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 SYSCALL_DEFINE3(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 SYSCALL_DEFINE2(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 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 351                 void __user *, arg)
 352 {
 353         char buffer[256];
 354
 355         /* We only trust the superuser with rebooting the system. */
 356         if (!capable(CAP_SYS_BOOT))
 357                 return -EPERM;
 358
 359         /* For safety, we require "magic" arguments. */
 360         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 361             (magic2 != LINUX_REBOOT_MAGIC2 &&
 362                         magic2 != LINUX_REBOOT_MAGIC2A &&
 363                         magic2 != LINUX_REBOOT_MAGIC2B &&
 364                         magic2 != LINUX_REBOOT_MAGIC2C))
 365                 return -EINVAL;
 366
 367         /* Instead of trying to make the power_off code look like
 368          * halt when pm_power_off is not set do it the easy way.
 369          */
 370         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 371                 cmd = LINUX_REBOOT_CMD_HALT;
 372
 373         lock_kernel();
 374         switch (cmd) {
 375         case LINUX_REBOOT_CMD_RESTART:
 376                 kernel_restart(NULL);
 377                 break;
 378
 379         case LINUX_REBOOT_CMD_CAD_ON:
 380                 C_A_D = 1;
 381                 break;
 382
 383         case LINUX_REBOOT_CMD_CAD_OFF:
 384                 C_A_D = 0;
 385                 break;
 386
 387         case LINUX_REBOOT_CMD_HALT:
 388                 kernel_halt();
 389                 unlock_kernel();
 390                 do_exit(0);
 391                 break;
 392
 393         case LINUX_REBOOT_CMD_POWER_OFF:
 394                 kernel_power_off();
 395                 unlock_kernel();
 396                 do_exit(0);
 397                 break;
 398
 399         case LINUX_REBOOT_CMD_RESTART2:
 400                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 401                         unlock_kernel();
 402                         return -EFAULT;
 403                 }
 404                 buffer[sizeof(buffer) - 1] = '\0';
 405
 406                 kernel_restart(buffer);
 407                 break;
 408
 409 #ifdef CONFIG_KEXEC
 410         case LINUX_REBOOT_CMD_KEXEC:
 411                 {
 412                         int ret;
 413                         ret = kernel_kexec();
 414                         unlock_kernel();
 415                         return ret;
 416                 }
 417 #endif
 418
 419 #ifdef CONFIG_HIBERNATION
 420         case LINUX_REBOOT_CMD_SW_SUSPEND:
 421                 {
 422                         int ret = hibernate();
 423                         unlock_kernel();
 424                         return ret;
 425                 }
 426 #endif
 427
 428         default:
 429                 unlock_kernel();
 430                 return -EINVAL;
 431         }
 432         unlock_kernel();
 433         return 0;
 434 }
 435
 436 static void deferred_cad(struct work_struct *dummy)
 437 {
 438         kernel_restart(NULL);
 439 }
 440
 441 /*
 442  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 443  * As it's called within an interrupt, it may NOT sync: the only choice
 444  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 445  */
 446 void ctrl_alt_del(void)
 447 {
 448         static DECLARE_WORK(cad_work, deferred_cad);
 449
 450         if (C_A_D)
 451                 schedule_work(&cad_work);
 452         else
 453                 kill_cad_pid(SIGINT, 1);
 454 }
 455
 456 /*
 457  * Unprivileged users may change the real gid to the effective gid
 458  * or vice versa.  (BSD-style)
 459  *
 460  * If you set the real gid at all, or set the effective gid to a value not
 461  * equal to the real gid, then the saved gid is set to the new effective gid.
 462  *
 463  * This makes it possible for a setgid program to completely drop its
 464  * privileges, which is often a useful assertion to make when you are doing
 465  * a security audit over a program.
 466  *
 467  * The general idea is that a program which uses just setregid() will be
 468  * 100% compatible with BSD.  A program which uses just setgid() will be
 469  * 100% compatible with POSIX with saved IDs.
 470  *
 471  * SMP: There are not races, the GIDs are checked only by filesystem
 472  *      operations (as far as semantic preservation is concerned).
 473  */
 474 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 475 {
 476         int old_rgid = current->gid;
 477         int old_egid = current->egid;
 478         int new_rgid = old_rgid;
 479         int new_egid = old_egid;
 480         int retval;
 481
 482         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 483         if (retval)
 484                 return retval;
 485
 486         if (rgid != (gid_t) -1) {
 487                 if ((old_rgid == rgid) ||
 488                     (current->egid==rgid) ||
 489                     capable(CAP_SETGID))
 490                         new_rgid = rgid;
 491                 else
 492                         return -EPERM;
 493         }
 494         if (egid != (gid_t) -1) {
 495                 if ((old_rgid == egid) ||
 496                     (current->egid == egid) ||
 497                     (current->sgid == egid) ||
 498                     capable(CAP_SETGID))
 499                         new_egid = egid;
 500                 else
 501                         return -EPERM;
 502         }
 503         if (new_egid != old_egid) {
 504                 set_dumpable(current->mm, suid_dumpable);
 505                 smp_wmb();
 506         }
 507         if (rgid != (gid_t) -1 ||
 508             (egid != (gid_t) -1 && egid != old_rgid))
 509                 current->sgid = new_egid;
 510         current->fsgid = new_egid;
 511         current->egid = new_egid;
 512         current->gid = new_rgid;
 513         key_fsgid_changed(current);
 514         proc_id_connector(current, PROC_EVENT_GID);
 515         return 0;
 516 }
 517
 518 /*
 519  * setgid() is implemented like SysV w/ SAVED_IDS
 520  *
 521  * SMP: Same implicit races as above.
 522  */
 523 SYSCALL_DEFINE1(setgid, gid_t, gid)
 524 {
 525         int old_egid = current->egid;
 526         int retval;
 527
 528         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 529         if (retval)
 530                 return retval;
 531
 532         if (capable(CAP_SETGID)) {
 533                 if (old_egid != gid) {
 534                         set_dumpable(current->mm, suid_dumpable);
 535                         smp_wmb();
 536                 }
 537                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 538         } else if ((gid == current->gid) || (gid == current->sgid)) {
 539                 if (old_egid != gid) {
 540                         set_dumpable(current->mm, suid_dumpable);
 541                         smp_wmb();
 542                 }
 543                 current->egid = current->fsgid = gid;
 544         }
 545         else
 546                 return -EPERM;
 547
 548         key_fsgid_changed(current);
 549         proc_id_connector(current, PROC_EVENT_GID);
 550         return 0;
 551 }
 552
 553 static int set_user(uid_t new_ruid, int dumpclear)
 554 {
 555         struct user_struct *new_user;
 556
 557         new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
 558         if (!new_user)
 559                 return -EAGAIN;
 560
 561         if (atomic_read(&new_user->processes) >=
 562                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 563                         new_user != current->nsproxy->user_ns->root_user) {
 564                 free_uid(new_user);
 565                 return -EAGAIN;
 566         }
 567
 568         switch_uid(new_user);
 569
 570         if (dumpclear) {
 571                 set_dumpable(current->mm, suid_dumpable);
 572                 smp_wmb();
 573         }
 574         current->uid = new_ruid;
 575         return 0;
 576 }
 577
 578 /*
 579  * Unprivileged users may change the real uid to the effective uid
 580  * or vice versa.  (BSD-style)
 581  *
 582  * If you set the real uid at all, or set the effective uid to a value not
 583  * equal to the real uid, then the saved uid is set to the new effective uid.
 584  *
 585  * This makes it possible for a setuid program to completely drop its
 586  * privileges, which is often a useful assertion to make when you are doing
 587  * a security audit over a program.
 588  *
 589  * The general idea is that a program which uses just setreuid() will be
 590  * 100% compatible with BSD.  A program which uses just setuid() will be
 591  * 100% compatible with POSIX with saved IDs.
 592  */
 593 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 594 {
 595         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 596         int retval;
 597
 598         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 599         if (retval)
 600                 return retval;
 601
 602         new_ruid = old_ruid = current->uid;
 603         new_euid = old_euid = current->euid;
 604         old_suid = current->suid;
 605
 606         if (ruid != (uid_t) -1) {
 607                 new_ruid = ruid;
 608                 if ((old_ruid != ruid) &&
 609                     (current->euid != ruid) &&
 610                     !capable(CAP_SETUID))
 611                         return -EPERM;
 612         }
 613
 614         if (euid != (uid_t) -1) {
 615                 new_euid = euid;
 616                 if ((old_ruid != euid) &&
 617                     (current->euid != euid) &&
 618                     (current->suid != euid) &&
 619                     !capable(CAP_SETUID))
 620                         return -EPERM;
 621         }
 622
 623         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 624                 return -EAGAIN;
 625
 626         if (new_euid != old_euid) {
 627                 set_dumpable(current->mm, suid_dumpable);
 628                 smp_wmb();
 629         }
 630         current->fsuid = current->euid = new_euid;
 631         if (ruid != (uid_t) -1 ||
 632             (euid != (uid_t) -1 && euid != old_ruid))
 633                 current->suid = current->euid;
 634         current->fsuid = current->euid;
 635
 636         key_fsuid_changed(current);
 637         proc_id_connector(current, PROC_EVENT_UID);
 638
 639         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 640 }
 641
 642
 643
 644 /*
 645  * setuid() is implemented like SysV with SAVED_IDS
 646  *
 647  * Note that SAVED_ID's is deficient in that a setuid root program
 648  * like sendmail, for example, cannot set its uid to be a normal
 649  * user and then switch back, because if you're root, setuid() sets
 650  * the saved uid too.  If you don't like this, blame the bright people
 651  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 652  * will allow a root program to temporarily drop privileges and be able to
 653  * regain them by swapping the real and effective uid.
 654  */
 655 SYSCALL_DEFINE1(setuid, uid_t, uid)
 656 {
 657         int old_euid = current->euid;
 658         int old_ruid, old_suid, new_suid;
 659         int retval;
 660
 661         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 662         if (retval)
 663                 return retval;
 664
 665         old_ruid = current->uid;
 666         old_suid = current->suid;
 667         new_suid = old_suid;
 668
 669         if (capable(CAP_SETUID)) {
 670                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 671                         return -EAGAIN;
 672                 new_suid = uid;
 673         } else if ((uid != current->uid) && (uid != new_suid))
 674                 return -EPERM;
 675
 676         if (old_euid != uid) {
 677                 set_dumpable(current->mm, suid_dumpable);
 678                 smp_wmb();
 679         }
 680         current->fsuid = current->euid = uid;
 681         current->suid = new_suid;
 682
 683         key_fsuid_changed(current);
 684         proc_id_connector(current, PROC_EVENT_UID);
 685
 686         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 687 }
 688
 689
 690 /*
 691  * This function implements a generic ability to update ruid, euid,
 692  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 693  */
 694 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 695 {
 696         int old_ruid = current->uid;
 697         int old_euid = current->euid;
 698         int old_suid = current->suid;
 699         int retval;
 700
 701         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 702         if (retval)
 703                 return retval;
 704
 705         if (!capable(CAP_SETUID)) {
 706                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 707                     (ruid != current->euid) && (ruid != current->suid))
 708                         return -EPERM;
 709                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 710                     (euid != current->euid) && (euid != current->suid))
 711                         return -EPERM;
 712                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 713                     (suid != current->euid) && (suid != current->suid))
 714                         return -EPERM;
 715         }
 716         if (ruid != (uid_t) -1) {
 717                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 718                         return -EAGAIN;
 719         }
 720         if (euid != (uid_t) -1) {
 721                 if (euid != current->euid) {
 722                         set_dumpable(current->mm, suid_dumpable);
 723                         smp_wmb();
 724                 }
 725                 current->euid = euid;
 726         }
 727         current->fsuid = current->euid;
 728         if (suid != (uid_t) -1)
 729                 current->suid = suid;
 730
 731         key_fsuid_changed(current);
 732         proc_id_connector(current, PROC_EVENT_UID);
 733
 734         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 735 }
 736
 737 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 738 {
 739         int retval;
 740
 741         if (!(retval = put_user(current->uid, ruid)) &&
 742             !(retval = put_user(current->euid, euid)))
 743                 retval = put_user(current->suid, suid);
 744
 745         return retval;
 746 }
 747
 748 /*
 749  * Same as above, but for rgid, egid, sgid.
 750  */
 751 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 752 {
 753         int retval;
 754
 755         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 756         if (retval)
 757                 return retval;
 758
 759         if (!capable(CAP_SETGID)) {
 760                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 761                     (rgid != current->egid) && (rgid != current->sgid))
 762                         return -EPERM;
 763                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 764                     (egid != current->egid) && (egid != current->sgid))
 765                         return -EPERM;
 766                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 767                     (sgid != current->egid) && (sgid != current->sgid))
 768                         return -EPERM;
 769         }
 770         if (egid != (gid_t) -1) {
 771                 if (egid != current->egid) {
 772                         set_dumpable(current->mm, suid_dumpable);
 773                         smp_wmb();
 774                 }
 775                 current->egid = egid;
 776         }
 777         current->fsgid = current->egid;
 778         if (rgid != (gid_t) -1)
 779                 current->gid = rgid;
 780         if (sgid != (gid_t) -1)
 781                 current->sgid = sgid;
 782
 783         key_fsgid_changed(current);
 784         proc_id_connector(current, PROC_EVENT_GID);
 785         return 0;
 786 }
 787
 788 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 789 {
 790         int retval;
 791
 792         if (!(retval = put_user(current->gid, rgid)) &&
 793             !(retval = put_user(current->egid, egid)))
 794                 retval = put_user(current->sgid, sgid);
 795
 796         return retval;
 797 }
 798
 799
 800 /*
 801  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 802  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 803  * whatever uid it wants to). It normally shadows "euid", except when
 804  * explicitly set by setfsuid() or for access..
 805  */
 806 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 807 {
 808         int old_fsuid;
 809
 810         old_fsuid = current->fsuid;
 811         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 812                 return old_fsuid;
 813
 814         if (uid == current->uid || uid == current->euid ||
 815             uid == current->suid || uid == current->fsuid ||
 816             capable(CAP_SETUID)) {
 817                 if (uid != old_fsuid) {
 818                         set_dumpable(current->mm, suid_dumpable);
 819                         smp_wmb();
 820                 }
 821                 current->fsuid = uid;
 822         }
 823
 824         key_fsuid_changed(current);
 825         proc_id_connector(current, PROC_EVENT_UID);
 826
 827         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 828
 829         return old_fsuid;
 830 }
 831
 832 /*
 833  * Samma på svenska..
 834  */
 835 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 836 {
 837         int old_fsgid;
 838
 839         old_fsgid = current->fsgid;
 840         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 841                 return old_fsgid;
 842
 843         if (gid == current->gid || gid == current->egid ||
 844             gid == current->sgid || gid == current->fsgid ||
 845             capable(CAP_SETGID)) {
 846                 if (gid != old_fsgid) {
 847                         set_dumpable(current->mm, suid_dumpable);
 848                         smp_wmb();
 849                 }
 850                 current->fsgid = gid;
 851                 key_fsgid_changed(current);
 852                 proc_id_connector(current, PROC_EVENT_GID);
 853         }
 854         return old_fsgid;
 855 }
 856
 857 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 858 {
 859         /*
 860          *      In the SMP world we might just be unlucky and have one of
 861          *      the times increment as we use it. Since the value is an
 862          *      atomically safe type this is just fine. Conceptually its
 863          *      as if the syscall took an instant longer to occur.
 864          */
 865         if (tbuf) {
 866                 struct tms tmp;
 867                 struct task_struct *tsk = current;
 868                 struct task_struct *t;
 869                 cputime_t utime, stime, cutime, cstime;
 870
 871                 spin_lock_irq(&tsk->sighand->siglock);
 872                 utime = tsk->signal->utime;
 873                 stime = tsk->signal->stime;
 874                 t = tsk;
 875                 do {
 876                         utime = cputime_add(utime, t->utime);
 877                         stime = cputime_add(stime, t->stime);
 878                         t = next_thread(t);
 879                 } while (t != tsk);
 880
 881                 cutime = tsk->signal->cutime;
 882                 cstime = tsk->signal->cstime;
 883                 spin_unlock_irq(&tsk->sighand->siglock);
 884
 885                 tmp.tms_utime = cputime_to_clock_t(utime);
 886                 tmp.tms_stime = cputime_to_clock_t(stime);
 887                 tmp.tms_cutime = cputime_to_clock_t(cutime);
 888                 tmp.tms_cstime = cputime_to_clock_t(cstime);
 889                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 890                         return -EFAULT;
 891         }
 892         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 893 }
 894
 895 /*
 896  * This needs some heavy checking ...
 897  * I just haven't the stomach for it. I also don't fully
 898  * understand sessions/pgrp etc. Let somebody who does explain it.
 899  *
 900  * OK, I think I have the protection semantics right.... this is really
 901  * only important on a multi-user system anyway, to make sure one user
 902  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 903  *
 904  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 905  * LBT 04.03.94
 906  */
 907 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 908 {
 909         struct task_struct *p;
 910         struct task_struct *group_leader = current->group_leader;
 911         struct pid *pgrp;
 912         int err;
 913
 914         if (!pid)
 915                 pid = task_pid_vnr(group_leader);
 916         if (!pgid)
 917                 pgid = pid;
 918         if (pgid < 0)
 919                 return -EINVAL;
 920
 921         /* From this point forward we keep holding onto the tasklist lock
 922          * so that our parent does not change from under us. -DaveM
 923          */
 924         write_lock_irq(&tasklist_lock);
 925
 926         err = -ESRCH;
 927         p = find_task_by_vpid(pid);
 928         if (!p)
 929                 goto out;
 930
 931         err = -EINVAL;
 932         if (!thread_group_leader(p))
 933                 goto out;
 934
 935         if (same_thread_group(p->real_parent, group_leader)) {
 936                 err = -EPERM;
 937                 if (task_session(p) != task_session(group_leader))
 938                         goto out;
 939                 err = -EACCES;
 940                 if (p->did_exec)
 941                         goto out;
 942         } else {
 943                 err = -ESRCH;
 944                 if (p != group_leader)
 945                         goto out;
 946         }
 947
 948         err = -EPERM;
 949         if (p->signal->leader)
 950                 goto out;
 951
 952         pgrp = task_pid(p);
 953         if (pgid != pid) {
 954                 struct task_struct *g;
 955
 956                 pgrp = find_vpid(pgid);
 957                 g = pid_task(pgrp, PIDTYPE_PGID);
 958                 if (!g || task_session(g) != task_session(group_leader))
 959                         goto out;
 960         }
 961
 962         err = security_task_setpgid(p, pgid);
 963         if (err)
 964                 goto out;
 965
 966         if (task_pgrp(p) != pgrp) {
 967                 change_pid(p, PIDTYPE_PGID, pgrp);
 968                 set_task_pgrp(p, pid_nr(pgrp));
 969         }
 970
 971         err = 0;
 972 out:
 973         /* All paths lead to here, thus we are safe. -DaveM */
 974         write_unlock_irq(&tasklist_lock);
 975         return err;
 976 }
 977
 978 SYSCALL_DEFINE1(getpgid, pid_t, pid)
 979 {
 980         struct task_struct *p;
 981         struct pid *grp;
 982         int retval;
 983
 984         rcu_read_lock();
 985         if (!pid)
 986                 grp = task_pgrp(current);
 987         else {
 988                 retval = -ESRCH;
 989                 p = find_task_by_vpid(pid);
 990                 if (!p)
 991                         goto out;
 992                 grp = task_pgrp(p);
 993                 if (!grp)
 994                         goto out;
 995
 996                 retval = security_task_getpgid(p);
 997                 if (retval)
 998                         goto out;
 999         }
1000         retval = pid_vnr(grp);
1001 out:
1002         rcu_read_unlock();
1003         return retval;
1004 }
1005
1006 #ifdef __ARCH_WANT_SYS_GETPGRP
1007
1008 SYSCALL_DEFINE0(getpgrp)
1009 {
1010         return sys_getpgid(0);
1011 }
1012
1013 #endif
1014
1015 SYSCALL_DEFINE1(getsid, pid_t, pid)
1016 {
1017         struct task_struct *p;
1018         struct pid *sid;
1019         int retval;
1020
1021         rcu_read_lock();
1022         if (!pid)
1023                 sid = task_session(current);
1024         else {
1025                 retval = -ESRCH;
1026                 p = find_task_by_vpid(pid);
1027                 if (!p)
1028                         goto out;
1029                 sid = task_session(p);
1030                 if (!sid)
1031                         goto out;
1032
1033                 retval = security_task_getsid(p);
1034                 if (retval)
1035                         goto out;
1036         }
1037         retval = pid_vnr(sid);
1038 out:
1039         rcu_read_unlock();
1040         return retval;
1041 }
1042
1043 SYSCALL_DEFINE0(setsid)
1044 {
1045         struct task_struct *group_leader = current->group_leader;
1046         struct pid *sid = task_pid(group_leader);
1047         pid_t session = pid_vnr(sid);
1048         int err = -EPERM;
1049
1050         write_lock_irq(&tasklist_lock);
1051         /* Fail if I am already a session leader */
1052         if (group_leader->signal->leader)
1053                 goto out;
1054
1055         /* Fail if a process group id already exists that equals the
1056          * proposed session id.
1057          */
1058         if (pid_task(sid, PIDTYPE_PGID))
1059                 goto out;
1060
1061         group_leader->signal->leader = 1;
1062         __set_special_pids(sid);
1063
1064         spin_lock(&group_leader->sighand->siglock);
1065         group_leader->signal->tty = NULL;
1066         spin_unlock(&group_leader->sighand->siglock);
1067
1068         err = session;
1069 out:
1070         write_unlock_irq(&tasklist_lock);
1071         return err;
1072 }
1073
1074 /*
1075  * Supplementary group IDs
1076  */
1077
1078 /* init to 2 - one for init_task, one to ensure it is never freed */
1079 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1080
1081 struct group_info *groups_alloc(int gidsetsize)
1082 {
1083         struct group_info *group_info;
1084         int nblocks;
1085         int i;
1086
1087         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1088         /* Make sure we always allocate at least one indirect block pointer */
1089         nblocks = nblocks ? : 1;
1090         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1091         if (!group_info)
1092                 return NULL;
1093         group_info->ngroups = gidsetsize;
1094         group_info->nblocks = nblocks;
1095         atomic_set(&group_info->usage, 1);
1096
1097         if (gidsetsize <= NGROUPS_SMALL)
1098                 group_info->blocks[0] = group_info->small_block;
1099         else {
1100                 for (i = 0; i < nblocks; i++) {
1101                         gid_t *b;
1102                         b = (void *)__get_free_page(GFP_USER);
1103                         if (!b)
1104                                 goto out_undo_partial_alloc;
1105                         group_info->blocks[i] = b;
1106                 }
1107         }
1108         return group_info;
1109
1110 out_undo_partial_alloc:
1111         while (--i >= 0) {
1112                 free_page((unsigned long)group_info->blocks[i]);
1113         }
1114         kfree(group_info);
1115         return NULL;
1116 }
1117
1118 EXPORT_SYMBOL(groups_alloc);
1119
1120 void groups_free(struct group_info *group_info)
1121 {
1122         if (group_info->blocks[0] != group_info->small_block) {
1123                 int i;
1124                 for (i = 0; i < group_info->nblocks; i++)
1125                         free_page((unsigned long)group_info->blocks[i]);
1126         }
1127         kfree(group_info);
1128 }
1129
1130 EXPORT_SYMBOL(groups_free);
1131
1132 /* export the group_info to a user-space array */
1133 static int groups_to_user(gid_t __user *grouplist,
1134     struct group_info *group_info)
1135 {
1136         int i;
1137         unsigned int count = group_info->ngroups;
1138
1139         for (i = 0; i < group_info->nblocks; i++) {
1140                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1141                 unsigned int len = cp_count * sizeof(*grouplist);
1142
1143                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1144                         return -EFAULT;
1145
1146                 grouplist += NGROUPS_PER_BLOCK;
1147                 count -= cp_count;
1148         }
1149         return 0;
1150 }
1151
1152 /* fill a group_info from a user-space array - it must be allocated already */
1153 static int groups_from_user(struct group_info *group_info,
1154     gid_t __user *grouplist)
1155 {
1156         int i;
1157         unsigned int count = group_info->ngroups;
1158
1159         for (i = 0; i < group_info->nblocks; i++) {
1160                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1161                 unsigned int len = cp_count * sizeof(*grouplist);
1162
1163                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1164                         return -EFAULT;
1165
1166                 grouplist += NGROUPS_PER_BLOCK;
1167                 count -= cp_count;
1168         }
1169         return 0;
1170 }
1171
1172 /* a simple Shell sort */
1173 static void groups_sort(struct group_info *group_info)
1174 {
1175         int base, max, stride;
1176         int gidsetsize = group_info->ngroups;
1177
1178         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1179                 ; /* nothing */
1180         stride /= 3;
1181
1182         while (stride) {
1183                 max = gidsetsize - stride;
1184                 for (base = 0; base < max; base++) {
1185                         int left = base;
1186                         int right = left + stride;
1187                         gid_t tmp = GROUP_AT(group_info, right);
1188
1189                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1190                                 GROUP_AT(group_info, right) =
1191                                     GROUP_AT(group_info, left);
1192                                 right = left;
1193                                 left -= stride;
1194                         }
1195                         GROUP_AT(group_info, right) = tmp;
1196                 }
1197                 stride /= 3;
1198         }
1199 }
1200
1201 /* a simple bsearch */
1202 int groups_search(struct group_info *group_info, gid_t grp)
1203 {
1204         unsigned int left, right;
1205
1206         if (!group_info)
1207                 return 0;
1208
1209         left = 0;
1210         right = group_info->ngroups;
1211         while (left < right) {
1212                 unsigned int mid = (left+right)/2;
1213                 int cmp = grp - GROUP_AT(group_info, mid);
1214                 if (cmp > 0)
1215                         left = mid + 1;
1216                 else if (cmp < 0)
1217                         right = mid;
1218                 else
1219                         return 1;
1220         }
1221         return 0;
1222 }
1223
1224 /* validate and set current->group_info */
1225 int set_current_groups(struct group_info *group_info)
1226 {
1227         int retval;
1228         struct group_info *old_info;
1229
1230         retval = security_task_setgroups(group_info);
1231         if (retval)
1232                 return retval;
1233
1234         groups_sort(group_info);
1235         get_group_info(group_info);
1236
1237         task_lock(current);
1238         old_info = current->group_info;
1239         current->group_info = group_info;
1240         task_unlock(current);
1241
1242         put_group_info(old_info);
1243
1244         return 0;
1245 }
1246
1247 EXPORT_SYMBOL(set_current_groups);
1248
1249 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1250 {
1251         int i = 0;
1252
1253         /*
1254          *      SMP: Nobody else can change our grouplist. Thus we are
1255          *      safe.
1256          */
1257
1258         if (gidsetsize < 0)
1259                 return -EINVAL;
1260
1261         /* no need to grab task_lock here; it cannot change */
1262         i = current->group_info->ngroups;
1263         if (gidsetsize) {
1264                 if (i > gidsetsize) {
1265                         i = -EINVAL;
1266                         goto out;
1267                 }
1268                 if (groups_to_user(grouplist, current->group_info)) {
1269                         i = -EFAULT;
1270                         goto out;
1271                 }
1272         }
1273 out:
1274         return i;
1275 }
1276
1277 /*
1278  *      SMP: Our groups are copy-on-write. We can set them safely
1279  *      without another task interfering.
1280  */
1281
1282 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1283 {
1284         struct group_info *group_info;
1285         int retval;
1286
1287         if (!capable(CAP_SETGID))
1288                 return -EPERM;
1289         if ((unsigned)gidsetsize > NGROUPS_MAX)
1290                 return -EINVAL;
1291
1292         group_info = groups_alloc(gidsetsize);
1293         if (!group_info)
1294                 return -ENOMEM;
1295         retval = groups_from_user(group_info, grouplist);
1296         if (retval) {
1297                 put_group_info(group_info);
1298                 return retval;
1299         }
1300
1301         retval = set_current_groups(group_info);
1302         put_group_info(group_info);
1303
1304         return retval;
1305 }
1306
1307 /*
1308  * Check whether we're fsgid/egid or in the supplemental group..
1309  */
1310 int in_group_p(gid_t grp)
1311 {
1312         int retval = 1;
1313         if (grp != current->fsgid)
1314                 retval = groups_search(current->group_info, grp);
1315         return retval;
1316 }
1317
1318 EXPORT_SYMBOL(in_group_p);
1319
1320 int in_egroup_p(gid_t grp)
1321 {
1322         int retval = 1;
1323         if (grp != current->egid)
1324                 retval = groups_search(current->group_info, grp);
1325         return retval;
1326 }
1327
1328 EXPORT_SYMBOL(in_egroup_p);
1329
1330 DECLARE_RWSEM(uts_sem);
1331
1332 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1333 {
1334         int errno = 0;
1335
1336         down_read(&uts_sem);
1337         if (copy_to_user(name, utsname(), sizeof *name))
1338                 errno = -EFAULT;
1339         up_read(&uts_sem);
1340         return errno;
1341 }
1342
1343 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1344 {
1345         int errno;
1346         char tmp[__NEW_UTS_LEN];
1347
1348         if (!capable(CAP_SYS_ADMIN))
1349                 return -EPERM;
1350         if (len < 0 || len > __NEW_UTS_LEN)
1351                 return -EINVAL;
1352         down_write(&uts_sem);
1353         errno = -EFAULT;
1354         if (!copy_from_user(tmp, name, len)) {
1355                 memcpy(utsname()->nodename, tmp, len);
1356                 utsname()->nodename[len] = 0;
1357                 errno = 0;
1358         }
1359         up_write(&uts_sem);
1360         return errno;
1361 }
1362
1363 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1364
1365 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1366 {
1367         int i, errno;
1368
1369         if (len < 0)
1370                 return -EINVAL;
1371         down_read(&uts_sem);
1372         i = 1 + strlen(utsname()->nodename);
1373         if (i > len)
1374                 i = len;
1375         errno = 0;
1376         if (copy_to_user(name, utsname()->nodename, i))
1377                 errno = -EFAULT;
1378         up_read(&uts_sem);
1379         return errno;
1380 }
1381
1382 #endif
1383
1384 /*
1385  * Only setdomainname; getdomainname can be implemented by calling
1386  * uname()
1387  */
1388 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1389 {
1390         int errno;
1391         char tmp[__NEW_UTS_LEN];
1392
1393         if (!capable(CAP_SYS_ADMIN))
1394                 return -EPERM;
1395         if (len < 0 || len > __NEW_UTS_LEN)
1396                 return -EINVAL;
1397
1398         down_write(&uts_sem);
1399         errno = -EFAULT;
1400         if (!copy_from_user(tmp, name, len)) {
1401                 memcpy(utsname()->domainname, tmp, len);
1402                 utsname()->domainname[len] = 0;
1403                 errno = 0;
1404         }
1405         up_write(&uts_sem);
1406         return errno;
1407 }
1408
1409 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1410 {
1411         if (resource >= RLIM_NLIMITS)
1412                 return -EINVAL;
1413         else {
1414                 struct rlimit value;
1415                 task_lock(current->group_leader);
1416                 value = current->signal->rlim[resource];
1417                 task_unlock(current->group_leader);
1418                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1419         }
1420 }
1421
1422 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1423
1424 /*
1425  *      Back compatibility for getrlimit. Needed for some apps.
1426  */
1427
1428 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1429                 struct rlimit __user *, rlim)
1430 {
1431         struct rlimit x;
1432         if (resource >= RLIM_NLIMITS)
1433                 return -EINVAL;
1434
1435         task_lock(current->group_leader);
1436         x = current->signal->rlim[resource];
1437         task_unlock(current->group_leader);
1438         if (x.rlim_cur > 0x7FFFFFFF)
1439                 x.rlim_cur = 0x7FFFFFFF;
1440         if (x.rlim_max > 0x7FFFFFFF)
1441                 x.rlim_max = 0x7FFFFFFF;
1442         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1443 }
1444
1445 #endif
1446
1447 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1448 {
1449         struct rlimit new_rlim, *old_rlim;
1450         unsigned long it_prof_secs;
1451         int retval;
1452
1453         if (resource >= RLIM_NLIMITS)
1454                 return -EINVAL;
1455         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1456                 return -EFAULT;
1457         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1458                 return -EINVAL;
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         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1464                 return -EPERM;
1465
1466         retval = security_task_setrlimit(resource, &new_rlim);
1467         if (retval)
1468                 return retval;
1469
1470         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1471                 /*
1472                  * The caller is asking for an immediate RLIMIT_CPU
1473                  * expiry.  But we use the zero value to mean "it was
1474                  * never set".  So let's cheat and make it one second
1475                  * instead
1476                  */
1477                 new_rlim.rlim_cur = 1;
1478         }
1479
1480         task_lock(current->group_leader);
1481         *old_rlim = new_rlim;
1482         task_unlock(current->group_leader);
1483
1484         if (resource != RLIMIT_CPU)
1485                 goto out;
1486
1487         /*
1488          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1489          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1490          * very long-standing error, and fixing it now risks breakage of
1491          * applications, so we live with it
1492          */
1493         if (new_rlim.rlim_cur == RLIM_INFINITY)
1494                 goto out;
1495
1496         it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1497         if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1498                 unsigned long rlim_cur = new_rlim.rlim_cur;
1499                 cputime_t cputime;
1500
1501                 cputime = secs_to_cputime(rlim_cur);
1502                 read_lock(&tasklist_lock);
1503                 spin_lock_irq(&current->sighand->siglock);
1504                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1505                 spin_unlock_irq(&current->sighand->siglock);
1506                 read_unlock(&tasklist_lock);
1507         }
1508 out:
1509         return 0;
1510 }
1511
1512 /*
1513  * It would make sense to put struct rusage in the task_struct,
1514  * except that would make the task_struct be *really big*.  After
1515  * task_struct gets moved into malloc'ed memory, it would
1516  * make sense to do this.  It will make moving the rest of the information
1517  * a lot simpler!  (Which we're not doing right now because we're not
1518  * measuring them yet).
1519  *
1520  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1521  * races with threads incrementing their own counters.  But since word
1522  * reads are atomic, we either get new values or old values and we don't
1523  * care which for the sums.  We always take the siglock to protect reading
1524  * the c* fields from p->signal from races with exit.c updating those
1525  * fields when reaping, so a sample either gets all the additions of a
1526  * given child after it's reaped, or none so this sample is before reaping.
1527  *
1528  * Locking:
1529  * We need to take the siglock for CHILDEREN, SELF and BOTH
1530  * for  the cases current multithreaded, non-current single threaded
1531  * non-current multithreaded.  Thread traversal is now safe with
1532  * the siglock held.
1533  * Strictly speaking, we donot need to take the siglock if we are current and
1534  * single threaded,  as no one else can take our signal_struct away, no one
1535  * else can  reap the  children to update signal->c* counters, and no one else
1536  * can race with the signal-> fields. If we do not take any lock, the
1537  * signal-> fields could be read out of order while another thread was just
1538  * exiting. So we should  place a read memory barrier when we avoid the lock.
1539  * On the writer side,  write memory barrier is implied in  __exit_signal
1540  * as __exit_signal releases  the siglock spinlock after updating the signal->
1541  * fields. But we don't do this yet to keep things simple.
1542  *
1543  */
1544
1545 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r,
1546                                      cputime_t *utimep, cputime_t *stimep)
1547 {
1548         *utimep = cputime_add(*utimep, t->utime);
1549         *stimep = cputime_add(*stimep, t->stime);
1550         r->ru_nvcsw += t->nvcsw;
1551         r->ru_nivcsw += t->nivcsw;
1552         r->ru_minflt += t->min_flt;
1553         r->ru_majflt += t->maj_flt;
1554         r->ru_inblock += task_io_get_inblock(t);
1555         r->ru_oublock += task_io_get_oublock(t);
1556 }
1557
1558 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1559 {
1560         struct task_struct *t;
1561         unsigned long flags;
1562         cputime_t utime, stime;
1563
1564         memset((char *) r, 0, sizeof *r);
1565         utime = stime = cputime_zero;
1566
1567         if (who == RUSAGE_THREAD) {
1568                 accumulate_thread_rusage(p, r, &utime, &stime);
1569                 goto out;
1570         }
1571
1572         if (!lock_task_sighand(p, &flags))
1573                 return;
1574
1575         switch (who) {
1576                 case RUSAGE_BOTH:
1577                 case RUSAGE_CHILDREN:
1578                         utime = p->signal->cutime;
1579                         stime = p->signal->cstime;
1580                         r->ru_nvcsw = p->signal->cnvcsw;
1581                         r->ru_nivcsw = p->signal->cnivcsw;
1582                         r->ru_minflt = p->signal->cmin_flt;
1583                         r->ru_majflt = p->signal->cmaj_flt;
1584                         r->ru_inblock = p->signal->cinblock;
1585                         r->ru_oublock = p->signal->coublock;
1586
1587                         if (who == RUSAGE_CHILDREN)
1588                                 break;
1589
1590                 case RUSAGE_SELF:
1591                         utime = cputime_add(utime, p->signal->utime);
1592                         stime = cputime_add(stime, p->signal->stime);
1593                         r->ru_nvcsw += p->signal->nvcsw;
1594                         r->ru_nivcsw += p->signal->nivcsw;
1595                         r->ru_minflt += p->signal->min_flt;
1596                         r->ru_majflt += p->signal->maj_flt;
1597                         r->ru_inblock += p->signal->inblock;
1598                         r->ru_oublock += p->signal->oublock;
1599                         t = p;
1600                         do {
1601                                 accumulate_thread_rusage(t, r, &utime, &stime);
1602                                 t = next_thread(t);
1603                         } while (t != p);
1604                         break;
1605
1606                 default:
1607                         BUG();
1608         }
1609         unlock_task_sighand(p, &flags);
1610
1611 out:
1612         cputime_to_timeval(utime, &r->ru_utime);
1613         cputime_to_timeval(stime, &r->ru_stime);
1614 }
1615
1616 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1617 {
1618         struct rusage r;
1619         k_getrusage(p, who, &r);
1620         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1621 }
1622
1623 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1624 {
1625         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1626             who != RUSAGE_THREAD)
1627                 return -EINVAL;
1628         return getrusage(current, who, ru);
1629 }
1630
1631 SYSCALL_DEFINE1(umask, int, mask)
1632 {
1633         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1634         return mask;
1635 }
1636
1637 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1638                 unsigned long, arg4, unsigned long, arg5)
1639 {
1640         long error = 0;
1641
1642         if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
1643                 return error;
1644
1645         switch (option) {
1646                 case PR_SET_PDEATHSIG:
1647                         if (!valid_signal(arg2)) {
1648                                 error = -EINVAL;
1649                                 break;
1650                         }
1651                         current->pdeath_signal = arg2;
1652                         break;
1653                 case PR_GET_PDEATHSIG:
1654                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1655                         break;
1656                 case PR_GET_DUMPABLE:
1657                         error = get_dumpable(current->mm);
1658                         break;
1659                 case PR_SET_DUMPABLE:
1660                         if (arg2 < 0 || arg2 > 1) {
1661                                 error = -EINVAL;
1662                                 break;
1663                         }
1664                         set_dumpable(current->mm, arg2);
1665                         break;
1666
1667                 case PR_SET_UNALIGN:
1668                         error = SET_UNALIGN_CTL(current, arg2);
1669                         break;
1670                 case PR_GET_UNALIGN:
1671                         error = GET_UNALIGN_CTL(current, arg2);
1672                         break;
1673                 case PR_SET_FPEMU:
1674                         error = SET_FPEMU_CTL(current, arg2);
1675                         break;
1676                 case PR_GET_FPEMU:
1677                         error = GET_FPEMU_CTL(current, arg2);
1678                         break;
1679                 case PR_SET_FPEXC:
1680                         error = SET_FPEXC_CTL(current, arg2);
1681                         break;
1682                 case PR_GET_FPEXC:
1683                         error = GET_FPEXC_CTL(current, arg2);
1684                         break;
1685                 case PR_GET_TIMING:
1686                         error = PR_TIMING_STATISTICAL;
1687                         break;
1688                 case PR_SET_TIMING:
1689                         if (arg2 != PR_TIMING_STATISTICAL)
1690                                 error = -EINVAL;
1691                         break;
1692
1693                 case PR_SET_NAME: {
1694                         struct task_struct *me = current;
1695                         unsigned char ncomm[sizeof(me->comm)];
1696
1697                         ncomm[sizeof(me->comm)-1] = 0;
1698                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1699                                                 sizeof(me->comm)-1) < 0)
1700                                 return -EFAULT;
1701                         set_task_comm(me, ncomm);
1702                         return 0;
1703                 }
1704                 case PR_GET_NAME: {
1705                         struct task_struct *me = current;
1706                         unsigned char tcomm[sizeof(me->comm)];
1707
1708                         get_task_comm(tcomm, me);
1709                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1710                                 return -EFAULT;
1711                         return 0;
1712                 }
1713                 case PR_GET_ENDIAN:
1714                         error = GET_ENDIAN(current, arg2);
1715                         break;
1716                 case PR_SET_ENDIAN:
1717                         error = SET_ENDIAN(current, arg2);
1718                         break;
1719
1720                 case PR_GET_SECCOMP:
1721                         error = prctl_get_seccomp();
1722                         break;
1723                 case PR_SET_SECCOMP:
1724                         error = prctl_set_seccomp(arg2);
1725                         break;
1726                 case PR_GET_TSC:
1727                         error = GET_TSC_CTL(arg2);
1728                         break;
1729                 case PR_SET_TSC:
1730                         error = SET_TSC_CTL(arg2);
1731                         break;
1732                 default:
1733                         error = -EINVAL;
1734                         break;
1735         }
1736         return error;
1737 }
1738
1739 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1740                 struct getcpu_cache __user *, unused)
1741 {
1742         int err = 0;
1743         int cpu = raw_smp_processor_id();
1744         if (cpup)
1745                 err |= put_user(cpu, cpup);
1746         if (nodep)
1747                 err |= put_user(cpu_to_node(cpu), nodep);
1748         return err ? -EFAULT : 0;
1749 }
1750
1751 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1752
1753 static void argv_cleanup(char **argv, char **envp)
1754 {
1755         argv_free(argv);
1756 }
1757
1758 /**
1759  * orderly_poweroff - Trigger an orderly system poweroff
1760  * @force: force poweroff if command execution fails
1761  *
1762  * This may be called from any context to trigger a system shutdown.
1763  * If the orderly shutdown fails, it will force an immediate shutdown.
1764  */
1765 int orderly_poweroff(bool force)
1766 {
1767         int argc;
1768         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1769         static char *envp[] = {
1770                 "HOME=/",
1771                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1772                 NULL
1773         };
1774         int ret = -ENOMEM;
1775         struct subprocess_info *info;
1776
1777         if (argv == NULL) {
1778                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1779                        __func__, poweroff_cmd);
1780                 goto out;
1781         }
1782
1783         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1784         if (info == NULL) {
1785                 argv_free(argv);
1786                 goto out;
1787         }
1788
1789         call_usermodehelper_setcleanup(info, argv_cleanup);
1790
1791         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1792
1793   out:
1794         if (ret && force) {
1795                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1796                        "forcing the issue\n");
1797
1798                 /* I guess this should try to kick off some daemon to
1799                    sync and poweroff asap.  Or not even bother syncing
1800                    if we're doing an emergency shutdown? */
1801                 emergency_sync();
1802                 kernel_power_off();
1803         }
1804
1805         return ret;
1806 }
1807 EXPORT_SYMBOL_GPL(orderly_poweroff);