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