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