git.oblomov.eu Git - 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 #include <linux/cn_proc.h>
  32
  33 #include <linux/compat.h>
  34 #include <linux/syscalls.h>
  35
  36 #include <asm/uaccess.h>
  37 #include <asm/io.h>
  38 #include <asm/unistd.h>
  39
  40 #ifndef SET_UNALIGN_CTL
  41 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  42 #endif
  43 #ifndef GET_UNALIGN_CTL
  44 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  45 #endif
  46 #ifndef SET_FPEMU_CTL
  47 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  48 #endif
  49 #ifndef GET_FPEMU_CTL
  50 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  51 #endif
  52 #ifndef SET_FPEXC_CTL
  53 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  54 #endif
  55 #ifndef GET_FPEXC_CTL
  56 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  57 #endif
  58
  59 /*
  60  * this is where the system-wide overflow UID and GID are defined, for
  61  * architectures that now have 32-bit UID/GID but didn't in the past
  62  */
  63
  64 int overflowuid = DEFAULT_OVERFLOWUID;
  65 int overflowgid = DEFAULT_OVERFLOWGID;
  66
  67 #ifdef CONFIG_UID16
  68 EXPORT_SYMBOL(overflowuid);
  69 EXPORT_SYMBOL(overflowgid);
  70 #endif
  71
  72 /*
  73  * the same as above, but for filesystems which can only store a 16-bit
  74  * UID and GID. as such, this is needed on all architectures
  75  */
  76
  77 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  78 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  79
  80 EXPORT_SYMBOL(fs_overflowuid);
  81 EXPORT_SYMBOL(fs_overflowgid);
  82
  83 /*
  84  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
  85  */
  86
  87 int C_A_D = 1;
  88 int cad_pid = 1;
  89
  90 /*
  91  *      Notifier list for kernel code which wants to be called
  92  *      at shutdown. This is used to stop any idling DMA operations
  93  *      and the like.
  94  */
  95
  96 static struct notifier_block *reboot_notifier_list;
  97 static DEFINE_RWLOCK(notifier_lock);
  98
  99 /**
 100  *      notifier_chain_register - Add notifier to a notifier chain
 101  *      @list: Pointer to root list pointer
 102  *      @n: New entry in notifier chain
 103  *
 104  *      Adds a notifier to a notifier chain.
 105  *
 106  *      Currently always returns zero.
 107  */
 108
 109 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
 110 {
 111         write_lock(&notifier_lock);
 112         while(*list)
 113         {
 114                 if(n->priority > (*list)->priority)
 115                         break;
 116                 list= &((*list)->next);
 117         }
 118         n->next = *list;
 119         *list=n;
 120         write_unlock(&notifier_lock);
 121         return 0;
 122 }
 123
 124 EXPORT_SYMBOL(notifier_chain_register);
 125
 126 /**
 127  *      notifier_chain_unregister - Remove notifier from a notifier chain
 128  *      @nl: Pointer to root list pointer
 129  *      @n: New entry in notifier chain
 130  *
 131  *      Removes a notifier from a notifier chain.
 132  *
 133  *      Returns zero on success, or %-ENOENT on failure.
 134  */
 135
 136 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
 137 {
 138         write_lock(&notifier_lock);
 139         while((*nl)!=NULL)
 140         {
 141                 if((*nl)==n)
 142                 {
 143                         *nl=n->next;
 144                         write_unlock(&notifier_lock);
 145                         return 0;
 146                 }
 147                 nl=&((*nl)->next);
 148         }
 149         write_unlock(&notifier_lock);
 150         return -ENOENT;
 151 }
 152
 153 EXPORT_SYMBOL(notifier_chain_unregister);
 154
 155 /**
 156  *      notifier_call_chain - Call functions in a notifier chain
 157  *      @n: Pointer to root pointer of notifier chain
 158  *      @val: Value passed unmodified to notifier function
 159  *      @v: Pointer passed unmodified to notifier function
 160  *
 161  *      Calls each function in a notifier chain in turn.
 162  *
 163  *      If the return value of the notifier can be and'd
 164  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
 165  *      will return immediately, with the return value of
 166  *      the notifier function which halted execution.
 167  *      Otherwise, the return value is the return value
 168  *      of the last notifier function called.
 169  */
 170
 171 int notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
 172 {
 173         int ret=NOTIFY_DONE;
 174         struct notifier_block *nb = *n;
 175
 176         while(nb)
 177         {
 178                 ret=nb->notifier_call(nb,val,v);
 179                 if(ret&NOTIFY_STOP_MASK)
 180                 {
 181                         return ret;
 182                 }
 183                 nb=nb->next;
 184         }
 185         return ret;
 186 }
 187
 188 EXPORT_SYMBOL(notifier_call_chain);
 189
 190 /**
 191  *      register_reboot_notifier - Register function to be called at reboot time
 192  *      @nb: Info about notifier function to be called
 193  *
 194  *      Registers a function with the list of functions
 195  *      to be called at reboot time.
 196  *
 197  *      Currently always returns zero, as notifier_chain_register
 198  *      always returns zero.
 199  */
 200
 201 int register_reboot_notifier(struct notifier_block * nb)
 202 {
 203         return notifier_chain_register(&reboot_notifier_list, nb);
 204 }
 205
 206 EXPORT_SYMBOL(register_reboot_notifier);
 207
 208 /**
 209  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 210  *      @nb: Hook to be unregistered
 211  *
 212  *      Unregisters a previously registered reboot
 213  *      notifier function.
 214  *
 215  *      Returns zero on success, or %-ENOENT on failure.
 216  */
 217
 218 int unregister_reboot_notifier(struct notifier_block * nb)
 219 {
 220         return notifier_chain_unregister(&reboot_notifier_list, nb);
 221 }
 222
 223 EXPORT_SYMBOL(unregister_reboot_notifier);
 224
 225 static int set_one_prio(struct task_struct *p, int niceval, int error)
 226 {
 227         int no_nice;
 228
 229         if (p->uid != current->euid &&
 230                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
 231                 error = -EPERM;
 232                 goto out;
 233         }
 234         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 235                 error = -EACCES;
 236                 goto out;
 237         }
 238         no_nice = security_task_setnice(p, niceval);
 239         if (no_nice) {
 240                 error = no_nice;
 241                 goto out;
 242         }
 243         if (error == -ESRCH)
 244                 error = 0;
 245         set_user_nice(p, niceval);
 246 out:
 247         return error;
 248 }
 249
 250 asmlinkage long sys_setpriority(int which, int who, int niceval)
 251 {
 252         struct task_struct *g, *p;
 253         struct user_struct *user;
 254         int error = -EINVAL;
 255
 256         if (which > 2 || which < 0)
 257                 goto out;
 258
 259         /* normalize: avoid signed division (rounding problems) */
 260         error = -ESRCH;
 261         if (niceval < -20)
 262                 niceval = -20;
 263         if (niceval > 19)
 264                 niceval = 19;
 265
 266         read_lock(&tasklist_lock);
 267         switch (which) {
 268                 case PRIO_PROCESS:
 269                         if (!who)
 270                                 who = current->pid;
 271                         p = find_task_by_pid(who);
 272                         if (p)
 273                                 error = set_one_prio(p, niceval, error);
 274                         break;
 275                 case PRIO_PGRP:
 276                         if (!who)
 277                                 who = process_group(current);
 278                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 279                                 error = set_one_prio(p, niceval, error);
 280                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 281                         break;
 282                 case PRIO_USER:
 283                         user = current->user;
 284                         if (!who)
 285                                 who = current->uid;
 286                         else
 287                                 if ((who != current->uid) && !(user = find_user(who)))
 288                                         goto out_unlock;        /* No processes for this user */
 289
 290                         do_each_thread(g, p)
 291                                 if (p->uid == who)
 292                                         error = set_one_prio(p, niceval, error);
 293                         while_each_thread(g, p);
 294                         if (who != current->uid)
 295                                 free_uid(user);         /* For find_user() */
 296                         break;
 297         }
 298 out_unlock:
 299         read_unlock(&tasklist_lock);
 300 out:
 301         return error;
 302 }
 303
 304 /*
 305  * Ugh. To avoid negative return values, "getpriority()" will
 306  * not return the normal nice-value, but a negated value that
 307  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 308  * to stay compatible.
 309  */
 310 asmlinkage long sys_getpriority(int which, int who)
 311 {
 312         struct task_struct *g, *p;
 313         struct user_struct *user;
 314         long niceval, retval = -ESRCH;
 315
 316         if (which > 2 || which < 0)
 317                 return -EINVAL;
 318
 319         read_lock(&tasklist_lock);
 320         switch (which) {
 321                 case PRIO_PROCESS:
 322                         if (!who)
 323                                 who = current->pid;
 324                         p = find_task_by_pid(who);
 325                         if (p) {
 326                                 niceval = 20 - task_nice(p);
 327                                 if (niceval > retval)
 328                                         retval = niceval;
 329                         }
 330                         break;
 331                 case PRIO_PGRP:
 332                         if (!who)
 333                                 who = process_group(current);
 334                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 335                                 niceval = 20 - task_nice(p);
 336                                 if (niceval > retval)
 337                                         retval = niceval;
 338                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 339                         break;
 340                 case PRIO_USER:
 341                         user = current->user;
 342                         if (!who)
 343                                 who = current->uid;
 344                         else
 345                                 if ((who != current->uid) && !(user = find_user(who)))
 346                                         goto out_unlock;        /* No processes for this user */
 347
 348                         do_each_thread(g, p)
 349                                 if (p->uid == who) {
 350                                         niceval = 20 - task_nice(p);
 351                                         if (niceval > retval)
 352                                                 retval = niceval;
 353                                 }
 354                         while_each_thread(g, p);
 355                         if (who != current->uid)
 356                                 free_uid(user);         /* for find_user() */
 357                         break;
 358         }
 359 out_unlock:
 360         read_unlock(&tasklist_lock);
 361
 362         return retval;
 363 }
 364
 365 /**
 366  *      emergency_restart - reboot the system
 367  *
 368  *      Without shutting down any hardware or taking any locks
 369  *      reboot the system.  This is called when we know we are in
 370  *      trouble so this is our best effort to reboot.  This is
 371  *      safe to call in interrupt context.
 372  */
 373 void emergency_restart(void)
 374 {
 375         machine_emergency_restart();
 376 }
 377 EXPORT_SYMBOL_GPL(emergency_restart);
 378
 379 void kernel_restart_prepare(char *cmd)
 380 {
 381         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 382         system_state = SYSTEM_RESTART;
 383         device_shutdown();
 384 }
 385
 386 /**
 387  *      kernel_restart - reboot the system
 388  *      @cmd: pointer to buffer containing command to execute for restart
 389  *              or %NULL
 390  *
 391  *      Shutdown everything and perform a clean reboot.
 392  *      This is not safe to call in interrupt context.
 393  */
 394 void kernel_restart(char *cmd)
 395 {
 396         kernel_restart_prepare(cmd);
 397         if (!cmd) {
 398                 printk(KERN_EMERG "Restarting system.\n");
 399         } else {
 400                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 401         }
 402         printk(".\n");
 403         machine_restart(cmd);
 404 }
 405 EXPORT_SYMBOL_GPL(kernel_restart);
 406
 407 /**
 408  *      kernel_kexec - reboot the system
 409  *
 410  *      Move into place and start executing a preloaded standalone
 411  *      executable.  If nothing was preloaded return an error.
 412  */
 413 void kernel_kexec(void)
 414 {
 415 #ifdef CONFIG_KEXEC
 416         struct kimage *image;
 417         image = xchg(&kexec_image, 0);
 418         if (!image) {
 419                 return;
 420         }
 421         kernel_restart_prepare(NULL);
 422         printk(KERN_EMERG "Starting new kernel\n");
 423         machine_shutdown();
 424         machine_kexec(image);
 425 #endif
 426 }
 427 EXPORT_SYMBOL_GPL(kernel_kexec);
 428
 429 /**
 430  *      kernel_halt - halt the system
 431  *
 432  *      Shutdown everything and perform a clean system halt.
 433  */
 434 void kernel_halt_prepare(void)
 435 {
 436         notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
 437         system_state = SYSTEM_HALT;
 438         device_shutdown();
 439 }
 440 void kernel_halt(void)
 441 {
 442         kernel_halt_prepare();
 443         printk(KERN_EMERG "System halted.\n");
 444         machine_halt();
 445 }
 446 EXPORT_SYMBOL_GPL(kernel_halt);
 447
 448 /**
 449  *      kernel_power_off - power_off the system
 450  *
 451  *      Shutdown everything and perform a clean system power_off.
 452  */
 453 void kernel_power_off_prepare(void)
 454 {
 455         notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
 456         system_state = SYSTEM_POWER_OFF;
 457         device_shutdown();
 458 }
 459 void kernel_power_off(void)
 460 {
 461         kernel_power_off_prepare();
 462         printk(KERN_EMERG "Power down.\n");
 463         machine_power_off();
 464 }
 465 EXPORT_SYMBOL_GPL(kernel_power_off);
 466
 467 /*
 468  * Reboot system call: for obvious reasons only root may call it,
 469  * and even root needs to set up some magic numbers in the registers
 470  * so that some mistake won't make this reboot the whole machine.
 471  * You can also set the meaning of the ctrl-alt-del-key here.
 472  *
 473  * reboot doesn't sync: do that yourself before calling this.
 474  */
 475 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 476 {
 477         char buffer[256];
 478
 479         /* We only trust the superuser with rebooting the system. */
 480         if (!capable(CAP_SYS_BOOT))
 481                 return -EPERM;
 482
 483         /* For safety, we require "magic" arguments. */
 484         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 485             (magic2 != LINUX_REBOOT_MAGIC2 &&
 486                         magic2 != LINUX_REBOOT_MAGIC2A &&
 487                         magic2 != LINUX_REBOOT_MAGIC2B &&
 488                         magic2 != LINUX_REBOOT_MAGIC2C))
 489                 return -EINVAL;
 490
 491         lock_kernel();
 492         switch (cmd) {
 493         case LINUX_REBOOT_CMD_RESTART:
 494                 kernel_restart(NULL);
 495                 break;
 496
 497         case LINUX_REBOOT_CMD_CAD_ON:
 498                 C_A_D = 1;
 499                 break;
 500
 501         case LINUX_REBOOT_CMD_CAD_OFF:
 502                 C_A_D = 0;
 503                 break;
 504
 505         case LINUX_REBOOT_CMD_HALT:
 506                 kernel_halt();
 507                 unlock_kernel();
 508                 do_exit(0);
 509                 break;
 510
 511         case LINUX_REBOOT_CMD_POWER_OFF:
 512                 kernel_power_off();
 513                 unlock_kernel();
 514                 do_exit(0);
 515                 break;
 516
 517         case LINUX_REBOOT_CMD_RESTART2:
 518                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 519                         unlock_kernel();
 520                         return -EFAULT;
 521                 }
 522                 buffer[sizeof(buffer) - 1] = '\0';
 523
 524                 kernel_restart(buffer);
 525                 break;
 526
 527         case LINUX_REBOOT_CMD_KEXEC:
 528                 kernel_kexec();
 529                 unlock_kernel();
 530                 return -EINVAL;
 531
 532 #ifdef CONFIG_SOFTWARE_SUSPEND
 533         case LINUX_REBOOT_CMD_SW_SUSPEND:
 534                 {
 535                         int ret = software_suspend();
 536                         unlock_kernel();
 537                         return ret;
 538                 }
 539 #endif
 540
 541         default:
 542                 unlock_kernel();
 543                 return -EINVAL;
 544         }
 545         unlock_kernel();
 546         return 0;
 547 }
 548
 549 static void deferred_cad(void *dummy)
 550 {
 551         kernel_restart(NULL);
 552 }
 553
 554 /*
 555  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 556  * As it's called within an interrupt, it may NOT sync: the only choice
 557  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 558  */
 559 void ctrl_alt_del(void)
 560 {
 561         static DECLARE_WORK(cad_work, deferred_cad, NULL);
 562
 563         if (C_A_D)
 564                 schedule_work(&cad_work);
 565         else
 566                 kill_proc(cad_pid, SIGINT, 1);
 567 }
 568
 569
 570 /*
 571  * Unprivileged users may change the real gid to the effective gid
 572  * or vice versa.  (BSD-style)
 573  *
 574  * If you set the real gid at all, or set the effective gid to a value not
 575  * equal to the real gid, then the saved gid is set to the new effective gid.
 576  *
 577  * This makes it possible for a setgid program to completely drop its
 578  * privileges, which is often a useful assertion to make when you are doing
 579  * a security audit over a program.
 580  *
 581  * The general idea is that a program which uses just setregid() will be
 582  * 100% compatible with BSD.  A program which uses just setgid() will be
 583  * 100% compatible with POSIX with saved IDs.
 584  *
 585  * SMP: There are not races, the GIDs are checked only by filesystem
 586  *      operations (as far as semantic preservation is concerned).
 587  */
 588 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 589 {
 590         int old_rgid = current->gid;
 591         int old_egid = current->egid;
 592         int new_rgid = old_rgid;
 593         int new_egid = old_egid;
 594         int retval;
 595
 596         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 597         if (retval)
 598                 return retval;
 599
 600         if (rgid != (gid_t) -1) {
 601                 if ((old_rgid == rgid) ||
 602                     (current->egid==rgid) ||
 603                     capable(CAP_SETGID))
 604                         new_rgid = rgid;
 605                 else
 606                         return -EPERM;
 607         }
 608         if (egid != (gid_t) -1) {
 609                 if ((old_rgid == egid) ||
 610                     (current->egid == egid) ||
 611                     (current->sgid == egid) ||
 612                     capable(CAP_SETGID))
 613                         new_egid = egid;
 614                 else {
 615                         return -EPERM;
 616                 }
 617         }
 618         if (new_egid != old_egid)
 619         {
 620                 current->mm->dumpable = suid_dumpable;
 621                 smp_wmb();
 622         }
 623         if (rgid != (gid_t) -1 ||
 624             (egid != (gid_t) -1 && egid != old_rgid))
 625                 current->sgid = new_egid;
 626         current->fsgid = new_egid;
 627         current->egid = new_egid;
 628         current->gid = new_rgid;
 629         key_fsgid_changed(current);
 630         proc_id_connector(current, PROC_EVENT_GID);
 631         return 0;
 632 }
 633
 634 /*
 635  * setgid() is implemented like SysV w/ SAVED_IDS
 636  *
 637  * SMP: Same implicit races as above.
 638  */
 639 asmlinkage long sys_setgid(gid_t gid)
 640 {
 641         int old_egid = current->egid;
 642         int retval;
 643
 644         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 645         if (retval)
 646                 return retval;
 647
 648         if (capable(CAP_SETGID))
 649         {
 650                 if(old_egid != gid)
 651                 {
 652                         current->mm->dumpable = suid_dumpable;
 653                         smp_wmb();
 654                 }
 655                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 656         }
 657         else if ((gid == current->gid) || (gid == current->sgid))
 658         {
 659                 if(old_egid != gid)
 660                 {
 661                         current->mm->dumpable = suid_dumpable;
 662                         smp_wmb();
 663                 }
 664                 current->egid = current->fsgid = gid;
 665         }
 666         else
 667                 return -EPERM;
 668
 669         key_fsgid_changed(current);
 670         proc_id_connector(current, PROC_EVENT_GID);
 671         return 0;
 672 }
 673
 674 static int set_user(uid_t new_ruid, int dumpclear)
 675 {
 676         struct user_struct *new_user;
 677
 678         new_user = alloc_uid(new_ruid);
 679         if (!new_user)
 680                 return -EAGAIN;
 681
 682         if (atomic_read(&new_user->processes) >=
 683                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 684                         new_user != &root_user) {
 685                 free_uid(new_user);
 686                 return -EAGAIN;
 687         }
 688
 689         switch_uid(new_user);
 690
 691         if(dumpclear)
 692         {
 693                 current->mm->dumpable = suid_dumpable;
 694                 smp_wmb();
 695         }
 696         current->uid = new_ruid;
 697         return 0;
 698 }
 699
 700 /*
 701  * Unprivileged users may change the real uid to the effective uid
 702  * or vice versa.  (BSD-style)
 703  *
 704  * If you set the real uid at all, or set the effective uid to a value not
 705  * equal to the real uid, then the saved uid is set to the new effective uid.
 706  *
 707  * This makes it possible for a setuid program to completely drop its
 708  * privileges, which is often a useful assertion to make when you are doing
 709  * a security audit over a program.
 710  *
 711  * The general idea is that a program which uses just setreuid() will be
 712  * 100% compatible with BSD.  A program which uses just setuid() will be
 713  * 100% compatible with POSIX with saved IDs.
 714  */
 715 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 716 {
 717         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 718         int retval;
 719
 720         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 721         if (retval)
 722                 return retval;
 723
 724         new_ruid = old_ruid = current->uid;
 725         new_euid = old_euid = current->euid;
 726         old_suid = current->suid;
 727
 728         if (ruid != (uid_t) -1) {
 729                 new_ruid = ruid;
 730                 if ((old_ruid != ruid) &&
 731                     (current->euid != ruid) &&
 732                     !capable(CAP_SETUID))
 733                         return -EPERM;
 734         }
 735
 736         if (euid != (uid_t) -1) {
 737                 new_euid = euid;
 738                 if ((old_ruid != euid) &&
 739                     (current->euid != euid) &&
 740                     (current->suid != euid) &&
 741                     !capable(CAP_SETUID))
 742                         return -EPERM;
 743         }
 744
 745         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 746                 return -EAGAIN;
 747
 748         if (new_euid != old_euid)
 749         {
 750                 current->mm->dumpable = suid_dumpable;
 751                 smp_wmb();
 752         }
 753         current->fsuid = current->euid = new_euid;
 754         if (ruid != (uid_t) -1 ||
 755             (euid != (uid_t) -1 && euid != old_ruid))
 756                 current->suid = current->euid;
 757         current->fsuid = current->euid;
 758
 759         key_fsuid_changed(current);
 760         proc_id_connector(current, PROC_EVENT_UID);
 761
 762         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 763 }
 764
 765
 766
 767 /*
 768  * setuid() is implemented like SysV with SAVED_IDS
 769  *
 770  * Note that SAVED_ID's is deficient in that a setuid root program
 771  * like sendmail, for example, cannot set its uid to be a normal
 772  * user and then switch back, because if you're root, setuid() sets
 773  * the saved uid too.  If you don't like this, blame the bright people
 774  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 775  * will allow a root program to temporarily drop privileges and be able to
 776  * regain them by swapping the real and effective uid.
 777  */
 778 asmlinkage long sys_setuid(uid_t uid)
 779 {
 780         int old_euid = current->euid;
 781         int old_ruid, old_suid, new_ruid, new_suid;
 782         int retval;
 783
 784         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 785         if (retval)
 786                 return retval;
 787
 788         old_ruid = new_ruid = current->uid;
 789         old_suid = current->suid;
 790         new_suid = old_suid;
 791
 792         if (capable(CAP_SETUID)) {
 793                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 794                         return -EAGAIN;
 795                 new_suid = uid;
 796         } else if ((uid != current->uid) && (uid != new_suid))
 797                 return -EPERM;
 798
 799         if (old_euid != uid)
 800         {
 801                 current->mm->dumpable = suid_dumpable;
 802                 smp_wmb();
 803         }
 804         current->fsuid = current->euid = uid;
 805         current->suid = new_suid;
 806
 807         key_fsuid_changed(current);
 808         proc_id_connector(current, PROC_EVENT_UID);
 809
 810         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 811 }
 812
 813
 814 /*
 815  * This function implements a generic ability to update ruid, euid,
 816  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 817  */
 818 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 819 {
 820         int old_ruid = current->uid;
 821         int old_euid = current->euid;
 822         int old_suid = current->suid;
 823         int retval;
 824
 825         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 826         if (retval)
 827                 return retval;
 828
 829         if (!capable(CAP_SETUID)) {
 830                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 831                     (ruid != current->euid) && (ruid != current->suid))
 832                         return -EPERM;
 833                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 834                     (euid != current->euid) && (euid != current->suid))
 835                         return -EPERM;
 836                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 837                     (suid != current->euid) && (suid != current->suid))
 838                         return -EPERM;
 839         }
 840         if (ruid != (uid_t) -1) {
 841                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 842                         return -EAGAIN;
 843         }
 844         if (euid != (uid_t) -1) {
 845                 if (euid != current->euid)
 846                 {
 847                         current->mm->dumpable = suid_dumpable;
 848                         smp_wmb();
 849                 }
 850                 current->euid = euid;
 851         }
 852         current->fsuid = current->euid;
 853         if (suid != (uid_t) -1)
 854                 current->suid = suid;
 855
 856         key_fsuid_changed(current);
 857         proc_id_connector(current, PROC_EVENT_UID);
 858
 859         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 860 }
 861
 862 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 863 {
 864         int retval;
 865
 866         if (!(retval = put_user(current->uid, ruid)) &&
 867             !(retval = put_user(current->euid, euid)))
 868                 retval = put_user(current->suid, suid);
 869
 870         return retval;
 871 }
 872
 873 /*
 874  * Same as above, but for rgid, egid, sgid.
 875  */
 876 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 877 {
 878         int retval;
 879
 880         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 881         if (retval)
 882                 return retval;
 883
 884         if (!capable(CAP_SETGID)) {
 885                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 886                     (rgid != current->egid) && (rgid != current->sgid))
 887                         return -EPERM;
 888                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 889                     (egid != current->egid) && (egid != current->sgid))
 890                         return -EPERM;
 891                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 892                     (sgid != current->egid) && (sgid != current->sgid))
 893                         return -EPERM;
 894         }
 895         if (egid != (gid_t) -1) {
 896                 if (egid != current->egid)
 897                 {
 898                         current->mm->dumpable = suid_dumpable;
 899                         smp_wmb();
 900                 }
 901                 current->egid = egid;
 902         }
 903         current->fsgid = current->egid;
 904         if (rgid != (gid_t) -1)
 905                 current->gid = rgid;
 906         if (sgid != (gid_t) -1)
 907                 current->sgid = sgid;
 908
 909         key_fsgid_changed(current);
 910         proc_id_connector(current, PROC_EVENT_GID);
 911         return 0;
 912 }
 913
 914 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 915 {
 916         int retval;
 917
 918         if (!(retval = put_user(current->gid, rgid)) &&
 919             !(retval = put_user(current->egid, egid)))
 920                 retval = put_user(current->sgid, sgid);
 921
 922         return retval;
 923 }
 924
 925
 926 /*
 927  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 928  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 929  * whatever uid it wants to). It normally shadows "euid", except when
 930  * explicitly set by setfsuid() or for access..
 931  */
 932 asmlinkage long sys_setfsuid(uid_t uid)
 933 {
 934         int old_fsuid;
 935
 936         old_fsuid = current->fsuid;
 937         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 938                 return old_fsuid;
 939
 940         if (uid == current->uid || uid == current->euid ||
 941             uid == current->suid || uid == current->fsuid ||
 942             capable(CAP_SETUID))
 943         {
 944                 if (uid != old_fsuid)
 945                 {
 946                         current->mm->dumpable = suid_dumpable;
 947                         smp_wmb();
 948                 }
 949                 current->fsuid = uid;
 950         }
 951
 952         key_fsuid_changed(current);
 953         proc_id_connector(current, PROC_EVENT_UID);
 954
 955         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 956
 957         return old_fsuid;
 958 }
 959
 960 /*
 961  * Samma på svenska..
 962  */
 963 asmlinkage long sys_setfsgid(gid_t gid)
 964 {
 965         int old_fsgid;
 966
 967         old_fsgid = current->fsgid;
 968         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 969                 return old_fsgid;
 970
 971         if (gid == current->gid || gid == current->egid ||
 972             gid == current->sgid || gid == current->fsgid ||
 973             capable(CAP_SETGID))
 974         {
 975                 if (gid != old_fsgid)
 976                 {
 977                         current->mm->dumpable = suid_dumpable;
 978                         smp_wmb();
 979                 }
 980                 current->fsgid = gid;
 981                 key_fsgid_changed(current);
 982                 proc_id_connector(current, PROC_EVENT_GID);
 983         }
 984         return old_fsgid;
 985 }
 986
 987 asmlinkage long sys_times(struct tms __user * tbuf)
 988 {
 989         /*
 990          *      In the SMP world we might just be unlucky and have one of
 991          *      the times increment as we use it. Since the value is an
 992          *      atomically safe type this is just fine. Conceptually its
 993          *      as if the syscall took an instant longer to occur.
 994          */
 995         if (tbuf) {
 996                 struct tms tmp;
 997                 cputime_t utime, stime, cutime, cstime;
 998
 999 #ifdef CONFIG_SMP
1000                 if (thread_group_empty(current)) {
1001                         /*
1002                          * Single thread case without the use of any locks.
1003                          *
1004                          * We may race with release_task if two threads are
1005                          * executing. However, release task first adds up the
1006                          * counters (__exit_signal) before  removing the task
1007                          * from the process tasklist (__unhash_process).
1008                          * __exit_signal also acquires and releases the
1009                          * siglock which results in the proper memory ordering
1010                          * so that the list modifications are always visible
1011                          * after the counters have been updated.
1012                          *
1013                          * If the counters have been updated by the second thread
1014                          * but the thread has not yet been removed from the list
1015                          * then the other branch will be executing which will
1016                          * block on tasklist_lock until the exit handling of the
1017                          * other task is finished.
1018                          *
1019                          * This also implies that the sighand->siglock cannot
1020                          * be held by another processor. So we can also
1021                          * skip acquiring that lock.
1022                          */
1023                         utime = cputime_add(current->signal->utime, current->utime);
1024                         stime = cputime_add(current->signal->utime, current->stime);
1025                         cutime = current->signal->cutime;
1026                         cstime = current->signal->cstime;
1027                 } else
1028 #endif
1029                 {
1030
1031                         /* Process with multiple threads */
1032                         struct task_struct *tsk = current;
1033                         struct task_struct *t;
1034
1035                         read_lock(&tasklist_lock);
1036                         utime = tsk->signal->utime;
1037                         stime = tsk->signal->stime;
1038                         t = tsk;
1039                         do {
1040                                 utime = cputime_add(utime, t->utime);
1041                                 stime = cputime_add(stime, t->stime);
1042                                 t = next_thread(t);
1043                         } while (t != tsk);
1044
1045                         /*
1046                          * While we have tasklist_lock read-locked, no dying thread
1047                          * can be updating current->signal->[us]time.  Instead,
1048                          * we got their counts included in the live thread loop.
1049                          * However, another thread can come in right now and
1050                          * do a wait call that updates current->signal->c[us]time.
1051                          * To make sure we always see that pair updated atomically,
1052                          * we take the siglock around fetching them.
1053                          */
1054                         spin_lock_irq(&tsk->sighand->siglock);
1055                         cutime = tsk->signal->cutime;
1056                         cstime = tsk->signal->cstime;
1057                         spin_unlock_irq(&tsk->sighand->siglock);
1058                         read_unlock(&tasklist_lock);
1059                 }
1060                 tmp.tms_utime = cputime_to_clock_t(utime);
1061                 tmp.tms_stime = cputime_to_clock_t(stime);
1062                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1063                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1064                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1065                         return -EFAULT;
1066         }
1067         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1068 }
1069
1070 /*
1071  * This needs some heavy checking ...
1072  * I just haven't the stomach for it. I also don't fully
1073  * understand sessions/pgrp etc. Let somebody who does explain it.
1074  *
1075  * OK, I think I have the protection semantics right.... this is really
1076  * only important on a multi-user system anyway, to make sure one user
1077  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1078  *
1079  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1080  * LBT 04.03.94
1081  */
1082
1083 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1084 {
1085         struct task_struct *p;
1086         int err = -EINVAL;
1087
1088         if (!pid)
1089                 pid = current->pid;
1090         if (!pgid)
1091                 pgid = pid;
1092         if (pgid < 0)
1093                 return -EINVAL;
1094
1095         /* From this point forward we keep holding onto the tasklist lock
1096          * so that our parent does not change from under us. -DaveM
1097          */
1098         write_lock_irq(&tasklist_lock);
1099
1100         err = -ESRCH;
1101         p = find_task_by_pid(pid);
1102         if (!p)
1103                 goto out;
1104
1105         err = -EINVAL;
1106         if (!thread_group_leader(p))
1107                 goto out;
1108
1109         if (p->parent == current || p->real_parent == current) {
1110                 err = -EPERM;
1111                 if (p->signal->session != current->signal->session)
1112                         goto out;
1113                 err = -EACCES;
1114                 if (p->did_exec)
1115                         goto out;
1116         } else {
1117                 err = -ESRCH;
1118                 if (p != current)
1119                         goto out;
1120         }
1121
1122         err = -EPERM;
1123         if (p->signal->leader)
1124                 goto out;
1125
1126         if (pgid != pid) {
1127                 struct task_struct *p;
1128
1129                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1130                         if (p->signal->session == current->signal->session)
1131                                 goto ok_pgid;
1132                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1133                 goto out;
1134         }
1135
1136 ok_pgid:
1137         err = security_task_setpgid(p, pgid);
1138         if (err)
1139                 goto out;
1140
1141         if (process_group(p) != pgid) {
1142                 detach_pid(p, PIDTYPE_PGID);
1143                 p->signal->pgrp = pgid;
1144                 attach_pid(p, PIDTYPE_PGID, pgid);
1145         }
1146
1147         err = 0;
1148 out:
1149         /* All paths lead to here, thus we are safe. -DaveM */
1150         write_unlock_irq(&tasklist_lock);
1151         return err;
1152 }
1153
1154 asmlinkage long sys_getpgid(pid_t pid)
1155 {
1156         if (!pid) {
1157                 return process_group(current);
1158         } else {
1159                 int retval;
1160                 struct task_struct *p;
1161
1162                 read_lock(&tasklist_lock);
1163                 p = find_task_by_pid(pid);
1164
1165                 retval = -ESRCH;
1166                 if (p) {
1167                         retval = security_task_getpgid(p);
1168                         if (!retval)
1169                                 retval = process_group(p);
1170                 }
1171                 read_unlock(&tasklist_lock);
1172                 return retval;
1173         }
1174 }
1175
1176 #ifdef __ARCH_WANT_SYS_GETPGRP
1177
1178 asmlinkage long sys_getpgrp(void)
1179 {
1180         /* SMP - assuming writes are word atomic this is fine */
1181         return process_group(current);
1182 }
1183
1184 #endif
1185
1186 asmlinkage long sys_getsid(pid_t pid)
1187 {
1188         if (!pid) {
1189                 return current->signal->session;
1190         } else {
1191                 int retval;
1192                 struct task_struct *p;
1193
1194                 read_lock(&tasklist_lock);
1195                 p = find_task_by_pid(pid);
1196
1197                 retval = -ESRCH;
1198                 if(p) {
1199                         retval = security_task_getsid(p);
1200                         if (!retval)
1201                                 retval = p->signal->session;
1202                 }
1203                 read_unlock(&tasklist_lock);
1204                 return retval;
1205         }
1206 }
1207
1208 asmlinkage long sys_setsid(void)
1209 {
1210         struct pid *pid;
1211         int err = -EPERM;
1212
1213         if (!thread_group_leader(current))
1214                 return -EINVAL;
1215
1216         down(&tty_sem);
1217         write_lock_irq(&tasklist_lock);
1218
1219         pid = find_pid(PIDTYPE_PGID, current->pid);
1220         if (pid)
1221                 goto out;
1222
1223         current->signal->leader = 1;
1224         __set_special_pids(current->pid, current->pid);
1225         current->signal->tty = NULL;
1226         current->signal->tty_old_pgrp = 0;
1227         err = process_group(current);
1228 out:
1229         write_unlock_irq(&tasklist_lock);
1230         up(&tty_sem);
1231         return err;
1232 }
1233
1234 /*
1235  * Supplementary group IDs
1236  */
1237
1238 /* init to 2 - one for init_task, one to ensure it is never freed */
1239 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1240
1241 struct group_info *groups_alloc(int gidsetsize)
1242 {
1243         struct group_info *group_info;
1244         int nblocks;
1245         int i;
1246
1247         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1248         /* Make sure we always allocate at least one indirect block pointer */
1249         nblocks = nblocks ? : 1;
1250         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1251         if (!group_info)
1252                 return NULL;
1253         group_info->ngroups = gidsetsize;
1254         group_info->nblocks = nblocks;
1255         atomic_set(&group_info->usage, 1);
1256
1257         if (gidsetsize <= NGROUPS_SMALL) {
1258                 group_info->blocks[0] = group_info->small_block;
1259         } else {
1260                 for (i = 0; i < nblocks; i++) {
1261                         gid_t *b;
1262                         b = (void *)__get_free_page(GFP_USER);
1263                         if (!b)
1264                                 goto out_undo_partial_alloc;
1265                         group_info->blocks[i] = b;
1266                 }
1267         }
1268         return group_info;
1269
1270 out_undo_partial_alloc:
1271         while (--i >= 0) {
1272                 free_page((unsigned long)group_info->blocks[i]);
1273         }
1274         kfree(group_info);
1275         return NULL;
1276 }
1277
1278 EXPORT_SYMBOL(groups_alloc);
1279
1280 void groups_free(struct group_info *group_info)
1281 {
1282         if (group_info->blocks[0] != group_info->small_block) {
1283                 int i;
1284                 for (i = 0; i < group_info->nblocks; i++)
1285                         free_page((unsigned long)group_info->blocks[i]);
1286         }
1287         kfree(group_info);
1288 }
1289
1290 EXPORT_SYMBOL(groups_free);
1291
1292 /* export the group_info to a user-space array */
1293 static int groups_to_user(gid_t __user *grouplist,
1294     struct group_info *group_info)
1295 {
1296         int i;
1297         int count = group_info->ngroups;
1298
1299         for (i = 0; i < group_info->nblocks; i++) {
1300                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1301                 int off = i * NGROUPS_PER_BLOCK;
1302                 int len = cp_count * sizeof(*grouplist);
1303
1304                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1305                         return -EFAULT;
1306
1307                 count -= cp_count;
1308         }
1309         return 0;
1310 }
1311
1312 /* fill a group_info from a user-space array - it must be allocated already */
1313 static int groups_from_user(struct group_info *group_info,
1314     gid_t __user *grouplist)
1315  {
1316         int i;
1317         int count = group_info->ngroups;
1318
1319         for (i = 0; i < group_info->nblocks; i++) {
1320                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1321                 int off = i * NGROUPS_PER_BLOCK;
1322                 int len = cp_count * sizeof(*grouplist);
1323
1324                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1325                         return -EFAULT;
1326
1327                 count -= cp_count;
1328         }
1329         return 0;
1330 }
1331
1332 /* a simple Shell sort */
1333 static void groups_sort(struct group_info *group_info)
1334 {
1335         int base, max, stride;
1336         int gidsetsize = group_info->ngroups;
1337
1338         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1339                 ; /* nothing */
1340         stride /= 3;
1341
1342         while (stride) {
1343                 max = gidsetsize - stride;
1344                 for (base = 0; base < max; base++) {
1345                         int left = base;
1346                         int right = left + stride;
1347                         gid_t tmp = GROUP_AT(group_info, right);
1348
1349                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1350                                 GROUP_AT(group_info, right) =
1351                                     GROUP_AT(group_info, left);
1352                                 right = left;
1353                                 left -= stride;
1354                         }
1355                         GROUP_AT(group_info, right) = tmp;
1356                 }
1357                 stride /= 3;
1358         }
1359 }
1360
1361 /* a simple bsearch */
1362 int groups_search(struct group_info *group_info, gid_t grp)
1363 {
1364         int left, right;
1365
1366         if (!group_info)
1367                 return 0;
1368
1369         left = 0;
1370         right = group_info->ngroups;
1371         while (left < right) {
1372                 int mid = (left+right)/2;
1373                 int cmp = grp - GROUP_AT(group_info, mid);
1374                 if (cmp > 0)
1375                         left = mid + 1;
1376                 else if (cmp < 0)
1377                         right = mid;
1378                 else
1379                         return 1;
1380         }
1381         return 0;
1382 }
1383
1384 /* validate and set current->group_info */
1385 int set_current_groups(struct group_info *group_info)
1386 {
1387         int retval;
1388         struct group_info *old_info;
1389
1390         retval = security_task_setgroups(group_info);
1391         if (retval)
1392                 return retval;
1393
1394         groups_sort(group_info);
1395         get_group_info(group_info);
1396
1397         task_lock(current);
1398         old_info = current->group_info;
1399         current->group_info = group_info;
1400         task_unlock(current);
1401
1402         put_group_info(old_info);
1403
1404         return 0;
1405 }
1406
1407 EXPORT_SYMBOL(set_current_groups);
1408
1409 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1410 {
1411         int i = 0;
1412
1413         /*
1414          *      SMP: Nobody else can change our grouplist. Thus we are
1415          *      safe.
1416          */
1417
1418         if (gidsetsize < 0)
1419                 return -EINVAL;
1420
1421         /* no need to grab task_lock here; it cannot change */
1422         get_group_info(current->group_info);
1423         i = current->group_info->ngroups;
1424         if (gidsetsize) {
1425                 if (i > gidsetsize) {
1426                         i = -EINVAL;
1427                         goto out;
1428                 }
1429                 if (groups_to_user(grouplist, current->group_info)) {
1430                         i = -EFAULT;
1431                         goto out;
1432                 }
1433         }
1434 out:
1435         put_group_info(current->group_info);
1436         return i;
1437 }
1438
1439 /*
1440  *      SMP: Our groups are copy-on-write. We can set them safely
1441  *      without another task interfering.
1442  */
1443
1444 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1445 {
1446         struct group_info *group_info;
1447         int retval;
1448
1449         if (!capable(CAP_SETGID))
1450                 return -EPERM;
1451         if ((unsigned)gidsetsize > NGROUPS_MAX)
1452                 return -EINVAL;
1453
1454         group_info = groups_alloc(gidsetsize);
1455         if (!group_info)
1456                 return -ENOMEM;
1457         retval = groups_from_user(group_info, grouplist);
1458         if (retval) {
1459                 put_group_info(group_info);
1460                 return retval;
1461         }
1462
1463         retval = set_current_groups(group_info);
1464         put_group_info(group_info);
1465
1466         return retval;
1467 }
1468
1469 /*
1470  * Check whether we're fsgid/egid or in the supplemental group..
1471  */
1472 int in_group_p(gid_t grp)
1473 {
1474         int retval = 1;
1475         if (grp != current->fsgid) {
1476                 get_group_info(current->group_info);
1477                 retval = groups_search(current->group_info, grp);
1478                 put_group_info(current->group_info);
1479         }
1480         return retval;
1481 }
1482
1483 EXPORT_SYMBOL(in_group_p);
1484
1485 int in_egroup_p(gid_t grp)
1486 {
1487         int retval = 1;
1488         if (grp != current->egid) {
1489                 get_group_info(current->group_info);
1490                 retval = groups_search(current->group_info, grp);
1491                 put_group_info(current->group_info);
1492         }
1493         return retval;
1494 }
1495
1496 EXPORT_SYMBOL(in_egroup_p);
1497
1498 DECLARE_RWSEM(uts_sem);
1499
1500 asmlinkage long sys_newuname(struct new_utsname __user * name)
1501 {
1502         int errno = 0;
1503
1504         down_read(&uts_sem);
1505         if (copy_to_user(name,&system_utsname,sizeof *name))
1506                 errno = -EFAULT;
1507         up_read(&uts_sem);
1508         return errno;
1509 }
1510
1511 asmlinkage long sys_sethostname(char __user *name, int len)
1512 {
1513         int errno;
1514         char tmp[__NEW_UTS_LEN];
1515
1516         if (!capable(CAP_SYS_ADMIN))
1517                 return -EPERM;
1518         if (len < 0 || len > __NEW_UTS_LEN)
1519                 return -EINVAL;
1520         down_write(&uts_sem);
1521         errno = -EFAULT;
1522         if (!copy_from_user(tmp, name, len)) {
1523                 memcpy(system_utsname.nodename, tmp, len);
1524                 system_utsname.nodename[len] = 0;
1525                 errno = 0;
1526         }
1527         up_write(&uts_sem);
1528         return errno;
1529 }
1530
1531 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1532
1533 asmlinkage long sys_gethostname(char __user *name, int len)
1534 {
1535         int i, errno;
1536
1537         if (len < 0)
1538                 return -EINVAL;
1539         down_read(&uts_sem);
1540         i = 1 + strlen(system_utsname.nodename);
1541         if (i > len)
1542                 i = len;
1543         errno = 0;
1544         if (copy_to_user(name, system_utsname.nodename, i))
1545                 errno = -EFAULT;
1546         up_read(&uts_sem);
1547         return errno;
1548 }
1549
1550 #endif
1551
1552 /*
1553  * Only setdomainname; getdomainname can be implemented by calling
1554  * uname()
1555  */
1556 asmlinkage long sys_setdomainname(char __user *name, int len)
1557 {
1558         int errno;
1559         char tmp[__NEW_UTS_LEN];
1560
1561         if (!capable(CAP_SYS_ADMIN))
1562                 return -EPERM;
1563         if (len < 0 || len > __NEW_UTS_LEN)
1564                 return -EINVAL;
1565
1566         down_write(&uts_sem);
1567         errno = -EFAULT;
1568         if (!copy_from_user(tmp, name, len)) {
1569                 memcpy(system_utsname.domainname, tmp, len);
1570                 system_utsname.domainname[len] = 0;
1571                 errno = 0;
1572         }
1573         up_write(&uts_sem);
1574         return errno;
1575 }
1576
1577 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1578 {
1579         if (resource >= RLIM_NLIMITS)
1580                 return -EINVAL;
1581         else {
1582                 struct rlimit value;
1583                 task_lock(current->group_leader);
1584                 value = current->signal->rlim[resource];
1585                 task_unlock(current->group_leader);
1586                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1587         }
1588 }
1589
1590 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1591
1592 /*
1593  *      Back compatibility for getrlimit. Needed for some apps.
1594  */
1595
1596 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1597 {
1598         struct rlimit x;
1599         if (resource >= RLIM_NLIMITS)
1600                 return -EINVAL;
1601
1602         task_lock(current->group_leader);
1603         x = current->signal->rlim[resource];
1604         task_unlock(current->group_leader);
1605         if(x.rlim_cur > 0x7FFFFFFF)
1606                 x.rlim_cur = 0x7FFFFFFF;
1607         if(x.rlim_max > 0x7FFFFFFF)
1608                 x.rlim_max = 0x7FFFFFFF;
1609         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1610 }
1611
1612 #endif
1613
1614 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1615 {
1616         struct rlimit new_rlim, *old_rlim;
1617         int retval;
1618
1619         if (resource >= RLIM_NLIMITS)
1620                 return -EINVAL;
1621         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1622                 return -EFAULT;
1623        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1624                return -EINVAL;
1625         old_rlim = current->signal->rlim + resource;
1626         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1627             !capable(CAP_SYS_RESOURCE))
1628                 return -EPERM;
1629         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1630                         return -EPERM;
1631
1632         retval = security_task_setrlimit(resource, &new_rlim);
1633         if (retval)
1634                 return retval;
1635
1636         task_lock(current->group_leader);
1637         *old_rlim = new_rlim;
1638         task_unlock(current->group_leader);
1639
1640         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1641             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1642              new_rlim.rlim_cur <= cputime_to_secs(
1643                      current->signal->it_prof_expires))) {
1644                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1645                 read_lock(&tasklist_lock);
1646                 spin_lock_irq(&current->sighand->siglock);
1647                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1648                                       &cputime, NULL);
1649                 spin_unlock_irq(&current->sighand->siglock);
1650                 read_unlock(&tasklist_lock);
1651         }
1652
1653         return 0;
1654 }
1655
1656 /*
1657  * It would make sense to put struct rusage in the task_struct,
1658  * except that would make the task_struct be *really big*.  After
1659  * task_struct gets moved into malloc'ed memory, it would
1660  * make sense to do this.  It will make moving the rest of the information
1661  * a lot simpler!  (Which we're not doing right now because we're not
1662  * measuring them yet).
1663  *
1664  * This expects to be called with tasklist_lock read-locked or better,
1665  * and the siglock not locked.  It may momentarily take the siglock.
1666  *
1667  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1668  * races with threads incrementing their own counters.  But since word
1669  * reads are atomic, we either get new values or old values and we don't
1670  * care which for the sums.  We always take the siglock to protect reading
1671  * the c* fields from p->signal from races with exit.c updating those
1672  * fields when reaping, so a sample either gets all the additions of a
1673  * given child after it's reaped, or none so this sample is before reaping.
1674  */
1675
1676 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1677 {
1678         struct task_struct *t;
1679         unsigned long flags;
1680         cputime_t utime, stime;
1681
1682         memset((char *) r, 0, sizeof *r);
1683
1684         if (unlikely(!p->signal))
1685                 return;
1686
1687         switch (who) {
1688                 case RUSAGE_CHILDREN:
1689                         spin_lock_irqsave(&p->sighand->siglock, flags);
1690                         utime = p->signal->cutime;
1691                         stime = p->signal->cstime;
1692                         r->ru_nvcsw = p->signal->cnvcsw;
1693                         r->ru_nivcsw = p->signal->cnivcsw;
1694                         r->ru_minflt = p->signal->cmin_flt;
1695                         r->ru_majflt = p->signal->cmaj_flt;
1696                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1697                         cputime_to_timeval(utime, &r->ru_utime);
1698                         cputime_to_timeval(stime, &r->ru_stime);
1699                         break;
1700                 case RUSAGE_SELF:
1701                         spin_lock_irqsave(&p->sighand->siglock, flags);
1702                         utime = stime = cputime_zero;
1703                         goto sum_group;
1704                 case RUSAGE_BOTH:
1705                         spin_lock_irqsave(&p->sighand->siglock, flags);
1706                         utime = p->signal->cutime;
1707                         stime = p->signal->cstime;
1708                         r->ru_nvcsw = p->signal->cnvcsw;
1709                         r->ru_nivcsw = p->signal->cnivcsw;
1710                         r->ru_minflt = p->signal->cmin_flt;
1711                         r->ru_majflt = p->signal->cmaj_flt;
1712                 sum_group:
1713                         utime = cputime_add(utime, p->signal->utime);
1714                         stime = cputime_add(stime, p->signal->stime);
1715                         r->ru_nvcsw += p->signal->nvcsw;
1716                         r->ru_nivcsw += p->signal->nivcsw;
1717                         r->ru_minflt += p->signal->min_flt;
1718                         r->ru_majflt += p->signal->maj_flt;
1719                         t = p;
1720                         do {
1721                                 utime = cputime_add(utime, t->utime);
1722                                 stime = cputime_add(stime, t->stime);
1723                                 r->ru_nvcsw += t->nvcsw;
1724                                 r->ru_nivcsw += t->nivcsw;
1725                                 r->ru_minflt += t->min_flt;
1726                                 r->ru_majflt += t->maj_flt;
1727                                 t = next_thread(t);
1728                         } while (t != p);
1729                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1730                         cputime_to_timeval(utime, &r->ru_utime);
1731                         cputime_to_timeval(stime, &r->ru_stime);
1732                         break;
1733                 default:
1734                         BUG();
1735         }
1736 }
1737
1738 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1739 {
1740         struct rusage r;
1741         read_lock(&tasklist_lock);
1742         k_getrusage(p, who, &r);
1743         read_unlock(&tasklist_lock);
1744         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1745 }
1746
1747 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1748 {
1749         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1750                 return -EINVAL;
1751         return getrusage(current, who, ru);
1752 }
1753
1754 asmlinkage long sys_umask(int mask)
1755 {
1756         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1757         return mask;
1758 }
1759
1760 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1761                           unsigned long arg4, unsigned long arg5)
1762 {
1763         long error;
1764
1765         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1766         if (error)
1767                 return error;
1768
1769         switch (option) {
1770                 case PR_SET_PDEATHSIG:
1771                         if (!valid_signal(arg2)) {
1772                                 error = -EINVAL;
1773                                 break;
1774                         }
1775                         current->pdeath_signal = arg2;
1776                         break;
1777                 case PR_GET_PDEATHSIG:
1778                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1779                         break;
1780                 case PR_GET_DUMPABLE:
1781                         error = current->mm->dumpable;
1782                         break;
1783                 case PR_SET_DUMPABLE:
1784                         if (arg2 < 0 || arg2 > 2) {
1785                                 error = -EINVAL;
1786                                 break;
1787                         }
1788                         current->mm->dumpable = arg2;
1789                         break;
1790
1791                 case PR_SET_UNALIGN:
1792                         error = SET_UNALIGN_CTL(current, arg2);
1793                         break;
1794                 case PR_GET_UNALIGN:
1795                         error = GET_UNALIGN_CTL(current, arg2);
1796                         break;
1797                 case PR_SET_FPEMU:
1798                         error = SET_FPEMU_CTL(current, arg2);
1799                         break;
1800                 case PR_GET_FPEMU:
1801                         error = GET_FPEMU_CTL(current, arg2);
1802                         break;
1803                 case PR_SET_FPEXC:
1804                         error = SET_FPEXC_CTL(current, arg2);
1805                         break;
1806                 case PR_GET_FPEXC:
1807                         error = GET_FPEXC_CTL(current, arg2);
1808                         break;
1809                 case PR_GET_TIMING:
1810                         error = PR_TIMING_STATISTICAL;
1811                         break;
1812                 case PR_SET_TIMING:
1813                         if (arg2 == PR_TIMING_STATISTICAL)
1814                                 error = 0;
1815                         else
1816                                 error = -EINVAL;
1817                         break;
1818
1819                 case PR_GET_KEEPCAPS:
1820                         if (current->keep_capabilities)
1821                                 error = 1;
1822                         break;
1823                 case PR_SET_KEEPCAPS:
1824                         if (arg2 != 0 && arg2 != 1) {
1825                                 error = -EINVAL;
1826                                 break;
1827                         }
1828                         current->keep_capabilities = arg2;
1829                         break;
1830                 case PR_SET_NAME: {
1831                         struct task_struct *me = current;
1832                         unsigned char ncomm[sizeof(me->comm)];
1833
1834                         ncomm[sizeof(me->comm)-1] = 0;
1835                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1836                                                 sizeof(me->comm)-1) < 0)
1837                                 return -EFAULT;
1838                         set_task_comm(me, ncomm);
1839                         return 0;
1840                 }
1841                 case PR_GET_NAME: {
1842                         struct task_struct *me = current;
1843                         unsigned char tcomm[sizeof(me->comm)];
1844
1845                         get_task_comm(tcomm, me);
1846                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1847                                 return -EFAULT;
1848                         return 0;
1849                 }
1850                 default:
1851                         error = -EINVAL;
1852                         break;
1853         }
1854         return error;
1855 }