kernel/sys.c

   1 /*
   2  *  linux/kernel/sys.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/mm.h>
   9 #include <linux/utsname.h>
  10 #include <linux/mman.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/notifier.h>
  13 #include <linux/reboot.h>
  14 #include <linux/prctl.h>
  15 #include <linux/highuid.h>
  16 #include <linux/fs.h>
  17 #include <linux/resource.h>
  18 #include <linux/kernel.h>
  19 #include <linux/kexec.h>
  20 #include <linux/workqueue.h>
  21 #include <linux/capability.h>
  22 #include <linux/device.h>
  23 #include <linux/key.h>
  24 #include <linux/times.h>
  25 #include <linux/posix-timers.h>
  26 #include <linux/security.h>
  27 #include <linux/dcookies.h>
  28 #include <linux/suspend.h>
  29 #include <linux/tty.h>
  30 #include <linux/signal.h>
  31 #include <linux/cn_proc.h>
  32 #include <linux/getcpu.h>
  33 #include <linux/task_io_accounting_ops.h>
  34 #include <linux/seccomp.h>
  35 #include <linux/cpu.h>
  36
  37 #include <linux/compat.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/kprobes.h>
  40 #include <linux/user_namespace.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/io.h>
  44 #include <asm/unistd.h>
  45
  46 #ifndef SET_UNALIGN_CTL
  47 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
  48 #endif
  49 #ifndef GET_UNALIGN_CTL
  50 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
  51 #endif
  52 #ifndef SET_FPEMU_CTL
  53 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
  54 #endif
  55 #ifndef GET_FPEMU_CTL
  56 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
  57 #endif
  58 #ifndef SET_FPEXC_CTL
  59 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
  60 #endif
  61 #ifndef GET_FPEXC_CTL
  62 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
  63 #endif
  64 #ifndef GET_ENDIAN
  65 # define GET_ENDIAN(a,b)        (-EINVAL)
  66 #endif
  67 #ifndef SET_ENDIAN
  68 # define SET_ENDIAN(a,b)        (-EINVAL)
  69 #endif
  70
  71 /*
  72  * this is where the system-wide overflow UID and GID are defined, for
  73  * architectures that now have 32-bit UID/GID but didn't in the past
  74  */
  75
  76 int overflowuid = DEFAULT_OVERFLOWUID;
  77 int overflowgid = DEFAULT_OVERFLOWGID;
  78
  79 #ifdef CONFIG_UID16
  80 EXPORT_SYMBOL(overflowuid);
  81 EXPORT_SYMBOL(overflowgid);
  82 #endif
  83
  84 /*
  85  * the same as above, but for filesystems which can only store a 16-bit
  86  * UID and GID. as such, this is needed on all architectures
  87  */
  88
  89 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
  90 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
  91
  92 EXPORT_SYMBOL(fs_overflowuid);
  93 EXPORT_SYMBOL(fs_overflowgid);
  94
  95 /*
  96  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
  97  */
  98
  99 int C_A_D = 1;
 100 struct pid *cad_pid;
 101 EXPORT_SYMBOL(cad_pid);
 102
 103 /*
 104  * If set, this is used for preparing the system to power off.
 105  */
 106
 107 void (*pm_power_off_prepare)(void);
 108
 109 static int set_one_prio(struct task_struct *p, int niceval, int error)
 110 {
 111         int no_nice;
 112
 113         if (p->uid != current->euid &&
 114                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
 115                 error = -EPERM;
 116                 goto out;
 117         }
 118         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 119                 error = -EACCES;
 120                 goto out;
 121         }
 122         no_nice = security_task_setnice(p, niceval);
 123         if (no_nice) {
 124                 error = no_nice;
 125                 goto out;
 126         }
 127         if (error == -ESRCH)
 128                 error = 0;
 129         set_user_nice(p, niceval);
 130 out:
 131         return error;
 132 }
 133
 134 asmlinkage long sys_setpriority(int which, int who, int niceval)
 135 {
 136         struct task_struct *g, *p;
 137         struct user_struct *user;
 138         int error = -EINVAL;
 139         struct pid *pgrp;
 140
 141         if (which > PRIO_USER || which < PRIO_PROCESS)
 142                 goto out;
 143
 144         /* normalize: avoid signed division (rounding problems) */
 145         error = -ESRCH;
 146         if (niceval < -20)
 147                 niceval = -20;
 148         if (niceval > 19)
 149                 niceval = 19;
 150
 151         read_lock(&tasklist_lock);
 152         switch (which) {
 153                 case PRIO_PROCESS:
 154                         if (who)
 155                                 p = find_task_by_pid_ns(who,
 156                                                 current->nsproxy->pid_ns);
 157                         else
 158                                 p = current;
 159                         if (p)
 160                                 error = set_one_prio(p, niceval, error);
 161                         break;
 162                 case PRIO_PGRP:
 163                         if (who)
 164                                 pgrp = find_vpid(who);
 165                         else
 166                                 pgrp = task_pgrp(current);
 167                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 168                                 error = set_one_prio(p, niceval, error);
 169                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 170                         break;
 171                 case PRIO_USER:
 172                         user = current->user;
 173                         if (!who)
 174                                 who = current->uid;
 175                         else
 176                                 if ((who != current->uid) && !(user = find_user(who)))
 177                                         goto out_unlock;        /* No processes for this user */
 178
 179                         do_each_thread(g, p)
 180                                 if (p->uid == who)
 181                                         error = set_one_prio(p, niceval, error);
 182                         while_each_thread(g, p);
 183                         if (who != current->uid)
 184                                 free_uid(user);         /* For find_user() */
 185                         break;
 186         }
 187 out_unlock:
 188         read_unlock(&tasklist_lock);
 189 out:
 190         return error;
 191 }
 192
 193 /*
 194  * Ugh. To avoid negative return values, "getpriority()" will
 195  * not return the normal nice-value, but a negated value that
 196  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 197  * to stay compatible.
 198  */
 199 asmlinkage long sys_getpriority(int which, int who)
 200 {
 201         struct task_struct *g, *p;
 202         struct user_struct *user;
 203         long niceval, retval = -ESRCH;
 204         struct pid *pgrp;
 205
 206         if (which > PRIO_USER || which < PRIO_PROCESS)
 207                 return -EINVAL;
 208
 209         read_lock(&tasklist_lock);
 210         switch (which) {
 211                 case PRIO_PROCESS:
 212                         if (who)
 213                                 p = find_task_by_pid_ns(who,
 214                                                 current->nsproxy->pid_ns);
 215                         else
 216                                 p = current;
 217                         if (p) {
 218                                 niceval = 20 - task_nice(p);
 219                                 if (niceval > retval)
 220                                         retval = niceval;
 221                         }
 222                         break;
 223                 case PRIO_PGRP:
 224                         if (who)
 225                                 pgrp = find_vpid(who);
 226                         else
 227                                 pgrp = task_pgrp(current);
 228                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 229                                 niceval = 20 - task_nice(p);
 230                                 if (niceval > retval)
 231                                         retval = niceval;
 232                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 233                         break;
 234                 case PRIO_USER:
 235                         user = current->user;
 236                         if (!who)
 237                                 who = current->uid;
 238                         else
 239                                 if ((who != current->uid) && !(user = find_user(who)))
 240                                         goto out_unlock;        /* No processes for this user */
 241
 242                         do_each_thread(g, p)
 243                                 if (p->uid == who) {
 244                                         niceval = 20 - task_nice(p);
 245                                         if (niceval > retval)
 246                                                 retval = niceval;
 247                                 }
 248                         while_each_thread(g, p);
 249                         if (who != current->uid)
 250                                 free_uid(user);         /* for find_user() */
 251                         break;
 252         }
 253 out_unlock:
 254         read_unlock(&tasklist_lock);
 255
 256         return retval;
 257 }
 258
 259 /**
 260  *      emergency_restart - reboot the system
 261  *
 262  *      Without shutting down any hardware or taking any locks
 263  *      reboot the system.  This is called when we know we are in
 264  *      trouble so this is our best effort to reboot.  This is
 265  *      safe to call in interrupt context.
 266  */
 267 void emergency_restart(void)
 268 {
 269         machine_emergency_restart();
 270 }
 271 EXPORT_SYMBOL_GPL(emergency_restart);
 272
 273 static void kernel_restart_prepare(char *cmd)
 274 {
 275         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
 276         system_state = SYSTEM_RESTART;
 277         device_shutdown();
 278         sysdev_shutdown();
 279 }
 280
 281 /**
 282  *      kernel_restart - reboot the system
 283  *      @cmd: pointer to buffer containing command to execute for restart
 284  *              or %NULL
 285  *
 286  *      Shutdown everything and perform a clean reboot.
 287  *      This is not safe to call in interrupt context.
 288  */
 289 void kernel_restart(char *cmd)
 290 {
 291         kernel_restart_prepare(cmd);
 292         if (!cmd)
 293                 printk(KERN_EMERG "Restarting system.\n");
 294         else
 295                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
 296         machine_restart(cmd);
 297 }
 298 EXPORT_SYMBOL_GPL(kernel_restart);
 299
 300 /**
 301  *      kernel_kexec - reboot the system
 302  *
 303  *      Move into place and start executing a preloaded standalone
 304  *      executable.  If nothing was preloaded return an error.
 305  */
 306 static void kernel_kexec(void)
 307 {
 308 #ifdef CONFIG_KEXEC
 309         struct kimage *image;
 310         image = xchg(&kexec_image, NULL);
 311         if (!image)
 312                 return;
 313         kernel_restart_prepare(NULL);
 314         printk(KERN_EMERG "Starting new kernel\n");
 315         machine_shutdown();
 316         machine_kexec(image);
 317 #endif
 318 }
 319
 320 void kernel_shutdown_prepare(enum system_states state)
 321 {
 322         blocking_notifier_call_chain(&reboot_notifier_list,
 323                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
 324         system_state = state;
 325         device_shutdown();
 326 }
 327 /**
 328  *      kernel_halt - halt the system
 329  *
 330  *      Shutdown everything and perform a clean system halt.
 331  */
 332 void kernel_halt(void)
 333 {
 334         kernel_shutdown_prepare(SYSTEM_HALT);
 335         sysdev_shutdown();
 336         printk(KERN_EMERG "System halted.\n");
 337         machine_halt();
 338 }
 339
 340 EXPORT_SYMBOL_GPL(kernel_halt);
 341
 342 /**
 343  *      kernel_power_off - power_off the system
 344  *
 345  *      Shutdown everything and perform a clean system power_off.
 346  */
 347 void kernel_power_off(void)
 348 {
 349         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
 350         if (pm_power_off_prepare)
 351                 pm_power_off_prepare();
 352         disable_nonboot_cpus();
 353         sysdev_shutdown();
 354         printk(KERN_EMERG "Power down.\n");
 355         machine_power_off();
 356 }
 357 EXPORT_SYMBOL_GPL(kernel_power_off);
 358 /*
 359  * Reboot system call: for obvious reasons only root may call it,
 360  * and even root needs to set up some magic numbers in the registers
 361  * so that some mistake won't make this reboot the whole machine.
 362  * You can also set the meaning of the ctrl-alt-del-key here.
 363  *
 364  * reboot doesn't sync: do that yourself before calling this.
 365  */
 366 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
 367 {
 368         char buffer[256];
 369
 370         /* We only trust the superuser with rebooting the system. */
 371         if (!capable(CAP_SYS_BOOT))
 372                 return -EPERM;
 373
 374         /* For safety, we require "magic" arguments. */
 375         if (magic1 != LINUX_REBOOT_MAGIC1 ||
 376             (magic2 != LINUX_REBOOT_MAGIC2 &&
 377                         magic2 != LINUX_REBOOT_MAGIC2A &&
 378                         magic2 != LINUX_REBOOT_MAGIC2B &&
 379                         magic2 != LINUX_REBOOT_MAGIC2C))
 380                 return -EINVAL;
 381
 382         /* Instead of trying to make the power_off code look like
 383          * halt when pm_power_off is not set do it the easy way.
 384          */
 385         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
 386                 cmd = LINUX_REBOOT_CMD_HALT;
 387
 388         lock_kernel();
 389         switch (cmd) {
 390         case LINUX_REBOOT_CMD_RESTART:
 391                 kernel_restart(NULL);
 392                 break;
 393
 394         case LINUX_REBOOT_CMD_CAD_ON:
 395                 C_A_D = 1;
 396                 break;
 397
 398         case LINUX_REBOOT_CMD_CAD_OFF:
 399                 C_A_D = 0;
 400                 break;
 401
 402         case LINUX_REBOOT_CMD_HALT:
 403                 kernel_halt();
 404                 unlock_kernel();
 405                 do_exit(0);
 406                 break;
 407
 408         case LINUX_REBOOT_CMD_POWER_OFF:
 409                 kernel_power_off();
 410                 unlock_kernel();
 411                 do_exit(0);
 412                 break;
 413
 414         case LINUX_REBOOT_CMD_RESTART2:
 415                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
 416                         unlock_kernel();
 417                         return -EFAULT;
 418                 }
 419                 buffer[sizeof(buffer) - 1] = '\0';
 420
 421                 kernel_restart(buffer);
 422                 break;
 423
 424         case LINUX_REBOOT_CMD_KEXEC:
 425                 kernel_kexec();
 426                 unlock_kernel();
 427                 return -EINVAL;
 428
 429 #ifdef CONFIG_HIBERNATION
 430         case LINUX_REBOOT_CMD_SW_SUSPEND:
 431                 {
 432                         int ret = hibernate();
 433                         unlock_kernel();
 434                         return ret;
 435                 }
 436 #endif
 437
 438         default:
 439                 unlock_kernel();
 440                 return -EINVAL;
 441         }
 442         unlock_kernel();
 443         return 0;
 444 }
 445
 446 static void deferred_cad(struct work_struct *dummy)
 447 {
 448         kernel_restart(NULL);
 449 }
 450
 451 /*
 452  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
 453  * As it's called within an interrupt, it may NOT sync: the only choice
 454  * is whether to reboot at once, or just ignore the ctrl-alt-del.
 455  */
 456 void ctrl_alt_del(void)
 457 {
 458         static DECLARE_WORK(cad_work, deferred_cad);
 459
 460         if (C_A_D)
 461                 schedule_work(&cad_work);
 462         else
 463                 kill_cad_pid(SIGINT, 1);
 464 }
 465
 466 /*
 467  * Unprivileged users may change the real gid to the effective gid
 468  * or vice versa.  (BSD-style)
 469  *
 470  * If you set the real gid at all, or set the effective gid to a value not
 471  * equal to the real gid, then the saved gid is set to the new effective gid.
 472  *
 473  * This makes it possible for a setgid program to completely drop its
 474  * privileges, which is often a useful assertion to make when you are doing
 475  * a security audit over a program.
 476  *
 477  * The general idea is that a program which uses just setregid() will be
 478  * 100% compatible with BSD.  A program which uses just setgid() will be
 479  * 100% compatible with POSIX with saved IDs.
 480  *
 481  * SMP: There are not races, the GIDs are checked only by filesystem
 482  *      operations (as far as semantic preservation is concerned).
 483  */
 484 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
 485 {
 486         int old_rgid = current->gid;
 487         int old_egid = current->egid;
 488         int new_rgid = old_rgid;
 489         int new_egid = old_egid;
 490         int retval;
 491
 492         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
 493         if (retval)
 494                 return retval;
 495
 496         if (rgid != (gid_t) -1) {
 497                 if ((old_rgid == rgid) ||
 498                     (current->egid==rgid) ||
 499                     capable(CAP_SETGID))
 500                         new_rgid = rgid;
 501                 else
 502                         return -EPERM;
 503         }
 504         if (egid != (gid_t) -1) {
 505                 if ((old_rgid == egid) ||
 506                     (current->egid == egid) ||
 507                     (current->sgid == egid) ||
 508                     capable(CAP_SETGID))
 509                         new_egid = egid;
 510                 else
 511                         return -EPERM;
 512         }
 513         if (new_egid != old_egid) {
 514                 set_dumpable(current->mm, suid_dumpable);
 515                 smp_wmb();
 516         }
 517         if (rgid != (gid_t) -1 ||
 518             (egid != (gid_t) -1 && egid != old_rgid))
 519                 current->sgid = new_egid;
 520         current->fsgid = new_egid;
 521         current->egid = new_egid;
 522         current->gid = new_rgid;
 523         key_fsgid_changed(current);
 524         proc_id_connector(current, PROC_EVENT_GID);
 525         return 0;
 526 }
 527
 528 /*
 529  * setgid() is implemented like SysV w/ SAVED_IDS
 530  *
 531  * SMP: Same implicit races as above.
 532  */
 533 asmlinkage long sys_setgid(gid_t gid)
 534 {
 535         int old_egid = current->egid;
 536         int retval;
 537
 538         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
 539         if (retval)
 540                 return retval;
 541
 542         if (capable(CAP_SETGID)) {
 543                 if (old_egid != gid) {
 544                         set_dumpable(current->mm, suid_dumpable);
 545                         smp_wmb();
 546                 }
 547                 current->gid = current->egid = current->sgid = current->fsgid = gid;
 548         } else if ((gid == current->gid) || (gid == current->sgid)) {
 549                 if (old_egid != gid) {
 550                         set_dumpable(current->mm, suid_dumpable);
 551                         smp_wmb();
 552                 }
 553                 current->egid = current->fsgid = gid;
 554         }
 555         else
 556                 return -EPERM;
 557
 558         key_fsgid_changed(current);
 559         proc_id_connector(current, PROC_EVENT_GID);
 560         return 0;
 561 }
 562
 563 static int set_user(uid_t new_ruid, int dumpclear)
 564 {
 565         struct user_struct *new_user;
 566
 567         new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
 568         if (!new_user)
 569                 return -EAGAIN;
 570
 571         if (atomic_read(&new_user->processes) >=
 572                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
 573                         new_user != current->nsproxy->user_ns->root_user) {
 574                 free_uid(new_user);
 575                 return -EAGAIN;
 576         }
 577
 578         switch_uid(new_user);
 579
 580         if (dumpclear) {
 581                 set_dumpable(current->mm, suid_dumpable);
 582                 smp_wmb();
 583         }
 584         current->uid = new_ruid;
 585         return 0;
 586 }
 587
 588 /*
 589  * Unprivileged users may change the real uid to the effective uid
 590  * or vice versa.  (BSD-style)
 591  *
 592  * If you set the real uid at all, or set the effective uid to a value not
 593  * equal to the real uid, then the saved uid is set to the new effective uid.
 594  *
 595  * This makes it possible for a setuid program to completely drop its
 596  * privileges, which is often a useful assertion to make when you are doing
 597  * a security audit over a program.
 598  *
 599  * The general idea is that a program which uses just setreuid() will be
 600  * 100% compatible with BSD.  A program which uses just setuid() will be
 601  * 100% compatible with POSIX with saved IDs.
 602  */
 603 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
 604 {
 605         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
 606         int retval;
 607
 608         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
 609         if (retval)
 610                 return retval;
 611
 612         new_ruid = old_ruid = current->uid;
 613         new_euid = old_euid = current->euid;
 614         old_suid = current->suid;
 615
 616         if (ruid != (uid_t) -1) {
 617                 new_ruid = ruid;
 618                 if ((old_ruid != ruid) &&
 619                     (current->euid != ruid) &&
 620                     !capable(CAP_SETUID))
 621                         return -EPERM;
 622         }
 623
 624         if (euid != (uid_t) -1) {
 625                 new_euid = euid;
 626                 if ((old_ruid != euid) &&
 627                     (current->euid != euid) &&
 628                     (current->suid != euid) &&
 629                     !capable(CAP_SETUID))
 630                         return -EPERM;
 631         }
 632
 633         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
 634                 return -EAGAIN;
 635
 636         if (new_euid != old_euid) {
 637                 set_dumpable(current->mm, suid_dumpable);
 638                 smp_wmb();
 639         }
 640         current->fsuid = current->euid = new_euid;
 641         if (ruid != (uid_t) -1 ||
 642             (euid != (uid_t) -1 && euid != old_ruid))
 643                 current->suid = current->euid;
 644         current->fsuid = current->euid;
 645
 646         key_fsuid_changed(current);
 647         proc_id_connector(current, PROC_EVENT_UID);
 648
 649         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
 650 }
 651
 652
 653
 654 /*
 655  * setuid() is implemented like SysV with SAVED_IDS
 656  *
 657  * Note that SAVED_ID's is deficient in that a setuid root program
 658  * like sendmail, for example, cannot set its uid to be a normal
 659  * user and then switch back, because if you're root, setuid() sets
 660  * the saved uid too.  If you don't like this, blame the bright people
 661  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 662  * will allow a root program to temporarily drop privileges and be able to
 663  * regain them by swapping the real and effective uid.
 664  */
 665 asmlinkage long sys_setuid(uid_t uid)
 666 {
 667         int old_euid = current->euid;
 668         int old_ruid, old_suid, new_suid;
 669         int retval;
 670
 671         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
 672         if (retval)
 673                 return retval;
 674
 675         old_ruid = current->uid;
 676         old_suid = current->suid;
 677         new_suid = old_suid;
 678
 679         if (capable(CAP_SETUID)) {
 680                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
 681                         return -EAGAIN;
 682                 new_suid = uid;
 683         } else if ((uid != current->uid) && (uid != new_suid))
 684                 return -EPERM;
 685
 686         if (old_euid != uid) {
 687                 set_dumpable(current->mm, suid_dumpable);
 688                 smp_wmb();
 689         }
 690         current->fsuid = current->euid = uid;
 691         current->suid = new_suid;
 692
 693         key_fsuid_changed(current);
 694         proc_id_connector(current, PROC_EVENT_UID);
 695
 696         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
 697 }
 698
 699
 700 /*
 701  * This function implements a generic ability to update ruid, euid,
 702  * and suid.  This allows you to implement the 4.4 compatible seteuid().
 703  */
 704 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 705 {
 706         int old_ruid = current->uid;
 707         int old_euid = current->euid;
 708         int old_suid = current->suid;
 709         int retval;
 710
 711         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
 712         if (retval)
 713                 return retval;
 714
 715         if (!capable(CAP_SETUID)) {
 716                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
 717                     (ruid != current->euid) && (ruid != current->suid))
 718                         return -EPERM;
 719                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
 720                     (euid != current->euid) && (euid != current->suid))
 721                         return -EPERM;
 722                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
 723                     (suid != current->euid) && (suid != current->suid))
 724                         return -EPERM;
 725         }
 726         if (ruid != (uid_t) -1) {
 727                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
 728                         return -EAGAIN;
 729         }
 730         if (euid != (uid_t) -1) {
 731                 if (euid != current->euid) {
 732                         set_dumpable(current->mm, suid_dumpable);
 733                         smp_wmb();
 734                 }
 735                 current->euid = euid;
 736         }
 737         current->fsuid = current->euid;
 738         if (suid != (uid_t) -1)
 739                 current->suid = suid;
 740
 741         key_fsuid_changed(current);
 742         proc_id_connector(current, PROC_EVENT_UID);
 743
 744         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
 745 }
 746
 747 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
 748 {
 749         int retval;
 750
 751         if (!(retval = put_user(current->uid, ruid)) &&
 752             !(retval = put_user(current->euid, euid)))
 753                 retval = put_user(current->suid, suid);
 754
 755         return retval;
 756 }
 757
 758 /*
 759  * Same as above, but for rgid, egid, sgid.
 760  */
 761 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 762 {
 763         int retval;
 764
 765         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
 766         if (retval)
 767                 return retval;
 768
 769         if (!capable(CAP_SETGID)) {
 770                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
 771                     (rgid != current->egid) && (rgid != current->sgid))
 772                         return -EPERM;
 773                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
 774                     (egid != current->egid) && (egid != current->sgid))
 775                         return -EPERM;
 776                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
 777                     (sgid != current->egid) && (sgid != current->sgid))
 778                         return -EPERM;
 779         }
 780         if (egid != (gid_t) -1) {
 781                 if (egid != current->egid) {
 782                         set_dumpable(current->mm, suid_dumpable);
 783                         smp_wmb();
 784                 }
 785                 current->egid = egid;
 786         }
 787         current->fsgid = current->egid;
 788         if (rgid != (gid_t) -1)
 789                 current->gid = rgid;
 790         if (sgid != (gid_t) -1)
 791                 current->sgid = sgid;
 792
 793         key_fsgid_changed(current);
 794         proc_id_connector(current, PROC_EVENT_GID);
 795         return 0;
 796 }
 797
 798 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
 799 {
 800         int retval;
 801
 802         if (!(retval = put_user(current->gid, rgid)) &&
 803             !(retval = put_user(current->egid, egid)))
 804                 retval = put_user(current->sgid, sgid);
 805
 806         return retval;
 807 }
 808
 809
 810 /*
 811  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 812  * is used for "access()" and for the NFS daemon (letting nfsd stay at
 813  * whatever uid it wants to). It normally shadows "euid", except when
 814  * explicitly set by setfsuid() or for access..
 815  */
 816 asmlinkage long sys_setfsuid(uid_t uid)
 817 {
 818         int old_fsuid;
 819
 820         old_fsuid = current->fsuid;
 821         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
 822                 return old_fsuid;
 823
 824         if (uid == current->uid || uid == current->euid ||
 825             uid == current->suid || uid == current->fsuid ||
 826             capable(CAP_SETUID)) {
 827                 if (uid != old_fsuid) {
 828                         set_dumpable(current->mm, suid_dumpable);
 829                         smp_wmb();
 830                 }
 831                 current->fsuid = uid;
 832         }
 833
 834         key_fsuid_changed(current);
 835         proc_id_connector(current, PROC_EVENT_UID);
 836
 837         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
 838
 839         return old_fsuid;
 840 }
 841
 842 /*
 843  * Samma på svenska..
 844  */
 845 asmlinkage long sys_setfsgid(gid_t gid)
 846 {
 847         int old_fsgid;
 848
 849         old_fsgid = current->fsgid;
 850         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
 851                 return old_fsgid;
 852
 853         if (gid == current->gid || gid == current->egid ||
 854             gid == current->sgid || gid == current->fsgid ||
 855             capable(CAP_SETGID)) {
 856                 if (gid != old_fsgid) {
 857                         set_dumpable(current->mm, suid_dumpable);
 858                         smp_wmb();
 859                 }
 860                 current->fsgid = gid;
 861                 key_fsgid_changed(current);
 862                 proc_id_connector(current, PROC_EVENT_GID);
 863         }
 864         return old_fsgid;
 865 }
 866
 867 asmlinkage long sys_times(struct tms __user * tbuf)
 868 {
 869         /*
 870          *      In the SMP world we might just be unlucky and have one of
 871          *      the times increment as we use it. Since the value is an
 872          *      atomically safe type this is just fine. Conceptually its
 873          *      as if the syscall took an instant longer to occur.
 874          */
 875         if (tbuf) {
 876                 struct tms tmp;
 877                 struct task_struct *tsk = current;
 878                 struct task_struct *t;
 879                 cputime_t utime, stime, cutime, cstime;
 880
 881                 spin_lock_irq(&tsk->sighand->siglock);
 882                 utime = tsk->signal->utime;
 883                 stime = tsk->signal->stime;
 884                 t = tsk;
 885                 do {
 886                         utime = cputime_add(utime, t->utime);
 887                         stime = cputime_add(stime, t->stime);
 888                         t = next_thread(t);
 889                 } while (t != tsk);
 890
 891                 cutime = tsk->signal->cutime;
 892                 cstime = tsk->signal->cstime;
 893                 spin_unlock_irq(&tsk->sighand->siglock);
 894
 895                 tmp.tms_utime = cputime_to_clock_t(utime);
 896                 tmp.tms_stime = cputime_to_clock_t(stime);
 897                 tmp.tms_cutime = cputime_to_clock_t(cutime);
 898                 tmp.tms_cstime = cputime_to_clock_t(cstime);
 899                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 900                         return -EFAULT;
 901         }
 902         return (long) jiffies_64_to_clock_t(get_jiffies_64());
 903 }
 904
 905 /*
 906  * This needs some heavy checking ...
 907  * I just haven't the stomach for it. I also don't fully
 908  * understand sessions/pgrp etc. Let somebody who does explain it.
 909  *
 910  * OK, I think I have the protection semantics right.... this is really
 911  * only important on a multi-user system anyway, to make sure one user
 912  * can't send a signal to a process owned by another.  -TYT, 12/12/91
 913  *
 914  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
 915  * LBT 04.03.94
 916  */
 917 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
 918 {
 919         struct task_struct *p;
 920         struct task_struct *group_leader = current->group_leader;
 921         int err = -EINVAL;
 922         struct pid_namespace *ns;
 923
 924         if (!pid)
 925                 pid = task_pid_vnr(group_leader);
 926         if (!pgid)
 927                 pgid = pid;
 928         if (pgid < 0)
 929                 return -EINVAL;
 930
 931         /* From this point forward we keep holding onto the tasklist lock
 932          * so that our parent does not change from under us. -DaveM
 933          */
 934         ns = current->nsproxy->pid_ns;
 935
 936         write_lock_irq(&tasklist_lock);
 937
 938         err = -ESRCH;
 939         p = find_task_by_pid_ns(pid, ns);
 940         if (!p)
 941                 goto out;
 942
 943         err = -EINVAL;
 944         if (!thread_group_leader(p))
 945                 goto out;
 946
 947         if (p->real_parent->tgid == group_leader->tgid) {
 948                 err = -EPERM;
 949                 if (task_session(p) != task_session(group_leader))
 950                         goto out;
 951                 err = -EACCES;
 952                 if (p->did_exec)
 953                         goto out;
 954         } else {
 955                 err = -ESRCH;
 956                 if (p != group_leader)
 957                         goto out;
 958         }
 959
 960         err = -EPERM;
 961         if (p->signal->leader)
 962                 goto out;
 963
 964         if (pgid != pid) {
 965                 struct task_struct *g;
 966
 967                 g = find_task_by_pid_type_ns(PIDTYPE_PGID, pgid, ns);
 968                 if (!g || task_session(g) != task_session(group_leader))
 969                         goto out;
 970         }
 971
 972         err = security_task_setpgid(p, pgid);
 973         if (err)
 974                 goto out;
 975
 976         if (task_pgrp_nr_ns(p, ns) != pgid) {
 977                 struct pid *pid;
 978
 979                 detach_pid(p, PIDTYPE_PGID);
 980                 pid = find_vpid(pgid);
 981                 attach_pid(p, PIDTYPE_PGID, pid);
 982                 p->signal->pgrp = pid_nr(pid);
 983         }
 984
 985         err = 0;
 986 out:
 987         /* All paths lead to here, thus we are safe. -DaveM */
 988         write_unlock_irq(&tasklist_lock);
 989         return err;
 990 }
 991
 992 asmlinkage long sys_getpgid(pid_t pid)
 993 {
 994         if (!pid)
 995                 return task_pgrp_vnr(current);
 996         else {
 997                 int retval;
 998                 struct task_struct *p;
 999                 struct pid_namespace *ns;
1000
1001                 ns = current->nsproxy->pid_ns;
1002
1003                 read_lock(&tasklist_lock);
1004                 p = find_task_by_pid_ns(pid, ns);
1005                 retval = -ESRCH;
1006                 if (p) {
1007                         retval = security_task_getpgid(p);
1008                         if (!retval)
1009                                 retval = task_pgrp_nr_ns(p, ns);
1010                 }
1011                 read_unlock(&tasklist_lock);
1012                 return retval;
1013         }
1014 }
1015
1016 #ifdef __ARCH_WANT_SYS_GETPGRP
1017
1018 asmlinkage long sys_getpgrp(void)
1019 {
1020         /* SMP - assuming writes are word atomic this is fine */
1021         return task_pgrp_vnr(current);
1022 }
1023
1024 #endif
1025
1026 asmlinkage long sys_getsid(pid_t pid)
1027 {
1028         if (!pid)
1029                 return task_session_vnr(current);
1030         else {
1031                 int retval;
1032                 struct task_struct *p;
1033                 struct pid_namespace *ns;
1034
1035                 ns = current->nsproxy->pid_ns;
1036
1037                 read_lock(&tasklist_lock);
1038                 p = find_task_by_pid_ns(pid, ns);
1039                 retval = -ESRCH;
1040                 if (p) {
1041                         retval = security_task_getsid(p);
1042                         if (!retval)
1043                                 retval = task_session_nr_ns(p, ns);
1044                 }
1045                 read_unlock(&tasklist_lock);
1046                 return retval;
1047         }
1048 }
1049
1050 asmlinkage long sys_setsid(void)
1051 {
1052         struct task_struct *group_leader = current->group_leader;
1053         pid_t session;
1054         int err = -EPERM;
1055
1056         write_lock_irq(&tasklist_lock);
1057
1058         /* Fail if I am already a session leader */
1059         if (group_leader->signal->leader)
1060                 goto out;
1061
1062         session = group_leader->pid;
1063         /* Fail if a process group id already exists that equals the
1064          * proposed session id.
1065          *
1066          * Don't check if session id == 1 because kernel threads use this
1067          * session id and so the check will always fail and make it so
1068          * init cannot successfully call setsid.
1069          */
1070         if (session > 1 && find_task_by_pid_type(PIDTYPE_PGID, session))
1071                 goto out;
1072
1073         group_leader->signal->leader = 1;
1074         __set_special_pids(session, session);
1075
1076         spin_lock(&group_leader->sighand->siglock);
1077         group_leader->signal->tty = NULL;
1078         spin_unlock(&group_leader->sighand->siglock);
1079
1080         err = task_pgrp_vnr(group_leader);
1081 out:
1082         write_unlock_irq(&tasklist_lock);
1083         return err;
1084 }
1085
1086 /*
1087  * Supplementary group IDs
1088  */
1089
1090 /* init to 2 - one for init_task, one to ensure it is never freed */
1091 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1092
1093 struct group_info *groups_alloc(int gidsetsize)
1094 {
1095         struct group_info *group_info;
1096         int nblocks;
1097         int i;
1098
1099         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1100         /* Make sure we always allocate at least one indirect block pointer */
1101         nblocks = nblocks ? : 1;
1102         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1103         if (!group_info)
1104                 return NULL;
1105         group_info->ngroups = gidsetsize;
1106         group_info->nblocks = nblocks;
1107         atomic_set(&group_info->usage, 1);
1108
1109         if (gidsetsize <= NGROUPS_SMALL)
1110                 group_info->blocks[0] = group_info->small_block;
1111         else {
1112                 for (i = 0; i < nblocks; i++) {
1113                         gid_t *b;
1114                         b = (void *)__get_free_page(GFP_USER);
1115                         if (!b)
1116                                 goto out_undo_partial_alloc;
1117                         group_info->blocks[i] = b;
1118                 }
1119         }
1120         return group_info;
1121
1122 out_undo_partial_alloc:
1123         while (--i >= 0) {
1124                 free_page((unsigned long)group_info->blocks[i]);
1125         }
1126         kfree(group_info);
1127         return NULL;
1128 }
1129
1130 EXPORT_SYMBOL(groups_alloc);
1131
1132 void groups_free(struct group_info *group_info)
1133 {
1134         if (group_info->blocks[0] != group_info->small_block) {
1135                 int i;
1136                 for (i = 0; i < group_info->nblocks; i++)
1137                         free_page((unsigned long)group_info->blocks[i]);
1138         }
1139         kfree(group_info);
1140 }
1141
1142 EXPORT_SYMBOL(groups_free);
1143
1144 /* export the group_info to a user-space array */
1145 static int groups_to_user(gid_t __user *grouplist,
1146     struct group_info *group_info)
1147 {
1148         int i;
1149         int count = group_info->ngroups;
1150
1151         for (i = 0; i < group_info->nblocks; i++) {
1152                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1153                 int off = i * NGROUPS_PER_BLOCK;
1154                 int len = cp_count * sizeof(*grouplist);
1155
1156                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1157                         return -EFAULT;
1158
1159                 count -= cp_count;
1160         }
1161         return 0;
1162 }
1163
1164 /* fill a group_info from a user-space array - it must be allocated already */
1165 static int groups_from_user(struct group_info *group_info,
1166     gid_t __user *grouplist)
1167 {
1168         int i;
1169         int count = group_info->ngroups;
1170
1171         for (i = 0; i < group_info->nblocks; i++) {
1172                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1173                 int off = i * NGROUPS_PER_BLOCK;
1174                 int len = cp_count * sizeof(*grouplist);
1175
1176                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1177                         return -EFAULT;
1178
1179                 count -= cp_count;
1180         }
1181         return 0;
1182 }
1183
1184 /* a simple Shell sort */
1185 static void groups_sort(struct group_info *group_info)
1186 {
1187         int base, max, stride;
1188         int gidsetsize = group_info->ngroups;
1189
1190         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1191                 ; /* nothing */
1192         stride /= 3;
1193
1194         while (stride) {
1195                 max = gidsetsize - stride;
1196                 for (base = 0; base < max; base++) {
1197                         int left = base;
1198                         int right = left + stride;
1199                         gid_t tmp = GROUP_AT(group_info, right);
1200
1201                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1202                                 GROUP_AT(group_info, right) =
1203                                     GROUP_AT(group_info, left);
1204                                 right = left;
1205                                 left -= stride;
1206                         }
1207                         GROUP_AT(group_info, right) = tmp;
1208                 }
1209                 stride /= 3;
1210         }
1211 }
1212
1213 /* a simple bsearch */
1214 int groups_search(struct group_info *group_info, gid_t grp)
1215 {
1216         unsigned int left, right;
1217
1218         if (!group_info)
1219                 return 0;
1220
1221         left = 0;
1222         right = group_info->ngroups;
1223         while (left < right) {
1224                 unsigned int mid = (left+right)/2;
1225                 int cmp = grp - GROUP_AT(group_info, mid);
1226                 if (cmp > 0)
1227                         left = mid + 1;
1228                 else if (cmp < 0)
1229                         right = mid;
1230                 else
1231                         return 1;
1232         }
1233         return 0;
1234 }
1235
1236 /* validate and set current->group_info */
1237 int set_current_groups(struct group_info *group_info)
1238 {
1239         int retval;
1240         struct group_info *old_info;
1241
1242         retval = security_task_setgroups(group_info);
1243         if (retval)
1244                 return retval;
1245
1246         groups_sort(group_info);
1247         get_group_info(group_info);
1248
1249         task_lock(current);
1250         old_info = current->group_info;
1251         current->group_info = group_info;
1252         task_unlock(current);
1253
1254         put_group_info(old_info);
1255
1256         return 0;
1257 }
1258
1259 EXPORT_SYMBOL(set_current_groups);
1260
1261 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1262 {
1263         int i = 0;
1264
1265         /*
1266          *      SMP: Nobody else can change our grouplist. Thus we are
1267          *      safe.
1268          */
1269
1270         if (gidsetsize < 0)
1271                 return -EINVAL;
1272
1273         /* no need to grab task_lock here; it cannot change */
1274         i = current->group_info->ngroups;
1275         if (gidsetsize) {
1276                 if (i > gidsetsize) {
1277                         i = -EINVAL;
1278                         goto out;
1279                 }
1280                 if (groups_to_user(grouplist, current->group_info)) {
1281                         i = -EFAULT;
1282                         goto out;
1283                 }
1284         }
1285 out:
1286         return i;
1287 }
1288
1289 /*
1290  *      SMP: Our groups are copy-on-write. We can set them safely
1291  *      without another task interfering.
1292  */
1293
1294 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1295 {
1296         struct group_info *group_info;
1297         int retval;
1298
1299         if (!capable(CAP_SETGID))
1300                 return -EPERM;
1301         if ((unsigned)gidsetsize > NGROUPS_MAX)
1302                 return -EINVAL;
1303
1304         group_info = groups_alloc(gidsetsize);
1305         if (!group_info)
1306                 return -ENOMEM;
1307         retval = groups_from_user(group_info, grouplist);
1308         if (retval) {
1309                 put_group_info(group_info);
1310                 return retval;
1311         }
1312
1313         retval = set_current_groups(group_info);
1314         put_group_info(group_info);
1315
1316         return retval;
1317 }
1318
1319 /*
1320  * Check whether we're fsgid/egid or in the supplemental group..
1321  */
1322 int in_group_p(gid_t grp)
1323 {
1324         int retval = 1;
1325         if (grp != current->fsgid)
1326                 retval = groups_search(current->group_info, grp);
1327         return retval;
1328 }
1329
1330 EXPORT_SYMBOL(in_group_p);
1331
1332 int in_egroup_p(gid_t grp)
1333 {
1334         int retval = 1;
1335         if (grp != current->egid)
1336                 retval = groups_search(current->group_info, grp);
1337         return retval;
1338 }
1339
1340 EXPORT_SYMBOL(in_egroup_p);
1341
1342 DECLARE_RWSEM(uts_sem);
1343
1344 EXPORT_SYMBOL(uts_sem);
1345
1346 asmlinkage long sys_newuname(struct new_utsname __user * name)
1347 {
1348         int errno = 0;
1349
1350         down_read(&uts_sem);
1351         if (copy_to_user(name, utsname(), sizeof *name))
1352                 errno = -EFAULT;
1353         up_read(&uts_sem);
1354         return errno;
1355 }
1356
1357 asmlinkage long sys_sethostname(char __user *name, int len)
1358 {
1359         int errno;
1360         char tmp[__NEW_UTS_LEN];
1361
1362         if (!capable(CAP_SYS_ADMIN))
1363                 return -EPERM;
1364         if (len < 0 || len > __NEW_UTS_LEN)
1365                 return -EINVAL;
1366         down_write(&uts_sem);
1367         errno = -EFAULT;
1368         if (!copy_from_user(tmp, name, len)) {
1369                 memcpy(utsname()->nodename, tmp, len);
1370                 utsname()->nodename[len] = 0;
1371                 errno = 0;
1372         }
1373         up_write(&uts_sem);
1374         return errno;
1375 }
1376
1377 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1378
1379 asmlinkage long sys_gethostname(char __user *name, int len)
1380 {
1381         int i, errno;
1382
1383         if (len < 0)
1384                 return -EINVAL;
1385         down_read(&uts_sem);
1386         i = 1 + strlen(utsname()->nodename);
1387         if (i > len)
1388                 i = len;
1389         errno = 0;
1390         if (copy_to_user(name, utsname()->nodename, i))
1391                 errno = -EFAULT;
1392         up_read(&uts_sem);
1393         return errno;
1394 }
1395
1396 #endif
1397
1398 /*
1399  * Only setdomainname; getdomainname can be implemented by calling
1400  * uname()
1401  */
1402 asmlinkage long sys_setdomainname(char __user *name, int len)
1403 {
1404         int errno;
1405         char tmp[__NEW_UTS_LEN];
1406
1407         if (!capable(CAP_SYS_ADMIN))
1408                 return -EPERM;
1409         if (len < 0 || len > __NEW_UTS_LEN)
1410                 return -EINVAL;
1411
1412         down_write(&uts_sem);
1413         errno = -EFAULT;
1414         if (!copy_from_user(tmp, name, len)) {
1415                 memcpy(utsname()->domainname, tmp, len);
1416                 utsname()->domainname[len] = 0;
1417                 errno = 0;
1418         }
1419         up_write(&uts_sem);
1420         return errno;
1421 }
1422
1423 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1424 {
1425         if (resource >= RLIM_NLIMITS)
1426                 return -EINVAL;
1427         else {
1428                 struct rlimit value;
1429                 task_lock(current->group_leader);
1430                 value = current->signal->rlim[resource];
1431                 task_unlock(current->group_leader);
1432                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1433         }
1434 }
1435
1436 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1437
1438 /*
1439  *      Back compatibility for getrlimit. Needed for some apps.
1440  */
1441
1442 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1443 {
1444         struct rlimit x;
1445         if (resource >= RLIM_NLIMITS)
1446                 return -EINVAL;
1447
1448         task_lock(current->group_leader);
1449         x = current->signal->rlim[resource];
1450         task_unlock(current->group_leader);
1451         if (x.rlim_cur > 0x7FFFFFFF)
1452                 x.rlim_cur = 0x7FFFFFFF;
1453         if (x.rlim_max > 0x7FFFFFFF)
1454                 x.rlim_max = 0x7FFFFFFF;
1455         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1456 }
1457
1458 #endif
1459
1460 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1461 {
1462         struct rlimit new_rlim, *old_rlim;
1463         unsigned long it_prof_secs;
1464         int retval;
1465
1466         if (resource >= RLIM_NLIMITS)
1467                 return -EINVAL;
1468         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1469                 return -EFAULT;
1470         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1471                 return -EINVAL;
1472         old_rlim = current->signal->rlim + resource;
1473         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1474             !capable(CAP_SYS_RESOURCE))
1475                 return -EPERM;
1476         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1477                 return -EPERM;
1478
1479         retval = security_task_setrlimit(resource, &new_rlim);
1480         if (retval)
1481                 return retval;
1482
1483         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1484                 /*
1485                  * The caller is asking for an immediate RLIMIT_CPU
1486                  * expiry.  But we use the zero value to mean "it was
1487                  * never set".  So let's cheat and make it one second
1488                  * instead
1489                  */
1490                 new_rlim.rlim_cur = 1;
1491         }
1492
1493         task_lock(current->group_leader);
1494         *old_rlim = new_rlim;
1495         task_unlock(current->group_leader);
1496
1497         if (resource != RLIMIT_CPU)
1498                 goto out;
1499
1500         /*
1501          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1502          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1503          * very long-standing error, and fixing it now risks breakage of
1504          * applications, so we live with it
1505          */
1506         if (new_rlim.rlim_cur == RLIM_INFINITY)
1507                 goto out;
1508
1509         it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1510         if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1511                 unsigned long rlim_cur = new_rlim.rlim_cur;
1512                 cputime_t cputime;
1513
1514                 cputime = secs_to_cputime(rlim_cur);
1515                 read_lock(&tasklist_lock);
1516                 spin_lock_irq(&current->sighand->siglock);
1517                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1518                 spin_unlock_irq(&current->sighand->siglock);
1519                 read_unlock(&tasklist_lock);
1520         }
1521 out:
1522         return 0;
1523 }
1524
1525 /*
1526  * It would make sense to put struct rusage in the task_struct,
1527  * except that would make the task_struct be *really big*.  After
1528  * task_struct gets moved into malloc'ed memory, it would
1529  * make sense to do this.  It will make moving the rest of the information
1530  * a lot simpler!  (Which we're not doing right now because we're not
1531  * measuring them yet).
1532  *
1533  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1534  * races with threads incrementing their own counters.  But since word
1535  * reads are atomic, we either get new values or old values and we don't
1536  * care which for the sums.  We always take the siglock to protect reading
1537  * the c* fields from p->signal from races with exit.c updating those
1538  * fields when reaping, so a sample either gets all the additions of a
1539  * given child after it's reaped, or none so this sample is before reaping.
1540  *
1541  * Locking:
1542  * We need to take the siglock for CHILDEREN, SELF and BOTH
1543  * for  the cases current multithreaded, non-current single threaded
1544  * non-current multithreaded.  Thread traversal is now safe with
1545  * the siglock held.
1546  * Strictly speaking, we donot need to take the siglock if we are current and
1547  * single threaded,  as no one else can take our signal_struct away, no one
1548  * else can  reap the  children to update signal->c* counters, and no one else
1549  * can race with the signal-> fields. If we do not take any lock, the
1550  * signal-> fields could be read out of order while another thread was just
1551  * exiting. So we should  place a read memory barrier when we avoid the lock.
1552  * On the writer side,  write memory barrier is implied in  __exit_signal
1553  * as __exit_signal releases  the siglock spinlock after updating the signal->
1554  * fields. But we don't do this yet to keep things simple.
1555  *
1556  */
1557
1558 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1559 {
1560         struct task_struct *t;
1561         unsigned long flags;
1562         cputime_t utime, stime;
1563
1564         memset((char *) r, 0, sizeof *r);
1565         utime = stime = cputime_zero;
1566
1567         rcu_read_lock();
1568         if (!lock_task_sighand(p, &flags)) {
1569                 rcu_read_unlock();
1570                 return;
1571         }
1572
1573         switch (who) {
1574                 case RUSAGE_BOTH:
1575                 case RUSAGE_CHILDREN:
1576                         utime = p->signal->cutime;
1577                         stime = p->signal->cstime;
1578                         r->ru_nvcsw = p->signal->cnvcsw;
1579                         r->ru_nivcsw = p->signal->cnivcsw;
1580                         r->ru_minflt = p->signal->cmin_flt;
1581                         r->ru_majflt = p->signal->cmaj_flt;
1582                         r->ru_inblock = p->signal->cinblock;
1583                         r->ru_oublock = p->signal->coublock;
1584
1585                         if (who == RUSAGE_CHILDREN)
1586                                 break;
1587
1588                 case RUSAGE_SELF:
1589                         utime = cputime_add(utime, p->signal->utime);
1590                         stime = cputime_add(stime, p->signal->stime);
1591                         r->ru_nvcsw += p->signal->nvcsw;
1592                         r->ru_nivcsw += p->signal->nivcsw;
1593                         r->ru_minflt += p->signal->min_flt;
1594                         r->ru_majflt += p->signal->maj_flt;
1595                         r->ru_inblock += p->signal->inblock;
1596                         r->ru_oublock += p->signal->oublock;
1597                         t = p;
1598                         do {
1599                                 utime = cputime_add(utime, t->utime);
1600                                 stime = cputime_add(stime, t->stime);
1601                                 r->ru_nvcsw += t->nvcsw;
1602                                 r->ru_nivcsw += t->nivcsw;
1603                                 r->ru_minflt += t->min_flt;
1604                                 r->ru_majflt += t->maj_flt;
1605                                 r->ru_inblock += task_io_get_inblock(t);
1606                                 r->ru_oublock += task_io_get_oublock(t);
1607                                 t = next_thread(t);
1608                         } while (t != p);
1609                         break;
1610
1611                 default:
1612                         BUG();
1613         }
1614
1615         unlock_task_sighand(p, &flags);
1616         rcu_read_unlock();
1617
1618         cputime_to_timeval(utime, &r->ru_utime);
1619         cputime_to_timeval(stime, &r->ru_stime);
1620 }
1621
1622 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1623 {
1624         struct rusage r;
1625         k_getrusage(p, who, &r);
1626         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1627 }
1628
1629 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1630 {
1631         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1632                 return -EINVAL;
1633         return getrusage(current, who, ru);
1634 }
1635
1636 asmlinkage long sys_umask(int mask)
1637 {
1638         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1639         return mask;
1640 }
1641
1642 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1643                           unsigned long arg4, unsigned long arg5)
1644 {
1645         long error;
1646
1647         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1648         if (error)
1649                 return error;
1650
1651         switch (option) {
1652                 case PR_SET_PDEATHSIG:
1653                         if (!valid_signal(arg2)) {
1654                                 error = -EINVAL;
1655                                 break;
1656                         }
1657                         current->pdeath_signal = arg2;
1658                         break;
1659                 case PR_GET_PDEATHSIG:
1660                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1661                         break;
1662                 case PR_GET_DUMPABLE:
1663                         error = get_dumpable(current->mm);
1664                         break;
1665                 case PR_SET_DUMPABLE:
1666                         if (arg2 < 0 || arg2 > 1) {
1667                                 error = -EINVAL;
1668                                 break;
1669                         }
1670                         set_dumpable(current->mm, arg2);
1671                         break;
1672
1673                 case PR_SET_UNALIGN:
1674                         error = SET_UNALIGN_CTL(current, arg2);
1675                         break;
1676                 case PR_GET_UNALIGN:
1677                         error = GET_UNALIGN_CTL(current, arg2);
1678                         break;
1679                 case PR_SET_FPEMU:
1680                         error = SET_FPEMU_CTL(current, arg2);
1681                         break;
1682                 case PR_GET_FPEMU:
1683                         error = GET_FPEMU_CTL(current, arg2);
1684                         break;
1685                 case PR_SET_FPEXC:
1686                         error = SET_FPEXC_CTL(current, arg2);
1687                         break;
1688                 case PR_GET_FPEXC:
1689                         error = GET_FPEXC_CTL(current, arg2);
1690                         break;
1691                 case PR_GET_TIMING:
1692                         error = PR_TIMING_STATISTICAL;
1693                         break;
1694                 case PR_SET_TIMING:
1695                         if (arg2 == PR_TIMING_STATISTICAL)
1696                                 error = 0;
1697                         else
1698                                 error = -EINVAL;
1699                         break;
1700
1701                 case PR_GET_KEEPCAPS:
1702                         if (current->keep_capabilities)
1703                                 error = 1;
1704                         break;
1705                 case PR_SET_KEEPCAPS:
1706                         if (arg2 != 0 && arg2 != 1) {
1707                                 error = -EINVAL;
1708                                 break;
1709                         }
1710                         current->keep_capabilities = arg2;
1711                         break;
1712                 case PR_SET_NAME: {
1713                         struct task_struct *me = current;
1714                         unsigned char ncomm[sizeof(me->comm)];
1715
1716                         ncomm[sizeof(me->comm)-1] = 0;
1717                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1718                                                 sizeof(me->comm)-1) < 0)
1719                                 return -EFAULT;
1720                         set_task_comm(me, ncomm);
1721                         return 0;
1722                 }
1723                 case PR_GET_NAME: {
1724                         struct task_struct *me = current;
1725                         unsigned char tcomm[sizeof(me->comm)];
1726
1727                         get_task_comm(tcomm, me);
1728                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1729                                 return -EFAULT;
1730                         return 0;
1731                 }
1732                 case PR_GET_ENDIAN:
1733                         error = GET_ENDIAN(current, arg2);
1734                         break;
1735                 case PR_SET_ENDIAN:
1736                         error = SET_ENDIAN(current, 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
1746                 default:
1747                         error = -EINVAL;
1748                         break;
1749         }
1750         return error;
1751 }
1752
1753 asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1754                            struct getcpu_cache __user *cache)
1755 {
1756         int err = 0;
1757         int cpu = raw_smp_processor_id();
1758         if (cpup)
1759                 err |= put_user(cpu, cpup);
1760         if (nodep)
1761                 err |= put_user(cpu_to_node(cpu), nodep);
1762         if (cache) {
1763                 /*
1764                  * The cache is not needed for this implementation,
1765                  * but make sure user programs pass something
1766                  * valid. vsyscall implementations can instead make
1767                  * good use of the cache. Only use t0 and t1 because
1768                  * these are available in both 32bit and 64bit ABI (no
1769                  * need for a compat_getcpu). 32bit has enough
1770                  * padding
1771                  */
1772                 unsigned long t0, t1;
1773                 get_user(t0, &cache->blob[0]);
1774                 get_user(t1, &cache->blob[1]);
1775                 t0++;
1776                 t1++;
1777                 put_user(t0, &cache->blob[0]);
1778                 put_user(t1, &cache->blob[1]);
1779         }
1780         return err ? -EFAULT : 0;
1781 }
1782
1783 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1784
1785 static void argv_cleanup(char **argv, char **envp)
1786 {
1787         argv_free(argv);
1788 }
1789
1790 /**
1791  * orderly_poweroff - Trigger an orderly system poweroff
1792  * @force: force poweroff if command execution fails
1793  *
1794  * This may be called from any context to trigger a system shutdown.
1795  * If the orderly shutdown fails, it will force an immediate shutdown.
1796  */
1797 int orderly_poweroff(bool force)
1798 {
1799         int argc;
1800         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1801         static char *envp[] = {
1802                 "HOME=/",
1803                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1804                 NULL
1805         };
1806         int ret = -ENOMEM;
1807         struct subprocess_info *info;
1808
1809         if (argv == NULL) {
1810                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1811                        __func__, poweroff_cmd);
1812                 goto out;
1813         }
1814
1815         info = call_usermodehelper_setup(argv[0], argv, envp);
1816         if (info == NULL) {
1817                 argv_free(argv);
1818                 goto out;
1819         }
1820
1821         call_usermodehelper_setcleanup(info, argv_cleanup);
1822
1823         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1824
1825   out:
1826         if (ret && force) {
1827                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1828                        "forcing the issue\n");
1829
1830                 /* I guess this should try to kick off some daemon to
1831                    sync and poweroff asap.  Or not even bother syncing
1832                    if we're doing an emergency shutdown? */
1833                 emergency_sync();
1834                 kernel_power_off();
1835         }
1836
1837         return ret;
1838 }
1839 EXPORT_SYMBOL_GPL(orderly_poweroff);