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/notifier.h>
  12 #include <linux/reboot.h>
  13 #include <linux/prctl.h>
  14 #include <linux/highuid.h>
  15 #include <linux/fs.h>
  16 #include <linux/perf_event.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 #include <linux/personality.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/fs_struct.h>
  39 #include <linux/gfp.h>
  40 #include <linux/syscore_ops.h>
  41 #include <linux/version.h>
  42 #include <linux/ctype.h>
  43
  44 #include <linux/compat.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/kprobes.h>
  47 #include <linux/user_namespace.h>
  48
  49 #include <linux/kmsg_dump.h>
  50 /* Move somewhere else to avoid recompiling? */
  51 #include <generated/utsrelease.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/io.h>
  55 #include <asm/unistd.h>
  56
  57 #ifndef SET_UNALIGN_CTL
  58 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  59 #endif
  60 #ifndef GET_UNALIGN_CTL
  61 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  62 #endif
  63 #ifndef SET_FPEMU_CTL
  64 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  65 #endif
  66 #ifndef GET_FPEMU_CTL
  67 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  68 #endif
  69 #ifndef SET_FPEXC_CTL
  70 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  71 #endif
  72 #ifndef GET_FPEXC_CTL
  73 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  74 #endif
  75 #ifndef GET_ENDIAN
  76 # define GET_ENDIAN(a,b)        (-EINVAL)
  77 #endif
  78 #ifndef SET_ENDIAN
  79 # define SET_ENDIAN(a,b)        (-EINVAL)
  80 #endif
  81 #ifndef GET_TSC_CTL
  82 # define GET_TSC_CTL(a)         (-EINVAL)
  83 #endif
  84 #ifndef SET_TSC_CTL
  85 # define SET_TSC_CTL(a)         (-EINVAL)
  86 #endif
  87
  88 /*
  89  * this is where the system-wide overflow UID and GID are defined, for
  90  * architectures that now have 32-bit UID/GID but didn't in the past
  91  */
  92
  93 int overflowuid = DEFAULT_OVERFLOWUID;
  94 int overflowgid = DEFAULT_OVERFLOWGID;
  95
  96 #ifdef CONFIG_UID16
  97 EXPORT_SYMBOL(overflowuid);
  98 EXPORT_SYMBOL(overflowgid);
  99 #endif
 100
 101 /*
 102  * the same as above, but for filesystems which can only store a 16-bit
 103  * UID and GID. as such, this is needed on all architectures
 104  */
 105
 106 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 107 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 108
 109 EXPORT_SYMBOL(fs_overflowuid);
 110 EXPORT_SYMBOL(fs_overflowgid);
 111
 112 /*
 113  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 114  */
 115
 116 int C_A_D = 1;
 117 struct pid *cad_pid;
 118 EXPORT_SYMBOL(cad_pid);
 119
 120 /*
 121  * If set, this is used for preparing the system to power off.
 122  */
 123
 124 void (*pm_power_off_prepare)(void);
 125
 126 /*
 127  * Returns true if current's euid is same as p's uid or euid,
 128  * or has CAP_SYS_NICE to p's user_ns.
 129  *
 130  * Called with rcu_read_lock, creds are safe
 131  */
 132 static bool set_one_prio_perm(struct task_struct *p)
 133 {
 134         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 135
 136         if (pcred->user->user_ns == cred->user->user_ns &&
 137             (pcred->uid  == cred->euid ||
 138              pcred->euid == cred->euid))
 139                 return true;
 140         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 141                 return true;
 142         return false;
 143 }
 144
 145 /*
 146  * set the priority of a task
 147  * - the caller must hold the RCU read lock
 148  */
 149 static int set_one_prio(struct task_struct *p, int niceval, int error)
 150 {
 151         int no_nice;
 152
 153         if (!set_one_prio_perm(p)) {
 154                 error = -EPERM;
 155                 goto out;
 156         }
 157         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 158                 error = -EACCES;
 159                 goto out;
 160         }
 161         no_nice = security_task_setnice(p, niceval);
 162         if (no_nice) {
 163                 error = no_nice;
 164                 goto out;
 165         }
 166         if (error == -ESRCH)
 167                 error = 0;
 168         set_user_nice(p, niceval);
 169 out:
 170         return error;
 171 }
 172
 173 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 174 {
 175         struct task_struct *g, *p;
 176         struct user_struct *user;
 177         const struct cred *cred = current_cred();
 178         int error = -EINVAL;
 179         struct pid *pgrp;
 180
 181         if (which > PRIO_USER || which < PRIO_PROCESS)
 182                 goto out;
 183
 184         /* normalize: avoid signed division (rounding problems) */
 185         error = -ESRCH;
 186         if (niceval < -20)
 187                 niceval = -20;
 188         if (niceval > 19)
 189                 niceval = 19;
 190
 191         rcu_read_lock();
 192         read_lock(&tasklist_lock);
 193         switch (which) {
 194                 case PRIO_PROCESS:
 195                         if (who)
 196                                 p = find_task_by_vpid(who);
 197                         else
 198                                 p = current;
 199                         if (p)
 200                                 error = set_one_prio(p, niceval, error);
 201                         break;
 202                 case PRIO_PGRP:
 203                         if (who)
 204                                 pgrp = find_vpid(who);
 205                         else
 206                                 pgrp = task_pgrp(current);
 207                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 208                                 error = set_one_prio(p, niceval, error);
 209                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 210                         break;
 211                 case PRIO_USER:
 212                         user = (struct user_struct *) cred->user;
 213                         if (!who)
 214                                 who = cred->uid;
 215                         else if ((who != cred->uid) &&
 216                                  !(user = find_user(who)))
 217                                 goto out_unlock;        /* No processes for this user */
 218
 219                         do_each_thread(g, p) {
 220                                 if (__task_cred(p)->uid == who)
 221                                         error = set_one_prio(p, niceval, error);
 222                         } while_each_thread(g, p);
 223                         if (who != cred->uid)
 224                                 free_uid(user);         /* For find_user() */
 225                         break;
 226         }
 227 out_unlock:
 228         read_unlock(&tasklist_lock);
 229         rcu_read_unlock();
 230 out:
 231         return error;
 232 }
 233
 234 /*
 235  * Ugh. To avoid negative return values, "getpriority()" will
 236  * not return the normal nice-value, but a negated value that
 237  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 238  * to stay compatible.
 239  */
 240 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 241 {
 242         struct task_struct *g, *p;
 243         struct user_struct *user;
 244         const struct cred *cred = current_cred();
 245         long niceval, retval = -ESRCH;
 246         struct pid *pgrp;
 247
 248         if (which > PRIO_USER || which < PRIO_PROCESS)
 249                 return -EINVAL;
 250
 251         rcu_read_lock();
 252         read_lock(&tasklist_lock);
 253         switch (which) {
 254                 case PRIO_PROCESS:
 255                         if (who)
 256                                 p = find_task_by_vpid(who);
 257                         else
 258                                 p = current;
 259                         if (p) {
 260                                 niceval = 20 - task_nice(p);
 261                                 if (niceval > retval)
 262                                         retval = niceval;
 263                         }
 264                         break;
 265                 case PRIO_PGRP:
 266                         if (who)
 267                                 pgrp = find_vpid(who);
 268                         else
 269                                 pgrp = task_pgrp(current);
 270                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 271                                 niceval = 20 - task_nice(p);
 272                                 if (niceval > retval)
 273                                         retval = niceval;
 274                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 275                         break;
 276                 case PRIO_USER:
 277                         user = (struct user_struct *) cred->user;
 278                         if (!who)
 279                                 who = cred->uid;
 280                         else if ((who != cred->uid) &&
 281                                  !(user = find_user(who)))
 282                                 goto out_unlock;        /* No processes for this user */
 283
 284                         do_each_thread(g, p) {
 285                                 if (__task_cred(p)->uid == who) {
 286                                         niceval = 20 - task_nice(p);
 287                                         if (niceval > retval)
 288                                                 retval = niceval;
 289                                 }
 290                         } while_each_thread(g, p);
 291                         if (who != cred->uid)
 292                                 free_uid(user);         /* for find_user() */
 293                         break;
 294         }
 295 out_unlock:
 296         read_unlock(&tasklist_lock);
 297         rcu_read_unlock();
 298
 299         return retval;
 300 }
 301
 302 /**
 303  *      emergency_restart - reboot the system
 304  *
 305  *      Without shutting down any hardware or taking any locks
 306  *      reboot the system.  This is called when we know we are in
 307  *      trouble so this is our best effort to reboot.  This is
 308  *      safe to call in interrupt context.
 309  */
 310 void emergency_restart(void)
 311 {
 312         kmsg_dump(KMSG_DUMP_EMERG);
 313         machine_emergency_restart();
 314 }
 315 EXPORT_SYMBOL_GPL(emergency_restart);
 316
 317 void kernel_restart_prepare(char *cmd)
 318 {
 319         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 320         system_state = SYSTEM_RESTART;
 321         usermodehelper_disable();
 322         device_shutdown();
 323 }
 324
 325 /**
 326  *      kernel_restart - reboot the system
 327  *      @cmd: pointer to buffer containing command to execute for restart
 328  *              or %NULL
 329  *
 330  *      Shutdown everything and perform a clean reboot.
 331  *      This is not safe to call in interrupt context.
 332  */
 333 void kernel_restart(char *cmd)
 334 {
 335         kernel_restart_prepare(cmd);
 336         if (pm_power_off_prepare)
 337                 pm_power_off_prepare();
 338         disable_nonboot_cpus();
 339         syscore_shutdown();
 340         if (!cmd)
 341                 printk(KERN_EMERG "Restarting system.\n");
 342         else{
 343                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 344                 printk(KERN_EMERG "pid = %d name:%s\n", task_tgid_vnr(current), current->comm);
 345         }
 346         kmsg_dump(KMSG_DUMP_RESTART);
 347         machine_restart(cmd);
 348 }
 349 EXPORT_SYMBOL_GPL(kernel_restart);
 350
 351 static void kernel_shutdown_prepare(enum system_states state)
 352 {
 353         blocking_notifier_call_chain(&reboot_notifier_list,
 354                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 355         system_state = state;
 356         usermodehelper_disable();
 357         device_shutdown();
 358 }
 359 /**
 360  *      kernel_halt - halt the system
 361  *
 362  *      Shutdown everything and perform a clean system halt.
 363  */
 364 void kernel_halt(void)
 365 {
 366         kernel_shutdown_prepare(SYSTEM_HALT);
 367         disable_nonboot_cpus();
 368         syscore_shutdown();
 369         printk(KERN_EMERG "System halted.\n");
 370         kmsg_dump(KMSG_DUMP_HALT);
 371         machine_halt();
 372 }
 373
 374 EXPORT_SYMBOL_GPL(kernel_halt);
 375
 376 /**
 377  *      kernel_power_off - power_off the system
 378  *
 379  *      Shutdown everything and perform a clean system power_off.
 380  */
 381 void kernel_power_off(void)
 382 {
 383         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 384         if (pm_power_off_prepare)
 385                 pm_power_off_prepare();
 386         disable_nonboot_cpus();
 387         syscore_shutdown();
 388         printk(KERN_EMERG "Power down.\n");
 389         kmsg_dump(KMSG_DUMP_POWEROFF);
 390         machine_power_off();
 391 }
 392 EXPORT_SYMBOL_GPL(kernel_power_off);
 393
 394 static DEFINE_MUTEX(reboot_mutex);
 395
 396 /*
 397  * Reboot system call: for obvious reasons only root may call it,
 398  * and even root needs to set up some magic numbers in the registers
 399  * so that some mistake won't make this reboot the whole machine.
 400  * You can also set the meaning of the ctrl-alt-del-key here.
 401  *
 402  * reboot doesn't sync: do that yourself before calling this.
 403  */
 404 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 405                 void __user *, arg)
 406 {
 407         char buffer[256];
 408         int ret = 0;
 409
 410         /* We only trust the superuser with rebooting the system. */
 411         if (!capable(CAP_SYS_BOOT))
 412                 return -EPERM;
 413
 414         /* For safety, we require "magic" arguments. */
 415         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 416             (magic2 != LINUX_REBOOT_MAGIC2 &&
 417                         magic2 != LINUX_REBOOT_MAGIC2A &&
 418                         magic2 != LINUX_REBOOT_MAGIC2B &&
 419                         magic2 != LINUX_REBOOT_MAGIC2C))
 420                 return -EINVAL;
 421
 422         /* Instead of trying to make the power_off code look like
 423          * halt when pm_power_off is not set do it the easy way.
 424          */
 425         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 426                 cmd = LINUX_REBOOT_CMD_HALT;
 427
 428         mutex_lock(&reboot_mutex);
 429         switch (cmd) {
 430         case LINUX_REBOOT_CMD_RESTART:
 431                 kernel_restart(NULL);
 432                 break;
 433
 434         case LINUX_REBOOT_CMD_CAD_ON:
 435                 C_A_D = 1;
 436                 break;
 437
 438         case LINUX_REBOOT_CMD_CAD_OFF:
 439                 C_A_D = 0;
 440                 break;
 441
 442         case LINUX_REBOOT_CMD_HALT:
 443                 kernel_halt();
 444                 do_exit(0);
 445                 panic("cannot halt");
 446
 447         case LINUX_REBOOT_CMD_POWER_OFF:
 448                 kernel_power_off();
 449                 do_exit(0);
 450                 break;
 451
 452         case LINUX_REBOOT_CMD_RESTART2:
 453                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 454                         ret = -EFAULT;
 455                         break;
 456                 }
 457                 buffer[sizeof(buffer) - 1] = '\0';
 458
 459                 kernel_restart(buffer);
 460                 break;
 461
 462 #ifdef CONFIG_KEXEC
 463         case LINUX_REBOOT_CMD_KEXEC:
 464                 ret = kernel_kexec();
 465                 break;
 466 #endif
 467
 468 #ifdef CONFIG_HIBERNATION
 469         case LINUX_REBOOT_CMD_SW_SUSPEND:
 470                 ret = hibernate();
 471                 break;
 472 #endif
 473
 474         default:
 475                 ret = -EINVAL;
 476                 break;
 477         }
 478         mutex_unlock(&reboot_mutex);
 479         return ret;
 480 }
 481
 482 static void deferred_cad(struct work_struct *dummy)
 483 {
 484         kernel_restart(NULL);
 485 }
 486
 487 /*
 488  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 489  * As it's called within an interrupt, it may NOT sync: the only choice
 490  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 491  */
 492 void ctrl_alt_del(void)
 493 {
 494         static DECLARE_WORK(cad_work, deferred_cad);
 495
 496         if (C_A_D)
 497                 schedule_work(&cad_work);
 498         else
 499                 kill_cad_pid(SIGINT, 1);
 500 }
 501
 502 /*
 503  * Unprivileged users may change the real gid to the effective gid
 504  * or vice versa.  (BSD-style)
 505  *
 506  * If you set the real gid at all, or set the effective gid to a value not
 507  * equal to the real gid, then the saved gid is set to the new effective gid.
 508  *
 509  * This makes it possible for a setgid program to completely drop its
 510  * privileges, which is often a useful assertion to make when you are doing
 511  * a security audit over a program.
 512  *
 513  * The general idea is that a program which uses just setregid() will be
 514  * 100% compatible with BSD.  A program which uses just setgid() will be
 515  * 100% compatible with POSIX with saved IDs.
 516  *
 517  * SMP: There are not races, the GIDs are checked only by filesystem
 518  *      operations (as far as semantic preservation is concerned).
 519  */
 520 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 521 {
 522         const struct cred *old;
 523         struct cred *new;
 524         int retval;
 525
 526         new = prepare_creds();
 527         if (!new)
 528                 return -ENOMEM;
 529         old = current_cred();
 530
 531         retval = -EPERM;
 532         if (rgid != (gid_t) -1) {
 533                 if (old->gid == rgid ||
 534                     old->egid == rgid ||
 535                     nsown_capable(CAP_SETGID))
 536                         new->gid = rgid;
 537                 else
 538                         goto error;
 539         }
 540         if (egid != (gid_t) -1) {
 541                 if (old->gid == egid ||
 542                     old->egid == egid ||
 543                     old->sgid == egid ||
 544                     nsown_capable(CAP_SETGID))
 545                         new->egid = egid;
 546                 else
 547                         goto error;
 548         }
 549
 550         if (rgid != (gid_t) -1 ||
 551             (egid != (gid_t) -1 && egid != old->gid))
 552                 new->sgid = new->egid;
 553         new->fsgid = new->egid;
 554
 555         return commit_creds(new);
 556
 557 error:
 558         abort_creds(new);
 559         return retval;
 560 }
 561
 562 /*
 563  * setgid() is implemented like SysV w/ SAVED_IDS
 564  *
 565  * SMP: Same implicit races as above.
 566  */
 567 SYSCALL_DEFINE1(setgid, gid_t, gid)
 568 {
 569         const struct cred *old;
 570         struct cred *new;
 571         int retval;
 572
 573         new = prepare_creds();
 574         if (!new)
 575                 return -ENOMEM;
 576         old = current_cred();
 577
 578         retval = -EPERM;
 579         if (nsown_capable(CAP_SETGID))
 580                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 581         else if (gid == old->gid || gid == old->sgid)
 582                 new->egid = new->fsgid = gid;
 583         else
 584                 goto error;
 585
 586         return commit_creds(new);
 587
 588 error:
 589         abort_creds(new);
 590         return retval;
 591 }
 592
 593 /*
 594  * change the user struct in a credentials set to match the new UID
 595  */
 596 static int set_user(struct cred *new)
 597 {
 598         struct user_struct *new_user;
 599
 600         new_user = alloc_uid(current_user_ns(), new->uid);
 601         if (!new_user)
 602                 return -EAGAIN;
 603
 604         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 605                         new_user != INIT_USER) {
 606                 free_uid(new_user);
 607                 return -EAGAIN;
 608         }
 609
 610         free_uid(new->user);
 611         new->user = new_user;
 612         return 0;
 613 }
 614
 615 /*
 616  * Unprivileged users may change the real uid to the effective uid
 617  * or vice versa.  (BSD-style)
 618  *
 619  * If you set the real uid at all, or set the effective uid to a value not
 620  * equal to the real uid, then the saved uid is set to the new effective uid.
 621  *
 622  * This makes it possible for a setuid program to completely drop its
 623  * privileges, which is often a useful assertion to make when you are doing
 624  * a security audit over a program.
 625  *
 626  * The general idea is that a program which uses just setreuid() will be
 627  * 100% compatible with BSD.  A program which uses just setuid() will be
 628  * 100% compatible with POSIX with saved IDs.
 629  */
 630 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 631 {
 632         const struct cred *old;
 633         struct cred *new;
 634         int retval;
 635
 636         new = prepare_creds();
 637         if (!new)
 638                 return -ENOMEM;
 639         old = current_cred();
 640
 641         retval = -EPERM;
 642         if (ruid != (uid_t) -1) {
 643                 new->uid = ruid;
 644                 if (old->uid != ruid &&
 645                     old->euid != ruid &&
 646                     !nsown_capable(CAP_SETUID))
 647                         goto error;
 648         }
 649
 650         if (euid != (uid_t) -1) {
 651                 new->euid = euid;
 652                 if (old->uid != euid &&
 653                     old->euid != euid &&
 654                     old->suid != euid &&
 655                     !nsown_capable(CAP_SETUID))
 656                         goto error;
 657         }
 658
 659         if (new->uid != old->uid) {
 660                 retval = set_user(new);
 661                 if (retval < 0)
 662                         goto error;
 663         }
 664         if (ruid != (uid_t) -1 ||
 665             (euid != (uid_t) -1 && euid != old->uid))
 666                 new->suid = new->euid;
 667         new->fsuid = new->euid;
 668
 669         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 670         if (retval < 0)
 671                 goto error;
 672
 673         return commit_creds(new);
 674
 675 error:
 676         abort_creds(new);
 677         return retval;
 678 }
 679
 680 /*
 681  * setuid() is implemented like SysV with SAVED_IDS
 682  *
 683  * Note that SAVED_ID's is deficient in that a setuid root program
 684  * like sendmail, for example, cannot set its uid to be a normal
 685  * user and then switch back, because if you're root, setuid() sets
 686  * the saved uid too.  If you don't like this, blame the bright people
 687  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 688  * will allow a root program to temporarily drop privileges and be able to
 689  * regain them by swapping the real and effective uid.
 690  */
 691 SYSCALL_DEFINE1(setuid, uid_t, uid)
 692 {
 693         const struct cred *old;
 694         struct cred *new;
 695         int retval;
 696
 697         new = prepare_creds();
 698         if (!new)
 699                 return -ENOMEM;
 700         old = current_cred();
 701
 702         retval = -EPERM;
 703         if (nsown_capable(CAP_SETUID)) {
 704                 new->suid = new->uid = uid;
 705                 if (uid != old->uid) {
 706                         retval = set_user(new);
 707                         if (retval < 0)
 708                                 goto error;
 709                 }
 710         } else if (uid != old->uid && uid != new->suid) {
 711                 goto error;
 712         }
 713
 714         new->fsuid = new->euid = uid;
 715
 716         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 717         if (retval < 0)
 718                 goto error;
 719
 720         return commit_creds(new);
 721
 722 error:
 723         abort_creds(new);
 724         return retval;
 725 }
 726
 727
 728 /*
 729  * This function implements a generic ability to update ruid, euid,
 730  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 731  */
 732 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 733 {
 734         const struct cred *old;
 735         struct cred *new;
 736         int retval;
 737
 738         new = prepare_creds();
 739         if (!new)
 740                 return -ENOMEM;
 741
 742         old = current_cred();
 743
 744         retval = -EPERM;
 745         if (!nsown_capable(CAP_SETUID)) {
 746                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 747                     ruid != old->euid  && ruid != old->suid)
 748                         goto error;
 749                 if (euid != (uid_t) -1 && euid != old->uid &&
 750                     euid != old->euid  && euid != old->suid)
 751                         goto error;
 752                 if (suid != (uid_t) -1 && suid != old->uid &&
 753                     suid != old->euid  && suid != old->suid)
 754                         goto error;
 755         }
 756
 757         if (ruid != (uid_t) -1) {
 758                 new->uid = ruid;
 759                 if (ruid != old->uid) {
 760                         retval = set_user(new);
 761                         if (retval < 0)
 762                                 goto error;
 763                 }
 764         }
 765         if (euid != (uid_t) -1)
 766                 new->euid = euid;
 767         if (suid != (uid_t) -1)
 768                 new->suid = suid;
 769         new->fsuid = new->euid;
 770
 771         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 772         if (retval < 0)
 773                 goto error;
 774
 775         return commit_creds(new);
 776
 777 error:
 778         abort_creds(new);
 779         return retval;
 780 }
 781
 782 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 783 {
 784         const struct cred *cred = current_cred();
 785         int retval;
 786
 787         if (!(retval   = put_user(cred->uid,  ruid)) &&
 788             !(retval   = put_user(cred->euid, euid)))
 789                 retval = put_user(cred->suid, suid);
 790
 791         return retval;
 792 }
 793
 794 /*
 795  * Same as above, but for rgid, egid, sgid.
 796  */
 797 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 798 {
 799         const struct cred *old;
 800         struct cred *new;
 801         int retval;
 802
 803         new = prepare_creds();
 804         if (!new)
 805                 return -ENOMEM;
 806         old = current_cred();
 807
 808         retval = -EPERM;
 809         if (!nsown_capable(CAP_SETGID)) {
 810                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 811                     rgid != old->egid  && rgid != old->sgid)
 812                         goto error;
 813                 if (egid != (gid_t) -1 && egid != old->gid &&
 814                     egid != old->egid  && egid != old->sgid)
 815                         goto error;
 816                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 817                     sgid != old->egid  && sgid != old->sgid)
 818                         goto error;
 819         }
 820
 821         if (rgid != (gid_t) -1)
 822                 new->gid = rgid;
 823         if (egid != (gid_t) -1)
 824                 new->egid = egid;
 825         if (sgid != (gid_t) -1)
 826                 new->sgid = sgid;
 827         new->fsgid = new->egid;
 828
 829         return commit_creds(new);
 830
 831 error:
 832         abort_creds(new);
 833         return retval;
 834 }
 835
 836 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 837 {
 838         const struct cred *cred = current_cred();
 839         int retval;
 840
 841         if (!(retval   = put_user(cred->gid,  rgid)) &&
 842             !(retval   = put_user(cred->egid, egid)))
 843                 retval = put_user(cred->sgid, sgid);
 844
 845         return retval;
 846 }
 847
 848
 849 /*
 850  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 851  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 852  * whatever uid it wants to). It normally shadows "euid", except when
 853  * explicitly set by setfsuid() or for access..
 854  */
 855 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 856 {
 857         const struct cred *old;
 858         struct cred *new;
 859         uid_t old_fsuid;
 860
 861         new = prepare_creds();
 862         if (!new)
 863                 return current_fsuid();
 864         old = current_cred();
 865         old_fsuid = old->fsuid;
 866
 867         if (uid == old->uid  || uid == old->euid  ||
 868             uid == old->suid || uid == old->fsuid ||
 869             nsown_capable(CAP_SETUID)) {
 870                 if (uid != old_fsuid) {
 871                         new->fsuid = uid;
 872                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 873                                 goto change_okay;
 874                 }
 875         }
 876
 877         abort_creds(new);
 878         return old_fsuid;
 879
 880 change_okay:
 881         commit_creds(new);
 882         return old_fsuid;
 883 }
 884
 885 /*
 886  * Samma på svenska..
 887  */
 888 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 889 {
 890         const struct cred *old;
 891         struct cred *new;
 892         gid_t old_fsgid;
 893
 894         new = prepare_creds();
 895         if (!new)
 896                 return current_fsgid();
 897         old = current_cred();
 898         old_fsgid = old->fsgid;
 899
 900         if (gid == old->gid  || gid == old->egid  ||
 901             gid == old->sgid || gid == old->fsgid ||
 902             nsown_capable(CAP_SETGID)) {
 903                 if (gid != old_fsgid) {
 904                         new->fsgid = gid;
 905                         goto change_okay;
 906                 }
 907         }
 908
 909         abort_creds(new);
 910         return old_fsgid;
 911
 912 change_okay:
 913         commit_creds(new);
 914         return old_fsgid;
 915 }
 916
 917 void do_sys_times(struct tms *tms)
 918 {
 919         cputime_t tgutime, tgstime, cutime, cstime;
 920
 921         spin_lock_irq(&current->sighand->siglock);
 922         thread_group_times(current, &tgutime, &tgstime);
 923         cutime = current->signal->cutime;
 924         cstime = current->signal->cstime;
 925         spin_unlock_irq(&current->sighand->siglock);
 926         tms->tms_utime = cputime_to_clock_t(tgutime);
 927         tms->tms_stime = cputime_to_clock_t(tgstime);
 928         tms->tms_cutime = cputime_to_clock_t(cutime);
 929         tms->tms_cstime = cputime_to_clock_t(cstime);
 930 }
 931
 932 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 933 {
 934         if (tbuf) {
 935                 struct tms tmp;
 936
 937                 do_sys_times(&tmp);
 938                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 939                         return -EFAULT;
 940         }
 941         force_successful_syscall_return();
 942         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 943 }
 944
 945 /*
 946  * This needs some heavy checking ...
 947  * I just haven't the stomach for it. I also don't fully
 948  * understand sessions/pgrp etc. Let somebody who does explain it.
 949  *
 950  * OK, I think I have the protection semantics right.... this is really
 951  * only important on a multi-user system anyway, to make sure one user
 952  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 953  *
 954  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 955  * LBT 04.03.94
 956  */
 957 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 958 {
 959         struct task_struct *p;
 960         struct task_struct *group_leader = current->group_leader;
 961         struct pid *pgrp;
 962         int err;
 963
 964         if (!pid)
 965                 pid = task_pid_vnr(group_leader);
 966         if (!pgid)
 967                 pgid = pid;
 968         if (pgid < 0)
 969                 return -EINVAL;
 970         rcu_read_lock();
 971
 972         /* From this point forward we keep holding onto the tasklist lock
 973          * so that our parent does not change from under us. -DaveM
 974          */
 975         write_lock_irq(&tasklist_lock);
 976
 977         err = -ESRCH;
 978         p = find_task_by_vpid(pid);
 979         if (!p)
 980                 goto out;
 981
 982         err = -EINVAL;
 983         if (!thread_group_leader(p))
 984                 goto out;
 985
 986         if (same_thread_group(p->real_parent, group_leader)) {
 987                 err = -EPERM;
 988                 if (task_session(p) != task_session(group_leader))
 989                         goto out;
 990                 err = -EACCES;
 991                 if (p->did_exec)
 992                         goto out;
 993         } else {
 994                 err = -ESRCH;
 995                 if (p != group_leader)
 996                         goto out;
 997         }
 998
 999         err = -EPERM;
1000         if (p->signal->leader)
1001                 goto out;
1002
1003         pgrp = task_pid(p);
1004         if (pgid != pid) {
1005                 struct task_struct *g;
1006
1007                 pgrp = find_vpid(pgid);
1008                 g = pid_task(pgrp, PIDTYPE_PGID);
1009                 if (!g || task_session(g) != task_session(group_leader))
1010                         goto out;
1011         }
1012
1013         err = security_task_setpgid(p, pgid);
1014         if (err)
1015                 goto out;
1016
1017         if (task_pgrp(p) != pgrp)
1018                 change_pid(p, PIDTYPE_PGID, pgrp);
1019
1020         err = 0;
1021 out:
1022         /* All paths lead to here, thus we are safe. -DaveM */
1023         write_unlock_irq(&tasklist_lock);
1024         rcu_read_unlock();
1025         return err;
1026 }
1027
1028 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1029 {
1030         struct task_struct *p;
1031         struct pid *grp;
1032         int retval;
1033
1034         rcu_read_lock();
1035         if (!pid)
1036                 grp = task_pgrp(current);
1037         else {
1038                 retval = -ESRCH;
1039                 p = find_task_by_vpid(pid);
1040                 if (!p)
1041                         goto out;
1042                 grp = task_pgrp(p);
1043                 if (!grp)
1044                         goto out;
1045
1046                 retval = security_task_getpgid(p);
1047                 if (retval)
1048                         goto out;
1049         }
1050         retval = pid_vnr(grp);
1051 out:
1052         rcu_read_unlock();
1053         return retval;
1054 }
1055
1056 #ifdef __ARCH_WANT_SYS_GETPGRP
1057
1058 SYSCALL_DEFINE0(getpgrp)
1059 {
1060         return sys_getpgid(0);
1061 }
1062
1063 #endif
1064
1065 SYSCALL_DEFINE1(getsid, pid_t, pid)
1066 {
1067         struct task_struct *p;
1068         struct pid *sid;
1069         int retval;
1070
1071         rcu_read_lock();
1072         if (!pid)
1073                 sid = task_session(current);
1074         else {
1075                 retval = -ESRCH;
1076                 p = find_task_by_vpid(pid);
1077                 if (!p)
1078                         goto out;
1079                 sid = task_session(p);
1080                 if (!sid)
1081                         goto out;
1082
1083                 retval = security_task_getsid(p);
1084                 if (retval)
1085                         goto out;
1086         }
1087         retval = pid_vnr(sid);
1088 out:
1089         rcu_read_unlock();
1090         return retval;
1091 }
1092
1093 SYSCALL_DEFINE0(setsid)
1094 {
1095         struct task_struct *group_leader = current->group_leader;
1096         struct pid *sid = task_pid(group_leader);
1097         pid_t session = pid_vnr(sid);
1098         int err = -EPERM;
1099
1100         write_lock_irq(&tasklist_lock);
1101         /* Fail if I am already a session leader */
1102         if (group_leader->signal->leader)
1103                 goto out;
1104
1105         /* Fail if a process group id already exists that equals the
1106          * proposed session id.
1107          */
1108         if (pid_task(sid, PIDTYPE_PGID))
1109                 goto out;
1110
1111         group_leader->signal->leader = 1;
1112         __set_special_pids(sid);
1113
1114         proc_clear_tty(group_leader);
1115
1116         err = session;
1117 out:
1118         write_unlock_irq(&tasklist_lock);
1119         if (err > 0) {
1120                 proc_sid_connector(group_leader);
1121                 sched_autogroup_create_attach(group_leader);
1122         }
1123         return err;
1124 }
1125
1126 DECLARE_RWSEM(uts_sem);
1127
1128 #ifdef COMPAT_UTS_MACHINE
1129 #define override_architecture(name) \
1130         (personality(current->personality) == PER_LINUX32 && \
1131          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1132                       sizeof(COMPAT_UTS_MACHINE)))
1133 #else
1134 #define override_architecture(name)     0
1135 #endif
1136
1137 /*
1138  * Work around broken programs that cannot handle "Linux 3.0".
1139  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1140  */
1141 static int override_release(char __user *release, size_t len)
1142 {
1143         int ret = 0;
1144
1145         if (current->personality & UNAME26) {
1146                 const char *rest = UTS_RELEASE;
1147                 char buf[65] = { 0 };
1148                 int ndots = 0;
1149                 unsigned v;
1150                 size_t copy;
1151
1152                 while (*rest) {
1153                         if (*rest == '.' && ++ndots >= 3)
1154                                 break;
1155                         if (!isdigit(*rest) && *rest != '.')
1156                                 break;
1157                         rest++;
1158                 }
1159                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1160                 copy = clamp_t(size_t, len, 1, sizeof(buf));
1161                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1162                 ret = copy_to_user(release, buf, copy + 1);
1163         }
1164         return ret;
1165 }
1166
1167 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1168 {
1169         int errno = 0;
1170
1171         down_read(&uts_sem);
1172         if (copy_to_user(name, utsname(), sizeof *name))
1173                 errno = -EFAULT;
1174         up_read(&uts_sem);
1175
1176         if (!errno && override_release(name->release, sizeof(name->release)))
1177                 errno = -EFAULT;
1178         if (!errno && override_architecture(name))
1179                 errno = -EFAULT;
1180         return errno;
1181 }
1182
1183 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1184 /*
1185  * Old cruft
1186  */
1187 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1188 {
1189         int error = 0;
1190
1191         if (!name)
1192                 return -EFAULT;
1193
1194         down_read(&uts_sem);
1195         if (copy_to_user(name, utsname(), sizeof(*name)))
1196                 error = -EFAULT;
1197         up_read(&uts_sem);
1198
1199         if (!error && override_release(name->release, sizeof(name->release)))
1200                 error = -EFAULT;
1201         if (!error && override_architecture(name))
1202                 error = -EFAULT;
1203         return error;
1204 }
1205
1206 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1207 {
1208         int error;
1209
1210         if (!name)
1211                 return -EFAULT;
1212         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1213                 return -EFAULT;
1214
1215         down_read(&uts_sem);
1216         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1217                                __OLD_UTS_LEN);
1218         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1219         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1220                                 __OLD_UTS_LEN);
1221         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1222         error |= __copy_to_user(&name->release, &utsname()->release,
1223                                 __OLD_UTS_LEN);
1224         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1225         error |= __copy_to_user(&name->version, &utsname()->version,
1226                                 __OLD_UTS_LEN);
1227         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1228         error |= __copy_to_user(&name->machine, &utsname()->machine,
1229                                 __OLD_UTS_LEN);
1230         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1231         up_read(&uts_sem);
1232
1233         if (!error && override_architecture(name))
1234                 error = -EFAULT;
1235         if (!error && override_release(name->release, sizeof(name->release)))
1236                 error = -EFAULT;
1237         return error ? -EFAULT : 0;
1238 }
1239 #endif
1240
1241 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1242 {
1243         int errno;
1244         char tmp[__NEW_UTS_LEN];
1245
1246         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1247                 return -EPERM;
1248
1249         if (len < 0 || len > __NEW_UTS_LEN)
1250                 return -EINVAL;
1251         down_write(&uts_sem);
1252         errno = -EFAULT;
1253         if (!copy_from_user(tmp, name, len)) {
1254                 struct new_utsname *u = utsname();
1255
1256                 memcpy(u->nodename, tmp, len);
1257                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1258                 errno = 0;
1259         }
1260         up_write(&uts_sem);
1261         return errno;
1262 }
1263
1264 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1265
1266 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1267 {
1268         int i, errno;
1269         struct new_utsname *u;
1270
1271         if (len < 0)
1272                 return -EINVAL;
1273         down_read(&uts_sem);
1274         u = utsname();
1275         i = 1 + strlen(u->nodename);
1276         if (i > len)
1277                 i = len;
1278         errno = 0;
1279         if (copy_to_user(name, u->nodename, i))
1280                 errno = -EFAULT;
1281         up_read(&uts_sem);
1282         return errno;
1283 }
1284
1285 #endif
1286
1287 /*
1288  * Only setdomainname; getdomainname can be implemented by calling
1289  * uname()
1290  */
1291 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1292 {
1293         int errno;
1294         char tmp[__NEW_UTS_LEN];
1295
1296         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1297                 return -EPERM;
1298         if (len < 0 || len > __NEW_UTS_LEN)
1299                 return -EINVAL;
1300
1301         down_write(&uts_sem);
1302         errno = -EFAULT;
1303         if (!copy_from_user(tmp, name, len)) {
1304                 struct new_utsname *u = utsname();
1305
1306                 memcpy(u->domainname, tmp, len);
1307                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1308                 errno = 0;
1309         }
1310         up_write(&uts_sem);
1311         return errno;
1312 }
1313
1314 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1315 {
1316         struct rlimit value;
1317         int ret;
1318
1319         ret = do_prlimit(current, resource, NULL, &value);
1320         if (!ret)
1321                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1322
1323         return ret;
1324 }
1325
1326 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1327
1328 /*
1329  *      Back compatibility for getrlimit. Needed for some apps.
1330  */
1331
1332 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1333                 struct rlimit __user *, rlim)
1334 {
1335         struct rlimit x;
1336         if (resource >= RLIM_NLIMITS)
1337                 return -EINVAL;
1338
1339         task_lock(current->group_leader);
1340         x = current->signal->rlim[resource];
1341         task_unlock(current->group_leader);
1342         if (x.rlim_cur > 0x7FFFFFFF)
1343                 x.rlim_cur = 0x7FFFFFFF;
1344         if (x.rlim_max > 0x7FFFFFFF)
1345                 x.rlim_max = 0x7FFFFFFF;
1346         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1347 }
1348
1349 #endif
1350
1351 static inline bool rlim64_is_infinity(__u64 rlim64)
1352 {
1353 #if BITS_PER_LONG < 64
1354         return rlim64 >= ULONG_MAX;
1355 #else
1356         return rlim64 == RLIM64_INFINITY;
1357 #endif
1358 }
1359
1360 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1361 {
1362         if (rlim->rlim_cur == RLIM_INFINITY)
1363                 rlim64->rlim_cur = RLIM64_INFINITY;
1364         else
1365                 rlim64->rlim_cur = rlim->rlim_cur;
1366         if (rlim->rlim_max == RLIM_INFINITY)
1367                 rlim64->rlim_max = RLIM64_INFINITY;
1368         else
1369                 rlim64->rlim_max = rlim->rlim_max;
1370 }
1371
1372 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1373 {
1374         if (rlim64_is_infinity(rlim64->rlim_cur))
1375                 rlim->rlim_cur = RLIM_INFINITY;
1376         else
1377                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1378         if (rlim64_is_infinity(rlim64->rlim_max))
1379                 rlim->rlim_max = RLIM_INFINITY;
1380         else
1381                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1382 }
1383
1384 /* make sure you are allowed to change @tsk limits before calling this */
1385 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1386                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1387 {
1388         struct rlimit *rlim;
1389         int retval = 0;
1390
1391         if (resource >= RLIM_NLIMITS)
1392                 return -EINVAL;
1393         if (new_rlim) {
1394                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1395                         return -EINVAL;
1396                 if (resource == RLIMIT_NOFILE &&
1397                                 new_rlim->rlim_max > sysctl_nr_open)
1398                         return -EPERM;
1399         }
1400
1401         /* protect tsk->signal and tsk->sighand from disappearing */
1402         read_lock(&tasklist_lock);
1403         if (!tsk->sighand) {
1404                 retval = -ESRCH;
1405                 goto out;
1406         }
1407
1408         rlim = tsk->signal->rlim + resource;
1409         task_lock(tsk->group_leader);
1410         if (new_rlim) {
1411                 /* Keep the capable check against init_user_ns until
1412                    cgroups can contain all limits */
1413                 if (new_rlim->rlim_max > rlim->rlim_max &&
1414                                 !capable(CAP_SYS_RESOURCE))
1415                         retval = -EPERM;
1416                 if (!retval)
1417                         retval = security_task_setrlimit(tsk->group_leader,
1418                                         resource, new_rlim);
1419                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1420                         /*
1421                          * The caller is asking for an immediate RLIMIT_CPU
1422                          * expiry.  But we use the zero value to mean "it was
1423                          * never set".  So let's cheat and make it one second
1424                          * instead
1425                          */
1426                         new_rlim->rlim_cur = 1;
1427                 }
1428         }
1429         if (!retval) {
1430                 if (old_rlim)
1431                         *old_rlim = *rlim;
1432                 if (new_rlim)
1433                         *rlim = *new_rlim;
1434         }
1435         task_unlock(tsk->group_leader);
1436
1437         /*
1438          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1439          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1440          * very long-standing error, and fixing it now risks breakage of
1441          * applications, so we live with it
1442          */
1443          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1444                          new_rlim->rlim_cur != RLIM_INFINITY)
1445                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1446 out:
1447         read_unlock(&tasklist_lock);
1448         return retval;
1449 }
1450
1451 /* rcu lock must be held */
1452 static int check_prlimit_permission(struct task_struct *task)
1453 {
1454         const struct cred *cred = current_cred(), *tcred;
1455
1456         if (current == task)
1457                 return 0;
1458
1459         tcred = __task_cred(task);
1460         if (cred->user->user_ns == tcred->user->user_ns &&
1461             (cred->uid == tcred->euid &&
1462              cred->uid == tcred->suid &&
1463              cred->uid == tcred->uid  &&
1464              cred->gid == tcred->egid &&
1465              cred->gid == tcred->sgid &&
1466              cred->gid == tcred->gid))
1467                 return 0;
1468         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1469                 return 0;
1470
1471         return -EPERM;
1472 }
1473
1474 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1475                 const struct rlimit64 __user *, new_rlim,
1476                 struct rlimit64 __user *, old_rlim)
1477 {
1478         struct rlimit64 old64, new64;
1479         struct rlimit old, new;
1480         struct task_struct *tsk;
1481         int ret;
1482
1483         if (new_rlim) {
1484                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1485                         return -EFAULT;
1486                 rlim64_to_rlim(&new64, &new);
1487         }
1488
1489         rcu_read_lock();
1490         tsk = pid ? find_task_by_vpid(pid) : current;
1491         if (!tsk) {
1492                 rcu_read_unlock();
1493                 return -ESRCH;
1494         }
1495         ret = check_prlimit_permission(tsk);
1496         if (ret) {
1497                 rcu_read_unlock();
1498                 return ret;
1499         }
1500         get_task_struct(tsk);
1501         rcu_read_unlock();
1502
1503         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1504                         old_rlim ? &old : NULL);
1505
1506         if (!ret && old_rlim) {
1507                 rlim_to_rlim64(&old, &old64);
1508                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1509                         ret = -EFAULT;
1510         }
1511
1512         put_task_struct(tsk);
1513         return ret;
1514 }
1515
1516 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1517 {
1518         struct rlimit new_rlim;
1519
1520         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1521                 return -EFAULT;
1522         return do_prlimit(current, resource, &new_rlim, NULL);
1523 }
1524
1525 /*
1526  * It would make sense to put struct rusage in the task_struct,
1527  * except that would make the task_struct be *really big*.  After
1528  * task_struct gets moved into malloc'ed memory, it would
1529  * make sense to do this.  It will make moving the rest of the information
1530  * a lot simpler!  (Which we're not doing right now because we're not
1531  * measuring them yet).
1532  *
1533  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1534  * races with threads incrementing their own counters.  But since word
1535  * reads are atomic, we either get new values or old values and we don't
1536  * care which for the sums.  We always take the siglock to protect reading
1537  * the c* fields from p->signal from races with exit.c updating those
1538  * fields when reaping, so a sample either gets all the additions of a
1539  * given child after it's reaped, or none so this sample is before reaping.
1540  *
1541  * Locking:
1542  * We need to take the siglock for CHILDEREN, SELF and BOTH
1543  * for  the cases current multithreaded, non-current single threaded
1544  * non-current multithreaded.  Thread traversal is now safe with
1545  * the siglock held.
1546  * Strictly speaking, we donot need to take the siglock if we are current and
1547  * single threaded,  as no one else can take our signal_struct away, no one
1548  * else can  reap the  children to update signal->c* counters, and no one else
1549  * can race with the signal-> fields. If we do not take any lock, the
1550  * signal-> fields could be read out of order while another thread was just
1551  * exiting. So we should  place a read memory barrier when we avoid the lock.
1552  * On the writer side,  write memory barrier is implied in  __exit_signal
1553  * as __exit_signal releases  the siglock spinlock after updating the signal->
1554  * fields. But we don't do this yet to keep things simple.
1555  *
1556  */
1557
1558 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1559 {
1560         r->ru_nvcsw += t->nvcsw;
1561         r->ru_nivcsw += t->nivcsw;
1562         r->ru_minflt += t->min_flt;
1563         r->ru_majflt += t->maj_flt;
1564         r->ru_inblock += task_io_get_inblock(t);
1565         r->ru_oublock += task_io_get_oublock(t);
1566 }
1567
1568 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1569 {
1570         struct task_struct *t;
1571         unsigned long flags;
1572         cputime_t tgutime, tgstime, utime, stime;
1573         unsigned long maxrss = 0;
1574
1575         memset((char *) r, 0, sizeof *r);
1576         utime = stime = cputime_zero;
1577
1578         if (who == RUSAGE_THREAD) {
1579                 task_times(current, &utime, &stime);
1580                 accumulate_thread_rusage(p, r);
1581                 maxrss = p->signal->maxrss;
1582                 goto out;
1583         }
1584
1585         if (!lock_task_sighand(p, &flags))
1586                 return;
1587
1588         switch (who) {
1589                 case RUSAGE_BOTH:
1590                 case RUSAGE_CHILDREN:
1591                         utime = p->signal->cutime;
1592                         stime = p->signal->cstime;
1593                         r->ru_nvcsw = p->signal->cnvcsw;
1594                         r->ru_nivcsw = p->signal->cnivcsw;
1595                         r->ru_minflt = p->signal->cmin_flt;
1596                         r->ru_majflt = p->signal->cmaj_flt;
1597                         r->ru_inblock = p->signal->cinblock;
1598                         r->ru_oublock = p->signal->coublock;
1599                         maxrss = p->signal->cmaxrss;
1600
1601                         if (who == RUSAGE_CHILDREN)
1602                                 break;
1603
1604                 case RUSAGE_SELF:
1605                         thread_group_times(p, &tgutime, &tgstime);
1606                         utime = cputime_add(utime, tgutime);
1607                         stime = cputime_add(stime, tgstime);
1608                         r->ru_nvcsw += p->signal->nvcsw;
1609                         r->ru_nivcsw += p->signal->nivcsw;
1610                         r->ru_minflt += p->signal->min_flt;
1611                         r->ru_majflt += p->signal->maj_flt;
1612                         r->ru_inblock += p->signal->inblock;
1613                         r->ru_oublock += p->signal->oublock;
1614                         if (maxrss < p->signal->maxrss)
1615                                 maxrss = p->signal->maxrss;
1616                         t = p;
1617                         do {
1618                                 accumulate_thread_rusage(t, r);
1619                                 t = next_thread(t);
1620                         } while (t != p);
1621                         break;
1622
1623                 default:
1624                         BUG();
1625         }
1626         unlock_task_sighand(p, &flags);
1627
1628 out:
1629         cputime_to_timeval(utime, &r->ru_utime);
1630         cputime_to_timeval(stime, &r->ru_stime);
1631
1632         if (who != RUSAGE_CHILDREN) {
1633                 struct mm_struct *mm = get_task_mm(p);
1634                 if (mm) {
1635                         setmax_mm_hiwater_rss(&maxrss, mm);
1636                         mmput(mm);
1637                 }
1638         }
1639         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1640 }
1641
1642 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1643 {
1644         struct rusage r;
1645         k_getrusage(p, who, &r);
1646         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1647 }
1648
1649 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1650 {
1651         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1652             who != RUSAGE_THREAD)
1653                 return -EINVAL;
1654         return getrusage(current, who, ru);
1655 }
1656
1657 SYSCALL_DEFINE1(umask, int, mask)
1658 {
1659         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1660         return mask;
1661 }
1662
1663 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1664                 unsigned long, arg4, unsigned long, arg5)
1665 {
1666         struct task_struct *me = current;
1667         unsigned char comm[sizeof(me->comm)];
1668         long error;
1669
1670         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1671         if (error != -ENOSYS)
1672                 return error;
1673
1674         error = 0;
1675         switch (option) {
1676                 case PR_SET_PDEATHSIG:
1677                         if (!valid_signal(arg2)) {
1678                                 error = -EINVAL;
1679                                 break;
1680                         }
1681                         me->pdeath_signal = arg2;
1682                         error = 0;
1683                         break;
1684                 case PR_GET_PDEATHSIG:
1685                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1686                         break;
1687                 case PR_GET_DUMPABLE:
1688                         error = get_dumpable(me->mm);
1689                         break;
1690                 case PR_SET_DUMPABLE:
1691                         if (arg2 < 0 || arg2 > 1) {
1692                                 error = -EINVAL;
1693                                 break;
1694                         }
1695                         set_dumpable(me->mm, arg2);
1696                         error = 0;
1697                         break;
1698
1699                 case PR_SET_UNALIGN:
1700                         error = SET_UNALIGN_CTL(me, arg2);
1701                         break;
1702                 case PR_GET_UNALIGN:
1703                         error = GET_UNALIGN_CTL(me, arg2);
1704                         break;
1705                 case PR_SET_FPEMU:
1706                         error = SET_FPEMU_CTL(me, arg2);
1707                         break;
1708                 case PR_GET_FPEMU:
1709                         error = GET_FPEMU_CTL(me, arg2);
1710                         break;
1711                 case PR_SET_FPEXC:
1712                         error = SET_FPEXC_CTL(me, arg2);
1713                         break;
1714                 case PR_GET_FPEXC:
1715                         error = GET_FPEXC_CTL(me, arg2);
1716                         break;
1717                 case PR_GET_TIMING:
1718                         error = PR_TIMING_STATISTICAL;
1719                         break;
1720                 case PR_SET_TIMING:
1721                         if (arg2 != PR_TIMING_STATISTICAL)
1722                                 error = -EINVAL;
1723                         else
1724                                 error = 0;
1725                         break;
1726
1727                 case PR_SET_NAME:
1728                         comm[sizeof(me->comm)-1] = 0;
1729                         if (strncpy_from_user(comm, (char __user *)arg2,
1730                                               sizeof(me->comm) - 1) < 0)
1731                                 return -EFAULT;
1732                         set_task_comm(me, comm);
1733                         return 0;
1734                 case PR_GET_NAME:
1735                         get_task_comm(comm, me);
1736                         if (copy_to_user((char __user *)arg2, comm,
1737                                          sizeof(comm)))
1738                                 return -EFAULT;
1739                         return 0;
1740                 case PR_GET_ENDIAN:
1741                         error = GET_ENDIAN(me, arg2);
1742                         break;
1743                 case PR_SET_ENDIAN:
1744                         error = SET_ENDIAN(me, arg2);
1745                         break;
1746
1747                 case PR_GET_SECCOMP:
1748                         error = prctl_get_seccomp();
1749                         break;
1750                 case PR_SET_SECCOMP:
1751                         error = prctl_set_seccomp(arg2);
1752                         break;
1753                 case PR_GET_TSC:
1754                         error = GET_TSC_CTL(arg2);
1755                         break;
1756                 case PR_SET_TSC:
1757                         error = SET_TSC_CTL(arg2);
1758                         break;
1759                 case PR_TASK_PERF_EVENTS_DISABLE:
1760                         error = perf_event_task_disable();
1761                         break;
1762                 case PR_TASK_PERF_EVENTS_ENABLE:
1763                         error = perf_event_task_enable();
1764                         break;
1765                 case PR_GET_TIMERSLACK:
1766                         error = current->timer_slack_ns;
1767                         break;
1768                 case PR_SET_TIMERSLACK:
1769                         if (arg2 <= 0)
1770                                 current->timer_slack_ns =
1771                                         current->default_timer_slack_ns;
1772                         else
1773                                 current->timer_slack_ns = arg2;
1774                         error = 0;
1775                         break;
1776                 case PR_MCE_KILL:
1777                         if (arg4 | arg5)
1778                                 return -EINVAL;
1779                         switch (arg2) {
1780                         case PR_MCE_KILL_CLEAR:
1781                                 if (arg3 != 0)
1782                                         return -EINVAL;
1783                                 current->flags &= ~PF_MCE_PROCESS;
1784                                 break;
1785                         case PR_MCE_KILL_SET:
1786                                 current->flags |= PF_MCE_PROCESS;
1787                                 if (arg3 == PR_MCE_KILL_EARLY)
1788                                         current->flags |= PF_MCE_EARLY;
1789                                 else if (arg3 == PR_MCE_KILL_LATE)
1790                                         current->flags &= ~PF_MCE_EARLY;
1791                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1792                                         current->flags &=
1793                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1794                                 else
1795                                         return -EINVAL;
1796                                 break;
1797                         default:
1798                                 return -EINVAL;
1799                         }
1800                         error = 0;
1801                         break;
1802                 case PR_MCE_KILL_GET:
1803                         if (arg2 | arg3 | arg4 | arg5)
1804                                 return -EINVAL;
1805                         if (current->flags & PF_MCE_PROCESS)
1806                                 error = (current->flags & PF_MCE_EARLY) ?
1807                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1808                         else
1809                                 error = PR_MCE_KILL_DEFAULT;
1810                         break;
1811                 default:
1812                         error = -EINVAL;
1813                         break;
1814         }
1815         return error;
1816 }
1817
1818 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1819                 struct getcpu_cache __user *, unused)
1820 {
1821         int err = 0;
1822         int cpu = raw_smp_processor_id();
1823         if (cpup)
1824                 err |= put_user(cpu, cpup);
1825         if (nodep)
1826                 err |= put_user(cpu_to_node(cpu), nodep);
1827         return err ? -EFAULT : 0;
1828 }
1829
1830 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1831
1832 static void argv_cleanup(struct subprocess_info *info)
1833 {
1834         argv_free(info->argv);
1835 }
1836
1837 /**
1838  * orderly_poweroff - Trigger an orderly system poweroff
1839  * @force: force poweroff if command execution fails
1840  *
1841  * This may be called from any context to trigger a system shutdown.
1842  * If the orderly shutdown fails, it will force an immediate shutdown.
1843  */
1844 int orderly_poweroff(bool force)
1845 {
1846         int argc;
1847         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1848         static char *envp[] = {
1849                 "HOME=/",
1850                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1851                 NULL
1852         };
1853         int ret = -ENOMEM;
1854         struct subprocess_info *info;
1855
1856         if (argv == NULL) {
1857                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1858                        __func__, poweroff_cmd);
1859                 goto out;
1860         }
1861
1862         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1863         if (info == NULL) {
1864                 argv_free(argv);
1865                 goto out;
1866         }
1867
1868         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1869
1870         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1871
1872   out:
1873         if (ret && force) {
1874                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1875                        "forcing the issue\n");
1876
1877                 /* I guess this should try to kick off some daemon to
1878                    sync and poweroff asap.  Or not even bother syncing
1879                    if we're doing an emergency shutdown? */
1880                 emergency_sync();
1881                 kernel_power_off();
1882         }
1883
1884         return ret;
1885 }
1886 EXPORT_SYMBOL_GPL(orderly_poweroff);