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/kmod.h>
  15 #include <linux/reboot.h>
  16 #include <linux/prctl.h>
  17 #include <linux/init.h>
  18 #include <linux/highuid.h>
  19 #include <linux/fs.h>
  20 #include <linux/kernel.h>
  21 #include <linux/kexec.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/capability.h>
  24 #include <linux/device.h>
  25 #include <linux/key.h>
  26 #include <linux/times.h>
  27 #include <linux/posix-timers.h>
  28 #include <linux/security.h>
  29 #include <linux/dcookies.h>
  30 #include <linux/suspend.h>
  31 #include <linux/tty.h>
  32 #include <linux/signal.h>
  33 #include <linux/cn_proc.h>
  34 #include <linux/vs_base.h>
  35 #include <linux/vs_cvirt.h>
  36
  37 #include <linux/compat.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/kprobes.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43 #include <asm/unistd.h>
  44
  45 #ifndef SET_UNALIGN_CTL
  46 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  47 #endif
  48 #ifndef GET_UNALIGN_CTL
  49 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  50 #endif
  51 #ifndef SET_FPEMU_CTL
  52 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  53 #endif
  54 #ifndef GET_FPEMU_CTL
  55 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  56 #endif
  57 #ifndef SET_FPEXC_CTL
  58 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  59 #endif
  60 #ifndef GET_FPEXC_CTL
  61 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  62 #endif
  63
  64 /*
  65  * this is where the system-wide overflow UID and GID are defined, for
  66  * architectures that now have 32-bit UID/GID but didn't in the past
  67  */
  68
  69 int overflowuid = DEFAULT_OVERFLOWUID;
  70 int overflowgid = DEFAULT_OVERFLOWGID;
  71
  72 #ifdef CONFIG_UID16
  73 EXPORT_SYMBOL(overflowuid);
  74 EXPORT_SYMBOL(overflowgid);
  75 #endif
  76
  77 /*
  78  * the same as above, but for filesystems which can only store a 16-bit
  79  * UID and GID. as such, this is needed on all architectures
  80  */
  81
  82 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  83 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  84
  85 EXPORT_SYMBOL(fs_overflowuid);
  86 EXPORT_SYMBOL(fs_overflowgid);
  87
  88 /*
  89  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
  90  */
  91
  92 int C_A_D = 1;
  93 int cad_pid = 1;
  94
  95 /*
  96  *      Notifier list for kernel code which wants to be called
  97  *      at shutdown. This is used to stop any idling DMA operations
  98  *      and the like.
  99  */
 100
 101 static struct notifier_block *reboot_notifier_list;
 102 static DEFINE_RWLOCK(notifier_lock);
 103
 104 /**
 105  *      notifier_chain_register - Add notifier to a notifier chain
 106  *      @list: Pointer to root list pointer
 107  *      @n: New entry in notifier chain
 108  *
 109  *      Adds a notifier to a notifier chain.
 110  *
 111  *      Currently always returns zero.
 112  */
 113
 114 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
 115 {
 116         write_lock(&notifier_lock);
 117         while(*list)
 118         {
 119                 if(n->priority > (*list)->priority)
 120                         break;
 121                 list= &((*list)->next);
 122         }
 123         n->next = *list;
 124         *list=n;
 125         write_unlock(&notifier_lock);
 126         return 0;
 127 }
 128
 129 EXPORT_SYMBOL(notifier_chain_register);
 130
 131 /**
 132  *      notifier_chain_unregister - Remove notifier from a notifier chain
 133  *      @nl: Pointer to root list pointer
 134  *      @n: New entry in notifier chain
 135  *
 136  *      Removes a notifier from a notifier chain.
 137  *
 138  *      Returns zero on success, or %-ENOENT on failure.
 139  */
 140
 141 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
 142 {
 143         write_lock(&notifier_lock);
 144         while((*nl)!=NULL)
 145         {
 146                 if((*nl)==n)
 147                 {
 148                         *nl=n->next;
 149                         write_unlock(&notifier_lock);
 150                         return 0;
 151                 }
 152                 nl=&((*nl)->next);
 153         }
 154         write_unlock(&notifier_lock);
 155         return -ENOENT;
 156 }
 157
 158 EXPORT_SYMBOL(notifier_chain_unregister);
 159
 160 /**
 161  *      notifier_call_chain - Call functions in a notifier chain
 162  *      @n: Pointer to root pointer of notifier chain
 163  *      @val: Value passed unmodified to notifier function
 164  *      @v: Pointer passed unmodified to notifier function
 165  *
 166  *      Calls each function in a notifier chain in turn.
 167  *
 168  *      If the return value of the notifier can be and'd
 169  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
 170  *      will return immediately, with the return value of
 171  *      the notifier function which halted execution.
 172  *      Otherwise, the return value is the return value
 173  *      of the last notifier function called.
 174  */
 175
 176 int __kprobes notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
 177 {
 178         int ret=NOTIFY_DONE;
 179         struct notifier_block *nb = *n;
 180
 181         while(nb)
 182         {
 183                 ret=nb->notifier_call(nb,val,v);
 184                 if(ret&NOTIFY_STOP_MASK)
 185                 {
 186                         return ret;
 187                 }
 188                 nb=nb->next;
 189         }
 190         return ret;
 191 }
 192
 193 EXPORT_SYMBOL(notifier_call_chain);
 194
 195 /**
 196  *      register_reboot_notifier - Register function to be called at reboot time
 197  *      @nb: Info about notifier function to be called
 198  *
 199  *      Registers a function with the list of functions
 200  *      to be called at reboot time.
 201  *
 202  *      Currently always returns zero, as notifier_chain_register
 203  *      always returns zero.
 204  */
 205
 206 int register_reboot_notifier(struct notifier_block * nb)
 207 {
 208         return notifier_chain_register(&reboot_notifier_list, nb);
 209 }
 210
 211 EXPORT_SYMBOL(register_reboot_notifier);
 212
 213 /**
 214  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
 215  *      @nb: Hook to be unregistered
 216  *
 217  *      Unregisters a previously registered reboot
 218  *      notifier function.
 219  *
 220  *      Returns zero on success, or %-ENOENT on failure.
 221  */
 222
 223 int unregister_reboot_notifier(struct notifier_block * nb)
 224 {
 225         return notifier_chain_unregister(&reboot_notifier_list, nb);
 226 }
 227
 228 EXPORT_SYMBOL(unregister_reboot_notifier);
 229
 230 #ifndef CONFIG_SECURITY
 231 int capable(int cap)
 232 {
 233         if (vx_check_bit(VXC_CAP_MASK, cap) && !vx_mcaps(1L << cap))
 234                 return 0;
 235         if (cap_raised(current->cap_effective, cap)) {
 236                current->flags |= PF_SUPERPRIV;
 237                return 1;
 238         }
 239         return 0;
 240 }
 241 EXPORT_SYMBOL(capable);
 242 #endif
 243
 244 static int set_one_prio(struct task_struct *p, int niceval, int error)
 245 {
 246         int no_nice;
 247
 248         if (p->uid != current->euid &&
 249                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
 250                 error = -EPERM;
 251                 goto out;
 252         }
 253         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 254                 if (vx_flags(VXF_IGNEG_NICE, 0))
 255                         error = 0;
 256                 else
 257                         error = -EACCES;
 258                 goto out;
 259         }
 260         no_nice = security_task_setnice(p, niceval);
 261         if (no_nice) {
 262                 error = no_nice;
 263                 goto out;
 264         }
 265         if (error == -ESRCH)
 266                 error = 0;
 267         set_user_nice(p, niceval);
 268 out:
 269         return error;
 270 }
 271
 272 asmlinkage long sys_setpriority(int which, int who, int niceval)
 273 {
 274         struct task_struct *g, *p;
 275         struct user_struct *user;
 276         int error = -EINVAL;
 277
 278         if (which > 2 || which < 0)
 279                 goto out;
 280
 281         /* normalize: avoid signed division (rounding problems) */
 282         error = -ESRCH;
 283         if (niceval < -20)
 284                 niceval = -20;
 285         if (niceval > 19)
 286                 niceval = 19;
 287
 288         read_lock(&tasklist_lock);
 289         switch (which) {
 290                 case PRIO_PROCESS:
 291                         if (!who)
 292                                 who = current->pid;
 293                         p = find_task_by_pid(who);
 294                         if (p)
 295                                 error = set_one_prio(p, niceval, error);
 296                         break;
 297                 case PRIO_PGRP:
 298                         if (!who)
 299                                 who = process_group(current);
 300                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 301                                 error = set_one_prio(p, niceval, error);
 302                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 303                         break;
 304                 case PRIO_USER:
 305                         user = current->user;
 306                         if (!who)
 307                                 who = current->uid;
 308                         else
 309                                 if ((who != current->uid) &&
 310                                         !(user = find_user(vx_current_xid(), who)))
 311                                         goto out_unlock;        /* No processes for this user */
 312
 313                         do_each_thread(g, p)
 314                                 if (p->uid == who)
 315                                         error = set_one_prio(p, niceval, error);
 316                         while_each_thread(g, p);
 317                         if (who != current->uid)
 318                                 free_uid(user);         /* For find_user() */
 319                         break;
 320         }
 321 out_unlock:
 322         read_unlock(&tasklist_lock);
 323 out:
 324         return error;
 325 }
 326
 327 /*
 328  * Ugh. To avoid negative return values, "getpriority()" will
 329  * not return the normal nice-value, but a negated value that
 330  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 331  * to stay compatible.
 332  */
 333 asmlinkage long sys_getpriority(int which, int who)
 334 {
 335         struct task_struct *g, *p;
 336         struct user_struct *user;
 337         long niceval, retval = -ESRCH;
 338
 339         if (which > 2 || which < 0)
 340                 return -EINVAL;
 341
 342         read_lock(&tasklist_lock);
 343         switch (which) {
 344                 case PRIO_PROCESS:
 345                         if (!who)
 346                                 who = current->pid;
 347                         p = find_task_by_pid(who);
 348                         if (p) {
 349                                 niceval = 20 - task_nice(p);
 350                                 if (niceval > retval)
 351                                         retval = niceval;
 352                         }
 353                         break;
 354                 case PRIO_PGRP:
 355                         if (!who)
 356                                 who = process_group(current);
 357                         do_each_task_pid(who, PIDTYPE_PGID, p) {
 358                                 niceval = 20 - task_nice(p);
 359                                 if (niceval > retval)
 360                                         retval = niceval;
 361                         } while_each_task_pid(who, PIDTYPE_PGID, p);
 362                         break;
 363                 case PRIO_USER:
 364                         user = current->user;
 365                         if (!who)
 366                                 who = current->uid;
 367                         else
 368                                 if ((who != current->uid) &&
 369                                         !(user = find_user(vx_current_xid(), who)))
 370                                         goto out_unlock;        /* No processes for this user */
 371
 372                         do_each_thread(g, p)
 373                                 if (p->uid == who) {
 374                                         niceval = 20 - task_nice(p);
 375                                         if (niceval > retval)
 376                                                 retval = niceval;
 377                                 }
 378                         while_each_thread(g, p);
 379                         if (who != current->uid)
 380                                 free_uid(user);         /* for find_user() */
 381                         break;
 382         }
 383 out_unlock:
 384         read_unlock(&tasklist_lock);
 385
 386         return retval;
 387 }
 388
 389 /**
 390  *      emergency_restart - reboot the system
 391  *
 392  *      Without shutting down any hardware or taking any locks
 393  *      reboot the system.  This is called when we know we are in
 394  *      trouble so this is our best effort to reboot.  This is
 395  *      safe to call in interrupt context.
 396  */
 397 void emergency_restart(void)
 398 {
 399         machine_emergency_restart();
 400 }
 401 EXPORT_SYMBOL_GPL(emergency_restart);
 402
 403 void kernel_restart_prepare(char *cmd)
 404 {
 405         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 406         system_state = SYSTEM_RESTART;
 407         device_shutdown();
 408 }
 409
 410 /**
 411  *      kernel_restart - reboot the system
 412  *      @cmd: pointer to buffer containing command to execute for restart
 413  *              or %NULL
 414  *
 415  *      Shutdown everything and perform a clean reboot.
 416  *      This is not safe to call in interrupt context.
 417  */
 418 void kernel_restart(char *cmd)
 419 {
 420         kernel_restart_prepare(cmd);
 421         if (!cmd) {
 422                 printk(KERN_EMERG "Restarting system.\n");
 423         } else {
 424                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 425         }
 426         printk(".\n");
 427         machine_restart(cmd);
 428 }
 429 EXPORT_SYMBOL_GPL(kernel_restart);
 430
 431 /**
 432  *      kernel_kexec - reboot the system
 433  *
 434  *      Move into place and start executing a preloaded standalone
 435  *      executable.  If nothing was preloaded return an error.
 436  */
 437 void kernel_kexec(void)
 438 {
 439 #ifdef CONFIG_KEXEC
 440         struct kimage *image;
 441         image = xchg(&kexec_image, NULL);
 442         if (!image) {
 443                 return;
 444         }
 445         kernel_restart_prepare(NULL);
 446         printk(KERN_EMERG "Starting new kernel\n");
 447         machine_shutdown();
 448         machine_kexec(image);
 449 #endif
 450 }
 451 EXPORT_SYMBOL_GPL(kernel_kexec);
 452
 453 void kernel_shutdown_prepare(enum system_states state)
 454 {
 455         notifier_call_chain(&reboot_notifier_list,
 456                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 457         system_state = state;
 458         device_shutdown();
 459 }
 460 /**
 461  *      kernel_halt - halt the system
 462  *
 463  *      Shutdown everything and perform a clean system halt.
 464  */
 465 void kernel_halt(void)
 466 {
 467         kernel_shutdown_prepare(SYSTEM_HALT);
 468         printk(KERN_EMERG "System halted.\n");
 469         machine_halt();
 470 }
 471
 472 EXPORT_SYMBOL_GPL(kernel_halt);
 473
 474 /**
 475  *      kernel_power_off - power_off the system
 476  *
 477  *      Shutdown everything and perform a clean system power_off.
 478  */
 479 void kernel_power_off(void)
 480 {
 481         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 482         printk(KERN_EMERG "Power down.\n");
 483         machine_power_off();
 484 }
 485 EXPORT_SYMBOL_GPL(kernel_power_off);
 486
 487 long vs_reboot(unsigned int, void __user *);
 488
 489 /*
 490  * Reboot system call: for obvious reasons only root may call it,
 491  * and even root needs to set up some magic numbers in the registers
 492  * so that some mistake won't make this reboot the whole machine.
 493  * You can also set the meaning of the ctrl-alt-del-key here.
 494  *
 495  * reboot doesn't sync: do that yourself before calling this.
 496  */
 497 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 498 {
 499         char buffer[256];
 500
 501         /* We only trust the superuser with rebooting the system. */
 502         if (!capable(CAP_SYS_BOOT))
 503                 return -EPERM;
 504
 505         /* For safety, we require "magic" arguments. */
 506         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 507             (magic2 != LINUX_REBOOT_MAGIC2 &&
 508                         magic2 != LINUX_REBOOT_MAGIC2A &&
 509                         magic2 != LINUX_REBOOT_MAGIC2B &&
 510                         magic2 != LINUX_REBOOT_MAGIC2C))
 511                 return -EINVAL;
 512
 513         /* Instead of trying to make the power_off code look like
 514          * halt when pm_power_off is not set do it the easy way.
 515          */
 516         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 517                 cmd = LINUX_REBOOT_CMD_HALT;
 518
 519         if (!vx_check(0, VX_ADMIN|VX_WATCH))
 520                 return vs_reboot(cmd, arg);
 521
 522         lock_kernel();
 523         switch (cmd) {
 524         case LINUX_REBOOT_CMD_RESTART:
 525                 kernel_restart(NULL);
 526                 break;
 527
 528         case LINUX_REBOOT_CMD_CAD_ON:
 529                 C_A_D = 1;
 530                 break;
 531
 532         case LINUX_REBOOT_CMD_CAD_OFF:
 533                 C_A_D = 0;
 534                 break;
 535
 536         case LINUX_REBOOT_CMD_HALT:
 537                 kernel_halt();
 538                 unlock_kernel();
 539                 do_exit(0);
 540                 break;
 541
 542         case LINUX_REBOOT_CMD_POWER_OFF:
 543                 kernel_power_off();
 544                 unlock_kernel();
 545                 do_exit(0);
 546                 break;
 547
 548         case LINUX_REBOOT_CMD_RESTART2:
 549                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 550                         unlock_kernel();
 551                         return -EFAULT;
 552                 }
 553                 buffer[sizeof(buffer) - 1] = '\0';
 554
 555                 kernel_restart(buffer);
 556                 break;
 557
 558         case LINUX_REBOOT_CMD_KEXEC:
 559                 kernel_kexec();
 560                 unlock_kernel();
 561                 return -EINVAL;
 562
 563 #ifdef CONFIG_SOFTWARE_SUSPEND
 564         case LINUX_REBOOT_CMD_SW_SUSPEND:
 565                 {
 566                         int ret = software_suspend();
 567                         unlock_kernel();
 568                         return ret;
 569                 }
 570 #endif
 571
 572         default:
 573                 unlock_kernel();
 574                 return -EINVAL;
 575         }
 576         unlock_kernel();
 577         return 0;
 578 }
 579
 580 static void deferred_cad(void *dummy)
 581 {
 582         kernel_restart(NULL);
 583 }
 584
 585 /*
 586  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 587  * As it's called within an interrupt, it may NOT sync: the only choice
 588  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 589  */
 590 void ctrl_alt_del(void)
 591 {
 592         static DECLARE_WORK(cad_work, deferred_cad, NULL);
 593
 594         if (C_A_D)
 595                 schedule_work(&cad_work);
 596         else
 597                 kill_proc(cad_pid, SIGINT, 1);
 598 }
 599
 600
 601 /*
 602  * Unprivileged users may change the real gid to the effective gid
 603  * or vice versa.  (BSD-style)
 604  *
 605  * If you set the real gid at all, or set the effective gid to a value not
 606  * equal to the real gid, then the saved gid is set to the new effective gid.
 607  *
 608  * This makes it possible for a setgid program to completely drop its
 609  * privileges, which is often a useful assertion to make when you are doing
 610  * a security audit over a program.
 611  *
 612  * The general idea is that a program which uses just setregid() will be
 613  * 100% compatible with BSD.  A program which uses just setgid() will be
 614  * 100% compatible with POSIX with saved IDs.
 615  *
 616  * SMP: There are not races, the GIDs are checked only by filesystem
 617  *      operations (as far as semantic preservation is concerned).
 618  */
 619 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 620 {
 621         int old_rgid = current->gid;
 622         int old_egid = current->egid;
 623         int new_rgid = old_rgid;
 624         int new_egid = old_egid;
 625         int retval;
 626
 627         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 628         if (retval)
 629                 return retval;
 630
 631         if (rgid != (gid_t) -1) {
 632                 if ((old_rgid == rgid) ||
 633                     (current->egid==rgid) ||
 634                     capable(CAP_SETGID))
 635                         new_rgid = rgid;
 636                 else
 637                         return -EPERM;
 638         }
 639         if (egid != (gid_t) -1) {
 640                 if ((old_rgid == egid) ||
 641                     (current->egid == egid) ||
 642                     (current->sgid == egid) ||
 643                     capable(CAP_SETGID))
 644                         new_egid = egid;
 645                 else {
 646                         return -EPERM;
 647                 }
 648         }
 649         if (new_egid != old_egid)
 650         {
 651                 current->mm->dumpable = suid_dumpable;
 652                 smp_wmb();
 653         }
 654         if (rgid != (gid_t) -1 ||
 655             (egid != (gid_t) -1 && egid != old_rgid))
 656                 current->sgid = new_egid;
 657         current->fsgid = new_egid;
 658         current->egid = new_egid;
 659         current->gid = new_rgid;
 660         key_fsgid_changed(current);
 661         proc_id_connector(current, PROC_EVENT_GID);
 662         return 0;
 663 }
 664
 665 /*
 666  * setgid() is implemented like SysV w/ SAVED_IDS
 667  *
 668  * SMP: Same implicit races as above.
 669  */
 670 asmlinkage long sys_setgid(gid_t gid)
 671 {
 672         int old_egid = current->egid;
 673         int retval;
 674
 675         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 676         if (retval)
 677                 return retval;
 678
 679         if (capable(CAP_SETGID))
 680         {
 681                 if(old_egid != gid)
 682                 {
 683                         current->mm->dumpable = suid_dumpable;
 684                         smp_wmb();
 685                 }
 686                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 687         }
 688         else if ((gid == current->gid) || (gid == current->sgid))
 689         {
 690                 if(old_egid != gid)
 691                 {
 692                         current->mm->dumpable = suid_dumpable;
 693                         smp_wmb();
 694                 }
 695                 current->egid = current->fsgid = gid;
 696         }
 697         else
 698                 return -EPERM;
 699
 700         key_fsgid_changed(current);
 701         proc_id_connector(current, PROC_EVENT_GID);
 702         return 0;
 703 }
 704
 705 static int set_user(uid_t new_ruid, int dumpclear)
 706 {
 707         struct user_struct *new_user;
 708
 709         new_user = alloc_uid(vx_current_xid(), new_ruid);
 710         if (!new_user)
 711                 return -EAGAIN;
 712
 713         if (atomic_read(&new_user->processes) >=
 714                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 715                         new_user != &root_user) {
 716                 free_uid(new_user);
 717                 return -EAGAIN;
 718         }
 719
 720         switch_uid(new_user);
 721
 722         if(dumpclear)
 723         {
 724                 current->mm->dumpable = suid_dumpable;
 725                 smp_wmb();
 726         }
 727         current->uid = new_ruid;
 728         return 0;
 729 }
 730
 731 /*
 732  * Unprivileged users may change the real uid to the effective uid
 733  * or vice versa.  (BSD-style)
 734  *
 735  * If you set the real uid at all, or set the effective uid to a value not
 736  * equal to the real uid, then the saved uid is set to the new effective uid.
 737  *
 738  * This makes it possible for a setuid program to completely drop its
 739  * privileges, which is often a useful assertion to make when you are doing
 740  * a security audit over a program.
 741  *
 742  * The general idea is that a program which uses just setreuid() will be
 743  * 100% compatible with BSD.  A program which uses just setuid() will be
 744  * 100% compatible with POSIX with saved IDs.
 745  */
 746 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 747 {
 748         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 749         int retval;
 750
 751         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 752         if (retval)
 753                 return retval;
 754
 755         new_ruid = old_ruid = current->uid;
 756         new_euid = old_euid = current->euid;
 757         old_suid = current->suid;
 758
 759         if (ruid != (uid_t) -1) {
 760                 new_ruid = ruid;
 761                 if ((old_ruid != ruid) &&
 762                     (current->euid != ruid) &&
 763                     !capable(CAP_SETUID))
 764                         return -EPERM;
 765         }
 766
 767         if (euid != (uid_t) -1) {
 768                 new_euid = euid;
 769                 if ((old_ruid != euid) &&
 770                     (current->euid != euid) &&
 771                     (current->suid != euid) &&
 772                     !capable(CAP_SETUID))
 773                         return -EPERM;
 774         }
 775
 776         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 777                 return -EAGAIN;
 778
 779         if (new_euid != old_euid)
 780         {
 781                 current->mm->dumpable = suid_dumpable;
 782                 smp_wmb();
 783         }
 784         current->fsuid = current->euid = new_euid;
 785         if (ruid != (uid_t) -1 ||
 786             (euid != (uid_t) -1 && euid != old_ruid))
 787                 current->suid = current->euid;
 788         current->fsuid = current->euid;
 789
 790         key_fsuid_changed(current);
 791         proc_id_connector(current, PROC_EVENT_UID);
 792
 793         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 794 }
 795
 796
 797
 798 /*
 799  * setuid() is implemented like SysV with SAVED_IDS
 800  *
 801  * Note that SAVED_ID's is deficient in that a setuid root program
 802  * like sendmail, for example, cannot set its uid to be a normal
 803  * user and then switch back, because if you're root, setuid() sets
 804  * the saved uid too.  If you don't like this, blame the bright people
 805  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 806  * will allow a root program to temporarily drop privileges and be able to
 807  * regain them by swapping the real and effective uid.
 808  */
 809 asmlinkage long sys_setuid(uid_t uid)
 810 {
 811         int old_euid = current->euid;
 812         int old_ruid, old_suid, new_ruid, new_suid;
 813         int retval;
 814
 815         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 816         if (retval)
 817                 return retval;
 818
 819         old_ruid = new_ruid = current->uid;
 820         old_suid = current->suid;
 821         new_suid = old_suid;
 822
 823         if (capable(CAP_SETUID)) {
 824                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 825                         return -EAGAIN;
 826                 new_suid = uid;
 827         } else if ((uid != current->uid) && (uid != new_suid))
 828                 return -EPERM;
 829
 830         if (old_euid != uid)
 831         {
 832                 current->mm->dumpable = suid_dumpable;
 833                 smp_wmb();
 834         }
 835         current->fsuid = current->euid = uid;
 836         current->suid = new_suid;
 837
 838         key_fsuid_changed(current);
 839         proc_id_connector(current, PROC_EVENT_UID);
 840
 841         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 842 }
 843
 844
 845 /*
 846  * This function implements a generic ability to update ruid, euid,
 847  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 848  */
 849 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 850 {
 851         int old_ruid = current->uid;
 852         int old_euid = current->euid;
 853         int old_suid = current->suid;
 854         int retval;
 855
 856         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 857         if (retval)
 858                 return retval;
 859
 860         if (!capable(CAP_SETUID)) {
 861                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 862                     (ruid != current->euid) && (ruid != current->suid))
 863                         return -EPERM;
 864                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 865                     (euid != current->euid) && (euid != current->suid))
 866                         return -EPERM;
 867                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 868                     (suid != current->euid) && (suid != current->suid))
 869                         return -EPERM;
 870         }
 871         if (ruid != (uid_t) -1) {
 872                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 873                         return -EAGAIN;
 874         }
 875         if (euid != (uid_t) -1) {
 876                 if (euid != current->euid)
 877                 {
 878                         current->mm->dumpable = suid_dumpable;
 879                         smp_wmb();
 880                 }
 881                 current->euid = euid;
 882         }
 883         current->fsuid = current->euid;
 884         if (suid != (uid_t) -1)
 885                 current->suid = suid;
 886
 887         key_fsuid_changed(current);
 888         proc_id_connector(current, PROC_EVENT_UID);
 889
 890         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 891 }
 892
 893 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 894 {
 895         int retval;
 896
 897         if (!(retval = put_user(current->uid, ruid)) &&
 898             !(retval = put_user(current->euid, euid)))
 899                 retval = put_user(current->suid, suid);
 900
 901         return retval;
 902 }
 903
 904 /*
 905  * Same as above, but for rgid, egid, sgid.
 906  */
 907 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 908 {
 909         int retval;
 910
 911         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 912         if (retval)
 913                 return retval;
 914
 915         if (!capable(CAP_SETGID)) {
 916                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 917                     (rgid != current->egid) && (rgid != current->sgid))
 918                         return -EPERM;
 919                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 920                     (egid != current->egid) && (egid != current->sgid))
 921                         return -EPERM;
 922                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 923                     (sgid != current->egid) && (sgid != current->sgid))
 924                         return -EPERM;
 925         }
 926         if (egid != (gid_t) -1) {
 927                 if (egid != current->egid)
 928                 {
 929                         current->mm->dumpable = suid_dumpable;
 930                         smp_wmb();
 931                 }
 932                 current->egid = egid;
 933         }
 934         current->fsgid = current->egid;
 935         if (rgid != (gid_t) -1)
 936                 current->gid = rgid;
 937         if (sgid != (gid_t) -1)
 938                 current->sgid = sgid;
 939
 940         key_fsgid_changed(current);
 941         proc_id_connector(current, PROC_EVENT_GID);
 942         return 0;
 943 }
 944
 945 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 946 {
 947         int retval;
 948
 949         if (!(retval = put_user(current->gid, rgid)) &&
 950             !(retval = put_user(current->egid, egid)))
 951                 retval = put_user(current->sgid, sgid);
 952
 953         return retval;
 954 }
 955
 956
 957 /*
 958  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 959  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 960  * whatever uid it wants to). It normally shadows "euid", except when
 961  * explicitly set by setfsuid() or for access..
 962  */
 963 asmlinkage long sys_setfsuid(uid_t uid)
 964 {
 965         int old_fsuid;
 966
 967         old_fsuid = current->fsuid;
 968         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 969                 return old_fsuid;
 970
 971         if (uid == current->uid || uid == current->euid ||
 972             uid == current->suid || uid == current->fsuid ||
 973             capable(CAP_SETUID))
 974         {
 975                 if (uid != old_fsuid)
 976                 {
 977                         current->mm->dumpable = suid_dumpable;
 978                         smp_wmb();
 979                 }
 980                 current->fsuid = uid;
 981         }
 982
 983         key_fsuid_changed(current);
 984         proc_id_connector(current, PROC_EVENT_UID);
 985
 986         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 987
 988         return old_fsuid;
 989 }
 990
 991 /*
 992  * Samma på svenska..
 993  */
 994 asmlinkage long sys_setfsgid(gid_t gid)
 995 {
 996         int old_fsgid;
 997
 998         old_fsgid = current->fsgid;
 999         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
1000                 return old_fsgid;
1001
1002         if (gid == current->gid || gid == current->egid ||
1003             gid == current->sgid || gid == current->fsgid ||
1004             capable(CAP_SETGID))
1005         {
1006                 if (gid != old_fsgid)
1007                 {
1008                         current->mm->dumpable = suid_dumpable;
1009                         smp_wmb();
1010                 }
1011                 current->fsgid = gid;
1012                 key_fsgid_changed(current);
1013                 proc_id_connector(current, PROC_EVENT_GID);
1014         }
1015         return old_fsgid;
1016 }
1017
1018 asmlinkage long sys_times(struct tms __user * tbuf)
1019 {
1020         /*
1021          *      In the SMP world we might just be unlucky and have one of
1022          *      the times increment as we use it. Since the value is an
1023          *      atomically safe type this is just fine. Conceptually its
1024          *      as if the syscall took an instant longer to occur.
1025          */
1026         if (tbuf) {
1027                 struct tms tmp;
1028                 cputime_t utime, stime, cutime, cstime;
1029
1030 #ifdef CONFIG_SMP
1031                 if (thread_group_empty(current)) {
1032                         /*
1033                          * Single thread case without the use of any locks.
1034                          *
1035                          * We may race with release_task if two threads are
1036                          * executing. However, release task first adds up the
1037                          * counters (__exit_signal) before  removing the task
1038                          * from the process tasklist (__unhash_process).
1039                          * __exit_signal also acquires and releases the
1040                          * siglock which results in the proper memory ordering
1041                          * so that the list modifications are always visible
1042                          * after the counters have been updated.
1043                          *
1044                          * If the counters have been updated by the second thread
1045                          * but the thread has not yet been removed from the list
1046                          * then the other branch will be executing which will
1047                          * block on tasklist_lock until the exit handling of the
1048                          * other task is finished.
1049                          *
1050                          * This also implies that the sighand->siglock cannot
1051                          * be held by another processor. So we can also
1052                          * skip acquiring that lock.
1053                          */
1054                         utime = cputime_add(current->signal->utime, current->utime);
1055                         stime = cputime_add(current->signal->utime, current->stime);
1056                         cutime = current->signal->cutime;
1057                         cstime = current->signal->cstime;
1058                 } else
1059 #endif
1060                 {
1061
1062                         /* Process with multiple threads */
1063                         struct task_struct *tsk = current;
1064                         struct task_struct *t;
1065
1066                         read_lock(&tasklist_lock);
1067                         utime = tsk->signal->utime;
1068                         stime = tsk->signal->stime;
1069                         t = tsk;
1070                         do {
1071                                 utime = cputime_add(utime, t->utime);
1072                                 stime = cputime_add(stime, t->stime);
1073                                 t = next_thread(t);
1074                         } while (t != tsk);
1075
1076                         /*
1077                          * While we have tasklist_lock read-locked, no dying thread
1078                          * can be updating current->signal->[us]time.  Instead,
1079                          * we got their counts included in the live thread loop.
1080                          * However, another thread can come in right now and
1081                          * do a wait call that updates current->signal->c[us]time.
1082                          * To make sure we always see that pair updated atomically,
1083                          * we take the siglock around fetching them.
1084                          */
1085                         spin_lock_irq(&tsk->sighand->siglock);
1086                         cutime = tsk->signal->cutime;
1087                         cstime = tsk->signal->cstime;
1088                         spin_unlock_irq(&tsk->sighand->siglock);
1089                         read_unlock(&tasklist_lock);
1090                 }
1091                 tmp.tms_utime = cputime_to_clock_t(utime);
1092                 tmp.tms_stime = cputime_to_clock_t(stime);
1093                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1094                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1095                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1096                         return -EFAULT;
1097         }
1098         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1099 }
1100
1101 /*
1102  * This needs some heavy checking ...
1103  * I just haven't the stomach for it. I also don't fully
1104  * understand sessions/pgrp etc. Let somebody who does explain it.
1105  *
1106  * OK, I think I have the protection semantics right.... this is really
1107  * only important on a multi-user system anyway, to make sure one user
1108  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1109  *
1110  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1111  * LBT 04.03.94
1112  */
1113
1114 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1115 {
1116         struct task_struct *p;
1117         struct task_struct *group_leader = current->group_leader;
1118         pid_t rpgid;
1119         int err = -EINVAL;
1120
1121         if (!pid)
1122                 pid = vx_map_pid(group_leader->pid);
1123         if (!pgid)
1124                 pgid = pid;
1125         if (pgid < 0)
1126                 return -EINVAL;
1127
1128         rpgid = vx_rmap_pid(pgid);
1129
1130         /* From this point forward we keep holding onto the tasklist lock
1131          * so that our parent does not change from under us. -DaveM
1132          */
1133         write_lock_irq(&tasklist_lock);
1134
1135         err = -ESRCH;
1136         p = find_task_by_pid(pid);
1137         if (!p)
1138                 goto out;
1139
1140         err = -EINVAL;
1141         if (!thread_group_leader(p))
1142                 goto out;
1143
1144         if (p->real_parent == group_leader) {
1145                 err = -EPERM;
1146                 if (p->signal->session != group_leader->signal->session)
1147                         goto out;
1148                 err = -EACCES;
1149                 if (p->did_exec)
1150                         goto out;
1151         } else {
1152                 err = -ESRCH;
1153                 if (p != group_leader)
1154                         goto out;
1155         }
1156
1157         err = -EPERM;
1158         if (p->signal->leader)
1159                 goto out;
1160
1161         if (pgid != pid) {
1162                 struct task_struct *p;
1163
1164                 do_each_task_pid(rpgid, PIDTYPE_PGID, p) {
1165                         if (p->signal->session == group_leader->signal->session)
1166                                 goto ok_pgid;
1167                 } while_each_task_pid(rpgid, PIDTYPE_PGID, p);
1168                 goto out;
1169         }
1170
1171 ok_pgid:
1172         err = security_task_setpgid(p, rpgid);
1173         if (err)
1174                 goto out;
1175
1176         if (process_group(p) != rpgid) {
1177                 detach_pid(p, PIDTYPE_PGID);
1178                 p->signal->pgrp = rpgid;
1179                 attach_pid(p, PIDTYPE_PGID, rpgid);
1180         }
1181
1182         err = 0;
1183 out:
1184         /* All paths lead to here, thus we are safe. -DaveM */
1185         write_unlock_irq(&tasklist_lock);
1186         return err;
1187 }
1188
1189 asmlinkage long sys_getpgid(pid_t pid)
1190 {
1191         if (!pid) {
1192                 return vx_rmap_pid(process_group(current));
1193         } else {
1194                 int retval;
1195                 struct task_struct *p;
1196
1197                 read_lock(&tasklist_lock);
1198                 p = find_task_by_pid(pid);
1199
1200                 retval = -ESRCH;
1201                 if (p) {
1202                         retval = security_task_getpgid(p);
1203                         if (!retval)
1204                                 retval = vx_rmap_pid(process_group(p));
1205                 }
1206                 read_unlock(&tasklist_lock);
1207                 return retval;
1208         }
1209 }
1210
1211 #ifdef __ARCH_WANT_SYS_GETPGRP
1212
1213 asmlinkage long sys_getpgrp(void)
1214 {
1215         /* SMP - assuming writes are word atomic this is fine */
1216         return process_group(current);
1217 }
1218
1219 #endif
1220
1221 asmlinkage long sys_getsid(pid_t pid)
1222 {
1223         if (!pid) {
1224                 return current->signal->session;
1225         } else {
1226                 int retval;
1227                 struct task_struct *p;
1228
1229                 read_lock(&tasklist_lock);
1230                 p = find_task_by_pid(pid);
1231
1232                 retval = -ESRCH;
1233                 if(p) {
1234                         retval = security_task_getsid(p);
1235                         if (!retval)
1236                                 retval = p->signal->session;
1237                 }
1238                 read_unlock(&tasklist_lock);
1239                 return retval;
1240         }
1241 }
1242
1243 asmlinkage long sys_setsid(void)
1244 {
1245         struct task_struct *group_leader = current->group_leader;
1246         struct pid *pid;
1247         int err = -EPERM;
1248
1249         down(&tty_sem);
1250         write_lock_irq(&tasklist_lock);
1251
1252         pid = find_pid(PIDTYPE_PGID, group_leader->pid);
1253         if (pid)
1254                 goto out;
1255
1256         group_leader->signal->leader = 1;
1257         __set_special_pids(group_leader->pid, group_leader->pid);
1258         group_leader->signal->tty = NULL;
1259         group_leader->signal->tty_old_pgrp = 0;
1260         err = process_group(group_leader);
1261 out:
1262         write_unlock_irq(&tasklist_lock);
1263         up(&tty_sem);
1264         return err;
1265 }
1266
1267 /*
1268  * Supplementary group IDs
1269  */
1270
1271 /* init to 2 - one for init_task, one to ensure it is never freed */
1272 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1273
1274 struct group_info *groups_alloc(int gidsetsize)
1275 {
1276         struct group_info *group_info;
1277         int nblocks;
1278         int i;
1279
1280         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1281         /* Make sure we always allocate at least one indirect block pointer */
1282         nblocks = nblocks ? : 1;
1283         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1284         if (!group_info)
1285                 return NULL;
1286         group_info->ngroups = gidsetsize;
1287         group_info->nblocks = nblocks;
1288         atomic_set(&group_info->usage, 1);
1289
1290         if (gidsetsize <= NGROUPS_SMALL) {
1291                 group_info->blocks[0] = group_info->small_block;
1292         } else {
1293                 for (i = 0; i < nblocks; i++) {
1294                         gid_t *b;
1295                         b = (void *)__get_free_page(GFP_USER);
1296                         if (!b)
1297                                 goto out_undo_partial_alloc;
1298                         group_info->blocks[i] = b;
1299                 }
1300         }
1301         return group_info;
1302
1303 out_undo_partial_alloc:
1304         while (--i >= 0) {
1305                 free_page((unsigned long)group_info->blocks[i]);
1306         }
1307         kfree(group_info);
1308         return NULL;
1309 }
1310
1311 EXPORT_SYMBOL(groups_alloc);
1312
1313 void groups_free(struct group_info *group_info)
1314 {
1315         if (group_info->blocks[0] != group_info->small_block) {
1316                 int i;
1317                 for (i = 0; i < group_info->nblocks; i++)
1318                         free_page((unsigned long)group_info->blocks[i]);
1319         }
1320         kfree(group_info);
1321 }
1322
1323 EXPORT_SYMBOL(groups_free);
1324
1325 /* export the group_info to a user-space array */
1326 static int groups_to_user(gid_t __user *grouplist,
1327     struct group_info *group_info)
1328 {
1329         int i;
1330         int count = group_info->ngroups;
1331
1332         for (i = 0; i < group_info->nblocks; i++) {
1333                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1334                 int off = i * NGROUPS_PER_BLOCK;
1335                 int len = cp_count * sizeof(*grouplist);
1336
1337                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1338                         return -EFAULT;
1339
1340                 count -= cp_count;
1341         }
1342         return 0;
1343 }
1344
1345 /* fill a group_info from a user-space array - it must be allocated already */
1346 static int groups_from_user(struct group_info *group_info,
1347     gid_t __user *grouplist)
1348  {
1349         int i;
1350         int count = group_info->ngroups;
1351
1352         for (i = 0; i < group_info->nblocks; i++) {
1353                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1354                 int off = i * NGROUPS_PER_BLOCK;
1355                 int len = cp_count * sizeof(*grouplist);
1356
1357                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1358                         return -EFAULT;
1359
1360                 count -= cp_count;
1361         }
1362         return 0;
1363 }
1364
1365 /* a simple Shell sort */
1366 static void groups_sort(struct group_info *group_info)
1367 {
1368         int base, max, stride;
1369         int gidsetsize = group_info->ngroups;
1370
1371         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1372                 ; /* nothing */
1373         stride /= 3;
1374
1375         while (stride) {
1376                 max = gidsetsize - stride;
1377                 for (base = 0; base < max; base++) {
1378                         int left = base;
1379                         int right = left + stride;
1380                         gid_t tmp = GROUP_AT(group_info, right);
1381
1382                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1383                                 GROUP_AT(group_info, right) =
1384                                     GROUP_AT(group_info, left);
1385                                 right = left;
1386                                 left -= stride;
1387                         }
1388                         GROUP_AT(group_info, right) = tmp;
1389                 }
1390                 stride /= 3;
1391         }
1392 }
1393
1394 /* a simple bsearch */
1395 int groups_search(struct group_info *group_info, gid_t grp)
1396 {
1397         int left, right;
1398
1399         if (!group_info)
1400                 return 0;
1401
1402         left = 0;
1403         right = group_info->ngroups;
1404         while (left < right) {
1405                 int mid = (left+right)/2;
1406                 int cmp = grp - GROUP_AT(group_info, mid);
1407                 if (cmp > 0)
1408                         left = mid + 1;
1409                 else if (cmp < 0)
1410                         right = mid;
1411                 else
1412                         return 1;
1413         }
1414         return 0;
1415 }
1416
1417 /* validate and set current->group_info */
1418 int set_current_groups(struct group_info *group_info)
1419 {
1420         int retval;
1421         struct group_info *old_info;
1422
1423         retval = security_task_setgroups(group_info);
1424         if (retval)
1425                 return retval;
1426
1427         groups_sort(group_info);
1428         get_group_info(group_info);
1429
1430         task_lock(current);
1431         old_info = current->group_info;
1432         current->group_info = group_info;
1433         task_unlock(current);
1434
1435         put_group_info(old_info);
1436
1437         return 0;
1438 }
1439
1440 EXPORT_SYMBOL(set_current_groups);
1441
1442 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1443 {
1444         int i = 0;
1445
1446         /*
1447          *      SMP: Nobody else can change our grouplist. Thus we are
1448          *      safe.
1449          */
1450
1451         if (gidsetsize < 0)
1452                 return -EINVAL;
1453
1454         /* no need to grab task_lock here; it cannot change */
1455         get_group_info(current->group_info);
1456         i = current->group_info->ngroups;
1457         if (gidsetsize) {
1458                 if (i > gidsetsize) {
1459                         i = -EINVAL;
1460                         goto out;
1461                 }
1462                 if (groups_to_user(grouplist, current->group_info)) {
1463                         i = -EFAULT;
1464                         goto out;
1465                 }
1466         }
1467 out:
1468         put_group_info(current->group_info);
1469         return i;
1470 }
1471
1472 /*
1473  *      SMP: Our groups are copy-on-write. We can set them safely
1474  *      without another task interfering.
1475  */
1476
1477 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1478 {
1479         struct group_info *group_info;
1480         int retval;
1481
1482         if (!capable(CAP_SETGID))
1483                 return -EPERM;
1484         if ((unsigned)gidsetsize > NGROUPS_MAX)
1485                 return -EINVAL;
1486
1487         group_info = groups_alloc(gidsetsize);
1488         if (!group_info)
1489                 return -ENOMEM;
1490         retval = groups_from_user(group_info, grouplist);
1491         if (retval) {
1492                 put_group_info(group_info);
1493                 return retval;
1494         }
1495
1496         retval = set_current_groups(group_info);
1497         put_group_info(group_info);
1498
1499         return retval;
1500 }
1501
1502 /*
1503  * Check whether we're fsgid/egid or in the supplemental group..
1504  */
1505 int in_group_p(gid_t grp)
1506 {
1507         int retval = 1;
1508         if (grp != current->fsgid) {
1509                 get_group_info(current->group_info);
1510                 retval = groups_search(current->group_info, grp);
1511                 put_group_info(current->group_info);
1512         }
1513         return retval;
1514 }
1515
1516 EXPORT_SYMBOL(in_group_p);
1517
1518 int in_egroup_p(gid_t grp)
1519 {
1520         int retval = 1;
1521         if (grp != current->egid) {
1522                 get_group_info(current->group_info);
1523                 retval = groups_search(current->group_info, grp);
1524                 put_group_info(current->group_info);
1525         }
1526         return retval;
1527 }
1528
1529 EXPORT_SYMBOL(in_egroup_p);
1530
1531 DECLARE_RWSEM(uts_sem);
1532
1533 EXPORT_SYMBOL(uts_sem);
1534
1535 asmlinkage long sys_newuname(struct new_utsname __user * name)
1536 {
1537         int errno = 0;
1538
1539         down_read(&uts_sem);
1540         if (copy_to_user(name, vx_new_utsname(), sizeof *name))
1541                 errno = -EFAULT;
1542         up_read(&uts_sem);
1543         return errno;
1544 }
1545
1546 asmlinkage long sys_sethostname(char __user *name, int len)
1547 {
1548         int errno;
1549         char tmp[__NEW_UTS_LEN];
1550
1551         if (!vx_capable(CAP_SYS_ADMIN, VXC_SET_UTSNAME))
1552                 return -EPERM;
1553         if (len < 0 || len > __NEW_UTS_LEN)
1554                 return -EINVAL;
1555         down_write(&uts_sem);
1556         errno = -EFAULT;
1557         if (!copy_from_user(tmp, name, len)) {
1558                 char *ptr = vx_new_uts(nodename);
1559
1560                 memcpy(ptr, tmp, len);
1561                 ptr[len] = 0;
1562                 errno = 0;
1563         }
1564         up_write(&uts_sem);
1565         return errno;
1566 }
1567
1568 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1569
1570 asmlinkage long sys_gethostname(char __user *name, int len)
1571 {
1572         int i, errno;
1573         char *ptr;
1574
1575         if (len < 0)
1576                 return -EINVAL;
1577         down_read(&uts_sem);
1578         ptr = vx_new_uts(nodename);
1579         i = 1 + strlen(ptr);
1580         if (i > len)
1581                 i = len;
1582         errno = 0;
1583         if (copy_to_user(name, ptr, i))
1584                 errno = -EFAULT;
1585         up_read(&uts_sem);
1586         return errno;
1587 }
1588
1589 #endif
1590
1591 /*
1592  * Only setdomainname; getdomainname can be implemented by calling
1593  * uname()
1594  */
1595 asmlinkage long sys_setdomainname(char __user *name, int len)
1596 {
1597         int errno;
1598         char tmp[__NEW_UTS_LEN];
1599
1600         if (!vx_capable(CAP_SYS_ADMIN, VXC_SET_UTSNAME))
1601                 return -EPERM;
1602         if (len < 0 || len > __NEW_UTS_LEN)
1603                 return -EINVAL;
1604
1605         down_write(&uts_sem);
1606         errno = -EFAULT;
1607         if (!copy_from_user(tmp, name, len)) {
1608                 char *ptr = vx_new_uts(domainname);
1609
1610                 memcpy(ptr, tmp, len);
1611                 ptr[len] = 0;
1612                 errno = 0;
1613         }
1614         up_write(&uts_sem);
1615         return errno;
1616 }
1617
1618 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1619 {
1620         if (resource >= RLIM_NLIMITS)
1621                 return -EINVAL;
1622         else {
1623                 struct rlimit value;
1624                 task_lock(current->group_leader);
1625                 value = current->signal->rlim[resource];
1626                 task_unlock(current->group_leader);
1627                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1628         }
1629 }
1630
1631 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1632
1633 /*
1634  *      Back compatibility for getrlimit. Needed for some apps.
1635  */
1636
1637 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1638 {
1639         struct rlimit x;
1640         if (resource >= RLIM_NLIMITS)
1641                 return -EINVAL;
1642
1643         task_lock(current->group_leader);
1644         x = current->signal->rlim[resource];
1645         task_unlock(current->group_leader);
1646         if(x.rlim_cur > 0x7FFFFFFF)
1647                 x.rlim_cur = 0x7FFFFFFF;
1648         if(x.rlim_max > 0x7FFFFFFF)
1649                 x.rlim_max = 0x7FFFFFFF;
1650         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1651 }
1652
1653 #endif
1654
1655 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1656 {
1657         struct rlimit new_rlim, *old_rlim;
1658         int retval;
1659
1660         if (resource >= RLIM_NLIMITS)
1661                 return -EINVAL;
1662         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1663                 return -EFAULT;
1664        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1665                return -EINVAL;
1666         old_rlim = current->signal->rlim + resource;
1667         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1668             !vx_capable(CAP_SYS_RESOURCE, VXC_SET_RLIMIT))
1669                 return -EPERM;
1670         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1671                         return -EPERM;
1672
1673         retval = security_task_setrlimit(resource, &new_rlim);
1674         if (retval)
1675                 return retval;
1676
1677         task_lock(current->group_leader);
1678         *old_rlim = new_rlim;
1679         task_unlock(current->group_leader);
1680
1681         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1682             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1683              new_rlim.rlim_cur <= cputime_to_secs(
1684                      current->signal->it_prof_expires))) {
1685                 unsigned long rlim_cur = new_rlim.rlim_cur;
1686                 cputime_t cputime;
1687
1688                 if (rlim_cur == 0) {
1689                         /*
1690                          * The caller is asking for an immediate RLIMIT_CPU
1691                          * expiry.  But we use the zero value to mean "it was
1692                          * never set".  So let's cheat and make it one second
1693                          * instead
1694                          */
1695                         rlim_cur = 1;
1696                 }
1697                 cputime = secs_to_cputime(rlim_cur);
1698                 read_lock(&tasklist_lock);
1699                 spin_lock_irq(&current->sighand->siglock);
1700                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1701                                       &cputime, NULL);
1702                 spin_unlock_irq(&current->sighand->siglock);
1703                 read_unlock(&tasklist_lock);
1704         }
1705
1706         return 0;
1707 }
1708
1709 /*
1710  * It would make sense to put struct rusage in the task_struct,
1711  * except that would make the task_struct be *really big*.  After
1712  * task_struct gets moved into malloc'ed memory, it would
1713  * make sense to do this.  It will make moving the rest of the information
1714  * a lot simpler!  (Which we're not doing right now because we're not
1715  * measuring them yet).
1716  *
1717  * This expects to be called with tasklist_lock read-locked or better,
1718  * and the siglock not locked.  It may momentarily take the siglock.
1719  *
1720  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1721  * races with threads incrementing their own counters.  But since word
1722  * reads are atomic, we either get new values or old values and we don't
1723  * care which for the sums.  We always take the siglock to protect reading
1724  * the c* fields from p->signal from races with exit.c updating those
1725  * fields when reaping, so a sample either gets all the additions of a
1726  * given child after it's reaped, or none so this sample is before reaping.
1727  */
1728
1729 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1730 {
1731         struct task_struct *t;
1732         unsigned long flags;
1733         cputime_t utime, stime;
1734
1735         memset((char *) r, 0, sizeof *r);
1736
1737         if (unlikely(!p->signal))
1738                 return;
1739
1740         utime = stime = cputime_zero;
1741
1742         switch (who) {
1743                 case RUSAGE_BOTH:
1744                 case RUSAGE_CHILDREN:
1745                         spin_lock_irqsave(&p->sighand->siglock, flags);
1746                         utime = p->signal->cutime;
1747                         stime = p->signal->cstime;
1748                         r->ru_nvcsw = p->signal->cnvcsw;
1749                         r->ru_nivcsw = p->signal->cnivcsw;
1750                         r->ru_minflt = p->signal->cmin_flt;
1751                         r->ru_majflt = p->signal->cmaj_flt;
1752                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1753
1754                         if (who == RUSAGE_CHILDREN)
1755                                 break;
1756
1757                 case RUSAGE_SELF:
1758                         utime = cputime_add(utime, p->signal->utime);
1759                         stime = cputime_add(stime, p->signal->stime);
1760                         r->ru_nvcsw += p->signal->nvcsw;
1761                         r->ru_nivcsw += p->signal->nivcsw;
1762                         r->ru_minflt += p->signal->min_flt;
1763                         r->ru_majflt += p->signal->maj_flt;
1764                         t = p;
1765                         do {
1766                                 utime = cputime_add(utime, t->utime);
1767                                 stime = cputime_add(stime, t->stime);
1768                                 r->ru_nvcsw += t->nvcsw;
1769                                 r->ru_nivcsw += t->nivcsw;
1770                                 r->ru_minflt += t->min_flt;
1771                                 r->ru_majflt += t->maj_flt;
1772                                 t = next_thread(t);
1773                         } while (t != p);
1774                         break;
1775
1776                 default:
1777                         BUG();
1778         }
1779
1780         cputime_to_timeval(utime, &r->ru_utime);
1781         cputime_to_timeval(stime, &r->ru_stime);
1782 }
1783
1784 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1785 {
1786         struct rusage r;
1787         read_lock(&tasklist_lock);
1788         k_getrusage(p, who, &r);
1789         read_unlock(&tasklist_lock);
1790         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1791 }
1792
1793 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1794 {
1795         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1796                 return -EINVAL;
1797         return getrusage(current, who, ru);
1798 }
1799
1800 asmlinkage long sys_umask(int mask)
1801 {
1802         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1803         return mask;
1804 }
1805
1806 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1807                           unsigned long arg4, unsigned long arg5)
1808 {
1809         long error;
1810
1811         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1812         if (error)
1813                 return error;
1814
1815         switch (option) {
1816                 case PR_SET_PDEATHSIG:
1817                         if (!valid_signal(arg2)) {
1818                                 error = -EINVAL;
1819                                 break;
1820                         }
1821                         current->pdeath_signal = arg2;
1822                         break;
1823                 case PR_GET_PDEATHSIG:
1824                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1825                         break;
1826                 case PR_GET_DUMPABLE:
1827                         error = current->mm->dumpable;
1828                         break;
1829                 case PR_SET_DUMPABLE:
1830                         if (arg2 < 0 || arg2 > 1) {
1831                                 error = -EINVAL;
1832                                 break;
1833                         }
1834                         current->mm->dumpable = arg2;
1835                         break;
1836
1837                 case PR_SET_UNALIGN:
1838                         error = SET_UNALIGN_CTL(current, arg2);
1839                         break;
1840                 case PR_GET_UNALIGN:
1841                         error = GET_UNALIGN_CTL(current, arg2);
1842                         break;
1843                 case PR_SET_FPEMU:
1844                         error = SET_FPEMU_CTL(current, arg2);
1845                         break;
1846                 case PR_GET_FPEMU:
1847                         error = GET_FPEMU_CTL(current, arg2);
1848                         break;
1849                 case PR_SET_FPEXC:
1850                         error = SET_FPEXC_CTL(current, arg2);
1851                         break;
1852                 case PR_GET_FPEXC:
1853                         error = GET_FPEXC_CTL(current, arg2);
1854                         break;
1855                 case PR_GET_TIMING:
1856                         error = PR_TIMING_STATISTICAL;
1857                         break;
1858                 case PR_SET_TIMING:
1859                         if (arg2 == PR_TIMING_STATISTICAL)
1860                                 error = 0;
1861                         else
1862                                 error = -EINVAL;
1863                         break;
1864
1865                 case PR_GET_KEEPCAPS:
1866                         if (current->keep_capabilities)
1867                                 error = 1;
1868                         break;
1869                 case PR_SET_KEEPCAPS:
1870                         if (arg2 != 0 && arg2 != 1) {
1871                                 error = -EINVAL;
1872                                 break;
1873                         }
1874                         current->keep_capabilities = arg2;
1875                         break;
1876                 case PR_SET_NAME: {
1877                         struct task_struct *me = current;
1878                         unsigned char ncomm[sizeof(me->comm)];
1879
1880                         ncomm[sizeof(me->comm)-1] = 0;
1881                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1882                                                 sizeof(me->comm)-1) < 0)
1883                                 return -EFAULT;
1884                         set_task_comm(me, ncomm);
1885                         return 0;
1886                 }
1887                 case PR_GET_NAME: {
1888                         struct task_struct *me = current;
1889                         unsigned char tcomm[sizeof(me->comm)];
1890
1891                         get_task_comm(tcomm, me);
1892                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1893                                 return -EFAULT;
1894                         return 0;
1895                 }
1896                 default:
1897                         error = -EINVAL;
1898                         break;
1899         }
1900         return error;
1901 }