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

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/mm.h>
   9 #include <linux/utsname.h>
  10 #include <linux/mman.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/notifier.h>
  13 #include <linux/reboot.h>
  14 #include <linux/prctl.h>
  15 #include <linux/highuid.h>
  16 #include <linux/fs.h>
  17 #include <linux/resource.h>
  18 #include <linux/kernel.h>
  19 #include <linux/kexec.h>
  20 #include <linux/workqueue.h>
  21 #include <linux/capability.h>
  22 #include <linux/device.h>
  23 #include <linux/key.h>
  24 #include <linux/times.h>
  25 #include <linux/posix-timers.h>
  26 #include <linux/security.h>
  27 #include <linux/dcookies.h>
  28 #include <linux/suspend.h>
  29 #include <linux/tty.h>
  30 #include <linux/signal.h>
  31 #include <linux/cn_proc.h>
  32 #include <linux/getcpu.h>
  33 #include <linux/task_io_accounting_ops.h>
  34 #include <linux/seccomp.h>
  35 #include <linux/cpu.h>
  36
  37 #include <linux/compat.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/kprobes.h>
  40 #include <linux/user_namespace.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/io.h>
  44 #include <asm/unistd.h>
  45
  46 #ifndef SET_UNALIGN_CTL
  47 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  48 #endif
  49 #ifndef GET_UNALIGN_CTL
  50 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  51 #endif
  52 #ifndef SET_FPEMU_CTL
  53 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  54 #endif
  55 #ifndef GET_FPEMU_CTL
  56 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  57 #endif
  58 #ifndef SET_FPEXC_CTL
  59 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  60 #endif
  61 #ifndef GET_FPEXC_CTL
  62 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  63 #endif
  64 #ifndef GET_ENDIAN
  65 # define GET_ENDIAN(a,b)        (-EINVAL)
  66 #endif
  67 #ifndef SET_ENDIAN
  68 # define SET_ENDIAN(a,b)        (-EINVAL)
  69 #endif
  70 #ifndef GET_TSC_CTL
  71 # define GET_TSC_CTL(a)         (-EINVAL)
  72 #endif
  73 #ifndef SET_TSC_CTL
  74 # define SET_TSC_CTL(a)         (-EINVAL)
  75 #endif
  76
  77 /*
  78  * this is where the system-wide overflow UID and GID are defined, for
  79  * architectures that now have 32-bit UID/GID but didn't in the past
  80  */
  81
  82 int overflowuid = DEFAULT_OVERFLOWUID;
  83 int overflowgid = DEFAULT_OVERFLOWGID;
  84
  85 #ifdef CONFIG_UID16
  86 EXPORT_SYMBOL(overflowuid);
  87 EXPORT_SYMBOL(overflowgid);
  88 #endif
  89
  90 /*
  91  * the same as above, but for filesystems which can only store a 16-bit
  92  * UID and GID. as such, this is needed on all architectures
  93  */
  94
  95 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  96 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  97
  98 EXPORT_SYMBOL(fs_overflowuid);
  99 EXPORT_SYMBOL(fs_overflowgid);
 100
 101 /*
 102  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 103  */
 104
 105 int C_A_D = 1;
 106 struct pid *cad_pid;
 107 EXPORT_SYMBOL(cad_pid);
 108
 109 /*
 110  * If set, this is used for preparing the system to power off.
 111  */
 112
 113 void (*pm_power_off_prepare)(void);
 114
 115 static int set_one_prio(struct task_struct *p, int niceval, int error)
 116 {
 117         int no_nice;
 118
 119         if (p->uid != current->euid &&
 120                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
 121                 error = -EPERM;
 122                 goto out;
 123         }
 124         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 125                 error = -EACCES;
 126                 goto out;
 127         }
 128         no_nice = security_task_setnice(p, niceval);
 129         if (no_nice) {
 130                 error = no_nice;
 131                 goto out;
 132         }
 133         if (error == -ESRCH)
 134                 error = 0;
 135         set_user_nice(p, niceval);
 136 out:
 137         return error;
 138 }
 139
 140 asmlinkage long sys_setpriority(int which, int who, int niceval)
 141 {
 142         struct task_struct *g, *p;
 143         struct user_struct *user;
 144         int error = -EINVAL;
 145         struct pid *pgrp;
 146
 147         if (which > PRIO_USER || which < PRIO_PROCESS)
 148                 goto out;
 149
 150         /* normalize: avoid signed division (rounding problems) */
 151         error = -ESRCH;
 152         if (niceval < -20)
 153                 niceval = -20;
 154         if (niceval > 19)
 155                 niceval = 19;
 156
 157         read_lock(&tasklist_lock);
 158         switch (which) {
 159                 case PRIO_PROCESS:
 160                         if (who)
 161                                 p = find_task_by_vpid(who);
 162                         else
 163                                 p = current;
 164                         if (p)
 165                                 error = set_one_prio(p, niceval, error);
 166                         break;
 167                 case PRIO_PGRP:
 168                         if (who)
 169                                 pgrp = find_vpid(who);
 170                         else
 171                                 pgrp = task_pgrp(current);
 172                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 173                                 error = set_one_prio(p, niceval, error);
 174                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 175                         break;
 176                 case PRIO_USER:
 177                         user = current->user;
 178                         if (!who)
 179                                 who = current->uid;
 180                         else
 181                                 if ((who != current->uid) && !(user = find_user(who)))
 182                                         goto out_unlock;        /* No processes for this user */
 183
 184                         do_each_thread(g, p)
 185                                 if (p->uid == who)
 186                                         error = set_one_prio(p, niceval, error);
 187                         while_each_thread(g, p);
 188                         if (who != current->uid)
 189                                 free_uid(user);         /* For find_user() */
 190                         break;
 191         }
 192 out_unlock:
 193         read_unlock(&tasklist_lock);
 194 out:
 195         return error;
 196 }
 197
 198 /*
 199  * Ugh. To avoid negative return values, "getpriority()" will
 200  * not return the normal nice-value, but a negated value that
 201  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 202  * to stay compatible.
 203  */
 204 asmlinkage long sys_getpriority(int which, int who)
 205 {
 206         struct task_struct *g, *p;
 207         struct user_struct *user;
 208         long niceval, retval = -ESRCH;
 209         struct pid *pgrp;
 210
 211         if (which > PRIO_USER || which < PRIO_PROCESS)
 212                 return -EINVAL;
 213
 214         read_lock(&tasklist_lock);
 215         switch (which) {
 216                 case PRIO_PROCESS:
 217                         if (who)
 218                                 p = find_task_by_vpid(who);
 219                         else
 220                                 p = current;
 221                         if (p) {
 222                                 niceval = 20 - task_nice(p);
 223                                 if (niceval > retval)
 224                                         retval = niceval;
 225                         }
 226                         break;
 227                 case PRIO_PGRP:
 228                         if (who)
 229                                 pgrp = find_vpid(who);
 230                         else
 231                                 pgrp = task_pgrp(current);
 232                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 233                                 niceval = 20 - task_nice(p);
 234                                 if (niceval > retval)
 235                                         retval = niceval;
 236                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 237                         break;
 238                 case PRIO_USER:
 239                         user = current->user;
 240                         if (!who)
 241                                 who = current->uid;
 242                         else
 243                                 if ((who != current->uid) && !(user = find_user(who)))
 244                                         goto out_unlock;        /* No processes for this user */
 245
 246                         do_each_thread(g, p)
 247                                 if (p->uid == who) {
 248                                         niceval = 20 - task_nice(p);
 249                                         if (niceval > retval)
 250                                                 retval = niceval;
 251                                 }
 252                         while_each_thread(g, p);
 253                         if (who != current->uid)
 254                                 free_uid(user);         /* for find_user() */
 255                         break;
 256         }
 257 out_unlock:
 258         read_unlock(&tasklist_lock);
 259
 260         return retval;
 261 }
 262
 263 /**
 264  *      emergency_restart - reboot the system
 265  *
 266  *      Without shutting down any hardware or taking any locks
 267  *      reboot the system.  This is called when we know we are in
 268  *      trouble so this is our best effort to reboot.  This is
 269  *      safe to call in interrupt context.
 270  */
 271 void emergency_restart(void)
 272 {
 273         machine_emergency_restart();
 274 }
 275 EXPORT_SYMBOL_GPL(emergency_restart);
 276
 277 void kernel_restart_prepare(char *cmd)
 278 {
 279         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 280         system_state = SYSTEM_RESTART;
 281         device_shutdown();
 282         sysdev_shutdown();
 283 }
 284
 285 /**
 286  *      kernel_restart - reboot the system
 287  *      @cmd: pointer to buffer containing command to execute for restart
 288  *              or %NULL
 289  *
 290  *      Shutdown everything and perform a clean reboot.
 291  *      This is not safe to call in interrupt context.
 292  */
 293 void kernel_restart(char *cmd)
 294 {
 295         kernel_restart_prepare(cmd);
 296         if (!cmd)
 297                 printk(KERN_EMERG "Restarting system.\n");
 298         else
 299                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 300         machine_restart(cmd);
 301 }
 302 EXPORT_SYMBOL_GPL(kernel_restart);
 303
 304 static void kernel_shutdown_prepare(enum system_states state)
 305 {
 306         blocking_notifier_call_chain(&reboot_notifier_list,
 307                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 308         system_state = state;
 309         device_shutdown();
 310 }
 311 /**
 312  *      kernel_halt - halt the system
 313  *
 314  *      Shutdown everything and perform a clean system halt.
 315  */
 316 void kernel_halt(void)
 317 {
 318         kernel_shutdown_prepare(SYSTEM_HALT);
 319         sysdev_shutdown();
 320         printk(KERN_EMERG "System halted.\n");
 321         machine_halt();
 322 }
 323
 324 EXPORT_SYMBOL_GPL(kernel_halt);
 325
 326 /**
 327  *      kernel_power_off - power_off the system
 328  *
 329  *      Shutdown everything and perform a clean system power_off.
 330  */
 331 void kernel_power_off(void)
 332 {
 333         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 334         if (pm_power_off_prepare)
 335                 pm_power_off_prepare();
 336         disable_nonboot_cpus();
 337         sysdev_shutdown();
 338         printk(KERN_EMERG "Power down.\n");
 339         machine_power_off();
 340 }
 341 EXPORT_SYMBOL_GPL(kernel_power_off);
 342 /*
 343  * Reboot system call: for obvious reasons only root may call it,
 344  * and even root needs to set up some magic numbers in the registers
 345  * so that some mistake won't make this reboot the whole machine.
 346  * You can also set the meaning of the ctrl-alt-del-key here.
 347  *
 348  * reboot doesn't sync: do that yourself before calling this.
 349  */
 350 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 351 {
 352         char buffer[256];
 353
 354         /* We only trust the superuser with rebooting the system. */
 355         if (!capable(CAP_SYS_BOOT))
 356                 return -EPERM;
 357
 358         /* For safety, we require "magic" arguments. */
 359         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 360             (magic2 != LINUX_REBOOT_MAGIC2 &&
 361                         magic2 != LINUX_REBOOT_MAGIC2A &&
 362                         magic2 != LINUX_REBOOT_MAGIC2B &&
 363                         magic2 != LINUX_REBOOT_MAGIC2C))
 364                 return -EINVAL;
 365
 366         /* Instead of trying to make the power_off code look like
 367          * halt when pm_power_off is not set do it the easy way.
 368          */
 369         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 370                 cmd = LINUX_REBOOT_CMD_HALT;
 371
 372         lock_kernel();
 373         switch (cmd) {
 374         case LINUX_REBOOT_CMD_RESTART:
 375                 kernel_restart(NULL);
 376                 break;
 377
 378         case LINUX_REBOOT_CMD_CAD_ON:
 379                 C_A_D = 1;
 380                 break;
 381
 382         case LINUX_REBOOT_CMD_CAD_OFF:
 383                 C_A_D = 0;
 384                 break;
 385
 386         case LINUX_REBOOT_CMD_HALT:
 387                 kernel_halt();
 388                 unlock_kernel();
 389                 do_exit(0);
 390                 break;
 391
 392         case LINUX_REBOOT_CMD_POWER_OFF:
 393                 kernel_power_off();
 394                 unlock_kernel();
 395                 do_exit(0);
 396                 break;
 397
 398         case LINUX_REBOOT_CMD_RESTART2:
 399                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 400                         unlock_kernel();
 401                         return -EFAULT;
 402                 }
 403                 buffer[sizeof(buffer) - 1] = '\0';
 404
 405                 kernel_restart(buffer);
 406                 break;
 407
 408 #ifdef CONFIG_KEXEC
 409         case LINUX_REBOOT_CMD_KEXEC:
 410                 {
 411                         int ret;
 412                         ret = kernel_kexec();
 413                         unlock_kernel();
 414                         return ret;
 415                 }
 416 #endif
 417
 418 #ifdef CONFIG_HIBERNATION
 419         case LINUX_REBOOT_CMD_SW_SUSPEND:
 420                 {
 421                         int ret = hibernate();
 422                         unlock_kernel();
 423                         return ret;
 424                 }
 425 #endif
 426
 427         default:
 428                 unlock_kernel();
 429                 return -EINVAL;
 430         }
 431         unlock_kernel();
 432         return 0;
 433 }
 434
 435 static void deferred_cad(struct work_struct *dummy)
 436 {
 437         kernel_restart(NULL);
 438 }
 439
 440 /*
 441  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 442  * As it's called within an interrupt, it may NOT sync: the only choice
 443  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 444  */
 445 void ctrl_alt_del(void)
 446 {
 447         static DECLARE_WORK(cad_work, deferred_cad);
 448
 449         if (C_A_D)
 450                 schedule_work(&cad_work);
 451         else
 452                 kill_cad_pid(SIGINT, 1);
 453 }
 454
 455 /*
 456  * Unprivileged users may change the real gid to the effective gid
 457  * or vice versa.  (BSD-style)
 458  *
 459  * If you set the real gid at all, or set the effective gid to a value not
 460  * equal to the real gid, then the saved gid is set to the new effective gid.
 461  *
 462  * This makes it possible for a setgid program to completely drop its
 463  * privileges, which is often a useful assertion to make when you are doing
 464  * a security audit over a program.
 465  *
 466  * The general idea is that a program which uses just setregid() will be
 467  * 100% compatible with BSD.  A program which uses just setgid() will be
 468  * 100% compatible with POSIX with saved IDs.
 469  *
 470  * SMP: There are not races, the GIDs are checked only by filesystem
 471  *      operations (as far as semantic preservation is concerned).
 472  */
 473 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 474 {
 475         int old_rgid = current->gid;
 476         int old_egid = current->egid;
 477         int new_rgid = old_rgid;
 478         int new_egid = old_egid;
 479         int retval;
 480
 481         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 482         if (retval)
 483                 return retval;
 484
 485         if (rgid != (gid_t) -1) {
 486                 if ((old_rgid == rgid) ||
 487                     (current->egid==rgid) ||
 488                     capable(CAP_SETGID))
 489                         new_rgid = rgid;
 490                 else
 491                         return -EPERM;
 492         }
 493         if (egid != (gid_t) -1) {
 494                 if ((old_rgid == egid) ||
 495                     (current->egid == egid) ||
 496                     (current->sgid == egid) ||
 497                     capable(CAP_SETGID))
 498                         new_egid = egid;
 499                 else
 500                         return -EPERM;
 501         }
 502         if (new_egid != old_egid) {
 503                 set_dumpable(current->mm, suid_dumpable);
 504                 smp_wmb();
 505         }
 506         if (rgid != (gid_t) -1 ||
 507             (egid != (gid_t) -1 && egid != old_rgid))
 508                 current->sgid = new_egid;
 509         current->fsgid = new_egid;
 510         current->egid = new_egid;
 511         current->gid = new_rgid;
 512         key_fsgid_changed(current);
 513         proc_id_connector(current, PROC_EVENT_GID);
 514         return 0;
 515 }
 516
 517 /*
 518  * setgid() is implemented like SysV w/ SAVED_IDS
 519  *
 520  * SMP: Same implicit races as above.
 521  */
 522 asmlinkage long sys_setgid(gid_t gid)
 523 {
 524         int old_egid = current->egid;
 525         int retval;
 526
 527         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 528         if (retval)
 529                 return retval;
 530
 531         if (capable(CAP_SETGID)) {
 532                 if (old_egid != gid) {
 533                         set_dumpable(current->mm, suid_dumpable);
 534                         smp_wmb();
 535                 }
 536                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 537         } else if ((gid == current->gid) || (gid == current->sgid)) {
 538                 if (old_egid != gid) {
 539                         set_dumpable(current->mm, suid_dumpable);
 540                         smp_wmb();
 541                 }
 542                 current->egid = current->fsgid = gid;
 543         }
 544         else
 545                 return -EPERM;
 546
 547         key_fsgid_changed(current);
 548         proc_id_connector(current, PROC_EVENT_GID);
 549         return 0;
 550 }
 551
 552 static int set_user(uid_t new_ruid, int dumpclear)
 553 {
 554         struct user_struct *new_user;
 555
 556         new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
 557         if (!new_user)
 558                 return -EAGAIN;
 559
 560         if (atomic_read(&new_user->processes) >=
 561                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 562                         new_user != current->nsproxy->user_ns->root_user) {
 563                 free_uid(new_user);
 564                 return -EAGAIN;
 565         }
 566
 567         switch_uid(new_user);
 568
 569         if (dumpclear) {
 570                 set_dumpable(current->mm, suid_dumpable);
 571                 smp_wmb();
 572         }
 573         current->uid = new_ruid;
 574         return 0;
 575 }
 576
 577 /*
 578  * Unprivileged users may change the real uid to the effective uid
 579  * or vice versa.  (BSD-style)
 580  *
 581  * If you set the real uid at all, or set the effective uid to a value not
 582  * equal to the real uid, then the saved uid is set to the new effective uid.
 583  *
 584  * This makes it possible for a setuid program to completely drop its
 585  * privileges, which is often a useful assertion to make when you are doing
 586  * a security audit over a program.
 587  *
 588  * The general idea is that a program which uses just setreuid() will be
 589  * 100% compatible with BSD.  A program which uses just setuid() will be
 590  * 100% compatible with POSIX with saved IDs.
 591  */
 592 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 593 {
 594         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 595         int retval;
 596
 597         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 598         if (retval)
 599                 return retval;
 600
 601         new_ruid = old_ruid = current->uid;
 602         new_euid = old_euid = current->euid;
 603         old_suid = current->suid;
 604
 605         if (ruid != (uid_t) -1) {
 606                 new_ruid = ruid;
 607                 if ((old_ruid != ruid) &&
 608                     (current->euid != ruid) &&
 609                     !capable(CAP_SETUID))
 610                         return -EPERM;
 611         }
 612
 613         if (euid != (uid_t) -1) {
 614                 new_euid = euid;
 615                 if ((old_ruid != euid) &&
 616                     (current->euid != euid) &&
 617                     (current->suid != euid) &&
 618                     !capable(CAP_SETUID))
 619                         return -EPERM;
 620         }
 621
 622         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 623                 return -EAGAIN;
 624
 625         if (new_euid != old_euid) {
 626                 set_dumpable(current->mm, suid_dumpable);
 627                 smp_wmb();
 628         }
 629         current->fsuid = current->euid = new_euid;
 630         if (ruid != (uid_t) -1 ||
 631             (euid != (uid_t) -1 && euid != old_ruid))
 632                 current->suid = current->euid;
 633         current->fsuid = current->euid;
 634
 635         key_fsuid_changed(current);
 636         proc_id_connector(current, PROC_EVENT_UID);
 637
 638         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 639 }
 640
 641
 642
 643 /*
 644  * setuid() is implemented like SysV with SAVED_IDS
 645  *
 646  * Note that SAVED_ID's is deficient in that a setuid root program
 647  * like sendmail, for example, cannot set its uid to be a normal
 648  * user and then switch back, because if you're root, setuid() sets
 649  * the saved uid too.  If you don't like this, blame the bright people
 650  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 651  * will allow a root program to temporarily drop privileges and be able to
 652  * regain them by swapping the real and effective uid.
 653  */
 654 asmlinkage long sys_setuid(uid_t uid)
 655 {
 656         int old_euid = current->euid;
 657         int old_ruid, old_suid, new_suid;
 658         int retval;
 659
 660         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 661         if (retval)
 662                 return retval;
 663
 664         old_ruid = current->uid;
 665         old_suid = current->suid;
 666         new_suid = old_suid;
 667
 668         if (capable(CAP_SETUID)) {
 669                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 670                         return -EAGAIN;
 671                 new_suid = uid;
 672         } else if ((uid != current->uid) && (uid != new_suid))
 673                 return -EPERM;
 674
 675         if (old_euid != uid) {
 676                 set_dumpable(current->mm, suid_dumpable);
 677                 smp_wmb();
 678         }
 679         current->fsuid = current->euid = uid;
 680         current->suid = new_suid;
 681
 682         key_fsuid_changed(current);
 683         proc_id_connector(current, PROC_EVENT_UID);
 684
 685         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 686 }
 687
 688
 689 /*
 690  * This function implements a generic ability to update ruid, euid,
 691  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 692  */
 693 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 694 {
 695         int old_ruid = current->uid;
 696         int old_euid = current->euid;
 697         int old_suid = current->suid;
 698         int retval;
 699
 700         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 701         if (retval)
 702                 return retval;
 703
 704         if (!capable(CAP_SETUID)) {
 705                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 706                     (ruid != current->euid) && (ruid != current->suid))
 707                         return -EPERM;
 708                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 709                     (euid != current->euid) && (euid != current->suid))
 710                         return -EPERM;
 711                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 712                     (suid != current->euid) && (suid != current->suid))
 713                         return -EPERM;
 714         }
 715         if (ruid != (uid_t) -1) {
 716                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 717                         return -EAGAIN;
 718         }
 719         if (euid != (uid_t) -1) {
 720                 if (euid != current->euid) {
 721                         set_dumpable(current->mm, suid_dumpable);
 722                         smp_wmb();
 723                 }
 724                 current->euid = euid;
 725         }
 726         current->fsuid = current->euid;
 727         if (suid != (uid_t) -1)
 728                 current->suid = suid;
 729
 730         key_fsuid_changed(current);
 731         proc_id_connector(current, PROC_EVENT_UID);
 732
 733         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 734 }
 735
 736 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 737 {
 738         int retval;
 739
 740         if (!(retval = put_user(current->uid, ruid)) &&
 741             !(retval = put_user(current->euid, euid)))
 742                 retval = put_user(current->suid, suid);
 743
 744         return retval;
 745 }
 746
 747 /*
 748  * Same as above, but for rgid, egid, sgid.
 749  */
 750 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 751 {
 752         int retval;
 753
 754         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 755         if (retval)
 756                 return retval;
 757
 758         if (!capable(CAP_SETGID)) {
 759                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 760                     (rgid != current->egid) && (rgid != current->sgid))
 761                         return -EPERM;
 762                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 763                     (egid != current->egid) && (egid != current->sgid))
 764                         return -EPERM;
 765                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 766                     (sgid != current->egid) && (sgid != current->sgid))
 767                         return -EPERM;
 768         }
 769         if (egid != (gid_t) -1) {
 770                 if (egid != current->egid) {
 771                         set_dumpable(current->mm, suid_dumpable);
 772                         smp_wmb();
 773                 }
 774                 current->egid = egid;
 775         }
 776         current->fsgid = current->egid;
 777         if (rgid != (gid_t) -1)
 778                 current->gid = rgid;
 779         if (sgid != (gid_t) -1)
 780                 current->sgid = sgid;
 781
 782         key_fsgid_changed(current);
 783         proc_id_connector(current, PROC_EVENT_GID);
 784         return 0;
 785 }
 786
 787 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 788 {
 789         int retval;
 790
 791         if (!(retval = put_user(current->gid, rgid)) &&
 792             !(retval = put_user(current->egid, egid)))
 793                 retval = put_user(current->sgid, sgid);
 794
 795         return retval;
 796 }
 797
 798
 799 /*
 800  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 801  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 802  * whatever uid it wants to). It normally shadows "euid", except when
 803  * explicitly set by setfsuid() or for access..
 804  */
 805 asmlinkage long sys_setfsuid(uid_t uid)
 806 {
 807         int old_fsuid;
 808
 809         old_fsuid = current->fsuid;
 810         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 811                 return old_fsuid;
 812
 813         if (uid == current->uid || uid == current->euid ||
 814             uid == current->suid || uid == current->fsuid ||
 815             capable(CAP_SETUID)) {
 816                 if (uid != old_fsuid) {
 817                         set_dumpable(current->mm, suid_dumpable);
 818                         smp_wmb();
 819                 }
 820                 current->fsuid = uid;
 821         }
 822
 823         key_fsuid_changed(current);
 824         proc_id_connector(current, PROC_EVENT_UID);
 825
 826         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 827
 828         return old_fsuid;
 829 }
 830
 831 /*
 832  * Samma på svenska..
 833  */
 834 asmlinkage long sys_setfsgid(gid_t gid)
 835 {
 836         int old_fsgid;
 837
 838         old_fsgid = current->fsgid;
 839         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 840                 return old_fsgid;
 841
 842         if (gid == current->gid || gid == current->egid ||
 843             gid == current->sgid || gid == current->fsgid ||
 844             capable(CAP_SETGID)) {
 845                 if (gid != old_fsgid) {
 846                         set_dumpable(current->mm, suid_dumpable);
 847                         smp_wmb();
 848                 }
 849                 current->fsgid = gid;
 850                 key_fsgid_changed(current);
 851                 proc_id_connector(current, PROC_EVENT_GID);
 852         }
 853         return old_fsgid;
 854 }
 855
 856 asmlinkage long sys_times(struct tms __user * tbuf)
 857 {
 858         /*
 859          *      In the SMP world we might just be unlucky and have one of
 860          *      the times increment as we use it. Since the value is an
 861          *      atomically safe type this is just fine. Conceptually its
 862          *      as if the syscall took an instant longer to occur.
 863          */
 864         if (tbuf) {
 865                 struct tms tmp;
 866                 struct task_struct *tsk = current;
 867                 struct task_struct *t;
 868                 cputime_t utime, stime, cutime, cstime;
 869
 870                 spin_lock_irq(&tsk->sighand->siglock);
 871                 utime = tsk->signal->utime;
 872                 stime = tsk->signal->stime;
 873                 t = tsk;
 874                 do {
 875                         utime = cputime_add(utime, t->utime);
 876                         stime = cputime_add(stime, t->stime);
 877                         t = next_thread(t);
 878                 } while (t != tsk);
 879
 880                 cutime = tsk->signal->cutime;
 881                 cstime = tsk->signal->cstime;
 882                 spin_unlock_irq(&tsk->sighand->siglock);
 883
 884                 tmp.tms_utime = cputime_to_clock_t(utime);
 885                 tmp.tms_stime = cputime_to_clock_t(stime);
 886                 tmp.tms_cutime = cputime_to_clock_t(cutime);
 887                 tmp.tms_cstime = cputime_to_clock_t(cstime);
 888                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 889                         return -EFAULT;
 890         }
 891         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 892 }
 893
 894 /*
 895  * This needs some heavy checking ...
 896  * I just haven't the stomach for it. I also don't fully
 897  * understand sessions/pgrp etc. Let somebody who does explain it.
 898  *
 899  * OK, I think I have the protection semantics right.... this is really
 900  * only important on a multi-user system anyway, to make sure one user
 901  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 902  *
 903  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 904  * LBT 04.03.94
 905  */
 906 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
 907 {
 908         struct task_struct *p;
 909         struct task_struct *group_leader = current->group_leader;
 910         struct pid *pgrp;
 911         int err;
 912
 913         if (!pid)
 914                 pid = task_pid_vnr(group_leader);
 915         if (!pgid)
 916                 pgid = pid;
 917         if (pgid < 0)
 918                 return -EINVAL;
 919
 920         /* From this point forward we keep holding onto the tasklist lock
 921          * so that our parent does not change from under us. -DaveM
 922          */
 923         write_lock_irq(&tasklist_lock);
 924
 925         err = -ESRCH;
 926         p = find_task_by_vpid(pid);
 927         if (!p)
 928                 goto out;
 929
 930         err = -EINVAL;
 931         if (!thread_group_leader(p))
 932                 goto out;
 933
 934         if (same_thread_group(p->real_parent, group_leader)) {
 935                 err = -EPERM;
 936                 if (task_session(p) != task_session(group_leader))
 937                         goto out;
 938                 err = -EACCES;
 939                 if (p->did_exec)
 940                         goto out;
 941         } else {
 942                 err = -ESRCH;
 943                 if (p != group_leader)
 944                         goto out;
 945         }
 946
 947         err = -EPERM;
 948         if (p->signal->leader)
 949                 goto out;
 950
 951         pgrp = task_pid(p);
 952         if (pgid != pid) {
 953                 struct task_struct *g;
 954
 955                 pgrp = find_vpid(pgid);
 956                 g = pid_task(pgrp, PIDTYPE_PGID);
 957                 if (!g || task_session(g) != task_session(group_leader))
 958                         goto out;
 959         }
 960
 961         err = security_task_setpgid(p, pgid);
 962         if (err)
 963                 goto out;
 964
 965         if (task_pgrp(p) != pgrp) {
 966                 change_pid(p, PIDTYPE_PGID, pgrp);
 967                 set_task_pgrp(p, pid_nr(pgrp));
 968         }
 969
 970         err = 0;
 971 out:
 972         /* All paths lead to here, thus we are safe. -DaveM */
 973         write_unlock_irq(&tasklist_lock);
 974         return err;
 975 }
 976
 977 asmlinkage long sys_getpgid(pid_t pid)
 978 {
 979         struct task_struct *p;
 980         struct pid *grp;
 981         int retval;
 982
 983         rcu_read_lock();
 984         if (!pid)
 985                 grp = task_pgrp(current);
 986         else {
 987                 retval = -ESRCH;
 988                 p = find_task_by_vpid(pid);
 989                 if (!p)
 990                         goto out;
 991                 grp = task_pgrp(p);
 992                 if (!grp)
 993                         goto out;
 994
 995                 retval = security_task_getpgid(p);
 996                 if (retval)
 997                         goto out;
 998         }
 999         retval = pid_vnr(grp);
1000 out:
1001         rcu_read_unlock();
1002         return retval;
1003 }
1004
1005 #ifdef __ARCH_WANT_SYS_GETPGRP
1006
1007 asmlinkage long sys_getpgrp(void)
1008 {
1009         return sys_getpgid(0);
1010 }
1011
1012 #endif
1013
1014 asmlinkage long sys_getsid(pid_t pid)
1015 {
1016         struct task_struct *p;
1017         struct pid *sid;
1018         int retval;
1019
1020         rcu_read_lock();
1021         if (!pid)
1022                 sid = task_session(current);
1023         else {
1024                 retval = -ESRCH;
1025                 p = find_task_by_vpid(pid);
1026                 if (!p)
1027                         goto out;
1028                 sid = task_session(p);
1029                 if (!sid)
1030                         goto out;
1031
1032                 retval = security_task_getsid(p);
1033                 if (retval)
1034                         goto out;
1035         }
1036         retval = pid_vnr(sid);
1037 out:
1038         rcu_read_unlock();
1039         return retval;
1040 }
1041
1042 asmlinkage long sys_setsid(void)
1043 {
1044         struct task_struct *group_leader = current->group_leader;
1045         struct pid *sid = task_pid(group_leader);
1046         pid_t session = pid_vnr(sid);
1047         int err = -EPERM;
1048
1049         write_lock_irq(&tasklist_lock);
1050         /* Fail if I am already a session leader */
1051         if (group_leader->signal->leader)
1052                 goto out;
1053
1054         /* Fail if a process group id already exists that equals the
1055          * proposed session id.
1056          */
1057         if (pid_task(sid, PIDTYPE_PGID))
1058                 goto out;
1059
1060         group_leader->signal->leader = 1;
1061         __set_special_pids(sid);
1062
1063         proc_clear_tty(group_leader);
1064
1065         err = session;
1066 out:
1067         write_unlock_irq(&tasklist_lock);
1068         return err;
1069 }
1070
1071 /*
1072  * Supplementary group IDs
1073  */
1074
1075 /* init to 2 - one for init_task, one to ensure it is never freed */
1076 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1077
1078 struct group_info *groups_alloc(int gidsetsize)
1079 {
1080         struct group_info *group_info;
1081         int nblocks;
1082         int i;
1083
1084         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1085         /* Make sure we always allocate at least one indirect block pointer */
1086         nblocks = nblocks ? : 1;
1087         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1088         if (!group_info)
1089                 return NULL;
1090         group_info->ngroups = gidsetsize;
1091         group_info->nblocks = nblocks;
1092         atomic_set(&group_info->usage, 1);
1093
1094         if (gidsetsize <= NGROUPS_SMALL)
1095                 group_info->blocks[0] = group_info->small_block;
1096         else {
1097                 for (i = 0; i < nblocks; i++) {
1098                         gid_t *b;
1099                         b = (void *)__get_free_page(GFP_USER);
1100                         if (!b)
1101                                 goto out_undo_partial_alloc;
1102                         group_info->blocks[i] = b;
1103                 }
1104         }
1105         return group_info;
1106
1107 out_undo_partial_alloc:
1108         while (--i >= 0) {
1109                 free_page((unsigned long)group_info->blocks[i]);
1110         }
1111         kfree(group_info);
1112         return NULL;
1113 }
1114
1115 EXPORT_SYMBOL(groups_alloc);
1116
1117 void groups_free(struct group_info *group_info)
1118 {
1119         if (group_info->blocks[0] != group_info->small_block) {
1120                 int i;
1121                 for (i = 0; i < group_info->nblocks; i++)
1122                         free_page((unsigned long)group_info->blocks[i]);
1123         }
1124         kfree(group_info);
1125 }
1126
1127 EXPORT_SYMBOL(groups_free);
1128
1129 /* export the group_info to a user-space array */
1130 static int groups_to_user(gid_t __user *grouplist,
1131     struct group_info *group_info)
1132 {
1133         int i;
1134         unsigned int count = group_info->ngroups;
1135
1136         for (i = 0; i < group_info->nblocks; i++) {
1137                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1138                 unsigned int len = cp_count * sizeof(*grouplist);
1139
1140                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1141                         return -EFAULT;
1142
1143                 grouplist += NGROUPS_PER_BLOCK;
1144                 count -= cp_count;
1145         }
1146         return 0;
1147 }
1148
1149 /* fill a group_info from a user-space array - it must be allocated already */
1150 static int groups_from_user(struct group_info *group_info,
1151     gid_t __user *grouplist)
1152 {
1153         int i;
1154         unsigned int count = group_info->ngroups;
1155
1156         for (i = 0; i < group_info->nblocks; i++) {
1157                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1158                 unsigned int len = cp_count * sizeof(*grouplist);
1159
1160                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1161                         return -EFAULT;
1162
1163                 grouplist += NGROUPS_PER_BLOCK;
1164                 count -= cp_count;
1165         }
1166         return 0;
1167 }
1168
1169 /* a simple Shell sort */
1170 static void groups_sort(struct group_info *group_info)
1171 {
1172         int base, max, stride;
1173         int gidsetsize = group_info->ngroups;
1174
1175         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1176                 ; /* nothing */
1177         stride /= 3;
1178
1179         while (stride) {
1180                 max = gidsetsize - stride;
1181                 for (base = 0; base < max; base++) {
1182                         int left = base;
1183                         int right = left + stride;
1184                         gid_t tmp = GROUP_AT(group_info, right);
1185
1186                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1187                                 GROUP_AT(group_info, right) =
1188                                     GROUP_AT(group_info, left);
1189                                 right = left;
1190                                 left -= stride;
1191                         }
1192                         GROUP_AT(group_info, right) = tmp;
1193                 }
1194                 stride /= 3;
1195         }
1196 }
1197
1198 /* a simple bsearch */
1199 int groups_search(struct group_info *group_info, gid_t grp)
1200 {
1201         unsigned int left, right;
1202
1203         if (!group_info)
1204                 return 0;
1205
1206         left = 0;
1207         right = group_info->ngroups;
1208         while (left < right) {
1209                 unsigned int mid = (left+right)/2;
1210                 int cmp = grp - GROUP_AT(group_info, mid);
1211                 if (cmp > 0)
1212                         left = mid + 1;
1213                 else if (cmp < 0)
1214                         right = mid;
1215                 else
1216                         return 1;
1217         }
1218         return 0;
1219 }
1220
1221 /* validate and set current->group_info */
1222 int set_current_groups(struct group_info *group_info)
1223 {
1224         int retval;
1225         struct group_info *old_info;
1226
1227         retval = security_task_setgroups(group_info);
1228         if (retval)
1229                 return retval;
1230
1231         groups_sort(group_info);
1232         get_group_info(group_info);
1233
1234         task_lock(current);
1235         old_info = current->group_info;
1236         current->group_info = group_info;
1237         task_unlock(current);
1238
1239         put_group_info(old_info);
1240
1241         return 0;
1242 }
1243
1244 EXPORT_SYMBOL(set_current_groups);
1245
1246 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1247 {
1248         int i = 0;
1249
1250         /*
1251          *      SMP: Nobody else can change our grouplist. Thus we are
1252          *      safe.
1253          */
1254
1255         if (gidsetsize < 0)
1256                 return -EINVAL;
1257
1258         /* no need to grab task_lock here; it cannot change */
1259         i = current->group_info->ngroups;
1260         if (gidsetsize) {
1261                 if (i > gidsetsize) {
1262                         i = -EINVAL;
1263                         goto out;
1264                 }
1265                 if (groups_to_user(grouplist, current->group_info)) {
1266                         i = -EFAULT;
1267                         goto out;
1268                 }
1269         }
1270 out:
1271         return i;
1272 }
1273
1274 /*
1275  *      SMP: Our groups are copy-on-write. We can set them safely
1276  *      without another task interfering.
1277  */
1278
1279 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1280 {
1281         struct group_info *group_info;
1282         int retval;
1283
1284         if (!capable(CAP_SETGID))
1285                 return -EPERM;
1286         if ((unsigned)gidsetsize > NGROUPS_MAX)
1287                 return -EINVAL;
1288
1289         group_info = groups_alloc(gidsetsize);
1290         if (!group_info)
1291                 return -ENOMEM;
1292         retval = groups_from_user(group_info, grouplist);
1293         if (retval) {
1294                 put_group_info(group_info);
1295                 return retval;
1296         }
1297
1298         retval = set_current_groups(group_info);
1299         put_group_info(group_info);
1300
1301         return retval;
1302 }
1303
1304 /*
1305  * Check whether we're fsgid/egid or in the supplemental group..
1306  */
1307 int in_group_p(gid_t grp)
1308 {
1309         int retval = 1;
1310         if (grp != current->fsgid)
1311                 retval = groups_search(current->group_info, grp);
1312         return retval;
1313 }
1314
1315 EXPORT_SYMBOL(in_group_p);
1316
1317 int in_egroup_p(gid_t grp)
1318 {
1319         int retval = 1;
1320         if (grp != current->egid)
1321                 retval = groups_search(current->group_info, grp);
1322         return retval;
1323 }
1324
1325 EXPORT_SYMBOL(in_egroup_p);
1326
1327 DECLARE_RWSEM(uts_sem);
1328
1329 asmlinkage long sys_newuname(struct new_utsname __user * name)
1330 {
1331         int errno = 0;
1332
1333         down_read(&uts_sem);
1334         if (copy_to_user(name, utsname(), sizeof *name))
1335                 errno = -EFAULT;
1336         up_read(&uts_sem);
1337         return errno;
1338 }
1339
1340 asmlinkage long sys_sethostname(char __user *name, int len)
1341 {
1342         int errno;
1343         char tmp[__NEW_UTS_LEN];
1344
1345         if (!capable(CAP_SYS_ADMIN))
1346                 return -EPERM;
1347         if (len < 0 || len > __NEW_UTS_LEN)
1348                 return -EINVAL;
1349         down_write(&uts_sem);
1350         errno = -EFAULT;
1351         if (!copy_from_user(tmp, name, len)) {
1352                 struct new_utsname *u = utsname();
1353
1354                 memcpy(u->nodename, tmp, len);
1355                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1356                 errno = 0;
1357         }
1358         up_write(&uts_sem);
1359         return errno;
1360 }
1361
1362 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1363
1364 asmlinkage long sys_gethostname(char __user *name, int len)
1365 {
1366         int i, errno;
1367         struct new_utsname *u;
1368
1369         if (len < 0)
1370                 return -EINVAL;
1371         down_read(&uts_sem);
1372         u = utsname();
1373         i = 1 + strlen(u->nodename);
1374         if (i > len)
1375                 i = len;
1376         errno = 0;
1377         if (copy_to_user(name, u->nodename, i))
1378                 errno = -EFAULT;
1379         up_read(&uts_sem);
1380         return errno;
1381 }
1382
1383 #endif
1384
1385 /*
1386  * Only setdomainname; getdomainname can be implemented by calling
1387  * uname()
1388  */
1389 asmlinkage long sys_setdomainname(char __user *name, int len)
1390 {
1391         int errno;
1392         char tmp[__NEW_UTS_LEN];
1393
1394         if (!capable(CAP_SYS_ADMIN))
1395                 return -EPERM;
1396         if (len < 0 || len > __NEW_UTS_LEN)
1397                 return -EINVAL;
1398
1399         down_write(&uts_sem);
1400         errno = -EFAULT;
1401         if (!copy_from_user(tmp, name, len)) {
1402                 struct new_utsname *u = utsname();
1403
1404                 memcpy(u->domainname, tmp, len);
1405                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1406                 errno = 0;
1407         }
1408         up_write(&uts_sem);
1409         return errno;
1410 }
1411
1412 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1413 {
1414         if (resource >= RLIM_NLIMITS)
1415                 return -EINVAL;
1416         else {
1417                 struct rlimit value;
1418                 task_lock(current->group_leader);
1419                 value = current->signal->rlim[resource];
1420                 task_unlock(current->group_leader);
1421                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1422         }
1423 }
1424
1425 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1426
1427 /*
1428  *      Back compatibility for getrlimit. Needed for some apps.
1429  */
1430
1431 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1432 {
1433         struct rlimit x;
1434         if (resource >= RLIM_NLIMITS)
1435                 return -EINVAL;
1436
1437         task_lock(current->group_leader);
1438         x = current->signal->rlim[resource];
1439         task_unlock(current->group_leader);
1440         if (x.rlim_cur > 0x7FFFFFFF)
1441                 x.rlim_cur = 0x7FFFFFFF;
1442         if (x.rlim_max > 0x7FFFFFFF)
1443                 x.rlim_max = 0x7FFFFFFF;
1444         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1445 }
1446
1447 #endif
1448
1449 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1450 {
1451         struct rlimit new_rlim, *old_rlim;
1452         unsigned long it_prof_secs;
1453         int retval;
1454
1455         if (resource >= RLIM_NLIMITS)
1456                 return -EINVAL;
1457         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1458                 return -EFAULT;
1459         old_rlim = current->signal->rlim + resource;
1460         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1461             !capable(CAP_SYS_RESOURCE))
1462                 return -EPERM;
1463
1464         if (resource == RLIMIT_NOFILE) {
1465                 if (new_rlim.rlim_max == RLIM_INFINITY)
1466                         new_rlim.rlim_max = sysctl_nr_open;
1467                 if (new_rlim.rlim_cur == RLIM_INFINITY)
1468                         new_rlim.rlim_cur = sysctl_nr_open;
1469                 if (new_rlim.rlim_max > sysctl_nr_open)
1470                         return -EPERM;
1471         }
1472
1473         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1474                 return -EINVAL;
1475
1476         retval = security_task_setrlimit(resource, &new_rlim);
1477         if (retval)
1478                 return retval;
1479
1480         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1481                 /*
1482                  * The caller is asking for an immediate RLIMIT_CPU
1483                  * expiry.  But we use the zero value to mean "it was
1484                  * never set".  So let's cheat and make it one second
1485                  * instead
1486                  */
1487                 new_rlim.rlim_cur = 1;
1488         }
1489
1490         task_lock(current->group_leader);
1491         *old_rlim = new_rlim;
1492         task_unlock(current->group_leader);
1493
1494         if (resource != RLIMIT_CPU)
1495                 goto out;
1496
1497         /*
1498          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1499          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1500          * very long-standing error, and fixing it now risks breakage of
1501          * applications, so we live with it
1502          */
1503         if (new_rlim.rlim_cur == RLIM_INFINITY)
1504                 goto out;
1505
1506         it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1507         if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1508                 unsigned long rlim_cur = new_rlim.rlim_cur;
1509                 cputime_t cputime;
1510
1511                 cputime = secs_to_cputime(rlim_cur);
1512                 read_lock(&tasklist_lock);
1513                 spin_lock_irq(&current->sighand->siglock);
1514                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1515                 spin_unlock_irq(&current->sighand->siglock);
1516                 read_unlock(&tasklist_lock);
1517         }
1518 out:
1519         return 0;
1520 }
1521
1522 /*
1523  * It would make sense to put struct rusage in the task_struct,
1524  * except that would make the task_struct be *really big*.  After
1525  * task_struct gets moved into malloc'ed memory, it would
1526  * make sense to do this.  It will make moving the rest of the information
1527  * a lot simpler!  (Which we're not doing right now because we're not
1528  * measuring them yet).
1529  *
1530  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1531  * races with threads incrementing their own counters.  But since word
1532  * reads are atomic, we either get new values or old values and we don't
1533  * care which for the sums.  We always take the siglock to protect reading
1534  * the c* fields from p->signal from races with exit.c updating those
1535  * fields when reaping, so a sample either gets all the additions of a
1536  * given child after it's reaped, or none so this sample is before reaping.
1537  *
1538  * Locking:
1539  * We need to take the siglock for CHILDEREN, SELF and BOTH
1540  * for  the cases current multithreaded, non-current single threaded
1541  * non-current multithreaded.  Thread traversal is now safe with
1542  * the siglock held.
1543  * Strictly speaking, we donot need to take the siglock if we are current and
1544  * single threaded,  as no one else can take our signal_struct away, no one
1545  * else can  reap the  children to update signal->c* counters, and no one else
1546  * can race with the signal-> fields. If we do not take any lock, the
1547  * signal-> fields could be read out of order while another thread was just
1548  * exiting. So we should  place a read memory barrier when we avoid the lock.
1549  * On the writer side,  write memory barrier is implied in  __exit_signal
1550  * as __exit_signal releases  the siglock spinlock after updating the signal->
1551  * fields. But we don't do this yet to keep things simple.
1552  *
1553  */
1554
1555 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r,
1556                                      cputime_t *utimep, cputime_t *stimep)
1557 {
1558         *utimep = cputime_add(*utimep, t->utime);
1559         *stimep = cputime_add(*stimep, t->stime);
1560         r->ru_nvcsw += t->nvcsw;
1561         r->ru_nivcsw += t->nivcsw;
1562         r->ru_minflt += t->min_flt;
1563         r->ru_majflt += t->maj_flt;
1564         r->ru_inblock += task_io_get_inblock(t);
1565         r->ru_oublock += task_io_get_oublock(t);
1566 }
1567
1568 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1569 {
1570         struct task_struct *t;
1571         unsigned long flags;
1572         cputime_t utime, stime;
1573
1574         memset((char *) r, 0, sizeof *r);
1575         utime = stime = cputime_zero;
1576
1577         if (who == RUSAGE_THREAD) {
1578                 accumulate_thread_rusage(p, r, &utime, &stime);
1579                 goto out;
1580         }
1581
1582         if (!lock_task_sighand(p, &flags))
1583                 return;
1584
1585         switch (who) {
1586                 case RUSAGE_BOTH:
1587                 case RUSAGE_CHILDREN:
1588                         utime = p->signal->cutime;
1589                         stime = p->signal->cstime;
1590                         r->ru_nvcsw = p->signal->cnvcsw;
1591                         r->ru_nivcsw = p->signal->cnivcsw;
1592                         r->ru_minflt = p->signal->cmin_flt;
1593                         r->ru_majflt = p->signal->cmaj_flt;
1594                         r->ru_inblock = p->signal->cinblock;
1595                         r->ru_oublock = p->signal->coublock;
1596
1597                         if (who == RUSAGE_CHILDREN)
1598                                 break;
1599
1600                 case RUSAGE_SELF:
1601                         utime = cputime_add(utime, p->signal->utime);
1602                         stime = cputime_add(stime, p->signal->stime);
1603                         r->ru_nvcsw += p->signal->nvcsw;
1604                         r->ru_nivcsw += p->signal->nivcsw;
1605                         r->ru_minflt += p->signal->min_flt;
1606                         r->ru_majflt += p->signal->maj_flt;
1607                         r->ru_inblock += p->signal->inblock;
1608                         r->ru_oublock += p->signal->oublock;
1609                         t = p;
1610                         do {
1611                                 accumulate_thread_rusage(t, r, &utime, &stime);
1612                                 t = next_thread(t);
1613                         } while (t != p);
1614                         break;
1615
1616                 default:
1617                         BUG();
1618         }
1619         unlock_task_sighand(p, &flags);
1620
1621 out:
1622         cputime_to_timeval(utime, &r->ru_utime);
1623         cputime_to_timeval(stime, &r->ru_stime);
1624 }
1625
1626 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1627 {
1628         struct rusage r;
1629         k_getrusage(p, who, &r);
1630         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1631 }
1632
1633 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1634 {
1635         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1636             who != RUSAGE_THREAD)
1637                 return -EINVAL;
1638         return getrusage(current, who, ru);
1639 }
1640
1641 asmlinkage long sys_umask(int mask)
1642 {
1643         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1644         return mask;
1645 }
1646
1647 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1648                           unsigned long arg4, unsigned long arg5)
1649 {
1650         long error = 0;
1651
1652         if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
1653                 return error;
1654
1655         switch (option) {
1656                 case PR_SET_PDEATHSIG:
1657                         if (!valid_signal(arg2)) {
1658                                 error = -EINVAL;
1659                                 break;
1660                         }
1661                         current->pdeath_signal = arg2;
1662                         break;
1663                 case PR_GET_PDEATHSIG:
1664                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1665                         break;
1666                 case PR_GET_DUMPABLE:
1667                         error = get_dumpable(current->mm);
1668                         break;
1669                 case PR_SET_DUMPABLE:
1670                         if (arg2 < 0 || arg2 > 1) {
1671                                 error = -EINVAL;
1672                                 break;
1673                         }
1674                         set_dumpable(current->mm, arg2);
1675                         break;
1676
1677                 case PR_SET_UNALIGN:
1678                         error = SET_UNALIGN_CTL(current, arg2);
1679                         break;
1680                 case PR_GET_UNALIGN:
1681                         error = GET_UNALIGN_CTL(current, arg2);
1682                         break;
1683                 case PR_SET_FPEMU:
1684                         error = SET_FPEMU_CTL(current, arg2);
1685                         break;
1686                 case PR_GET_FPEMU:
1687                         error = GET_FPEMU_CTL(current, arg2);
1688                         break;
1689                 case PR_SET_FPEXC:
1690                         error = SET_FPEXC_CTL(current, arg2);
1691                         break;
1692                 case PR_GET_FPEXC:
1693                         error = GET_FPEXC_CTL(current, arg2);
1694                         break;
1695                 case PR_GET_TIMING:
1696                         error = PR_TIMING_STATISTICAL;
1697                         break;
1698                 case PR_SET_TIMING:
1699                         if (arg2 != PR_TIMING_STATISTICAL)
1700                                 error = -EINVAL;
1701                         break;
1702
1703                 case PR_SET_NAME: {
1704                         struct task_struct *me = current;
1705                         unsigned char ncomm[sizeof(me->comm)];
1706
1707                         ncomm[sizeof(me->comm)-1] = 0;
1708                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1709                                                 sizeof(me->comm)-1) < 0)
1710                                 return -EFAULT;
1711                         set_task_comm(me, ncomm);
1712                         return 0;
1713                 }
1714                 case PR_GET_NAME: {
1715                         struct task_struct *me = current;
1716                         unsigned char tcomm[sizeof(me->comm)];
1717
1718                         get_task_comm(tcomm, me);
1719                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1720                                 return -EFAULT;
1721                         return 0;
1722                 }
1723                 case PR_GET_ENDIAN:
1724                         error = GET_ENDIAN(current, arg2);
1725                         break;
1726                 case PR_SET_ENDIAN:
1727                         error = SET_ENDIAN(current, arg2);
1728                         break;
1729
1730                 case PR_GET_SECCOMP:
1731                         error = prctl_get_seccomp();
1732                         break;
1733                 case PR_SET_SECCOMP:
1734                         error = prctl_set_seccomp(arg2);
1735                         break;
1736                 case PR_GET_TSC:
1737                         error = GET_TSC_CTL(arg2);
1738                         break;
1739                 case PR_SET_TSC:
1740                         error = SET_TSC_CTL(arg2);
1741                         break;
1742                 default:
1743                         error = -EINVAL;
1744                         break;
1745         }
1746         return error;
1747 }
1748
1749 asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1750                            struct getcpu_cache __user *unused)
1751 {
1752         int err = 0;
1753         int cpu = raw_smp_processor_id();
1754         if (cpup)
1755                 err |= put_user(cpu, cpup);
1756         if (nodep)
1757                 err |= put_user(cpu_to_node(cpu), nodep);
1758         return err ? -EFAULT : 0;
1759 }
1760
1761 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1762
1763 static void argv_cleanup(char **argv, char **envp)
1764 {
1765         argv_free(argv);
1766 }
1767
1768 /**
1769  * orderly_poweroff - Trigger an orderly system poweroff
1770  * @force: force poweroff if command execution fails
1771  *
1772  * This may be called from any context to trigger a system shutdown.
1773  * If the orderly shutdown fails, it will force an immediate shutdown.
1774  */
1775 int orderly_poweroff(bool force)
1776 {
1777         int argc;
1778         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1779         static char *envp[] = {
1780                 "HOME=/",
1781                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1782                 NULL
1783         };
1784         int ret = -ENOMEM;
1785         struct subprocess_info *info;
1786
1787         if (argv == NULL) {
1788                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1789                        __func__, poweroff_cmd);
1790                 goto out;
1791         }
1792
1793         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1794         if (info == NULL) {
1795                 argv_free(argv);
1796                 goto out;
1797         }
1798
1799         call_usermodehelper_setcleanup(info, argv_cleanup);
1800
1801         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1802
1803   out:
1804         if (ret && force) {
1805                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1806                        "forcing the issue\n");
1807
1808                 /* I guess this should try to kick off some daemon to
1809                    sync and poweroff asap.  Or not even bother syncing
1810                    if we're doing an emergency shutdown? */
1811                 emergency_sync();
1812                 kernel_power_off();
1813         }
1814
1815         return ret;
1816 }
1817 EXPORT_SYMBOL_GPL(orderly_poweroff);