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_task(pgrp, PIDTYPE_PGID, p) {
 173                                 error = set_one_prio(p, niceval, error);
 174                         } while_each_pid_task(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_task(pgrp, PIDTYPE_PGID, p) {
 233                                 niceval = 20 - task_nice(p);
 234                                 if (niceval > retval)
 235                                         retval = niceval;
 236                         } while_each_pid_task(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 static 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 /**
 305  *      kernel_kexec - reboot the system
 306  *
 307  *      Move into place and start executing a preloaded standalone
 308  *      executable.  If nothing was preloaded return an error.
 309  */
 310 static void kernel_kexec(void)
 311 {
 312 #ifdef CONFIG_KEXEC
 313         struct kimage *image;
 314         image = xchg(&kexec_image, NULL);
 315         if (!image)
 316                 return;
 317         kernel_restart_prepare(NULL);
 318         printk(KERN_EMERG "Starting new kernel\n");
 319         machine_shutdown();
 320         machine_kexec(image);
 321 #endif
 322 }
 323
 324 static void kernel_shutdown_prepare(enum system_states state)
 325 {
 326         blocking_notifier_call_chain(&reboot_notifier_list,
 327                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 328         system_state = state;
 329         device_shutdown();
 330 }
 331 /**
 332  *      kernel_halt - halt the system
 333  *
 334  *      Shutdown everything and perform a clean system halt.
 335  */
 336 void kernel_halt(void)
 337 {
 338         kernel_shutdown_prepare(SYSTEM_HALT);
 339         sysdev_shutdown();
 340         printk(KERN_EMERG "System halted.\n");
 341         machine_halt();
 342 }
 343
 344 EXPORT_SYMBOL_GPL(kernel_halt);
 345
 346 /**
 347  *      kernel_power_off - power_off the system
 348  *
 349  *      Shutdown everything and perform a clean system power_off.
 350  */
 351 void kernel_power_off(void)
 352 {
 353         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 354         if (pm_power_off_prepare)
 355                 pm_power_off_prepare();
 356         disable_nonboot_cpus();
 357         sysdev_shutdown();
 358         printk(KERN_EMERG "Power down.\n");
 359         machine_power_off();
 360 }
 361 EXPORT_SYMBOL_GPL(kernel_power_off);
 362 /*
 363  * Reboot system call: for obvious reasons only root may call it,
 364  * and even root needs to set up some magic numbers in the registers
 365  * so that some mistake won't make this reboot the whole machine.
 366  * You can also set the meaning of the ctrl-alt-del-key here.
 367  *
 368  * reboot doesn't sync: do that yourself before calling this.
 369  */
 370 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 371 {
 372         char buffer[256];
 373
 374         /* We only trust the superuser with rebooting the system. */
 375         if (!capable(CAP_SYS_BOOT))
 376                 return -EPERM;
 377
 378         /* For safety, we require "magic" arguments. */
 379         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 380             (magic2 != LINUX_REBOOT_MAGIC2 &&
 381                         magic2 != LINUX_REBOOT_MAGIC2A &&
 382                         magic2 != LINUX_REBOOT_MAGIC2B &&
 383                         magic2 != LINUX_REBOOT_MAGIC2C))
 384                 return -EINVAL;
 385
 386         /* Instead of trying to make the power_off code look like
 387          * halt when pm_power_off is not set do it the easy way.
 388          */
 389         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 390                 cmd = LINUX_REBOOT_CMD_HALT;
 391
 392         lock_kernel();
 393         switch (cmd) {
 394         case LINUX_REBOOT_CMD_RESTART:
 395                 kernel_restart(NULL);
 396                 break;
 397
 398         case LINUX_REBOOT_CMD_CAD_ON:
 399                 C_A_D = 1;
 400                 break;
 401
 402         case LINUX_REBOOT_CMD_CAD_OFF:
 403                 C_A_D = 0;
 404                 break;
 405
 406         case LINUX_REBOOT_CMD_HALT:
 407                 kernel_halt();
 408                 unlock_kernel();
 409                 do_exit(0);
 410                 break;
 411
 412         case LINUX_REBOOT_CMD_POWER_OFF:
 413                 kernel_power_off();
 414                 unlock_kernel();
 415                 do_exit(0);
 416                 break;
 417
 418         case LINUX_REBOOT_CMD_RESTART2:
 419                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 420                         unlock_kernel();
 421                         return -EFAULT;
 422                 }
 423                 buffer[sizeof(buffer) - 1] = '\0';
 424
 425                 kernel_restart(buffer);
 426                 break;
 427
 428         case LINUX_REBOOT_CMD_KEXEC:
 429                 kernel_kexec();
 430                 unlock_kernel();
 431                 return -EINVAL;
 432
 433 #ifdef CONFIG_HIBERNATION
 434         case LINUX_REBOOT_CMD_SW_SUSPEND:
 435                 {
 436                         int ret = hibernate();
 437                         unlock_kernel();
 438                         return ret;
 439                 }
 440 #endif
 441
 442         default:
 443                 unlock_kernel();
 444                 return -EINVAL;
 445         }
 446         unlock_kernel();
 447         return 0;
 448 }
 449
 450 static void deferred_cad(struct work_struct *dummy)
 451 {
 452         kernel_restart(NULL);
 453 }
 454
 455 /*
 456  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 457  * As it's called within an interrupt, it may NOT sync: the only choice
 458  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 459  */
 460 void ctrl_alt_del(void)
 461 {
 462         static DECLARE_WORK(cad_work, deferred_cad);
 463
 464         if (C_A_D)
 465                 schedule_work(&cad_work);
 466         else
 467                 kill_cad_pid(SIGINT, 1);
 468 }
 469
 470 /*
 471  * Unprivileged users may change the real gid to the effective gid
 472  * or vice versa.  (BSD-style)
 473  *
 474  * If you set the real gid at all, or set the effective gid to a value not
 475  * equal to the real gid, then the saved gid is set to the new effective gid.
 476  *
 477  * This makes it possible for a setgid program to completely drop its
 478  * privileges, which is often a useful assertion to make when you are doing
 479  * a security audit over a program.
 480  *
 481  * The general idea is that a program which uses just setregid() will be
 482  * 100% compatible with BSD.  A program which uses just setgid() will be
 483  * 100% compatible with POSIX with saved IDs.
 484  *
 485  * SMP: There are not races, the GIDs are checked only by filesystem
 486  *      operations (as far as semantic preservation is concerned).
 487  */
 488 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 489 {
 490         int old_rgid = current->gid;
 491         int old_egid = current->egid;
 492         int new_rgid = old_rgid;
 493         int new_egid = old_egid;
 494         int retval;
 495
 496         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 497         if (retval)
 498                 return retval;
 499
 500         if (rgid != (gid_t) -1) {
 501                 if ((old_rgid == rgid) ||
 502                     (current->egid==rgid) ||
 503                     capable(CAP_SETGID))
 504                         new_rgid = rgid;
 505                 else
 506                         return -EPERM;
 507         }
 508         if (egid != (gid_t) -1) {
 509                 if ((old_rgid == egid) ||
 510                     (current->egid == egid) ||
 511                     (current->sgid == egid) ||
 512                     capable(CAP_SETGID))
 513                         new_egid = egid;
 514                 else
 515                         return -EPERM;
 516         }
 517         if (new_egid != old_egid) {
 518                 set_dumpable(current->mm, suid_dumpable);
 519                 smp_wmb();
 520         }
 521         if (rgid != (gid_t) -1 ||
 522             (egid != (gid_t) -1 && egid != old_rgid))
 523                 current->sgid = new_egid;
 524         current->fsgid = new_egid;
 525         current->egid = new_egid;
 526         current->gid = new_rgid;
 527         key_fsgid_changed(current);
 528         proc_id_connector(current, PROC_EVENT_GID);
 529         return 0;
 530 }
 531
 532 /*
 533  * setgid() is implemented like SysV w/ SAVED_IDS
 534  *
 535  * SMP: Same implicit races as above.
 536  */
 537 asmlinkage long sys_setgid(gid_t gid)
 538 {
 539         int old_egid = current->egid;
 540         int retval;
 541
 542         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 543         if (retval)
 544                 return retval;
 545
 546         if (capable(CAP_SETGID)) {
 547                 if (old_egid != gid) {
 548                         set_dumpable(current->mm, suid_dumpable);
 549                         smp_wmb();
 550                 }
 551                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 552         } else if ((gid == current->gid) || (gid == current->sgid)) {
 553                 if (old_egid != gid) {
 554                         set_dumpable(current->mm, suid_dumpable);
 555                         smp_wmb();
 556                 }
 557                 current->egid = current->fsgid = gid;
 558         }
 559         else
 560                 return -EPERM;
 561
 562         key_fsgid_changed(current);
 563         proc_id_connector(current, PROC_EVENT_GID);
 564         return 0;
 565 }
 566
 567 static int set_user(uid_t new_ruid, int dumpclear)
 568 {
 569         struct user_struct *new_user;
 570
 571         new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
 572         if (!new_user)
 573                 return -EAGAIN;
 574
 575         if (atomic_read(&new_user->processes) >=
 576                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 577                         new_user != current->nsproxy->user_ns->root_user) {
 578                 free_uid(new_user);
 579                 return -EAGAIN;
 580         }
 581
 582         switch_uid(new_user);
 583
 584         if (dumpclear) {
 585                 set_dumpable(current->mm, suid_dumpable);
 586                 smp_wmb();
 587         }
 588         current->uid = new_ruid;
 589         return 0;
 590 }
 591
 592 /*
 593  * Unprivileged users may change the real uid to the effective uid
 594  * or vice versa.  (BSD-style)
 595  *
 596  * If you set the real uid at all, or set the effective uid to a value not
 597  * equal to the real uid, then the saved uid is set to the new effective uid.
 598  *
 599  * This makes it possible for a setuid program to completely drop its
 600  * privileges, which is often a useful assertion to make when you are doing
 601  * a security audit over a program.
 602  *
 603  * The general idea is that a program which uses just setreuid() will be
 604  * 100% compatible with BSD.  A program which uses just setuid() will be
 605  * 100% compatible with POSIX with saved IDs.
 606  */
 607 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 608 {
 609         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 610         int retval;
 611
 612         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 613         if (retval)
 614                 return retval;
 615
 616         new_ruid = old_ruid = current->uid;
 617         new_euid = old_euid = current->euid;
 618         old_suid = current->suid;
 619
 620         if (ruid != (uid_t) -1) {
 621                 new_ruid = ruid;
 622                 if ((old_ruid != ruid) &&
 623                     (current->euid != ruid) &&
 624                     !capable(CAP_SETUID))
 625                         return -EPERM;
 626         }
 627
 628         if (euid != (uid_t) -1) {
 629                 new_euid = euid;
 630                 if ((old_ruid != euid) &&
 631                     (current->euid != euid) &&
 632                     (current->suid != euid) &&
 633                     !capable(CAP_SETUID))
 634                         return -EPERM;
 635         }
 636
 637         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 638                 return -EAGAIN;
 639
 640         if (new_euid != old_euid) {
 641                 set_dumpable(current->mm, suid_dumpable);
 642                 smp_wmb();
 643         }
 644         current->fsuid = current->euid = new_euid;
 645         if (ruid != (uid_t) -1 ||
 646             (euid != (uid_t) -1 && euid != old_ruid))
 647                 current->suid = current->euid;
 648         current->fsuid = current->euid;
 649
 650         key_fsuid_changed(current);
 651         proc_id_connector(current, PROC_EVENT_UID);
 652
 653         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 654 }
 655
 656
 657
 658 /*
 659  * setuid() is implemented like SysV with SAVED_IDS
 660  *
 661  * Note that SAVED_ID's is deficient in that a setuid root program
 662  * like sendmail, for example, cannot set its uid to be a normal
 663  * user and then switch back, because if you're root, setuid() sets
 664  * the saved uid too.  If you don't like this, blame the bright people
 665  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 666  * will allow a root program to temporarily drop privileges and be able to
 667  * regain them by swapping the real and effective uid.
 668  */
 669 asmlinkage long sys_setuid(uid_t uid)
 670 {
 671         int old_euid = current->euid;
 672         int old_ruid, old_suid, new_suid;
 673         int retval;
 674
 675         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 676         if (retval)
 677                 return retval;
 678
 679         old_ruid = current->uid;
 680         old_suid = current->suid;
 681         new_suid = old_suid;
 682
 683         if (capable(CAP_SETUID)) {
 684                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 685                         return -EAGAIN;
 686                 new_suid = uid;
 687         } else if ((uid != current->uid) && (uid != new_suid))
 688                 return -EPERM;
 689
 690         if (old_euid != uid) {
 691                 set_dumpable(current->mm, suid_dumpable);
 692                 smp_wmb();
 693         }
 694         current->fsuid = current->euid = uid;
 695         current->suid = new_suid;
 696
 697         key_fsuid_changed(current);
 698         proc_id_connector(current, PROC_EVENT_UID);
 699
 700         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 701 }
 702
 703
 704 /*
 705  * This function implements a generic ability to update ruid, euid,
 706  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 707  */
 708 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 709 {
 710         int old_ruid = current->uid;
 711         int old_euid = current->euid;
 712         int old_suid = current->suid;
 713         int retval;
 714
 715         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 716         if (retval)
 717                 return retval;
 718
 719         if (!capable(CAP_SETUID)) {
 720                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 721                     (ruid != current->euid) && (ruid != current->suid))
 722                         return -EPERM;
 723                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 724                     (euid != current->euid) && (euid != current->suid))
 725                         return -EPERM;
 726                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 727                     (suid != current->euid) && (suid != current->suid))
 728                         return -EPERM;
 729         }
 730         if (ruid != (uid_t) -1) {
 731                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 732                         return -EAGAIN;
 733         }
 734         if (euid != (uid_t) -1) {
 735                 if (euid != current->euid) {
 736                         set_dumpable(current->mm, suid_dumpable);
 737                         smp_wmb();
 738                 }
 739                 current->euid = euid;
 740         }
 741         current->fsuid = current->euid;
 742         if (suid != (uid_t) -1)
 743                 current->suid = suid;
 744
 745         key_fsuid_changed(current);
 746         proc_id_connector(current, PROC_EVENT_UID);
 747
 748         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 749 }
 750
 751 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 752 {
 753         int retval;
 754
 755         if (!(retval = put_user(current->uid, ruid)) &&
 756             !(retval = put_user(current->euid, euid)))
 757                 retval = put_user(current->suid, suid);
 758
 759         return retval;
 760 }
 761
 762 /*
 763  * Same as above, but for rgid, egid, sgid.
 764  */
 765 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 766 {
 767         int retval;
 768
 769         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 770         if (retval)
 771                 return retval;
 772
 773         if (!capable(CAP_SETGID)) {
 774                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 775                     (rgid != current->egid) && (rgid != current->sgid))
 776                         return -EPERM;
 777                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 778                     (egid != current->egid) && (egid != current->sgid))
 779                         return -EPERM;
 780                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 781                     (sgid != current->egid) && (sgid != current->sgid))
 782                         return -EPERM;
 783         }
 784         if (egid != (gid_t) -1) {
 785                 if (egid != current->egid) {
 786                         set_dumpable(current->mm, suid_dumpable);
 787                         smp_wmb();
 788                 }
 789                 current->egid = egid;
 790         }
 791         current->fsgid = current->egid;
 792         if (rgid != (gid_t) -1)
 793                 current->gid = rgid;
 794         if (sgid != (gid_t) -1)
 795                 current->sgid = sgid;
 796
 797         key_fsgid_changed(current);
 798         proc_id_connector(current, PROC_EVENT_GID);
 799         return 0;
 800 }
 801
 802 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 803 {
 804         int retval;
 805
 806         if (!(retval = put_user(current->gid, rgid)) &&
 807             !(retval = put_user(current->egid, egid)))
 808                 retval = put_user(current->sgid, sgid);
 809
 810         return retval;
 811 }
 812
 813
 814 /*
 815  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 816  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 817  * whatever uid it wants to). It normally shadows "euid", except when
 818  * explicitly set by setfsuid() or for access..
 819  */
 820 asmlinkage long sys_setfsuid(uid_t uid)
 821 {
 822         int old_fsuid;
 823
 824         old_fsuid = current->fsuid;
 825         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 826                 return old_fsuid;
 827
 828         if (uid == current->uid || uid == current->euid ||
 829             uid == current->suid || uid == current->fsuid ||
 830             capable(CAP_SETUID)) {
 831                 if (uid != old_fsuid) {
 832                         set_dumpable(current->mm, suid_dumpable);
 833                         smp_wmb();
 834                 }
 835                 current->fsuid = uid;
 836         }
 837
 838         key_fsuid_changed(current);
 839         proc_id_connector(current, PROC_EVENT_UID);
 840
 841         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 842
 843         return old_fsuid;
 844 }
 845
 846 /*
 847  * Samma på svenska..
 848  */
 849 asmlinkage long sys_setfsgid(gid_t gid)
 850 {
 851         int old_fsgid;
 852
 853         old_fsgid = current->fsgid;
 854         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 855                 return old_fsgid;
 856
 857         if (gid == current->gid || gid == current->egid ||
 858             gid == current->sgid || gid == current->fsgid ||
 859             capable(CAP_SETGID)) {
 860                 if (gid != old_fsgid) {
 861                         set_dumpable(current->mm, suid_dumpable);
 862                         smp_wmb();
 863                 }
 864                 current->fsgid = gid;
 865                 key_fsgid_changed(current);
 866                 proc_id_connector(current, PROC_EVENT_GID);
 867         }
 868         return old_fsgid;
 869 }
 870
 871 asmlinkage long sys_times(struct tms __user * tbuf)
 872 {
 873         /*
 874          *      In the SMP world we might just be unlucky and have one of
 875          *      the times increment as we use it. Since the value is an
 876          *      atomically safe type this is just fine. Conceptually its
 877          *      as if the syscall took an instant longer to occur.
 878          */
 879         if (tbuf) {
 880                 struct tms tmp;
 881                 struct task_struct *tsk = current;
 882                 struct task_struct *t;
 883                 cputime_t utime, stime, cutime, cstime;
 884
 885                 spin_lock_irq(&tsk->sighand->siglock);
 886                 utime = tsk->signal->utime;
 887                 stime = tsk->signal->stime;
 888                 t = tsk;
 889                 do {
 890                         utime = cputime_add(utime, t->utime);
 891                         stime = cputime_add(stime, t->stime);
 892                         t = next_thread(t);
 893                 } while (t != tsk);
 894
 895                 cutime = tsk->signal->cutime;
 896                 cstime = tsk->signal->cstime;
 897                 spin_unlock_irq(&tsk->sighand->siglock);
 898
 899                 tmp.tms_utime = cputime_to_clock_t(utime);
 900                 tmp.tms_stime = cputime_to_clock_t(stime);
 901                 tmp.tms_cutime = cputime_to_clock_t(cutime);
 902                 tmp.tms_cstime = cputime_to_clock_t(cstime);
 903                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 904                         return -EFAULT;
 905         }
 906         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 907 }
 908
 909 /*
 910  * This needs some heavy checking ...
 911  * I just haven't the stomach for it. I also don't fully
 912  * understand sessions/pgrp etc. Let somebody who does explain it.
 913  *
 914  * OK, I think I have the protection semantics right.... this is really
 915  * only important on a multi-user system anyway, to make sure one user
 916  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 917  *
 918  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 919  * LBT 04.03.94
 920  */
 921 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
 922 {
 923         struct task_struct *p;
 924         struct task_struct *group_leader = current->group_leader;
 925         struct pid *pgrp;
 926         int err;
 927
 928         if (!pid)
 929                 pid = task_pid_vnr(group_leader);
 930         if (!pgid)
 931                 pgid = pid;
 932         if (pgid < 0)
 933                 return -EINVAL;
 934
 935         /* From this point forward we keep holding onto the tasklist lock
 936          * so that our parent does not change from under us. -DaveM
 937          */
 938         write_lock_irq(&tasklist_lock);
 939
 940         err = -ESRCH;
 941         p = find_task_by_vpid(pid);
 942         if (!p)
 943                 goto out;
 944
 945         err = -EINVAL;
 946         if (!thread_group_leader(p))
 947                 goto out;
 948
 949         if (same_thread_group(p->real_parent, group_leader)) {
 950                 err = -EPERM;
 951                 if (task_session(p) != task_session(group_leader))
 952                         goto out;
 953                 err = -EACCES;
 954                 if (p->did_exec)
 955                         goto out;
 956         } else {
 957                 err = -ESRCH;
 958                 if (p != group_leader)
 959                         goto out;
 960         }
 961
 962         err = -EPERM;
 963         if (p->signal->leader)
 964                 goto out;
 965
 966         pgrp = task_pid(p);
 967         if (pgid != pid) {
 968                 struct task_struct *g;
 969
 970                 pgrp = find_vpid(pgid);
 971                 g = pid_task(pgrp, PIDTYPE_PGID);
 972                 if (!g || task_session(g) != task_session(group_leader))
 973                         goto out;
 974         }
 975
 976         err = security_task_setpgid(p, pgid);
 977         if (err)
 978                 goto out;
 979
 980         if (task_pgrp(p) != pgrp) {
 981                 change_pid(p, PIDTYPE_PGID, pgrp);
 982                 set_task_pgrp(p, pid_nr(pgrp));
 983         }
 984
 985         err = 0;
 986 out:
 987         /* All paths lead to here, thus we are safe. -DaveM */
 988         write_unlock_irq(&tasklist_lock);
 989         return err;
 990 }
 991
 992 asmlinkage long sys_getpgid(pid_t pid)
 993 {
 994         if (!pid)
 995                 return task_pgrp_vnr(current);
 996         else {
 997                 int retval;
 998                 struct task_struct *p;
 999
1000                 read_lock(&tasklist_lock);
1001                 p = find_task_by_vpid(pid);
1002                 retval = -ESRCH;
1003                 if (p) {
1004                         retval = security_task_getpgid(p);
1005                         if (!retval)
1006                                 retval = task_pgrp_vnr(p);
1007                 }
1008                 read_unlock(&tasklist_lock);
1009                 return retval;
1010         }
1011 }
1012
1013 #ifdef __ARCH_WANT_SYS_GETPGRP
1014
1015 asmlinkage long sys_getpgrp(void)
1016 {
1017         /* SMP - assuming writes are word atomic this is fine */
1018         return task_pgrp_vnr(current);
1019 }
1020
1021 #endif
1022
1023 asmlinkage long sys_getsid(pid_t pid)
1024 {
1025         struct task_struct *p;
1026         struct pid *sid;
1027         int retval;
1028
1029         rcu_read_lock();
1030         if (!pid)
1031                 sid = task_session(current);
1032         else {
1033                 retval = -ESRCH;
1034                 p = find_task_by_vpid(pid);
1035                 if (!p)
1036                         goto out;
1037                 sid = task_session(p);
1038                 if (!sid)
1039                         goto out;
1040
1041                 retval = security_task_getsid(p);
1042                 if (retval)
1043                         goto out;
1044         }
1045         retval = pid_vnr(sid);
1046 out:
1047         rcu_read_unlock();
1048         return retval;
1049 }
1050
1051 asmlinkage long sys_setsid(void)
1052 {
1053         struct task_struct *group_leader = current->group_leader;
1054         struct pid *sid = task_pid(group_leader);
1055         pid_t session = pid_vnr(sid);
1056         int err = -EPERM;
1057
1058         write_lock_irq(&tasklist_lock);
1059         /* Fail if I am already a session leader */
1060         if (group_leader->signal->leader)
1061                 goto out;
1062
1063         /* Fail if a process group id already exists that equals the
1064          * proposed session id.
1065          */
1066         if (pid_task(sid, PIDTYPE_PGID))
1067                 goto out;
1068
1069         group_leader->signal->leader = 1;
1070         __set_special_pids(sid);
1071
1072         spin_lock(&group_leader->sighand->siglock);
1073         group_leader->signal->tty = NULL;
1074         spin_unlock(&group_leader->sighand->siglock);
1075
1076         err = session;
1077 out:
1078         write_unlock_irq(&tasklist_lock);
1079         return err;
1080 }
1081
1082 /*
1083  * Supplementary group IDs
1084  */
1085
1086 /* init to 2 - one for init_task, one to ensure it is never freed */
1087 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1088
1089 struct group_info *groups_alloc(int gidsetsize)
1090 {
1091         struct group_info *group_info;
1092         int nblocks;
1093         int i;
1094
1095         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1096         /* Make sure we always allocate at least one indirect block pointer */
1097         nblocks = nblocks ? : 1;
1098         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1099         if (!group_info)
1100                 return NULL;
1101         group_info->ngroups = gidsetsize;
1102         group_info->nblocks = nblocks;
1103         atomic_set(&group_info->usage, 1);
1104
1105         if (gidsetsize <= NGROUPS_SMALL)
1106                 group_info->blocks[0] = group_info->small_block;
1107         else {
1108                 for (i = 0; i < nblocks; i++) {
1109                         gid_t *b;
1110                         b = (void *)__get_free_page(GFP_USER);
1111                         if (!b)
1112                                 goto out_undo_partial_alloc;
1113                         group_info->blocks[i] = b;
1114                 }
1115         }
1116         return group_info;
1117
1118 out_undo_partial_alloc:
1119         while (--i >= 0) {
1120                 free_page((unsigned long)group_info->blocks[i]);
1121         }
1122         kfree(group_info);
1123         return NULL;
1124 }
1125
1126 EXPORT_SYMBOL(groups_alloc);
1127
1128 void groups_free(struct group_info *group_info)
1129 {
1130         if (group_info->blocks[0] != group_info->small_block) {
1131                 int i;
1132                 for (i = 0; i < group_info->nblocks; i++)
1133                         free_page((unsigned long)group_info->blocks[i]);
1134         }
1135         kfree(group_info);
1136 }
1137
1138 EXPORT_SYMBOL(groups_free);
1139
1140 /* export the group_info to a user-space array */
1141 static int groups_to_user(gid_t __user *grouplist,
1142     struct group_info *group_info)
1143 {
1144         int i;
1145         unsigned int count = group_info->ngroups;
1146
1147         for (i = 0; i < group_info->nblocks; i++) {
1148                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1149                 unsigned int len = cp_count * sizeof(*grouplist);
1150
1151                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1152                         return -EFAULT;
1153
1154                 grouplist += NGROUPS_PER_BLOCK;
1155                 count -= cp_count;
1156         }
1157         return 0;
1158 }
1159
1160 /* fill a group_info from a user-space array - it must be allocated already */
1161 static int groups_from_user(struct group_info *group_info,
1162     gid_t __user *grouplist)
1163 {
1164         int i;
1165         unsigned int count = group_info->ngroups;
1166
1167         for (i = 0; i < group_info->nblocks; i++) {
1168                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1169                 unsigned int len = cp_count * sizeof(*grouplist);
1170
1171                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1172                         return -EFAULT;
1173
1174                 grouplist += NGROUPS_PER_BLOCK;
1175                 count -= cp_count;
1176         }
1177         return 0;
1178 }
1179
1180 /* a simple Shell sort */
1181 static void groups_sort(struct group_info *group_info)
1182 {
1183         int base, max, stride;
1184         int gidsetsize = group_info->ngroups;
1185
1186         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1187                 ; /* nothing */
1188         stride /= 3;
1189
1190         while (stride) {
1191                 max = gidsetsize - stride;
1192                 for (base = 0; base < max; base++) {
1193                         int left = base;
1194                         int right = left + stride;
1195                         gid_t tmp = GROUP_AT(group_info, right);
1196
1197                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1198                                 GROUP_AT(group_info, right) =
1199                                     GROUP_AT(group_info, left);
1200                                 right = left;
1201                                 left -= stride;
1202                         }
1203                         GROUP_AT(group_info, right) = tmp;
1204                 }
1205                 stride /= 3;
1206         }
1207 }
1208
1209 /* a simple bsearch */
1210 int groups_search(struct group_info *group_info, gid_t grp)
1211 {
1212         unsigned int left, right;
1213
1214         if (!group_info)
1215                 return 0;
1216
1217         left = 0;
1218         right = group_info->ngroups;
1219         while (left < right) {
1220                 unsigned int mid = (left+right)/2;
1221                 int cmp = grp - GROUP_AT(group_info, mid);
1222                 if (cmp > 0)
1223                         left = mid + 1;
1224                 else if (cmp < 0)
1225                         right = mid;
1226                 else
1227                         return 1;
1228         }
1229         return 0;
1230 }
1231
1232 /* validate and set current->group_info */
1233 int set_current_groups(struct group_info *group_info)
1234 {
1235         int retval;
1236         struct group_info *old_info;
1237
1238         retval = security_task_setgroups(group_info);
1239         if (retval)
1240                 return retval;
1241
1242         groups_sort(group_info);
1243         get_group_info(group_info);
1244
1245         task_lock(current);
1246         old_info = current->group_info;
1247         current->group_info = group_info;
1248         task_unlock(current);
1249
1250         put_group_info(old_info);
1251
1252         return 0;
1253 }
1254
1255 EXPORT_SYMBOL(set_current_groups);
1256
1257 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1258 {
1259         int i = 0;
1260
1261         /*
1262          *      SMP: Nobody else can change our grouplist. Thus we are
1263          *      safe.
1264          */
1265
1266         if (gidsetsize < 0)
1267                 return -EINVAL;
1268
1269         /* no need to grab task_lock here; it cannot change */
1270         i = current->group_info->ngroups;
1271         if (gidsetsize) {
1272                 if (i > gidsetsize) {
1273                         i = -EINVAL;
1274                         goto out;
1275                 }
1276                 if (groups_to_user(grouplist, current->group_info)) {
1277                         i = -EFAULT;
1278                         goto out;
1279                 }
1280         }
1281 out:
1282         return i;
1283 }
1284
1285 /*
1286  *      SMP: Our groups are copy-on-write. We can set them safely
1287  *      without another task interfering.
1288  */
1289
1290 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1291 {
1292         struct group_info *group_info;
1293         int retval;
1294
1295         if (!capable(CAP_SETGID))
1296                 return -EPERM;
1297         if ((unsigned)gidsetsize > NGROUPS_MAX)
1298                 return -EINVAL;
1299
1300         group_info = groups_alloc(gidsetsize);
1301         if (!group_info)
1302                 return -ENOMEM;
1303         retval = groups_from_user(group_info, grouplist);
1304         if (retval) {
1305                 put_group_info(group_info);
1306                 return retval;
1307         }
1308
1309         retval = set_current_groups(group_info);
1310         put_group_info(group_info);
1311
1312         return retval;
1313 }
1314
1315 /*
1316  * Check whether we're fsgid/egid or in the supplemental group..
1317  */
1318 int in_group_p(gid_t grp)
1319 {
1320         int retval = 1;
1321         if (grp != current->fsgid)
1322                 retval = groups_search(current->group_info, grp);
1323         return retval;
1324 }
1325
1326 EXPORT_SYMBOL(in_group_p);
1327
1328 int in_egroup_p(gid_t grp)
1329 {
1330         int retval = 1;
1331         if (grp != current->egid)
1332                 retval = groups_search(current->group_info, grp);
1333         return retval;
1334 }
1335
1336 EXPORT_SYMBOL(in_egroup_p);
1337
1338 DECLARE_RWSEM(uts_sem);
1339
1340 EXPORT_SYMBOL(uts_sem);
1341
1342 asmlinkage long sys_newuname(struct new_utsname __user * name)
1343 {
1344         int errno = 0;
1345
1346         down_read(&uts_sem);
1347         if (copy_to_user(name, utsname(), sizeof *name))
1348                 errno = -EFAULT;
1349         up_read(&uts_sem);
1350         return errno;
1351 }
1352
1353 asmlinkage long sys_sethostname(char __user *name, int len)
1354 {
1355         int errno;
1356         char tmp[__NEW_UTS_LEN];
1357
1358         if (!capable(CAP_SYS_ADMIN))
1359                 return -EPERM;
1360         if (len < 0 || len > __NEW_UTS_LEN)
1361                 return -EINVAL;
1362         down_write(&uts_sem);
1363         errno = -EFAULT;
1364         if (!copy_from_user(tmp, name, len)) {
1365                 memcpy(utsname()->nodename, tmp, len);
1366                 utsname()->nodename[len] = 0;
1367                 errno = 0;
1368         }
1369         up_write(&uts_sem);
1370         return errno;
1371 }
1372
1373 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1374
1375 asmlinkage long sys_gethostname(char __user *name, int len)
1376 {
1377         int i, errno;
1378
1379         if (len < 0)
1380                 return -EINVAL;
1381         down_read(&uts_sem);
1382         i = 1 + strlen(utsname()->nodename);
1383         if (i > len)
1384                 i = len;
1385         errno = 0;
1386         if (copy_to_user(name, utsname()->nodename, i))
1387                 errno = -EFAULT;
1388         up_read(&uts_sem);
1389         return errno;
1390 }
1391
1392 #endif
1393
1394 /*
1395  * Only setdomainname; getdomainname can be implemented by calling
1396  * uname()
1397  */
1398 asmlinkage long sys_setdomainname(char __user *name, int len)
1399 {
1400         int errno;
1401         char tmp[__NEW_UTS_LEN];
1402
1403         if (!capable(CAP_SYS_ADMIN))
1404                 return -EPERM;
1405         if (len < 0 || len > __NEW_UTS_LEN)
1406                 return -EINVAL;
1407
1408         down_write(&uts_sem);
1409         errno = -EFAULT;
1410         if (!copy_from_user(tmp, name, len)) {
1411                 memcpy(utsname()->domainname, tmp, len);
1412                 utsname()->domainname[len] = 0;
1413                 errno = 0;
1414         }
1415         up_write(&uts_sem);
1416         return errno;
1417 }
1418
1419 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1420 {
1421         if (resource >= RLIM_NLIMITS)
1422                 return -EINVAL;
1423         else {
1424                 struct rlimit value;
1425                 task_lock(current->group_leader);
1426                 value = current->signal->rlim[resource];
1427                 task_unlock(current->group_leader);
1428                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1429         }
1430 }
1431
1432 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1433
1434 /*
1435  *      Back compatibility for getrlimit. Needed for some apps.
1436  */
1437
1438 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1439 {
1440         struct rlimit x;
1441         if (resource >= RLIM_NLIMITS)
1442                 return -EINVAL;
1443
1444         task_lock(current->group_leader);
1445         x = current->signal->rlim[resource];
1446         task_unlock(current->group_leader);
1447         if (x.rlim_cur > 0x7FFFFFFF)
1448                 x.rlim_cur = 0x7FFFFFFF;
1449         if (x.rlim_max > 0x7FFFFFFF)
1450                 x.rlim_max = 0x7FFFFFFF;
1451         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1452 }
1453
1454 #endif
1455
1456 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1457 {
1458         struct rlimit new_rlim, *old_rlim;
1459         unsigned long it_prof_secs;
1460         int retval;
1461
1462         if (resource >= RLIM_NLIMITS)
1463                 return -EINVAL;
1464         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1465                 return -EFAULT;
1466         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1467                 return -EINVAL;
1468         old_rlim = current->signal->rlim + resource;
1469         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1470             !capable(CAP_SYS_RESOURCE))
1471                 return -EPERM;
1472         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1473                 return -EPERM;
1474
1475         retval = security_task_setrlimit(resource, &new_rlim);
1476         if (retval)
1477                 return retval;
1478
1479         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1480                 /*
1481                  * The caller is asking for an immediate RLIMIT_CPU
1482                  * expiry.  But we use the zero value to mean "it was
1483                  * never set".  So let's cheat and make it one second
1484                  * instead
1485                  */
1486                 new_rlim.rlim_cur = 1;
1487         }
1488
1489         task_lock(current->group_leader);
1490         *old_rlim = new_rlim;
1491         task_unlock(current->group_leader);
1492
1493         if (resource != RLIMIT_CPU)
1494                 goto out;
1495
1496         /*
1497          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1498          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1499          * very long-standing error, and fixing it now risks breakage of
1500          * applications, so we live with it
1501          */
1502         if (new_rlim.rlim_cur == RLIM_INFINITY)
1503                 goto out;
1504
1505         it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1506         if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1507                 unsigned long rlim_cur = new_rlim.rlim_cur;
1508                 cputime_t cputime;
1509
1510                 cputime = secs_to_cputime(rlim_cur);
1511                 read_lock(&tasklist_lock);
1512                 spin_lock_irq(&current->sighand->siglock);
1513                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1514                 spin_unlock_irq(&current->sighand->siglock);
1515                 read_unlock(&tasklist_lock);
1516         }
1517 out:
1518         return 0;
1519 }
1520
1521 /*
1522  * It would make sense to put struct rusage in the task_struct,
1523  * except that would make the task_struct be *really big*.  After
1524  * task_struct gets moved into malloc'ed memory, it would
1525  * make sense to do this.  It will make moving the rest of the information
1526  * a lot simpler!  (Which we're not doing right now because we're not
1527  * measuring them yet).
1528  *
1529  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1530  * races with threads incrementing their own counters.  But since word
1531  * reads are atomic, we either get new values or old values and we don't
1532  * care which for the sums.  We always take the siglock to protect reading
1533  * the c* fields from p->signal from races with exit.c updating those
1534  * fields when reaping, so a sample either gets all the additions of a
1535  * given child after it's reaped, or none so this sample is before reaping.
1536  *
1537  * Locking:
1538  * We need to take the siglock for CHILDEREN, SELF and BOTH
1539  * for  the cases current multithreaded, non-current single threaded
1540  * non-current multithreaded.  Thread traversal is now safe with
1541  * the siglock held.
1542  * Strictly speaking, we donot need to take the siglock if we are current and
1543  * single threaded,  as no one else can take our signal_struct away, no one
1544  * else can  reap the  children to update signal->c* counters, and no one else
1545  * can race with the signal-> fields. If we do not take any lock, the
1546  * signal-> fields could be read out of order while another thread was just
1547  * exiting. So we should  place a read memory barrier when we avoid the lock.
1548  * On the writer side,  write memory barrier is implied in  __exit_signal
1549  * as __exit_signal releases  the siglock spinlock after updating the signal->
1550  * fields. But we don't do this yet to keep things simple.
1551  *
1552  */
1553
1554 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r,
1555                                      cputime_t *utimep, cputime_t *stimep)
1556 {
1557         *utimep = cputime_add(*utimep, t->utime);
1558         *stimep = cputime_add(*stimep, t->stime);
1559         r->ru_nvcsw += t->nvcsw;
1560         r->ru_nivcsw += t->nivcsw;
1561         r->ru_minflt += t->min_flt;
1562         r->ru_majflt += t->maj_flt;
1563         r->ru_inblock += task_io_get_inblock(t);
1564         r->ru_oublock += task_io_get_oublock(t);
1565 }
1566
1567 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1568 {
1569         struct task_struct *t;
1570         unsigned long flags;
1571         cputime_t utime, stime;
1572
1573         memset((char *) r, 0, sizeof *r);
1574         utime = stime = cputime_zero;
1575
1576         if (who == RUSAGE_THREAD) {
1577                 accumulate_thread_rusage(p, r, &utime, &stime);
1578                 goto out;
1579         }
1580
1581         if (!lock_task_sighand(p, &flags))
1582                 return;
1583
1584         switch (who) {
1585                 case RUSAGE_BOTH:
1586                 case RUSAGE_CHILDREN:
1587                         utime = p->signal->cutime;
1588                         stime = p->signal->cstime;
1589                         r->ru_nvcsw = p->signal->cnvcsw;
1590                         r->ru_nivcsw = p->signal->cnivcsw;
1591                         r->ru_minflt = p->signal->cmin_flt;
1592                         r->ru_majflt = p->signal->cmaj_flt;
1593                         r->ru_inblock = p->signal->cinblock;
1594                         r->ru_oublock = p->signal->coublock;
1595
1596                         if (who == RUSAGE_CHILDREN)
1597                                 break;
1598
1599                 case RUSAGE_SELF:
1600                         utime = cputime_add(utime, p->signal->utime);
1601                         stime = cputime_add(stime, p->signal->stime);
1602                         r->ru_nvcsw += p->signal->nvcsw;
1603                         r->ru_nivcsw += p->signal->nivcsw;
1604                         r->ru_minflt += p->signal->min_flt;
1605                         r->ru_majflt += p->signal->maj_flt;
1606                         r->ru_inblock += p->signal->inblock;
1607                         r->ru_oublock += p->signal->oublock;
1608                         t = p;
1609                         do {
1610                                 accumulate_thread_rusage(t, r, &utime, &stime);
1611                                 t = next_thread(t);
1612                         } while (t != p);
1613                         break;
1614
1615                 default:
1616                         BUG();
1617         }
1618         unlock_task_sighand(p, &flags);
1619
1620 out:
1621         cputime_to_timeval(utime, &r->ru_utime);
1622         cputime_to_timeval(stime, &r->ru_stime);
1623 }
1624
1625 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1626 {
1627         struct rusage r;
1628         k_getrusage(p, who, &r);
1629         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1630 }
1631
1632 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1633 {
1634         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1635             who != RUSAGE_THREAD)
1636                 return -EINVAL;
1637         return getrusage(current, who, ru);
1638 }
1639
1640 asmlinkage long sys_umask(int mask)
1641 {
1642         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1643         return mask;
1644 }
1645
1646 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1647                           unsigned long arg4, unsigned long arg5)
1648 {
1649         long uninitialized_var(error);
1650
1651         if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
1652                 return error;
1653
1654         switch (option) {
1655                 case PR_SET_PDEATHSIG:
1656                         if (!valid_signal(arg2)) {
1657                                 error = -EINVAL;
1658                                 break;
1659                         }
1660                         current->pdeath_signal = arg2;
1661                         break;
1662                 case PR_GET_PDEATHSIG:
1663                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1664                         break;
1665                 case PR_GET_DUMPABLE:
1666                         error = get_dumpable(current->mm);
1667                         break;
1668                 case PR_SET_DUMPABLE:
1669                         if (arg2 < 0 || arg2 > 1) {
1670                                 error = -EINVAL;
1671                                 break;
1672                         }
1673                         set_dumpable(current->mm, arg2);
1674                         break;
1675
1676                 case PR_SET_UNALIGN:
1677                         error = SET_UNALIGN_CTL(current, arg2);
1678                         break;
1679                 case PR_GET_UNALIGN:
1680                         error = GET_UNALIGN_CTL(current, arg2);
1681                         break;
1682                 case PR_SET_FPEMU:
1683                         error = SET_FPEMU_CTL(current, arg2);
1684                         break;
1685                 case PR_GET_FPEMU:
1686                         error = GET_FPEMU_CTL(current, arg2);
1687                         break;
1688                 case PR_SET_FPEXC:
1689                         error = SET_FPEXC_CTL(current, arg2);
1690                         break;
1691                 case PR_GET_FPEXC:
1692                         error = GET_FPEXC_CTL(current, arg2);
1693                         break;
1694                 case PR_GET_TIMING:
1695                         error = PR_TIMING_STATISTICAL;
1696                         break;
1697                 case PR_SET_TIMING:
1698                         if (arg2 == PR_TIMING_STATISTICAL)
1699                                 error = 0;
1700                         else
1701                                 error = -EINVAL;
1702                         break;
1703
1704                 case PR_SET_NAME: {
1705                         struct task_struct *me = current;
1706                         unsigned char ncomm[sizeof(me->comm)];
1707
1708                         ncomm[sizeof(me->comm)-1] = 0;
1709                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1710                                                 sizeof(me->comm)-1) < 0)
1711                                 return -EFAULT;
1712                         set_task_comm(me, ncomm);
1713                         return 0;
1714                 }
1715                 case PR_GET_NAME: {
1716                         struct task_struct *me = current;
1717                         unsigned char tcomm[sizeof(me->comm)];
1718
1719                         get_task_comm(tcomm, me);
1720                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1721                                 return -EFAULT;
1722                         return 0;
1723                 }
1724                 case PR_GET_ENDIAN:
1725                         error = GET_ENDIAN(current, arg2);
1726                         break;
1727                 case PR_SET_ENDIAN:
1728                         error = SET_ENDIAN(current, arg2);
1729                         break;
1730
1731                 case PR_GET_SECCOMP:
1732                         error = prctl_get_seccomp();
1733                         break;
1734                 case PR_SET_SECCOMP:
1735                         error = prctl_set_seccomp(arg2);
1736                         break;
1737                 case PR_GET_TSC:
1738                         error = GET_TSC_CTL(arg2);
1739                         break;
1740                 case PR_SET_TSC:
1741                         error = SET_TSC_CTL(arg2);
1742                         break;
1743                 default:
1744                         error = -EINVAL;
1745                         break;
1746         }
1747         return error;
1748 }
1749
1750 asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1751                            struct getcpu_cache __user *unused)
1752 {
1753         int err = 0;
1754         int cpu = raw_smp_processor_id();
1755         if (cpup)
1756                 err |= put_user(cpu, cpup);
1757         if (nodep)
1758                 err |= put_user(cpu_to_node(cpu), nodep);
1759         return err ? -EFAULT : 0;
1760 }
1761
1762 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1763
1764 static void argv_cleanup(char **argv, char **envp)
1765 {
1766         argv_free(argv);
1767 }
1768
1769 /**
1770  * orderly_poweroff - Trigger an orderly system poweroff
1771  * @force: force poweroff if command execution fails
1772  *
1773  * This may be called from any context to trigger a system shutdown.
1774  * If the orderly shutdown fails, it will force an immediate shutdown.
1775  */
1776 int orderly_poweroff(bool force)
1777 {
1778         int argc;
1779         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1780         static char *envp[] = {
1781                 "HOME=/",
1782                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1783                 NULL
1784         };
1785         int ret = -ENOMEM;
1786         struct subprocess_info *info;
1787
1788         if (argv == NULL) {
1789                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1790                        __func__, poweroff_cmd);
1791                 goto out;
1792         }
1793
1794         info = call_usermodehelper_setup(argv[0], argv, envp);
1795         if (info == NULL) {
1796                 argv_free(argv);
1797                 goto out;
1798         }
1799
1800         call_usermodehelper_setcleanup(info, argv_cleanup);
1801
1802         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1803
1804   out:
1805         if (ret && force) {
1806                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1807                        "forcing the issue\n");
1808
1809                 /* I guess this should try to kick off some daemon to
1810                    sync and poweroff asap.  Or not even bother syncing
1811                    if we're doing an emergency shutdown? */
1812                 emergency_sync();
1813                 kernel_power_off();
1814         }
1815
1816         return ret;
1817 }
1818 EXPORT_SYMBOL_GPL(orderly_poweroff);