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 *        DEBUG
  58 ****************/
  59 #define RESTART_DEBUG
  60 #ifdef RESTART_DEBUG
  61 #define restart_dbg(format, arg...) \
  62         printk("RESTART_DEBUG : " format "\n" , ## arg)
  63 #else
  64 #define restart_dbg(format, arg...) do {} while (0)
  65 #endif
  66
  67
  68
  69 #ifndef SET_UNALIGN_CTL
  70 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  71 #endif
  72 #ifndef GET_UNALIGN_CTL
  73 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  74 #endif
  75 #ifndef SET_FPEMU_CTL
  76 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  77 #endif
  78 #ifndef GET_FPEMU_CTL
  79 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  80 #endif
  81 #ifndef SET_FPEXC_CTL
  82 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  83 #endif
  84 #ifndef GET_FPEXC_CTL
  85 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  86 #endif
  87 #ifndef GET_ENDIAN
  88 # define GET_ENDIAN(a,b)        (-EINVAL)
  89 #endif
  90 #ifndef SET_ENDIAN
  91 # define SET_ENDIAN(a,b)        (-EINVAL)
  92 #endif
  93 #ifndef GET_TSC_CTL
  94 # define GET_TSC_CTL(a)         (-EINVAL)
  95 #endif
  96 #ifndef SET_TSC_CTL
  97 # define SET_TSC_CTL(a)         (-EINVAL)
  98 #endif
  99
 100 /*
 101  * this is where the system-wide overflow UID and GID are defined, for
 102  * architectures that now have 32-bit UID/GID but didn't in the past
 103  */
 104
 105 int overflowuid = DEFAULT_OVERFLOWUID;
 106 int overflowgid = DEFAULT_OVERFLOWGID;
 107
 108 #ifdef CONFIG_UID16
 109 EXPORT_SYMBOL(overflowuid);
 110 EXPORT_SYMBOL(overflowgid);
 111 #endif
 112
 113 /*
 114  * the same as above, but for filesystems which can only store a 16-bit
 115  * UID and GID. as such, this is needed on all architectures
 116  */
 117
 118 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 119 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
 120
 121 EXPORT_SYMBOL(fs_overflowuid);
 122 EXPORT_SYMBOL(fs_overflowgid);
 123
 124 /*
 125  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
 126  */
 127
 128 int C_A_D = 1;
 129 struct pid *cad_pid;
 130 EXPORT_SYMBOL(cad_pid);
 131
 132 /*
 133  * If set, this is used for preparing the system to power off.
 134  */
 135
 136 void (*pm_power_off_prepare)(void);
 137
 138 /*
 139  * Returns true if current's euid is same as p's uid or euid,
 140  * or has CAP_SYS_NICE to p's user_ns.
 141  *
 142  * Called with rcu_read_lock, creds are safe
 143  */
 144 static bool set_one_prio_perm(struct task_struct *p)
 145 {
 146         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 147
 148         if (pcred->user->user_ns == cred->user->user_ns &&
 149             (pcred->uid  == cred->euid ||
 150              pcred->euid == cred->euid))
 151                 return true;
 152         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
 153                 return true;
 154         return false;
 155 }
 156
 157 /*
 158  * set the priority of a task
 159  * - the caller must hold the RCU read lock
 160  */
 161 static int set_one_prio(struct task_struct *p, int niceval, int error)
 162 {
 163         int no_nice;
 164
 165         if (!set_one_prio_perm(p)) {
 166                 error = -EPERM;
 167                 goto out;
 168         }
 169         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 170                 error = -EACCES;
 171                 goto out;
 172         }
 173         no_nice = security_task_setnice(p, niceval);
 174         if (no_nice) {
 175                 error = no_nice;
 176                 goto out;
 177         }
 178         if (error == -ESRCH)
 179                 error = 0;
 180         set_user_nice(p, niceval);
 181 out:
 182         return error;
 183 }
 184
 185 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 186 {
 187         struct task_struct *g, *p;
 188         struct user_struct *user;
 189         const struct cred *cred = current_cred();
 190         int error = -EINVAL;
 191         struct pid *pgrp;
 192
 193         if (which > PRIO_USER || which < PRIO_PROCESS)
 194                 goto out;
 195
 196         /* normalize: avoid signed division (rounding problems) */
 197         error = -ESRCH;
 198         if (niceval < -20)
 199                 niceval = -20;
 200         if (niceval > 19)
 201                 niceval = 19;
 202
 203         rcu_read_lock();
 204         read_lock(&tasklist_lock);
 205         switch (which) {
 206                 case PRIO_PROCESS:
 207                         if (who)
 208                                 p = find_task_by_vpid(who);
 209                         else
 210                                 p = current;
 211                         if (p)
 212                                 error = set_one_prio(p, niceval, error);
 213                         break;
 214                 case PRIO_PGRP:
 215                         if (who)
 216                                 pgrp = find_vpid(who);
 217                         else
 218                                 pgrp = task_pgrp(current);
 219                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 220                                 error = set_one_prio(p, niceval, error);
 221                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 222                         break;
 223                 case PRIO_USER:
 224                         user = (struct user_struct *) cred->user;
 225                         if (!who)
 226                                 who = cred->uid;
 227                         else if ((who != cred->uid) &&
 228                                  !(user = find_user(who)))
 229                                 goto out_unlock;        /* No processes for this user */
 230
 231                         do_each_thread(g, p) {
 232                                 if (__task_cred(p)->uid == who)
 233                                         error = set_one_prio(p, niceval, error);
 234                         } while_each_thread(g, p);
 235                         if (who != cred->uid)
 236                                 free_uid(user);         /* For find_user() */
 237                         break;
 238         }
 239 out_unlock:
 240         read_unlock(&tasklist_lock);
 241         rcu_read_unlock();
 242 out:
 243         return error;
 244 }
 245
 246 /*
 247  * Ugh. To avoid negative return values, "getpriority()" will
 248  * not return the normal nice-value, but a negated value that
 249  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 250  * to stay compatible.
 251  */
 252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
 253 {
 254         struct task_struct *g, *p;
 255         struct user_struct *user;
 256         const struct cred *cred = current_cred();
 257         long niceval, retval = -ESRCH;
 258         struct pid *pgrp;
 259
 260         if (which > PRIO_USER || which < PRIO_PROCESS)
 261                 return -EINVAL;
 262
 263         rcu_read_lock();
 264         read_lock(&tasklist_lock);
 265         switch (which) {
 266                 case PRIO_PROCESS:
 267                         if (who)
 268                                 p = find_task_by_vpid(who);
 269                         else
 270                                 p = current;
 271                         if (p) {
 272                                 niceval = 20 - task_nice(p);
 273                                 if (niceval > retval)
 274                                         retval = niceval;
 275                         }
 276                         break;
 277                 case PRIO_PGRP:
 278                         if (who)
 279                                 pgrp = find_vpid(who);
 280                         else
 281                                 pgrp = task_pgrp(current);
 282                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 283                                 niceval = 20 - task_nice(p);
 284                                 if (niceval > retval)
 285                                         retval = niceval;
 286                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 287                         break;
 288                 case PRIO_USER:
 289                         user = (struct user_struct *) cred->user;
 290                         if (!who)
 291                                 who = cred->uid;
 292                         else if ((who != cred->uid) &&
 293                                  !(user = find_user(who)))
 294                                 goto out_unlock;        /* No processes for this user */
 295
 296                         do_each_thread(g, p) {
 297                                 if (__task_cred(p)->uid == who) {
 298                                         niceval = 20 - task_nice(p);
 299                                         if (niceval > retval)
 300                                                 retval = niceval;
 301                                 }
 302                         } while_each_thread(g, p);
 303                         if (who != cred->uid)
 304                                 free_uid(user);         /* for find_user() */
 305                         break;
 306         }
 307 out_unlock:
 308         read_unlock(&tasklist_lock);
 309         rcu_read_unlock();
 310
 311         return retval;
 312 }
 313
 314 /**
 315  *      emergency_restart - reboot the system
 316  *
 317  *      Without shutting down any hardware or taking any locks
 318  *      reboot the system.  This is called when we know we are in
 319  *      trouble so this is our best effort to reboot.  This is
 320  *      safe to call in interrupt context.
 321  */
 322 void emergency_restart(void)
 323 {
 324         kmsg_dump(KMSG_DUMP_EMERG);
 325         machine_emergency_restart();
 326 }
 327 EXPORT_SYMBOL_GPL(emergency_restart);
 328
 329 void kernel_restart_prepare(char *cmd)
 330 {
 331         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 332         system_state = SYSTEM_RESTART;
 333         usermodehelper_disable();
 334         device_shutdown();
 335         syscore_shutdown();
 336 }
 337
 338 /**
 339  *      kernel_restart - reboot the system
 340  *      @cmd: pointer to buffer containing command to execute for restart
 341  *              or %NULL
 342  *
 343  *      Shutdown everything and perform a clean reboot.
 344  *      This is not safe to call in interrupt context.
 345  */
 346 void kernel_restart(char *cmd)
 347 {
 348         /*
 349         *  debug trace
 350         */
 351         restart_dbg("%s->%d->cmd=%s",__FUNCTION__,__LINE__,cmd);
 352
 353         kernel_restart_prepare(cmd);
 354 #ifndef CONFIG_PLAT_RK
 355         disable_nonboot_cpus();
 356 #endif
 357         if (!cmd)
 358                 printk(KERN_EMERG "Restarting system.\n");
 359         else
 360                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 361         kmsg_dump(KMSG_DUMP_RESTART);
 362         machine_restart(cmd);
 363 }
 364 EXPORT_SYMBOL_GPL(kernel_restart);
 365
 366 static void kernel_shutdown_prepare(enum system_states state)
 367 {
 368         blocking_notifier_call_chain(&reboot_notifier_list,
 369                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 370         system_state = state;
 371         usermodehelper_disable();
 372         device_shutdown();
 373 }
 374 /**
 375  *      kernel_halt - halt the system
 376  *
 377  *      Shutdown everything and perform a clean system halt.
 378  */
 379 void kernel_halt(void)
 380 {
 381         kernel_shutdown_prepare(SYSTEM_HALT);
 382         syscore_shutdown();
 383         printk(KERN_EMERG "System halted.\n");
 384         kmsg_dump(KMSG_DUMP_HALT);
 385         machine_halt();
 386 }
 387
 388 EXPORT_SYMBOL_GPL(kernel_halt);
 389
 390 /**
 391  *      kernel_power_off - power_off the system
 392  *
 393  *      Shutdown everything and perform a clean system power_off.
 394  */
 395 void kernel_power_off(void)
 396 {
 397         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 398         if (pm_power_off_prepare)
 399                 pm_power_off_prepare();
 400         disable_nonboot_cpus();
 401         syscore_shutdown();
 402         printk(KERN_EMERG "Power down.\n");
 403         kmsg_dump(KMSG_DUMP_POWEROFF);
 404         machine_power_off();
 405 }
 406 EXPORT_SYMBOL_GPL(kernel_power_off);
 407
 408 static DEFINE_MUTEX(reboot_mutex);
 409
 410 /*
 411  * Reboot system call: for obvious reasons only root may call it,
 412  * and even root needs to set up some magic numbers in the registers
 413  * so that some mistake won't make this reboot the whole machine.
 414  * You can also set the meaning of the ctrl-alt-del-key here.
 415  *
 416  * reboot doesn't sync: do that yourself before calling this.
 417  */
 418 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
 419                 void __user *, arg)
 420 {
 421         char buffer[256];
 422         int ret = 0;
 423
 424         /* We only trust the superuser with rebooting the system. */
 425         if (!capable(CAP_SYS_BOOT))
 426                 return -EPERM;
 427
 428         /* For safety, we require "magic" arguments. */
 429         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 430             (magic2 != LINUX_REBOOT_MAGIC2 &&
 431                         magic2 != LINUX_REBOOT_MAGIC2A &&
 432                         magic2 != LINUX_REBOOT_MAGIC2B &&
 433                         magic2 != LINUX_REBOOT_MAGIC2C))
 434                 return -EINVAL;
 435
 436         /* Instead of trying to make the power_off code look like
 437          * halt when pm_power_off is not set do it the easy way.
 438          */
 439         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 440                 cmd = LINUX_REBOOT_CMD_HALT;
 441
 442         mutex_lock(&reboot_mutex);
 443         switch (cmd) {
 444         case LINUX_REBOOT_CMD_RESTART:
 445                 /*
 446                 *  debug trace
 447                 */
 448                 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
 449
 450                 kernel_restart(NULL);
 451                 break;
 452
 453         case LINUX_REBOOT_CMD_CAD_ON:
 454                 C_A_D = 1;
 455                 break;
 456
 457         case LINUX_REBOOT_CMD_CAD_OFF:
 458                 C_A_D = 0;
 459                 break;
 460
 461         case LINUX_REBOOT_CMD_HALT:
 462                 kernel_halt();
 463                 do_exit(0);
 464                 panic("cannot halt");
 465
 466         case LINUX_REBOOT_CMD_POWER_OFF:
 467                 /*
 468                 *  debug trace
 469                 */
 470                 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
 471
 472                 kernel_power_off();
 473                 do_exit(0);
 474                 break;
 475
 476         case LINUX_REBOOT_CMD_RESTART2:
 477                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 478                         ret = -EFAULT;
 479                         break;
 480                 }
 481                 buffer[sizeof(buffer) - 1] = '\0';
 482                 /*
 483                 *  debug trace
 484                 */
 485                 restart_dbg("%s->%d->cmd=%x args=%s",__FUNCTION__,__LINE__,cmd,buffer);
 486
 487                 kernel_restart(buffer);
 488                 break;
 489
 490 #ifdef CONFIG_KEXEC
 491         case LINUX_REBOOT_CMD_KEXEC:
 492                 ret = kernel_kexec();
 493                 break;
 494 #endif
 495
 496 #ifdef CONFIG_HIBERNATION
 497         case LINUX_REBOOT_CMD_SW_SUSPEND:
 498                 ret = hibernate();
 499                 break;
 500 #endif
 501
 502         default:
 503                 ret = -EINVAL;
 504                 break;
 505         }
 506         mutex_unlock(&reboot_mutex);
 507         return ret;
 508 }
 509
 510 static void deferred_cad(struct work_struct *dummy)
 511 {
 512         kernel_restart(NULL);
 513 }
 514
 515 /*
 516  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 517  * As it's called within an interrupt, it may NOT sync: the only choice
 518  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 519  */
 520 void ctrl_alt_del(void)
 521 {
 522         static DECLARE_WORK(cad_work, deferred_cad);
 523
 524         if (C_A_D)
 525                 schedule_work(&cad_work);
 526         else
 527                 kill_cad_pid(SIGINT, 1);
 528 }
 529
 530 /*
 531  * Unprivileged users may change the real gid to the effective gid
 532  * or vice versa.  (BSD-style)
 533  *
 534  * If you set the real gid at all, or set the effective gid to a value not
 535  * equal to the real gid, then the saved gid is set to the new effective gid.
 536  *
 537  * This makes it possible for a setgid program to completely drop its
 538  * privileges, which is often a useful assertion to make when you are doing
 539  * a security audit over a program.
 540  *
 541  * The general idea is that a program which uses just setregid() will be
 542  * 100% compatible with BSD.  A program which uses just setgid() will be
 543  * 100% compatible with POSIX with saved IDs.
 544  *
 545  * SMP: There are not races, the GIDs are checked only by filesystem
 546  *      operations (as far as semantic preservation is concerned).
 547  */
 548 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 549 {
 550         const struct cred *old;
 551         struct cred *new;
 552         int retval;
 553
 554         new = prepare_creds();
 555         if (!new)
 556                 return -ENOMEM;
 557         old = current_cred();
 558
 559         retval = -EPERM;
 560         if (rgid != (gid_t) -1) {
 561                 if (old->gid == rgid ||
 562                     old->egid == rgid ||
 563                     nsown_capable(CAP_SETGID))
 564                         new->gid = rgid;
 565                 else
 566                         goto error;
 567         }
 568         if (egid != (gid_t) -1) {
 569                 if (old->gid == egid ||
 570                     old->egid == egid ||
 571                     old->sgid == egid ||
 572                     nsown_capable(CAP_SETGID))
 573                         new->egid = egid;
 574                 else
 575                         goto error;
 576         }
 577
 578         if (rgid != (gid_t) -1 ||
 579             (egid != (gid_t) -1 && egid != old->gid))
 580                 new->sgid = new->egid;
 581         new->fsgid = new->egid;
 582
 583         return commit_creds(new);
 584
 585 error:
 586         abort_creds(new);
 587         return retval;
 588 }
 589
 590 /*
 591  * setgid() is implemented like SysV w/ SAVED_IDS
 592  *
 593  * SMP: Same implicit races as above.
 594  */
 595 SYSCALL_DEFINE1(setgid, gid_t, gid)
 596 {
 597         const struct cred *old;
 598         struct cred *new;
 599         int retval;
 600
 601         new = prepare_creds();
 602         if (!new)
 603                 return -ENOMEM;
 604         old = current_cred();
 605
 606         retval = -EPERM;
 607         if (nsown_capable(CAP_SETGID))
 608                 new->gid = new->egid = new->sgid = new->fsgid = gid;
 609         else if (gid == old->gid || gid == old->sgid)
 610                 new->egid = new->fsgid = gid;
 611         else
 612                 goto error;
 613
 614         return commit_creds(new);
 615
 616 error:
 617         abort_creds(new);
 618         return retval;
 619 }
 620
 621 /*
 622  * change the user struct in a credentials set to match the new UID
 623  */
 624 static int set_user(struct cred *new)
 625 {
 626         struct user_struct *new_user;
 627
 628         new_user = alloc_uid(current_user_ns(), new->uid);
 629         if (!new_user)
 630                 return -EAGAIN;
 631
 632         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 633                         new_user != INIT_USER) {
 634                 free_uid(new_user);
 635                 return -EAGAIN;
 636         }
 637
 638         free_uid(new->user);
 639         new->user = new_user;
 640         return 0;
 641 }
 642
 643 /*
 644  * Unprivileged users may change the real uid to the effective uid
 645  * or vice versa.  (BSD-style)
 646  *
 647  * If you set the real uid at all, or set the effective uid to a value not
 648  * equal to the real uid, then the saved uid is set to the new effective uid.
 649  *
 650  * This makes it possible for a setuid program to completely drop its
 651  * privileges, which is often a useful assertion to make when you are doing
 652  * a security audit over a program.
 653  *
 654  * The general idea is that a program which uses just setreuid() will be
 655  * 100% compatible with BSD.  A program which uses just setuid() will be
 656  * 100% compatible with POSIX with saved IDs.
 657  */
 658 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 659 {
 660         const struct cred *old;
 661         struct cred *new;
 662         int retval;
 663
 664         new = prepare_creds();
 665         if (!new)
 666                 return -ENOMEM;
 667         old = current_cred();
 668
 669         retval = -EPERM;
 670         if (ruid != (uid_t) -1) {
 671                 new->uid = ruid;
 672                 if (old->uid != ruid &&
 673                     old->euid != ruid &&
 674                     !nsown_capable(CAP_SETUID))
 675                         goto error;
 676         }
 677
 678         if (euid != (uid_t) -1) {
 679                 new->euid = euid;
 680                 if (old->uid != euid &&
 681                     old->euid != euid &&
 682                     old->suid != euid &&
 683                     !nsown_capable(CAP_SETUID))
 684                         goto error;
 685         }
 686
 687         if (new->uid != old->uid) {
 688                 retval = set_user(new);
 689                 if (retval < 0)
 690                         goto error;
 691         }
 692         if (ruid != (uid_t) -1 ||
 693             (euid != (uid_t) -1 && euid != old->uid))
 694                 new->suid = new->euid;
 695         new->fsuid = new->euid;
 696
 697         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 698         if (retval < 0)
 699                 goto error;
 700
 701         return commit_creds(new);
 702
 703 error:
 704         abort_creds(new);
 705         return retval;
 706 }
 707
 708 /*
 709  * setuid() is implemented like SysV with SAVED_IDS
 710  *
 711  * Note that SAVED_ID's is deficient in that a setuid root program
 712  * like sendmail, for example, cannot set its uid to be a normal
 713  * user and then switch back, because if you're root, setuid() sets
 714  * the saved uid too.  If you don't like this, blame the bright people
 715  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 716  * will allow a root program to temporarily drop privileges and be able to
 717  * regain them by swapping the real and effective uid.
 718  */
 719 SYSCALL_DEFINE1(setuid, uid_t, uid)
 720 {
 721         const struct cred *old;
 722         struct cred *new;
 723         int retval;
 724
 725         new = prepare_creds();
 726         if (!new)
 727                 return -ENOMEM;
 728         old = current_cred();
 729
 730         retval = -EPERM;
 731         if (nsown_capable(CAP_SETUID)) {
 732                 new->suid = new->uid = uid;
 733                 if (uid != old->uid) {
 734                         retval = set_user(new);
 735                         if (retval < 0)
 736                                 goto error;
 737                 }
 738         } else if (uid != old->uid && uid != new->suid) {
 739                 goto error;
 740         }
 741
 742         new->fsuid = new->euid = uid;
 743
 744         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 745         if (retval < 0)
 746                 goto error;
 747
 748         return commit_creds(new);
 749
 750 error:
 751         abort_creds(new);
 752         return retval;
 753 }
 754
 755
 756 /*
 757  * This function implements a generic ability to update ruid, euid,
 758  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 759  */
 760 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 761 {
 762         const struct cred *old;
 763         struct cred *new;
 764         int retval;
 765
 766         new = prepare_creds();
 767         if (!new)
 768                 return -ENOMEM;
 769
 770         old = current_cred();
 771
 772         retval = -EPERM;
 773         if (!nsown_capable(CAP_SETUID)) {
 774                 if (ruid != (uid_t) -1 && ruid != old->uid &&
 775                     ruid != old->euid  && ruid != old->suid)
 776                         goto error;
 777                 if (euid != (uid_t) -1 && euid != old->uid &&
 778                     euid != old->euid  && euid != old->suid)
 779                         goto error;
 780                 if (suid != (uid_t) -1 && suid != old->uid &&
 781                     suid != old->euid  && suid != old->suid)
 782                         goto error;
 783         }
 784
 785         if (ruid != (uid_t) -1) {
 786                 new->uid = ruid;
 787                 if (ruid != old->uid) {
 788                         retval = set_user(new);
 789                         if (retval < 0)
 790                                 goto error;
 791                 }
 792         }
 793         if (euid != (uid_t) -1)
 794                 new->euid = euid;
 795         if (suid != (uid_t) -1)
 796                 new->suid = suid;
 797         new->fsuid = new->euid;
 798
 799         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 800         if (retval < 0)
 801                 goto error;
 802
 803         return commit_creds(new);
 804
 805 error:
 806         abort_creds(new);
 807         return retval;
 808 }
 809
 810 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
 811 {
 812         const struct cred *cred = current_cred();
 813         int retval;
 814
 815         if (!(retval   = put_user(cred->uid,  ruid)) &&
 816             !(retval   = put_user(cred->euid, euid)))
 817                 retval = put_user(cred->suid, suid);
 818
 819         return retval;
 820 }
 821
 822 /*
 823  * Same as above, but for rgid, egid, sgid.
 824  */
 825 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 826 {
 827         const struct cred *old;
 828         struct cred *new;
 829         int retval;
 830
 831         new = prepare_creds();
 832         if (!new)
 833                 return -ENOMEM;
 834         old = current_cred();
 835
 836         retval = -EPERM;
 837         if (!nsown_capable(CAP_SETGID)) {
 838                 if (rgid != (gid_t) -1 && rgid != old->gid &&
 839                     rgid != old->egid  && rgid != old->sgid)
 840                         goto error;
 841                 if (egid != (gid_t) -1 && egid != old->gid &&
 842                     egid != old->egid  && egid != old->sgid)
 843                         goto error;
 844                 if (sgid != (gid_t) -1 && sgid != old->gid &&
 845                     sgid != old->egid  && sgid != old->sgid)
 846                         goto error;
 847         }
 848
 849         if (rgid != (gid_t) -1)
 850                 new->gid = rgid;
 851         if (egid != (gid_t) -1)
 852                 new->egid = egid;
 853         if (sgid != (gid_t) -1)
 854                 new->sgid = sgid;
 855         new->fsgid = new->egid;
 856
 857         return commit_creds(new);
 858
 859 error:
 860         abort_creds(new);
 861         return retval;
 862 }
 863
 864 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
 865 {
 866         const struct cred *cred = current_cred();
 867         int retval;
 868
 869         if (!(retval   = put_user(cred->gid,  rgid)) &&
 870             !(retval   = put_user(cred->egid, egid)))
 871                 retval = put_user(cred->sgid, sgid);
 872
 873         return retval;
 874 }
 875
 876
 877 /*
 878  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 879  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 880  * whatever uid it wants to). It normally shadows "euid", except when
 881  * explicitly set by setfsuid() or for access..
 882  */
 883 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 884 {
 885         const struct cred *old;
 886         struct cred *new;
 887         uid_t old_fsuid;
 888
 889         new = prepare_creds();
 890         if (!new)
 891                 return current_fsuid();
 892         old = current_cred();
 893         old_fsuid = old->fsuid;
 894
 895         if (uid == old->uid  || uid == old->euid  ||
 896             uid == old->suid || uid == old->fsuid ||
 897             nsown_capable(CAP_SETUID)) {
 898                 if (uid != old_fsuid) {
 899                         new->fsuid = uid;
 900                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 901                                 goto change_okay;
 902                 }
 903         }
 904
 905         abort_creds(new);
 906         return old_fsuid;
 907
 908 change_okay:
 909         commit_creds(new);
 910         return old_fsuid;
 911 }
 912
 913 /*
 914  * Samma på svenska..
 915  */
 916 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 917 {
 918         const struct cred *old;
 919         struct cred *new;
 920         gid_t old_fsgid;
 921
 922         new = prepare_creds();
 923         if (!new)
 924                 return current_fsgid();
 925         old = current_cred();
 926         old_fsgid = old->fsgid;
 927
 928         if (gid == old->gid  || gid == old->egid  ||
 929             gid == old->sgid || gid == old->fsgid ||
 930             nsown_capable(CAP_SETGID)) {
 931                 if (gid != old_fsgid) {
 932                         new->fsgid = gid;
 933                         goto change_okay;
 934                 }
 935         }
 936
 937         abort_creds(new);
 938         return old_fsgid;
 939
 940 change_okay:
 941         commit_creds(new);
 942         return old_fsgid;
 943 }
 944
 945 void do_sys_times(struct tms *tms)
 946 {
 947         cputime_t tgutime, tgstime, cutime, cstime;
 948
 949         spin_lock_irq(&current->sighand->siglock);
 950         thread_group_times(current, &tgutime, &tgstime);
 951         cutime = current->signal->cutime;
 952         cstime = current->signal->cstime;
 953         spin_unlock_irq(&current->sighand->siglock);
 954         tms->tms_utime = cputime_to_clock_t(tgutime);
 955         tms->tms_stime = cputime_to_clock_t(tgstime);
 956         tms->tms_cutime = cputime_to_clock_t(cutime);
 957         tms->tms_cstime = cputime_to_clock_t(cstime);
 958 }
 959
 960 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 961 {
 962         if (tbuf) {
 963                 struct tms tmp;
 964
 965                 do_sys_times(&tmp);
 966                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 967                         return -EFAULT;
 968         }
 969         force_successful_syscall_return();
 970         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 971 }
 972
 973 /*
 974  * This needs some heavy checking ...
 975  * I just haven't the stomach for it. I also don't fully
 976  * understand sessions/pgrp etc. Let somebody who does explain it.
 977  *
 978  * OK, I think I have the protection semantics right.... this is really
 979  * only important on a multi-user system anyway, to make sure one user
 980  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 981  *
 982  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 983  * LBT 04.03.94
 984  */
 985 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
 986 {
 987         struct task_struct *p;
 988         struct task_struct *group_leader = current->group_leader;
 989         struct pid *pgrp;
 990         int err;
 991
 992         if (!pid)
 993                 pid = task_pid_vnr(group_leader);
 994         if (!pgid)
 995                 pgid = pid;
 996         if (pgid < 0)
 997                 return -EINVAL;
 998         rcu_read_lock();
 999
1000         /* From this point forward we keep holding onto the tasklist lock
1001          * so that our parent does not change from under us. -DaveM
1002          */
1003         write_lock_irq(&tasklist_lock);
1004
1005         err = -ESRCH;
1006         p = find_task_by_vpid(pid);
1007         if (!p)
1008                 goto out;
1009
1010         err = -EINVAL;
1011         if (!thread_group_leader(p))
1012                 goto out;
1013
1014         if (same_thread_group(p->real_parent, group_leader)) {
1015                 err = -EPERM;
1016                 if (task_session(p) != task_session(group_leader))
1017                         goto out;
1018                 err = -EACCES;
1019                 if (p->did_exec)
1020                         goto out;
1021         } else {
1022                 err = -ESRCH;
1023                 if (p != group_leader)
1024                         goto out;
1025         }
1026
1027         err = -EPERM;
1028         if (p->signal->leader)
1029                 goto out;
1030
1031         pgrp = task_pid(p);
1032         if (pgid != pid) {
1033                 struct task_struct *g;
1034
1035                 pgrp = find_vpid(pgid);
1036                 g = pid_task(pgrp, PIDTYPE_PGID);
1037                 if (!g || task_session(g) != task_session(group_leader))
1038                         goto out;
1039         }
1040
1041         err = security_task_setpgid(p, pgid);
1042         if (err)
1043                 goto out;
1044
1045         if (task_pgrp(p) != pgrp)
1046                 change_pid(p, PIDTYPE_PGID, pgrp);
1047
1048         err = 0;
1049 out:
1050         /* All paths lead to here, thus we are safe. -DaveM */
1051         write_unlock_irq(&tasklist_lock);
1052         rcu_read_unlock();
1053         return err;
1054 }
1055
1056 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1057 {
1058         struct task_struct *p;
1059         struct pid *grp;
1060         int retval;
1061
1062         rcu_read_lock();
1063         if (!pid)
1064                 grp = task_pgrp(current);
1065         else {
1066                 retval = -ESRCH;
1067                 p = find_task_by_vpid(pid);
1068                 if (!p)
1069                         goto out;
1070                 grp = task_pgrp(p);
1071                 if (!grp)
1072                         goto out;
1073
1074                 retval = security_task_getpgid(p);
1075                 if (retval)
1076                         goto out;
1077         }
1078         retval = pid_vnr(grp);
1079 out:
1080         rcu_read_unlock();
1081         return retval;
1082 }
1083
1084 #ifdef __ARCH_WANT_SYS_GETPGRP
1085
1086 SYSCALL_DEFINE0(getpgrp)
1087 {
1088         return sys_getpgid(0);
1089 }
1090
1091 #endif
1092
1093 SYSCALL_DEFINE1(getsid, pid_t, pid)
1094 {
1095         struct task_struct *p;
1096         struct pid *sid;
1097         int retval;
1098
1099         rcu_read_lock();
1100         if (!pid)
1101                 sid = task_session(current);
1102         else {
1103                 retval = -ESRCH;
1104                 p = find_task_by_vpid(pid);
1105                 if (!p)
1106                         goto out;
1107                 sid = task_session(p);
1108                 if (!sid)
1109                         goto out;
1110
1111                 retval = security_task_getsid(p);
1112                 if (retval)
1113                         goto out;
1114         }
1115         retval = pid_vnr(sid);
1116 out:
1117         rcu_read_unlock();
1118         return retval;
1119 }
1120
1121 SYSCALL_DEFINE0(setsid)
1122 {
1123         struct task_struct *group_leader = current->group_leader;
1124         struct pid *sid = task_pid(group_leader);
1125         pid_t session = pid_vnr(sid);
1126         int err = -EPERM;
1127
1128         write_lock_irq(&tasklist_lock);
1129         /* Fail if I am already a session leader */
1130         if (group_leader->signal->leader)
1131                 goto out;
1132
1133         /* Fail if a process group id already exists that equals the
1134          * proposed session id.
1135          */
1136         if (pid_task(sid, PIDTYPE_PGID))
1137                 goto out;
1138
1139         group_leader->signal->leader = 1;
1140         __set_special_pids(sid);
1141
1142         proc_clear_tty(group_leader);
1143
1144         err = session;
1145 out:
1146         write_unlock_irq(&tasklist_lock);
1147         if (err > 0) {
1148                 proc_sid_connector(group_leader);
1149                 sched_autogroup_create_attach(group_leader);
1150         }
1151         return err;
1152 }
1153
1154 DECLARE_RWSEM(uts_sem);
1155
1156 #ifdef COMPAT_UTS_MACHINE
1157 #define override_architecture(name) \
1158         (personality(current->personality) == PER_LINUX32 && \
1159          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1160                       sizeof(COMPAT_UTS_MACHINE)))
1161 #else
1162 #define override_architecture(name)     0
1163 #endif
1164
1165 /*
1166  * Work around broken programs that cannot handle "Linux 3.0".
1167  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1168  */
1169 static int override_release(char __user *release, size_t len)
1170 {
1171         int ret = 0;
1172
1173         if (current->personality & UNAME26) {
1174                 const char *rest = UTS_RELEASE;
1175                 char buf[65] = { 0 };
1176                 int ndots = 0;
1177                 unsigned v;
1178                 size_t copy;
1179
1180                 while (*rest) {
1181                         if (*rest == '.' && ++ndots >= 3)
1182                                 break;
1183                         if (!isdigit(*rest) && *rest != '.')
1184                                 break;
1185                         rest++;
1186                 }
1187                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1188                 copy = clamp_t(size_t, len, 1, sizeof(buf));
1189                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1190                 ret = copy_to_user(release, buf, copy + 1);
1191         }
1192         return ret;
1193 }
1194
1195 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1196 {
1197         int errno = 0;
1198
1199         down_read(&uts_sem);
1200         if (copy_to_user(name, utsname(), sizeof *name))
1201                 errno = -EFAULT;
1202         up_read(&uts_sem);
1203
1204         if (!errno && override_release(name->release, sizeof(name->release)))
1205                 errno = -EFAULT;
1206         if (!errno && override_architecture(name))
1207                 errno = -EFAULT;
1208         return errno;
1209 }
1210
1211 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1212 /*
1213  * Old cruft
1214  */
1215 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1216 {
1217         int error = 0;
1218
1219         if (!name)
1220                 return -EFAULT;
1221
1222         down_read(&uts_sem);
1223         if (copy_to_user(name, utsname(), sizeof(*name)))
1224                 error = -EFAULT;
1225         up_read(&uts_sem);
1226
1227         if (!error && override_release(name->release, sizeof(name->release)))
1228                 error = -EFAULT;
1229         if (!error && override_architecture(name))
1230                 error = -EFAULT;
1231         return error;
1232 }
1233
1234 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1235 {
1236         int error;
1237
1238         if (!name)
1239                 return -EFAULT;
1240         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1241                 return -EFAULT;
1242
1243         down_read(&uts_sem);
1244         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1245                                __OLD_UTS_LEN);
1246         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1247         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1248                                 __OLD_UTS_LEN);
1249         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1250         error |= __copy_to_user(&name->release, &utsname()->release,
1251                                 __OLD_UTS_LEN);
1252         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1253         error |= __copy_to_user(&name->version, &utsname()->version,
1254                                 __OLD_UTS_LEN);
1255         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1256         error |= __copy_to_user(&name->machine, &utsname()->machine,
1257                                 __OLD_UTS_LEN);
1258         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1259         up_read(&uts_sem);
1260
1261         if (!error && override_architecture(name))
1262                 error = -EFAULT;
1263         if (!error && override_release(name->release, sizeof(name->release)))
1264                 error = -EFAULT;
1265         return error ? -EFAULT : 0;
1266 }
1267 #endif
1268
1269 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1270 {
1271         int errno;
1272         char tmp[__NEW_UTS_LEN];
1273
1274         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1275                 return -EPERM;
1276
1277         if (len < 0 || len > __NEW_UTS_LEN)
1278                 return -EINVAL;
1279         down_write(&uts_sem);
1280         errno = -EFAULT;
1281         if (!copy_from_user(tmp, name, len)) {
1282                 struct new_utsname *u = utsname();
1283
1284                 memcpy(u->nodename, tmp, len);
1285                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1286                 errno = 0;
1287         }
1288         up_write(&uts_sem);
1289         return errno;
1290 }
1291
1292 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1293
1294 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1295 {
1296         int i, errno;
1297         struct new_utsname *u;
1298
1299         if (len < 0)
1300                 return -EINVAL;
1301         down_read(&uts_sem);
1302         u = utsname();
1303         i = 1 + strlen(u->nodename);
1304         if (i > len)
1305                 i = len;
1306         errno = 0;
1307         if (copy_to_user(name, u->nodename, i))
1308                 errno = -EFAULT;
1309         up_read(&uts_sem);
1310         return errno;
1311 }
1312
1313 #endif
1314
1315 /*
1316  * Only setdomainname; getdomainname can be implemented by calling
1317  * uname()
1318  */
1319 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1320 {
1321         int errno;
1322         char tmp[__NEW_UTS_LEN];
1323
1324         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1325                 return -EPERM;
1326         if (len < 0 || len > __NEW_UTS_LEN)
1327                 return -EINVAL;
1328
1329         down_write(&uts_sem);
1330         errno = -EFAULT;
1331         if (!copy_from_user(tmp, name, len)) {
1332                 struct new_utsname *u = utsname();
1333
1334                 memcpy(u->domainname, tmp, len);
1335                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1336                 errno = 0;
1337         }
1338         up_write(&uts_sem);
1339         return errno;
1340 }
1341
1342 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1343 {
1344         struct rlimit value;
1345         int ret;
1346
1347         ret = do_prlimit(current, resource, NULL, &value);
1348         if (!ret)
1349                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1350
1351         return ret;
1352 }
1353
1354 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1355
1356 /*
1357  *      Back compatibility for getrlimit. Needed for some apps.
1358  */
1359
1360 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1361                 struct rlimit __user *, rlim)
1362 {
1363         struct rlimit x;
1364         if (resource >= RLIM_NLIMITS)
1365                 return -EINVAL;
1366
1367         task_lock(current->group_leader);
1368         x = current->signal->rlim[resource];
1369         task_unlock(current->group_leader);
1370         if (x.rlim_cur > 0x7FFFFFFF)
1371                 x.rlim_cur = 0x7FFFFFFF;
1372         if (x.rlim_max > 0x7FFFFFFF)
1373                 x.rlim_max = 0x7FFFFFFF;
1374         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1375 }
1376
1377 #endif
1378
1379 static inline bool rlim64_is_infinity(__u64 rlim64)
1380 {
1381 #if BITS_PER_LONG < 64
1382         return rlim64 >= ULONG_MAX;
1383 #else
1384         return rlim64 == RLIM64_INFINITY;
1385 #endif
1386 }
1387
1388 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1389 {
1390         if (rlim->rlim_cur == RLIM_INFINITY)
1391                 rlim64->rlim_cur = RLIM64_INFINITY;
1392         else
1393                 rlim64->rlim_cur = rlim->rlim_cur;
1394         if (rlim->rlim_max == RLIM_INFINITY)
1395                 rlim64->rlim_max = RLIM64_INFINITY;
1396         else
1397                 rlim64->rlim_max = rlim->rlim_max;
1398 }
1399
1400 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1401 {
1402         if (rlim64_is_infinity(rlim64->rlim_cur))
1403                 rlim->rlim_cur = RLIM_INFINITY;
1404         else
1405                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1406         if (rlim64_is_infinity(rlim64->rlim_max))
1407                 rlim->rlim_max = RLIM_INFINITY;
1408         else
1409                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1410 }
1411
1412 /* make sure you are allowed to change @tsk limits before calling this */
1413 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1414                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1415 {
1416         struct rlimit *rlim;
1417         int retval = 0;
1418
1419         if (resource >= RLIM_NLIMITS)
1420                 return -EINVAL;
1421         if (new_rlim) {
1422                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1423                         return -EINVAL;
1424                 if (resource == RLIMIT_NOFILE &&
1425                                 new_rlim->rlim_max > sysctl_nr_open)
1426                         return -EPERM;
1427         }
1428
1429         /* protect tsk->signal and tsk->sighand from disappearing */
1430         read_lock(&tasklist_lock);
1431         if (!tsk->sighand) {
1432                 retval = -ESRCH;
1433                 goto out;
1434         }
1435
1436         rlim = tsk->signal->rlim + resource;
1437         task_lock(tsk->group_leader);
1438         if (new_rlim) {
1439                 /* Keep the capable check against init_user_ns until
1440                    cgroups can contain all limits */
1441                 if (new_rlim->rlim_max > rlim->rlim_max &&
1442                                 !capable(CAP_SYS_RESOURCE))
1443                         retval = -EPERM;
1444                 if (!retval)
1445                         retval = security_task_setrlimit(tsk->group_leader,
1446                                         resource, new_rlim);
1447                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1448                         /*
1449                          * The caller is asking for an immediate RLIMIT_CPU
1450                          * expiry.  But we use the zero value to mean "it was
1451                          * never set".  So let's cheat and make it one second
1452                          * instead
1453                          */
1454                         new_rlim->rlim_cur = 1;
1455                 }
1456         }
1457         if (!retval) {
1458                 if (old_rlim)
1459                         *old_rlim = *rlim;
1460                 if (new_rlim)
1461                         *rlim = *new_rlim;
1462         }
1463         task_unlock(tsk->group_leader);
1464
1465         /*
1466          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1467          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1468          * very long-standing error, and fixing it now risks breakage of
1469          * applications, so we live with it
1470          */
1471          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1472                          new_rlim->rlim_cur != RLIM_INFINITY)
1473                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1474 out:
1475         read_unlock(&tasklist_lock);
1476         return retval;
1477 }
1478
1479 /* rcu lock must be held */
1480 static int check_prlimit_permission(struct task_struct *task)
1481 {
1482         const struct cred *cred = current_cred(), *tcred;
1483
1484         if (current == task)
1485                 return 0;
1486
1487         tcred = __task_cred(task);
1488         if (cred->user->user_ns == tcred->user->user_ns &&
1489             (cred->uid == tcred->euid &&
1490              cred->uid == tcred->suid &&
1491              cred->uid == tcred->uid  &&
1492              cred->gid == tcred->egid &&
1493              cred->gid == tcred->sgid &&
1494              cred->gid == tcred->gid))
1495                 return 0;
1496         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1497                 return 0;
1498
1499         return -EPERM;
1500 }
1501
1502 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1503                 const struct rlimit64 __user *, new_rlim,
1504                 struct rlimit64 __user *, old_rlim)
1505 {
1506         struct rlimit64 old64, new64;
1507         struct rlimit old, new;
1508         struct task_struct *tsk;
1509         int ret;
1510
1511         if (new_rlim) {
1512                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1513                         return -EFAULT;
1514                 rlim64_to_rlim(&new64, &new);
1515         }
1516
1517         rcu_read_lock();
1518         tsk = pid ? find_task_by_vpid(pid) : current;
1519         if (!tsk) {
1520                 rcu_read_unlock();
1521                 return -ESRCH;
1522         }
1523         ret = check_prlimit_permission(tsk);
1524         if (ret) {
1525                 rcu_read_unlock();
1526                 return ret;
1527         }
1528         get_task_struct(tsk);
1529         rcu_read_unlock();
1530
1531         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1532                         old_rlim ? &old : NULL);
1533
1534         if (!ret && old_rlim) {
1535                 rlim_to_rlim64(&old, &old64);
1536                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1537                         ret = -EFAULT;
1538         }
1539
1540         put_task_struct(tsk);
1541         return ret;
1542 }
1543
1544 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1545 {
1546         struct rlimit new_rlim;
1547
1548         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1549                 return -EFAULT;
1550         return do_prlimit(current, resource, &new_rlim, NULL);
1551 }
1552
1553 /*
1554  * It would make sense to put struct rusage in the task_struct,
1555  * except that would make the task_struct be *really big*.  After
1556  * task_struct gets moved into malloc'ed memory, it would
1557  * make sense to do this.  It will make moving the rest of the information
1558  * a lot simpler!  (Which we're not doing right now because we're not
1559  * measuring them yet).
1560  *
1561  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1562  * races with threads incrementing their own counters.  But since word
1563  * reads are atomic, we either get new values or old values and we don't
1564  * care which for the sums.  We always take the siglock to protect reading
1565  * the c* fields from p->signal from races with exit.c updating those
1566  * fields when reaping, so a sample either gets all the additions of a
1567  * given child after it's reaped, or none so this sample is before reaping.
1568  *
1569  * Locking:
1570  * We need to take the siglock for CHILDEREN, SELF and BOTH
1571  * for  the cases current multithreaded, non-current single threaded
1572  * non-current multithreaded.  Thread traversal is now safe with
1573  * the siglock held.
1574  * Strictly speaking, we donot need to take the siglock if we are current and
1575  * single threaded,  as no one else can take our signal_struct away, no one
1576  * else can  reap the  children to update signal->c* counters, and no one else
1577  * can race with the signal-> fields. If we do not take any lock, the
1578  * signal-> fields could be read out of order while another thread was just
1579  * exiting. So we should  place a read memory barrier when we avoid the lock.
1580  * On the writer side,  write memory barrier is implied in  __exit_signal
1581  * as __exit_signal releases  the siglock spinlock after updating the signal->
1582  * fields. But we don't do this yet to keep things simple.
1583  *
1584  */
1585
1586 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1587 {
1588         r->ru_nvcsw += t->nvcsw;
1589         r->ru_nivcsw += t->nivcsw;
1590         r->ru_minflt += t->min_flt;
1591         r->ru_majflt += t->maj_flt;
1592         r->ru_inblock += task_io_get_inblock(t);
1593         r->ru_oublock += task_io_get_oublock(t);
1594 }
1595
1596 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1597 {
1598         struct task_struct *t;
1599         unsigned long flags;
1600         cputime_t tgutime, tgstime, utime, stime;
1601         unsigned long maxrss = 0;
1602
1603         memset((char *) r, 0, sizeof *r);
1604         utime = stime = cputime_zero;
1605
1606         if (who == RUSAGE_THREAD) {
1607                 task_times(current, &utime, &stime);
1608                 accumulate_thread_rusage(p, r);
1609                 maxrss = p->signal->maxrss;
1610                 goto out;
1611         }
1612
1613         if (!lock_task_sighand(p, &flags))
1614                 return;
1615
1616         switch (who) {
1617                 case RUSAGE_BOTH:
1618                 case RUSAGE_CHILDREN:
1619                         utime = p->signal->cutime;
1620                         stime = p->signal->cstime;
1621                         r->ru_nvcsw = p->signal->cnvcsw;
1622                         r->ru_nivcsw = p->signal->cnivcsw;
1623                         r->ru_minflt = p->signal->cmin_flt;
1624                         r->ru_majflt = p->signal->cmaj_flt;
1625                         r->ru_inblock = p->signal->cinblock;
1626                         r->ru_oublock = p->signal->coublock;
1627                         maxrss = p->signal->cmaxrss;
1628
1629                         if (who == RUSAGE_CHILDREN)
1630                                 break;
1631
1632                 case RUSAGE_SELF:
1633                         thread_group_times(p, &tgutime, &tgstime);
1634                         utime = cputime_add(utime, tgutime);
1635                         stime = cputime_add(stime, tgstime);
1636                         r->ru_nvcsw += p->signal->nvcsw;
1637                         r->ru_nivcsw += p->signal->nivcsw;
1638                         r->ru_minflt += p->signal->min_flt;
1639                         r->ru_majflt += p->signal->maj_flt;
1640                         r->ru_inblock += p->signal->inblock;
1641                         r->ru_oublock += p->signal->oublock;
1642                         if (maxrss < p->signal->maxrss)
1643                                 maxrss = p->signal->maxrss;
1644                         t = p;
1645                         do {
1646                                 accumulate_thread_rusage(t, r);
1647                                 t = next_thread(t);
1648                         } while (t != p);
1649                         break;
1650
1651                 default:
1652                         BUG();
1653         }
1654         unlock_task_sighand(p, &flags);
1655
1656 out:
1657         cputime_to_timeval(utime, &r->ru_utime);
1658         cputime_to_timeval(stime, &r->ru_stime);
1659
1660         if (who != RUSAGE_CHILDREN) {
1661                 struct mm_struct *mm = get_task_mm(p);
1662                 if (mm) {
1663                         setmax_mm_hiwater_rss(&maxrss, mm);
1664                         mmput(mm);
1665                 }
1666         }
1667         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1668 }
1669
1670 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1671 {
1672         struct rusage r;
1673         k_getrusage(p, who, &r);
1674         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1675 }
1676
1677 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1678 {
1679         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1680             who != RUSAGE_THREAD)
1681                 return -EINVAL;
1682         return getrusage(current, who, ru);
1683 }
1684
1685 SYSCALL_DEFINE1(umask, int, mask)
1686 {
1687         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1688         return mask;
1689 }
1690
1691 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1692                 unsigned long, arg4, unsigned long, arg5)
1693 {
1694         struct task_struct *me = current;
1695         unsigned char comm[sizeof(me->comm)];
1696         long error;
1697
1698         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1699         if (error != -ENOSYS)
1700                 return error;
1701
1702         error = 0;
1703         switch (option) {
1704                 case PR_SET_PDEATHSIG:
1705                         if (!valid_signal(arg2)) {
1706                                 error = -EINVAL;
1707                                 break;
1708                         }
1709                         me->pdeath_signal = arg2;
1710                         error = 0;
1711                         break;
1712                 case PR_GET_PDEATHSIG:
1713                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1714                         break;
1715                 case PR_GET_DUMPABLE:
1716                         error = get_dumpable(me->mm);
1717                         break;
1718                 case PR_SET_DUMPABLE:
1719                         if (arg2 < 0 || arg2 > 1) {
1720                                 error = -EINVAL;
1721                                 break;
1722                         }
1723                         set_dumpable(me->mm, arg2);
1724                         error = 0;
1725                         break;
1726
1727                 case PR_SET_UNALIGN:
1728                         error = SET_UNALIGN_CTL(me, arg2);
1729                         break;
1730                 case PR_GET_UNALIGN:
1731                         error = GET_UNALIGN_CTL(me, arg2);
1732                         break;
1733                 case PR_SET_FPEMU:
1734                         error = SET_FPEMU_CTL(me, arg2);
1735                         break;
1736                 case PR_GET_FPEMU:
1737                         error = GET_FPEMU_CTL(me, arg2);
1738                         break;
1739                 case PR_SET_FPEXC:
1740                         error = SET_FPEXC_CTL(me, arg2);
1741                         break;
1742                 case PR_GET_FPEXC:
1743                         error = GET_FPEXC_CTL(me, arg2);
1744                         break;
1745                 case PR_GET_TIMING:
1746                         error = PR_TIMING_STATISTICAL;
1747                         break;
1748                 case PR_SET_TIMING:
1749                         if (arg2 != PR_TIMING_STATISTICAL)
1750                                 error = -EINVAL;
1751                         else
1752                                 error = 0;
1753                         break;
1754
1755                 case PR_SET_NAME:
1756                         comm[sizeof(me->comm)-1] = 0;
1757                         if (strncpy_from_user(comm, (char __user *)arg2,
1758                                               sizeof(me->comm) - 1) < 0)
1759                                 return -EFAULT;
1760                         set_task_comm(me, comm);
1761                         return 0;
1762                 case PR_GET_NAME:
1763                         get_task_comm(comm, me);
1764                         if (copy_to_user((char __user *)arg2, comm,
1765                                          sizeof(comm)))
1766                                 return -EFAULT;
1767                         return 0;
1768                 case PR_GET_ENDIAN:
1769                         error = GET_ENDIAN(me, arg2);
1770                         break;
1771                 case PR_SET_ENDIAN:
1772                         error = SET_ENDIAN(me, arg2);
1773                         break;
1774
1775                 case PR_GET_SECCOMP:
1776                         error = prctl_get_seccomp();
1777                         break;
1778                 case PR_SET_SECCOMP:
1779                         error = prctl_set_seccomp(arg2);
1780                         break;
1781                 case PR_GET_TSC:
1782                         error = GET_TSC_CTL(arg2);
1783                         break;
1784                 case PR_SET_TSC:
1785                         error = SET_TSC_CTL(arg2);
1786                         break;
1787                 case PR_TASK_PERF_EVENTS_DISABLE:
1788                         error = perf_event_task_disable();
1789                         break;
1790                 case PR_TASK_PERF_EVENTS_ENABLE:
1791                         error = perf_event_task_enable();
1792                         break;
1793                 case PR_GET_TIMERSLACK:
1794                         error = current->timer_slack_ns;
1795                         break;
1796                 case PR_SET_TIMERSLACK:
1797                         if (arg2 <= 0)
1798                                 current->timer_slack_ns =
1799                                         current->default_timer_slack_ns;
1800                         else
1801                                 current->timer_slack_ns = arg2;
1802                         error = 0;
1803                         break;
1804                 case PR_MCE_KILL:
1805                         if (arg4 | arg5)
1806                                 return -EINVAL;
1807                         switch (arg2) {
1808                         case PR_MCE_KILL_CLEAR:
1809                                 if (arg3 != 0)
1810                                         return -EINVAL;
1811                                 current->flags &= ~PF_MCE_PROCESS;
1812                                 break;
1813                         case PR_MCE_KILL_SET:
1814                                 current->flags |= PF_MCE_PROCESS;
1815                                 if (arg3 == PR_MCE_KILL_EARLY)
1816                                         current->flags |= PF_MCE_EARLY;
1817                                 else if (arg3 == PR_MCE_KILL_LATE)
1818                                         current->flags &= ~PF_MCE_EARLY;
1819                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1820                                         current->flags &=
1821                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1822                                 else
1823                                         return -EINVAL;
1824                                 break;
1825                         default:
1826                                 return -EINVAL;
1827                         }
1828                         error = 0;
1829                         break;
1830                 case PR_MCE_KILL_GET:
1831                         if (arg2 | arg3 | arg4 | arg5)
1832                                 return -EINVAL;
1833                         if (current->flags & PF_MCE_PROCESS)
1834                                 error = (current->flags & PF_MCE_EARLY) ?
1835                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1836                         else
1837                                 error = PR_MCE_KILL_DEFAULT;
1838                         break;
1839                 default:
1840                         error = -EINVAL;
1841                         break;
1842         }
1843         return error;
1844 }
1845
1846 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1847                 struct getcpu_cache __user *, unused)
1848 {
1849         int err = 0;
1850         int cpu = raw_smp_processor_id();
1851         if (cpup)
1852                 err |= put_user(cpu, cpup);
1853         if (nodep)
1854                 err |= put_user(cpu_to_node(cpu), nodep);
1855         return err ? -EFAULT : 0;
1856 }
1857
1858 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1859
1860 static void argv_cleanup(struct subprocess_info *info)
1861 {
1862         argv_free(info->argv);
1863 }
1864
1865 /**
1866  * orderly_poweroff - Trigger an orderly system poweroff
1867  * @force: force poweroff if command execution fails
1868  *
1869  * This may be called from any context to trigger a system shutdown.
1870  * If the orderly shutdown fails, it will force an immediate shutdown.
1871  */
1872 int orderly_poweroff(bool force)
1873 {
1874         int argc;
1875         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1876         static char *envp[] = {
1877                 "HOME=/",
1878                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1879                 NULL
1880         };
1881         int ret = -ENOMEM;
1882         struct subprocess_info *info;
1883
1884         if (argv == NULL) {
1885                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1886                        __func__, poweroff_cmd);
1887                 goto out;
1888         }
1889
1890         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1891         if (info == NULL) {
1892                 argv_free(argv);
1893                 goto out;
1894         }
1895
1896         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1897
1898         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1899
1900   out:
1901         if (ret && force) {
1902                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1903                        "forcing the issue\n");
1904
1905                 /* I guess this should try to kick off some daemon to
1906                    sync and poweroff asap.  Or not even bother syncing
1907                    if we're doing an emergency shutdown? */
1908                 emergency_sync();
1909                 kernel_power_off();
1910         }
1911
1912         return ret;
1913 }
1914 EXPORT_SYMBOL_GPL(orderly_poweroff);