1 /* 2 * linux/kernel/sys.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 #include <linux/export.h> 8 #include <linux/mm.h> 9 #include <linux/utsname.h> 10 #include <linux/mman.h> 11 #include <linux/reboot.h> 12 #include <linux/prctl.h> 13 #include <linux/highuid.h> 14 #include <linux/fs.h> 15 #include <linux/kmod.h> 16 #include <linux/perf_event.h> 17 #include <linux/resource.h> 18 #include <linux/kernel.h> 19 #include <linux/kexec.h> 20 #include <linux/workqueue.h> 21 #include <linux/capability.h> 22 #include <linux/device.h> 23 #include <linux/key.h> 24 #include <linux/times.h> 25 #include <linux/posix-timers.h> 26 #include <linux/security.h> 27 #include <linux/dcookies.h> 28 #include <linux/suspend.h> 29 #include <linux/tty.h> 30 #include <linux/signal.h> 31 #include <linux/cn_proc.h> 32 #include <linux/getcpu.h> 33 #include <linux/task_io_accounting_ops.h> 34 #include <linux/seccomp.h> 35 #include <linux/cpu.h> 36 #include <linux/personality.h> 37 #include <linux/ptrace.h> 38 #include <linux/fs_struct.h> 39 #include <linux/file.h> 40 #include <linux/mount.h> 41 #include <linux/gfp.h> 42 #include <linux/syscore_ops.h> 43 #include <linux/version.h> 44 #include <linux/ctype.h> 45 46 #include <linux/compat.h> 47 #include <linux/syscalls.h> 48 #include <linux/kprobes.h> 49 #include <linux/user_namespace.h> 50 #include <linux/binfmts.h> 51 52 #include <linux/kmsg_dump.h> 53 /* Move somewhere else to avoid recompiling? */ 54 #include <generated/utsrelease.h> 55 56 #include <asm/uaccess.h> 57 #include <asm/io.h> 58 #include <asm/unistd.h> 59 60 #ifndef SET_UNALIGN_CTL 61 # define SET_UNALIGN_CTL(a,b) (-EINVAL) 62 #endif 63 #ifndef GET_UNALIGN_CTL 64 # define GET_UNALIGN_CTL(a,b) (-EINVAL) 65 #endif 66 #ifndef SET_FPEMU_CTL 67 # define SET_FPEMU_CTL(a,b) (-EINVAL) 68 #endif 69 #ifndef GET_FPEMU_CTL 70 # define GET_FPEMU_CTL(a,b) (-EINVAL) 71 #endif 72 #ifndef SET_FPEXC_CTL 73 # define SET_FPEXC_CTL(a,b) (-EINVAL) 74 #endif 75 #ifndef GET_FPEXC_CTL 76 # define GET_FPEXC_CTL(a,b) (-EINVAL) 77 #endif 78 #ifndef GET_ENDIAN 79 # define GET_ENDIAN(a,b) (-EINVAL) 80 #endif 81 #ifndef SET_ENDIAN 82 # define SET_ENDIAN(a,b) (-EINVAL) 83 #endif 84 #ifndef GET_TSC_CTL 85 # define GET_TSC_CTL(a) (-EINVAL) 86 #endif 87 #ifndef SET_TSC_CTL 88 # define SET_TSC_CTL(a) (-EINVAL) 89 #endif 90 91 /* 92 * this is where the system-wide overflow UID and GID are defined, for 93 * architectures that now have 32-bit UID/GID but didn't in the past 94 */ 95 96 int overflowuid = DEFAULT_OVERFLOWUID; 97 int overflowgid = DEFAULT_OVERFLOWGID; 98 99 EXPORT_SYMBOL(overflowuid); 100 EXPORT_SYMBOL(overflowgid); 101 102 /* 103 * the same as above, but for filesystems which can only store a 16-bit 104 * UID and GID. as such, this is needed on all architectures 105 */ 106 107 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID; 108 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID; 109 110 EXPORT_SYMBOL(fs_overflowuid); 111 EXPORT_SYMBOL(fs_overflowgid); 112 113 /* 114 * this indicates whether you can reboot with ctrl-alt-del: the default is yes 115 */ 116 117 int C_A_D = 1; 118 struct pid *cad_pid; 119 EXPORT_SYMBOL(cad_pid); 120 121 /* 122 * If set, this is used for preparing the system to power off. 123 */ 124 125 void (*pm_power_off_prepare)(void); 126 127 /* 128 * Returns true if current's euid is same as p's uid or euid, 129 * or has CAP_SYS_NICE to p's user_ns. 130 * 131 * Called with rcu_read_lock, creds are safe 132 */ 133 static bool set_one_prio_perm(struct task_struct *p) 134 { 135 const struct cred *cred = current_cred(), *pcred = __task_cred(p); 136 137 if (uid_eq(pcred->uid, cred->euid) || 138 uid_eq(pcred->euid, cred->euid)) 139 return true; 140 if (ns_capable(pcred->user_ns, CAP_SYS_NICE)) 141 return true; 142 return false; 143 } 144 145 /* 146 * set the priority of a task 147 * - the caller must hold the RCU read lock 148 */ 149 static int set_one_prio(struct task_struct *p, int niceval, int error) 150 { 151 int no_nice; 152 153 if (!set_one_prio_perm(p)) { 154 error = -EPERM; 155 goto out; 156 } 157 if (niceval < task_nice(p) && !can_nice(p, niceval)) { 158 error = -EACCES; 159 goto out; 160 } 161 no_nice = security_task_setnice(p, niceval); 162 if (no_nice) { 163 error = no_nice; 164 goto out; 165 } 166 if (error == -ESRCH) 167 error = 0; 168 set_user_nice(p, niceval); 169 out: 170 return error; 171 } 172 173 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval) 174 { 175 struct task_struct *g, *p; 176 struct user_struct *user; 177 const struct cred *cred = current_cred(); 178 int error = -EINVAL; 179 struct pid *pgrp; 180 kuid_t uid; 181 182 if (which > PRIO_USER || which < PRIO_PROCESS) 183 goto out; 184 185 /* normalize: avoid signed division (rounding problems) */ 186 error = -ESRCH; 187 if (niceval < -20) 188 niceval = -20; 189 if (niceval > 19) 190 niceval = 19; 191 192 rcu_read_lock(); 193 read_lock(&tasklist_lock); 194 switch (which) { 195 case PRIO_PROCESS: 196 if (who) 197 p = find_task_by_vpid(who); 198 else 199 p = current; 200 if (p) 201 error = set_one_prio(p, niceval, error); 202 break; 203 case PRIO_PGRP: 204 if (who) 205 pgrp = find_vpid(who); 206 else 207 pgrp = task_pgrp(current); 208 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 209 error = set_one_prio(p, niceval, error); 210 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 211 break; 212 case PRIO_USER: 213 uid = make_kuid(cred->user_ns, who); 214 user = cred->user; 215 if (!who) 216 uid = cred->uid; 217 else if (!uid_eq(uid, cred->uid) && 218 !(user = find_user(uid))) 219 goto out_unlock; /* No processes for this user */ 220 221 do_each_thread(g, p) { 222 if (uid_eq(task_uid(p), uid)) 223 error = set_one_prio(p, niceval, error); 224 } while_each_thread(g, p); 225 if (!uid_eq(uid, cred->uid)) 226 free_uid(user); /* For find_user() */ 227 break; 228 } 229 out_unlock: 230 read_unlock(&tasklist_lock); 231 rcu_read_unlock(); 232 out: 233 return error; 234 } 235 236 /* 237 * Ugh. To avoid negative return values, "getpriority()" will 238 * not return the normal nice-value, but a negated value that 239 * has been offset by 20 (ie it returns 40..1 instead of -20..19) 240 * to stay compatible. 241 */ 242 SYSCALL_DEFINE2(getpriority, int, which, int, who) 243 { 244 struct task_struct *g, *p; 245 struct user_struct *user; 246 const struct cred *cred = current_cred(); 247 long niceval, retval = -ESRCH; 248 struct pid *pgrp; 249 kuid_t uid; 250 251 if (which > PRIO_USER || which < PRIO_PROCESS) 252 return -EINVAL; 253 254 rcu_read_lock(); 255 read_lock(&tasklist_lock); 256 switch (which) { 257 case PRIO_PROCESS: 258 if (who) 259 p = find_task_by_vpid(who); 260 else 261 p = current; 262 if (p) { 263 niceval = 20 - task_nice(p); 264 if (niceval > retval) 265 retval = niceval; 266 } 267 break; 268 case PRIO_PGRP: 269 if (who) 270 pgrp = find_vpid(who); 271 else 272 pgrp = task_pgrp(current); 273 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 274 niceval = 20 - task_nice(p); 275 if (niceval > retval) 276 retval = niceval; 277 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 278 break; 279 case PRIO_USER: 280 uid = make_kuid(cred->user_ns, who); 281 user = cred->user; 282 if (!who) 283 uid = cred->uid; 284 else if (!uid_eq(uid, cred->uid) && 285 !(user = find_user(uid))) 286 goto out_unlock; /* No processes for this user */ 287 288 do_each_thread(g, p) { 289 if (uid_eq(task_uid(p), uid)) { 290 niceval = 20 - task_nice(p); 291 if (niceval > retval) 292 retval = niceval; 293 } 294 } while_each_thread(g, p); 295 if (!uid_eq(uid, cred->uid)) 296 free_uid(user); /* for find_user() */ 297 break; 298 } 299 out_unlock: 300 read_unlock(&tasklist_lock); 301 rcu_read_unlock(); 302 303 return retval; 304 } 305 306 /** 307 * emergency_restart - reboot the system 308 * 309 * Without shutting down any hardware or taking any locks 310 * reboot the system. This is called when we know we are in 311 * trouble so this is our best effort to reboot. This is 312 * safe to call in interrupt context. 313 */ 314 void emergency_restart(void) 315 { 316 kmsg_dump(KMSG_DUMP_EMERG); 317 machine_emergency_restart(); 318 } 319 EXPORT_SYMBOL_GPL(emergency_restart); 320 321 void kernel_restart_prepare(char *cmd) 322 { 323 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd); 324 system_state = SYSTEM_RESTART; 325 usermodehelper_disable(); 326 device_shutdown(); 327 syscore_shutdown(); 328 } 329 330 /** 331 * register_reboot_notifier - Register function to be called at reboot time 332 * @nb: Info about notifier function to be called 333 * 334 * Registers a function with the list of functions 335 * to be called at reboot time. 336 * 337 * Currently always returns zero, as blocking_notifier_chain_register() 338 * always returns zero. 339 */ 340 int register_reboot_notifier(struct notifier_block *nb) 341 { 342 return blocking_notifier_chain_register(&reboot_notifier_list, nb); 343 } 344 EXPORT_SYMBOL(register_reboot_notifier); 345 346 /** 347 * unregister_reboot_notifier - Unregister previously registered reboot notifier 348 * @nb: Hook to be unregistered 349 * 350 * Unregisters a previously registered reboot 351 * notifier function. 352 * 353 * Returns zero on success, or %-ENOENT on failure. 354 */ 355 int unregister_reboot_notifier(struct notifier_block *nb) 356 { 357 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb); 358 } 359 EXPORT_SYMBOL(unregister_reboot_notifier); 360 361 /** 362 * kernel_restart - reboot the system 363 * @cmd: pointer to buffer containing command to execute for restart 364 * or %NULL 365 * 366 * Shutdown everything and perform a clean reboot. 367 * This is not safe to call in interrupt context. 368 */ 369 void kernel_restart(char *cmd) 370 { 371 kernel_restart_prepare(cmd); 372 disable_nonboot_cpus(); 373 if (!cmd) 374 printk(KERN_EMERG "Restarting system.\n"); 375 else 376 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd); 377 kmsg_dump(KMSG_DUMP_RESTART); 378 machine_restart(cmd); 379 } 380 EXPORT_SYMBOL_GPL(kernel_restart); 381 382 static void kernel_shutdown_prepare(enum system_states state) 383 { 384 blocking_notifier_call_chain(&reboot_notifier_list, 385 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL); 386 system_state = state; 387 usermodehelper_disable(); 388 device_shutdown(); 389 } 390 /** 391 * kernel_halt - halt the system 392 * 393 * Shutdown everything and perform a clean system halt. 394 */ 395 void kernel_halt(void) 396 { 397 kernel_shutdown_prepare(SYSTEM_HALT); 398 syscore_shutdown(); 399 printk(KERN_EMERG "System halted.\n"); 400 kmsg_dump(KMSG_DUMP_HALT); 401 machine_halt(); 402 } 403 404 EXPORT_SYMBOL_GPL(kernel_halt); 405 406 /** 407 * kernel_power_off - power_off the system 408 * 409 * Shutdown everything and perform a clean system power_off. 410 */ 411 void kernel_power_off(void) 412 { 413 kernel_shutdown_prepare(SYSTEM_POWER_OFF); 414 if (pm_power_off_prepare) 415 pm_power_off_prepare(); 416 disable_nonboot_cpus(); 417 syscore_shutdown(); 418 printk(KERN_EMERG "Power down.\n"); 419 kmsg_dump(KMSG_DUMP_POWEROFF); 420 machine_power_off(); 421 } 422 EXPORT_SYMBOL_GPL(kernel_power_off); 423 424 static DEFINE_MUTEX(reboot_mutex); 425 426 /* 427 * Reboot system call: for obvious reasons only root may call it, 428 * and even root needs to set up some magic numbers in the registers 429 * so that some mistake won't make this reboot the whole machine. 430 * You can also set the meaning of the ctrl-alt-del-key here. 431 * 432 * reboot doesn't sync: do that yourself before calling this. 433 */ 434 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd, 435 void __user *, arg) 436 { 437 char buffer[256]; 438 int ret = 0; 439 440 /* We only trust the superuser with rebooting the system. */ 441 if (!capable(CAP_SYS_BOOT)) 442 return -EPERM; 443 444 /* For safety, we require "magic" arguments. */ 445 if (magic1 != LINUX_REBOOT_MAGIC1 || 446 (magic2 != LINUX_REBOOT_MAGIC2 && 447 magic2 != LINUX_REBOOT_MAGIC2A && 448 magic2 != LINUX_REBOOT_MAGIC2B && 449 magic2 != LINUX_REBOOT_MAGIC2C)) 450 return -EINVAL; 451 452 /* 453 * If pid namespaces are enabled and the current task is in a child 454 * pid_namespace, the command is handled by reboot_pid_ns() which will 455 * call do_exit(). 456 */ 457 ret = reboot_pid_ns(task_active_pid_ns(current), cmd); 458 if (ret) 459 return ret; 460 461 /* Instead of trying to make the power_off code look like 462 * halt when pm_power_off is not set do it the easy way. 463 */ 464 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off) 465 cmd = LINUX_REBOOT_CMD_HALT; 466 467 mutex_lock(&reboot_mutex); 468 switch (cmd) { 469 case LINUX_REBOOT_CMD_RESTART: 470 kernel_restart(NULL); 471 break; 472 473 case LINUX_REBOOT_CMD_CAD_ON: 474 C_A_D = 1; 475 break; 476 477 case LINUX_REBOOT_CMD_CAD_OFF: 478 C_A_D = 0; 479 break; 480 481 case LINUX_REBOOT_CMD_HALT: 482 kernel_halt(); 483 do_exit(0); 484 panic("cannot halt"); 485 486 case LINUX_REBOOT_CMD_POWER_OFF: 487 kernel_power_off(); 488 do_exit(0); 489 break; 490 491 case LINUX_REBOOT_CMD_RESTART2: 492 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) { 493 ret = -EFAULT; 494 break; 495 } 496 buffer[sizeof(buffer) - 1] = '\0'; 497 498 kernel_restart(buffer); 499 break; 500 501 #ifdef CONFIG_KEXEC 502 case LINUX_REBOOT_CMD_KEXEC: 503 ret = kernel_kexec(); 504 break; 505 #endif 506 507 #ifdef CONFIG_HIBERNATION 508 case LINUX_REBOOT_CMD_SW_SUSPEND: 509 ret = hibernate(); 510 break; 511 #endif 512 513 default: 514 ret = -EINVAL; 515 break; 516 } 517 mutex_unlock(&reboot_mutex); 518 return ret; 519 } 520 521 static void deferred_cad(struct work_struct *dummy) 522 { 523 kernel_restart(NULL); 524 } 525 526 /* 527 * This function gets called by ctrl-alt-del - ie the keyboard interrupt. 528 * As it's called within an interrupt, it may NOT sync: the only choice 529 * is whether to reboot at once, or just ignore the ctrl-alt-del. 530 */ 531 void ctrl_alt_del(void) 532 { 533 static DECLARE_WORK(cad_work, deferred_cad); 534 535 if (C_A_D) 536 schedule_work(&cad_work); 537 else 538 kill_cad_pid(SIGINT, 1); 539 } 540 541 /* 542 * Unprivileged users may change the real gid to the effective gid 543 * or vice versa. (BSD-style) 544 * 545 * If you set the real gid at all, or set the effective gid to a value not 546 * equal to the real gid, then the saved gid is set to the new effective gid. 547 * 548 * This makes it possible for a setgid program to completely drop its 549 * privileges, which is often a useful assertion to make when you are doing 550 * a security audit over a program. 551 * 552 * The general idea is that a program which uses just setregid() will be 553 * 100% compatible with BSD. A program which uses just setgid() will be 554 * 100% compatible with POSIX with saved IDs. 555 * 556 * SMP: There are not races, the GIDs are checked only by filesystem 557 * operations (as far as semantic preservation is concerned). 558 */ 559 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid) 560 { 561 struct user_namespace *ns = current_user_ns(); 562 const struct cred *old; 563 struct cred *new; 564 int retval; 565 kgid_t krgid, kegid; 566 567 krgid = make_kgid(ns, rgid); 568 kegid = make_kgid(ns, egid); 569 570 if ((rgid != (gid_t) -1) && !gid_valid(krgid)) 571 return -EINVAL; 572 if ((egid != (gid_t) -1) && !gid_valid(kegid)) 573 return -EINVAL; 574 575 new = prepare_creds(); 576 if (!new) 577 return -ENOMEM; 578 old = current_cred(); 579 580 retval = -EPERM; 581 if (rgid != (gid_t) -1) { 582 if (gid_eq(old->gid, krgid) || 583 gid_eq(old->egid, krgid) || 584 nsown_capable(CAP_SETGID)) 585 new->gid = krgid; 586 else 587 goto error; 588 } 589 if (egid != (gid_t) -1) { 590 if (gid_eq(old->gid, kegid) || 591 gid_eq(old->egid, kegid) || 592 gid_eq(old->sgid, kegid) || 593 nsown_capable(CAP_SETGID)) 594 new->egid = kegid; 595 else 596 goto error; 597 } 598 599 if (rgid != (gid_t) -1 || 600 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid))) 601 new->sgid = new->egid; 602 new->fsgid = new->egid; 603 604 return commit_creds(new); 605 606 error: 607 abort_creds(new); 608 return retval; 609 } 610 611 /* 612 * setgid() is implemented like SysV w/ SAVED_IDS 613 * 614 * SMP: Same implicit races as above. 615 */ 616 SYSCALL_DEFINE1(setgid, gid_t, gid) 617 { 618 struct user_namespace *ns = current_user_ns(); 619 const struct cred *old; 620 struct cred *new; 621 int retval; 622 kgid_t kgid; 623 624 kgid = make_kgid(ns, gid); 625 if (!gid_valid(kgid)) 626 return -EINVAL; 627 628 new = prepare_creds(); 629 if (!new) 630 return -ENOMEM; 631 old = current_cred(); 632 633 retval = -EPERM; 634 if (nsown_capable(CAP_SETGID)) 635 new->gid = new->egid = new->sgid = new->fsgid = kgid; 636 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid)) 637 new->egid = new->fsgid = kgid; 638 else 639 goto error; 640 641 return commit_creds(new); 642 643 error: 644 abort_creds(new); 645 return retval; 646 } 647 648 /* 649 * change the user struct in a credentials set to match the new UID 650 */ 651 static int set_user(struct cred *new) 652 { 653 struct user_struct *new_user; 654 655 new_user = alloc_uid(new->uid); 656 if (!new_user) 657 return -EAGAIN; 658 659 /* 660 * We don't fail in case of NPROC limit excess here because too many 661 * poorly written programs don't check set*uid() return code, assuming 662 * it never fails if called by root. We may still enforce NPROC limit 663 * for programs doing set*uid()+execve() by harmlessly deferring the 664 * failure to the execve() stage. 665 */ 666 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) && 667 new_user != INIT_USER) 668 current->flags |= PF_NPROC_EXCEEDED; 669 else 670 current->flags &= ~PF_NPROC_EXCEEDED; 671 672 free_uid(new->user); 673 new->user = new_user; 674 return 0; 675 } 676 677 /* 678 * Unprivileged users may change the real uid to the effective uid 679 * or vice versa. (BSD-style) 680 * 681 * If you set the real uid at all, or set the effective uid to a value not 682 * equal to the real uid, then the saved uid is set to the new effective uid. 683 * 684 * This makes it possible for a setuid program to completely drop its 685 * privileges, which is often a useful assertion to make when you are doing 686 * a security audit over a program. 687 * 688 * The general idea is that a program which uses just setreuid() will be 689 * 100% compatible with BSD. A program which uses just setuid() will be 690 * 100% compatible with POSIX with saved IDs. 691 */ 692 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid) 693 { 694 struct user_namespace *ns = current_user_ns(); 695 const struct cred *old; 696 struct cred *new; 697 int retval; 698 kuid_t kruid, keuid; 699 700 kruid = make_kuid(ns, ruid); 701 keuid = make_kuid(ns, euid); 702 703 if ((ruid != (uid_t) -1) && !uid_valid(kruid)) 704 return -EINVAL; 705 if ((euid != (uid_t) -1) && !uid_valid(keuid)) 706 return -EINVAL; 707 708 new = prepare_creds(); 709 if (!new) 710 return -ENOMEM; 711 old = current_cred(); 712 713 retval = -EPERM; 714 if (ruid != (uid_t) -1) { 715 new->uid = kruid; 716 if (!uid_eq(old->uid, kruid) && 717 !uid_eq(old->euid, kruid) && 718 !nsown_capable(CAP_SETUID)) 719 goto error; 720 } 721 722 if (euid != (uid_t) -1) { 723 new->euid = keuid; 724 if (!uid_eq(old->uid, keuid) && 725 !uid_eq(old->euid, keuid) && 726 !uid_eq(old->suid, keuid) && 727 !nsown_capable(CAP_SETUID)) 728 goto error; 729 } 730 731 if (!uid_eq(new->uid, old->uid)) { 732 retval = set_user(new); 733 if (retval < 0) 734 goto error; 735 } 736 if (ruid != (uid_t) -1 || 737 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid))) 738 new->suid = new->euid; 739 new->fsuid = new->euid; 740 741 retval = security_task_fix_setuid(new, old, LSM_SETID_RE); 742 if (retval < 0) 743 goto error; 744 745 return commit_creds(new); 746 747 error: 748 abort_creds(new); 749 return retval; 750 } 751 752 /* 753 * setuid() is implemented like SysV with SAVED_IDS 754 * 755 * Note that SAVED_ID's is deficient in that a setuid root program 756 * like sendmail, for example, cannot set its uid to be a normal 757 * user and then switch back, because if you're root, setuid() sets 758 * the saved uid too. If you don't like this, blame the bright people 759 * in the POSIX committee and/or USG. Note that the BSD-style setreuid() 760 * will allow a root program to temporarily drop privileges and be able to 761 * regain them by swapping the real and effective uid. 762 */ 763 SYSCALL_DEFINE1(setuid, uid_t, uid) 764 { 765 struct user_namespace *ns = current_user_ns(); 766 const struct cred *old; 767 struct cred *new; 768 int retval; 769 kuid_t kuid; 770 771 kuid = make_kuid(ns, uid); 772 if (!uid_valid(kuid)) 773 return -EINVAL; 774 775 new = prepare_creds(); 776 if (!new) 777 return -ENOMEM; 778 old = current_cred(); 779 780 retval = -EPERM; 781 if (nsown_capable(CAP_SETUID)) { 782 new->suid = new->uid = kuid; 783 if (!uid_eq(kuid, old->uid)) { 784 retval = set_user(new); 785 if (retval < 0) 786 goto error; 787 } 788 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) { 789 goto error; 790 } 791 792 new->fsuid = new->euid = kuid; 793 794 retval = security_task_fix_setuid(new, old, LSM_SETID_ID); 795 if (retval < 0) 796 goto error; 797 798 return commit_creds(new); 799 800 error: 801 abort_creds(new); 802 return retval; 803 } 804 805 806 /* 807 * This function implements a generic ability to update ruid, euid, 808 * and suid. This allows you to implement the 4.4 compatible seteuid(). 809 */ 810 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid) 811 { 812 struct user_namespace *ns = current_user_ns(); 813 const struct cred *old; 814 struct cred *new; 815 int retval; 816 kuid_t kruid, keuid, ksuid; 817 818 kruid = make_kuid(ns, ruid); 819 keuid = make_kuid(ns, euid); 820 ksuid = make_kuid(ns, suid); 821 822 if ((ruid != (uid_t) -1) && !uid_valid(kruid)) 823 return -EINVAL; 824 825 if ((euid != (uid_t) -1) && !uid_valid(keuid)) 826 return -EINVAL; 827 828 if ((suid != (uid_t) -1) && !uid_valid(ksuid)) 829 return -EINVAL; 830 831 new = prepare_creds(); 832 if (!new) 833 return -ENOMEM; 834 835 old = current_cred(); 836 837 retval = -EPERM; 838 if (!nsown_capable(CAP_SETUID)) { 839 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) && 840 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid)) 841 goto error; 842 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) && 843 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid)) 844 goto error; 845 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) && 846 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid)) 847 goto error; 848 } 849 850 if (ruid != (uid_t) -1) { 851 new->uid = kruid; 852 if (!uid_eq(kruid, old->uid)) { 853 retval = set_user(new); 854 if (retval < 0) 855 goto error; 856 } 857 } 858 if (euid != (uid_t) -1) 859 new->euid = keuid; 860 if (suid != (uid_t) -1) 861 new->suid = ksuid; 862 new->fsuid = new->euid; 863 864 retval = security_task_fix_setuid(new, old, LSM_SETID_RES); 865 if (retval < 0) 866 goto error; 867 868 return commit_creds(new); 869 870 error: 871 abort_creds(new); 872 return retval; 873 } 874 875 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp) 876 { 877 const struct cred *cred = current_cred(); 878 int retval; 879 uid_t ruid, euid, suid; 880 881 ruid = from_kuid_munged(cred->user_ns, cred->uid); 882 euid = from_kuid_munged(cred->user_ns, cred->euid); 883 suid = from_kuid_munged(cred->user_ns, cred->suid); 884 885 if (!(retval = put_user(ruid, ruidp)) && 886 !(retval = put_user(euid, euidp))) 887 retval = put_user(suid, suidp); 888 889 return retval; 890 } 891 892 /* 893 * Same as above, but for rgid, egid, sgid. 894 */ 895 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid) 896 { 897 struct user_namespace *ns = current_user_ns(); 898 const struct cred *old; 899 struct cred *new; 900 int retval; 901 kgid_t krgid, kegid, ksgid; 902 903 krgid = make_kgid(ns, rgid); 904 kegid = make_kgid(ns, egid); 905 ksgid = make_kgid(ns, sgid); 906 907 if ((rgid != (gid_t) -1) && !gid_valid(krgid)) 908 return -EINVAL; 909 if ((egid != (gid_t) -1) && !gid_valid(kegid)) 910 return -EINVAL; 911 if ((sgid != (gid_t) -1) && !gid_valid(ksgid)) 912 return -EINVAL; 913 914 new = prepare_creds(); 915 if (!new) 916 return -ENOMEM; 917 old = current_cred(); 918 919 retval = -EPERM; 920 if (!nsown_capable(CAP_SETGID)) { 921 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) && 922 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid)) 923 goto error; 924 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) && 925 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid)) 926 goto error; 927 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) && 928 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid)) 929 goto error; 930 } 931 932 if (rgid != (gid_t) -1) 933 new->gid = krgid; 934 if (egid != (gid_t) -1) 935 new->egid = kegid; 936 if (sgid != (gid_t) -1) 937 new->sgid = ksgid; 938 new->fsgid = new->egid; 939 940 return commit_creds(new); 941 942 error: 943 abort_creds(new); 944 return retval; 945 } 946 947 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp) 948 { 949 const struct cred *cred = current_cred(); 950 int retval; 951 gid_t rgid, egid, sgid; 952 953 rgid = from_kgid_munged(cred->user_ns, cred->gid); 954 egid = from_kgid_munged(cred->user_ns, cred->egid); 955 sgid = from_kgid_munged(cred->user_ns, cred->sgid); 956 957 if (!(retval = put_user(rgid, rgidp)) && 958 !(retval = put_user(egid, egidp))) 959 retval = put_user(sgid, sgidp); 960 961 return retval; 962 } 963 964 965 /* 966 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This 967 * is used for "access()" and for the NFS daemon (letting nfsd stay at 968 * whatever uid it wants to). It normally shadows "euid", except when 969 * explicitly set by setfsuid() or for access.. 970 */ 971 SYSCALL_DEFINE1(setfsuid, uid_t, uid) 972 { 973 const struct cred *old; 974 struct cred *new; 975 uid_t old_fsuid; 976 kuid_t kuid; 977 978 old = current_cred(); 979 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid); 980 981 kuid = make_kuid(old->user_ns, uid); 982 if (!uid_valid(kuid)) 983 return old_fsuid; 984 985 new = prepare_creds(); 986 if (!new) 987 return old_fsuid; 988 989 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) || 990 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) || 991 nsown_capable(CAP_SETUID)) { 992 if (!uid_eq(kuid, old->fsuid)) { 993 new->fsuid = kuid; 994 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0) 995 goto change_okay; 996 } 997 } 998 999 abort_creds(new); 1000 return old_fsuid; 1001 1002 change_okay: 1003 commit_creds(new); 1004 return old_fsuid; 1005 } 1006 1007 /* 1008 * Samma på svenska.. 1009 */ 1010 SYSCALL_DEFINE1(setfsgid, gid_t, gid) 1011 { 1012 const struct cred *old; 1013 struct cred *new; 1014 gid_t old_fsgid; 1015 kgid_t kgid; 1016 1017 old = current_cred(); 1018 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid); 1019 1020 kgid = make_kgid(old->user_ns, gid); 1021 if (!gid_valid(kgid)) 1022 return old_fsgid; 1023 1024 new = prepare_creds(); 1025 if (!new) 1026 return old_fsgid; 1027 1028 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) || 1029 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) || 1030 nsown_capable(CAP_SETGID)) { 1031 if (!gid_eq(kgid, old->fsgid)) { 1032 new->fsgid = kgid; 1033 goto change_okay; 1034 } 1035 } 1036 1037 abort_creds(new); 1038 return old_fsgid; 1039 1040 change_okay: 1041 commit_creds(new); 1042 return old_fsgid; 1043 } 1044 1045 void do_sys_times(struct tms *tms) 1046 { 1047 cputime_t tgutime, tgstime, cutime, cstime; 1048 1049 spin_lock_irq(¤t->sighand->siglock); 1050 thread_group_cputime_adjusted(current, &tgutime, &tgstime); 1051 cutime = current->signal->cutime; 1052 cstime = current->signal->cstime; 1053 spin_unlock_irq(¤t->sighand->siglock); 1054 tms->tms_utime = cputime_to_clock_t(tgutime); 1055 tms->tms_stime = cputime_to_clock_t(tgstime); 1056 tms->tms_cutime = cputime_to_clock_t(cutime); 1057 tms->tms_cstime = cputime_to_clock_t(cstime); 1058 } 1059 1060 SYSCALL_DEFINE1(times, struct tms __user *, tbuf) 1061 { 1062 if (tbuf) { 1063 struct tms tmp; 1064 1065 do_sys_times(&tmp); 1066 if (copy_to_user(tbuf, &tmp, sizeof(struct tms))) 1067 return -EFAULT; 1068 } 1069 force_successful_syscall_return(); 1070 return (long) jiffies_64_to_clock_t(get_jiffies_64()); 1071 } 1072 1073 /* 1074 * This needs some heavy checking ... 1075 * I just haven't the stomach for it. I also don't fully 1076 * understand sessions/pgrp etc. Let somebody who does explain it. 1077 * 1078 * OK, I think I have the protection semantics right.... this is really 1079 * only important on a multi-user system anyway, to make sure one user 1080 * can't send a signal to a process owned by another. -TYT, 12/12/91 1081 * 1082 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX. 1083 * LBT 04.03.94 1084 */ 1085 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid) 1086 { 1087 struct task_struct *p; 1088 struct task_struct *group_leader = current->group_leader; 1089 struct pid *pgrp; 1090 int err; 1091 1092 if (!pid) 1093 pid = task_pid_vnr(group_leader); 1094 if (!pgid) 1095 pgid = pid; 1096 if (pgid < 0) 1097 return -EINVAL; 1098 rcu_read_lock(); 1099 1100 /* From this point forward we keep holding onto the tasklist lock 1101 * so that our parent does not change from under us. -DaveM 1102 */ 1103 write_lock_irq(&tasklist_lock); 1104 1105 err = -ESRCH; 1106 p = find_task_by_vpid(pid); 1107 if (!p) 1108 goto out; 1109 1110 err = -EINVAL; 1111 if (!thread_group_leader(p)) 1112 goto out; 1113 1114 if (same_thread_group(p->real_parent, group_leader)) { 1115 err = -EPERM; 1116 if (task_session(p) != task_session(group_leader)) 1117 goto out; 1118 err = -EACCES; 1119 if (p->did_exec) 1120 goto out; 1121 } else { 1122 err = -ESRCH; 1123 if (p != group_leader) 1124 goto out; 1125 } 1126 1127 err = -EPERM; 1128 if (p->signal->leader) 1129 goto out; 1130 1131 pgrp = task_pid(p); 1132 if (pgid != pid) { 1133 struct task_struct *g; 1134 1135 pgrp = find_vpid(pgid); 1136 g = pid_task(pgrp, PIDTYPE_PGID); 1137 if (!g || task_session(g) != task_session(group_leader)) 1138 goto out; 1139 } 1140 1141 err = security_task_setpgid(p, pgid); 1142 if (err) 1143 goto out; 1144 1145 if (task_pgrp(p) != pgrp) 1146 change_pid(p, PIDTYPE_PGID, pgrp); 1147 1148 err = 0; 1149 out: 1150 /* All paths lead to here, thus we are safe. -DaveM */ 1151 write_unlock_irq(&tasklist_lock); 1152 rcu_read_unlock(); 1153 return err; 1154 } 1155 1156 SYSCALL_DEFINE1(getpgid, pid_t, pid) 1157 { 1158 struct task_struct *p; 1159 struct pid *grp; 1160 int retval; 1161 1162 rcu_read_lock(); 1163 if (!pid) 1164 grp = task_pgrp(current); 1165 else { 1166 retval = -ESRCH; 1167 p = find_task_by_vpid(pid); 1168 if (!p) 1169 goto out; 1170 grp = task_pgrp(p); 1171 if (!grp) 1172 goto out; 1173 1174 retval = security_task_getpgid(p); 1175 if (retval) 1176 goto out; 1177 } 1178 retval = pid_vnr(grp); 1179 out: 1180 rcu_read_unlock(); 1181 return retval; 1182 } 1183 1184 #ifdef __ARCH_WANT_SYS_GETPGRP 1185 1186 SYSCALL_DEFINE0(getpgrp) 1187 { 1188 return sys_getpgid(0); 1189 } 1190 1191 #endif 1192 1193 SYSCALL_DEFINE1(getsid, pid_t, pid) 1194 { 1195 struct task_struct *p; 1196 struct pid *sid; 1197 int retval; 1198 1199 rcu_read_lock(); 1200 if (!pid) 1201 sid = task_session(current); 1202 else { 1203 retval = -ESRCH; 1204 p = find_task_by_vpid(pid); 1205 if (!p) 1206 goto out; 1207 sid = task_session(p); 1208 if (!sid) 1209 goto out; 1210 1211 retval = security_task_getsid(p); 1212 if (retval) 1213 goto out; 1214 } 1215 retval = pid_vnr(sid); 1216 out: 1217 rcu_read_unlock(); 1218 return retval; 1219 } 1220 1221 SYSCALL_DEFINE0(setsid) 1222 { 1223 struct task_struct *group_leader = current->group_leader; 1224 struct pid *sid = task_pid(group_leader); 1225 pid_t session = pid_vnr(sid); 1226 int err = -EPERM; 1227 1228 write_lock_irq(&tasklist_lock); 1229 /* Fail if I am already a session leader */ 1230 if (group_leader->signal->leader) 1231 goto out; 1232 1233 /* Fail if a process group id already exists that equals the 1234 * proposed session id. 1235 */ 1236 if (pid_task(sid, PIDTYPE_PGID)) 1237 goto out; 1238 1239 group_leader->signal->leader = 1; 1240 __set_special_pids(sid); 1241 1242 proc_clear_tty(group_leader); 1243 1244 err = session; 1245 out: 1246 write_unlock_irq(&tasklist_lock); 1247 if (err > 0) { 1248 proc_sid_connector(group_leader); 1249 sched_autogroup_create_attach(group_leader); 1250 } 1251 return err; 1252 } 1253 1254 DECLARE_RWSEM(uts_sem); 1255 1256 #ifdef COMPAT_UTS_MACHINE 1257 #define override_architecture(name) \ 1258 (personality(current->personality) == PER_LINUX32 && \ 1259 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \ 1260 sizeof(COMPAT_UTS_MACHINE))) 1261 #else 1262 #define override_architecture(name) 0 1263 #endif 1264 1265 /* 1266 * Work around broken programs that cannot handle "Linux 3.0". 1267 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40 1268 */ 1269 static int override_release(char __user *release, size_t len) 1270 { 1271 int ret = 0; 1272 1273 if (current->personality & UNAME26) { 1274 const char *rest = UTS_RELEASE; 1275 char buf[65] = { 0 }; 1276 int ndots = 0; 1277 unsigned v; 1278 size_t copy; 1279 1280 while (*rest) { 1281 if (*rest == '.' && ++ndots >= 3) 1282 break; 1283 if (!isdigit(*rest) && *rest != '.') 1284 break; 1285 rest++; 1286 } 1287 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40; 1288 copy = clamp_t(size_t, len, 1, sizeof(buf)); 1289 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest); 1290 ret = copy_to_user(release, buf, copy + 1); 1291 } 1292 return ret; 1293 } 1294 1295 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name) 1296 { 1297 int errno = 0; 1298 1299 down_read(&uts_sem); 1300 if (copy_to_user(name, utsname(), sizeof *name)) 1301 errno = -EFAULT; 1302 up_read(&uts_sem); 1303 1304 if (!errno && override_release(name->release, sizeof(name->release))) 1305 errno = -EFAULT; 1306 if (!errno && override_architecture(name)) 1307 errno = -EFAULT; 1308 return errno; 1309 } 1310 1311 #ifdef __ARCH_WANT_SYS_OLD_UNAME 1312 /* 1313 * Old cruft 1314 */ 1315 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name) 1316 { 1317 int error = 0; 1318 1319 if (!name) 1320 return -EFAULT; 1321 1322 down_read(&uts_sem); 1323 if (copy_to_user(name, utsname(), sizeof(*name))) 1324 error = -EFAULT; 1325 up_read(&uts_sem); 1326 1327 if (!error && override_release(name->release, sizeof(name->release))) 1328 error = -EFAULT; 1329 if (!error && override_architecture(name)) 1330 error = -EFAULT; 1331 return error; 1332 } 1333 1334 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name) 1335 { 1336 int error; 1337 1338 if (!name) 1339 return -EFAULT; 1340 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname))) 1341 return -EFAULT; 1342 1343 down_read(&uts_sem); 1344 error = __copy_to_user(&name->sysname, &utsname()->sysname, 1345 __OLD_UTS_LEN); 1346 error |= __put_user(0, name->sysname + __OLD_UTS_LEN); 1347 error |= __copy_to_user(&name->nodename, &utsname()->nodename, 1348 __OLD_UTS_LEN); 1349 error |= __put_user(0, name->nodename + __OLD_UTS_LEN); 1350 error |= __copy_to_user(&name->release, &utsname()->release, 1351 __OLD_UTS_LEN); 1352 error |= __put_user(0, name->release + __OLD_UTS_LEN); 1353 error |= __copy_to_user(&name->version, &utsname()->version, 1354 __OLD_UTS_LEN); 1355 error |= __put_user(0, name->version + __OLD_UTS_LEN); 1356 error |= __copy_to_user(&name->machine, &utsname()->machine, 1357 __OLD_UTS_LEN); 1358 error |= __put_user(0, name->machine + __OLD_UTS_LEN); 1359 up_read(&uts_sem); 1360 1361 if (!error && override_architecture(name)) 1362 error = -EFAULT; 1363 if (!error && override_release(name->release, sizeof(name->release))) 1364 error = -EFAULT; 1365 return error ? -EFAULT : 0; 1366 } 1367 #endif 1368 1369 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len) 1370 { 1371 int errno; 1372 char tmp[__NEW_UTS_LEN]; 1373 1374 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN)) 1375 return -EPERM; 1376 1377 if (len < 0 || len > __NEW_UTS_LEN) 1378 return -EINVAL; 1379 down_write(&uts_sem); 1380 errno = -EFAULT; 1381 if (!copy_from_user(tmp, name, len)) { 1382 struct new_utsname *u = utsname(); 1383 1384 memcpy(u->nodename, tmp, len); 1385 memset(u->nodename + len, 0, sizeof(u->nodename) - len); 1386 errno = 0; 1387 uts_proc_notify(UTS_PROC_HOSTNAME); 1388 } 1389 up_write(&uts_sem); 1390 return errno; 1391 } 1392 1393 #ifdef __ARCH_WANT_SYS_GETHOSTNAME 1394 1395 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len) 1396 { 1397 int i, errno; 1398 struct new_utsname *u; 1399 1400 if (len < 0) 1401 return -EINVAL; 1402 down_read(&uts_sem); 1403 u = utsname(); 1404 i = 1 + strlen(u->nodename); 1405 if (i > len) 1406 i = len; 1407 errno = 0; 1408 if (copy_to_user(name, u->nodename, i)) 1409 errno = -EFAULT; 1410 up_read(&uts_sem); 1411 return errno; 1412 } 1413 1414 #endif 1415 1416 /* 1417 * Only setdomainname; getdomainname can be implemented by calling 1418 * uname() 1419 */ 1420 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len) 1421 { 1422 int errno; 1423 char tmp[__NEW_UTS_LEN]; 1424 1425 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN)) 1426 return -EPERM; 1427 if (len < 0 || len > __NEW_UTS_LEN) 1428 return -EINVAL; 1429 1430 down_write(&uts_sem); 1431 errno = -EFAULT; 1432 if (!copy_from_user(tmp, name, len)) { 1433 struct new_utsname *u = utsname(); 1434 1435 memcpy(u->domainname, tmp, len); 1436 memset(u->domainname + len, 0, sizeof(u->domainname) - len); 1437 errno = 0; 1438 uts_proc_notify(UTS_PROC_DOMAINNAME); 1439 } 1440 up_write(&uts_sem); 1441 return errno; 1442 } 1443 1444 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim) 1445 { 1446 struct rlimit value; 1447 int ret; 1448 1449 ret = do_prlimit(current, resource, NULL, &value); 1450 if (!ret) 1451 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0; 1452 1453 return ret; 1454 } 1455 1456 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT 1457 1458 /* 1459 * Back compatibility for getrlimit. Needed for some apps. 1460 */ 1461 1462 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource, 1463 struct rlimit __user *, rlim) 1464 { 1465 struct rlimit x; 1466 if (resource >= RLIM_NLIMITS) 1467 return -EINVAL; 1468 1469 task_lock(current->group_leader); 1470 x = current->signal->rlim[resource]; 1471 task_unlock(current->group_leader); 1472 if (x.rlim_cur > 0x7FFFFFFF) 1473 x.rlim_cur = 0x7FFFFFFF; 1474 if (x.rlim_max > 0x7FFFFFFF) 1475 x.rlim_max = 0x7FFFFFFF; 1476 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0; 1477 } 1478 1479 #endif 1480 1481 static inline bool rlim64_is_infinity(__u64 rlim64) 1482 { 1483 #if BITS_PER_LONG < 64 1484 return rlim64 >= ULONG_MAX; 1485 #else 1486 return rlim64 == RLIM64_INFINITY; 1487 #endif 1488 } 1489 1490 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64) 1491 { 1492 if (rlim->rlim_cur == RLIM_INFINITY) 1493 rlim64->rlim_cur = RLIM64_INFINITY; 1494 else 1495 rlim64->rlim_cur = rlim->rlim_cur; 1496 if (rlim->rlim_max == RLIM_INFINITY) 1497 rlim64->rlim_max = RLIM64_INFINITY; 1498 else 1499 rlim64->rlim_max = rlim->rlim_max; 1500 } 1501 1502 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim) 1503 { 1504 if (rlim64_is_infinity(rlim64->rlim_cur)) 1505 rlim->rlim_cur = RLIM_INFINITY; 1506 else 1507 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur; 1508 if (rlim64_is_infinity(rlim64->rlim_max)) 1509 rlim->rlim_max = RLIM_INFINITY; 1510 else 1511 rlim->rlim_max = (unsigned long)rlim64->rlim_max; 1512 } 1513 1514 /* make sure you are allowed to change @tsk limits before calling this */ 1515 int do_prlimit(struct task_struct *tsk, unsigned int resource, 1516 struct rlimit *new_rlim, struct rlimit *old_rlim) 1517 { 1518 struct rlimit *rlim; 1519 int retval = 0; 1520 1521 if (resource >= RLIM_NLIMITS) 1522 return -EINVAL; 1523 if (new_rlim) { 1524 if (new_rlim->rlim_cur > new_rlim->rlim_max) 1525 return -EINVAL; 1526 if (resource == RLIMIT_NOFILE && 1527 new_rlim->rlim_max > sysctl_nr_open) 1528 return -EPERM; 1529 } 1530 1531 /* protect tsk->signal and tsk->sighand from disappearing */ 1532 read_lock(&tasklist_lock); 1533 if (!tsk->sighand) { 1534 retval = -ESRCH; 1535 goto out; 1536 } 1537 1538 rlim = tsk->signal->rlim + resource; 1539 task_lock(tsk->group_leader); 1540 if (new_rlim) { 1541 /* Keep the capable check against init_user_ns until 1542 cgroups can contain all limits */ 1543 if (new_rlim->rlim_max > rlim->rlim_max && 1544 !capable(CAP_SYS_RESOURCE)) 1545 retval = -EPERM; 1546 if (!retval) 1547 retval = security_task_setrlimit(tsk->group_leader, 1548 resource, new_rlim); 1549 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) { 1550 /* 1551 * The caller is asking for an immediate RLIMIT_CPU 1552 * expiry. But we use the zero value to mean "it was 1553 * never set". So let's cheat and make it one second 1554 * instead 1555 */ 1556 new_rlim->rlim_cur = 1; 1557 } 1558 } 1559 if (!retval) { 1560 if (old_rlim) 1561 *old_rlim = *rlim; 1562 if (new_rlim) 1563 *rlim = *new_rlim; 1564 } 1565 task_unlock(tsk->group_leader); 1566 1567 /* 1568 * RLIMIT_CPU handling. Note that the kernel fails to return an error 1569 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a 1570 * very long-standing error, and fixing it now risks breakage of 1571 * applications, so we live with it 1572 */ 1573 if (!retval && new_rlim && resource == RLIMIT_CPU && 1574 new_rlim->rlim_cur != RLIM_INFINITY) 1575 update_rlimit_cpu(tsk, new_rlim->rlim_cur); 1576 out: 1577 read_unlock(&tasklist_lock); 1578 return retval; 1579 } 1580 1581 /* rcu lock must be held */ 1582 static int check_prlimit_permission(struct task_struct *task) 1583 { 1584 const struct cred *cred = current_cred(), *tcred; 1585 1586 if (current == task) 1587 return 0; 1588 1589 tcred = __task_cred(task); 1590 if (uid_eq(cred->uid, tcred->euid) && 1591 uid_eq(cred->uid, tcred->suid) && 1592 uid_eq(cred->uid, tcred->uid) && 1593 gid_eq(cred->gid, tcred->egid) && 1594 gid_eq(cred->gid, tcred->sgid) && 1595 gid_eq(cred->gid, tcred->gid)) 1596 return 0; 1597 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE)) 1598 return 0; 1599 1600 return -EPERM; 1601 } 1602 1603 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource, 1604 const struct rlimit64 __user *, new_rlim, 1605 struct rlimit64 __user *, old_rlim) 1606 { 1607 struct rlimit64 old64, new64; 1608 struct rlimit old, new; 1609 struct task_struct *tsk; 1610 int ret; 1611 1612 if (new_rlim) { 1613 if (copy_from_user(&new64, new_rlim, sizeof(new64))) 1614 return -EFAULT; 1615 rlim64_to_rlim(&new64, &new); 1616 } 1617 1618 rcu_read_lock(); 1619 tsk = pid ? find_task_by_vpid(pid) : current; 1620 if (!tsk) { 1621 rcu_read_unlock(); 1622 return -ESRCH; 1623 } 1624 ret = check_prlimit_permission(tsk); 1625 if (ret) { 1626 rcu_read_unlock(); 1627 return ret; 1628 } 1629 get_task_struct(tsk); 1630 rcu_read_unlock(); 1631 1632 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL, 1633 old_rlim ? &old : NULL); 1634 1635 if (!ret && old_rlim) { 1636 rlim_to_rlim64(&old, &old64); 1637 if (copy_to_user(old_rlim, &old64, sizeof(old64))) 1638 ret = -EFAULT; 1639 } 1640 1641 put_task_struct(tsk); 1642 return ret; 1643 } 1644 1645 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim) 1646 { 1647 struct rlimit new_rlim; 1648 1649 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim))) 1650 return -EFAULT; 1651 return do_prlimit(current, resource, &new_rlim, NULL); 1652 } 1653 1654 /* 1655 * It would make sense to put struct rusage in the task_struct, 1656 * except that would make the task_struct be *really big*. After 1657 * task_struct gets moved into malloc'ed memory, it would 1658 * make sense to do this. It will make moving the rest of the information 1659 * a lot simpler! (Which we're not doing right now because we're not 1660 * measuring them yet). 1661 * 1662 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have 1663 * races with threads incrementing their own counters. But since word 1664 * reads are atomic, we either get new values or old values and we don't 1665 * care which for the sums. We always take the siglock to protect reading 1666 * the c* fields from p->signal from races with exit.c updating those 1667 * fields when reaping, so a sample either gets all the additions of a 1668 * given child after it's reaped, or none so this sample is before reaping. 1669 * 1670 * Locking: 1671 * We need to take the siglock for CHILDEREN, SELF and BOTH 1672 * for the cases current multithreaded, non-current single threaded 1673 * non-current multithreaded. Thread traversal is now safe with 1674 * the siglock held. 1675 * Strictly speaking, we donot need to take the siglock if we are current and 1676 * single threaded, as no one else can take our signal_struct away, no one 1677 * else can reap the children to update signal->c* counters, and no one else 1678 * can race with the signal-> fields. If we do not take any lock, the 1679 * signal-> fields could be read out of order while another thread was just 1680 * exiting. So we should place a read memory barrier when we avoid the lock. 1681 * On the writer side, write memory barrier is implied in __exit_signal 1682 * as __exit_signal releases the siglock spinlock after updating the signal-> 1683 * fields. But we don't do this yet to keep things simple. 1684 * 1685 */ 1686 1687 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r) 1688 { 1689 r->ru_nvcsw += t->nvcsw; 1690 r->ru_nivcsw += t->nivcsw; 1691 r->ru_minflt += t->min_flt; 1692 r->ru_majflt += t->maj_flt; 1693 r->ru_inblock += task_io_get_inblock(t); 1694 r->ru_oublock += task_io_get_oublock(t); 1695 } 1696 1697 static void k_getrusage(struct task_struct *p, int who, struct rusage *r) 1698 { 1699 struct task_struct *t; 1700 unsigned long flags; 1701 cputime_t tgutime, tgstime, utime, stime; 1702 unsigned long maxrss = 0; 1703 1704 memset((char *) r, 0, sizeof *r); 1705 utime = stime = 0; 1706 1707 if (who == RUSAGE_THREAD) { 1708 task_cputime_adjusted(current, &utime, &stime); 1709 accumulate_thread_rusage(p, r); 1710 maxrss = p->signal->maxrss; 1711 goto out; 1712 } 1713 1714 if (!lock_task_sighand(p, &flags)) 1715 return; 1716 1717 switch (who) { 1718 case RUSAGE_BOTH: 1719 case RUSAGE_CHILDREN: 1720 utime = p->signal->cutime; 1721 stime = p->signal->cstime; 1722 r->ru_nvcsw = p->signal->cnvcsw; 1723 r->ru_nivcsw = p->signal->cnivcsw; 1724 r->ru_minflt = p->signal->cmin_flt; 1725 r->ru_majflt = p->signal->cmaj_flt; 1726 r->ru_inblock = p->signal->cinblock; 1727 r->ru_oublock = p->signal->coublock; 1728 maxrss = p->signal->cmaxrss; 1729 1730 if (who == RUSAGE_CHILDREN) 1731 break; 1732 1733 case RUSAGE_SELF: 1734 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1735 utime += tgutime; 1736 stime += tgstime; 1737 r->ru_nvcsw += p->signal->nvcsw; 1738 r->ru_nivcsw += p->signal->nivcsw; 1739 r->ru_minflt += p->signal->min_flt; 1740 r->ru_majflt += p->signal->maj_flt; 1741 r->ru_inblock += p->signal->inblock; 1742 r->ru_oublock += p->signal->oublock; 1743 if (maxrss < p->signal->maxrss) 1744 maxrss = p->signal->maxrss; 1745 t = p; 1746 do { 1747 accumulate_thread_rusage(t, r); 1748 t = next_thread(t); 1749 } while (t != p); 1750 break; 1751 1752 default: 1753 BUG(); 1754 } 1755 unlock_task_sighand(p, &flags); 1756 1757 out: 1758 cputime_to_timeval(utime, &r->ru_utime); 1759 cputime_to_timeval(stime, &r->ru_stime); 1760 1761 if (who != RUSAGE_CHILDREN) { 1762 struct mm_struct *mm = get_task_mm(p); 1763 if (mm) { 1764 setmax_mm_hiwater_rss(&maxrss, mm); 1765 mmput(mm); 1766 } 1767 } 1768 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */ 1769 } 1770 1771 int getrusage(struct task_struct *p, int who, struct rusage __user *ru) 1772 { 1773 struct rusage r; 1774 k_getrusage(p, who, &r); 1775 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0; 1776 } 1777 1778 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru) 1779 { 1780 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN && 1781 who != RUSAGE_THREAD) 1782 return -EINVAL; 1783 return getrusage(current, who, ru); 1784 } 1785 1786 SYSCALL_DEFINE1(umask, int, mask) 1787 { 1788 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO); 1789 return mask; 1790 } 1791 1792 #ifdef CONFIG_CHECKPOINT_RESTORE 1793 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd) 1794 { 1795 struct fd exe; 1796 struct dentry *dentry; 1797 int err; 1798 1799 exe = fdget(fd); 1800 if (!exe.file) 1801 return -EBADF; 1802 1803 dentry = exe.file->f_path.dentry; 1804 1805 /* 1806 * Because the original mm->exe_file points to executable file, make 1807 * sure that this one is executable as well, to avoid breaking an 1808 * overall picture. 1809 */ 1810 err = -EACCES; 1811 if (!S_ISREG(dentry->d_inode->i_mode) || 1812 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC) 1813 goto exit; 1814 1815 err = inode_permission(dentry->d_inode, MAY_EXEC); 1816 if (err) 1817 goto exit; 1818 1819 down_write(&mm->mmap_sem); 1820 1821 /* 1822 * Forbid mm->exe_file change if old file still mapped. 1823 */ 1824 err = -EBUSY; 1825 if (mm->exe_file) { 1826 struct vm_area_struct *vma; 1827 1828 for (vma = mm->mmap; vma; vma = vma->vm_next) 1829 if (vma->vm_file && 1830 path_equal(&vma->vm_file->f_path, 1831 &mm->exe_file->f_path)) 1832 goto exit_unlock; 1833 } 1834 1835 /* 1836 * The symlink can be changed only once, just to disallow arbitrary 1837 * transitions malicious software might bring in. This means one 1838 * could make a snapshot over all processes running and monitor 1839 * /proc/pid/exe changes to notice unusual activity if needed. 1840 */ 1841 err = -EPERM; 1842 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags)) 1843 goto exit_unlock; 1844 1845 err = 0; 1846 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */ 1847 exit_unlock: 1848 up_write(&mm->mmap_sem); 1849 1850 exit: 1851 fdput(exe); 1852 return err; 1853 } 1854 1855 static int prctl_set_mm(int opt, unsigned long addr, 1856 unsigned long arg4, unsigned long arg5) 1857 { 1858 unsigned long rlim = rlimit(RLIMIT_DATA); 1859 struct mm_struct *mm = current->mm; 1860 struct vm_area_struct *vma; 1861 int error; 1862 1863 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV)) 1864 return -EINVAL; 1865 1866 if (!capable(CAP_SYS_RESOURCE)) 1867 return -EPERM; 1868 1869 if (opt == PR_SET_MM_EXE_FILE) 1870 return prctl_set_mm_exe_file(mm, (unsigned int)addr); 1871 1872 if (addr >= TASK_SIZE || addr < mmap_min_addr) 1873 return -EINVAL; 1874 1875 error = -EINVAL; 1876 1877 down_read(&mm->mmap_sem); 1878 vma = find_vma(mm, addr); 1879 1880 switch (opt) { 1881 case PR_SET_MM_START_CODE: 1882 mm->start_code = addr; 1883 break; 1884 case PR_SET_MM_END_CODE: 1885 mm->end_code = addr; 1886 break; 1887 case PR_SET_MM_START_DATA: 1888 mm->start_data = addr; 1889 break; 1890 case PR_SET_MM_END_DATA: 1891 mm->end_data = addr; 1892 break; 1893 1894 case PR_SET_MM_START_BRK: 1895 if (addr <= mm->end_data) 1896 goto out; 1897 1898 if (rlim < RLIM_INFINITY && 1899 (mm->brk - addr) + 1900 (mm->end_data - mm->start_data) > rlim) 1901 goto out; 1902 1903 mm->start_brk = addr; 1904 break; 1905 1906 case PR_SET_MM_BRK: 1907 if (addr <= mm->end_data) 1908 goto out; 1909 1910 if (rlim < RLIM_INFINITY && 1911 (addr - mm->start_brk) + 1912 (mm->end_data - mm->start_data) > rlim) 1913 goto out; 1914 1915 mm->brk = addr; 1916 break; 1917 1918 /* 1919 * If command line arguments and environment 1920 * are placed somewhere else on stack, we can 1921 * set them up here, ARG_START/END to setup 1922 * command line argumets and ENV_START/END 1923 * for environment. 1924 */ 1925 case PR_SET_MM_START_STACK: 1926 case PR_SET_MM_ARG_START: 1927 case PR_SET_MM_ARG_END: 1928 case PR_SET_MM_ENV_START: 1929 case PR_SET_MM_ENV_END: 1930 if (!vma) { 1931 error = -EFAULT; 1932 goto out; 1933 } 1934 if (opt == PR_SET_MM_START_STACK) 1935 mm->start_stack = addr; 1936 else if (opt == PR_SET_MM_ARG_START) 1937 mm->arg_start = addr; 1938 else if (opt == PR_SET_MM_ARG_END) 1939 mm->arg_end = addr; 1940 else if (opt == PR_SET_MM_ENV_START) 1941 mm->env_start = addr; 1942 else if (opt == PR_SET_MM_ENV_END) 1943 mm->env_end = addr; 1944 break; 1945 1946 /* 1947 * This doesn't move auxiliary vector itself 1948 * since it's pinned to mm_struct, but allow 1949 * to fill vector with new values. It's up 1950 * to a caller to provide sane values here 1951 * otherwise user space tools which use this 1952 * vector might be unhappy. 1953 */ 1954 case PR_SET_MM_AUXV: { 1955 unsigned long user_auxv[AT_VECTOR_SIZE]; 1956 1957 if (arg4 > sizeof(user_auxv)) 1958 goto out; 1959 up_read(&mm->mmap_sem); 1960 1961 if (copy_from_user(user_auxv, (const void __user *)addr, arg4)) 1962 return -EFAULT; 1963 1964 /* Make sure the last entry is always AT_NULL */ 1965 user_auxv[AT_VECTOR_SIZE - 2] = 0; 1966 user_auxv[AT_VECTOR_SIZE - 1] = 0; 1967 1968 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv)); 1969 1970 task_lock(current); 1971 memcpy(mm->saved_auxv, user_auxv, arg4); 1972 task_unlock(current); 1973 1974 return 0; 1975 } 1976 default: 1977 goto out; 1978 } 1979 1980 error = 0; 1981 out: 1982 up_read(&mm->mmap_sem); 1983 return error; 1984 } 1985 1986 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr) 1987 { 1988 return put_user(me->clear_child_tid, tid_addr); 1989 } 1990 1991 #else /* CONFIG_CHECKPOINT_RESTORE */ 1992 static int prctl_set_mm(int opt, unsigned long addr, 1993 unsigned long arg4, unsigned long arg5) 1994 { 1995 return -EINVAL; 1996 } 1997 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr) 1998 { 1999 return -EINVAL; 2000 } 2001 #endif 2002 2003 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, 2004 unsigned long, arg4, unsigned long, arg5) 2005 { 2006 struct task_struct *me = current; 2007 unsigned char comm[sizeof(me->comm)]; 2008 long error; 2009 2010 error = security_task_prctl(option, arg2, arg3, arg4, arg5); 2011 if (error != -ENOSYS) 2012 return error; 2013 2014 error = 0; 2015 switch (option) { 2016 case PR_SET_PDEATHSIG: 2017 if (!valid_signal(arg2)) { 2018 error = -EINVAL; 2019 break; 2020 } 2021 me->pdeath_signal = arg2; 2022 break; 2023 case PR_GET_PDEATHSIG: 2024 error = put_user(me->pdeath_signal, (int __user *)arg2); 2025 break; 2026 case PR_GET_DUMPABLE: 2027 error = get_dumpable(me->mm); 2028 break; 2029 case PR_SET_DUMPABLE: 2030 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) { 2031 error = -EINVAL; 2032 break; 2033 } 2034 set_dumpable(me->mm, arg2); 2035 break; 2036 2037 case PR_SET_UNALIGN: 2038 error = SET_UNALIGN_CTL(me, arg2); 2039 break; 2040 case PR_GET_UNALIGN: 2041 error = GET_UNALIGN_CTL(me, arg2); 2042 break; 2043 case PR_SET_FPEMU: 2044 error = SET_FPEMU_CTL(me, arg2); 2045 break; 2046 case PR_GET_FPEMU: 2047 error = GET_FPEMU_CTL(me, arg2); 2048 break; 2049 case PR_SET_FPEXC: 2050 error = SET_FPEXC_CTL(me, arg2); 2051 break; 2052 case PR_GET_FPEXC: 2053 error = GET_FPEXC_CTL(me, arg2); 2054 break; 2055 case PR_GET_TIMING: 2056 error = PR_TIMING_STATISTICAL; 2057 break; 2058 case PR_SET_TIMING: 2059 if (arg2 != PR_TIMING_STATISTICAL) 2060 error = -EINVAL; 2061 break; 2062 case PR_SET_NAME: 2063 comm[sizeof(me->comm) - 1] = 0; 2064 if (strncpy_from_user(comm, (char __user *)arg2, 2065 sizeof(me->comm) - 1) < 0) 2066 return -EFAULT; 2067 set_task_comm(me, comm); 2068 proc_comm_connector(me); 2069 break; 2070 case PR_GET_NAME: 2071 get_task_comm(comm, me); 2072 if (copy_to_user((char __user *)arg2, comm, sizeof(comm))) 2073 return -EFAULT; 2074 break; 2075 case PR_GET_ENDIAN: 2076 error = GET_ENDIAN(me, arg2); 2077 break; 2078 case PR_SET_ENDIAN: 2079 error = SET_ENDIAN(me, arg2); 2080 break; 2081 case PR_GET_SECCOMP: 2082 error = prctl_get_seccomp(); 2083 break; 2084 case PR_SET_SECCOMP: 2085 error = prctl_set_seccomp(arg2, (char __user *)arg3); 2086 break; 2087 case PR_GET_TSC: 2088 error = GET_TSC_CTL(arg2); 2089 break; 2090 case PR_SET_TSC: 2091 error = SET_TSC_CTL(arg2); 2092 break; 2093 case PR_TASK_PERF_EVENTS_DISABLE: 2094 error = perf_event_task_disable(); 2095 break; 2096 case PR_TASK_PERF_EVENTS_ENABLE: 2097 error = perf_event_task_enable(); 2098 break; 2099 case PR_GET_TIMERSLACK: 2100 error = current->timer_slack_ns; 2101 break; 2102 case PR_SET_TIMERSLACK: 2103 if (arg2 <= 0) 2104 current->timer_slack_ns = 2105 current->default_timer_slack_ns; 2106 else 2107 current->timer_slack_ns = arg2; 2108 break; 2109 case PR_MCE_KILL: 2110 if (arg4 | arg5) 2111 return -EINVAL; 2112 switch (arg2) { 2113 case PR_MCE_KILL_CLEAR: 2114 if (arg3 != 0) 2115 return -EINVAL; 2116 current->flags &= ~PF_MCE_PROCESS; 2117 break; 2118 case PR_MCE_KILL_SET: 2119 current->flags |= PF_MCE_PROCESS; 2120 if (arg3 == PR_MCE_KILL_EARLY) 2121 current->flags |= PF_MCE_EARLY; 2122 else if (arg3 == PR_MCE_KILL_LATE) 2123 current->flags &= ~PF_MCE_EARLY; 2124 else if (arg3 == PR_MCE_KILL_DEFAULT) 2125 current->flags &= 2126 ~(PF_MCE_EARLY|PF_MCE_PROCESS); 2127 else 2128 return -EINVAL; 2129 break; 2130 default: 2131 return -EINVAL; 2132 } 2133 break; 2134 case PR_MCE_KILL_GET: 2135 if (arg2 | arg3 | arg4 | arg5) 2136 return -EINVAL; 2137 if (current->flags & PF_MCE_PROCESS) 2138 error = (current->flags & PF_MCE_EARLY) ? 2139 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE; 2140 else 2141 error = PR_MCE_KILL_DEFAULT; 2142 break; 2143 case PR_SET_MM: 2144 error = prctl_set_mm(arg2, arg3, arg4, arg5); 2145 break; 2146 case PR_GET_TID_ADDRESS: 2147 error = prctl_get_tid_address(me, (int __user **)arg2); 2148 break; 2149 case PR_SET_CHILD_SUBREAPER: 2150 me->signal->is_child_subreaper = !!arg2; 2151 break; 2152 case PR_GET_CHILD_SUBREAPER: 2153 error = put_user(me->signal->is_child_subreaper, 2154 (int __user *)arg2); 2155 break; 2156 case PR_SET_NO_NEW_PRIVS: 2157 if (arg2 != 1 || arg3 || arg4 || arg5) 2158 return -EINVAL; 2159 2160 current->no_new_privs = 1; 2161 break; 2162 case PR_GET_NO_NEW_PRIVS: 2163 if (arg2 || arg3 || arg4 || arg5) 2164 return -EINVAL; 2165 return current->no_new_privs ? 1 : 0; 2166 default: 2167 error = -EINVAL; 2168 break; 2169 } 2170 return error; 2171 } 2172 2173 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep, 2174 struct getcpu_cache __user *, unused) 2175 { 2176 int err = 0; 2177 int cpu = raw_smp_processor_id(); 2178 if (cpup) 2179 err |= put_user(cpu, cpup); 2180 if (nodep) 2181 err |= put_user(cpu_to_node(cpu), nodep); 2182 return err ? -EFAULT : 0; 2183 } 2184 2185 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff"; 2186 2187 static void argv_cleanup(struct subprocess_info *info) 2188 { 2189 argv_free(info->argv); 2190 } 2191 2192 static int __orderly_poweroff(void) 2193 { 2194 int argc; 2195 char **argv; 2196 static char *envp[] = { 2197 "HOME=/", 2198 "PATH=/sbin:/bin:/usr/sbin:/usr/bin", 2199 NULL 2200 }; 2201 int ret; 2202 2203 argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc); 2204 if (argv == NULL) { 2205 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n", 2206 __func__, poweroff_cmd); 2207 return -ENOMEM; 2208 } 2209 2210 ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_WAIT_EXEC, 2211 NULL, argv_cleanup, NULL); 2212 if (ret == -ENOMEM) 2213 argv_free(argv); 2214 2215 return ret; 2216 } 2217 2218 /** 2219 * orderly_poweroff - Trigger an orderly system poweroff 2220 * @force: force poweroff if command execution fails 2221 * 2222 * This may be called from any context to trigger a system shutdown. 2223 * If the orderly shutdown fails, it will force an immediate shutdown. 2224 */ 2225 int orderly_poweroff(bool force) 2226 { 2227 int ret = __orderly_poweroff(); 2228 2229 if (ret && force) { 2230 printk(KERN_WARNING "Failed to start orderly shutdown: " 2231 "forcing the issue\n"); 2232 2233 /* 2234 * I guess this should try to kick off some daemon to sync and 2235 * poweroff asap. Or not even bother syncing if we're doing an 2236 * emergency shutdown? 2237 */ 2238 emergency_sync(); 2239 kernel_power_off(); 2240 } 2241 2242 return ret; 2243 } 2244 EXPORT_SYMBOL_GPL(orderly_poweroff); 2245