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