unlock_kernel();
 488                 return -EINVAL;
 489
 490 #ifdef CONFIG_SOFTWARE_SUSPEND
 491         case LINUX_REBOOT_CMD_SW_SUSPEND:
 492                 {
 493                         int ret = software_suspend();
 494                         unlock_kernel();
 495                         return ret;
 496                 }
 497 #endif
 498
 499         default:
 500                 unlock_kernel();
 501                 return -EINVAL;
 502         }
 503         unlock_kernel();
 504         return 0;
 505 }
 506
 507 static void deferred_cad(void *dummy)
 508 {
 509         kernel_restart(NULL);
 510 }
 511
 512 /*
 513  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 514  * As it's called within an interrupt, it may NOT sync: the only choice
 515  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 516  */
 517 void ctrl_alt_del(void)
 518 {
 519         static DECLARE_WORK(cad_work, deferred_cad, NULL);
 520
 521         if (C_A_D)
 522                 schedule_work(&cad_work);
 523         else
 524                 kill_proc(cad_pid, SIGINT, 1);
 525 }
 526
 527
 528 /*
 529  * Unprivileged users may change the real gid to the effective gid
 530  * or vice versa.  (BSD-style)
 531  *
 532  * If you set the real gid at all, or set the effective gid to a value not
 533  * equal to the real gid, then the saved gid is set to the new effective gid.
 534  *
 535  * This makes it possible for a setgid program to completely drop its
 536  * privileges, which is often a useful assertion to make when you are doing
 537  * a security audit over a program.
 538  *
 539  * The general idea is that a program which uses just setregid() will be
 540  * 100% compatible with BSD.  A program which uses just setgid() will be
 541  * 100% compatible with POSIX with saved IDs.
 542  *
 543  * SMP: There are not races, the GIDs are checked only by filesystem
 544  *      operations (as far as semantic preservation is concerned).
 545  */
 546 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 547 {
 548         int old_rgid = current->gid;
 549         int old_egid = current->egid;
 550         int new_rgid = old_rgid;
 551         int new_egid = old_egid;
 552         int retval;
 553
 554         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 555         if (retval)
 556                 return retval;
 557
 558         if (rgid != (gid_t) -1) {
 559                 if ((old_rgid == rgid) ||
 560                     (current->egid==rgid) ||
 561                     capable(CAP_SETGID))
 562                         new_rgid = rgid;
 563                 else
 564                         return -EPERM;
 565         }
 566         if (egid != (gid_t) -1) {
 567                 if ((old_rgid == egid) ||
 568                     (current->egid == egid) ||
 569                     (current->sgid == egid) ||
 570                     capable(CAP_SETGID))
 571                         new_egid = egid;
 572                 else {
 573                         return -EPERM;
 574                 }
 575         }
 576         if (new_egid != old_egid)
 577         {
 578                 current->mm->dumpable = suid_dumpable;
 579                 smp_wmb();
 580         }
 581         if (rgid != (gid_t) -1 ||
 582             (egid != (gid_t) -1 && egid != old_rgid))
 583                 current->sgid = new_egid;
 584         current->fsgid = new_egid;
 585         current->egid = new_egid;
 586         current->gid = new_rgid;
 587         key_fsgid_changed(current);
 588         return 0;
 589 }
 590
 591 /*
 592  * setgid() is implemented like SysV w/ SAVED_IDS
 593  *
 594  * SMP: Same implicit races as above.
 595  */
 596 asmlinkage long sys_setgid(gid_t gid)
 597 {
 598         int old_egid = current->egid;
 599         int retval;
 600
 601         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 602         if (retval)
 603                 return retval;
 604
 605         if (capable(CAP_SETGID))
 606         {
 607                 if(old_egid != gid)
 608                 {
 609                         current->mm->dumpable = suid_dumpable;
 610                         smp_wmb();
 611                 }
 612                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 613         }
 614         else if ((gid == current->gid) || (gid == current->sgid))
 615         {
 616                 if(old_egid != gid)
 617                 {
 618                         current->mm->dumpable = suid_dumpable;
 619                         smp_wmb();
 620                 }
 621                 current->egid = current->fsgid = gid;
 622         }
 623         else
 624                 return -EPERM;
 625
 626         key_fsgid_changed(current);
 627         return 0;
 628 }
 629
 630 static int set_user(uid_t new_ruid, int dumpclear)
 631 {
 632         struct user_struct *new_user;
 633
 634         new_user = alloc_uid(new_ruid);
 635         if (!new_user)
 636                 return -EAGAIN;
 637
 638         if (atomic_read(&new_user->processes) >=
 639                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 640                         new_user != &root_user) {
 641                 free_uid(new_user);
 642                 return -EAGAIN;
 643         }
 644
 645         switch_uid(new_user);
 646
 647         if(dumpclear)
 648         {
 649                 current->mm->dumpable = suid_dumpable;
 650                 smp_wmb();
 651         }
 652         current->uid = new_ruid;
 653         return 0;
 654 }
 655
 656 /*
 657  * Unprivileged users may change the real uid to the effective uid
 658  * or vice versa.  (BSD-style)
 659  *
 660  * If you set the real uid at all, or set the effective uid to a value not
 661  * equal to the real uid, then the saved uid is set to the new effective uid.
 662  *
 663  * This makes it possible for a setuid program to completely drop its
 664  * privileges, which is often a useful assertion to make when you are doing
 665  * a security audit over a program.
 666  *
 667  * The general idea is that a program which uses just setreuid() will be
 668  * 100% compatible with BSD.  A program which uses just setuid() will be
 669  * 100% compatible with POSIX with saved IDs.
 670  */
 671 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 672 {
 673         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 674         int retval;
 675
 676         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 677         if (retval)
 678                 return retval;
 679
 680         new_ruid = old_ruid = current->uid;
 681         new_euid = old_euid = current->euid;
 682         old_suid = current->suid;
 683
 684         if (ruid != (uid_t) -1) {
 685                 new_ruid = ruid;
 686                 if ((old_ruid != ruid) &&
 687                     (current->euid != ruid) &&
 688                     !capable(CAP_SETUID))
 689                         return -EPERM;
 690         }
 691
 692         if (euid != (uid_t) -1) {
 693                 new_euid = euid;
 694                 if ((old_ruid != euid) &&
 695                     (current->euid != euid) &&
 696                     (current->suid != euid) &&
 697                     !capable(CAP_SETUID))
 698                         return -EPERM;
 699         }
 700
 701         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 702                 return -EAGAIN;
 703
 704         if (new_euid != old_euid)
 705         {
 706                 current->mm->dumpable = suid_dumpable;
 707                 smp_wmb();
 708         }
 709         current->fsuid = current->euid = new_euid;
 710         if (ruid != (uid_t) -1 ||
 711             (euid != (uid_t) -1 && euid != old_ruid))
 712                 current->suid = current->euid;
 713         current->fsuid = current->euid;
 714
 715         key_fsuid_changed(current);
 716
 717         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 718 }
 719
 720
 721
 722 /*
 723  * setuid() is implemented like SysV with SAVED_IDS
 724  *
 725  * Note that SAVED_ID's is deficient in that a setuid root program
 726  * like sendmail, for example, cannot set its uid to be a normal
 727  * user and then switch back, because if you're root, setuid() sets
 728  * the saved uid too.  If you don't like this, blame the bright people
 729  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 730  * will allow a root program to temporarily drop privileges and be able to
 731  * regain them by swapping the real and effective uid.
 732  */
 733 asmlinkage long sys_setuid(uid_t uid)
 734 {
 735         int old_euid = current->euid;
 736         int old_ruid, old_suid, new_ruid, new_suid;
 737         int retval;
 738
 739         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 740         if (retval)
 741                 return retval;
 742
 743         old_ruid = new_ruid = current->uid;
 744         old_suid = current->suid;
 745         new_suid = old_suid;
 746
 747         if (capable(CAP_SETUID)) {
 748                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 749                         return -EAGAIN;
 750                 new_suid = uid;
 751         } else if ((uid != current->uid) && (uid != new_suid))
 752                 return -EPERM;
 753
 754         if (old_euid != uid)
 755         {
 756                 current->mm->dumpable = suid_dumpable;
 757                 smp_wmb();
 758         }
 759         current->fsuid = current->euid = uid;
 760         current->suid = new_suid;
 761
 762         key_fsuid_changed(current);
 763
 764         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 765 }
 766
 767
 768 /*
 769  * This function implements a generic ability to update ruid, euid,
 770  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 771  */
 772 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 773 {
 774         int old_ruid = current->uid;
 775         int old_euid = current->euid;
 776         int old_suid = current->suid;
 777         int retval;
 778
 779         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 780         if (retval)
 781                 return retval;
 782
 783         if (!capable(CAP_SETUID)) {
 784                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 785                     (ruid != current->euid) && (ruid != current->suid))
 786                         return -EPERM;
 787                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 788                     (euid != current->euid) && (euid != current->suid))
 789                         return -EPERM;
 790                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 791                     (suid != current->euid) && (suid != current->suid))
 792                         return -EPERM;
 793         }
 794         if (ruid != (uid_t) -1) {
 795                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 796                         return -EAGAIN;
 797         }
 798         if (euid != (uid_t) -1) {
 799                 if (euid != current->euid)
 800                 {
 801                         current->mm->dumpable = suid_dumpable;
 802                         smp_wmb();
 803                 }
 804                 current->euid = euid;
 805         }
 806         current->fsuid = current->euid;
 807         if (suid != (uid_t) -1)
 808                 current->suid = suid;
 809
 810         key_fsuid_changed(current);
 811
 812         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 813 }
 814
 815 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 816 {
 817         int retval;
 818
 819         if (!(retval = put_user(current->uid, ruid)) &&
 820             !(retval = put_user(current->euid, euid)))
 821                 retval = put_user(current->suid, suid);
 822
 823         return retval;
 824 }
 825
 826 /*
 827  * Same as above, but for rgid, egid, sgid.
 828  */
 829 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 830 {
 831         int retval;
 832
 833         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 834         if (retval)
 835                 return retval;
 836
 837         if (!capable(CAP_SETGID)) {
 838                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 839                     (rgid != current->egid) && (rgid != current->sgid))
 840                         return -EPERM;
 841                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 842                     (egid != current->egid) && (egid != current->sgid))
 843                         return -EPERM;
 844                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 845                     (sgid != current->egid) && (sgid != current->sgid))
 846                         return -EPERM;
 847         }
 848         if (egid != (gid_t) -1) {
 849                 if (egid != current->egid)
 850                 {
 851                         current->mm->dumpable = suid_dumpable;
 852                         smp_wmb();
 853                 }
 854                 current->egid = egid;
 855         }
 856         current->fsgid = current->egid;
 857         if (rgid != (gid_t) -1)
 858                 current->gid = rgid;
 859         if (sgid != (gid_t) -1)
 860                 current->sgid = sgid;
 861
 862         key_fsgid_changed(current);
 863         return 0;
 864 }
 865
 866 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 867 {
 868         int retval;
 869
 870         if (!(retval = put_user(current->gid, rgid)) &&
 871             !(retval = put_user(current->egid, egid)))
 872                 retval = put_user(current->sgid, sgid);
 873
 874         return retval;
 875 }
 876
 877
 878 /*
 879  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 880  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 881  * whatever uid it wants to). It normally shadows "euid", except when
 882  * explicitly set by setfsuid() or for access..
 883  */
 884 asmlinkage long sys_setfsuid(uid_t uid)
 885 {
 886         int old_fsuid;
 887
 888         old_fsuid = current->fsuid;
 889         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 890                 return old_fsuid;
 891
 892         if (uid == current->uid || uid == current->euid ||
 893             uid == current->suid || uid == current->fsuid ||
 894             capable(CAP_SETUID))
 895         {
 896                 if (uid != old_fsuid)
 897                 {
 898                         current->mm->dumpable = suid_dumpable;
 899                         smp_wmb();
 900                 }
 901                 current->fsuid = uid;
 902         }
 903
 904         key_fsuid_changed(current);
 905
 906         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 907
 908         return old_fsuid;
 909 }
 910
 911 /*
 912  * Samma på svenska..
 913  */
 914 asmlinkage long sys_setfsgid(gid_t gid)
 915 {
 916         int old_fsgid;
 917
 918         old_fsgid = current->fsgid;
 919         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 920                 return old_fsgid;
 921
 922         if (gid == current->gid || gid == current->egid ||
 923             gid == current->sgid || gid == current->fsgid ||
 924             capable(CAP_SETGID))
 925         {
 926                 if (gid != old_fsgid)
 927                 {
 928                         current->mm->dumpable = suid_dumpable;
 929                         smp_wmb();
 930                 }
 931                 current->fsgid = gid;
 932                 key_fsgid_changed(current);
 933         }
 934         return old_fsgid;
 935 }
 936
 937 asmlinkage long sys_times(struct tms __user * tbuf)
 938 {
 939         /*
 940          *      In the SMP world we might just be unlucky and have one of
 941          *      the times increment as we use it. Since the value is an
 942          *      atomically safe type this is just fine. Conceptually its
 943          *      as if the syscall took an instant longer to occur.
 944          */
 945         if (tbuf) {
 946                 struct tms tmp;
 947                 cputime_t utime, stime, cutime, cstime;
 948
 949 #ifdef CONFIG_SMP
 950                 if (thread_group_empty(current)) {
 951                         /*
 952                          * Single thread case without the use of any locks.
 953                          *
 954                          * We may race with release_task if two threads are
 955                          * executing. However, release task first adds up the
 956                          * counters (__exit_signal) before  removing the task
 957                          * from the process tasklist (__unhash_process).
 958                          * __exit_signal also acquires and releases the
 959                          * siglock which results in the proper memory ordering
 960                          * so that the list modifications are always visible
 961                          * after the counters have been updated.
 962                          *
 963                          * If the counters have been updated by the second thread
 964                          * but the thread has not yet been removed from the list
 965                          * then the other branch will be executing which will
 966                          * block on tasklist_lock until the exit handling of the
 967                          * other task is finished.
 968                          *
 969                          * This also implies that the sighand->siglock cannot
 970                          * be held by another processor. So we can also
 971                          * skip acquiring that lock.
 972                          */
 973                         utime = cputime_add(current->signal->utime, current->utime);
 974                         stime = cputime_add(current->signal->utime, current->stime);
 975                         cutime = current->signal->cutime;
 976                         cstime = current->signal->cstime;
 977                 } else
 978 #endif
 979                 {
 980
 981                         /* Process with multiple threads */
 982                         struct task_struct *tsk = current;
 983                         struct task_struct *t;
 984
 985                         read_lock(&tasklist_lock);
 986                         utime = tsk->signal->utime;
 987                         stime = tsk->signal->stime;
 988                         t = tsk;
 989                         do {
 990                                 utime = cputime_add(utime, t->utime);
 991                                 stime = cputime_add(stime, t->stime);
 992                                 t = next_thread(t);
 993                         } while (t != tsk);
 994
 995                         /*
 996                          * While we have tasklist_lock read-locked, no dying thread
 997                          * can be updating current->signal->[us]time.  Instead,
 998                          * we got their counts included in the live thread loop.
 999                          * However, another thread can come in right now and
1000                          * do a wait call that updates current->signal->c[us]time.
1001                          * To make sure we always see that pair updated atomically,
1002                          * we take the siglock around fetching them.
1003                          */
1004                         spin_lock_irq(&tsk->sighand->siglock);
1005                         cutime = tsk->signal->cutime;
1006                         cstime = tsk->signal->cstime;
1007                         spin_unlock_irq(&tsk->sighand->siglock);
1008                         read_unlock(&tasklist_lock);
1009                 }
1010                 tmp.tms_utime = cputime_to_clock_t(utime);
1011                 tmp.tms_stime = cputime_to_clock_t(stime);
1012                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1013                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1014                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1015                         return -EFAULT;
1016         }
1017         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1018 }
1019
1020 /*
1021  * This needs some heavy checking ...
1022  * I just haven't the stomach for it. I also don't fully
1023  * understand sessions/pgrp etc. Let somebody who does explain it.
1024  *
1025  * OK, I think I have the protection semantics right.... this is really
1026  * only important on a multi-user system anyway, to make sure one user
1027  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1028  *
1029  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1030  * LBT 04.03.94
1031  */
1032
1033 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1034 {
1035         struct task_struct *p;
1036         int err = -EINVAL;
1037
1038         if (!pid)
1039                 pid = current->pid;
1040         if (!pgid)
1041                 pgid = pid;
1042         if (pgid < 0)
1043                 return -EINVAL;
1044
1045         /* From this point forward we keep holding onto the tasklist lock
1046          * so that our parent does not change from under us. -DaveM
1047          */
1048         write_lock_irq(&tasklist_lock);
1049
1050         err = -ESRCH;
1051         p = find_task_by_pid(pid);
1052         if (!p)
1053                 goto out;
1054
1055         err = -EINVAL;
1056         if (!thread_group_leader(p))
1057                 goto out;
1058
1059         if (p->parent == current || p->real_parent == current) {
1060                 err = -EPERM;
1061                 if (p->signal->session != current->signal->session)
1062                         goto out;
1063                 err = -EACCES;
1064                 if (p->did_exec)
1065                         goto out;
1066         } else {
1067                 err = -ESRCH;
1068                 if (p != current)
1069                         goto out;
1070         }
1071
1072         err = -EPERM;
1073         if (p->signal->leader)
1074                 goto out;
1075
1076         if (pgid != pid) {
1077                 struct task_struct *p;
1078
1079                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1080                         if (p->signal->session == current->signal->session)
1081                                 goto ok_pgid;
1082                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1083                 goto out;
1084         }
1085
1086 ok_pgid:
1087         err = security_task_setpgid(p, pgid);
1088         if (err)
1089                 goto out;
1090
1091         if (process_group(p) != pgid) {
1092                 detach_pid(p, PIDTYPE_PGID);
1093                 p->signal->pgrp = pgid;
1094                 attach_pid(p, PIDTYPE_PGID, pgid);
1095         }
1096
1097         err = 0;
1098 out:
1099         /* All paths lead to here, thus we are safe. -DaveM */
1100         write_unlock_irq(&tasklist_lock);
1101         return err;
1102 }
1103
1104 asmlinkage long sys_getpgid(pid_t pid)
1105 {
1106         if (!pid) {
1107                 return process_group(current);
1108         } else {
1109                 int retval;
1110                 struct task_struct *p;
1111
1112                 read_lock(&tasklist_lock);
1113                 p = find_task_by_pid(pid);
1114
1115                 retval = -ESRCH;
1116                 if (p) {
1117                         retval = security_task_getpgid(p);
1118                         if (!retval)
1119                                 retval = process_group(p);
1120                 }
1121                 read_unlock(&tasklist_lock);
1122                 return retval;
1123         }
1124 }
1125
1126 #ifdef __ARCH_WANT_SYS_GETPGRP
1127
1128 asmlinkage long sys_getpgrp(void)
1129 {
1130         /* SMP - assuming writes are word atomic this is fine */
1131         return process_group(current);
1132 }
1133
1134 #endif
1135
1136 asmlinkage long sys_getsid(pid_t pid)
1137 {
1138         if (!pid) {
1139                 return current->signal->session;
1140         } else {
1141                 int retval;
1142                 struct task_struct *p;
1143
1144                 read_lock(&tasklist_lock);
1145                 p = find_task_by_pid(pid);
1146
1147                 retval = -ESRCH;
1148                 if(p) {
1149                         retval = security_task_getsid(p);
1150                         if (!retval)
1151                                 retval = p->signal->session;
1152                 }
1153                 read_unlock(&tasklist_lock);
1154                 return retval;
1155         }
1156 }
1157
1158 asmlinkage long sys_setsid(void)
1159 {
1160         struct pid *pid;
1161         int err = -EPERM;
1162
1163         if (!thread_group_leader(current))
1164                 return -EINVAL;
1165
1166         down(&tty_sem);
1167         write_lock_irq(&tasklist_lock);
1168
1169         pid = find_pid(PIDTYPE_PGID, current->pid);
1170         if (pid)
1171                 goto out;
1172
1173         current->signal->leader = 1;
1174         __set_special_pids(current->pid, current->pid);
1175         current->signal->tty = NULL;
1176         current->signal->tty_old_pgrp = 0;
1177         err = process_group(current);
1178 out:
1179         write_unlock_irq(&tasklist_lock);
1180         up(&tty_sem);
1181         return err;
1182 }
1183
1184 /*
1185  * Supplementary group IDs
1186  */
1187
1188 /* init to 2 - one for init_task, one to ensure it is never freed */
1189 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1190
1191 struct group_info *groups_alloc(int gidsetsize)
1192 {
1193         struct group_info *group_info;
1194         int nblocks;
1195         int i;
1196
1197         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1198         /* Make sure we always allocate at least one indirect block pointer */
1199         nblocks = nblocks ? : 1;
1200         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1201         if (!group_info)
1202                 return NULL;
1203         group_info->ngroups = gidsetsize;
1204         group_info->nblocks = nblocks;
1205         atomic_set(&group_info->usage, 1);
1206
1207         if (gidsetsize <= NGROUPS_SMALL) {
1208                 group_info->blocks[0] = group_info->small_block;
1209         } else {
1210                 for (i = 0; i < nblocks; i++) {
1211                         gid_t *b;
1212                         b = (void *)__get_free_page(GFP_USER);
1213                         if (!b)
1214                                 goto out_undo_partial_alloc;
1215                         group_info->blocks[i] = b;
1216                 }
1217         }
1218         return group_info;
1219
1220 out_undo_partial_alloc:
1221         while (--i >= 0) {
1222                 free_page((unsigned long)group_info->blocks[i]);
1223         }
1224         kfree(group_info);
1225         return NULL;
1226 }
1227
1228 EXPORT_SYMBOL(groups_alloc);
1229
1230 void groups_free(struct group_info *group_info)
1231 {
1232         if (group_info->blocks[0] != group_info->small_block) {
1233                 int i;
1234                 for (i = 0; i < group_info->nblocks; i++)
1235                         free_page((unsigned long)group_info->blocks[i]);
1236         }
1237         kfree(group_info);
1238 }
1239
1240 EXPORT_SYMBOL(groups_free);
1241
1242 /* export the group_info to a user-space array */
1243 static int groups_to_user(gid_t __user *grouplist,
1244     struct group_info *group_info)
1245 {
1246         int i;
1247         int count = group_info->ngroups;
1248
1249         for (i = 0; i < group_info->nblocks; i++) {
1250                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1251                 int off = i * NGROUPS_PER_BLOCK;
1252                 int len = cp_count * sizeof(*grouplist);
1253
1254                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1255                         return -EFAULT;
1256
1257                 count -= cp_count;
1258         }
1259         return 0;
1260 }
1261
1262 /* fill a group_info from a user-space array - it must be allocated already */
1263 static int groups_from_user(struct group_info *group_info,
1264     gid_t __user *grouplist)
1265  {
1266         int i;
1267         int count = group_info->ngroups;
1268
1269         for (i = 0; i < group_info->nblocks; i++) {
1270                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1271                 int off = i * NGROUPS_PER_BLOCK;
1272                 int len = cp_count * sizeof(*grouplist);
1273
1274                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1275                         return -EFAULT;
1276
1277                 count -= cp_count;
1278         }
1279         return 0;
1280 }
1281
1282 /* a simple Shell sort */
1283 static void groups_sort(struct group_info *group_info)
1284 {
1285         int base, max, stride;
1286         int gidsetsize = group_info->ngroups;
1287
1288         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1289                 ; /* nothing */
1290         stride /= 3;
1291
1292         while (stride) {
1293                 max = gidsetsize - stride;
1294                 for (base = 0; base < max; base++) {
1295                         int left = base;
1296                         int right = left + stride;
1297                         gid_t tmp = GROUP_AT(group_info, right);
1298
1299                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1300                                 GROUP_AT(group_info, right) =
1301                                     GROUP_AT(group_info, left);
1302                                 right = left;
1303                                 left -= stride;
1304                         }
1305                         GROUP_AT(group_info, right) = tmp;
1306                 }
1307                 stride /= 3;
1308         }
1309 }
1310
1311 /* a simple bsearch */
1312 int groups_search(struct group_info *group_info, gid_t grp)
1313 {
1314         int left, right;
1315
1316         if (!group_info)
1317                 return 0;
1318
1319         left = 0;
1320         right = group_info->ngroups;
1321         while (left < right) {
1322                 int mid = (left+right)/2;
1323                 int cmp = grp - GROUP_AT(group_info, mid);
1324                 if (cmp > 0)
1325                         left = mid + 1;
1326                 else if (cmp < 0)
1327                         right = mid;
1328                 else
1329                         return 1;
1330         }
1331         return 0;
1332 }
1333
1334 /* validate and set current->group_info */
1335 int set_current_groups(struct group_info *group_info)
1336 {
1337         int retval;
1338         struct group_info *old_info;
1339
1340         retval = security_task_setgroups(group_info);
1341         if (retval)
1342                 return retval;
1343
1344         groups_sort(group_info);
1345         get_group_info(group_info);
1346
1347         task_lock(current);
1348         old_info = current->group_info;
1349         current->group_info = group_info;
1350         task_unlock(current);
1351
1352         put_group_info(old_info);
1353
1354         return 0;
1355 }
1356
1357 EXPORT_SYMBOL(set_current_groups);
1358
1359 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1360 {
1361         int i = 0;
1362
1363         /*
1364          *      SMP: Nobody else can change our grouplist. Thus we are
1365          *      safe.
1366          */
1367
1368         if (gidsetsize < 0)
1369                 return -EINVAL;
1370
1371         /* no need to grab task_lock here; it cannot change */
1372         get_group_info(current->group_info);
1373         i = current->group_info->ngroups;
1374         if (gidsetsize) {
1375                 if (i > gidsetsize) {
1376                         i = -EINVAL;
1377                         goto out;
1378                 }
1379                 if (groups_to_user(grouplist, current->group_info)) {
1380                         i = -EFAULT;
1381                         goto out;
1382                 }
1383         }
1384 out:
1385         put_group_info(current->group_info);
1386         return i;
1387 }
1388
1389 /*
1390  *      SMP: Our groups are copy-on-write. We can set them safely
1391  *      without another task interfering.
1392  */
1393
1394 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1395 {
1396         struct group_info *group_info;
1397         int retval;
1398
1399         if (!capable(CAP_SETGID))
1400                 return -EPERM;
1401         if ((unsigned)gidsetsize > NGROUPS_MAX)
1402                 return -EINVAL;
1403
1404         group_info = groups_alloc(gidsetsize);
1405         if (!group_info)
1406                 return -ENOMEM;
1407         retval = groups_from_user(group_info, grouplist);
1408         if (retval) {
1409                 put_group_info(group_info);
1410                 return retval;
1411         }
1412
1413         retval = set_current_groups(group_info);
1414         put_group_info(group_info);
1415
1416         return retval;
1417 }
1418
1419 /*
1420  * Check whether we're fsgid/egid or in the supplemental group..
1421  */
1422 int in_group_p(gid_t grp)
1423 {
1424         int retval = 1;
1425         if (grp != current->fsgid) {
1426                 get_group_info(current->group_info);
1427                 retval = groups_search(current->group_info, grp);
1428                 put_group_info(current->group_info);
1429         }
1430         return retval;
1431 }
1432
1433 EXPORT_SYMBOL(in_group_p);
1434
1435 int in_egroup_p(gid_t grp)
1436 {
1437         int retval = 1;
1438         if (grp != current->egid) {
1439                 get_group_info(current->group_info);
1440                 retval = groups_search(current->group_info, grp);
1441                 put_group_info(current->group_info);
1442         }
1443         return retval;
1444 }
1445
1446 EXPORT_SYMBOL(in_egroup_p);
1447
1448 DECLARE_RWSEM(uts_sem);
1449
1450 EXPORT_SYMBOL(uts_sem);
1451
1452 asmlinkage long sys_newuname(struct new_utsname __user * name)
1453 {
1454         int errno = 0;
1455
1456         down_read(&uts_sem);
1457         if (copy_to_user(name,&system_utsname,sizeof *name))
1458                 errno = -EFAULT;
1459         up_read(&uts_sem);
1460         return errno;
1461 }
1462
1463 asmlinkage long sys_sethostname(char __user *name, int len)
1464 {
1465         int errno;
1466         char tmp[__NEW_UTS_LEN];
1467
1468         if (!capable(CAP_SYS_ADMIN))
1469                 return -EPERM;
1470         if (len < 0 || len > __NEW_UTS_LEN)
1471                 return -EINVAL;
1472         down_write(&uts_sem);
1473         errno = -EFAULT;
1474         if (!copy_from_user(tmp, name, len)) {
1475                 memcpy(system_utsname.nodename, tmp, len);
1476                 system_utsname.nodename[len] = 0;
1477                 errno = 0;
1478         }
1479         up_write(&uts_sem);
1480         return errno;
1481 }
1482
1483 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1484
1485 asmlinkage long sys_gethostname(char __user *name, int len)
1486 {
1487         int i, errno;
1488
1489         if (len < 0)
1490                 return -EINVAL;
1491         down_read(&uts_sem);
1492         i = 1 + strlen(system_utsname.nodename);
1493         if (i > len)
1494                 i = len;
1495         errno = 0;
1496         if (copy_to_user(name, system_utsname.nodename, i))
1497                 errno = -EFAULT;
1498         up_read(&uts_sem);
1499         return errno;
1500 }
1501
1502 #endif
1503
1504 /*
1505  * Only setdomainname; getdomainname can be implemented by calling
1506  * uname()
1507  */
1508 asmlinkage long sys_setdomainname(char __user *name, int len)
1509 {
1510         int errno;
1511         char tmp[__NEW_UTS_LEN];
1512
1513         if (!capable(CAP_SYS_ADMIN))
1514                 return -EPERM;
1515         if (len < 0 || len > __NEW_UTS_LEN)
1516                 return -EINVAL;
1517
1518         down_write(&uts_sem);
1519         errno = -EFAULT;
1520         if (!copy_from_user(tmp, name, len)) {
1521                 memcpy(system_utsname.domainname, tmp, len);
1522                 system_utsname.domainname[len] = 0;
1523                 errno = 0;
1524         }
1525         up_write(&uts_sem);
1526         return errno;
1527 }
1528
1529 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1530 {
1531         if (resource >= RLIM_NLIMITS)
1532                 return -EINVAL;
1533         else {
1534                 struct rlimit value;
1535                 task_lock(current->group_leader);
1536                 value = current->signal->rlim[resource];
1537                 task_unlock(current->group_leader);
1538                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1539         }
1540 }
1541
1542 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1543
1544 /*
1545  *      Back compatibility for getrlimit. Needed for some apps.
1546  */
1547
1548 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1549 {
1550         struct rlimit x;
1551         if (resource >= RLIM_NLIMITS)
1552                 return -EINVAL;
1553
1554         task_lock(current->group_leader);
1555         x = current->signal->rlim[resource];
1556         task_unlock(current->group_leader);
1557         if(x.rlim_cur > 0x7FFFFFFF)
1558                 x.rlim_cur = 0x7FFFFFFF;
1559         if(x.rlim_max > 0x7FFFFFFF)
1560                 x.rlim_max = 0x7FFFFFFF;
1561         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1562 }
1563
1564 #endif
1565
1566 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1567 {
1568         struct rlimit new_rlim, *old_rlim;
1569         int retval;
1570
1571         if (resource >= RLIM_NLIMITS)
1572                 return -EINVAL;
1573         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1574                 return -EFAULT;
1575        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1576                return -EINVAL;
1577         old_rlim = current->signal->rlim + resource;
1578         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1579             !capable(CAP_SYS_RESOURCE))
1580                 return -EPERM;
1581         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1582                         return -EPERM;
1583
1584         retval = security_task_setrlimit(resource, &new_rlim);
1585         if (retval)
1586                 return retval;
1587
1588         task_lock(current->group_leader);
1589         *old_rlim = new_rlim;
1590         task_unlock(current->group_leader);
1591
1592         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1593             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1594              new_rlim.rlim_cur <= cputime_to_secs(
1595                      current->signal->it_prof_expires))) {
1596                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1597                 read_lock(&tasklist_lock);
1598                 spin_lock_irq(&current->sighand->siglock);
1599                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1600                                       &cputime, NULL);
1601                 spin_unlock_irq(&current->sighand->siglock);
1602                 read_unlock(&tasklist_lock);
1603         }
1604
1605         return 0;
1606 }
1607
1608 /*
1609  * It would make sense to put struct rusage in the task_struct,
1610  * except that would make the task_struct be *really big*.  After
1611  * task_struct gets moved into malloc'ed memory, it would
1612  * make sense to do this.  It will make moving the rest of the information
1613  * a lot simpler!  (Which we're not doing right now because we're not
1614  * measuring them yet).
1615  *
1616  * This expects to be called with tasklist_lock read-locked or better,
1617  * and the siglock not locked.  It may momentarily take the siglock.
1618  *
1619  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1620  * races with threads incrementing their own counters.  But since word
1621  * reads are atomic, we either get new values or old values and we don't
1622  * care which for the sums.  We always take the siglock to protect reading
1623  * the c* fields from p->signal from races with exit.c updating those
1624  * fields when reaping, so a sample either gets all the additions of a
1625  * given child after it's reaped, or none so this sample is before reaping.
1626  */
1627
1628 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1629 {
1630         struct task_struct *t;
1631         unsigned long flags;
1632         cputime_t utime, stime;
1633
1634         memset((char *) r, 0, sizeof *r);
1635
1636         if (unlikely(!p->signal))
1637                 return;
1638
1639         switch (who) {
1640                 case RUSAGE_CHILDREN:
1641                         spin_lock_irqsave(&p->sighand->siglock, flags);
1642                         utime = p->signal->cutime;
1643                         stime = p->signal->cstime;
1644                         r->ru_nvcsw = p->signal->cnvcsw;
1645                         r->ru_nivcsw = p->signal->cnivcsw;
1646                         r->ru_minflt = p->signal->cmin_flt;
1647                         r->ru_majflt = p->signal->cmaj_flt;
1648                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1649                         cputime_to_timeval(utime, &r->ru_utime);
1650                         cputime_to_timeval(stime, &r->ru_stime);
1651                         break;
1652                 case RUSAGE_SELF:
1653                         spin_lock_irqsave(&p->sighand->siglock, flags);
1654                         utime = stime = cputime_zero;
1655                         goto sum_group;
1656                 case RUSAGE_BOTH:
1657                         spin_lock_irqsave(&p->sighand->siglock, flags);
1658                         utime = p->signal->cutime;
1659                         stime = p->signal->cstime;
1660                         r->ru_nvcsw = p->signal->cnvcsw;
1661                         r->ru_nivcsw = p->signal->cnivcsw;
1662                         r->ru_minflt = p->signal->cmin_flt;
1663                         r->ru_majflt = p->signal->cmaj_flt;
1664                 sum_group:
1665                         utime = cputime_add(utime, p->signal->utime);
1666                         stime = cputime_add(stime, p->signal->stime);
1667                         r->ru_nvcsw += p->signal->nvcsw;
1668                         r->ru_nivcsw += p->signal->nivcsw;
1669                         r->ru_minflt += p->signal->min_flt;
1670                         r->ru_majflt += p->signal->maj_flt;
1671                         t = p;
1672                         do {
1673                                 utime = cputime_add(utime, t->utime);
1674                                 stime = cputime_add(stime, t->stime);
1675                                 r->ru_nvcsw += t->nvcsw;
1676                                 r->ru_nivcsw += t->nivcsw;
1677                                 r->ru_minflt += t->min_flt;
1678                                 r->ru_majflt += t->maj_flt;
1679                                 t = next_thread(t);
1680                         } while (t != p);
1681                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1682                         cputime_to_timeval(utime, &r->ru_utime);
1683                         cputime_to_timeval(stime, &r->ru_stime);
1684                         break;
1685                 default:
1686                         BUG();
1687         }
1688 }
1689
1690 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1691 {
1692         struct rusage r;
1693         read_lock(&tasklist_lock);
1694         k_getrusage(p, who, &r);
1695         read_unlock(&tasklist_lock);
1696         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1697 }
1698
1699 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1700 {
1701         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1702                 return -EINVAL;
1703         return getrusage(current, who, ru);
1704 }
1705
1706 asmlinkage long sys_umask(int mask)
1707 {
1708         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1709         return mask;
1710 }
1711
1712 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1713                           unsigned long arg4, unsigned long arg5)
1714 {
1715         long error;
1716         int sig;
1717
1718         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1719         if (error)
1720                 return error;
1721
1722         switch (option) {
1723                 case PR_SET_PDEATHSIG:
1724                         sig = arg2;
1725                         if (!valid_signal(sig)) {
1726                                 error = -EINVAL;
1727                                 break;
1728                         }
1729                         current->pdeath_signal = sig;
1730                         break;
1731                 case PR_GET_PDEATHSIG:
1732                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1733                         break;
1734                 case PR_GET_DUMPABLE:
1735                         if (current->mm->dumpable)
1736                                 error = 1;
1737                         break;
1738                 case PR_SET_DUMPABLE:
1739                         if (arg2 < 0 || arg2 > 2) {
1740                                 error = -EINVAL;
1741                                 break;
1742                         }
1743                         current->mm->dumpable = arg2;
1744                         break;
1745
1746                 case PR_SET_UNALIGN:
1747                         error = SET_UNALIGN_CTL(current, arg2);
1748                         break;
1749                 case PR_GET_UNALIGN:
1750                         error = GET_UNALIGN_CTL(current, arg2);
1751                         break;
1752                 case PR_SET_FPEMU:
1753                         error = SET_FPEMU_CTL(current, arg2);
1754                         break;
1755                 case PR_GET_FPEMU:
1756                         error = GET_FPEMU_CTL(current, arg2);
1757                         break;
1758                 case PR_SET_FPEXC:
1759                         error = SET_FPEXC_CTL(current, arg2);
1760                         break;
1761                 case PR_GET_FPEXC:
1762                         error = GET_FPEXC_CTL(current, arg2);
1763                         break;
1764                 case PR_GET_TIMING:
1765                         error = PR_TIMING_STATISTICAL;
1766                         break;
1767                 case PR_SET_TIMING:
1768                         if (arg2 == PR_TIMING_STATISTICAL)
1769                                 error = 0;
1770                         else
1771                                 error = -EINVAL;
1772                         break;
1773
1774                 case PR_GET_KEEPCAPS:
1775                         if (current->keep_capabilities)
1776                                 error = 1;
1777                         break;
1778                 case PR_SET_KEEPCAPS:
1779                         if (arg2 != 0 && arg2 != 1) {
1780                                 error = -EINVAL;
1781                                 break;
1782                         }
1783                         current->keep_capabilities = arg2;
1784                         break;
1785                 case PR_SET_NAME: {
1786                         struct task_struct *me = current;
1787                         unsigned char ncomm[sizeof(me->comm)];
1788
1789                         ncomm[sizeof(me->comm)-1] = 0;
1790                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1791                                                 sizeof(me->comm)-1) < 0)
1792                                 return -EFAULT;
1793                         set_task_comm(me, ncomm);
1794                         return 0;
1795                 }
1796                 case PR_GET_NAME: {
1797                         struct task_struct *me = current;
1798                         unsigned char tcomm[sizeof(me->comm)];
1799
1800                         get_task_comm(tcomm, me);
1801                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1802                                 return -EFAULT;
1803                         return 0;
1804                 }
1805                 default:
1806                         error = -EINVAL;
1807                         break;
1808         }
1809         return error;
1810 }