1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/kernel/sys.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8#include <linux/export.h>
   9#include <linux/mm.h>
  10#include <linux/utsname.h>
  11#include <linux/mman.h>
  12#include <linux/reboot.h>
  13#include <linux/prctl.h>
  14#include <linux/highuid.h>
  15#include <linux/fs.h>
  16#include <linux/kmod.h>
  17#include <linux/perf_event.h>
  18#include <linux/resource.h>
  19#include <linux/kernel.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/suspend.h>
  28#include <linux/tty.h>
  29#include <linux/signal.h>
  30#include <linux/cn_proc.h>
  31#include <linux/getcpu.h>
  32#include <linux/task_io_accounting_ops.h>
  33#include <linux/seccomp.h>
  34#include <linux/cpu.h>
  35#include <linux/personality.h>
  36#include <linux/ptrace.h>
  37#include <linux/fs_struct.h>
  38#include <linux/file.h>
  39#include <linux/mount.h>
  40#include <linux/gfp.h>
  41#include <linux/syscore_ops.h>
  42#include <linux/version.h>
  43#include <linux/ctype.h>
  44#include <linux/syscall_user_dispatch.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/time_namespace.h>
  51#include <linux/binfmts.h>
  52
  53#include <linux/sched.h>
  54#include <linux/sched/autogroup.h>
  55#include <linux/sched/loadavg.h>
  56#include <linux/sched/stat.h>
  57#include <linux/sched/mm.h>
  58#include <linux/sched/coredump.h>
  59#include <linux/sched/task.h>
  60#include <linux/sched/cputime.h>
  61#include <linux/rcupdate.h>
  62#include <linux/uidgid.h>
  63#include <linux/cred.h>
  64
  65#include <linux/nospec.h>
  66
  67#include <linux/kmsg_dump.h>
  68/* Move somewhere else to avoid recompiling? */
  69#include <generated/utsrelease.h>
  70
  71#include <linux/uaccess.h>
  72#include <asm/io.h>
  73#include <asm/unistd.h>
  74
  75#include "uid16.h"
  76
  77#ifndef SET_UNALIGN_CTL
  78# define SET_UNALIGN_CTL(a, b)	(-EINVAL)
  79#endif
  80#ifndef GET_UNALIGN_CTL
  81# define GET_UNALIGN_CTL(a, b)	(-EINVAL)
  82#endif
  83#ifndef SET_FPEMU_CTL
  84# define SET_FPEMU_CTL(a, b)	(-EINVAL)
  85#endif
  86#ifndef GET_FPEMU_CTL
  87# define GET_FPEMU_CTL(a, b)	(-EINVAL)
  88#endif
  89#ifndef SET_FPEXC_CTL
  90# define SET_FPEXC_CTL(a, b)	(-EINVAL)
  91#endif
  92#ifndef GET_FPEXC_CTL
  93# define GET_FPEXC_CTL(a, b)	(-EINVAL)
  94#endif
  95#ifndef GET_ENDIAN
  96# define GET_ENDIAN(a, b)	(-EINVAL)
  97#endif
  98#ifndef SET_ENDIAN
  99# define SET_ENDIAN(a, b)	(-EINVAL)
 100#endif
 101#ifndef GET_TSC_CTL
 102# define GET_TSC_CTL(a)		(-EINVAL)
 103#endif
 104#ifndef SET_TSC_CTL
 105# define SET_TSC_CTL(a)		(-EINVAL)
 106#endif
 107#ifndef GET_FP_MODE
 108# define GET_FP_MODE(a)		(-EINVAL)
 109#endif
 110#ifndef SET_FP_MODE
 111# define SET_FP_MODE(a,b)	(-EINVAL)
 112#endif
 113#ifndef SVE_SET_VL
 114# define SVE_SET_VL(a)		(-EINVAL)
 115#endif
 116#ifndef SVE_GET_VL
 117# define SVE_GET_VL()		(-EINVAL)
 118#endif
 119#ifndef PAC_RESET_KEYS
 120# define PAC_RESET_KEYS(a, b)	(-EINVAL)
 121#endif
 122#ifndef PAC_SET_ENABLED_KEYS
 123# define PAC_SET_ENABLED_KEYS(a, b, c)	(-EINVAL)
 124#endif
 125#ifndef PAC_GET_ENABLED_KEYS
 126# define PAC_GET_ENABLED_KEYS(a)	(-EINVAL)
 127#endif
 128#ifndef SET_TAGGED_ADDR_CTRL
 129# define SET_TAGGED_ADDR_CTRL(a)	(-EINVAL)
 130#endif
 131#ifndef GET_TAGGED_ADDR_CTRL
 132# define GET_TAGGED_ADDR_CTRL()		(-EINVAL)
 133#endif
 134
 135/*
 136 * this is where the system-wide overflow UID and GID are defined, for
 137 * architectures that now have 32-bit UID/GID but didn't in the past
 138 */
 139
 140int overflowuid = DEFAULT_OVERFLOWUID;
 141int overflowgid = DEFAULT_OVERFLOWGID;
 142
 143EXPORT_SYMBOL(overflowuid);
 144EXPORT_SYMBOL(overflowgid);
 145
 146/*
 147 * the same as above, but for filesystems which can only store a 16-bit
 148 * UID and GID. as such, this is needed on all architectures
 149 */
 150
 151int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 152int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
 153
 154EXPORT_SYMBOL(fs_overflowuid);
 155EXPORT_SYMBOL(fs_overflowgid);
 156
 157/*
 158 * Returns true if current's euid is same as p's uid or euid,
 159 * or has CAP_SYS_NICE to p's user_ns.
 160 *
 161 * Called with rcu_read_lock, creds are safe
 162 */
 163static bool set_one_prio_perm(struct task_struct *p)
 164{
 165	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 166
 167	if (uid_eq(pcred->uid,  cred->euid) ||
 168	    uid_eq(pcred->euid, cred->euid))
 169		return true;
 170	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
 171		return true;
 172	return false;
 173}
 174
 175/*
 176 * set the priority of a task
 177 * - the caller must hold the RCU read lock
 178 */
 179static int set_one_prio(struct task_struct *p, int niceval, int error)
 180{
 181	int no_nice;
 182
 183	if (!set_one_prio_perm(p)) {
 184		error = -EPERM;
 185		goto out;
 186	}
 187	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 188		error = -EACCES;
 189		goto out;
 190	}
 191	no_nice = security_task_setnice(p, niceval);
 192	if (no_nice) {
 193		error = no_nice;
 194		goto out;
 195	}
 196	if (error == -ESRCH)
 197		error = 0;
 198	set_user_nice(p, niceval);
 199out:
 200	return error;
 201}
 202
 203SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 204{
 205	struct task_struct *g, *p;
 206	struct user_struct *user;
 207	const struct cred *cred = current_cred();
 208	int error = -EINVAL;
 209	struct pid *pgrp;
 210	kuid_t uid;
 211
 212	if (which > PRIO_USER || which < PRIO_PROCESS)
 213		goto out;
 214
 215	/* normalize: avoid signed division (rounding problems) */
 216	error = -ESRCH;
 217	if (niceval < MIN_NICE)
 218		niceval = MIN_NICE;
 219	if (niceval > MAX_NICE)
 220		niceval = MAX_NICE;
 221
 222	rcu_read_lock();
 223	read_lock(&tasklist_lock);
 224	switch (which) {
 225	case PRIO_PROCESS:
 226		if (who)
 227			p = find_task_by_vpid(who);
 228		else
 229			p = current;
 230		if (p)
 231			error = set_one_prio(p, niceval, error);
 232		break;
 233	case PRIO_PGRP:
 234		if (who)
 235			pgrp = find_vpid(who);
 236		else
 237			pgrp = task_pgrp(current);
 238		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 239			error = set_one_prio(p, niceval, error);
 240		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 241		break;
 242	case PRIO_USER:
 243		uid = make_kuid(cred->user_ns, who);
 244		user = cred->user;
 245		if (!who)
 246			uid = cred->uid;
 247		else if (!uid_eq(uid, cred->uid)) {
 248			user = find_user(uid);
 249			if (!user)
 250				goto out_unlock;	/* No processes for this user */
 251		}
 252		do_each_thread(g, p) {
 253			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
 254				error = set_one_prio(p, niceval, error);
 255		} while_each_thread(g, p);
 256		if (!uid_eq(uid, cred->uid))
 257			free_uid(user);		/* For find_user() */
 258		break;
 259	}
 260out_unlock:
 261	read_unlock(&tasklist_lock);
 262	rcu_read_unlock();
 263out:
 264	return error;
 265}
 266
 267/*
 268 * Ugh. To avoid negative return values, "getpriority()" will
 269 * not return the normal nice-value, but a negated value that
 270 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 271 * to stay compatible.
 272 */
 273SYSCALL_DEFINE2(getpriority, int, which, int, who)
 274{
 275	struct task_struct *g, *p;
 276	struct user_struct *user;
 277	const struct cred *cred = current_cred();
 278	long niceval, retval = -ESRCH;
 279	struct pid *pgrp;
 280	kuid_t uid;
 281
 282	if (which > PRIO_USER || which < PRIO_PROCESS)
 283		return -EINVAL;
 284
 285	rcu_read_lock();
 286	read_lock(&tasklist_lock);
 287	switch (which) {
 288	case PRIO_PROCESS:
 289		if (who)
 290			p = find_task_by_vpid(who);
 291		else
 292			p = current;
 293		if (p) {
 294			niceval = nice_to_rlimit(task_nice(p));
 295			if (niceval > retval)
 296				retval = niceval;
 297		}
 298		break;
 299	case PRIO_PGRP:
 300		if (who)
 301			pgrp = find_vpid(who);
 302		else
 303			pgrp = task_pgrp(current);
 304		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 305			niceval = nice_to_rlimit(task_nice(p));
 306			if (niceval > retval)
 307				retval = niceval;
 308		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 309		break;
 310	case PRIO_USER:
 311		uid = make_kuid(cred->user_ns, who);
 312		user = cred->user;
 313		if (!who)
 314			uid = cred->uid;
 315		else if (!uid_eq(uid, cred->uid)) {
 316			user = find_user(uid);
 317			if (!user)
 318				goto out_unlock;	/* No processes for this user */
 319		}
 320		do_each_thread(g, p) {
 321			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
 322				niceval = nice_to_rlimit(task_nice(p));
 323				if (niceval > retval)
 324					retval = niceval;
 325			}
 326		} while_each_thread(g, p);
 327		if (!uid_eq(uid, cred->uid))
 328			free_uid(user);		/* for find_user() */
 329		break;
 330	}
 331out_unlock:
 332	read_unlock(&tasklist_lock);
 333	rcu_read_unlock();
 334
 335	return retval;
 336}
 337
 338/*
 339 * Unprivileged users may change the real gid to the effective gid
 340 * or vice versa.  (BSD-style)
 341 *
 342 * If you set the real gid at all, or set the effective gid to a value not
 343 * equal to the real gid, then the saved gid is set to the new effective gid.
 344 *
 345 * This makes it possible for a setgid program to completely drop its
 346 * privileges, which is often a useful assertion to make when you are doing
 347 * a security audit over a program.
 348 *
 349 * The general idea is that a program which uses just setregid() will be
 350 * 100% compatible with BSD.  A program which uses just setgid() will be
 351 * 100% compatible with POSIX with saved IDs.
 352 *
 353 * SMP: There are not races, the GIDs are checked only by filesystem
 354 *      operations (as far as semantic preservation is concerned).
 355 */
 356#ifdef CONFIG_MULTIUSER
 357long __sys_setregid(gid_t rgid, gid_t egid)
 358{
 359	struct user_namespace *ns = current_user_ns();
 360	const struct cred *old;
 361	struct cred *new;
 362	int retval;
 363	kgid_t krgid, kegid;
 364
 365	krgid = make_kgid(ns, rgid);
 366	kegid = make_kgid(ns, egid);
 367
 368	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 369		return -EINVAL;
 370	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 371		return -EINVAL;
 372
 373	new = prepare_creds();
 374	if (!new)
 375		return -ENOMEM;
 376	old = current_cred();
 377
 378	retval = -EPERM;
 379	if (rgid != (gid_t) -1) {
 380		if (gid_eq(old->gid, krgid) ||
 381		    gid_eq(old->egid, krgid) ||
 382		    ns_capable_setid(old->user_ns, CAP_SETGID))
 383			new->gid = krgid;
 384		else
 385			goto error;
 386	}
 387	if (egid != (gid_t) -1) {
 388		if (gid_eq(old->gid, kegid) ||
 389		    gid_eq(old->egid, kegid) ||
 390		    gid_eq(old->sgid, kegid) ||
 391		    ns_capable_setid(old->user_ns, CAP_SETGID))
 392			new->egid = kegid;
 393		else
 394			goto error;
 395	}
 396
 397	if (rgid != (gid_t) -1 ||
 398	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
 399		new->sgid = new->egid;
 400	new->fsgid = new->egid;
 401
 402	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
 403	if (retval < 0)
 404		goto error;
 405
 406	return commit_creds(new);
 407
 408error:
 409	abort_creds(new);
 410	return retval;
 411}
 412
 413SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 414{
 415	return __sys_setregid(rgid, egid);
 416}
 417
 418/*
 419 * setgid() is implemented like SysV w/ SAVED_IDS
 420 *
 421 * SMP: Same implicit races as above.
 422 */
 423long __sys_setgid(gid_t gid)
 424{
 425	struct user_namespace *ns = current_user_ns();
 426	const struct cred *old;
 427	struct cred *new;
 428	int retval;
 429	kgid_t kgid;
 430
 431	kgid = make_kgid(ns, gid);
 432	if (!gid_valid(kgid))
 433		return -EINVAL;
 434
 435	new = prepare_creds();
 436	if (!new)
 437		return -ENOMEM;
 438	old = current_cred();
 439
 440	retval = -EPERM;
 441	if (ns_capable_setid(old->user_ns, CAP_SETGID))
 442		new->gid = new->egid = new->sgid = new->fsgid = kgid;
 443	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
 444		new->egid = new->fsgid = kgid;
 445	else
 446		goto error;
 447
 448	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
 449	if (retval < 0)
 450		goto error;
 451
 452	return commit_creds(new);
 453
 454error:
 455	abort_creds(new);
 456	return retval;
 457}
 458
 459SYSCALL_DEFINE1(setgid, gid_t, gid)
 460{
 461	return __sys_setgid(gid);
 462}
 463
 464/*
 465 * change the user struct in a credentials set to match the new UID
 466 */
 467static int set_user(struct cred *new)
 468{
 469	struct user_struct *new_user;
 470
 471	new_user = alloc_uid(new->uid);
 472	if (!new_user)
 473		return -EAGAIN;
 474
 475	/*
 476	 * We don't fail in case of NPROC limit excess here because too many
 477	 * poorly written programs don't check set*uid() return code, assuming
 478	 * it never fails if called by root.  We may still enforce NPROC limit
 479	 * for programs doing set*uid()+execve() by harmlessly deferring the
 480	 * failure to the execve() stage.
 481	 */
 482	if (is_ucounts_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
 483			new_user != INIT_USER)
 484		current->flags |= PF_NPROC_EXCEEDED;
 485	else
 486		current->flags &= ~PF_NPROC_EXCEEDED;
 487
 488	free_uid(new->user);
 489	new->user = new_user;
 490	return 0;
 491}
 492
 493/*
 494 * Unprivileged users may change the real uid to the effective uid
 495 * or vice versa.  (BSD-style)
 496 *
 497 * If you set the real uid at all, or set the effective uid to a value not
 498 * equal to the real uid, then the saved uid is set to the new effective uid.
 499 *
 500 * This makes it possible for a setuid program to completely drop its
 501 * privileges, which is often a useful assertion to make when you are doing
 502 * a security audit over a program.
 503 *
 504 * The general idea is that a program which uses just setreuid() will be
 505 * 100% compatible with BSD.  A program which uses just setuid() will be
 506 * 100% compatible with POSIX with saved IDs.
 507 */
 508long __sys_setreuid(uid_t ruid, uid_t euid)
 509{
 510	struct user_namespace *ns = current_user_ns();
 511	const struct cred *old;
 512	struct cred *new;
 513	int retval;
 514	kuid_t kruid, keuid;
 515
 516	kruid = make_kuid(ns, ruid);
 517	keuid = make_kuid(ns, euid);
 518
 519	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 520		return -EINVAL;
 521	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 522		return -EINVAL;
 523
 524	new = prepare_creds();
 525	if (!new)
 526		return -ENOMEM;
 527	old = current_cred();
 528
 529	retval = -EPERM;
 530	if (ruid != (uid_t) -1) {
 531		new->uid = kruid;
 532		if (!uid_eq(old->uid, kruid) &&
 533		    !uid_eq(old->euid, kruid) &&
 534		    !ns_capable_setid(old->user_ns, CAP_SETUID))
 535			goto error;
 536	}
 537
 538	if (euid != (uid_t) -1) {
 539		new->euid = keuid;
 540		if (!uid_eq(old->uid, keuid) &&
 541		    !uid_eq(old->euid, keuid) &&
 542		    !uid_eq(old->suid, keuid) &&
 543		    !ns_capable_setid(old->user_ns, CAP_SETUID))
 544			goto error;
 545	}
 546
 547	if (!uid_eq(new->uid, old->uid)) {
 548		retval = set_user(new);
 549		if (retval < 0)
 550			goto error;
 551	}
 552	if (ruid != (uid_t) -1 ||
 553	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
 554		new->suid = new->euid;
 555	new->fsuid = new->euid;
 556
 557	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 558	if (retval < 0)
 559		goto error;
 560
 561	retval = set_cred_ucounts(new);
 562	if (retval < 0)
 563		goto error;
 564
 565	return commit_creds(new);
 566
 567error:
 568	abort_creds(new);
 569	return retval;
 570}
 571
 572SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 573{
 574	return __sys_setreuid(ruid, euid);
 575}
 576
 577/*
 578 * setuid() is implemented like SysV with SAVED_IDS
 579 *
 580 * Note that SAVED_ID's is deficient in that a setuid root program
 581 * like sendmail, for example, cannot set its uid to be a normal
 582 * user and then switch back, because if you're root, setuid() sets
 583 * the saved uid too.  If you don't like this, blame the bright people
 584 * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 585 * will allow a root program to temporarily drop privileges and be able to
 586 * regain them by swapping the real and effective uid.
 587 */
 588long __sys_setuid(uid_t uid)
 589{
 590	struct user_namespace *ns = current_user_ns();
 591	const struct cred *old;
 592	struct cred *new;
 593	int retval;
 594	kuid_t kuid;
 595
 596	kuid = make_kuid(ns, uid);
 597	if (!uid_valid(kuid))
 598		return -EINVAL;
 599
 600	new = prepare_creds();
 601	if (!new)
 602		return -ENOMEM;
 603	old = current_cred();
 604
 605	retval = -EPERM;
 606	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
 607		new->suid = new->uid = kuid;
 608		if (!uid_eq(kuid, old->uid)) {
 609			retval = set_user(new);
 610			if (retval < 0)
 611				goto error;
 612		}
 613	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
 614		goto error;
 615	}
 616
 617	new->fsuid = new->euid = kuid;
 618
 619	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 620	if (retval < 0)
 621		goto error;
 622
 623	retval = set_cred_ucounts(new);
 624	if (retval < 0)
 625		goto error;
 626
 627	return commit_creds(new);
 628
 629error:
 630	abort_creds(new);
 631	return retval;
 632}
 633
 634SYSCALL_DEFINE1(setuid, uid_t, uid)
 635{
 636	return __sys_setuid(uid);
 637}
 638
 639
 640/*
 641 * This function implements a generic ability to update ruid, euid,
 642 * and suid.  This allows you to implement the 4.4 compatible seteuid().
 643 */
 644long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 645{
 646	struct user_namespace *ns = current_user_ns();
 647	const struct cred *old;
 648	struct cred *new;
 649	int retval;
 650	kuid_t kruid, keuid, ksuid;
 651
 652	kruid = make_kuid(ns, ruid);
 653	keuid = make_kuid(ns, euid);
 654	ksuid = make_kuid(ns, suid);
 655
 656	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 657		return -EINVAL;
 658
 659	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 660		return -EINVAL;
 661
 662	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
 663		return -EINVAL;
 664
 665	new = prepare_creds();
 666	if (!new)
 667		return -ENOMEM;
 668
 669	old = current_cred();
 670
 671	retval = -EPERM;
 672	if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
 673		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
 674		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
 675			goto error;
 676		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
 677		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
 678			goto error;
 679		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
 680		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
 681			goto error;
 682	}
 683
 684	if (ruid != (uid_t) -1) {
 685		new->uid = kruid;
 686		if (!uid_eq(kruid, old->uid)) {
 687			retval = set_user(new);
 688			if (retval < 0)
 689				goto error;
 690		}
 691	}
 692	if (euid != (uid_t) -1)
 693		new->euid = keuid;
 694	if (suid != (uid_t) -1)
 695		new->suid = ksuid;
 696	new->fsuid = new->euid;
 697
 698	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 699	if (retval < 0)
 700		goto error;
 701
 702	retval = set_cred_ucounts(new);
 703	if (retval < 0)
 704		goto error;
 705
 706	return commit_creds(new);
 707
 708error:
 709	abort_creds(new);
 710	return retval;
 711}
 712
 713SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 714{
 715	return __sys_setresuid(ruid, euid, suid);
 716}
 717
 718SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
 719{
 720	const struct cred *cred = current_cred();
 721	int retval;
 722	uid_t ruid, euid, suid;
 723
 724	ruid = from_kuid_munged(cred->user_ns, cred->uid);
 725	euid = from_kuid_munged(cred->user_ns, cred->euid);
 726	suid = from_kuid_munged(cred->user_ns, cred->suid);
 727
 728	retval = put_user(ruid, ruidp);
 729	if (!retval) {
 730		retval = put_user(euid, euidp);
 731		if (!retval)
 732			return put_user(suid, suidp);
 733	}
 734	return retval;
 735}
 736
 737/*
 738 * Same as above, but for rgid, egid, sgid.
 739 */
 740long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 741{
 742	struct user_namespace *ns = current_user_ns();
 743	const struct cred *old;
 744	struct cred *new;
 745	int retval;
 746	kgid_t krgid, kegid, ksgid;
 747
 748	krgid = make_kgid(ns, rgid);
 749	kegid = make_kgid(ns, egid);
 750	ksgid = make_kgid(ns, sgid);
 751
 752	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 753		return -EINVAL;
 754	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 755		return -EINVAL;
 756	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
 757		return -EINVAL;
 758
 759	new = prepare_creds();
 760	if (!new)
 761		return -ENOMEM;
 762	old = current_cred();
 763
 764	retval = -EPERM;
 765	if (!ns_capable_setid(old->user_ns, CAP_SETGID)) {
 766		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
 767		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
 768			goto error;
 769		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
 770		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
 771			goto error;
 772		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
 773		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
 774			goto error;
 775	}
 776
 777	if (rgid != (gid_t) -1)
 778		new->gid = krgid;
 779	if (egid != (gid_t) -1)
 780		new->egid = kegid;
 781	if (sgid != (gid_t) -1)
 782		new->sgid = ksgid;
 783	new->fsgid = new->egid;
 784
 785	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
 786	if (retval < 0)
 787		goto error;
 788
 789	return commit_creds(new);
 790
 791error:
 792	abort_creds(new);
 793	return retval;
 794}
 795
 796SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 797{
 798	return __sys_setresgid(rgid, egid, sgid);
 799}
 800
 801SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
 802{
 803	const struct cred *cred = current_cred();
 804	int retval;
 805	gid_t rgid, egid, sgid;
 806
 807	rgid = from_kgid_munged(cred->user_ns, cred->gid);
 808	egid = from_kgid_munged(cred->user_ns, cred->egid);
 809	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
 810
 811	retval = put_user(rgid, rgidp);
 812	if (!retval) {
 813		retval = put_user(egid, egidp);
 814		if (!retval)
 815			retval = put_user(sgid, sgidp);
 816	}
 817
 818	return retval;
 819}
 820
 821
 822/*
 823 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 824 * is used for "access()" and for the NFS daemon (letting nfsd stay at
 825 * whatever uid it wants to). It normally shadows "euid", except when
 826 * explicitly set by setfsuid() or for access..
 827 */
 828long __sys_setfsuid(uid_t uid)
 829{
 830	const struct cred *old;
 831	struct cred *new;
 832	uid_t old_fsuid;
 833	kuid_t kuid;
 834
 835	old = current_cred();
 836	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
 837
 838	kuid = make_kuid(old->user_ns, uid);
 839	if (!uid_valid(kuid))
 840		return old_fsuid;
 841
 842	new = prepare_creds();
 843	if (!new)
 844		return old_fsuid;
 845
 846	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
 847	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
 848	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
 849		if (!uid_eq(kuid, old->fsuid)) {
 850			new->fsuid = kuid;
 851			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 852				goto change_okay;
 853		}
 854	}
 855
 856	abort_creds(new);
 857	return old_fsuid;
 858
 859change_okay:
 860	commit_creds(new);
 861	return old_fsuid;
 862}
 863
 864SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 865{
 866	return __sys_setfsuid(uid);
 867}
 868
 869/*
 870 * Samma på svenska..
 871 */
 872long __sys_setfsgid(gid_t gid)
 873{
 874	const struct cred *old;
 875	struct cred *new;
 876	gid_t old_fsgid;
 877	kgid_t kgid;
 878
 879	old = current_cred();
 880	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
 881
 882	kgid = make_kgid(old->user_ns, gid);
 883	if (!gid_valid(kgid))
 884		return old_fsgid;
 885
 886	new = prepare_creds();
 887	if (!new)
 888		return old_fsgid;
 889
 890	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
 891	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
 892	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
 893		if (!gid_eq(kgid, old->fsgid)) {
 894			new->fsgid = kgid;
 895			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
 896				goto change_okay;
 897		}
 898	}
 899
 900	abort_creds(new);
 901	return old_fsgid;
 902
 903change_okay:
 904	commit_creds(new);
 905	return old_fsgid;
 906}
 907
 908SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 909{
 910	return __sys_setfsgid(gid);
 911}
 912#endif /* CONFIG_MULTIUSER */
 913
 914/**
 915 * sys_getpid - return the thread group id of the current process
 916 *
 917 * Note, despite the name, this returns the tgid not the pid.  The tgid and
 918 * the pid are identical unless CLONE_THREAD was specified on clone() in
 919 * which case the tgid is the same in all threads of the same group.
 920 *
 921 * This is SMP safe as current->tgid does not change.
 922 */
 923SYSCALL_DEFINE0(getpid)
 924{
 925	return task_tgid_vnr(current);
 926}
 927
 928/* Thread ID - the internal kernel "pid" */
 929SYSCALL_DEFINE0(gettid)
 930{
 931	return task_pid_vnr(current);
 932}
 933
 934/*
 935 * Accessing ->real_parent is not SMP-safe, it could
 936 * change from under us. However, we can use a stale
 937 * value of ->real_parent under rcu_read_lock(), see
 938 * release_task()->call_rcu(delayed_put_task_struct).
 939 */
 940SYSCALL_DEFINE0(getppid)
 941{
 942	int pid;
 943
 944	rcu_read_lock();
 945	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
 946	rcu_read_unlock();
 947
 948	return pid;
 949}
 950
 951SYSCALL_DEFINE0(getuid)
 952{
 953	/* Only we change this so SMP safe */
 954	return from_kuid_munged(current_user_ns(), current_uid());
 955}
 956
 957SYSCALL_DEFINE0(geteuid)
 958{
 959	/* Only we change this so SMP safe */
 960	return from_kuid_munged(current_user_ns(), current_euid());
 961}
 962
 963SYSCALL_DEFINE0(getgid)
 964{
 965	/* Only we change this so SMP safe */
 966	return from_kgid_munged(current_user_ns(), current_gid());
 967}
 968
 969SYSCALL_DEFINE0(getegid)
 970{
 971	/* Only we change this so SMP safe */
 972	return from_kgid_munged(current_user_ns(), current_egid());
 973}
 974
 975static void do_sys_times(struct tms *tms)
 976{
 977	u64 tgutime, tgstime, cutime, cstime;
 978
 979	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
 980	cutime = current->signal->cutime;
 981	cstime = current->signal->cstime;
 982	tms->tms_utime = nsec_to_clock_t(tgutime);
 983	tms->tms_stime = nsec_to_clock_t(tgstime);
 984	tms->tms_cutime = nsec_to_clock_t(cutime);
 985	tms->tms_cstime = nsec_to_clock_t(cstime);
 986}
 987
 988SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 989{
 990	if (tbuf) {
 991		struct tms tmp;
 992
 993		do_sys_times(&tmp);
 994		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 995			return -EFAULT;
 996	}
 997	force_successful_syscall_return();
 998	return (long) jiffies_64_to_clock_t(get_jiffies_64());
 999}
1000
1001#ifdef CONFIG_COMPAT
1002static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1003{
1004	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1005}
1006
1007COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1008{
1009	if (tbuf) {
1010		struct tms tms;
1011		struct compat_tms tmp;
1012
1013		do_sys_times(&tms);
1014		/* Convert our struct tms to the compat version. */
1015		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1016		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1017		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1018		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1019		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1020			return -EFAULT;
1021	}
1022	force_successful_syscall_return();
1023	return compat_jiffies_to_clock_t(jiffies);
1024}
1025#endif
1026
1027/*
1028 * This needs some heavy checking ...
1029 * I just haven't the stomach for it. I also don't fully
1030 * understand sessions/pgrp etc. Let somebody who does explain it.
1031 *
1032 * OK, I think I have the protection semantics right.... this is really
1033 * only important on a multi-user system anyway, to make sure one user
1034 * can't send a signal to a process owned by another.  -TYT, 12/12/91
1035 *
1036 * !PF_FORKNOEXEC check to conform completely to POSIX.
1037 */
1038SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1039{
1040	struct task_struct *p;
1041	struct task_struct *group_leader = current->group_leader;
1042	struct pid *pgrp;
1043	int err;
1044
1045	if (!pid)
1046		pid = task_pid_vnr(group_leader);
1047	if (!pgid)
1048		pgid = pid;
1049	if (pgid < 0)
1050		return -EINVAL;
1051	rcu_read_lock();
1052
1053	/* From this point forward we keep holding onto the tasklist lock
1054	 * so that our parent does not change from under us. -DaveM
1055	 */
1056	write_lock_irq(&tasklist_lock);
1057
1058	err = -ESRCH;
1059	p = find_task_by_vpid(pid);
1060	if (!p)
1061		goto out;
1062
1063	err = -EINVAL;
1064	if (!thread_group_leader(p))
1065		goto out;
1066
1067	if (same_thread_group(p->real_parent, group_leader)) {
1068		err = -EPERM;
1069		if (task_session(p) != task_session(group_leader))
1070			goto out;
1071		err = -EACCES;
1072		if (!(p->flags & PF_FORKNOEXEC))
1073			goto out;
1074	} else {
1075		err = -ESRCH;
1076		if (p != group_leader)
1077			goto out;
1078	}
1079
1080	err = -EPERM;
1081	if (p->signal->leader)
1082		goto out;
1083
1084	pgrp = task_pid(p);
1085	if (pgid != pid) {
1086		struct task_struct *g;
1087
1088		pgrp = find_vpid(pgid);
1089		g = pid_task(pgrp, PIDTYPE_PGID);
1090		if (!g || task_session(g) != task_session(group_leader))
1091			goto out;
1092	}
1093
1094	err = security_task_setpgid(p, pgid);
1095	if (err)
1096		goto out;
1097
1098	if (task_pgrp(p) != pgrp)
1099		change_pid(p, PIDTYPE_PGID, pgrp);
1100
1101	err = 0;
1102out:
1103	/* All paths lead to here, thus we are safe. -DaveM */
1104	write_unlock_irq(&tasklist_lock);
1105	rcu_read_unlock();
1106	return err;
1107}
1108
1109static int do_getpgid(pid_t pid)
1110{
1111	struct task_struct *p;
1112	struct pid *grp;
1113	int retval;
1114
1115	rcu_read_lock();
1116	if (!pid)
1117		grp = task_pgrp(current);
1118	else {
1119		retval = -ESRCH;
1120		p = find_task_by_vpid(pid);
1121		if (!p)
1122			goto out;
1123		grp = task_pgrp(p);
1124		if (!grp)
1125			goto out;
1126
1127		retval = security_task_getpgid(p);
1128		if (retval)
1129			goto out;
1130	}
1131	retval = pid_vnr(grp);
1132out:
1133	rcu_read_unlock();
1134	return retval;
1135}
1136
1137SYSCALL_DEFINE1(getpgid, pid_t, pid)
1138{
1139	return do_getpgid(pid);
1140}
1141
1142#ifdef __ARCH_WANT_SYS_GETPGRP
1143
1144SYSCALL_DEFINE0(getpgrp)
1145{
1146	return do_getpgid(0);
1147}
1148
1149#endif
1150
1151SYSCALL_DEFINE1(getsid, pid_t, pid)
1152{
1153	struct task_struct *p;
1154	struct pid *sid;
1155	int retval;
1156
1157	rcu_read_lock();
1158	if (!pid)
1159		sid = task_session(current);
1160	else {
1161		retval = -ESRCH;
1162		p = find_task_by_vpid(pid);
1163		if (!p)
1164			goto out;
1165		sid = task_session(p);
1166		if (!sid)
1167			goto out;
1168
1169		retval = security_task_getsid(p);
1170		if (retval)
1171			goto out;
1172	}
1173	retval = pid_vnr(sid);
1174out:
1175	rcu_read_unlock();
1176	return retval;
1177}
1178
1179static void set_special_pids(struct pid *pid)
1180{
1181	struct task_struct *curr = current->group_leader;
1182
1183	if (task_session(curr) != pid)
1184		change_pid(curr, PIDTYPE_SID, pid);
1185
1186	if (task_pgrp(curr) != pid)
1187		change_pid(curr, PIDTYPE_PGID, pid);
1188}
1189
1190int ksys_setsid(void)
1191{
1192	struct task_struct *group_leader = current->group_leader;
1193	struct pid *sid = task_pid(group_leader);
1194	pid_t session = pid_vnr(sid);
1195	int err = -EPERM;
1196
1197	write_lock_irq(&tasklist_lock);
1198	/* Fail if I am already a session leader */
1199	if (group_leader->signal->leader)
1200		goto out;
1201
1202	/* Fail if a process group id already exists that equals the
1203	 * proposed session id.
1204	 */
1205	if (pid_task(sid, PIDTYPE_PGID))
1206		goto out;
1207
1208	group_leader->signal->leader = 1;
1209	set_special_pids(sid);
1210
1211	proc_clear_tty(group_leader);
1212
1213	err = session;
1214out:
1215	write_unlock_irq(&tasklist_lock);
1216	if (err > 0) {
1217		proc_sid_connector(group_leader);
1218		sched_autogroup_create_attach(group_leader);
1219	}
1220	return err;
1221}
1222
1223SYSCALL_DEFINE0(setsid)
1224{
1225	return ksys_setsid();
1226}
1227
1228DECLARE_RWSEM(uts_sem);
1229
1230#ifdef COMPAT_UTS_MACHINE
1231#define override_architecture(name) \
1232	(personality(current->personality) == PER_LINUX32 && \
1233	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1234		      sizeof(COMPAT_UTS_MACHINE)))
1235#else
1236#define override_architecture(name)	0
1237#endif
1238
1239/*
1240 * Work around broken programs that cannot handle "Linux 3.0".
1241 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1242 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1243 * 2.6.60.
1244 */
1245static int override_release(char __user *release, size_t len)
1246{
1247	int ret = 0;
1248
1249	if (current->personality & UNAME26) {
1250		const char *rest = UTS_RELEASE;
1251		char buf[65] = { 0 };
1252		int ndots = 0;
1253		unsigned v;
1254		size_t copy;
1255
1256		while (*rest) {
1257			if (*rest == '.' && ++ndots >= 3)
1258				break;
1259			if (!isdigit(*rest) && *rest != '.')
1260				break;
1261			rest++;
1262		}
1263		v = LINUX_VERSION_PATCHLEVEL + 60;
1264		copy = clamp_t(size_t, len, 1, sizeof(buf));
1265		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1266		ret = copy_to_user(release, buf, copy + 1);
1267	}
1268	return ret;
1269}
1270
1271SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1272{
1273	struct new_utsname tmp;
1274
1275	down_read(&uts_sem);
1276	memcpy(&tmp, utsname(), sizeof(tmp));
1277	up_read(&uts_sem);
1278	if (copy_to_user(name, &tmp, sizeof(tmp)))
1279		return -EFAULT;
1280
1281	if (override_release(name->release, sizeof(name->release)))
1282		return -EFAULT;
1283	if (override_architecture(name))
1284		return -EFAULT;
1285	return 0;
1286}
1287
1288#ifdef __ARCH_WANT_SYS_OLD_UNAME
1289/*
1290 * Old cruft
1291 */
1292SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1293{
1294	struct old_utsname tmp;
1295
1296	if (!name)
1297		return -EFAULT;
1298
1299	down_read(&uts_sem);
1300	memcpy(&tmp, utsname(), sizeof(tmp));
1301	up_read(&uts_sem);
1302	if (copy_to_user(name, &tmp, sizeof(tmp)))
1303		return -EFAULT;
1304
1305	if (override_release(name->release, sizeof(name->release)))
1306		return -EFAULT;
1307	if (override_architecture(name))
1308		return -EFAULT;
1309	return 0;
1310}
1311
1312SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1313{
1314	struct oldold_utsname tmp;
1315
1316	if (!name)
1317		return -EFAULT;
1318
1319	memset(&tmp, 0, sizeof(tmp));
1320
1321	down_read(&uts_sem);
1322	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1323	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1324	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1325	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1326	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1327	up_read(&uts_sem);
1328	if (copy_to_user(name, &tmp, sizeof(tmp)))
1329		return -EFAULT;
1330
1331	if (override_architecture(name))
1332		return -EFAULT;
1333	if (override_release(name->release, sizeof(name->release)))
1334		return -EFAULT;
1335	return 0;
1336}
1337#endif
1338
1339SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1340{
1341	int errno;
1342	char tmp[__NEW_UTS_LEN];
1343
1344	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1345		return -EPERM;
1346
1347	if (len < 0 || len > __NEW_UTS_LEN)
1348		return -EINVAL;
1349	errno = -EFAULT;
1350	if (!copy_from_user(tmp, name, len)) {
1351		struct new_utsname *u;
1352
1353		down_write(&uts_sem);
1354		u = utsname();
1355		memcpy(u->nodename, tmp, len);
1356		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1357		errno = 0;
1358		uts_proc_notify(UTS_PROC_HOSTNAME);
1359		up_write(&uts_sem);
1360	}
1361	return errno;
1362}
1363
1364#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1365
1366SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1367{
1368	int i;
1369	struct new_utsname *u;
1370	char tmp[__NEW_UTS_LEN + 1];
1371
1372	if (len < 0)
1373		return -EINVAL;
1374	down_read(&uts_sem);
1375	u = utsname();
1376	i = 1 + strlen(u->nodename);
1377	if (i > len)
1378		i = len;
1379	memcpy(tmp, u->nodename, i);
1380	up_read(&uts_sem);
1381	if (copy_to_user(name, tmp, i))
1382		return -EFAULT;
1383	return 0;
1384}
1385
1386#endif
1387
1388/*
1389 * Only setdomainname; getdomainname can be implemented by calling
1390 * uname()
1391 */
1392SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1393{
1394	int errno;
1395	char tmp[__NEW_UTS_LEN];
1396
1397	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1398		return -EPERM;
1399	if (len < 0 || len > __NEW_UTS_LEN)
1400		return -EINVAL;
1401
1402	errno = -EFAULT;
1403	if (!copy_from_user(tmp, name, len)) {
1404		struct new_utsname *u;
1405
1406		down_write(&uts_sem);
1407		u = utsname();
1408		memcpy(u->domainname, tmp, len);
1409		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1410		errno = 0;
1411		uts_proc_notify(UTS_PROC_DOMAINNAME);
1412		up_write(&uts_sem);
1413	}
1414	return errno;
1415}
1416
1417SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1418{
1419	struct rlimit value;
1420	int ret;
1421
1422	ret = do_prlimit(current, resource, NULL, &value);
1423	if (!ret)
1424		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1425
1426	return ret;
1427}
1428
1429#ifdef CONFIG_COMPAT
1430
1431COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1432		       struct compat_rlimit __user *, rlim)
1433{
1434	struct rlimit r;
1435	struct compat_rlimit r32;
1436
1437	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1438		return -EFAULT;
1439
1440	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1441		r.rlim_cur = RLIM_INFINITY;
1442	else
1443		r.rlim_cur = r32.rlim_cur;
1444	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1445		r.rlim_max = RLIM_INFINITY;
1446	else
1447		r.rlim_max = r32.rlim_max;
1448	return do_prlimit(current, resource, &r, NULL);
1449}
1450
1451COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1452		       struct compat_rlimit __user *, rlim)
1453{
1454	struct rlimit r;
1455	int ret;
1456
1457	ret = do_prlimit(current, resource, NULL, &r);
1458	if (!ret) {
1459		struct compat_rlimit r32;
1460		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1461			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1462		else
1463			r32.rlim_cur = r.rlim_cur;
1464		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1465			r32.rlim_max = COMPAT_RLIM_INFINITY;
1466		else
1467			r32.rlim_max = r.rlim_max;
1468
1469		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1470			return -EFAULT;
1471	}
1472	return ret;
1473}
1474
1475#endif
1476
1477#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1478
1479/*
1480 *	Back compatibility for getrlimit. Needed for some apps.
1481 */
1482SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1483		struct rlimit __user *, rlim)
1484{
1485	struct rlimit x;
1486	if (resource >= RLIM_NLIMITS)
1487		return -EINVAL;
1488
1489	resource = array_index_nospec(resource, RLIM_NLIMITS);
1490	task_lock(current->group_leader);
1491	x = current->signal->rlim[resource];
1492	task_unlock(current->group_leader);
1493	if (x.rlim_cur > 0x7FFFFFFF)
1494		x.rlim_cur = 0x7FFFFFFF;
1495	if (x.rlim_max > 0x7FFFFFFF)
1496		x.rlim_max = 0x7FFFFFFF;
1497	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1498}
1499
1500#ifdef CONFIG_COMPAT
1501COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1502		       struct compat_rlimit __user *, rlim)
1503{
1504	struct rlimit r;
1505
1506	if (resource >= RLIM_NLIMITS)
1507		return -EINVAL;
1508
1509	resource = array_index_nospec(resource, RLIM_NLIMITS);
1510	task_lock(current->group_leader);
1511	r = current->signal->rlim[resource];
1512	task_unlock(current->group_leader);
1513	if (r.rlim_cur > 0x7FFFFFFF)
1514		r.rlim_cur = 0x7FFFFFFF;
1515	if (r.rlim_max > 0x7FFFFFFF)
1516		r.rlim_max = 0x7FFFFFFF;
1517
1518	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1519	    put_user(r.rlim_max, &rlim->rlim_max))
1520		return -EFAULT;
1521	return 0;
1522}
1523#endif
1524
1525#endif
1526
1527static inline bool rlim64_is_infinity(__u64 rlim64)
1528{
1529#if BITS_PER_LONG < 64
1530	return rlim64 >= ULONG_MAX;
1531#else
1532	return rlim64 == RLIM64_INFINITY;
1533#endif
1534}
1535
1536static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1537{
1538	if (rlim->rlim_cur == RLIM_INFINITY)
1539		rlim64->rlim_cur = RLIM64_INFINITY;
1540	else
1541		rlim64->rlim_cur = rlim->rlim_cur;
1542	if (rlim->rlim_max == RLIM_INFINITY)
1543		rlim64->rlim_max = RLIM64_INFINITY;
1544	else
1545		rlim64->rlim_max = rlim->rlim_max;
1546}
1547
1548static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1549{
1550	if (rlim64_is_infinity(rlim64->rlim_cur))
1551		rlim->rlim_cur = RLIM_INFINITY;
1552	else
1553		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1554	if (rlim64_is_infinity(rlim64->rlim_max))
1555		rlim->rlim_max = RLIM_INFINITY;
1556	else
1557		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1558}
1559
1560/* make sure you are allowed to change @tsk limits before calling this */
1561int do_prlimit(struct task_struct *tsk, unsigned int resource,
1562		struct rlimit *new_rlim, struct rlimit *old_rlim)
1563{
1564	struct rlimit *rlim;
1565	int retval = 0;
1566
1567	if (resource >= RLIM_NLIMITS)
1568		return -EINVAL;
1569	if (new_rlim) {
1570		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1571			return -EINVAL;
1572		if (resource == RLIMIT_NOFILE &&
1573				new_rlim->rlim_max > sysctl_nr_open)
1574			return -EPERM;
1575	}
1576
1577	/* protect tsk->signal and tsk->sighand from disappearing */
1578	read_lock(&tasklist_lock);
1579	if (!tsk->sighand) {
1580		retval = -ESRCH;
1581		goto out;
1582	}
1583
1584	rlim = tsk->signal->rlim + resource;
1585	task_lock(tsk->group_leader);
1586	if (new_rlim) {
1587		/* Keep the capable check against init_user_ns until
1588		   cgroups can contain all limits */
1589		if (new_rlim->rlim_max > rlim->rlim_max &&
1590				!capable(CAP_SYS_RESOURCE))
1591			retval = -EPERM;
1592		if (!retval)
1593			retval = security_task_setrlimit(tsk, resource, new_rlim);
1594	}
1595	if (!retval) {
1596		if (old_rlim)
1597			*old_rlim = *rlim;
1598		if (new_rlim)
1599			*rlim = *new_rlim;
1600	}
1601	task_unlock(tsk->group_leader);
1602
1603	/*
1604	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1605	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1606	 * ignores the rlimit.
1607	 */
1608	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1609	     new_rlim->rlim_cur != RLIM_INFINITY &&
1610	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1611		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1612out:
1613	read_unlock(&tasklist_lock);
1614	return retval;
1615}
1616
1617/* rcu lock must be held */
1618static int check_prlimit_permission(struct task_struct *task,
1619				    unsigned int flags)
1620{
1621	const struct cred *cred = current_cred(), *tcred;
1622	bool id_match;
1623
1624	if (current == task)
1625		return 0;
1626
1627	tcred = __task_cred(task);
1628	id_match = (uid_eq(cred->uid, tcred->euid) &&
1629		    uid_eq(cred->uid, tcred->suid) &&
1630		    uid_eq(cred->uid, tcred->uid)  &&
1631		    gid_eq(cred->gid, tcred->egid) &&
1632		    gid_eq(cred->gid, tcred->sgid) &&
1633		    gid_eq(cred->gid, tcred->gid));
1634	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1635		return -EPERM;
1636
1637	return security_task_prlimit(cred, tcred, flags);
1638}
1639
1640SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1641		const struct rlimit64 __user *, new_rlim,
1642		struct rlimit64 __user *, old_rlim)
1643{
1644	struct rlimit64 old64, new64;
1645	struct rlimit old, new;
1646	struct task_struct *tsk;
1647	unsigned int checkflags = 0;
1648	int ret;
1649
1650	if (old_rlim)
1651		checkflags |= LSM_PRLIMIT_READ;
1652
1653	if (new_rlim) {
1654		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1655			return -EFAULT;
1656		rlim64_to_rlim(&new64, &new);
1657		checkflags |= LSM_PRLIMIT_WRITE;
1658	}
1659
1660	rcu_read_lock();
1661	tsk = pid ? find_task_by_vpid(pid) : current;
1662	if (!tsk) {
1663		rcu_read_unlock();
1664		return -ESRCH;
1665	}
1666	ret = check_prlimit_permission(tsk, checkflags);
1667	if (ret) {
1668		rcu_read_unlock();
1669		return ret;
1670	}
1671	get_task_struct(tsk);
1672	rcu_read_unlock();
1673
1674	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1675			old_rlim ? &old : NULL);
1676
1677	if (!ret && old_rlim) {
1678		rlim_to_rlim64(&old, &old64);
1679		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1680			ret = -EFAULT;
1681	}
1682
1683	put_task_struct(tsk);
1684	return ret;
1685}
1686
1687SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1688{
1689	struct rlimit new_rlim;
1690
1691	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1692		return -EFAULT;
1693	return do_prlimit(current, resource, &new_rlim, NULL);
1694}
1695
1696/*
1697 * It would make sense to put struct rusage in the task_struct,
1698 * except that would make the task_struct be *really big*.  After
1699 * task_struct gets moved into malloc'ed memory, it would
1700 * make sense to do this.  It will make moving the rest of the information
1701 * a lot simpler!  (Which we're not doing right now because we're not
1702 * measuring them yet).
1703 *
1704 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1705 * races with threads incrementing their own counters.  But since word
1706 * reads are atomic, we either get new values or old values and we don't
1707 * care which for the sums.  We always take the siglock to protect reading
1708 * the c* fields from p->signal from races with exit.c updating those
1709 * fields when reaping, so a sample either gets all the additions of a
1710 * given child after it's reaped, or none so this sample is before reaping.
1711 *
1712 * Locking:
1713 * We need to take the siglock for CHILDEREN, SELF and BOTH
1714 * for  the cases current multithreaded, non-current single threaded
1715 * non-current multithreaded.  Thread traversal is now safe with
1716 * the siglock held.
1717 * Strictly speaking, we donot need to take the siglock if we are current and
1718 * single threaded,  as no one else can take our signal_struct away, no one
1719 * else can  reap the  children to update signal->c* counters, and no one else
1720 * can race with the signal-> fields. If we do not take any lock, the
1721 * signal-> fields could be read out of order while another thread was just
1722 * exiting. So we should  place a read memory barrier when we avoid the lock.
1723 * On the writer side,  write memory barrier is implied in  __exit_signal
1724 * as __exit_signal releases  the siglock spinlock after updating the signal->
1725 * fields. But we don't do this yet to keep things simple.
1726 *
1727 */
1728
1729static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1730{
1731	r->ru_nvcsw += t->nvcsw;
1732	r->ru_nivcsw += t->nivcsw;
1733	r->ru_minflt += t->min_flt;
1734	r->ru_majflt += t->maj_flt;
1735	r->ru_inblock += task_io_get_inblock(t);
1736	r->ru_oublock += task_io_get_oublock(t);
1737}
1738
1739void getrusage(struct task_struct *p, int who, struct rusage *r)
1740{
1741	struct task_struct *t;
1742	unsigned long flags;
1743	u64 tgutime, tgstime, utime, stime;
1744	unsigned long maxrss = 0;
1745
1746	memset((char *)r, 0, sizeof (*r));
1747	utime = stime = 0;
1748
1749	if (who == RUSAGE_THREAD) {
1750		task_cputime_adjusted(current, &utime, &stime);
1751		accumulate_thread_rusage(p, r);
1752		maxrss = p->signal->maxrss;
1753		goto out;
1754	}
1755
1756	if (!lock_task_sighand(p, &flags))
1757		return;
1758
1759	switch (who) {
1760	case RUSAGE_BOTH:
1761	case RUSAGE_CHILDREN:
1762		utime = p->signal->cutime;
1763		stime = p->signal->cstime;
1764		r->ru_nvcsw = p->signal->cnvcsw;
1765		r->ru_nivcsw = p->signal->cnivcsw;
1766		r->ru_minflt = p->signal->cmin_flt;
1767		r->ru_majflt = p->signal->cmaj_flt;
1768		r->ru_inblock = p->signal->cinblock;
1769		r->ru_oublock = p->signal->coublock;
1770		maxrss = p->signal->cmaxrss;
1771
1772		if (who == RUSAGE_CHILDREN)
1773			break;
1774		fallthrough;
1775
1776	case RUSAGE_SELF:
1777		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1778		utime += tgutime;
1779		stime += tgstime;
1780		r->ru_nvcsw += p->signal->nvcsw;
1781		r->ru_nivcsw += p->signal->nivcsw;
1782		r->ru_minflt += p->signal->min_flt;
1783		r->ru_majflt += p->signal->maj_flt;
1784		r->ru_inblock += p->signal->inblock;
1785		r->ru_oublock += p->signal->oublock;
1786		if (maxrss < p->signal->maxrss)
1787			maxrss = p->signal->maxrss;
1788		t = p;
1789		do {
1790			accumulate_thread_rusage(t, r);
1791		} while_each_thread(p, t);
1792		break;
1793
1794	default:
1795		BUG();
1796	}
1797	unlock_task_sighand(p, &flags);
1798
1799out:
1800	r->ru_utime = ns_to_kernel_old_timeval(utime);
1801	r->ru_stime = ns_to_kernel_old_timeval(stime);
1802
1803	if (who != RUSAGE_CHILDREN) {
1804		struct mm_struct *mm = get_task_mm(p);
1805
1806		if (mm) {
1807			setmax_mm_hiwater_rss(&maxrss, mm);
1808			mmput(mm);
1809		}
1810	}
1811	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1812}
1813
1814SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1815{
1816	struct rusage r;
1817
1818	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1819	    who != RUSAGE_THREAD)
1820		return -EINVAL;
1821
1822	getrusage(current, who, &r);
1823	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1824}
1825
1826#ifdef CONFIG_COMPAT
1827COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1828{
1829	struct rusage r;
1830
1831	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1832	    who != RUSAGE_THREAD)
1833		return -EINVAL;
1834
1835	getrusage(current, who, &r);
1836	return put_compat_rusage(&r, ru);
1837}
1838#endif
1839
1840SYSCALL_DEFINE1(umask, int, mask)
1841{
1842	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1843	return mask;
1844}
1845
1846static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1847{
1848	struct fd exe;
1849	struct file *old_exe, *exe_file;
1850	struct inode *inode;
1851	int err;
1852
1853	exe = fdget(fd);
1854	if (!exe.file)
1855		return -EBADF;
1856
1857	inode = file_inode(exe.file);
1858
1859	/*
1860	 * Because the original mm->exe_file points to executable file, make
1861	 * sure that this one is executable as well, to avoid breaking an
1862	 * overall picture.
1863	 */
1864	err = -EACCES;
1865	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1866		goto exit;
1867
1868	err = file_permission(exe.file, MAY_EXEC);
1869	if (err)
1870		goto exit;
1871
1872	/*
1873	 * Forbid mm->exe_file change if old file still mapped.
1874	 */
1875	exe_file = get_mm_exe_file(mm);
1876	err = -EBUSY;
1877	if (exe_file) {
1878		struct vm_area_struct *vma;
1879
1880		mmap_read_lock(mm);
1881		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1882			if (!vma->vm_file)
1883				continue;
1884			if (path_equal(&vma->vm_file->f_path,
1885				       &exe_file->f_path))
1886				goto exit_err;
1887		}
1888
1889		mmap_read_unlock(mm);
1890		fput(exe_file);
1891	}
1892
1893	err = 0;
1894	/* set the new file, lockless */
1895	get_file(exe.file);
1896	old_exe = xchg(&mm->exe_file, exe.file);
1897	if (old_exe)
1898		fput(old_exe);
1899exit:
1900	fdput(exe);
1901	return err;
1902exit_err:
1903	mmap_read_unlock(mm);
1904	fput(exe_file);
1905	goto exit;
1906}
1907
1908/*
1909 * Check arithmetic relations of passed addresses.
1910 *
1911 * WARNING: we don't require any capability here so be very careful
1912 * in what is allowed for modification from userspace.
1913 */
1914static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1915{
1916	unsigned long mmap_max_addr = TASK_SIZE;
1917	int error = -EINVAL, i;
1918
1919	static const unsigned char offsets[] = {
1920		offsetof(struct prctl_mm_map, start_code),
1921		offsetof(struct prctl_mm_map, end_code),
1922		offsetof(struct prctl_mm_map, start_data),
1923		offsetof(struct prctl_mm_map, end_data),
1924		offsetof(struct prctl_mm_map, start_brk),
1925		offsetof(struct prctl_mm_map, brk),
1926		offsetof(struct prctl_mm_map, start_stack),
1927		offsetof(struct prctl_mm_map, arg_start),
1928		offsetof(struct prctl_mm_map, arg_end),
1929		offsetof(struct prctl_mm_map, env_start),
1930		offsetof(struct prctl_mm_map, env_end),
1931	};
1932
1933	/*
1934	 * Make sure the members are not somewhere outside
1935	 * of allowed address space.
1936	 */
1937	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1938		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1939
1940		if ((unsigned long)val >= mmap_max_addr ||
1941		    (unsigned long)val < mmap_min_addr)
1942			goto out;
1943	}
1944
1945	/*
1946	 * Make sure the pairs are ordered.
1947	 */
1948#define __prctl_check_order(__m1, __op, __m2)				\
1949	((unsigned long)prctl_map->__m1 __op				\
1950	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1951	error  = __prctl_check_order(start_code, <, end_code);
1952	error |= __prctl_check_order(start_data,<=, end_data);
1953	error |= __prctl_check_order(start_brk, <=, brk);
1954	error |= __prctl_check_order(arg_start, <=, arg_end);
1955	error |= __prctl_check_order(env_start, <=, env_end);
1956	if (error)
1957		goto out;
1958#undef __prctl_check_order
1959
1960	error = -EINVAL;
1961
1962	/*
 
 
 
 
 
 
 
1963	 * Neither we should allow to override limits if they set.
1964	 */
1965	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1966			      prctl_map->start_brk, prctl_map->end_data,
1967			      prctl_map->start_data))
1968			goto out;
1969
1970	error = 0;
1971out:
1972	return error;
1973}
1974
1975#ifdef CONFIG_CHECKPOINT_RESTORE
1976static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1977{
1978	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1979	unsigned long user_auxv[AT_VECTOR_SIZE];
1980	struct mm_struct *mm = current->mm;
1981	int error;
1982
1983	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1984	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1985
1986	if (opt == PR_SET_MM_MAP_SIZE)
1987		return put_user((unsigned int)sizeof(prctl_map),
1988				(unsigned int __user *)addr);
1989
1990	if (data_size != sizeof(prctl_map))
1991		return -EINVAL;
1992
1993	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1994		return -EFAULT;
1995
1996	error = validate_prctl_map_addr(&prctl_map);
1997	if (error)
1998		return error;
1999
2000	if (prctl_map.auxv_size) {
2001		/*
2002		 * Someone is trying to cheat the auxv vector.
2003		 */
2004		if (!prctl_map.auxv ||
2005				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2006			return -EINVAL;
2007
2008		memset(user_auxv, 0, sizeof(user_auxv));
2009		if (copy_from_user(user_auxv,
2010				   (const void __user *)prctl_map.auxv,
2011				   prctl_map.auxv_size))
2012			return -EFAULT;
2013
2014		/* Last entry must be AT_NULL as specification requires */
2015		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2016		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2017	}
2018
2019	if (prctl_map.exe_fd != (u32)-1) {
2020		/*
2021		 * Check if the current user is checkpoint/restore capable.
2022		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2023		 * or CAP_CHECKPOINT_RESTORE.
2024		 * Note that a user with access to ptrace can masquerade an
2025		 * arbitrary program as any executable, even setuid ones.
2026		 * This may have implications in the tomoyo subsystem.
2027		 */
2028		if (!checkpoint_restore_ns_capable(current_user_ns()))
2029			return -EPERM;
2030
2031		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2032		if (error)
2033			return error;
2034	}
2035
2036	/*
2037	 * arg_lock protects concurrent updates but we still need mmap_lock for
2038	 * read to exclude races with sys_brk.
2039	 */
2040	mmap_read_lock(mm);
2041
2042	/*
2043	 * We don't validate if these members are pointing to
2044	 * real present VMAs because application may have correspond
2045	 * VMAs already unmapped and kernel uses these members for statistics
2046	 * output in procfs mostly, except
2047	 *
2048	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2049	 *    for VMAs when updating these members so anything wrong written
2050	 *    here cause kernel to swear at userspace program but won't lead
2051	 *    to any problem in kernel itself
2052	 */
2053
2054	spin_lock(&mm->arg_lock);
2055	mm->start_code	= prctl_map.start_code;
2056	mm->end_code	= prctl_map.end_code;
2057	mm->start_data	= prctl_map.start_data;
2058	mm->end_data	= prctl_map.end_data;
2059	mm->start_brk	= prctl_map.start_brk;
2060	mm->brk		= prctl_map.brk;
2061	mm->start_stack	= prctl_map.start_stack;
2062	mm->arg_start	= prctl_map.arg_start;
2063	mm->arg_end	= prctl_map.arg_end;
2064	mm->env_start	= prctl_map.env_start;
2065	mm->env_end	= prctl_map.env_end;
2066	spin_unlock(&mm->arg_lock);
2067
2068	/*
2069	 * Note this update of @saved_auxv is lockless thus
2070	 * if someone reads this member in procfs while we're
2071	 * updating -- it may get partly updated results. It's
2072	 * known and acceptable trade off: we leave it as is to
2073	 * not introduce additional locks here making the kernel
2074	 * more complex.
2075	 */
2076	if (prctl_map.auxv_size)
2077		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2078
2079	mmap_read_unlock(mm);
2080	return 0;
2081}
2082#endif /* CONFIG_CHECKPOINT_RESTORE */
2083
2084static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2085			  unsigned long len)
2086{
2087	/*
2088	 * This doesn't move the auxiliary vector itself since it's pinned to
2089	 * mm_struct, but it permits filling the vector with new values.  It's
2090	 * up to the caller to provide sane values here, otherwise userspace
2091	 * tools which use this vector might be unhappy.
2092	 */
2093	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2094
2095	if (len > sizeof(user_auxv))
2096		return -EINVAL;
2097
2098	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2099		return -EFAULT;
2100
2101	/* Make sure the last entry is always AT_NULL */
2102	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2103	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2104
2105	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2106
2107	task_lock(current);
2108	memcpy(mm->saved_auxv, user_auxv, len);
2109	task_unlock(current);
2110
2111	return 0;
2112}
2113
2114static int prctl_set_mm(int opt, unsigned long addr,
2115			unsigned long arg4, unsigned long arg5)
2116{
2117	struct mm_struct *mm = current->mm;
2118	struct prctl_mm_map prctl_map = {
2119		.auxv = NULL,
2120		.auxv_size = 0,
2121		.exe_fd = -1,
2122	};
2123	struct vm_area_struct *vma;
2124	int error;
2125
2126	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2127			      opt != PR_SET_MM_MAP &&
2128			      opt != PR_SET_MM_MAP_SIZE)))
2129		return -EINVAL;
2130
2131#ifdef CONFIG_CHECKPOINT_RESTORE
2132	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2133		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2134#endif
2135
2136	if (!capable(CAP_SYS_RESOURCE))
2137		return -EPERM;
2138
2139	if (opt == PR_SET_MM_EXE_FILE)
2140		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2141
2142	if (opt == PR_SET_MM_AUXV)
2143		return prctl_set_auxv(mm, addr, arg4);
2144
2145	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2146		return -EINVAL;
2147
2148	error = -EINVAL;
2149
2150	/*
2151	 * arg_lock protects concurrent updates of arg boundaries, we need
2152	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2153	 * validation.
2154	 */
2155	mmap_read_lock(mm);
2156	vma = find_vma(mm, addr);
2157
2158	spin_lock(&mm->arg_lock);
2159	prctl_map.start_code	= mm->start_code;
2160	prctl_map.end_code	= mm->end_code;
2161	prctl_map.start_data	= mm->start_data;
2162	prctl_map.end_data	= mm->end_data;
2163	prctl_map.start_brk	= mm->start_brk;
2164	prctl_map.brk		= mm->brk;
2165	prctl_map.start_stack	= mm->start_stack;
2166	prctl_map.arg_start	= mm->arg_start;
2167	prctl_map.arg_end	= mm->arg_end;
2168	prctl_map.env_start	= mm->env_start;
2169	prctl_map.env_end	= mm->env_end;
2170
2171	switch (opt) {
2172	case PR_SET_MM_START_CODE:
2173		prctl_map.start_code = addr;
2174		break;
2175	case PR_SET_MM_END_CODE:
2176		prctl_map.end_code = addr;
2177		break;
2178	case PR_SET_MM_START_DATA:
2179		prctl_map.start_data = addr;
2180		break;
2181	case PR_SET_MM_END_DATA:
2182		prctl_map.end_data = addr;
2183		break;
2184	case PR_SET_MM_START_STACK:
2185		prctl_map.start_stack = addr;
2186		break;
2187	case PR_SET_MM_START_BRK:
2188		prctl_map.start_brk = addr;
2189		break;
2190	case PR_SET_MM_BRK:
2191		prctl_map.brk = addr;
2192		break;
2193	case PR_SET_MM_ARG_START:
2194		prctl_map.arg_start = addr;
2195		break;
2196	case PR_SET_MM_ARG_END:
2197		prctl_map.arg_end = addr;
2198		break;
2199	case PR_SET_MM_ENV_START:
2200		prctl_map.env_start = addr;
2201		break;
2202	case PR_SET_MM_ENV_END:
2203		prctl_map.env_end = addr;
2204		break;
2205	default:
2206		goto out;
2207	}
2208
2209	error = validate_prctl_map_addr(&prctl_map);
2210	if (error)
2211		goto out;
2212
2213	switch (opt) {
2214	/*
2215	 * If command line arguments and environment
2216	 * are placed somewhere else on stack, we can
2217	 * set them up here, ARG_START/END to setup
2218	 * command line arguments and ENV_START/END
2219	 * for environment.
2220	 */
2221	case PR_SET_MM_START_STACK:
2222	case PR_SET_MM_ARG_START:
2223	case PR_SET_MM_ARG_END:
2224	case PR_SET_MM_ENV_START:
2225	case PR_SET_MM_ENV_END:
2226		if (!vma) {
2227			error = -EFAULT;
2228			goto out;
2229		}
2230	}
2231
2232	mm->start_code	= prctl_map.start_code;
2233	mm->end_code	= prctl_map.end_code;
2234	mm->start_data	= prctl_map.start_data;
2235	mm->end_data	= prctl_map.end_data;
2236	mm->start_brk	= prctl_map.start_brk;
2237	mm->brk		= prctl_map.brk;
2238	mm->start_stack	= prctl_map.start_stack;
2239	mm->arg_start	= prctl_map.arg_start;
2240	mm->arg_end	= prctl_map.arg_end;
2241	mm->env_start	= prctl_map.env_start;
2242	mm->env_end	= prctl_map.env_end;
2243
2244	error = 0;
2245out:
2246	spin_unlock(&mm->arg_lock);
2247	mmap_read_unlock(mm);
2248	return error;
2249}
2250
2251#ifdef CONFIG_CHECKPOINT_RESTORE
2252static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2253{
2254	return put_user(me->clear_child_tid, tid_addr);
2255}
2256#else
2257static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2258{
2259	return -EINVAL;
2260}
2261#endif
2262
2263static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2264{
2265	/*
2266	 * If task has has_child_subreaper - all its descendants
2267	 * already have these flag too and new descendants will
2268	 * inherit it on fork, skip them.
2269	 *
2270	 * If we've found child_reaper - skip descendants in
2271	 * it's subtree as they will never get out pidns.
2272	 */
2273	if (p->signal->has_child_subreaper ||
2274	    is_child_reaper(task_pid(p)))
2275		return 0;
2276
2277	p->signal->has_child_subreaper = 1;
2278	return 1;
2279}
2280
2281int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2282{
2283	return -EINVAL;
2284}
2285
2286int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2287				    unsigned long ctrl)
2288{
2289	return -EINVAL;
2290}
2291
2292#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2293
2294SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2295		unsigned long, arg4, unsigned long, arg5)
2296{
2297	struct task_struct *me = current;
2298	unsigned char comm[sizeof(me->comm)];
2299	long error;
2300
2301	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2302	if (error != -ENOSYS)
2303		return error;
2304
2305	error = 0;
2306	switch (option) {
2307	case PR_SET_PDEATHSIG:
2308		if (!valid_signal(arg2)) {
2309			error = -EINVAL;
2310			break;
2311		}
2312		me->pdeath_signal = arg2;
2313		break;
2314	case PR_GET_PDEATHSIG:
2315		error = put_user(me->pdeath_signal, (int __user *)arg2);
2316		break;
2317	case PR_GET_DUMPABLE:
2318		error = get_dumpable(me->mm);
2319		break;
2320	case PR_SET_DUMPABLE:
2321		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2322			error = -EINVAL;
2323			break;
2324		}
2325		set_dumpable(me->mm, arg2);
2326		break;
2327
2328	case PR_SET_UNALIGN:
2329		error = SET_UNALIGN_CTL(me, arg2);
2330		break;
2331	case PR_GET_UNALIGN:
2332		error = GET_UNALIGN_CTL(me, arg2);
2333		break;
2334	case PR_SET_FPEMU:
2335		error = SET_FPEMU_CTL(me, arg2);
2336		break;
2337	case PR_GET_FPEMU:
2338		error = GET_FPEMU_CTL(me, arg2);
2339		break;
2340	case PR_SET_FPEXC:
2341		error = SET_FPEXC_CTL(me, arg2);
2342		break;
2343	case PR_GET_FPEXC:
2344		error = GET_FPEXC_CTL(me, arg2);
2345		break;
2346	case PR_GET_TIMING:
2347		error = PR_TIMING_STATISTICAL;
2348		break;
2349	case PR_SET_TIMING:
2350		if (arg2 != PR_TIMING_STATISTICAL)
2351			error = -EINVAL;
2352		break;
2353	case PR_SET_NAME:
2354		comm[sizeof(me->comm) - 1] = 0;
2355		if (strncpy_from_user(comm, (char __user *)arg2,
2356				      sizeof(me->comm) - 1) < 0)
2357			return -EFAULT;
2358		set_task_comm(me, comm);
2359		proc_comm_connector(me);
2360		break;
2361	case PR_GET_NAME:
2362		get_task_comm(comm, me);
2363		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2364			return -EFAULT;
2365		break;
2366	case PR_GET_ENDIAN:
2367		error = GET_ENDIAN(me, arg2);
2368		break;
2369	case PR_SET_ENDIAN:
2370		error = SET_ENDIAN(me, arg2);
2371		break;
2372	case PR_GET_SECCOMP:
2373		error = prctl_get_seccomp();
2374		break;
2375	case PR_SET_SECCOMP:
2376		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2377		break;
2378	case PR_GET_TSC:
2379		error = GET_TSC_CTL(arg2);
2380		break;
2381	case PR_SET_TSC:
2382		error = SET_TSC_CTL(arg2);
2383		break;
2384	case PR_TASK_PERF_EVENTS_DISABLE:
2385		error = perf_event_task_disable();
2386		break;
2387	case PR_TASK_PERF_EVENTS_ENABLE:
2388		error = perf_event_task_enable();
2389		break;
2390	case PR_GET_TIMERSLACK:
2391		if (current->timer_slack_ns > ULONG_MAX)
2392			error = ULONG_MAX;
2393		else
2394			error = current->timer_slack_ns;
2395		break;
2396	case PR_SET_TIMERSLACK:
2397		if (arg2 <= 0)
2398			current->timer_slack_ns =
2399					current->default_timer_slack_ns;
2400		else
2401			current->timer_slack_ns = arg2;
2402		break;
2403	case PR_MCE_KILL:
2404		if (arg4 | arg5)
2405			return -EINVAL;
2406		switch (arg2) {
2407		case PR_MCE_KILL_CLEAR:
2408			if (arg3 != 0)
2409				return -EINVAL;
2410			current->flags &= ~PF_MCE_PROCESS;
2411			break;
2412		case PR_MCE_KILL_SET:
2413			current->flags |= PF_MCE_PROCESS;
2414			if (arg3 == PR_MCE_KILL_EARLY)
2415				current->flags |= PF_MCE_EARLY;
2416			else if (arg3 == PR_MCE_KILL_LATE)
2417				current->flags &= ~PF_MCE_EARLY;
2418			else if (arg3 == PR_MCE_KILL_DEFAULT)
2419				current->flags &=
2420						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2421			else
2422				return -EINVAL;
2423			break;
2424		default:
2425			return -EINVAL;
2426		}
2427		break;
2428	case PR_MCE_KILL_GET:
2429		if (arg2 | arg3 | arg4 | arg5)
2430			return -EINVAL;
2431		if (current->flags & PF_MCE_PROCESS)
2432			error = (current->flags & PF_MCE_EARLY) ?
2433				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2434		else
2435			error = PR_MCE_KILL_DEFAULT;
2436		break;
2437	case PR_SET_MM:
2438		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2439		break;
2440	case PR_GET_TID_ADDRESS:
2441		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2442		break;
2443	case PR_SET_CHILD_SUBREAPER:
2444		me->signal->is_child_subreaper = !!arg2;
2445		if (!arg2)
2446			break;
2447
2448		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2449		break;
2450	case PR_GET_CHILD_SUBREAPER:
2451		error = put_user(me->signal->is_child_subreaper,
2452				 (int __user *)arg2);
2453		break;
2454	case PR_SET_NO_NEW_PRIVS:
2455		if (arg2 != 1 || arg3 || arg4 || arg5)
2456			return -EINVAL;
2457
2458		task_set_no_new_privs(current);
2459		break;
2460	case PR_GET_NO_NEW_PRIVS:
2461		if (arg2 || arg3 || arg4 || arg5)
2462			return -EINVAL;
2463		return task_no_new_privs(current) ? 1 : 0;
2464	case PR_GET_THP_DISABLE:
2465		if (arg2 || arg3 || arg4 || arg5)
2466			return -EINVAL;
2467		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2468		break;
2469	case PR_SET_THP_DISABLE:
2470		if (arg3 || arg4 || arg5)
2471			return -EINVAL;
2472		if (mmap_write_lock_killable(me->mm))
2473			return -EINTR;
2474		if (arg2)
2475			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2476		else
2477			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2478		mmap_write_unlock(me->mm);
2479		break;
2480	case PR_MPX_ENABLE_MANAGEMENT:
2481	case PR_MPX_DISABLE_MANAGEMENT:
2482		/* No longer implemented: */
2483		return -EINVAL;
2484	case PR_SET_FP_MODE:
2485		error = SET_FP_MODE(me, arg2);
2486		break;
2487	case PR_GET_FP_MODE:
2488		error = GET_FP_MODE(me);
2489		break;
2490	case PR_SVE_SET_VL:
2491		error = SVE_SET_VL(arg2);
2492		break;
2493	case PR_SVE_GET_VL:
2494		error = SVE_GET_VL();
2495		break;
2496	case PR_GET_SPECULATION_CTRL:
2497		if (arg3 || arg4 || arg5)
2498			return -EINVAL;
2499		error = arch_prctl_spec_ctrl_get(me, arg2);
2500		break;
2501	case PR_SET_SPECULATION_CTRL:
2502		if (arg4 || arg5)
2503			return -EINVAL;
2504		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2505		break;
2506	case PR_PAC_RESET_KEYS:
2507		if (arg3 || arg4 || arg5)
2508			return -EINVAL;
2509		error = PAC_RESET_KEYS(me, arg2);
2510		break;
2511	case PR_PAC_SET_ENABLED_KEYS:
2512		if (arg4 || arg5)
2513			return -EINVAL;
2514		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2515		break;
2516	case PR_PAC_GET_ENABLED_KEYS:
2517		if (arg2 || arg3 || arg4 || arg5)
2518			return -EINVAL;
2519		error = PAC_GET_ENABLED_KEYS(me);
2520		break;
2521	case PR_SET_TAGGED_ADDR_CTRL:
2522		if (arg3 || arg4 || arg5)
2523			return -EINVAL;
2524		error = SET_TAGGED_ADDR_CTRL(arg2);
2525		break;
2526	case PR_GET_TAGGED_ADDR_CTRL:
2527		if (arg2 || arg3 || arg4 || arg5)
2528			return -EINVAL;
2529		error = GET_TAGGED_ADDR_CTRL();
2530		break;
2531	case PR_SET_IO_FLUSHER:
2532		if (!capable(CAP_SYS_RESOURCE))
2533			return -EPERM;
2534
2535		if (arg3 || arg4 || arg5)
2536			return -EINVAL;
2537
2538		if (arg2 == 1)
2539			current->flags |= PR_IO_FLUSHER;
2540		else if (!arg2)
2541			current->flags &= ~PR_IO_FLUSHER;
2542		else
2543			return -EINVAL;
2544		break;
2545	case PR_GET_IO_FLUSHER:
2546		if (!capable(CAP_SYS_RESOURCE))
2547			return -EPERM;
2548
2549		if (arg2 || arg3 || arg4 || arg5)
2550			return -EINVAL;
2551
2552		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2553		break;
2554	case PR_SET_SYSCALL_USER_DISPATCH:
2555		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2556						  (char __user *) arg5);
2557		break;
2558#ifdef CONFIG_SCHED_CORE
2559	case PR_SCHED_CORE:
2560		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2561		break;
2562#endif
2563	default:
2564		error = -EINVAL;
2565		break;
2566	}
2567	return error;
2568}
2569
2570SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2571		struct getcpu_cache __user *, unused)
2572{
2573	int err = 0;
2574	int cpu = raw_smp_processor_id();
2575
2576	if (cpup)
2577		err |= put_user(cpu, cpup);
2578	if (nodep)
2579		err |= put_user(cpu_to_node(cpu), nodep);
2580	return err ? -EFAULT : 0;
2581}
2582
2583/**
2584 * do_sysinfo - fill in sysinfo struct
2585 * @info: pointer to buffer to fill
2586 */
2587static int do_sysinfo(struct sysinfo *info)
2588{
2589	unsigned long mem_total, sav_total;
2590	unsigned int mem_unit, bitcount;
2591	struct timespec64 tp;
2592
2593	memset(info, 0, sizeof(struct sysinfo));
2594
2595	ktime_get_boottime_ts64(&tp);
2596	timens_add_boottime(&tp);
2597	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2598
2599	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2600
2601	info->procs = nr_threads;
2602
2603	si_meminfo(info);
2604	si_swapinfo(info);
2605
2606	/*
2607	 * If the sum of all the available memory (i.e. ram + swap)
2608	 * is less than can be stored in a 32 bit unsigned long then
2609	 * we can be binary compatible with 2.2.x kernels.  If not,
2610	 * well, in that case 2.2.x was broken anyways...
2611	 *
2612	 *  -Erik Andersen <andersee@debian.org>
2613	 */
2614
2615	mem_total = info->totalram + info->totalswap;
2616	if (mem_total < info->totalram || mem_total < info->totalswap)
2617		goto out;
2618	bitcount = 0;
2619	mem_unit = info->mem_unit;
2620	while (mem_unit > 1) {
2621		bitcount++;
2622		mem_unit >>= 1;
2623		sav_total = mem_total;
2624		mem_total <<= 1;
2625		if (mem_total < sav_total)
2626			goto out;
2627	}
2628
2629	/*
2630	 * If mem_total did not overflow, multiply all memory values by
2631	 * info->mem_unit and set it to 1.  This leaves things compatible
2632	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2633	 * kernels...
2634	 */
2635
2636	info->mem_unit = 1;
2637	info->totalram <<= bitcount;
2638	info->freeram <<= bitcount;
2639	info->sharedram <<= bitcount;
2640	info->bufferram <<= bitcount;
2641	info->totalswap <<= bitcount;
2642	info->freeswap <<= bitcount;
2643	info->totalhigh <<= bitcount;
2644	info->freehigh <<= bitcount;
2645
2646out:
2647	return 0;
2648}
2649
2650SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2651{
2652	struct sysinfo val;
2653
2654	do_sysinfo(&val);
2655
2656	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2657		return -EFAULT;
2658
2659	return 0;
2660}
2661
2662#ifdef CONFIG_COMPAT
2663struct compat_sysinfo {
2664	s32 uptime;
2665	u32 loads[3];
2666	u32 totalram;
2667	u32 freeram;
2668	u32 sharedram;
2669	u32 bufferram;
2670	u32 totalswap;
2671	u32 freeswap;
2672	u16 procs;
2673	u16 pad;
2674	u32 totalhigh;
2675	u32 freehigh;
2676	u32 mem_unit;
2677	char _f[20-2*sizeof(u32)-sizeof(int)];
2678};
2679
2680COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2681{
2682	struct sysinfo s;
2683	struct compat_sysinfo s_32;
2684
2685	do_sysinfo(&s);
2686
2687	/* Check to see if any memory value is too large for 32-bit and scale
2688	 *  down if needed
2689	 */
2690	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2691		int bitcount = 0;
2692
2693		while (s.mem_unit < PAGE_SIZE) {
2694			s.mem_unit <<= 1;
2695			bitcount++;
2696		}
2697
2698		s.totalram >>= bitcount;
2699		s.freeram >>= bitcount;
2700		s.sharedram >>= bitcount;
2701		s.bufferram >>= bitcount;
2702		s.totalswap >>= bitcount;
2703		s.freeswap >>= bitcount;
2704		s.totalhigh >>= bitcount;
2705		s.freehigh >>= bitcount;
2706	}
2707
2708	memset(&s_32, 0, sizeof(s_32));
2709	s_32.uptime = s.uptime;
2710	s_32.loads[0] = s.loads[0];
2711	s_32.loads[1] = s.loads[1];
2712	s_32.loads[2] = s.loads[2];
2713	s_32.totalram = s.totalram;
2714	s_32.freeram = s.freeram;
2715	s_32.sharedram = s.sharedram;
2716	s_32.bufferram = s.bufferram;
2717	s_32.totalswap = s.totalswap;
2718	s_32.freeswap = s.freeswap;
2719	s_32.procs = s.procs;
2720	s_32.totalhigh = s.totalhigh;
2721	s_32.freehigh = s.freehigh;
2722	s_32.mem_unit = s.mem_unit;
2723	if (copy_to_user(info, &s_32, sizeof(s_32)))
2724		return -EFAULT;
2725	return 0;
2726}
2727#endif /* CONFIG_COMPAT */