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/dcookies.h>
  28#include <linux/suspend.h>
  29#include <linux/tty.h>
  30#include <linux/signal.h>
  31#include <linux/cn_proc.h>
  32#include <linux/getcpu.h>
  33#include <linux/task_io_accounting_ops.h>
  34#include <linux/seccomp.h>
  35#include <linux/cpu.h>
  36#include <linux/personality.h>
  37#include <linux/ptrace.h>
  38#include <linux/fs_struct.h>
  39#include <linux/file.h>
  40#include <linux/mount.h>
  41#include <linux/gfp.h>
  42#include <linux/syscore_ops.h>
  43#include <linux/version.h>
  44#include <linux/ctype.h>
  45
  46#include <linux/compat.h>
  47#include <linux/syscalls.h>
  48#include <linux/kprobes.h>
  49#include <linux/user_namespace.h>
  50#include <linux/binfmts.h>
  51
  52#include <linux/sched.h>
  53#include <linux/sched/autogroup.h>
  54#include <linux/sched/loadavg.h>
  55#include <linux/sched/stat.h>
  56#include <linux/sched/mm.h>
  57#include <linux/sched/coredump.h>
  58#include <linux/sched/task.h>
  59#include <linux/sched/cputime.h>
  60#include <linux/rcupdate.h>
  61#include <linux/uidgid.h>
  62#include <linux/cred.h>
  63
  64#include <linux/nospec.h>
  65
  66#include <linux/kmsg_dump.h>
  67/* Move somewhere else to avoid recompiling? */
  68#include <generated/utsrelease.h>
  69
  70#include <linux/uaccess.h>
  71#include <asm/io.h>
  72#include <asm/unistd.h>
  73
  74/* Hardening for Spectre-v1 */
  75#include <linux/nospec.h>
  76
  77#include "uid16.h"
  78
  79#ifndef SET_UNALIGN_CTL
  80# define SET_UNALIGN_CTL(a, b)	(-EINVAL)
  81#endif
  82#ifndef GET_UNALIGN_CTL
  83# define GET_UNALIGN_CTL(a, b)	(-EINVAL)
  84#endif
  85#ifndef SET_FPEMU_CTL
  86# define SET_FPEMU_CTL(a, b)	(-EINVAL)
  87#endif
  88#ifndef GET_FPEMU_CTL
  89# define GET_FPEMU_CTL(a, b)	(-EINVAL)
  90#endif
  91#ifndef SET_FPEXC_CTL
  92# define SET_FPEXC_CTL(a, b)	(-EINVAL)
  93#endif
  94#ifndef GET_FPEXC_CTL
  95# define GET_FPEXC_CTL(a, b)	(-EINVAL)
  96#endif
  97#ifndef GET_ENDIAN
  98# define GET_ENDIAN(a, b)	(-EINVAL)
  99#endif
 100#ifndef SET_ENDIAN
 101# define SET_ENDIAN(a, b)	(-EINVAL)
 102#endif
 103#ifndef GET_TSC_CTL
 104# define GET_TSC_CTL(a)		(-EINVAL)
 105#endif
 106#ifndef SET_TSC_CTL
 107# define SET_TSC_CTL(a)		(-EINVAL)
 108#endif
 109#ifndef MPX_ENABLE_MANAGEMENT
 110# define MPX_ENABLE_MANAGEMENT()	(-EINVAL)
 111#endif
 112#ifndef MPX_DISABLE_MANAGEMENT
 113# define MPX_DISABLE_MANAGEMENT()	(-EINVAL)
 114#endif
 115#ifndef GET_FP_MODE
 116# define GET_FP_MODE(a)		(-EINVAL)
 117#endif
 118#ifndef SET_FP_MODE
 119# define SET_FP_MODE(a,b)	(-EINVAL)
 120#endif
 121#ifndef SVE_SET_VL
 122# define SVE_SET_VL(a)		(-EINVAL)
 123#endif
 124#ifndef SVE_GET_VL
 125# define SVE_GET_VL()		(-EINVAL)
 126#endif
 127
 128/*
 129 * this is where the system-wide overflow UID and GID are defined, for
 130 * architectures that now have 32-bit UID/GID but didn't in the past
 131 */
 132
 133int overflowuid = DEFAULT_OVERFLOWUID;
 134int overflowgid = DEFAULT_OVERFLOWGID;
 135
 136EXPORT_SYMBOL(overflowuid);
 137EXPORT_SYMBOL(overflowgid);
 138
 139/*
 140 * the same as above, but for filesystems which can only store a 16-bit
 141 * UID and GID. as such, this is needed on all architectures
 142 */
 143
 144int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
 145int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
 146
 147EXPORT_SYMBOL(fs_overflowuid);
 148EXPORT_SYMBOL(fs_overflowgid);
 149
 150/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 151 * Returns true if current's euid is same as p's uid or euid,
 152 * or has CAP_SYS_NICE to p's user_ns.
 153 *
 154 * Called with rcu_read_lock, creds are safe
 155 */
 156static bool set_one_prio_perm(struct task_struct *p)
 157{
 158	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
 159
 160	if (uid_eq(pcred->uid,  cred->euid) ||
 161	    uid_eq(pcred->euid, cred->euid))
 162		return true;
 163	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
 164		return true;
 165	return false;
 166}
 167
 168/*
 169 * set the priority of a task
 170 * - the caller must hold the RCU read lock
 171 */
 172static int set_one_prio(struct task_struct *p, int niceval, int error)
 173{
 174	int no_nice;
 175
 176	if (!set_one_prio_perm(p)) {
 177		error = -EPERM;
 178		goto out;
 179	}
 180	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
 181		error = -EACCES;
 182		goto out;
 183	}
 184	no_nice = security_task_setnice(p, niceval);
 185	if (no_nice) {
 186		error = no_nice;
 187		goto out;
 188	}
 189	if (error == -ESRCH)
 190		error = 0;
 191	set_user_nice(p, niceval);
 192out:
 193	return error;
 194}
 195
 196SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
 197{
 198	struct task_struct *g, *p;
 199	struct user_struct *user;
 200	const struct cred *cred = current_cred();
 201	int error = -EINVAL;
 202	struct pid *pgrp;
 203	kuid_t uid;
 204
 205	if (which > PRIO_USER || which < PRIO_PROCESS)
 206		goto out;
 207
 208	/* normalize: avoid signed division (rounding problems) */
 209	error = -ESRCH;
 210	if (niceval < MIN_NICE)
 211		niceval = MIN_NICE;
 212	if (niceval > MAX_NICE)
 213		niceval = MAX_NICE;
 214
 215	rcu_read_lock();
 216	read_lock(&tasklist_lock);
 217	switch (which) {
 218	case PRIO_PROCESS:
 219		if (who)
 220			p = find_task_by_vpid(who);
 221		else
 222			p = current;
 223		if (p)
 224			error = set_one_prio(p, niceval, error);
 225		break;
 226	case PRIO_PGRP:
 227		if (who)
 228			pgrp = find_vpid(who);
 229		else
 230			pgrp = task_pgrp(current);
 231		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 232			error = set_one_prio(p, niceval, error);
 233		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 234		break;
 235	case PRIO_USER:
 236		uid = make_kuid(cred->user_ns, who);
 237		user = cred->user;
 238		if (!who)
 239			uid = cred->uid;
 240		else if (!uid_eq(uid, cred->uid)) {
 241			user = find_user(uid);
 242			if (!user)
 243				goto out_unlock;	/* No processes for this user */
 244		}
 245		do_each_thread(g, p) {
 246			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
 247				error = set_one_prio(p, niceval, error);
 248		} while_each_thread(g, p);
 249		if (!uid_eq(uid, cred->uid))
 250			free_uid(user);		/* For find_user() */
 251		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 252	}
 253out_unlock:
 254	read_unlock(&tasklist_lock);
 255	rcu_read_unlock();
 256out:
 257	return error;
 258}
 259
 260/*
 261 * Ugh. To avoid negative return values, "getpriority()" will
 262 * not return the normal nice-value, but a negated value that
 263 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
 264 * to stay compatible.
 265 */
 266SYSCALL_DEFINE2(getpriority, int, which, int, who)
 267{
 268	struct task_struct *g, *p;
 269	struct user_struct *user;
 270	const struct cred *cred = current_cred();
 271	long niceval, retval = -ESRCH;
 272	struct pid *pgrp;
 273	kuid_t uid;
 274
 275	if (which > PRIO_USER || which < PRIO_PROCESS)
 276		return -EINVAL;
 277
 278	rcu_read_lock();
 279	read_lock(&tasklist_lock);
 280	switch (which) {
 281	case PRIO_PROCESS:
 282		if (who)
 283			p = find_task_by_vpid(who);
 284		else
 285			p = current;
 286		if (p) {
 287			niceval = nice_to_rlimit(task_nice(p));
 288			if (niceval > retval)
 289				retval = niceval;
 290		}
 291		break;
 292	case PRIO_PGRP:
 293		if (who)
 294			pgrp = find_vpid(who);
 295		else
 296			pgrp = task_pgrp(current);
 297		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 298			niceval = nice_to_rlimit(task_nice(p));
 299			if (niceval > retval)
 300				retval = niceval;
 301		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 302		break;
 303	case PRIO_USER:
 304		uid = make_kuid(cred->user_ns, who);
 305		user = cred->user;
 306		if (!who)
 307			uid = cred->uid;
 308		else if (!uid_eq(uid, cred->uid)) {
 309			user = find_user(uid);
 310			if (!user)
 311				goto out_unlock;	/* No processes for this user */
 312		}
 313		do_each_thread(g, p) {
 314			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
 315				niceval = nice_to_rlimit(task_nice(p));
 316				if (niceval > retval)
 317					retval = niceval;
 318			}
 319		} while_each_thread(g, p);
 320		if (!uid_eq(uid, cred->uid))
 321			free_uid(user);		/* for find_user() */
 322		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 323	}
 324out_unlock:
 325	read_unlock(&tasklist_lock);
 326	rcu_read_unlock();
 327
 328	return retval;
 329}
 330
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 331/*
 332 * Unprivileged users may change the real gid to the effective gid
 333 * or vice versa.  (BSD-style)
 334 *
 335 * If you set the real gid at all, or set the effective gid to a value not
 336 * equal to the real gid, then the saved gid is set to the new effective gid.
 337 *
 338 * This makes it possible for a setgid program to completely drop its
 339 * privileges, which is often a useful assertion to make when you are doing
 340 * a security audit over a program.
 341 *
 342 * The general idea is that a program which uses just setregid() will be
 343 * 100% compatible with BSD.  A program which uses just setgid() will be
 344 * 100% compatible with POSIX with saved IDs.
 345 *
 346 * SMP: There are not races, the GIDs are checked only by filesystem
 347 *      operations (as far as semantic preservation is concerned).
 348 */
 349#ifdef CONFIG_MULTIUSER
 350long __sys_setregid(gid_t rgid, gid_t egid)
 351{
 352	struct user_namespace *ns = current_user_ns();
 353	const struct cred *old;
 354	struct cred *new;
 355	int retval;
 356	kgid_t krgid, kegid;
 357
 358	krgid = make_kgid(ns, rgid);
 359	kegid = make_kgid(ns, egid);
 360
 361	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 362		return -EINVAL;
 363	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 364		return -EINVAL;
 365
 366	new = prepare_creds();
 367	if (!new)
 368		return -ENOMEM;
 369	old = current_cred();
 370
 371	retval = -EPERM;
 372	if (rgid != (gid_t) -1) {
 373		if (gid_eq(old->gid, krgid) ||
 374		    gid_eq(old->egid, krgid) ||
 375		    ns_capable(old->user_ns, CAP_SETGID))
 376			new->gid = krgid;
 377		else
 378			goto error;
 379	}
 380	if (egid != (gid_t) -1) {
 381		if (gid_eq(old->gid, kegid) ||
 382		    gid_eq(old->egid, kegid) ||
 383		    gid_eq(old->sgid, kegid) ||
 384		    ns_capable(old->user_ns, CAP_SETGID))
 385			new->egid = kegid;
 386		else
 387			goto error;
 388	}
 389
 390	if (rgid != (gid_t) -1 ||
 391	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
 392		new->sgid = new->egid;
 393	new->fsgid = new->egid;
 394
 395	return commit_creds(new);
 396
 397error:
 398	abort_creds(new);
 399	return retval;
 400}
 401
 402SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
 403{
 404	return __sys_setregid(rgid, egid);
 405}
 406
 407/*
 408 * setgid() is implemented like SysV w/ SAVED_IDS
 409 *
 410 * SMP: Same implicit races as above.
 411 */
 412long __sys_setgid(gid_t gid)
 413{
 414	struct user_namespace *ns = current_user_ns();
 415	const struct cred *old;
 416	struct cred *new;
 417	int retval;
 418	kgid_t kgid;
 419
 420	kgid = make_kgid(ns, gid);
 421	if (!gid_valid(kgid))
 422		return -EINVAL;
 423
 424	new = prepare_creds();
 425	if (!new)
 426		return -ENOMEM;
 427	old = current_cred();
 428
 429	retval = -EPERM;
 430	if (ns_capable(old->user_ns, CAP_SETGID))
 431		new->gid = new->egid = new->sgid = new->fsgid = kgid;
 432	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
 433		new->egid = new->fsgid = kgid;
 434	else
 435		goto error;
 436
 437	return commit_creds(new);
 438
 439error:
 440	abort_creds(new);
 441	return retval;
 442}
 443
 444SYSCALL_DEFINE1(setgid, gid_t, gid)
 445{
 446	return __sys_setgid(gid);
 447}
 448
 449/*
 450 * change the user struct in a credentials set to match the new UID
 451 */
 452static int set_user(struct cred *new)
 453{
 454	struct user_struct *new_user;
 455
 456	new_user = alloc_uid(new->uid);
 457	if (!new_user)
 458		return -EAGAIN;
 459
 460	/*
 461	 * We don't fail in case of NPROC limit excess here because too many
 462	 * poorly written programs don't check set*uid() return code, assuming
 463	 * it never fails if called by root.  We may still enforce NPROC limit
 464	 * for programs doing set*uid()+execve() by harmlessly deferring the
 465	 * failure to the execve() stage.
 466	 */
 467	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
 468			new_user != INIT_USER)
 469		current->flags |= PF_NPROC_EXCEEDED;
 470	else
 471		current->flags &= ~PF_NPROC_EXCEEDED;
 472
 473	free_uid(new->user);
 474	new->user = new_user;
 475	return 0;
 476}
 477
 478/*
 479 * Unprivileged users may change the real uid to the effective uid
 480 * or vice versa.  (BSD-style)
 481 *
 482 * If you set the real uid at all, or set the effective uid to a value not
 483 * equal to the real uid, then the saved uid is set to the new effective uid.
 484 *
 485 * This makes it possible for a setuid program to completely drop its
 486 * privileges, which is often a useful assertion to make when you are doing
 487 * a security audit over a program.
 488 *
 489 * The general idea is that a program which uses just setreuid() will be
 490 * 100% compatible with BSD.  A program which uses just setuid() will be
 491 * 100% compatible with POSIX with saved IDs.
 492 */
 493long __sys_setreuid(uid_t ruid, uid_t euid)
 494{
 495	struct user_namespace *ns = current_user_ns();
 496	const struct cred *old;
 497	struct cred *new;
 498	int retval;
 499	kuid_t kruid, keuid;
 500
 501	kruid = make_kuid(ns, ruid);
 502	keuid = make_kuid(ns, euid);
 503
 504	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 505		return -EINVAL;
 506	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 507		return -EINVAL;
 508
 509	new = prepare_creds();
 510	if (!new)
 511		return -ENOMEM;
 512	old = current_cred();
 513
 514	retval = -EPERM;
 515	if (ruid != (uid_t) -1) {
 516		new->uid = kruid;
 517		if (!uid_eq(old->uid, kruid) &&
 518		    !uid_eq(old->euid, kruid) &&
 519		    !ns_capable(old->user_ns, CAP_SETUID))
 520			goto error;
 521	}
 522
 523	if (euid != (uid_t) -1) {
 524		new->euid = keuid;
 525		if (!uid_eq(old->uid, keuid) &&
 526		    !uid_eq(old->euid, keuid) &&
 527		    !uid_eq(old->suid, keuid) &&
 528		    !ns_capable(old->user_ns, CAP_SETUID))
 529			goto error;
 530	}
 531
 532	if (!uid_eq(new->uid, old->uid)) {
 533		retval = set_user(new);
 534		if (retval < 0)
 535			goto error;
 536	}
 537	if (ruid != (uid_t) -1 ||
 538	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
 539		new->suid = new->euid;
 540	new->fsuid = new->euid;
 541
 542	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
 543	if (retval < 0)
 544		goto error;
 545
 546	return commit_creds(new);
 547
 548error:
 549	abort_creds(new);
 550	return retval;
 551}
 552
 553SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
 554{
 555	return __sys_setreuid(ruid, euid);
 556}
 557
 558/*
 559 * setuid() is implemented like SysV with SAVED_IDS
 560 *
 561 * Note that SAVED_ID's is deficient in that a setuid root program
 562 * like sendmail, for example, cannot set its uid to be a normal
 563 * user and then switch back, because if you're root, setuid() sets
 564 * the saved uid too.  If you don't like this, blame the bright people
 565 * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
 566 * will allow a root program to temporarily drop privileges and be able to
 567 * regain them by swapping the real and effective uid.
 568 */
 569long __sys_setuid(uid_t uid)
 570{
 571	struct user_namespace *ns = current_user_ns();
 572	const struct cred *old;
 573	struct cred *new;
 574	int retval;
 575	kuid_t kuid;
 576
 577	kuid = make_kuid(ns, uid);
 578	if (!uid_valid(kuid))
 579		return -EINVAL;
 580
 581	new = prepare_creds();
 582	if (!new)
 583		return -ENOMEM;
 584	old = current_cred();
 585
 586	retval = -EPERM;
 587	if (ns_capable(old->user_ns, CAP_SETUID)) {
 588		new->suid = new->uid = kuid;
 589		if (!uid_eq(kuid, old->uid)) {
 590			retval = set_user(new);
 591			if (retval < 0)
 592				goto error;
 593		}
 594	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
 595		goto error;
 596	}
 597
 598	new->fsuid = new->euid = kuid;
 599
 600	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
 601	if (retval < 0)
 602		goto error;
 603
 604	return commit_creds(new);
 605
 606error:
 607	abort_creds(new);
 608	return retval;
 609}
 610
 611SYSCALL_DEFINE1(setuid, uid_t, uid)
 612{
 613	return __sys_setuid(uid);
 614}
 615
 616
 617/*
 618 * This function implements a generic ability to update ruid, euid,
 619 * and suid.  This allows you to implement the 4.4 compatible seteuid().
 620 */
 621long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
 622{
 623	struct user_namespace *ns = current_user_ns();
 624	const struct cred *old;
 625	struct cred *new;
 626	int retval;
 627	kuid_t kruid, keuid, ksuid;
 628
 629	kruid = make_kuid(ns, ruid);
 630	keuid = make_kuid(ns, euid);
 631	ksuid = make_kuid(ns, suid);
 632
 633	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
 634		return -EINVAL;
 635
 636	if ((euid != (uid_t) -1) && !uid_valid(keuid))
 637		return -EINVAL;
 638
 639	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
 640		return -EINVAL;
 641
 642	new = prepare_creds();
 643	if (!new)
 644		return -ENOMEM;
 645
 646	old = current_cred();
 647
 648	retval = -EPERM;
 649	if (!ns_capable(old->user_ns, CAP_SETUID)) {
 650		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
 651		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
 652			goto error;
 653		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
 654		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
 655			goto error;
 656		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
 657		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
 658			goto error;
 659	}
 660
 661	if (ruid != (uid_t) -1) {
 662		new->uid = kruid;
 663		if (!uid_eq(kruid, old->uid)) {
 664			retval = set_user(new);
 665			if (retval < 0)
 666				goto error;
 667		}
 668	}
 669	if (euid != (uid_t) -1)
 670		new->euid = keuid;
 671	if (suid != (uid_t) -1)
 672		new->suid = ksuid;
 673	new->fsuid = new->euid;
 674
 675	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
 676	if (retval < 0)
 677		goto error;
 678
 679	return commit_creds(new);
 680
 681error:
 682	abort_creds(new);
 683	return retval;
 684}
 685
 686SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
 687{
 688	return __sys_setresuid(ruid, euid, suid);
 689}
 690
 691SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
 692{
 693	const struct cred *cred = current_cred();
 694	int retval;
 695	uid_t ruid, euid, suid;
 696
 697	ruid = from_kuid_munged(cred->user_ns, cred->uid);
 698	euid = from_kuid_munged(cred->user_ns, cred->euid);
 699	suid = from_kuid_munged(cred->user_ns, cred->suid);
 700
 701	retval = put_user(ruid, ruidp);
 702	if (!retval) {
 703		retval = put_user(euid, euidp);
 704		if (!retval)
 705			return put_user(suid, suidp);
 706	}
 707	return retval;
 708}
 709
 710/*
 711 * Same as above, but for rgid, egid, sgid.
 712 */
 713long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
 714{
 715	struct user_namespace *ns = current_user_ns();
 716	const struct cred *old;
 717	struct cred *new;
 718	int retval;
 719	kgid_t krgid, kegid, ksgid;
 720
 721	krgid = make_kgid(ns, rgid);
 722	kegid = make_kgid(ns, egid);
 723	ksgid = make_kgid(ns, sgid);
 724
 725	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
 726		return -EINVAL;
 727	if ((egid != (gid_t) -1) && !gid_valid(kegid))
 728		return -EINVAL;
 729	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
 730		return -EINVAL;
 731
 732	new = prepare_creds();
 733	if (!new)
 734		return -ENOMEM;
 735	old = current_cred();
 736
 737	retval = -EPERM;
 738	if (!ns_capable(old->user_ns, CAP_SETGID)) {
 739		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
 740		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
 741			goto error;
 742		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
 743		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
 744			goto error;
 745		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
 746		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
 747			goto error;
 748	}
 749
 750	if (rgid != (gid_t) -1)
 751		new->gid = krgid;
 752	if (egid != (gid_t) -1)
 753		new->egid = kegid;
 754	if (sgid != (gid_t) -1)
 755		new->sgid = ksgid;
 756	new->fsgid = new->egid;
 757
 758	return commit_creds(new);
 759
 760error:
 761	abort_creds(new);
 762	return retval;
 763}
 764
 765SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
 766{
 767	return __sys_setresgid(rgid, egid, sgid);
 768}
 769
 770SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
 771{
 772	const struct cred *cred = current_cred();
 773	int retval;
 774	gid_t rgid, egid, sgid;
 775
 776	rgid = from_kgid_munged(cred->user_ns, cred->gid);
 777	egid = from_kgid_munged(cred->user_ns, cred->egid);
 778	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
 779
 780	retval = put_user(rgid, rgidp);
 781	if (!retval) {
 782		retval = put_user(egid, egidp);
 783		if (!retval)
 784			retval = put_user(sgid, sgidp);
 785	}
 786
 787	return retval;
 788}
 789
 790
 791/*
 792 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
 793 * is used for "access()" and for the NFS daemon (letting nfsd stay at
 794 * whatever uid it wants to). It normally shadows "euid", except when
 795 * explicitly set by setfsuid() or for access..
 796 */
 797long __sys_setfsuid(uid_t uid)
 798{
 799	const struct cred *old;
 800	struct cred *new;
 801	uid_t old_fsuid;
 802	kuid_t kuid;
 803
 804	old = current_cred();
 805	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
 806
 807	kuid = make_kuid(old->user_ns, uid);
 808	if (!uid_valid(kuid))
 809		return old_fsuid;
 810
 811	new = prepare_creds();
 812	if (!new)
 813		return old_fsuid;
 814
 815	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
 816	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
 817	    ns_capable(old->user_ns, CAP_SETUID)) {
 818		if (!uid_eq(kuid, old->fsuid)) {
 819			new->fsuid = kuid;
 820			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
 821				goto change_okay;
 822		}
 823	}
 824
 825	abort_creds(new);
 826	return old_fsuid;
 827
 828change_okay:
 829	commit_creds(new);
 830	return old_fsuid;
 831}
 832
 833SYSCALL_DEFINE1(setfsuid, uid_t, uid)
 834{
 835	return __sys_setfsuid(uid);
 836}
 837
 838/*
 839 * Samma på svenska..
 840 */
 841long __sys_setfsgid(gid_t gid)
 842{
 843	const struct cred *old;
 844	struct cred *new;
 845	gid_t old_fsgid;
 846	kgid_t kgid;
 847
 848	old = current_cred();
 849	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
 850
 851	kgid = make_kgid(old->user_ns, gid);
 852	if (!gid_valid(kgid))
 853		return old_fsgid;
 854
 855	new = prepare_creds();
 856	if (!new)
 857		return old_fsgid;
 858
 859	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
 860	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
 861	    ns_capable(old->user_ns, CAP_SETGID)) {
 862		if (!gid_eq(kgid, old->fsgid)) {
 863			new->fsgid = kgid;
 864			goto change_okay;
 865		}
 866	}
 867
 868	abort_creds(new);
 869	return old_fsgid;
 870
 871change_okay:
 872	commit_creds(new);
 873	return old_fsgid;
 874}
 875
 876SYSCALL_DEFINE1(setfsgid, gid_t, gid)
 877{
 878	return __sys_setfsgid(gid);
 879}
 880#endif /* CONFIG_MULTIUSER */
 881
 882/**
 883 * sys_getpid - return the thread group id of the current process
 884 *
 885 * Note, despite the name, this returns the tgid not the pid.  The tgid and
 886 * the pid are identical unless CLONE_THREAD was specified on clone() in
 887 * which case the tgid is the same in all threads of the same group.
 888 *
 889 * This is SMP safe as current->tgid does not change.
 890 */
 891SYSCALL_DEFINE0(getpid)
 892{
 893	return task_tgid_vnr(current);
 894}
 895
 896/* Thread ID - the internal kernel "pid" */
 897SYSCALL_DEFINE0(gettid)
 898{
 899	return task_pid_vnr(current);
 900}
 901
 902/*
 903 * Accessing ->real_parent is not SMP-safe, it could
 904 * change from under us. However, we can use a stale
 905 * value of ->real_parent under rcu_read_lock(), see
 906 * release_task()->call_rcu(delayed_put_task_struct).
 907 */
 908SYSCALL_DEFINE0(getppid)
 909{
 910	int pid;
 911
 912	rcu_read_lock();
 913	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
 914	rcu_read_unlock();
 915
 916	return pid;
 917}
 918
 919SYSCALL_DEFINE0(getuid)
 920{
 921	/* Only we change this so SMP safe */
 922	return from_kuid_munged(current_user_ns(), current_uid());
 923}
 924
 925SYSCALL_DEFINE0(geteuid)
 926{
 927	/* Only we change this so SMP safe */
 928	return from_kuid_munged(current_user_ns(), current_euid());
 929}
 930
 931SYSCALL_DEFINE0(getgid)
 932{
 933	/* Only we change this so SMP safe */
 934	return from_kgid_munged(current_user_ns(), current_gid());
 935}
 936
 937SYSCALL_DEFINE0(getegid)
 938{
 939	/* Only we change this so SMP safe */
 940	return from_kgid_munged(current_user_ns(), current_egid());
 941}
 942
 943static void do_sys_times(struct tms *tms)
 944{
 945	u64 tgutime, tgstime, cutime, cstime;
 946
 947	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
 948	cutime = current->signal->cutime;
 949	cstime = current->signal->cstime;
 950	tms->tms_utime = nsec_to_clock_t(tgutime);
 951	tms->tms_stime = nsec_to_clock_t(tgstime);
 952	tms->tms_cutime = nsec_to_clock_t(cutime);
 953	tms->tms_cstime = nsec_to_clock_t(cstime);
 
 954}
 955
 956SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 957{
 958	if (tbuf) {
 959		struct tms tmp;
 960
 961		do_sys_times(&tmp);
 962		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
 963			return -EFAULT;
 964	}
 965	force_successful_syscall_return();
 966	return (long) jiffies_64_to_clock_t(get_jiffies_64());
 967}
 968
 969#ifdef CONFIG_COMPAT
 970static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
 971{
 972	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
 973}
 974
 975COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
 976{
 977	if (tbuf) {
 978		struct tms tms;
 979		struct compat_tms tmp;
 980
 981		do_sys_times(&tms);
 982		/* Convert our struct tms to the compat version. */
 983		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
 984		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
 985		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
 986		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
 987		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
 988			return -EFAULT;
 989	}
 990	force_successful_syscall_return();
 991	return compat_jiffies_to_clock_t(jiffies);
 992}
 993#endif
 994
 995/*
 996 * This needs some heavy checking ...
 997 * I just haven't the stomach for it. I also don't fully
 998 * understand sessions/pgrp etc. Let somebody who does explain it.
 999 *
1000 * OK, I think I have the protection semantics right.... this is really
1001 * only important on a multi-user system anyway, to make sure one user
1002 * can't send a signal to a process owned by another.  -TYT, 12/12/91
1003 *
1004 * !PF_FORKNOEXEC check to conform completely to POSIX.
 
1005 */
1006SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1007{
1008	struct task_struct *p;
1009	struct task_struct *group_leader = current->group_leader;
1010	struct pid *pgrp;
1011	int err;
1012
1013	if (!pid)
1014		pid = task_pid_vnr(group_leader);
1015	if (!pgid)
1016		pgid = pid;
1017	if (pgid < 0)
1018		return -EINVAL;
1019	rcu_read_lock();
1020
1021	/* From this point forward we keep holding onto the tasklist lock
1022	 * so that our parent does not change from under us. -DaveM
1023	 */
1024	write_lock_irq(&tasklist_lock);
1025
1026	err = -ESRCH;
1027	p = find_task_by_vpid(pid);
1028	if (!p)
1029		goto out;
1030
1031	err = -EINVAL;
1032	if (!thread_group_leader(p))
1033		goto out;
1034
1035	if (same_thread_group(p->real_parent, group_leader)) {
1036		err = -EPERM;
1037		if (task_session(p) != task_session(group_leader))
1038			goto out;
1039		err = -EACCES;
1040		if (!(p->flags & PF_FORKNOEXEC))
1041			goto out;
1042	} else {
1043		err = -ESRCH;
1044		if (p != group_leader)
1045			goto out;
1046	}
1047
1048	err = -EPERM;
1049	if (p->signal->leader)
1050		goto out;
1051
1052	pgrp = task_pid(p);
1053	if (pgid != pid) {
1054		struct task_struct *g;
1055
1056		pgrp = find_vpid(pgid);
1057		g = pid_task(pgrp, PIDTYPE_PGID);
1058		if (!g || task_session(g) != task_session(group_leader))
1059			goto out;
1060	}
1061
1062	err = security_task_setpgid(p, pgid);
1063	if (err)
1064		goto out;
1065
1066	if (task_pgrp(p) != pgrp)
1067		change_pid(p, PIDTYPE_PGID, pgrp);
1068
1069	err = 0;
1070out:
1071	/* All paths lead to here, thus we are safe. -DaveM */
1072	write_unlock_irq(&tasklist_lock);
1073	rcu_read_unlock();
1074	return err;
1075}
1076
1077static int do_getpgid(pid_t pid)
1078{
1079	struct task_struct *p;
1080	struct pid *grp;
1081	int retval;
1082
1083	rcu_read_lock();
1084	if (!pid)
1085		grp = task_pgrp(current);
1086	else {
1087		retval = -ESRCH;
1088		p = find_task_by_vpid(pid);
1089		if (!p)
1090			goto out;
1091		grp = task_pgrp(p);
1092		if (!grp)
1093			goto out;
1094
1095		retval = security_task_getpgid(p);
1096		if (retval)
1097			goto out;
1098	}
1099	retval = pid_vnr(grp);
1100out:
1101	rcu_read_unlock();
1102	return retval;
1103}
1104
1105SYSCALL_DEFINE1(getpgid, pid_t, pid)
1106{
1107	return do_getpgid(pid);
1108}
1109
1110#ifdef __ARCH_WANT_SYS_GETPGRP
1111
1112SYSCALL_DEFINE0(getpgrp)
1113{
1114	return do_getpgid(0);
1115}
1116
1117#endif
1118
1119SYSCALL_DEFINE1(getsid, pid_t, pid)
1120{
1121	struct task_struct *p;
1122	struct pid *sid;
1123	int retval;
1124
1125	rcu_read_lock();
1126	if (!pid)
1127		sid = task_session(current);
1128	else {
1129		retval = -ESRCH;
1130		p = find_task_by_vpid(pid);
1131		if (!p)
1132			goto out;
1133		sid = task_session(p);
1134		if (!sid)
1135			goto out;
1136
1137		retval = security_task_getsid(p);
1138		if (retval)
1139			goto out;
1140	}
1141	retval = pid_vnr(sid);
1142out:
1143	rcu_read_unlock();
1144	return retval;
1145}
1146
1147static void set_special_pids(struct pid *pid)
1148{
1149	struct task_struct *curr = current->group_leader;
1150
1151	if (task_session(curr) != pid)
1152		change_pid(curr, PIDTYPE_SID, pid);
1153
1154	if (task_pgrp(curr) != pid)
1155		change_pid(curr, PIDTYPE_PGID, pid);
1156}
1157
1158int ksys_setsid(void)
1159{
1160	struct task_struct *group_leader = current->group_leader;
1161	struct pid *sid = task_pid(group_leader);
1162	pid_t session = pid_vnr(sid);
1163	int err = -EPERM;
1164
1165	write_lock_irq(&tasklist_lock);
1166	/* Fail if I am already a session leader */
1167	if (group_leader->signal->leader)
1168		goto out;
1169
1170	/* Fail if a process group id already exists that equals the
1171	 * proposed session id.
1172	 */
1173	if (pid_task(sid, PIDTYPE_PGID))
1174		goto out;
1175
1176	group_leader->signal->leader = 1;
1177	set_special_pids(sid);
1178
1179	proc_clear_tty(group_leader);
1180
1181	err = session;
1182out:
1183	write_unlock_irq(&tasklist_lock);
1184	if (err > 0) {
1185		proc_sid_connector(group_leader);
1186		sched_autogroup_create_attach(group_leader);
1187	}
1188	return err;
1189}
1190
1191SYSCALL_DEFINE0(setsid)
1192{
1193	return ksys_setsid();
1194}
1195
1196DECLARE_RWSEM(uts_sem);
1197
1198#ifdef COMPAT_UTS_MACHINE
1199#define override_architecture(name) \
1200	(personality(current->personality) == PER_LINUX32 && \
1201	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1202		      sizeof(COMPAT_UTS_MACHINE)))
1203#else
1204#define override_architecture(name)	0
1205#endif
1206
1207/*
1208 * Work around broken programs that cannot handle "Linux 3.0".
1209 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1210 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1211 */
1212static int override_release(char __user *release, size_t len)
1213{
1214	int ret = 0;
 
1215
1216	if (current->personality & UNAME26) {
1217		const char *rest = UTS_RELEASE;
1218		char buf[65] = { 0 };
1219		int ndots = 0;
1220		unsigned v;
1221		size_t copy;
1222
1223		while (*rest) {
1224			if (*rest == '.' && ++ndots >= 3)
1225				break;
1226			if (!isdigit(*rest) && *rest != '.')
1227				break;
1228			rest++;
1229		}
1230		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1231		copy = clamp_t(size_t, len, 1, sizeof(buf));
1232		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1233		ret = copy_to_user(release, buf, copy + 1);
1234	}
1235	return ret;
1236}
1237
1238SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1239{
1240	int errno = 0;
1241
1242	down_read(&uts_sem);
1243	if (copy_to_user(name, utsname(), sizeof *name))
1244		errno = -EFAULT;
1245	up_read(&uts_sem);
1246
1247	if (!errno && override_release(name->release, sizeof(name->release)))
1248		errno = -EFAULT;
1249	if (!errno && override_architecture(name))
1250		errno = -EFAULT;
1251	return errno;
1252}
1253
1254#ifdef __ARCH_WANT_SYS_OLD_UNAME
1255/*
1256 * Old cruft
1257 */
1258SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1259{
1260	int error = 0;
1261
1262	if (!name)
1263		return -EFAULT;
1264
1265	down_read(&uts_sem);
1266	if (copy_to_user(name, utsname(), sizeof(*name)))
1267		error = -EFAULT;
1268	up_read(&uts_sem);
1269
1270	if (!error && override_release(name->release, sizeof(name->release)))
1271		error = -EFAULT;
1272	if (!error && override_architecture(name))
1273		error = -EFAULT;
1274	return error;
1275}
1276
1277SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1278{
1279	int error;
1280
1281	if (!name)
1282		return -EFAULT;
1283	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1284		return -EFAULT;
1285
1286	down_read(&uts_sem);
1287	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1288			       __OLD_UTS_LEN);
1289	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1290	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1291				__OLD_UTS_LEN);
1292	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1293	error |= __copy_to_user(&name->release, &utsname()->release,
1294				__OLD_UTS_LEN);
1295	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1296	error |= __copy_to_user(&name->version, &utsname()->version,
1297				__OLD_UTS_LEN);
1298	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1299	error |= __copy_to_user(&name->machine, &utsname()->machine,
1300				__OLD_UTS_LEN);
1301	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1302	up_read(&uts_sem);
1303
1304	if (!error && override_architecture(name))
1305		error = -EFAULT;
1306	if (!error && override_release(name->release, sizeof(name->release)))
1307		error = -EFAULT;
1308	return error ? -EFAULT : 0;
1309}
1310#endif
1311
1312SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1313{
1314	int errno;
1315	char tmp[__NEW_UTS_LEN];
1316
1317	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1318		return -EPERM;
1319
1320	if (len < 0 || len > __NEW_UTS_LEN)
1321		return -EINVAL;
1322	down_write(&uts_sem);
1323	errno = -EFAULT;
1324	if (!copy_from_user(tmp, name, len)) {
1325		struct new_utsname *u = utsname();
1326
1327		memcpy(u->nodename, tmp, len);
1328		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1329		errno = 0;
1330		uts_proc_notify(UTS_PROC_HOSTNAME);
1331	}
1332	up_write(&uts_sem);
1333	return errno;
1334}
1335
1336#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1337
1338SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1339{
1340	int i, errno;
1341	struct new_utsname *u;
1342
1343	if (len < 0)
1344		return -EINVAL;
1345	down_read(&uts_sem);
1346	u = utsname();
1347	i = 1 + strlen(u->nodename);
1348	if (i > len)
1349		i = len;
1350	errno = 0;
1351	if (copy_to_user(name, u->nodename, i))
1352		errno = -EFAULT;
1353	up_read(&uts_sem);
1354	return errno;
1355}
1356
1357#endif
1358
1359/*
1360 * Only setdomainname; getdomainname can be implemented by calling
1361 * uname()
1362 */
1363SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1364{
1365	int errno;
1366	char tmp[__NEW_UTS_LEN];
1367
1368	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1369		return -EPERM;
1370	if (len < 0 || len > __NEW_UTS_LEN)
1371		return -EINVAL;
1372
1373	down_write(&uts_sem);
1374	errno = -EFAULT;
1375	if (!copy_from_user(tmp, name, len)) {
1376		struct new_utsname *u = utsname();
1377
1378		memcpy(u->domainname, tmp, len);
1379		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1380		errno = 0;
1381		uts_proc_notify(UTS_PROC_DOMAINNAME);
1382	}
1383	up_write(&uts_sem);
1384	return errno;
1385}
1386
1387SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1388{
1389	struct rlimit value;
1390	int ret;
1391
1392	ret = do_prlimit(current, resource, NULL, &value);
1393	if (!ret)
1394		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1395
1396	return ret;
1397}
1398
1399#ifdef CONFIG_COMPAT
1400
1401COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1402		       struct compat_rlimit __user *, rlim)
1403{
1404	struct rlimit r;
1405	struct compat_rlimit r32;
1406
1407	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1408		return -EFAULT;
1409
1410	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1411		r.rlim_cur = RLIM_INFINITY;
1412	else
1413		r.rlim_cur = r32.rlim_cur;
1414	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1415		r.rlim_max = RLIM_INFINITY;
1416	else
1417		r.rlim_max = r32.rlim_max;
1418	return do_prlimit(current, resource, &r, NULL);
1419}
1420
1421COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1422		       struct compat_rlimit __user *, rlim)
1423{
1424	struct rlimit r;
1425	int ret;
1426
1427	ret = do_prlimit(current, resource, NULL, &r);
1428	if (!ret) {
1429		struct compat_rlimit r32;
1430		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1431			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1432		else
1433			r32.rlim_cur = r.rlim_cur;
1434		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1435			r32.rlim_max = COMPAT_RLIM_INFINITY;
1436		else
1437			r32.rlim_max = r.rlim_max;
1438
1439		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1440			return -EFAULT;
1441	}
1442	return ret;
1443}
1444
1445#endif
1446
1447#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1448
1449/*
1450 *	Back compatibility for getrlimit. Needed for some apps.
1451 */
 
1452SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1453		struct rlimit __user *, rlim)
1454{
1455	struct rlimit x;
1456	if (resource >= RLIM_NLIMITS)
1457		return -EINVAL;
1458
1459	resource = array_index_nospec(resource, RLIM_NLIMITS);
1460	task_lock(current->group_leader);
1461	x = current->signal->rlim[resource];
1462	task_unlock(current->group_leader);
1463	if (x.rlim_cur > 0x7FFFFFFF)
1464		x.rlim_cur = 0x7FFFFFFF;
1465	if (x.rlim_max > 0x7FFFFFFF)
1466		x.rlim_max = 0x7FFFFFFF;
1467	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1468}
1469
1470#ifdef CONFIG_COMPAT
1471COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1472		       struct compat_rlimit __user *, rlim)
1473{
1474	struct rlimit r;
1475
1476	if (resource >= RLIM_NLIMITS)
1477		return -EINVAL;
1478
1479	resource = array_index_nospec(resource, RLIM_NLIMITS);
1480	task_lock(current->group_leader);
1481	r = current->signal->rlim[resource];
1482	task_unlock(current->group_leader);
1483	if (r.rlim_cur > 0x7FFFFFFF)
1484		r.rlim_cur = 0x7FFFFFFF;
1485	if (r.rlim_max > 0x7FFFFFFF)
1486		r.rlim_max = 0x7FFFFFFF;
1487
1488	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1489	    put_user(r.rlim_max, &rlim->rlim_max))
1490		return -EFAULT;
1491	return 0;
1492}
1493#endif
1494
1495#endif
1496
1497static inline bool rlim64_is_infinity(__u64 rlim64)
1498{
1499#if BITS_PER_LONG < 64
1500	return rlim64 >= ULONG_MAX;
1501#else
1502	return rlim64 == RLIM64_INFINITY;
1503#endif
1504}
1505
1506static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1507{
1508	if (rlim->rlim_cur == RLIM_INFINITY)
1509		rlim64->rlim_cur = RLIM64_INFINITY;
1510	else
1511		rlim64->rlim_cur = rlim->rlim_cur;
1512	if (rlim->rlim_max == RLIM_INFINITY)
1513		rlim64->rlim_max = RLIM64_INFINITY;
1514	else
1515		rlim64->rlim_max = rlim->rlim_max;
1516}
1517
1518static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1519{
1520	if (rlim64_is_infinity(rlim64->rlim_cur))
1521		rlim->rlim_cur = RLIM_INFINITY;
1522	else
1523		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1524	if (rlim64_is_infinity(rlim64->rlim_max))
1525		rlim->rlim_max = RLIM_INFINITY;
1526	else
1527		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1528}
1529
1530/* make sure you are allowed to change @tsk limits before calling this */
1531int do_prlimit(struct task_struct *tsk, unsigned int resource,
1532		struct rlimit *new_rlim, struct rlimit *old_rlim)
1533{
1534	struct rlimit *rlim;
1535	int retval = 0;
1536
1537	if (resource >= RLIM_NLIMITS)
1538		return -EINVAL;
1539	if (new_rlim) {
1540		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1541			return -EINVAL;
1542		if (resource == RLIMIT_NOFILE &&
1543				new_rlim->rlim_max > sysctl_nr_open)
1544			return -EPERM;
1545	}
1546
1547	/* protect tsk->signal and tsk->sighand from disappearing */
1548	read_lock(&tasklist_lock);
1549	if (!tsk->sighand) {
1550		retval = -ESRCH;
1551		goto out;
1552	}
1553
1554	rlim = tsk->signal->rlim + resource;
1555	task_lock(tsk->group_leader);
1556	if (new_rlim) {
1557		/* Keep the capable check against init_user_ns until
1558		   cgroups can contain all limits */
1559		if (new_rlim->rlim_max > rlim->rlim_max &&
1560				!capable(CAP_SYS_RESOURCE))
1561			retval = -EPERM;
1562		if (!retval)
1563			retval = security_task_setrlimit(tsk, resource, new_rlim);
 
1564		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1565			/*
1566			 * The caller is asking for an immediate RLIMIT_CPU
1567			 * expiry.  But we use the zero value to mean "it was
1568			 * never set".  So let's cheat and make it one second
1569			 * instead
1570			 */
1571			new_rlim->rlim_cur = 1;
1572		}
1573	}
1574	if (!retval) {
1575		if (old_rlim)
1576			*old_rlim = *rlim;
1577		if (new_rlim)
1578			*rlim = *new_rlim;
1579	}
1580	task_unlock(tsk->group_leader);
1581
1582	/*
1583	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1584	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1585	 * very long-standing error, and fixing it now risks breakage of
1586	 * applications, so we live with it
1587	 */
1588	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1589	     new_rlim->rlim_cur != RLIM_INFINITY &&
1590	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1591		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1592out:
1593	read_unlock(&tasklist_lock);
1594	return retval;
1595}
1596
1597/* rcu lock must be held */
1598static int check_prlimit_permission(struct task_struct *task,
1599				    unsigned int flags)
1600{
1601	const struct cred *cred = current_cred(), *tcred;
1602	bool id_match;
1603
1604	if (current == task)
1605		return 0;
1606
1607	tcred = __task_cred(task);
1608	id_match = (uid_eq(cred->uid, tcred->euid) &&
1609		    uid_eq(cred->uid, tcred->suid) &&
1610		    uid_eq(cred->uid, tcred->uid)  &&
1611		    gid_eq(cred->gid, tcred->egid) &&
1612		    gid_eq(cred->gid, tcred->sgid) &&
1613		    gid_eq(cred->gid, tcred->gid));
1614	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1615		return -EPERM;
 
1616
1617	return security_task_prlimit(cred, tcred, flags);
1618}
1619
1620SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1621		const struct rlimit64 __user *, new_rlim,
1622		struct rlimit64 __user *, old_rlim)
1623{
1624	struct rlimit64 old64, new64;
1625	struct rlimit old, new;
1626	struct task_struct *tsk;
1627	unsigned int checkflags = 0;
1628	int ret;
1629
1630	if (old_rlim)
1631		checkflags |= LSM_PRLIMIT_READ;
1632
1633	if (new_rlim) {
1634		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1635			return -EFAULT;
1636		rlim64_to_rlim(&new64, &new);
1637		checkflags |= LSM_PRLIMIT_WRITE;
1638	}
1639
1640	rcu_read_lock();
1641	tsk = pid ? find_task_by_vpid(pid) : current;
1642	if (!tsk) {
1643		rcu_read_unlock();
1644		return -ESRCH;
1645	}
1646	ret = check_prlimit_permission(tsk, checkflags);
1647	if (ret) {
1648		rcu_read_unlock();
1649		return ret;
1650	}
1651	get_task_struct(tsk);
1652	rcu_read_unlock();
1653
1654	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1655			old_rlim ? &old : NULL);
1656
1657	if (!ret && old_rlim) {
1658		rlim_to_rlim64(&old, &old64);
1659		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1660			ret = -EFAULT;
1661	}
1662
1663	put_task_struct(tsk);
1664	return ret;
1665}
1666
1667SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1668{
1669	struct rlimit new_rlim;
1670
1671	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1672		return -EFAULT;
1673	return do_prlimit(current, resource, &new_rlim, NULL);
1674}
1675
1676/*
1677 * It would make sense to put struct rusage in the task_struct,
1678 * except that would make the task_struct be *really big*.  After
1679 * task_struct gets moved into malloc'ed memory, it would
1680 * make sense to do this.  It will make moving the rest of the information
1681 * a lot simpler!  (Which we're not doing right now because we're not
1682 * measuring them yet).
1683 *
1684 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1685 * races with threads incrementing their own counters.  But since word
1686 * reads are atomic, we either get new values or old values and we don't
1687 * care which for the sums.  We always take the siglock to protect reading
1688 * the c* fields from p->signal from races with exit.c updating those
1689 * fields when reaping, so a sample either gets all the additions of a
1690 * given child after it's reaped, or none so this sample is before reaping.
1691 *
1692 * Locking:
1693 * We need to take the siglock for CHILDEREN, SELF and BOTH
1694 * for  the cases current multithreaded, non-current single threaded
1695 * non-current multithreaded.  Thread traversal is now safe with
1696 * the siglock held.
1697 * Strictly speaking, we donot need to take the siglock if we are current and
1698 * single threaded,  as no one else can take our signal_struct away, no one
1699 * else can  reap the  children to update signal->c* counters, and no one else
1700 * can race with the signal-> fields. If we do not take any lock, the
1701 * signal-> fields could be read out of order while another thread was just
1702 * exiting. So we should  place a read memory barrier when we avoid the lock.
1703 * On the writer side,  write memory barrier is implied in  __exit_signal
1704 * as __exit_signal releases  the siglock spinlock after updating the signal->
1705 * fields. But we don't do this yet to keep things simple.
1706 *
1707 */
1708
1709static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1710{
1711	r->ru_nvcsw += t->nvcsw;
1712	r->ru_nivcsw += t->nivcsw;
1713	r->ru_minflt += t->min_flt;
1714	r->ru_majflt += t->maj_flt;
1715	r->ru_inblock += task_io_get_inblock(t);
1716	r->ru_oublock += task_io_get_oublock(t);
1717}
1718
1719void getrusage(struct task_struct *p, int who, struct rusage *r)
1720{
1721	struct task_struct *t;
1722	unsigned long flags;
1723	u64 tgutime, tgstime, utime, stime;
1724	unsigned long maxrss = 0;
1725
1726	memset((char *)r, 0, sizeof (*r));
1727	utime = stime = 0;
1728
1729	if (who == RUSAGE_THREAD) {
1730		task_cputime_adjusted(current, &utime, &stime);
1731		accumulate_thread_rusage(p, r);
1732		maxrss = p->signal->maxrss;
1733		goto out;
1734	}
1735
1736	if (!lock_task_sighand(p, &flags))
1737		return;
1738
1739	switch (who) {
1740	case RUSAGE_BOTH:
1741	case RUSAGE_CHILDREN:
1742		utime = p->signal->cutime;
1743		stime = p->signal->cstime;
1744		r->ru_nvcsw = p->signal->cnvcsw;
1745		r->ru_nivcsw = p->signal->cnivcsw;
1746		r->ru_minflt = p->signal->cmin_flt;
1747		r->ru_majflt = p->signal->cmaj_flt;
1748		r->ru_inblock = p->signal->cinblock;
1749		r->ru_oublock = p->signal->coublock;
1750		maxrss = p->signal->cmaxrss;
1751
1752		if (who == RUSAGE_CHILDREN)
1753			break;
1754
1755	case RUSAGE_SELF:
1756		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1757		utime += tgutime;
1758		stime += tgstime;
1759		r->ru_nvcsw += p->signal->nvcsw;
1760		r->ru_nivcsw += p->signal->nivcsw;
1761		r->ru_minflt += p->signal->min_flt;
1762		r->ru_majflt += p->signal->maj_flt;
1763		r->ru_inblock += p->signal->inblock;
1764		r->ru_oublock += p->signal->oublock;
1765		if (maxrss < p->signal->maxrss)
1766			maxrss = p->signal->maxrss;
1767		t = p;
1768		do {
1769			accumulate_thread_rusage(t, r);
1770		} while_each_thread(p, t);
1771		break;
1772
1773	default:
1774		BUG();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1775	}
1776	unlock_task_sighand(p, &flags);
1777
1778out:
1779	r->ru_utime = ns_to_timeval(utime);
1780	r->ru_stime = ns_to_timeval(stime);
1781
1782	if (who != RUSAGE_CHILDREN) {
1783		struct mm_struct *mm = get_task_mm(p);
1784
1785		if (mm) {
1786			setmax_mm_hiwater_rss(&maxrss, mm);
1787			mmput(mm);
1788		}
1789	}
1790	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1791}
1792
1793SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1794{
1795	struct rusage r;
1796
1797	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1798	    who != RUSAGE_THREAD)
1799		return -EINVAL;
1800
1801	getrusage(current, who, &r);
1802	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1803}
1804
1805#ifdef CONFIG_COMPAT
1806COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1807{
1808	struct rusage r;
1809
1810	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1811	    who != RUSAGE_THREAD)
1812		return -EINVAL;
1813
1814	getrusage(current, who, &r);
1815	return put_compat_rusage(&r, ru);
1816}
1817#endif
1818
1819SYSCALL_DEFINE1(umask, int, mask)
1820{
1821	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1822	return mask;
1823}
1824
 
1825static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1826{
1827	struct fd exe;
1828	struct file *old_exe, *exe_file;
1829	struct inode *inode;
1830	int err;
1831
1832	exe = fdget(fd);
1833	if (!exe.file)
1834		return -EBADF;
1835
1836	inode = file_inode(exe.file);
1837
1838	/*
1839	 * Because the original mm->exe_file points to executable file, make
1840	 * sure that this one is executable as well, to avoid breaking an
1841	 * overall picture.
1842	 */
1843	err = -EACCES;
1844	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
 
1845		goto exit;
1846
1847	err = inode_permission(inode, MAY_EXEC);
1848	if (err)
1849		goto exit;
1850
 
 
1851	/*
1852	 * Forbid mm->exe_file change if old file still mapped.
1853	 */
1854	exe_file = get_mm_exe_file(mm);
1855	err = -EBUSY;
1856	if (exe_file) {
1857		struct vm_area_struct *vma;
1858
1859		down_read(&mm->mmap_sem);
1860		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1861			if (!vma->vm_file)
1862				continue;
1863			if (path_equal(&vma->vm_file->f_path,
1864				       &exe_file->f_path))
1865				goto exit_err;
1866		}
1867
1868		up_read(&mm->mmap_sem);
1869		fput(exe_file);
1870	}
1871
1872	err = 0;
1873	/* set the new file, lockless */
1874	get_file(exe.file);
1875	old_exe = xchg(&mm->exe_file, exe.file);
1876	if (old_exe)
1877		fput(old_exe);
1878exit:
1879	fdput(exe);
1880	return err;
1881exit_err:
1882	up_read(&mm->mmap_sem);
1883	fput(exe_file);
1884	goto exit;
1885}
1886
1887/*
1888 * WARNING: we don't require any capability here so be very careful
1889 * in what is allowed for modification from userspace.
1890 */
1891static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1892{
1893	unsigned long mmap_max_addr = TASK_SIZE;
1894	struct mm_struct *mm = current->mm;
1895	int error = -EINVAL, i;
1896
1897	static const unsigned char offsets[] = {
1898		offsetof(struct prctl_mm_map, start_code),
1899		offsetof(struct prctl_mm_map, end_code),
1900		offsetof(struct prctl_mm_map, start_data),
1901		offsetof(struct prctl_mm_map, end_data),
1902		offsetof(struct prctl_mm_map, start_brk),
1903		offsetof(struct prctl_mm_map, brk),
1904		offsetof(struct prctl_mm_map, start_stack),
1905		offsetof(struct prctl_mm_map, arg_start),
1906		offsetof(struct prctl_mm_map, arg_end),
1907		offsetof(struct prctl_mm_map, env_start),
1908		offsetof(struct prctl_mm_map, env_end),
1909	};
1910
1911	/*
1912	 * Make sure the members are not somewhere outside
1913	 * of allowed address space.
1914	 */
1915	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1916		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1917
1918		if ((unsigned long)val >= mmap_max_addr ||
1919		    (unsigned long)val < mmap_min_addr)
1920			goto out;
1921	}
1922
1923	/*
1924	 * Make sure the pairs are ordered.
1925	 */
1926#define __prctl_check_order(__m1, __op, __m2)				\
1927	((unsigned long)prctl_map->__m1 __op				\
1928	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1929	error  = __prctl_check_order(start_code, <, end_code);
1930	error |= __prctl_check_order(start_data, <, end_data);
1931	error |= __prctl_check_order(start_brk, <=, brk);
1932	error |= __prctl_check_order(arg_start, <=, arg_end);
1933	error |= __prctl_check_order(env_start, <=, env_end);
1934	if (error)
1935		goto out;
1936#undef __prctl_check_order
1937
1938	error = -EINVAL;
1939
1940	/*
1941	 * @brk should be after @end_data in traditional maps.
1942	 */
1943	if (prctl_map->start_brk <= prctl_map->end_data ||
1944	    prctl_map->brk <= prctl_map->end_data)
1945		goto out;
1946
1947	/*
1948	 * Neither we should allow to override limits if they set.
1949	 */
1950	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1951			      prctl_map->start_brk, prctl_map->end_data,
1952			      prctl_map->start_data))
1953			goto out;
1954
1955	/*
1956	 * Someone is trying to cheat the auxv vector.
1957	 */
1958	if (prctl_map->auxv_size) {
1959		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1960			goto out;
1961	}
1962
1963	/*
1964	 * Finally, make sure the caller has the rights to
1965	 * change /proc/pid/exe link: only local sys admin should
1966	 * be allowed to.
 
1967	 */
1968	if (prctl_map->exe_fd != (u32)-1) {
1969		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1970			goto out;
1971	}
1972
1973	error = 0;
1974out:
1975	return error;
1976}
1977
1978#ifdef CONFIG_CHECKPOINT_RESTORE
1979static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1980{
1981	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1982	unsigned long user_auxv[AT_VECTOR_SIZE];
1983	struct mm_struct *mm = current->mm;
1984	int error;
1985
1986	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1987	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1988
1989	if (opt == PR_SET_MM_MAP_SIZE)
1990		return put_user((unsigned int)sizeof(prctl_map),
1991				(unsigned int __user *)addr);
1992
1993	if (data_size != sizeof(prctl_map))
1994		return -EINVAL;
1995
1996	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1997		return -EFAULT;
1998
1999	error = validate_prctl_map(&prctl_map);
2000	if (error)
2001		return error;
2002
2003	if (prctl_map.auxv_size) {
2004		memset(user_auxv, 0, sizeof(user_auxv));
2005		if (copy_from_user(user_auxv,
2006				   (const void __user *)prctl_map.auxv,
2007				   prctl_map.auxv_size))
2008			return -EFAULT;
2009
2010		/* Last entry must be AT_NULL as specification requires */
2011		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2012		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2013	}
2014
2015	if (prctl_map.exe_fd != (u32)-1) {
2016		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2017		if (error)
2018			return error;
2019	}
2020
2021	down_write(&mm->mmap_sem);
2022
2023	/*
2024	 * We don't validate if these members are pointing to
2025	 * real present VMAs because application may have correspond
2026	 * VMAs already unmapped and kernel uses these members for statistics
2027	 * output in procfs mostly, except
2028	 *
2029	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
2030	 *    for VMAs when updating these memvers so anything wrong written
2031	 *    here cause kernel to swear at userspace program but won't lead
2032	 *    to any problem in kernel itself
2033	 */
2034
2035	mm->start_code	= prctl_map.start_code;
2036	mm->end_code	= prctl_map.end_code;
2037	mm->start_data	= prctl_map.start_data;
2038	mm->end_data	= prctl_map.end_data;
2039	mm->start_brk	= prctl_map.start_brk;
2040	mm->brk		= prctl_map.brk;
2041	mm->start_stack	= prctl_map.start_stack;
2042	mm->arg_start	= prctl_map.arg_start;
2043	mm->arg_end	= prctl_map.arg_end;
2044	mm->env_start	= prctl_map.env_start;
2045	mm->env_end	= prctl_map.env_end;
2046
2047	/*
2048	 * Note this update of @saved_auxv is lockless thus
2049	 * if someone reads this member in procfs while we're
2050	 * updating -- it may get partly updated results. It's
2051	 * known and acceptable trade off: we leave it as is to
2052	 * not introduce additional locks here making the kernel
2053	 * more complex.
2054	 */
2055	if (prctl_map.auxv_size)
2056		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2057
 
 
 
2058	up_write(&mm->mmap_sem);
2059	return 0;
2060}
2061#endif /* CONFIG_CHECKPOINT_RESTORE */
2062
2063static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2064			  unsigned long len)
2065{
2066	/*
2067	 * This doesn't move the auxiliary vector itself since it's pinned to
2068	 * mm_struct, but it permits filling the vector with new values.  It's
2069	 * up to the caller to provide sane values here, otherwise userspace
2070	 * tools which use this vector might be unhappy.
2071	 */
2072	unsigned long user_auxv[AT_VECTOR_SIZE];
2073
2074	if (len > sizeof(user_auxv))
2075		return -EINVAL;
2076
2077	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2078		return -EFAULT;
2079
2080	/* Make sure the last entry is always AT_NULL */
2081	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2082	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2083
2084	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2085
2086	task_lock(current);
2087	memcpy(mm->saved_auxv, user_auxv, len);
2088	task_unlock(current);
2089
2090	return 0;
2091}
2092
2093static int prctl_set_mm(int opt, unsigned long addr,
2094			unsigned long arg4, unsigned long arg5)
2095{
 
2096	struct mm_struct *mm = current->mm;
2097	struct prctl_mm_map prctl_map;
2098	struct vm_area_struct *vma;
2099	int error;
2100
2101	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2102			      opt != PR_SET_MM_MAP &&
2103			      opt != PR_SET_MM_MAP_SIZE)))
2104		return -EINVAL;
2105
2106#ifdef CONFIG_CHECKPOINT_RESTORE
2107	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2108		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2109#endif
2110
2111	if (!capable(CAP_SYS_RESOURCE))
2112		return -EPERM;
2113
2114	if (opt == PR_SET_MM_EXE_FILE)
2115		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2116
2117	if (opt == PR_SET_MM_AUXV)
2118		return prctl_set_auxv(mm, addr, arg4);
2119
2120	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2121		return -EINVAL;
2122
2123	error = -EINVAL;
2124
2125	down_write(&mm->mmap_sem);
2126	vma = find_vma(mm, addr);
2127
2128	prctl_map.start_code	= mm->start_code;
2129	prctl_map.end_code	= mm->end_code;
2130	prctl_map.start_data	= mm->start_data;
2131	prctl_map.end_data	= mm->end_data;
2132	prctl_map.start_brk	= mm->start_brk;
2133	prctl_map.brk		= mm->brk;
2134	prctl_map.start_stack	= mm->start_stack;
2135	prctl_map.arg_start	= mm->arg_start;
2136	prctl_map.arg_end	= mm->arg_end;
2137	prctl_map.env_start	= mm->env_start;
2138	prctl_map.env_end	= mm->env_end;
2139	prctl_map.auxv		= NULL;
2140	prctl_map.auxv_size	= 0;
2141	prctl_map.exe_fd	= -1;
2142
2143	switch (opt) {
2144	case PR_SET_MM_START_CODE:
2145		prctl_map.start_code = addr;
2146		break;
2147	case PR_SET_MM_END_CODE:
2148		prctl_map.end_code = addr;
2149		break;
2150	case PR_SET_MM_START_DATA:
2151		prctl_map.start_data = addr;
2152		break;
2153	case PR_SET_MM_END_DATA:
2154		prctl_map.end_data = addr;
2155		break;
2156	case PR_SET_MM_START_STACK:
2157		prctl_map.start_stack = addr;
2158		break;
 
2159	case PR_SET_MM_START_BRK:
2160		prctl_map.start_brk = addr;
 
 
 
 
 
 
 
 
2161		break;
 
2162	case PR_SET_MM_BRK:
2163		prctl_map.brk = addr;
2164		break;
2165	case PR_SET_MM_ARG_START:
2166		prctl_map.arg_start = addr;
2167		break;
2168	case PR_SET_MM_ARG_END:
2169		prctl_map.arg_end = addr;
2170		break;
2171	case PR_SET_MM_ENV_START:
2172		prctl_map.env_start = addr;
2173		break;
2174	case PR_SET_MM_ENV_END:
2175		prctl_map.env_end = addr;
2176		break;
2177	default:
2178		goto out;
2179	}
2180
2181	error = validate_prctl_map(&prctl_map);
2182	if (error)
2183		goto out;
 
 
 
 
2184
2185	switch (opt) {
2186	/*
2187	 * If command line arguments and environment
2188	 * are placed somewhere else on stack, we can
2189	 * set them up here, ARG_START/END to setup
2190	 * command line argumets and ENV_START/END
2191	 * for environment.
2192	 */
2193	case PR_SET_MM_START_STACK:
2194	case PR_SET_MM_ARG_START:
2195	case PR_SET_MM_ARG_END:
2196	case PR_SET_MM_ENV_START:
2197	case PR_SET_MM_ENV_END:
2198		if (!vma) {
2199			error = -EFAULT;
2200			goto out;
2201		}
2202	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2203
2204	mm->start_code	= prctl_map.start_code;
2205	mm->end_code	= prctl_map.end_code;
2206	mm->start_data	= prctl_map.start_data;
2207	mm->end_data	= prctl_map.end_data;
2208	mm->start_brk	= prctl_map.start_brk;
2209	mm->brk		= prctl_map.brk;
2210	mm->start_stack	= prctl_map.start_stack;
2211	mm->arg_start	= prctl_map.arg_start;
2212	mm->arg_end	= prctl_map.arg_end;
2213	mm->env_start	= prctl_map.env_start;
2214	mm->env_end	= prctl_map.env_end;
 
 
 
 
2215
2216	error = 0;
2217out:
2218	up_write(&mm->mmap_sem);
2219	return error;
2220}
2221
2222#ifdef CONFIG_CHECKPOINT_RESTORE
2223static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2224{
2225	return put_user(me->clear_child_tid, tid_addr);
2226}
2227#else
2228static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2229{
2230	return -EINVAL;
2231}
2232#endif
2233
2234static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2235{
2236	/*
2237	 * If task has has_child_subreaper - all its decendants
2238	 * already have these flag too and new decendants will
2239	 * inherit it on fork, skip them.
2240	 *
2241	 * If we've found child_reaper - skip descendants in
2242	 * it's subtree as they will never get out pidns.
2243	 */
2244	if (p->signal->has_child_subreaper ||
2245	    is_child_reaper(task_pid(p)))
2246		return 0;
2247
2248	p->signal->has_child_subreaper = 1;
2249	return 1;
2250}
2251
2252int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2253{
2254	return -EINVAL;
2255}
2256
2257int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2258				    unsigned long ctrl)
2259{
2260	return -EINVAL;
2261}
 
2262
2263SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2264		unsigned long, arg4, unsigned long, arg5)
2265{
2266	struct task_struct *me = current;
2267	unsigned char comm[sizeof(me->comm)];
2268	long error;
2269
2270	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2271	if (error != -ENOSYS)
2272		return error;
2273
2274	error = 0;
2275	switch (option) {
2276	case PR_SET_PDEATHSIG:
2277		if (!valid_signal(arg2)) {
2278			error = -EINVAL;
 
 
 
 
2279			break;
2280		}
2281		me->pdeath_signal = arg2;
2282		break;
2283	case PR_GET_PDEATHSIG:
2284		error = put_user(me->pdeath_signal, (int __user *)arg2);
2285		break;
2286	case PR_GET_DUMPABLE:
2287		error = get_dumpable(me->mm);
2288		break;
2289	case PR_SET_DUMPABLE:
2290		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2291			error = -EINVAL;
 
2292			break;
2293		}
2294		set_dumpable(me->mm, arg2);
2295		break;
2296
2297	case PR_SET_UNALIGN:
2298		error = SET_UNALIGN_CTL(me, arg2);
2299		break;
2300	case PR_GET_UNALIGN:
2301		error = GET_UNALIGN_CTL(me, arg2);
2302		break;
2303	case PR_SET_FPEMU:
2304		error = SET_FPEMU_CTL(me, arg2);
2305		break;
2306	case PR_GET_FPEMU:
2307		error = GET_FPEMU_CTL(me, arg2);
2308		break;
2309	case PR_SET_FPEXC:
2310		error = SET_FPEXC_CTL(me, arg2);
2311		break;
2312	case PR_GET_FPEXC:
2313		error = GET_FPEXC_CTL(me, arg2);
2314		break;
2315	case PR_GET_TIMING:
2316		error = PR_TIMING_STATISTICAL;
2317		break;
2318	case PR_SET_TIMING:
2319		if (arg2 != PR_TIMING_STATISTICAL)
2320			error = -EINVAL;
2321		break;
2322	case PR_SET_NAME:
2323		comm[sizeof(me->comm) - 1] = 0;
2324		if (strncpy_from_user(comm, (char __user *)arg2,
2325				      sizeof(me->comm) - 1) < 0)
2326			return -EFAULT;
2327		set_task_comm(me, comm);
2328		proc_comm_connector(me);
2329		break;
2330	case PR_GET_NAME:
2331		get_task_comm(comm, me);
2332		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2333			return -EFAULT;
2334		break;
2335	case PR_GET_ENDIAN:
2336		error = GET_ENDIAN(me, arg2);
2337		break;
2338	case PR_SET_ENDIAN:
2339		error = SET_ENDIAN(me, arg2);
2340		break;
2341	case PR_GET_SECCOMP:
2342		error = prctl_get_seccomp();
2343		break;
2344	case PR_SET_SECCOMP:
2345		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2346		break;
2347	case PR_GET_TSC:
2348		error = GET_TSC_CTL(arg2);
2349		break;
2350	case PR_SET_TSC:
2351		error = SET_TSC_CTL(arg2);
2352		break;
2353	case PR_TASK_PERF_EVENTS_DISABLE:
2354		error = perf_event_task_disable();
2355		break;
2356	case PR_TASK_PERF_EVENTS_ENABLE:
2357		error = perf_event_task_enable();
2358		break;
2359	case PR_GET_TIMERSLACK:
2360		if (current->timer_slack_ns > ULONG_MAX)
2361			error = ULONG_MAX;
2362		else
 
 
2363			error = current->timer_slack_ns;
2364		break;
2365	case PR_SET_TIMERSLACK:
2366		if (arg2 <= 0)
2367			current->timer_slack_ns =
2368					current->default_timer_slack_ns;
2369		else
2370			current->timer_slack_ns = arg2;
2371		break;
2372	case PR_MCE_KILL:
2373		if (arg4 | arg5)
2374			return -EINVAL;
2375		switch (arg2) {
2376		case PR_MCE_KILL_CLEAR:
2377			if (arg3 != 0)
2378				return -EINVAL;
2379			current->flags &= ~PF_MCE_PROCESS;
2380			break;
2381		case PR_MCE_KILL_SET:
2382			current->flags |= PF_MCE_PROCESS;
2383			if (arg3 == PR_MCE_KILL_EARLY)
2384				current->flags |= PF_MCE_EARLY;
2385			else if (arg3 == PR_MCE_KILL_LATE)
2386				current->flags &= ~PF_MCE_EARLY;
2387			else if (arg3 == PR_MCE_KILL_DEFAULT)
2388				current->flags &=
 
 
 
 
 
 
 
 
 
2389						~(PF_MCE_EARLY|PF_MCE_PROCESS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2390			else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2391				return -EINVAL;
 
 
2392			break;
 
 
 
 
2393		default:
2394			return -EINVAL;
2395		}
2396		break;
2397	case PR_MCE_KILL_GET:
2398		if (arg2 | arg3 | arg4 | arg5)
2399			return -EINVAL;
2400		if (current->flags & PF_MCE_PROCESS)
2401			error = (current->flags & PF_MCE_EARLY) ?
2402				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2403		else
2404			error = PR_MCE_KILL_DEFAULT;
2405		break;
2406	case PR_SET_MM:
2407		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2408		break;
2409	case PR_GET_TID_ADDRESS:
2410		error = prctl_get_tid_address(me, (int __user **)arg2);
2411		break;
2412	case PR_SET_CHILD_SUBREAPER:
2413		me->signal->is_child_subreaper = !!arg2;
2414		if (!arg2)
2415			break;
2416
2417		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2418		break;
2419	case PR_GET_CHILD_SUBREAPER:
2420		error = put_user(me->signal->is_child_subreaper,
2421				 (int __user *)arg2);
2422		break;
2423	case PR_SET_NO_NEW_PRIVS:
2424		if (arg2 != 1 || arg3 || arg4 || arg5)
2425			return -EINVAL;
2426
2427		task_set_no_new_privs(current);
2428		break;
2429	case PR_GET_NO_NEW_PRIVS:
2430		if (arg2 || arg3 || arg4 || arg5)
2431			return -EINVAL;
2432		return task_no_new_privs(current) ? 1 : 0;
2433	case PR_GET_THP_DISABLE:
2434		if (arg2 || arg3 || arg4 || arg5)
2435			return -EINVAL;
2436		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2437		break;
2438	case PR_SET_THP_DISABLE:
2439		if (arg3 || arg4 || arg5)
2440			return -EINVAL;
2441		if (down_write_killable(&me->mm->mmap_sem))
2442			return -EINTR;
2443		if (arg2)
2444			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2445		else
2446			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2447		up_write(&me->mm->mmap_sem);
2448		break;
2449	case PR_MPX_ENABLE_MANAGEMENT:
2450		if (arg2 || arg3 || arg4 || arg5)
2451			return -EINVAL;
2452		error = MPX_ENABLE_MANAGEMENT();
2453		break;
2454	case PR_MPX_DISABLE_MANAGEMENT:
2455		if (arg2 || arg3 || arg4 || arg5)
2456			return -EINVAL;
2457		error = MPX_DISABLE_MANAGEMENT();
2458		break;
2459	case PR_SET_FP_MODE:
2460		error = SET_FP_MODE(me, arg2);
2461		break;
2462	case PR_GET_FP_MODE:
2463		error = GET_FP_MODE(me);
2464		break;
2465	case PR_SVE_SET_VL:
2466		error = SVE_SET_VL(arg2);
2467		break;
2468	case PR_SVE_GET_VL:
2469		error = SVE_GET_VL();
2470		break;
2471	case PR_GET_SPECULATION_CTRL:
2472		if (arg3 || arg4 || arg5)
2473			return -EINVAL;
2474		error = arch_prctl_spec_ctrl_get(me, arg2);
2475		break;
2476	case PR_SET_SPECULATION_CTRL:
2477		if (arg4 || arg5)
2478			return -EINVAL;
2479		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2480		break;
2481	default:
2482		error = -EINVAL;
2483		break;
2484	}
2485	return error;
2486}
2487
2488SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2489		struct getcpu_cache __user *, unused)
2490{
2491	int err = 0;
2492	int cpu = raw_smp_processor_id();
2493
2494	if (cpup)
2495		err |= put_user(cpu, cpup);
2496	if (nodep)
2497		err |= put_user(cpu_to_node(cpu), nodep);
2498	return err ? -EFAULT : 0;
2499}
2500
 
 
 
 
 
 
 
2501/**
2502 * do_sysinfo - fill in sysinfo struct
2503 * @info: pointer to buffer to fill
 
 
 
2504 */
2505static int do_sysinfo(struct sysinfo *info)
2506{
2507	unsigned long mem_total, sav_total;
2508	unsigned int mem_unit, bitcount;
2509	struct timespec tp;
2510
2511	memset(info, 0, sizeof(struct sysinfo));
2512
2513	get_monotonic_boottime(&tp);
2514	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2515
2516	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2517
2518	info->procs = nr_threads;
2519
2520	si_meminfo(info);
2521	si_swapinfo(info);
2522
2523	/*
2524	 * If the sum of all the available memory (i.e. ram + swap)
2525	 * is less than can be stored in a 32 bit unsigned long then
2526	 * we can be binary compatible with 2.2.x kernels.  If not,
2527	 * well, in that case 2.2.x was broken anyways...
2528	 *
2529	 *  -Erik Andersen <andersee@debian.org>
2530	 */
2531
2532	mem_total = info->totalram + info->totalswap;
2533	if (mem_total < info->totalram || mem_total < info->totalswap)
 
2534		goto out;
2535	bitcount = 0;
2536	mem_unit = info->mem_unit;
2537	while (mem_unit > 1) {
2538		bitcount++;
2539		mem_unit >>= 1;
2540		sav_total = mem_total;
2541		mem_total <<= 1;
2542		if (mem_total < sav_total)
2543			goto out;
2544	}
2545
2546	/*
2547	 * If mem_total did not overflow, multiply all memory values by
2548	 * info->mem_unit and set it to 1.  This leaves things compatible
2549	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2550	 * kernels...
2551	 */
2552
2553	info->mem_unit = 1;
2554	info->totalram <<= bitcount;
2555	info->freeram <<= bitcount;
2556	info->sharedram <<= bitcount;
2557	info->bufferram <<= bitcount;
2558	info->totalswap <<= bitcount;
2559	info->freeswap <<= bitcount;
2560	info->totalhigh <<= bitcount;
2561	info->freehigh <<= bitcount;
2562
2563out:
2564	return 0;
2565}
2566
2567SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2568{
2569	struct sysinfo val;
2570
2571	do_sysinfo(&val);
2572
2573	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2574		return -EFAULT;
2575
2576	return 0;
2577}
2578
2579#ifdef CONFIG_COMPAT
2580struct compat_sysinfo {
2581	s32 uptime;
2582	u32 loads[3];
2583	u32 totalram;
2584	u32 freeram;
2585	u32 sharedram;
2586	u32 bufferram;
2587	u32 totalswap;
2588	u32 freeswap;
2589	u16 procs;
2590	u16 pad;
2591	u32 totalhigh;
2592	u32 freehigh;
2593	u32 mem_unit;
2594	char _f[20-2*sizeof(u32)-sizeof(int)];
2595};
2596
2597COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2598{
2599	struct sysinfo s;
2600
2601	do_sysinfo(&s);
2602
2603	/* Check to see if any memory value is too large for 32-bit and scale
2604	 *  down if needed
2605	 */
2606	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2607		int bitcount = 0;
2608
2609		while (s.mem_unit < PAGE_SIZE) {
2610			s.mem_unit <<= 1;
2611			bitcount++;
2612		}
2613
2614		s.totalram >>= bitcount;
2615		s.freeram >>= bitcount;
2616		s.sharedram >>= bitcount;
2617		s.bufferram >>= bitcount;
2618		s.totalswap >>= bitcount;
2619		s.freeswap >>= bitcount;
2620		s.totalhigh >>= bitcount;
2621		s.freehigh >>= bitcount;
2622	}
2623
2624	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2625	    __put_user(s.uptime, &info->uptime) ||
2626	    __put_user(s.loads[0], &info->loads[0]) ||
2627	    __put_user(s.loads[1], &info->loads[1]) ||
2628	    __put_user(s.loads[2], &info->loads[2]) ||
2629	    __put_user(s.totalram, &info->totalram) ||
2630	    __put_user(s.freeram, &info->freeram) ||
2631	    __put_user(s.sharedram, &info->sharedram) ||
2632	    __put_user(s.bufferram, &info->bufferram) ||
2633	    __put_user(s.totalswap, &info->totalswap) ||
2634	    __put_user(s.freeswap, &info->freeswap) ||
2635	    __put_user(s.procs, &info->procs) ||
2636	    __put_user(s.totalhigh, &info->totalhigh) ||
2637	    __put_user(s.freehigh, &info->freehigh) ||
2638	    __put_user(s.mem_unit, &info->mem_unit))
2639		return -EFAULT;
2640
2641	return 0;
2642}
2643#endif /* CONFIG_COMPAT */