sys.c

   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/mm_inline.h>
  11#include <linux/utsname.h>
  12#include <linux/mman.h>
  13#include <linux/reboot.h>
  14#include <linux/prctl.h>
  15#include <linux/highuid.h>
  16#include <linux/fs.h>
  17#include <linux/kmod.h>
  18#include <linux/ksm.h>
  19#include <linux/perf_event.h>
  20#include <linux/resource.h>
  21#include <linux/kernel.h>
  22#include <linux/workqueue.h>
  23#include <linux/capability.h>
  24#include <linux/device.h>
  25#include <linux/key.h>
  26#include <linux/times.h>
  27#include <linux/posix-timers.h>
  28#include <linux/security.h>
  29#include <linux/random.h>
  30#include <linux/suspend.h>
  31#include <linux/tty.h>
  32#include <linux/signal.h>
  33#include <linux/cn_proc.h>
  34#include <linux/getcpu.h>
  35#include <linux/task_io_accounting_ops.h>
  36#include <linux/seccomp.h>
  37#include <linux/cpu.h>
  38#include <linux/personality.h>
  39#include <linux/ptrace.h>
  40#include <linux/fs_struct.h>
  41#include <linux/file.h>
  42#include <linux/mount.h>
  43#include <linux/gfp.h>
  44#include <linux/syscore_ops.h>
  45#include <linux/version.h>
  46#include <linux/ctype.h>
  47#include <linux/syscall_user_dispatch.h>
  48
  49#include <linux/compat.h>
  50#include <linux/syscalls.h>
  51#include <linux/kprobes.h>
  52#include <linux/user_namespace.h>
  53#include <linux/time_namespace.h>
  54#include <linux/binfmts.h>
  55#include <linux/futex.h>
  56
  57#include <linux/sched.h>
  58#include <linux/sched/autogroup.h>
  59#include <linux/sched/loadavg.h>
  60#include <linux/sched/stat.h>
  61#include <linux/sched/mm.h>
  62#include <linux/sched/coredump.h>
  63#include <linux/sched/task.h>
  64#include <linux/sched/cputime.h>
  65#include <linux/rcupdate.h>
  66#include <linux/uidgid.h>
  67#include <linux/cred.h>
  68
  69#include <linux/nospec.h>
  70
  71#include <linux/kmsg_dump.h>
  72/* Move somewhere else to avoid recompiling? */
  73#include <generated/utsrelease.h>
  74
  75#include <linux/uaccess.h>
  76#include <asm/io.h>
  77#include <asm/unistd.h>
  78
  79#include <trace/events/task.h>
  80
  81#include "uid16.h"
  82
  83#ifndef SET_UNALIGN_CTL
  84# define SET_UNALIGN_CTL(a, b)	(-EINVAL)
  85#endif
  86#ifndef GET_UNALIGN_CTL
  87# define GET_UNALIGN_CTL(a, b)	(-EINVAL)
  88#endif
  89#ifndef SET_FPEMU_CTL
  90# define SET_FPEMU_CTL(a, b)	(-EINVAL)
  91#endif
  92#ifndef GET_FPEMU_CTL
  93# define GET_FPEMU_CTL(a, b)	(-EINVAL)
  94#endif
  95#ifndef SET_FPEXC_CTL
  96# define SET_FPEXC_CTL(a, b)	(-EINVAL)
  97#endif
  98#ifndef GET_FPEXC_CTL
  99# define GET_FPEXC_CTL(a, b)	(-EINVAL)
 100#endif
 101#ifndef GET_ENDIAN
 102# define GET_ENDIAN(a, b)	(-EINVAL)
 103#endif
 104#ifndef SET_ENDIAN
 105# define SET_ENDIAN(a, b)	(-EINVAL)
 106#endif
 107#ifndef GET_TSC_CTL
 108# define GET_TSC_CTL(a)		(-EINVAL)
 109#endif
 110#ifndef SET_TSC_CTL
 111# define SET_TSC_CTL(a)		(-EINVAL)
 112#endif
 113#ifndef GET_FP_MODE
 114# define GET_FP_MODE(a)		(-EINVAL)
 115#endif
 116#ifndef SET_FP_MODE
 117# define SET_FP_MODE(a,b)	(-EINVAL)
 118#endif
 119#ifndef SVE_SET_VL
 120# define SVE_SET_VL(a)		(-EINVAL)
 121#endif
 122#ifndef SVE_GET_VL
 123# define SVE_GET_VL()		(-EINVAL)
 124#endif
 125#ifndef SME_SET_VL
 126# define SME_SET_VL(a)		(-EINVAL)
 127#endif
 128#ifndef SME_GET_VL
 129# define SME_GET_VL()		(-EINVAL)
 130#endif
 131#ifndef PAC_RESET_KEYS
 132# define PAC_RESET_KEYS(a, b)	(-EINVAL)
 133#endif
 134#ifndef PAC_SET_ENABLED_KEYS
 135# define PAC_SET_ENABLED_KEYS(a, b, c)	(-EINVAL)
 136#endif
 137#ifndef PAC_GET_ENABLED_KEYS
 138# define PAC_GET_ENABLED_KEYS(a)	(-EINVAL)
 139#endif
 140#ifndef SET_TAGGED_ADDR_CTRL
 141# define SET_TAGGED_ADDR_CTRL(a)	(-EINVAL)
 142#endif
 143#ifndef GET_TAGGED_ADDR_CTRL
 144# define GET_TAGGED_ADDR_CTRL()		(-EINVAL)
 145#endif
 146#ifndef RISCV_V_SET_CONTROL
 147# define RISCV_V_SET_CONTROL(a)		(-EINVAL)
 148#endif
 149#ifndef RISCV_V_GET_CONTROL
 150# define RISCV_V_GET_CONTROL()		(-EINVAL)
 151#endif
 152#ifndef RISCV_SET_ICACHE_FLUSH_CTX
 153# define RISCV_SET_ICACHE_FLUSH_CTX(a, b)	(-EINVAL)
 154#endif
 155#ifndef PPC_GET_DEXCR_ASPECT
 156# define PPC_GET_DEXCR_ASPECT(a, b)	(-EINVAL)
 157#endif
 158#ifndef PPC_SET_DEXCR_ASPECT
 159# define PPC_SET_DEXCR_ASPECT(a, b, c)	(-EINVAL)
 160#endif
 161
 162/*
 163 * this is where the system-wide overflow UID and GID are defined, for
 164 * architectures that now have 32-bit UID/GID but didn't in the past
 165 */
 166
 167int overflowuid = DEFAULT_OVERFLOWUID;
 168int overflowgid = DEFAULT_OVERFLOWGID;
 169
 170EXPORT_SYMBOL(overflowuid);
 171EXPORT_SYMBOL(overflowgid);
 172
 173/*
 174 * the same as above, but for filesystems which can only store a 16-bit
 175 * UID and GID. as such, this is needed on all architectures
 176 */
 177
 178int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 179int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
 180
 181EXPORT_SYMBOL(fs_overflowuid);
 182EXPORT_SYMBOL(fs_overflowgid);
 183
 184static const struct ctl_table overflow_sysctl_table[] = {
 185	{
 186		.procname	= "overflowuid",
 187		.data		= &overflowuid,
 188		.maxlen		= sizeof(int),
 189		.mode		= 0644,
 190		.proc_handler	= proc_dointvec_minmax,
 191		.extra1		= SYSCTL_ZERO,
 192		.extra2		= SYSCTL_MAXOLDUID,
 193	},
 194	{
 195		.procname	= "overflowgid",
 196		.data		= &overflowgid,
 197		.maxlen		= sizeof(int),
 198		.mode		= 0644,
 199		.proc_handler	= proc_dointvec_minmax,
 200		.extra1		= SYSCTL_ZERO,
 201		.extra2		= SYSCTL_MAXOLDUID,
 202	},
 203};
 204
 205static int __init init_overflow_sysctl(void)
 206{
 207	register_sysctl_init("kernel", overflow_sysctl_table);
 208	return 0;
 209}
 210
 211postcore_initcall(init_overflow_sysctl);
 212
 213/*
 214 * Returns true if current's euid is same as p's uid or euid,
 215 * or has CAP_SYS_NICE to p's user_ns.
 216 *
 217 * Called with rcu_read_lock, creds are safe
 218 */
 219static bool set_one_prio_perm(struct task_struct *p)
 220{
 221	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 222
 223	if (uid_eq(pcred->uid,  cred->euid) ||
 224	    uid_eq(pcred->euid, cred->euid))
 225		return true;
 226	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
 227		return true;
 228	return false;
 229}
 230
 231/*
 232 * set the priority of a task
 233 * - the caller must hold the RCU read lock
 234 */
 235static int set_one_prio(struct task_struct *p, int niceval, int error)
 236{
 237	int no_nice;
 238
 239	if (!set_one_prio_perm(p)) {
 240		error = -EPERM;
 241		goto out;
 242	}
 243	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 244		error = -EACCES;
 245		goto out;
 246	}
 247	no_nice = security_task_setnice(p, niceval);
 248	if (no_nice) {
 249		error = no_nice;
 250		goto out;
 251	}
 252	if (error == -ESRCH)
 253		error = 0;
 254	set_user_nice(p, niceval);
 255out:
 256	return error;
 257}
 258
 259SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 260{
 261	struct task_struct *g, *p;
 262	struct user_struct *user;
 263	const struct cred *cred = current_cred();
 264	int error = -EINVAL;
 265	struct pid *pgrp;
 266	kuid_t uid;
 267
 268	if (which > PRIO_USER || which < PRIO_PROCESS)
 269		goto out;
 270
 271	/* normalize: avoid signed division (rounding problems) */
 272	error = -ESRCH;
 273	if (niceval < MIN_NICE)
 274		niceval = MIN_NICE;
 275	if (niceval > MAX_NICE)
 276		niceval = MAX_NICE;
 277
 278	rcu_read_lock();
 279	switch (which) {
 280	case PRIO_PROCESS:
 281		if (who)
 282			p = find_task_by_vpid(who);
 283		else
 284			p = current;
 285		if (p)
 286			error = set_one_prio(p, niceval, error);
 287		break;
 288	case PRIO_PGRP:
 289		if (who)
 290			pgrp = find_vpid(who);
 291		else
 292			pgrp = task_pgrp(current);
 293		read_lock(&tasklist_lock);
 294		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 295			error = set_one_prio(p, niceval, error);
 296		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 297		read_unlock(&tasklist_lock);
 298		break;
 299	case PRIO_USER:
 300		uid = make_kuid(cred->user_ns, who);
 301		user = cred->user;
 302		if (!who)
 303			uid = cred->uid;
 304		else if (!uid_eq(uid, cred->uid)) {
 305			user = find_user(uid);
 306			if (!user)
 307				goto out_unlock;	/* No processes for this user */
 308		}
 309		for_each_process_thread(g, p) {
 310			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
 311				error = set_one_prio(p, niceval, error);
 312		}
 313		if (!uid_eq(uid, cred->uid))
 314			free_uid(user);		/* For find_user() */
 315		break;
 316	}
 317out_unlock:
 318	rcu_read_unlock();
 319out:
 320	return error;
 321}
 322
 323/*
 324 * Ugh. To avoid negative return values, "getpriority()" will
 325 * not return the normal nice-value, but a negated value that
 326 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 327 * to stay compatible.
 328 */
 329SYSCALL_DEFINE2(getpriority, int, which, int, who)
 330{
 331	struct task_struct *g, *p;
 332	struct user_struct *user;
 333	const struct cred *cred = current_cred();
 334	long niceval, retval = -ESRCH;
 335	struct pid *pgrp;
 336	kuid_t uid;
 337
 338	if (which > PRIO_USER || which < PRIO_PROCESS)
 339		return -EINVAL;
 340
 341	rcu_read_lock();
 342	switch (which) {
 343	case PRIO_PROCESS:
 344		if (who)
 345			p = find_task_by_vpid(who);
 346		else
 347			p = current;
 348		if (p) {
 349			niceval = nice_to_rlimit(task_nice(p));
 350			if (niceval > retval)
 351				retval = niceval;
 352		}
 353		break;
 354	case PRIO_PGRP:
 355		if (who)
 356			pgrp = find_vpid(who);
 357		else
 358			pgrp = task_pgrp(current);
 359		read_lock(&tasklist_lock);
 360		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 361			niceval = nice_to_rlimit(task_nice(p));
 362			if (niceval > retval)
 363				retval = niceval;
 364		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 365		read_unlock(&tasklist_lock);
 366		break;
 367	case PRIO_USER:
 368		uid = make_kuid(cred->user_ns, who);
 369		user = cred->user;
 370		if (!who)
 371			uid = cred->uid;
 372		else if (!uid_eq(uid, cred->uid)) {
 373			user = find_user(uid);
 374			if (!user)
 375				goto out_unlock;	/* No processes for this user */
 376		}
 377		for_each_process_thread(g, p) {
 378			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
 379				niceval = nice_to_rlimit(task_nice(p));
 380				if (niceval > retval)
 381					retval = niceval;
 382			}
 383		}
 384		if (!uid_eq(uid, cred->uid))
 385			free_uid(user);		/* for find_user() */
 386		break;
 387	}
 388out_unlock:
 389	rcu_read_unlock();
 390
 391	return retval;
 392}
 393
 394/*
 395 * Unprivileged users may change the real gid to the effective gid
 396 * or vice versa.  (BSD-style)
 397 *
 398 * If you set the real gid at all, or set the effective gid to a value not
 399 * equal to the real gid, then the saved gid is set to the new effective gid.
 400 *
 401 * This makes it possible for a setgid program to completely drop its
 402 * privileges, which is often a useful assertion to make when you are doing
 403 * a security audit over a program.
 404 *
 405 * The general idea is that a program which uses just setregid() will be
 406 * 100% compatible with BSD.  A program which uses just setgid() will be
 407 * 100% compatible with POSIX with saved IDs.
 408 *
 409 * SMP: There are not races, the GIDs are checked only by filesystem
 410 *      operations (as far as semantic preservation is concerned).
 411 */
 412#ifdef CONFIG_MULTIUSER
 413long __sys_setregid(gid_t rgid, gid_t egid)
 414{
 415	struct user_namespace *ns = current_user_ns();
 416	const struct cred *old;
 417	struct cred *new;
 418	int retval;
 419	kgid_t krgid, kegid;
 420
 421	krgid = make_kgid(ns, rgid);
 422	kegid = make_kgid(ns, egid);
 423
 424	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 425		return -EINVAL;
 426	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 427		return -EINVAL;
 428
 429	new = prepare_creds();
 430	if (!new)
 431		return -ENOMEM;
 432	old = current_cred();
 433
 434	retval = -EPERM;
 435	if (rgid != (gid_t) -1) {
 436		if (gid_eq(old->gid, krgid) ||
 437		    gid_eq(old->egid, krgid) ||
 438		    ns_capable_setid(old->user_ns, CAP_SETGID))
 439			new->gid = krgid;
 440		else
 441			goto error;
 442	}
 443	if (egid != (gid_t) -1) {
 444		if (gid_eq(old->gid, kegid) ||
 445		    gid_eq(old->egid, kegid) ||
 446		    gid_eq(old->sgid, kegid) ||
 447		    ns_capable_setid(old->user_ns, CAP_SETGID))
 448			new->egid = kegid;
 449		else
 450			goto error;
 451	}
 452
 453	if (rgid != (gid_t) -1 ||
 454	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
 455		new->sgid = new->egid;
 456	new->fsgid = new->egid;
 457
 458	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
 459	if (retval < 0)
 460		goto error;
 461
 462	return commit_creds(new);
 463
 464error:
 465	abort_creds(new);
 466	return retval;
 467}
 468
 469SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 470{
 471	return __sys_setregid(rgid, egid);
 472}
 473
 474/*
 475 * setgid() is implemented like SysV w/ SAVED_IDS
 476 *
 477 * SMP: Same implicit races as above.
 478 */
 479long __sys_setgid(gid_t gid)
 480{
 481	struct user_namespace *ns = current_user_ns();
 482	const struct cred *old;
 483	struct cred *new;
 484	int retval;
 485	kgid_t kgid;
 486
 487	kgid = make_kgid(ns, gid);
 488	if (!gid_valid(kgid))
 489		return -EINVAL;
 490
 491	new = prepare_creds();
 492	if (!new)
 493		return -ENOMEM;
 494	old = current_cred();
 495
 496	retval = -EPERM;
 497	if (ns_capable_setid(old->user_ns, CAP_SETGID))
 498		new->gid = new->egid = new->sgid = new->fsgid = kgid;
 499	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
 500		new->egid = new->fsgid = kgid;
 501	else
 502		goto error;
 503
 504	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
 505	if (retval < 0)
 506		goto error;
 507
 508	return commit_creds(new);
 509
 510error:
 511	abort_creds(new);
 512	return retval;
 513}
 514
 515SYSCALL_DEFINE1(setgid, gid_t, gid)
 516{
 517	return __sys_setgid(gid);
 518}
 519
 520/*
 521 * change the user struct in a credentials set to match the new UID
 522 */
 523static int set_user(struct cred *new)
 524{
 525	struct user_struct *new_user;
 526
 527	new_user = alloc_uid(new->uid);
 528	if (!new_user)
 529		return -EAGAIN;
 530
 531	free_uid(new->user);
 532	new->user = new_user;
 533	return 0;
 534}
 535
 536static void flag_nproc_exceeded(struct cred *new)
 537{
 538	if (new->ucounts == current_ucounts())
 539		return;
 540
 541	/*
 542	 * We don't fail in case of NPROC limit excess here because too many
 543	 * poorly written programs don't check set*uid() return code, assuming
 544	 * it never fails if called by root.  We may still enforce NPROC limit
 545	 * for programs doing set*uid()+execve() by harmlessly deferring the
 546	 * failure to the execve() stage.
 547	 */
 548	if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
 549			new->user != INIT_USER)
 550		current->flags |= PF_NPROC_EXCEEDED;
 551	else
 552		current->flags &= ~PF_NPROC_EXCEEDED;
 553}
 554
 555/*
 556 * Unprivileged users may change the real uid to the effective uid
 557 * or vice versa.  (BSD-style)
 558 *
 559 * If you set the real uid at all, or set the effective uid to a value not
 560 * equal to the real uid, then the saved uid is set to the new effective uid.
 561 *
 562 * This makes it possible for a setuid program to completely drop its
 563 * privileges, which is often a useful assertion to make when you are doing
 564 * a security audit over a program.
 565 *
 566 * The general idea is that a program which uses just setreuid() will be
 567 * 100% compatible with BSD.  A program which uses just setuid() will be
 568 * 100% compatible with POSIX with saved IDs.
 569 */
 570long __sys_setreuid(uid_t ruid, uid_t euid)
 571{
 572	struct user_namespace *ns = current_user_ns();
 573	const struct cred *old;
 574	struct cred *new;
 575	int retval;
 576	kuid_t kruid, keuid;
 577
 578	kruid = make_kuid(ns, ruid);
 579	keuid = make_kuid(ns, euid);
 580
 581	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 582		return -EINVAL;
 583	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 584		return -EINVAL;
 585
 586	new = prepare_creds();
 587	if (!new)
 588		return -ENOMEM;
 589	old = current_cred();
 590
 591	retval = -EPERM;
 592	if (ruid != (uid_t) -1) {
 593		new->uid = kruid;
 594		if (!uid_eq(old->uid, kruid) &&
 595		    !uid_eq(old->euid, kruid) &&
 596		    !ns_capable_setid(old->user_ns, CAP_SETUID))
 597			goto error;
 598	}
 599
 600	if (euid != (uid_t) -1) {
 601		new->euid = keuid;
 602		if (!uid_eq(old->uid, keuid) &&
 603		    !uid_eq(old->euid, keuid) &&
 604		    !uid_eq(old->suid, keuid) &&
 605		    !ns_capable_setid(old->user_ns, CAP_SETUID))
 606			goto error;
 607	}
 608
 609	if (!uid_eq(new->uid, old->uid)) {
 610		retval = set_user(new);
 611		if (retval < 0)
 612			goto error;
 613	}
 614	if (ruid != (uid_t) -1 ||
 615	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
 616		new->suid = new->euid;
 617	new->fsuid = new->euid;
 618
 619	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 620	if (retval < 0)
 621		goto error;
 622
 623	retval = set_cred_ucounts(new);
 624	if (retval < 0)
 625		goto error;
 626
 627	flag_nproc_exceeded(new);
 628	return commit_creds(new);
 629
 630error:
 631	abort_creds(new);
 632	return retval;
 633}
 634
 635SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 636{
 637	return __sys_setreuid(ruid, euid);
 638}
 639
 640/*
 641 * setuid() is implemented like SysV with SAVED_IDS
 642 *
 643 * Note that SAVED_ID's is deficient in that a setuid root program
 644 * like sendmail, for example, cannot set its uid to be a normal
 645 * user and then switch back, because if you're root, setuid() sets
 646 * the saved uid too.  If you don't like this, blame the bright people
 647 * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 648 * will allow a root program to temporarily drop privileges and be able to
 649 * regain them by swapping the real and effective uid.
 650 */
 651long __sys_setuid(uid_t uid)
 652{
 653	struct user_namespace *ns = current_user_ns();
 654	const struct cred *old;
 655	struct cred *new;
 656	int retval;
 657	kuid_t kuid;
 658
 659	kuid = make_kuid(ns, uid);
 660	if (!uid_valid(kuid))
 661		return -EINVAL;
 662
 663	new = prepare_creds();
 664	if (!new)
 665		return -ENOMEM;
 666	old = current_cred();
 667
 668	retval = -EPERM;
 669	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
 670		new->suid = new->uid = kuid;
 671		if (!uid_eq(kuid, old->uid)) {
 672			retval = set_user(new);
 673			if (retval < 0)
 674				goto error;
 675		}
 676	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
 677		goto error;
 678	}
 679
 680	new->fsuid = new->euid = kuid;
 681
 682	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 683	if (retval < 0)
 684		goto error;
 685
 686	retval = set_cred_ucounts(new);
 687	if (retval < 0)
 688		goto error;
 689
 690	flag_nproc_exceeded(new);
 691	return commit_creds(new);
 692
 693error:
 694	abort_creds(new);
 695	return retval;
 696}
 697
 698SYSCALL_DEFINE1(setuid, uid_t, uid)
 699{
 700	return __sys_setuid(uid);
 701}
 702
 703
 704/*
 705 * This function implements a generic ability to update ruid, euid,
 706 * and suid.  This allows you to implement the 4.4 compatible seteuid().
 707 */
 708long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 709{
 710	struct user_namespace *ns = current_user_ns();
 711	const struct cred *old;
 712	struct cred *new;
 713	int retval;
 714	kuid_t kruid, keuid, ksuid;
 715	bool ruid_new, euid_new, suid_new;
 716
 717	kruid = make_kuid(ns, ruid);
 718	keuid = make_kuid(ns, euid);
 719	ksuid = make_kuid(ns, suid);
 720
 721	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 722		return -EINVAL;
 723
 724	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 725		return -EINVAL;
 726
 727	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
 728		return -EINVAL;
 729
 730	old = current_cred();
 731
 732	/* check for no-op */
 733	if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
 734	    (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
 735				    uid_eq(keuid, old->fsuid))) &&
 736	    (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
 737		return 0;
 738
 739	ruid_new = ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
 740		   !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
 741	euid_new = euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
 742		   !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
 743	suid_new = suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
 744		   !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
 745	if ((ruid_new || euid_new || suid_new) &&
 746	    !ns_capable_setid(old->user_ns, CAP_SETUID))
 747		return -EPERM;
 748
 749	new = prepare_creds();
 750	if (!new)
 751		return -ENOMEM;
 752
 753	if (ruid != (uid_t) -1) {
 754		new->uid = kruid;
 755		if (!uid_eq(kruid, old->uid)) {
 756			retval = set_user(new);
 757			if (retval < 0)
 758				goto error;
 759		}
 760	}
 761	if (euid != (uid_t) -1)
 762		new->euid = keuid;
 763	if (suid != (uid_t) -1)
 764		new->suid = ksuid;
 765	new->fsuid = new->euid;
 766
 767	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 768	if (retval < 0)
 769		goto error;
 770
 771	retval = set_cred_ucounts(new);
 772	if (retval < 0)
 773		goto error;
 774
 775	flag_nproc_exceeded(new);
 776	return commit_creds(new);
 777
 778error:
 779	abort_creds(new);
 780	return retval;
 781}
 782
 783SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 784{
 785	return __sys_setresuid(ruid, euid, suid);
 786}
 787
 788SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
 789{
 790	const struct cred *cred = current_cred();
 791	int retval;
 792	uid_t ruid, euid, suid;
 793
 794	ruid = from_kuid_munged(cred->user_ns, cred->uid);
 795	euid = from_kuid_munged(cred->user_ns, cred->euid);
 796	suid = from_kuid_munged(cred->user_ns, cred->suid);
 797
 798	retval = put_user(ruid, ruidp);
 799	if (!retval) {
 800		retval = put_user(euid, euidp);
 801		if (!retval)
 802			return put_user(suid, suidp);
 803	}
 804	return retval;
 805}
 806
 807/*
 808 * Same as above, but for rgid, egid, sgid.
 809 */
 810long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 811{
 812	struct user_namespace *ns = current_user_ns();
 813	const struct cred *old;
 814	struct cred *new;
 815	int retval;
 816	kgid_t krgid, kegid, ksgid;
 817	bool rgid_new, egid_new, sgid_new;
 818
 819	krgid = make_kgid(ns, rgid);
 820	kegid = make_kgid(ns, egid);
 821	ksgid = make_kgid(ns, sgid);
 822
 823	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 824		return -EINVAL;
 825	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 826		return -EINVAL;
 827	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
 828		return -EINVAL;
 829
 830	old = current_cred();
 831
 832	/* check for no-op */
 833	if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
 834	    (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
 835				    gid_eq(kegid, old->fsgid))) &&
 836	    (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
 837		return 0;
 838
 839	rgid_new = rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
 840		   !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
 841	egid_new = egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
 842		   !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
 843	sgid_new = sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
 844		   !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
 845	if ((rgid_new || egid_new || sgid_new) &&
 846	    !ns_capable_setid(old->user_ns, CAP_SETGID))
 847		return -EPERM;
 848
 849	new = prepare_creds();
 850	if (!new)
 851		return -ENOMEM;
 852
 853	if (rgid != (gid_t) -1)
 854		new->gid = krgid;
 855	if (egid != (gid_t) -1)
 856		new->egid = kegid;
 857	if (sgid != (gid_t) -1)
 858		new->sgid = ksgid;
 859	new->fsgid = new->egid;
 860
 861	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
 862	if (retval < 0)
 863		goto error;
 864
 865	return commit_creds(new);
 866
 867error:
 868	abort_creds(new);
 869	return retval;
 870}
 871
 872SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 873{
 874	return __sys_setresgid(rgid, egid, sgid);
 875}
 876
 877SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
 878{
 879	const struct cred *cred = current_cred();
 880	int retval;
 881	gid_t rgid, egid, sgid;
 882
 883	rgid = from_kgid_munged(cred->user_ns, cred->gid);
 884	egid = from_kgid_munged(cred->user_ns, cred->egid);
 885	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
 886
 887	retval = put_user(rgid, rgidp);
 888	if (!retval) {
 889		retval = put_user(egid, egidp);
 890		if (!retval)
 891			retval = put_user(sgid, sgidp);
 892	}
 893
 894	return retval;
 895}
 896
 897
 898/*
 899 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 900 * is used for "access()" and for the NFS daemon (letting nfsd stay at
 901 * whatever uid it wants to). It normally shadows "euid", except when
 902 * explicitly set by setfsuid() or for access..
 903 */
 904long __sys_setfsuid(uid_t uid)
 905{
 906	const struct cred *old;
 907	struct cred *new;
 908	uid_t old_fsuid;
 909	kuid_t kuid;
 910
 911	old = current_cred();
 912	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
 913
 914	kuid = make_kuid(old->user_ns, uid);
 915	if (!uid_valid(kuid))
 916		return old_fsuid;
 917
 918	new = prepare_creds();
 919	if (!new)
 920		return old_fsuid;
 921
 922	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
 923	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
 924	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
 925		if (!uid_eq(kuid, old->fsuid)) {
 926			new->fsuid = kuid;
 927			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 928				goto change_okay;
 929		}
 930	}
 931
 932	abort_creds(new);
 933	return old_fsuid;
 934
 935change_okay:
 936	commit_creds(new);
 937	return old_fsuid;
 938}
 939
 940SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 941{
 942	return __sys_setfsuid(uid);
 943}
 944
 945/*
 946 * Samma på svenska..
 947 */
 948long __sys_setfsgid(gid_t gid)
 949{
 950	const struct cred *old;
 951	struct cred *new;
 952	gid_t old_fsgid;
 953	kgid_t kgid;
 954
 955	old = current_cred();
 956	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
 957
 958	kgid = make_kgid(old->user_ns, gid);
 959	if (!gid_valid(kgid))
 960		return old_fsgid;
 961
 962	new = prepare_creds();
 963	if (!new)
 964		return old_fsgid;
 965
 966	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
 967	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
 968	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
 969		if (!gid_eq(kgid, old->fsgid)) {
 970			new->fsgid = kgid;
 971			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
 972				goto change_okay;
 973		}
 974	}
 975
 976	abort_creds(new);
 977	return old_fsgid;
 978
 979change_okay:
 980	commit_creds(new);
 981	return old_fsgid;
 982}
 983
 984SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 985{
 986	return __sys_setfsgid(gid);
 987}
 988#endif /* CONFIG_MULTIUSER */
 989
 990/**
 991 * sys_getpid - return the thread group id of the current process
 992 *
 993 * Note, despite the name, this returns the tgid not the pid.  The tgid and
 994 * the pid are identical unless CLONE_THREAD was specified on clone() in
 995 * which case the tgid is the same in all threads of the same group.
 996 *
 997 * This is SMP safe as current->tgid does not change.
 998 */
 999SYSCALL_DEFINE0(getpid)
1000{
1001	return task_tgid_vnr(current);
1002}
1003
1004/* Thread ID - the internal kernel "pid" */
1005SYSCALL_DEFINE0(gettid)
1006{
1007	return task_pid_vnr(current);
1008}
1009
1010/*
1011 * Accessing ->real_parent is not SMP-safe, it could
1012 * change from under us. However, we can use a stale
1013 * value of ->real_parent under rcu_read_lock(), see
1014 * release_task()->call_rcu(delayed_put_task_struct).
1015 */
1016SYSCALL_DEFINE0(getppid)
1017{
1018	int pid;
1019
1020	rcu_read_lock();
1021	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1022	rcu_read_unlock();
1023
1024	return pid;
1025}
1026
1027SYSCALL_DEFINE0(getuid)
1028{
1029	/* Only we change this so SMP safe */
1030	return from_kuid_munged(current_user_ns(), current_uid());
1031}
1032
1033SYSCALL_DEFINE0(geteuid)
1034{
1035	/* Only we change this so SMP safe */
1036	return from_kuid_munged(current_user_ns(), current_euid());
1037}
1038
1039SYSCALL_DEFINE0(getgid)
1040{
1041	/* Only we change this so SMP safe */
1042	return from_kgid_munged(current_user_ns(), current_gid());
1043}
1044
1045SYSCALL_DEFINE0(getegid)
1046{
1047	/* Only we change this so SMP safe */
1048	return from_kgid_munged(current_user_ns(), current_egid());
1049}
1050
1051static void do_sys_times(struct tms *tms)
1052{
1053	u64 tgutime, tgstime, cutime, cstime;
1054
1055	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1056	cutime = current->signal->cutime;
1057	cstime = current->signal->cstime;
1058	tms->tms_utime = nsec_to_clock_t(tgutime);
1059	tms->tms_stime = nsec_to_clock_t(tgstime);
1060	tms->tms_cutime = nsec_to_clock_t(cutime);
1061	tms->tms_cstime = nsec_to_clock_t(cstime);
1062}
1063
1064SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1065{
1066	if (tbuf) {
1067		struct tms tmp;
1068
1069		do_sys_times(&tmp);
1070		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1071			return -EFAULT;
1072	}
1073	force_successful_syscall_return();
1074	return (long) jiffies_64_to_clock_t(get_jiffies_64());
1075}
1076
1077#ifdef CONFIG_COMPAT
1078static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1079{
1080	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1081}
1082
1083COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1084{
1085	if (tbuf) {
1086		struct tms tms;
1087		struct compat_tms tmp;
1088
1089		do_sys_times(&tms);
1090		/* Convert our struct tms to the compat version. */
1091		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1092		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1093		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1094		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1095		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1096			return -EFAULT;
1097	}
1098	force_successful_syscall_return();
1099	return compat_jiffies_to_clock_t(jiffies);
1100}
1101#endif
1102
1103/*
1104 * This needs some heavy checking ...
1105 * I just haven't the stomach for it. I also don't fully
1106 * understand sessions/pgrp etc. Let somebody who does explain it.
1107 *
1108 * OK, I think I have the protection semantics right.... this is really
1109 * only important on a multi-user system anyway, to make sure one user
1110 * can't send a signal to a process owned by another.  -TYT, 12/12/91
1111 *
1112 * !PF_FORKNOEXEC check to conform completely to POSIX.
1113 */
1114SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1115{
1116	struct task_struct *p;
1117	struct task_struct *group_leader = current->group_leader;
1118	struct pid *pids[PIDTYPE_MAX] = { 0 };
1119	struct pid *pgrp;
1120	int err;
1121
1122	if (!pid)
1123		pid = task_pid_vnr(group_leader);
1124	if (!pgid)
1125		pgid = pid;
1126	if (pgid < 0)
1127		return -EINVAL;
1128	rcu_read_lock();
1129
1130	/* From this point forward we keep holding onto the tasklist lock
1131	 * so that our parent does not change from under us. -DaveM
1132	 */
1133	write_lock_irq(&tasklist_lock);
1134
1135	err = -ESRCH;
1136	p = find_task_by_vpid(pid);
1137	if (!p)
1138		goto out;
1139
1140	err = -EINVAL;
1141	if (!thread_group_leader(p))
1142		goto out;
1143
1144	if (same_thread_group(p->real_parent, group_leader)) {
1145		err = -EPERM;
1146		if (task_session(p) != task_session(group_leader))
1147			goto out;
1148		err = -EACCES;
1149		if (!(p->flags & PF_FORKNOEXEC))
1150			goto out;
1151	} else {
1152		err = -ESRCH;
1153		if (p != group_leader)
1154			goto out;
1155	}
1156
1157	err = -EPERM;
1158	if (p->signal->leader)
1159		goto out;
1160
1161	pgrp = task_pid(p);
1162	if (pgid != pid) {
1163		struct task_struct *g;
1164
1165		pgrp = find_vpid(pgid);
1166		g = pid_task(pgrp, PIDTYPE_PGID);
1167		if (!g || task_session(g) != task_session(group_leader))
1168			goto out;
1169	}
1170
1171	err = security_task_setpgid(p, pgid);
1172	if (err)
1173		goto out;
1174
1175	if (task_pgrp(p) != pgrp)
1176		change_pid(pids, p, PIDTYPE_PGID, pgrp);
1177
1178	err = 0;
1179out:
1180	/* All paths lead to here, thus we are safe. -DaveM */
1181	write_unlock_irq(&tasklist_lock);
1182	rcu_read_unlock();
1183	free_pids(pids);
1184	return err;
1185}
1186
1187static int do_getpgid(pid_t pid)
1188{
1189	struct task_struct *p;
1190	struct pid *grp;
1191	int retval;
1192
1193	rcu_read_lock();
1194	if (!pid)
1195		grp = task_pgrp(current);
1196	else {
1197		retval = -ESRCH;
1198		p = find_task_by_vpid(pid);
1199		if (!p)
1200			goto out;
1201		grp = task_pgrp(p);
1202		if (!grp)
1203			goto out;
1204
1205		retval = security_task_getpgid(p);
1206		if (retval)
1207			goto out;
1208	}
1209	retval = pid_vnr(grp);
1210out:
1211	rcu_read_unlock();
1212	return retval;
1213}
1214
1215SYSCALL_DEFINE1(getpgid, pid_t, pid)
1216{
1217	return do_getpgid(pid);
1218}
1219
1220#ifdef __ARCH_WANT_SYS_GETPGRP
1221
1222SYSCALL_DEFINE0(getpgrp)
1223{
1224	return do_getpgid(0);
1225}
1226
1227#endif
1228
1229SYSCALL_DEFINE1(getsid, pid_t, pid)
1230{
1231	struct task_struct *p;
1232	struct pid *sid;
1233	int retval;
1234
1235	rcu_read_lock();
1236	if (!pid)
1237		sid = task_session(current);
1238	else {
1239		retval = -ESRCH;
1240		p = find_task_by_vpid(pid);
1241		if (!p)
1242			goto out;
1243		sid = task_session(p);
1244		if (!sid)
1245			goto out;
1246
1247		retval = security_task_getsid(p);
1248		if (retval)
1249			goto out;
1250	}
1251	retval = pid_vnr(sid);
1252out:
1253	rcu_read_unlock();
1254	return retval;
1255}
1256
1257static void set_special_pids(struct pid **pids, struct pid *pid)
1258{
1259	struct task_struct *curr = current->group_leader;
1260
1261	if (task_session(curr) != pid)
1262		change_pid(pids, curr, PIDTYPE_SID, pid);
1263
1264	if (task_pgrp(curr) != pid)
1265		change_pid(pids, curr, PIDTYPE_PGID, pid);
1266}
1267
1268int ksys_setsid(void)
1269{
1270	struct task_struct *group_leader = current->group_leader;
1271	struct pid *sid = task_pid(group_leader);
1272	struct pid *pids[PIDTYPE_MAX] = { 0 };
1273	pid_t session = pid_vnr(sid);
1274	int err = -EPERM;
1275
1276	write_lock_irq(&tasklist_lock);
1277	/* Fail if I am already a session leader */
1278	if (group_leader->signal->leader)
1279		goto out;
1280
1281	/* Fail if a process group id already exists that equals the
1282	 * proposed session id.
1283	 */
1284	if (pid_task(sid, PIDTYPE_PGID))
1285		goto out;
1286
1287	group_leader->signal->leader = 1;
1288	set_special_pids(pids, sid);
1289
1290	proc_clear_tty(group_leader);
1291
1292	err = session;
1293out:
1294	write_unlock_irq(&tasklist_lock);
1295	free_pids(pids);
1296	if (err > 0) {
1297		proc_sid_connector(group_leader);
1298		sched_autogroup_create_attach(group_leader);
1299	}
1300	return err;
1301}
1302
1303SYSCALL_DEFINE0(setsid)
1304{
1305	return ksys_setsid();
1306}
1307
1308DECLARE_RWSEM(uts_sem);
1309
1310#ifdef COMPAT_UTS_MACHINE
1311#define override_architecture(name) \
1312	(personality(current->personality) == PER_LINUX32 && \
1313	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1314		      sizeof(COMPAT_UTS_MACHINE)))
1315#else
1316#define override_architecture(name)	0
1317#endif
1318
1319/*
1320 * Work around broken programs that cannot handle "Linux 3.0".
1321 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1322 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1323 * 2.6.60.
1324 */
1325static int override_release(char __user *release, size_t len)
1326{
1327	int ret = 0;
1328
1329	if (current->personality & UNAME26) {
1330		const char *rest = UTS_RELEASE;
1331		char buf[65] = { 0 };
1332		int ndots = 0;
1333		unsigned v;
1334		size_t copy;
1335
1336		while (*rest) {
1337			if (*rest == '.' && ++ndots >= 3)
1338				break;
1339			if (!isdigit(*rest) && *rest != '.')
1340				break;
1341			rest++;
1342		}
1343		v = LINUX_VERSION_PATCHLEVEL + 60;
1344		copy = clamp_t(size_t, len, 1, sizeof(buf));
1345		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1346		ret = copy_to_user(release, buf, copy + 1);
1347	}
1348	return ret;
1349}
1350
1351SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1352{
1353	struct new_utsname tmp;
1354
1355	down_read(&uts_sem);
1356	memcpy(&tmp, utsname(), sizeof(tmp));
1357	up_read(&uts_sem);
1358	if (copy_to_user(name, &tmp, sizeof(tmp)))
1359		return -EFAULT;
1360
1361	if (override_release(name->release, sizeof(name->release)))
1362		return -EFAULT;
1363	if (override_architecture(name))
1364		return -EFAULT;
1365	return 0;
1366}
1367
1368#ifdef __ARCH_WANT_SYS_OLD_UNAME
1369/*
1370 * Old cruft
1371 */
1372SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1373{
1374	struct old_utsname tmp;
1375
1376	if (!name)
1377		return -EFAULT;
1378
1379	down_read(&uts_sem);
1380	memcpy(&tmp, utsname(), sizeof(tmp));
1381	up_read(&uts_sem);
1382	if (copy_to_user(name, &tmp, sizeof(tmp)))
1383		return -EFAULT;
1384
1385	if (override_release(name->release, sizeof(name->release)))
1386		return -EFAULT;
1387	if (override_architecture(name))
1388		return -EFAULT;
1389	return 0;
1390}
1391
1392SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1393{
1394	struct oldold_utsname tmp;
1395
1396	if (!name)
1397		return -EFAULT;
1398
1399	memset(&tmp, 0, sizeof(tmp));
1400
1401	down_read(&uts_sem);
1402	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1403	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1404	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1405	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1406	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1407	up_read(&uts_sem);
1408	if (copy_to_user(name, &tmp, sizeof(tmp)))
1409		return -EFAULT;
1410
1411	if (override_architecture(name))
1412		return -EFAULT;
1413	if (override_release(name->release, sizeof(name->release)))
1414		return -EFAULT;
1415	return 0;
1416}
1417#endif
1418
1419SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1420{
1421	int errno;
1422	char tmp[__NEW_UTS_LEN];
1423
1424	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1425		return -EPERM;
1426
1427	if (len < 0 || len > __NEW_UTS_LEN)
1428		return -EINVAL;
1429	errno = -EFAULT;
1430	if (!copy_from_user(tmp, name, len)) {
1431		struct new_utsname *u;
1432
1433		add_device_randomness(tmp, len);
1434		down_write(&uts_sem);
1435		u = utsname();
1436		memcpy(u->nodename, tmp, len);
1437		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1438		errno = 0;
1439		uts_proc_notify(UTS_PROC_HOSTNAME);
1440		up_write(&uts_sem);
1441	}
1442	return errno;
1443}
1444
1445#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1446
1447SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1448{
1449	int i;
1450	struct new_utsname *u;
1451	char tmp[__NEW_UTS_LEN + 1];
1452
1453	if (len < 0)
1454		return -EINVAL;
1455	down_read(&uts_sem);
1456	u = utsname();
1457	i = 1 + strlen(u->nodename);
1458	if (i > len)
1459		i = len;
1460	memcpy(tmp, u->nodename, i);
1461	up_read(&uts_sem);
1462	if (copy_to_user(name, tmp, i))
1463		return -EFAULT;
1464	return 0;
1465}
1466
1467#endif
1468
1469/*
1470 * Only setdomainname; getdomainname can be implemented by calling
1471 * uname()
1472 */
1473SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1474{
1475	int errno;
1476	char tmp[__NEW_UTS_LEN];
1477
1478	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1479		return -EPERM;
1480	if (len < 0 || len > __NEW_UTS_LEN)
1481		return -EINVAL;
1482
1483	errno = -EFAULT;
1484	if (!copy_from_user(tmp, name, len)) {
1485		struct new_utsname *u;
1486
1487		add_device_randomness(tmp, len);
1488		down_write(&uts_sem);
1489		u = utsname();
1490		memcpy(u->domainname, tmp, len);
1491		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1492		errno = 0;
1493		uts_proc_notify(UTS_PROC_DOMAINNAME);
1494		up_write(&uts_sem);
1495	}
1496	return errno;
1497}
1498
1499/* make sure you are allowed to change @tsk limits before calling this */
1500static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1501		      struct rlimit *new_rlim, struct rlimit *old_rlim)
1502{
1503	struct rlimit *rlim;
1504	int retval = 0;
1505
1506	if (resource >= RLIM_NLIMITS)
1507		return -EINVAL;
1508	resource = array_index_nospec(resource, RLIM_NLIMITS);
1509
1510	if (new_rlim) {
1511		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1512			return -EINVAL;
1513		if (resource == RLIMIT_NOFILE &&
1514				new_rlim->rlim_max > sysctl_nr_open)
1515			return -EPERM;
1516	}
1517
1518	/* Holding a refcount on tsk protects tsk->signal from disappearing. */
1519	rlim = tsk->signal->rlim + resource;
1520	task_lock(tsk->group_leader);
1521	if (new_rlim) {
1522		/*
1523		 * Keep the capable check against init_user_ns until cgroups can
1524		 * contain all limits.
1525		 */
1526		if (new_rlim->rlim_max > rlim->rlim_max &&
1527				!capable(CAP_SYS_RESOURCE))
1528			retval = -EPERM;
1529		if (!retval)
1530			retval = security_task_setrlimit(tsk, resource, new_rlim);
1531	}
1532	if (!retval) {
1533		if (old_rlim)
1534			*old_rlim = *rlim;
1535		if (new_rlim)
1536			*rlim = *new_rlim;
1537	}
1538	task_unlock(tsk->group_leader);
1539
1540	/*
1541	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1542	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1543	 * ignores the rlimit.
1544	 */
1545	if (!retval && new_rlim && resource == RLIMIT_CPU &&
1546	    new_rlim->rlim_cur != RLIM_INFINITY &&
1547	    IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1548		/*
1549		 * update_rlimit_cpu can fail if the task is exiting, but there
1550		 * may be other tasks in the thread group that are not exiting,
1551		 * and they need their cpu timers adjusted.
1552		 *
1553		 * The group_leader is the last task to be released, so if we
1554		 * cannot update_rlimit_cpu on it, then the entire process is
1555		 * exiting and we do not need to update at all.
1556		 */
1557		update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1558	}
1559
1560	return retval;
1561}
1562
1563SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1564{
1565	struct rlimit value;
1566	int ret;
1567
1568	ret = do_prlimit(current, resource, NULL, &value);
1569	if (!ret)
1570		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1571
1572	return ret;
1573}
1574
1575#ifdef CONFIG_COMPAT
1576
1577COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1578		       struct compat_rlimit __user *, rlim)
1579{
1580	struct rlimit r;
1581	struct compat_rlimit r32;
1582
1583	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1584		return -EFAULT;
1585
1586	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1587		r.rlim_cur = RLIM_INFINITY;
1588	else
1589		r.rlim_cur = r32.rlim_cur;
1590	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1591		r.rlim_max = RLIM_INFINITY;
1592	else
1593		r.rlim_max = r32.rlim_max;
1594	return do_prlimit(current, resource, &r, NULL);
1595}
1596
1597COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1598		       struct compat_rlimit __user *, rlim)
1599{
1600	struct rlimit r;
1601	int ret;
1602
1603	ret = do_prlimit(current, resource, NULL, &r);
1604	if (!ret) {
1605		struct compat_rlimit r32;
1606		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1607			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1608		else
1609			r32.rlim_cur = r.rlim_cur;
1610		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1611			r32.rlim_max = COMPAT_RLIM_INFINITY;
1612		else
1613			r32.rlim_max = r.rlim_max;
1614
1615		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1616			return -EFAULT;
1617	}
1618	return ret;
1619}
1620
1621#endif
1622
1623#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1624
1625/*
1626 *	Back compatibility for getrlimit. Needed for some apps.
1627 */
1628SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1629		struct rlimit __user *, rlim)
1630{
1631	struct rlimit x;
1632	if (resource >= RLIM_NLIMITS)
1633		return -EINVAL;
1634
1635	resource = array_index_nospec(resource, RLIM_NLIMITS);
1636	task_lock(current->group_leader);
1637	x = current->signal->rlim[resource];
1638	task_unlock(current->group_leader);
1639	if (x.rlim_cur > 0x7FFFFFFF)
1640		x.rlim_cur = 0x7FFFFFFF;
1641	if (x.rlim_max > 0x7FFFFFFF)
1642		x.rlim_max = 0x7FFFFFFF;
1643	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1644}
1645
1646#ifdef CONFIG_COMPAT
1647COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1648		       struct compat_rlimit __user *, rlim)
1649{
1650	struct rlimit r;
1651
1652	if (resource >= RLIM_NLIMITS)
1653		return -EINVAL;
1654
1655	resource = array_index_nospec(resource, RLIM_NLIMITS);
1656	task_lock(current->group_leader);
1657	r = current->signal->rlim[resource];
1658	task_unlock(current->group_leader);
1659	if (r.rlim_cur > 0x7FFFFFFF)
1660		r.rlim_cur = 0x7FFFFFFF;
1661	if (r.rlim_max > 0x7FFFFFFF)
1662		r.rlim_max = 0x7FFFFFFF;
1663
1664	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1665	    put_user(r.rlim_max, &rlim->rlim_max))
1666		return -EFAULT;
1667	return 0;
1668}
1669#endif
1670
1671#endif
1672
1673static inline bool rlim64_is_infinity(__u64 rlim64)
1674{
1675#if BITS_PER_LONG < 64
1676	return rlim64 >= ULONG_MAX;
1677#else
1678	return rlim64 == RLIM64_INFINITY;
1679#endif
1680}
1681
1682static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1683{
1684	if (rlim->rlim_cur == RLIM_INFINITY)
1685		rlim64->rlim_cur = RLIM64_INFINITY;
1686	else
1687		rlim64->rlim_cur = rlim->rlim_cur;
1688	if (rlim->rlim_max == RLIM_INFINITY)
1689		rlim64->rlim_max = RLIM64_INFINITY;
1690	else
1691		rlim64->rlim_max = rlim->rlim_max;
1692}
1693
1694static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1695{
1696	if (rlim64_is_infinity(rlim64->rlim_cur))
1697		rlim->rlim_cur = RLIM_INFINITY;
1698	else
1699		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1700	if (rlim64_is_infinity(rlim64->rlim_max))
1701		rlim->rlim_max = RLIM_INFINITY;
1702	else
1703		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1704}
1705
1706/* rcu lock must be held */
1707static int check_prlimit_permission(struct task_struct *task,
1708				    unsigned int flags)
1709{
1710	const struct cred *cred = current_cred(), *tcred;
1711	bool id_match;
1712
1713	if (current == task)
1714		return 0;
1715
1716	tcred = __task_cred(task);
1717	id_match = (uid_eq(cred->uid, tcred->euid) &&
1718		    uid_eq(cred->uid, tcred->suid) &&
1719		    uid_eq(cred->uid, tcred->uid)  &&
1720		    gid_eq(cred->gid, tcred->egid) &&
1721		    gid_eq(cred->gid, tcred->sgid) &&
1722		    gid_eq(cred->gid, tcred->gid));
1723	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1724		return -EPERM;
1725
1726	return security_task_prlimit(cred, tcred, flags);
1727}
1728
1729SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1730		const struct rlimit64 __user *, new_rlim,
1731		struct rlimit64 __user *, old_rlim)
1732{
1733	struct rlimit64 old64, new64;
1734	struct rlimit old, new;
1735	struct task_struct *tsk;
1736	unsigned int checkflags = 0;
1737	bool need_tasklist;
1738	int ret;
1739
1740	if (old_rlim)
1741		checkflags |= LSM_PRLIMIT_READ;
1742
1743	if (new_rlim) {
1744		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1745			return -EFAULT;
1746		rlim64_to_rlim(&new64, &new);
1747		checkflags |= LSM_PRLIMIT_WRITE;
1748	}
1749
1750	rcu_read_lock();
1751	tsk = pid ? find_task_by_vpid(pid) : current;
1752	if (!tsk) {
1753		rcu_read_unlock();
1754		return -ESRCH;
1755	}
1756	ret = check_prlimit_permission(tsk, checkflags);
1757	if (ret) {
1758		rcu_read_unlock();
1759		return ret;
1760	}
1761	get_task_struct(tsk);
1762	rcu_read_unlock();
1763
1764	need_tasklist = !same_thread_group(tsk, current);
1765	if (need_tasklist) {
1766		/*
1767		 * Ensure we can't race with group exit or de_thread(),
1768		 * so tsk->group_leader can't be freed or changed until
1769		 * read_unlock(tasklist_lock) below.
1770		 */
1771		read_lock(&tasklist_lock);
1772		if (!pid_alive(tsk))
1773			ret = -ESRCH;
1774	}
1775
1776	if (!ret) {
1777		ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1778				old_rlim ? &old : NULL);
1779	}
1780
1781	if (need_tasklist)
1782		read_unlock(&tasklist_lock);
1783
1784	if (!ret && old_rlim) {
1785		rlim_to_rlim64(&old, &old64);
1786		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1787			ret = -EFAULT;
1788	}
1789
1790	put_task_struct(tsk);
1791	return ret;
1792}
1793
1794SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1795{
1796	struct rlimit new_rlim;
1797
1798	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1799		return -EFAULT;
1800	return do_prlimit(current, resource, &new_rlim, NULL);
1801}
1802
1803/*
1804 * It would make sense to put struct rusage in the task_struct,
1805 * except that would make the task_struct be *really big*.  After
1806 * task_struct gets moved into malloc'ed memory, it would
1807 * make sense to do this.  It will make moving the rest of the information
1808 * a lot simpler!  (Which we're not doing right now because we're not
1809 * measuring them yet).
1810 *
1811 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1812 * races with threads incrementing their own counters.  But since word
1813 * reads are atomic, we either get new values or old values and we don't
1814 * care which for the sums.  We always take the siglock to protect reading
1815 * the c* fields from p->signal from races with exit.c updating those
1816 * fields when reaping, so a sample either gets all the additions of a
1817 * given child after it's reaped, or none so this sample is before reaping.
1818 *
1819 * Locking:
1820 * We need to take the siglock for CHILDEREN, SELF and BOTH
1821 * for  the cases current multithreaded, non-current single threaded
1822 * non-current multithreaded.  Thread traversal is now safe with
1823 * the siglock held.
1824 * Strictly speaking, we donot need to take the siglock if we are current and
1825 * single threaded,  as no one else can take our signal_struct away, no one
1826 * else can  reap the  children to update signal->c* counters, and no one else
1827 * can race with the signal-> fields. If we do not take any lock, the
1828 * signal-> fields could be read out of order while another thread was just
1829 * exiting. So we should  place a read memory barrier when we avoid the lock.
1830 * On the writer side,  write memory barrier is implied in  __exit_signal
1831 * as __exit_signal releases  the siglock spinlock after updating the signal->
1832 * fields. But we don't do this yet to keep things simple.
1833 *
1834 */
1835
1836static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1837{
1838	r->ru_nvcsw += t->nvcsw;
1839	r->ru_nivcsw += t->nivcsw;
1840	r->ru_minflt += t->min_flt;
1841	r->ru_majflt += t->maj_flt;
1842	r->ru_inblock += task_io_get_inblock(t);
1843	r->ru_oublock += task_io_get_oublock(t);
1844}
1845
1846void getrusage(struct task_struct *p, int who, struct rusage *r)
1847{
1848	struct task_struct *t;
1849	unsigned long flags;
1850	u64 tgutime, tgstime, utime, stime;
1851	unsigned long maxrss;
1852	struct mm_struct *mm;
1853	struct signal_struct *sig = p->signal;
1854	unsigned int seq = 0;
1855
1856retry:
1857	memset(r, 0, sizeof(*r));
1858	utime = stime = 0;
1859	maxrss = 0;
1860
1861	if (who == RUSAGE_THREAD) {
1862		task_cputime_adjusted(current, &utime, &stime);
1863		accumulate_thread_rusage(p, r);
1864		maxrss = sig->maxrss;
1865		goto out_thread;
1866	}
1867
1868	flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1869
1870	switch (who) {
1871	case RUSAGE_BOTH:
1872	case RUSAGE_CHILDREN:
1873		utime = sig->cutime;
1874		stime = sig->cstime;
1875		r->ru_nvcsw = sig->cnvcsw;
1876		r->ru_nivcsw = sig->cnivcsw;
1877		r->ru_minflt = sig->cmin_flt;
1878		r->ru_majflt = sig->cmaj_flt;
1879		r->ru_inblock = sig->cinblock;
1880		r->ru_oublock = sig->coublock;
1881		maxrss = sig->cmaxrss;
1882
1883		if (who == RUSAGE_CHILDREN)
1884			break;
1885		fallthrough;
1886
1887	case RUSAGE_SELF:
1888		r->ru_nvcsw += sig->nvcsw;
1889		r->ru_nivcsw += sig->nivcsw;
1890		r->ru_minflt += sig->min_flt;
1891		r->ru_majflt += sig->maj_flt;
1892		r->ru_inblock += sig->inblock;
1893		r->ru_oublock += sig->oublock;
1894		if (maxrss < sig->maxrss)
1895			maxrss = sig->maxrss;
1896
1897		rcu_read_lock();
1898		__for_each_thread(sig, t)
1899			accumulate_thread_rusage(t, r);
1900		rcu_read_unlock();
1901
1902		break;
1903
1904	default:
1905		BUG();
1906	}
1907
1908	if (need_seqretry(&sig->stats_lock, seq)) {
1909		seq = 1;
1910		goto retry;
1911	}
1912	done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1913
1914	if (who == RUSAGE_CHILDREN)
1915		goto out_children;
1916
1917	thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1918	utime += tgutime;
1919	stime += tgstime;
1920
1921out_thread:
1922	mm = get_task_mm(p);
1923	if (mm) {
1924		setmax_mm_hiwater_rss(&maxrss, mm);
1925		mmput(mm);
1926	}
1927
1928out_children:
1929	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1930	r->ru_utime = ns_to_kernel_old_timeval(utime);
1931	r->ru_stime = ns_to_kernel_old_timeval(stime);
1932}
1933
1934SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1935{
1936	struct rusage r;
1937
1938	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1939	    who != RUSAGE_THREAD)
1940		return -EINVAL;
1941
1942	getrusage(current, who, &r);
1943	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1944}
1945
1946#ifdef CONFIG_COMPAT
1947COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1948{
1949	struct rusage r;
1950
1951	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1952	    who != RUSAGE_THREAD)
1953		return -EINVAL;
1954
1955	getrusage(current, who, &r);
1956	return put_compat_rusage(&r, ru);
1957}
1958#endif
1959
1960SYSCALL_DEFINE1(umask, int, mask)
1961{
1962	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1963	return mask;
1964}
1965
1966static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1967{
1968	CLASS(fd, exe)(fd);
1969	struct inode *inode;
1970	int err;
1971
1972	if (fd_empty(exe))
1973		return -EBADF;
1974
1975	inode = file_inode(fd_file(exe));
1976
1977	/*
1978	 * Because the original mm->exe_file points to executable file, make
1979	 * sure that this one is executable as well, to avoid breaking an
1980	 * overall picture.
1981	 */
1982	if (!S_ISREG(inode->i_mode) || path_noexec(&fd_file(exe)->f_path))
1983		return -EACCES;
1984
1985	err = file_permission(fd_file(exe), MAY_EXEC);
1986	if (err)
1987		return err;
1988
1989	return replace_mm_exe_file(mm, fd_file(exe));
1990}
1991
1992/*
1993 * Check arithmetic relations of passed addresses.
1994 *
1995 * WARNING: we don't require any capability here so be very careful
1996 * in what is allowed for modification from userspace.
1997 */
1998static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1999{
2000	unsigned long mmap_max_addr = TASK_SIZE;
2001	int error = -EINVAL, i;
2002
2003	static const unsigned char offsets[] = {
2004		offsetof(struct prctl_mm_map, start_code),
2005		offsetof(struct prctl_mm_map, end_code),
2006		offsetof(struct prctl_mm_map, start_data),
2007		offsetof(struct prctl_mm_map, end_data),
2008		offsetof(struct prctl_mm_map, start_brk),
2009		offsetof(struct prctl_mm_map, brk),
2010		offsetof(struct prctl_mm_map, start_stack),
2011		offsetof(struct prctl_mm_map, arg_start),
2012		offsetof(struct prctl_mm_map, arg_end),
2013		offsetof(struct prctl_mm_map, env_start),
2014		offsetof(struct prctl_mm_map, env_end),
2015	};
2016
2017	/*
2018	 * Make sure the members are not somewhere outside
2019	 * of allowed address space.
2020	 */
2021	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
2022		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
2023
2024		if ((unsigned long)val >= mmap_max_addr ||
2025		    (unsigned long)val < mmap_min_addr)
2026			goto out;
2027	}
2028
2029	/*
2030	 * Make sure the pairs are ordered.
2031	 */
2032#define __prctl_check_order(__m1, __op, __m2)				\
2033	((unsigned long)prctl_map->__m1 __op				\
2034	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
2035	error  = __prctl_check_order(start_code, <, end_code);
2036	error |= __prctl_check_order(start_data,<=, end_data);
2037	error |= __prctl_check_order(start_brk, <=, brk);
2038	error |= __prctl_check_order(arg_start, <=, arg_end);
2039	error |= __prctl_check_order(env_start, <=, env_end);
2040	if (error)
2041		goto out;
2042#undef __prctl_check_order
2043
2044	error = -EINVAL;
2045
2046	/*
2047	 * Neither we should allow to override limits if they set.
2048	 */
2049	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
2050			      prctl_map->start_brk, prctl_map->end_data,
2051			      prctl_map->start_data))
2052			goto out;
2053
2054	error = 0;
2055out:
2056	return error;
2057}
2058
2059#ifdef CONFIG_CHECKPOINT_RESTORE
2060static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2061{
2062	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2063	unsigned long user_auxv[AT_VECTOR_SIZE];
2064	struct mm_struct *mm = current->mm;
2065	int error;
2066
2067	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2068	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2069
2070	if (opt == PR_SET_MM_MAP_SIZE)
2071		return put_user((unsigned int)sizeof(prctl_map),
2072				(unsigned int __user *)addr);
2073
2074	if (data_size != sizeof(prctl_map))
2075		return -EINVAL;
2076
2077	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2078		return -EFAULT;
2079
2080	error = validate_prctl_map_addr(&prctl_map);
2081	if (error)
2082		return error;
2083
2084	if (prctl_map.auxv_size) {
2085		/*
2086		 * Someone is trying to cheat the auxv vector.
2087		 */
2088		if (!prctl_map.auxv ||
2089				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2090			return -EINVAL;
2091
2092		memset(user_auxv, 0, sizeof(user_auxv));
2093		if (copy_from_user(user_auxv,
2094				   (const void __user *)prctl_map.auxv,
2095				   prctl_map.auxv_size))
2096			return -EFAULT;
2097
2098		/* Last entry must be AT_NULL as specification requires */
2099		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2100		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2101	}
2102
2103	if (prctl_map.exe_fd != (u32)-1) {
2104		/*
2105		 * Check if the current user is checkpoint/restore capable.
2106		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2107		 * or CAP_CHECKPOINT_RESTORE.
2108		 * Note that a user with access to ptrace can masquerade an
2109		 * arbitrary program as any executable, even setuid ones.
2110		 * This may have implications in the tomoyo subsystem.
2111		 */
2112		if (!checkpoint_restore_ns_capable(current_user_ns()))
2113			return -EPERM;
2114
2115		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2116		if (error)
2117			return error;
2118	}
2119
2120	/*
2121	 * arg_lock protects concurrent updates but we still need mmap_lock for
2122	 * read to exclude races with sys_brk.
2123	 */
2124	mmap_read_lock(mm);
2125
2126	/*
2127	 * We don't validate if these members are pointing to
2128	 * real present VMAs because application may have correspond
2129	 * VMAs already unmapped and kernel uses these members for statistics
2130	 * output in procfs mostly, except
2131	 *
2132	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2133	 *    for VMAs when updating these members so anything wrong written
2134	 *    here cause kernel to swear at userspace program but won't lead
2135	 *    to any problem in kernel itself
2136	 */
2137
2138	spin_lock(&mm->arg_lock);
2139	mm->start_code	= prctl_map.start_code;
2140	mm->end_code	= prctl_map.end_code;
2141	mm->start_data	= prctl_map.start_data;
2142	mm->end_data	= prctl_map.end_data;
2143	mm->start_brk	= prctl_map.start_brk;
2144	mm->brk		= prctl_map.brk;
2145	mm->start_stack	= prctl_map.start_stack;
2146	mm->arg_start	= prctl_map.arg_start;
2147	mm->arg_end	= prctl_map.arg_end;
2148	mm->env_start	= prctl_map.env_start;
2149	mm->env_end	= prctl_map.env_end;
2150	spin_unlock(&mm->arg_lock);
2151
2152	/*
2153	 * Note this update of @saved_auxv is lockless thus
2154	 * if someone reads this member in procfs while we're
2155	 * updating -- it may get partly updated results. It's
2156	 * known and acceptable trade off: we leave it as is to
2157	 * not introduce additional locks here making the kernel
2158	 * more complex.
2159	 */
2160	if (prctl_map.auxv_size)
2161		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2162
2163	mmap_read_unlock(mm);
2164	return 0;
2165}
2166#endif /* CONFIG_CHECKPOINT_RESTORE */
2167
2168static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2169			  unsigned long len)
2170{
2171	/*
2172	 * This doesn't move the auxiliary vector itself since it's pinned to
2173	 * mm_struct, but it permits filling the vector with new values.  It's
2174	 * up to the caller to provide sane values here, otherwise userspace
2175	 * tools which use this vector might be unhappy.
2176	 */
2177	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2178
2179	if (len > sizeof(user_auxv))
2180		return -EINVAL;
2181
2182	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2183		return -EFAULT;
2184
2185	/* Make sure the last entry is always AT_NULL */
2186	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2187	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2188
2189	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2190
2191	task_lock(current);
2192	memcpy(mm->saved_auxv, user_auxv, len);
2193	task_unlock(current);
2194
2195	return 0;
2196}
2197
2198static int prctl_set_mm(int opt, unsigned long addr,
2199			unsigned long arg4, unsigned long arg5)
2200{
2201	struct mm_struct *mm = current->mm;
2202	struct prctl_mm_map prctl_map = {
2203		.auxv = NULL,
2204		.auxv_size = 0,
2205		.exe_fd = -1,
2206	};
2207	struct vm_area_struct *vma;
2208	int error;
2209
2210	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2211			      opt != PR_SET_MM_MAP &&
2212			      opt != PR_SET_MM_MAP_SIZE)))
2213		return -EINVAL;
2214
2215#ifdef CONFIG_CHECKPOINT_RESTORE
2216	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2217		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2218#endif
2219
2220	if (!capable(CAP_SYS_RESOURCE))
2221		return -EPERM;
2222
2223	if (opt == PR_SET_MM_EXE_FILE)
2224		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2225
2226	if (opt == PR_SET_MM_AUXV)
2227		return prctl_set_auxv(mm, addr, arg4);
2228
2229	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2230		return -EINVAL;
2231
2232	error = -EINVAL;
2233
2234	/*
2235	 * arg_lock protects concurrent updates of arg boundaries, we need
2236	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2237	 * validation.
2238	 */
2239	mmap_read_lock(mm);
2240	vma = find_vma(mm, addr);
2241
2242	spin_lock(&mm->arg_lock);
2243	prctl_map.start_code	= mm->start_code;
2244	prctl_map.end_code	= mm->end_code;
2245	prctl_map.start_data	= mm->start_data;
2246	prctl_map.end_data	= mm->end_data;
2247	prctl_map.start_brk	= mm->start_brk;
2248	prctl_map.brk		= mm->brk;
2249	prctl_map.start_stack	= mm->start_stack;
2250	prctl_map.arg_start	= mm->arg_start;
2251	prctl_map.arg_end	= mm->arg_end;
2252	prctl_map.env_start	= mm->env_start;
2253	prctl_map.env_end	= mm->env_end;
2254
2255	switch (opt) {
2256	case PR_SET_MM_START_CODE:
2257		prctl_map.start_code = addr;
2258		break;
2259	case PR_SET_MM_END_CODE:
2260		prctl_map.end_code = addr;
2261		break;
2262	case PR_SET_MM_START_DATA:
2263		prctl_map.start_data = addr;
2264		break;
2265	case PR_SET_MM_END_DATA:
2266		prctl_map.end_data = addr;
2267		break;
2268	case PR_SET_MM_START_STACK:
2269		prctl_map.start_stack = addr;
2270		break;
2271	case PR_SET_MM_START_BRK:
2272		prctl_map.start_brk = addr;
2273		break;
2274	case PR_SET_MM_BRK:
2275		prctl_map.brk = addr;
2276		break;
2277	case PR_SET_MM_ARG_START:
2278		prctl_map.arg_start = addr;
2279		break;
2280	case PR_SET_MM_ARG_END:
2281		prctl_map.arg_end = addr;
2282		break;
2283	case PR_SET_MM_ENV_START:
2284		prctl_map.env_start = addr;
2285		break;
2286	case PR_SET_MM_ENV_END:
2287		prctl_map.env_end = addr;
2288		break;
2289	default:
2290		goto out;
2291	}
2292
2293	error = validate_prctl_map_addr(&prctl_map);
2294	if (error)
2295		goto out;
2296
2297	switch (opt) {
2298	/*
2299	 * If command line arguments and environment
2300	 * are placed somewhere else on stack, we can
2301	 * set them up here, ARG_START/END to setup
2302	 * command line arguments and ENV_START/END
2303	 * for environment.
2304	 */
2305	case PR_SET_MM_START_STACK:
2306	case PR_SET_MM_ARG_START:
2307	case PR_SET_MM_ARG_END:
2308	case PR_SET_MM_ENV_START:
2309	case PR_SET_MM_ENV_END:
2310		if (!vma) {
2311			error = -EFAULT;
2312			goto out;
2313		}
2314	}
2315
2316	mm->start_code	= prctl_map.start_code;
2317	mm->end_code	= prctl_map.end_code;
2318	mm->start_data	= prctl_map.start_data;
2319	mm->end_data	= prctl_map.end_data;
2320	mm->start_brk	= prctl_map.start_brk;
2321	mm->brk		= prctl_map.brk;
2322	mm->start_stack	= prctl_map.start_stack;
2323	mm->arg_start	= prctl_map.arg_start;
2324	mm->arg_end	= prctl_map.arg_end;
2325	mm->env_start	= prctl_map.env_start;
2326	mm->env_end	= prctl_map.env_end;
2327
2328	error = 0;
2329out:
2330	spin_unlock(&mm->arg_lock);
2331	mmap_read_unlock(mm);
2332	return error;
2333}
2334
2335#ifdef CONFIG_CHECKPOINT_RESTORE
2336static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2337{
2338	return put_user(me->clear_child_tid, tid_addr);
2339}
2340#else
2341static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2342{
2343	return -EINVAL;
2344}
2345#endif
2346
2347static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2348{
2349	/*
2350	 * If task has has_child_subreaper - all its descendants
2351	 * already have these flag too and new descendants will
2352	 * inherit it on fork, skip them.
2353	 *
2354	 * If we've found child_reaper - skip descendants in
2355	 * it's subtree as they will never get out pidns.
2356	 */
2357	if (p->signal->has_child_subreaper ||
2358	    is_child_reaper(task_pid(p)))
2359		return 0;
2360
2361	p->signal->has_child_subreaper = 1;
2362	return 1;
2363}
2364
2365int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2366{
2367	return -EINVAL;
2368}
2369
2370int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2371				    unsigned long ctrl)
2372{
2373	return -EINVAL;
2374}
2375
2376int __weak arch_get_shadow_stack_status(struct task_struct *t, unsigned long __user *status)
2377{
2378	return -EINVAL;
2379}
2380
2381int __weak arch_set_shadow_stack_status(struct task_struct *t, unsigned long status)
2382{
2383	return -EINVAL;
2384}
2385
2386int __weak arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status)
2387{
2388	return -EINVAL;
2389}
2390
2391#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2392
2393static int prctl_set_vma(unsigned long opt, unsigned long addr,
2394			 unsigned long size, unsigned long arg)
2395{
2396	int error;
2397
2398	switch (opt) {
2399	case PR_SET_VMA_ANON_NAME:
2400		error = set_anon_vma_name(addr, size, (const char __user *)arg);
2401		break;
2402	default:
2403		error = -EINVAL;
2404	}
2405
2406	return error;
2407}
2408
2409static inline unsigned long get_current_mdwe(void)
2410{
2411	unsigned long ret = 0;
2412
2413	if (mm_flags_test(MMF_HAS_MDWE, current->mm))
2414		ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2415	if (mm_flags_test(MMF_HAS_MDWE_NO_INHERIT, current->mm))
2416		ret |= PR_MDWE_NO_INHERIT;
2417
2418	return ret;
2419}
2420
2421static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2422				 unsigned long arg4, unsigned long arg5)
2423{
2424	unsigned long current_bits;
2425
2426	if (arg3 || arg4 || arg5)
2427		return -EINVAL;
2428
2429	if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2430		return -EINVAL;
2431
2432	/* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2433	if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2434		return -EINVAL;
2435
2436	/*
2437	 * EOPNOTSUPP might be more appropriate here in principle, but
2438	 * existing userspace depends on EINVAL specifically.
2439	 */
2440	if (!arch_memory_deny_write_exec_supported())
2441		return -EINVAL;
2442
2443	current_bits = get_current_mdwe();
2444	if (current_bits && current_bits != bits)
2445		return -EPERM; /* Cannot unset the flags */
2446
2447	if (bits & PR_MDWE_NO_INHERIT)
2448		mm_flags_set(MMF_HAS_MDWE_NO_INHERIT, current->mm);
2449	if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2450		mm_flags_set(MMF_HAS_MDWE, current->mm);
2451
2452	return 0;
2453}
2454
2455static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2456				 unsigned long arg4, unsigned long arg5)
2457{
2458	if (arg2 || arg3 || arg4 || arg5)
2459		return -EINVAL;
2460	return get_current_mdwe();
2461}
2462
2463static int prctl_get_auxv(void __user *addr, unsigned long len)
2464{
2465	struct mm_struct *mm = current->mm;
2466	unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2467
2468	if (size && copy_to_user(addr, mm->saved_auxv, size))
2469		return -EFAULT;
2470	return sizeof(mm->saved_auxv);
2471}
2472
2473static int prctl_get_thp_disable(unsigned long arg2, unsigned long arg3,
2474				 unsigned long arg4, unsigned long arg5)
2475{
2476	struct mm_struct *mm = current->mm;
2477
2478	if (arg2 || arg3 || arg4 || arg5)
2479		return -EINVAL;
2480
2481	/* If disabled, we return "1 | flags", otherwise 0. */
2482	if (mm_flags_test(MMF_DISABLE_THP_COMPLETELY, mm))
2483		return 1;
2484	else if (mm_flags_test(MMF_DISABLE_THP_EXCEPT_ADVISED, mm))
2485		return 1 | PR_THP_DISABLE_EXCEPT_ADVISED;
2486	return 0;
2487}
2488
2489static int prctl_set_thp_disable(bool thp_disable, unsigned long flags,
2490				 unsigned long arg4, unsigned long arg5)
2491{
2492	struct mm_struct *mm = current->mm;
2493
2494	if (arg4 || arg5)
2495		return -EINVAL;
2496
2497	/* Flags are only allowed when disabling. */
2498	if ((!thp_disable && flags) || (flags & ~PR_THP_DISABLE_EXCEPT_ADVISED))
2499		return -EINVAL;
2500	if (mmap_write_lock_killable(current->mm))
2501		return -EINTR;
2502	if (thp_disable) {
2503		if (flags & PR_THP_DISABLE_EXCEPT_ADVISED) {
2504			mm_flags_clear(MMF_DISABLE_THP_COMPLETELY, mm);
2505			mm_flags_set(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2506		} else {
2507			mm_flags_set(MMF_DISABLE_THP_COMPLETELY, mm);
2508			mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2509		}
2510	} else {
2511		mm_flags_clear(MMF_DISABLE_THP_COMPLETELY, mm);
2512		mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2513	}
2514	mmap_write_unlock(current->mm);
2515	return 0;
2516}
2517
2518SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2519		unsigned long, arg4, unsigned long, arg5)
2520{
2521	struct task_struct *me = current;
2522	unsigned char comm[sizeof(me->comm)];
2523	long error;
2524
2525	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2526	if (error != -ENOSYS)
2527		return error;
2528
2529	error = 0;
2530	switch (option) {
2531	case PR_SET_PDEATHSIG:
2532		if (!valid_signal(arg2)) {
2533			error = -EINVAL;
2534			break;
2535		}
2536		/*
2537		 * Ensure that either:
2538		 *
2539		 * 1. Subsequent getppid() calls reflect the parent process having died.
2540		 * 2. forget_original_parent() will send the new me->pdeath_signal.
2541		 *
2542		 * Also prevent the read of me->pdeath_signal from being a data race.
2543		 */
2544		read_lock(&tasklist_lock);
2545		me->pdeath_signal = arg2;
2546		read_unlock(&tasklist_lock);
2547		break;
2548	case PR_GET_PDEATHSIG:
2549		error = put_user(me->pdeath_signal, (int __user *)arg2);
2550		break;
2551	case PR_GET_DUMPABLE:
2552		error = get_dumpable(me->mm);
2553		break;
2554	case PR_SET_DUMPABLE:
2555		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2556			error = -EINVAL;
2557			break;
2558		}
2559		set_dumpable(me->mm, arg2);
2560		break;
2561
2562	case PR_SET_UNALIGN:
2563		error = SET_UNALIGN_CTL(me, arg2);
2564		break;
2565	case PR_GET_UNALIGN:
2566		error = GET_UNALIGN_CTL(me, arg2);
2567		break;
2568	case PR_SET_FPEMU:
2569		error = SET_FPEMU_CTL(me, arg2);
2570		break;
2571	case PR_GET_FPEMU:
2572		error = GET_FPEMU_CTL(me, arg2);
2573		break;
2574	case PR_SET_FPEXC:
2575		error = SET_FPEXC_CTL(me, arg2);
2576		break;
2577	case PR_GET_FPEXC:
2578		error = GET_FPEXC_CTL(me, arg2);
2579		break;
2580	case PR_GET_TIMING:
2581		error = PR_TIMING_STATISTICAL;
2582		break;
2583	case PR_SET_TIMING:
2584		if (arg2 != PR_TIMING_STATISTICAL)
2585			error = -EINVAL;
2586		break;
2587	case PR_SET_NAME:
2588		comm[sizeof(me->comm) - 1] = 0;
2589		if (strncpy_from_user(comm, (char __user *)arg2,
2590				      sizeof(me->comm) - 1) < 0)
2591			return -EFAULT;
2592		set_task_comm(me, comm);
2593		proc_comm_connector(me);
2594		break;
2595	case PR_GET_NAME:
2596		get_task_comm(comm, me);
2597		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2598			return -EFAULT;
2599		break;
2600	case PR_GET_ENDIAN:
2601		error = GET_ENDIAN(me, arg2);
2602		break;
2603	case PR_SET_ENDIAN:
2604		error = SET_ENDIAN(me, arg2);
2605		break;
2606	case PR_GET_SECCOMP:
2607		error = prctl_get_seccomp();
2608		break;
2609	case PR_SET_SECCOMP:
2610		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2611		break;
2612	case PR_GET_TSC:
2613		error = GET_TSC_CTL(arg2);
2614		break;
2615	case PR_SET_TSC:
2616		error = SET_TSC_CTL(arg2);
2617		break;
2618	case PR_TASK_PERF_EVENTS_DISABLE:
2619		error = perf_event_task_disable();
2620		break;
2621	case PR_TASK_PERF_EVENTS_ENABLE:
2622		error = perf_event_task_enable();
2623		break;
2624	case PR_GET_TIMERSLACK:
2625		if (current->timer_slack_ns > ULONG_MAX)
2626			error = ULONG_MAX;
2627		else
2628			error = current->timer_slack_ns;
2629		break;
2630	case PR_SET_TIMERSLACK:
2631		if (rt_or_dl_task_policy(current))
2632			break;
2633		if (arg2 <= 0)
2634			current->timer_slack_ns =
2635					current->default_timer_slack_ns;
2636		else
2637			current->timer_slack_ns = arg2;
2638		break;
2639	case PR_MCE_KILL:
2640		if (arg4 | arg5)
2641			return -EINVAL;
2642		switch (arg2) {
2643		case PR_MCE_KILL_CLEAR:
2644			if (arg3 != 0)
2645				return -EINVAL;
2646			current->flags &= ~PF_MCE_PROCESS;
2647			break;
2648		case PR_MCE_KILL_SET:
2649			current->flags |= PF_MCE_PROCESS;
2650			if (arg3 == PR_MCE_KILL_EARLY)
2651				current->flags |= PF_MCE_EARLY;
2652			else if (arg3 == PR_MCE_KILL_LATE)
2653				current->flags &= ~PF_MCE_EARLY;
2654			else if (arg3 == PR_MCE_KILL_DEFAULT)
2655				current->flags &=
2656						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2657			else
2658				return -EINVAL;
2659			break;
2660		default:
2661			return -EINVAL;
2662		}
2663		break;
2664	case PR_MCE_KILL_GET:
2665		if (arg2 | arg3 | arg4 | arg5)
2666			return -EINVAL;
2667		if (current->flags & PF_MCE_PROCESS)
2668			error = (current->flags & PF_MCE_EARLY) ?
2669				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2670		else
2671			error = PR_MCE_KILL_DEFAULT;
2672		break;
2673	case PR_SET_MM:
2674		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2675		break;
2676	case PR_GET_TID_ADDRESS:
2677		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2678		break;
2679	case PR_SET_CHILD_SUBREAPER:
2680		me->signal->is_child_subreaper = !!arg2;
2681		if (!arg2)
2682			break;
2683
2684		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2685		break;
2686	case PR_GET_CHILD_SUBREAPER:
2687		error = put_user(me->signal->is_child_subreaper,
2688				 (int __user *)arg2);
2689		break;
2690	case PR_SET_NO_NEW_PRIVS:
2691		if (arg2 != 1 || arg3 || arg4 || arg5)
2692			return -EINVAL;
2693
2694		task_set_no_new_privs(current);
2695		break;
2696	case PR_GET_NO_NEW_PRIVS:
2697		if (arg2 || arg3 || arg4 || arg5)
2698			return -EINVAL;
2699		return task_no_new_privs(current) ? 1 : 0;
2700	case PR_GET_THP_DISABLE:
2701		error = prctl_get_thp_disable(arg2, arg3, arg4, arg5);
2702		break;
2703	case PR_SET_THP_DISABLE:
2704		error = prctl_set_thp_disable(arg2, arg3, arg4, arg5);
2705		break;
2706	case PR_MPX_ENABLE_MANAGEMENT:
2707	case PR_MPX_DISABLE_MANAGEMENT:
2708		/* No longer implemented: */
2709		return -EINVAL;
2710	case PR_SET_FP_MODE:
2711		error = SET_FP_MODE(me, arg2);
2712		break;
2713	case PR_GET_FP_MODE:
2714		error = GET_FP_MODE(me);
2715		break;
2716	case PR_SVE_SET_VL:
2717		error = SVE_SET_VL(arg2);
2718		break;
2719	case PR_SVE_GET_VL:
2720		error = SVE_GET_VL();
2721		break;
2722	case PR_SME_SET_VL:
2723		error = SME_SET_VL(arg2);
2724		break;
2725	case PR_SME_GET_VL:
2726		error = SME_GET_VL();
2727		break;
2728	case PR_GET_SPECULATION_CTRL:
2729		if (arg3 || arg4 || arg5)
2730			return -EINVAL;
2731		error = arch_prctl_spec_ctrl_get(me, arg2);
2732		break;
2733	case PR_SET_SPECULATION_CTRL:
2734		if (arg4 || arg5)
2735			return -EINVAL;
2736		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2737		break;
2738	case PR_PAC_RESET_KEYS:
2739		if (arg3 || arg4 || arg5)
2740			return -EINVAL;
2741		error = PAC_RESET_KEYS(me, arg2);
2742		break;
2743	case PR_PAC_SET_ENABLED_KEYS:
2744		if (arg4 || arg5)
2745			return -EINVAL;
2746		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2747		break;
2748	case PR_PAC_GET_ENABLED_KEYS:
2749		if (arg2 || arg3 || arg4 || arg5)
2750			return -EINVAL;
2751		error = PAC_GET_ENABLED_KEYS(me);
2752		break;
2753	case PR_SET_TAGGED_ADDR_CTRL:
2754		if (arg3 || arg4 || arg5)
2755			return -EINVAL;
2756		error = SET_TAGGED_ADDR_CTRL(arg2);
2757		break;
2758	case PR_GET_TAGGED_ADDR_CTRL:
2759		if (arg2 || arg3 || arg4 || arg5)
2760			return -EINVAL;
2761		error = GET_TAGGED_ADDR_CTRL();
2762		break;
2763	case PR_SET_IO_FLUSHER:
2764		if (!capable(CAP_SYS_RESOURCE))
2765			return -EPERM;
2766
2767		if (arg3 || arg4 || arg5)
2768			return -EINVAL;
2769
2770		if (arg2 == 1)
2771			current->flags |= PR_IO_FLUSHER;
2772		else if (!arg2)
2773			current->flags &= ~PR_IO_FLUSHER;
2774		else
2775			return -EINVAL;
2776		break;
2777	case PR_GET_IO_FLUSHER:
2778		if (!capable(CAP_SYS_RESOURCE))
2779			return -EPERM;
2780
2781		if (arg2 || arg3 || arg4 || arg5)
2782			return -EINVAL;
2783
2784		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2785		break;
2786	case PR_SET_SYSCALL_USER_DISPATCH:
2787		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2788						  (char __user *) arg5);
2789		break;
2790#ifdef CONFIG_SCHED_CORE
2791	case PR_SCHED_CORE:
2792		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2793		break;
2794#endif
2795	case PR_SET_MDWE:
2796		error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2797		break;
2798	case PR_GET_MDWE:
2799		error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2800		break;
2801	case PR_PPC_GET_DEXCR:
2802		if (arg3 || arg4 || arg5)
2803			return -EINVAL;
2804		error = PPC_GET_DEXCR_ASPECT(me, arg2);
2805		break;
2806	case PR_PPC_SET_DEXCR:
2807		if (arg4 || arg5)
2808			return -EINVAL;
2809		error = PPC_SET_DEXCR_ASPECT(me, arg2, arg3);
2810		break;
2811	case PR_SET_VMA:
2812		error = prctl_set_vma(arg2, arg3, arg4, arg5);
2813		break;
2814	case PR_GET_AUXV:
2815		if (arg4 || arg5)
2816			return -EINVAL;
2817		error = prctl_get_auxv((void __user *)arg2, arg3);
2818		break;
2819#ifdef CONFIG_KSM
2820	case PR_SET_MEMORY_MERGE:
2821		if (arg3 || arg4 || arg5)
2822			return -EINVAL;
2823		if (mmap_write_lock_killable(me->mm))
2824			return -EINTR;
2825
2826		if (arg2)
2827			error = ksm_enable_merge_any(me->mm);
2828		else
2829			error = ksm_disable_merge_any(me->mm);
2830		mmap_write_unlock(me->mm);
2831		break;
2832	case PR_GET_MEMORY_MERGE:
2833		if (arg2 || arg3 || arg4 || arg5)
2834			return -EINVAL;
2835
2836		error = !!mm_flags_test(MMF_VM_MERGE_ANY, me->mm);
2837		break;
2838#endif
2839	case PR_RISCV_V_SET_CONTROL:
2840		error = RISCV_V_SET_CONTROL(arg2);
2841		break;
2842	case PR_RISCV_V_GET_CONTROL:
2843		error = RISCV_V_GET_CONTROL();
2844		break;
2845	case PR_RISCV_SET_ICACHE_FLUSH_CTX:
2846		error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
2847		break;
2848	case PR_GET_SHADOW_STACK_STATUS:
2849		if (arg3 || arg4 || arg5)
2850			return -EINVAL;
2851		error = arch_get_shadow_stack_status(me, (unsigned long __user *) arg2);
2852		break;
2853	case PR_SET_SHADOW_STACK_STATUS:
2854		if (arg3 || arg4 || arg5)
2855			return -EINVAL;
2856		error = arch_set_shadow_stack_status(me, arg2);
2857		break;
2858	case PR_LOCK_SHADOW_STACK_STATUS:
2859		if (arg3 || arg4 || arg5)
2860			return -EINVAL;
2861		error = arch_lock_shadow_stack_status(me, arg2);
2862		break;
2863	case PR_TIMER_CREATE_RESTORE_IDS:
2864		if (arg3 || arg4 || arg5)
2865			return -EINVAL;
2866		error = posixtimer_create_prctl(arg2);
2867		break;
2868	case PR_FUTEX_HASH:
2869		error = futex_hash_prctl(arg2, arg3, arg4);
2870		break;
2871	default:
2872		trace_task_prctl_unknown(option, arg2, arg3, arg4, arg5);
2873		error = -EINVAL;
2874		break;
2875	}
2876	return error;
2877}
2878
2879SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2880		struct getcpu_cache __user *, unused)
2881{
2882	int err = 0;
2883	int cpu = raw_smp_processor_id();
2884
2885	if (cpup)
2886		err |= put_user(cpu, cpup);
2887	if (nodep)
2888		err |= put_user(cpu_to_node(cpu), nodep);
2889	return err ? -EFAULT : 0;
2890}
2891
2892/**
2893 * do_sysinfo - fill in sysinfo struct
2894 * @info: pointer to buffer to fill
2895 */
2896static int do_sysinfo(struct sysinfo *info)
2897{
2898	unsigned long mem_total, sav_total;
2899	unsigned int mem_unit, bitcount;
2900	struct timespec64 tp;
2901
2902	memset(info, 0, sizeof(struct sysinfo));
2903
2904	ktime_get_boottime_ts64(&tp);
2905	timens_add_boottime(&tp);
2906	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2907
2908	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2909
2910	info->procs = nr_threads;
2911
2912	si_meminfo(info);
2913	si_swapinfo(info);
2914
2915	/*
2916	 * If the sum of all the available memory (i.e. ram + swap)
2917	 * is less than can be stored in a 32 bit unsigned long then
2918	 * we can be binary compatible with 2.2.x kernels.  If not,
2919	 * well, in that case 2.2.x was broken anyways...
2920	 *
2921	 *  -Erik Andersen <andersee@debian.org>
2922	 */
2923
2924	mem_total = info->totalram + info->totalswap;
2925	if (mem_total < info->totalram || mem_total < info->totalswap)
2926		goto out;
2927	bitcount = 0;
2928	mem_unit = info->mem_unit;
2929	while (mem_unit > 1) {
2930		bitcount++;
2931		mem_unit >>= 1;
2932		sav_total = mem_total;
2933		mem_total <<= 1;
2934		if (mem_total < sav_total)
2935			goto out;
2936	}
2937
2938	/*
2939	 * If mem_total did not overflow, multiply all memory values by
2940	 * info->mem_unit and set it to 1.  This leaves things compatible
2941	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2942	 * kernels...
2943	 */
2944
2945	info->mem_unit = 1;
2946	info->totalram <<= bitcount;
2947	info->freeram <<= bitcount;
2948	info->sharedram <<= bitcount;
2949	info->bufferram <<= bitcount;
2950	info->totalswap <<= bitcount;
2951	info->freeswap <<= bitcount;
2952	info->totalhigh <<= bitcount;
2953	info->freehigh <<= bitcount;
2954
2955out:
2956	return 0;
2957}
2958
2959SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2960{
2961	struct sysinfo val;
2962
2963	do_sysinfo(&val);
2964
2965	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2966		return -EFAULT;
2967
2968	return 0;
2969}
2970
2971#ifdef CONFIG_COMPAT
2972struct compat_sysinfo {
2973	s32 uptime;
2974	u32 loads[3];
2975	u32 totalram;
2976	u32 freeram;
2977	u32 sharedram;
2978	u32 bufferram;
2979	u32 totalswap;
2980	u32 freeswap;
2981	u16 procs;
2982	u16 pad;
2983	u32 totalhigh;
2984	u32 freehigh;
2985	u32 mem_unit;
2986	char _f[20-2*sizeof(u32)-sizeof(int)];
2987};
2988
2989COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2990{
2991	struct sysinfo s;
2992	struct compat_sysinfo s_32;
2993
2994	do_sysinfo(&s);
2995
2996	/* Check to see if any memory value is too large for 32-bit and scale
2997	 *  down if needed
2998	 */
2999	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
3000		int bitcount = 0;
3001
3002		while (s.mem_unit < PAGE_SIZE) {
3003			s.mem_unit <<= 1;
3004			bitcount++;
3005		}
3006
3007		s.totalram >>= bitcount;
3008		s.freeram >>= bitcount;
3009		s.sharedram >>= bitcount;
3010		s.bufferram >>= bitcount;
3011		s.totalswap >>= bitcount;
3012		s.freeswap >>= bitcount;
3013		s.totalhigh >>= bitcount;
3014		s.freehigh >>= bitcount;
3015	}
3016
3017	memset(&s_32, 0, sizeof(s_32));
3018	s_32.uptime = s.uptime;
3019	s_32.loads[0] = s.loads[0];
3020	s_32.loads[1] = s.loads[1];
3021	s_32.loads[2] = s.loads[2];
3022	s_32.totalram = s.totalram;
3023	s_32.freeram = s.freeram;
3024	s_32.sharedram = s.sharedram;
3025	s_32.bufferram = s.bufferram;
3026	s_32.totalswap = s.totalswap;
3027	s_32.freeswap = s.freeswap;
3028	s_32.procs = s.procs;
3029	s_32.totalhigh = s.totalhigh;
3030	s_32.freehigh = s.freehigh;
3031	s_32.mem_unit = s.mem_unit;
3032	if (copy_to_user(info, &s_32, sizeof(s_32)))
3033		return -EFAULT;
3034	return 0;
3035}
3036#endif /* CONFIG_COMPAT */