Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/open.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8#include <linux/string.h>
   9#include <linux/mm.h>
  10#include <linux/file.h>
  11#include <linux/fdtable.h>
  12#include <linux/fsnotify.h>
  13#include <linux/module.h>
  14#include <linux/tty.h>
  15#include <linux/namei.h>
  16#include <linux/backing-dev.h>
  17#include <linux/capability.h>
  18#include <linux/securebits.h>
  19#include <linux/security.h>
  20#include <linux/mount.h>
  21#include <linux/fcntl.h>
  22#include <linux/slab.h>
  23#include <linux/uaccess.h>
  24#include <linux/fs.h>
  25#include <linux/personality.h>
  26#include <linux/pagemap.h>
  27#include <linux/syscalls.h>
  28#include <linux/rcupdate.h>
  29#include <linux/audit.h>
  30#include <linux/falloc.h>
  31#include <linux/fs_struct.h>
  32#include <linux/ima.h>
  33#include <linux/dnotify.h>
  34#include <linux/compat.h>
  35#include <linux/mnt_idmapping.h>
 
  36
  37#include "internal.h"
  38
  39int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
  40		loff_t length, unsigned int time_attrs, struct file *filp)
  41{
  42	int ret;
  43	struct iattr newattrs;
  44
  45	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  46	if (length < 0)
  47		return -EINVAL;
  48
  49	newattrs.ia_size = length;
  50	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  51	if (filp) {
  52		newattrs.ia_file = filp;
  53		newattrs.ia_valid |= ATTR_FILE;
  54	}
  55
  56	/* Remove suid, sgid, and file capabilities on truncate too */
  57	ret = dentry_needs_remove_privs(mnt_userns, dentry);
  58	if (ret < 0)
  59		return ret;
  60	if (ret)
  61		newattrs.ia_valid |= ret | ATTR_FORCE;
  62
  63	inode_lock(dentry->d_inode);
  64	/* Note any delegations or leases have already been broken: */
  65	ret = notify_change(mnt_userns, dentry, &newattrs, NULL);
  66	inode_unlock(dentry->d_inode);
  67	return ret;
  68}
  69
  70long vfs_truncate(const struct path *path, loff_t length)
  71{
  72	struct user_namespace *mnt_userns;
  73	struct inode *inode;
  74	long error;
  75
  76	inode = path->dentry->d_inode;
  77
  78	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  79	if (S_ISDIR(inode->i_mode))
  80		return -EISDIR;
  81	if (!S_ISREG(inode->i_mode))
  82		return -EINVAL;
  83
  84	error = mnt_want_write(path->mnt);
  85	if (error)
  86		goto out;
  87
  88	mnt_userns = mnt_user_ns(path->mnt);
  89	error = inode_permission(mnt_userns, inode, MAY_WRITE);
  90	if (error)
  91		goto mnt_drop_write_and_out;
  92
  93	error = -EPERM;
  94	if (IS_APPEND(inode))
  95		goto mnt_drop_write_and_out;
  96
  97	error = get_write_access(inode);
  98	if (error)
  99		goto mnt_drop_write_and_out;
 100
 101	/*
 102	 * Make sure that there are no leases.  get_write_access() protects
 103	 * against the truncate racing with a lease-granting setlease().
 104	 */
 105	error = break_lease(inode, O_WRONLY);
 106	if (error)
 107		goto put_write_and_out;
 108
 109	error = security_path_truncate(path);
 110	if (!error)
 111		error = do_truncate(mnt_userns, path->dentry, length, 0, NULL);
 112
 113put_write_and_out:
 114	put_write_access(inode);
 115mnt_drop_write_and_out:
 116	mnt_drop_write(path->mnt);
 117out:
 118	return error;
 119}
 120EXPORT_SYMBOL_GPL(vfs_truncate);
 121
 122long do_sys_truncate(const char __user *pathname, loff_t length)
 123{
 124	unsigned int lookup_flags = LOOKUP_FOLLOW;
 125	struct path path;
 126	int error;
 127
 128	if (length < 0)	/* sorry, but loff_t says... */
 129		return -EINVAL;
 130
 131retry:
 132	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 133	if (!error) {
 134		error = vfs_truncate(&path, length);
 135		path_put(&path);
 136	}
 137	if (retry_estale(error, lookup_flags)) {
 138		lookup_flags |= LOOKUP_REVAL;
 139		goto retry;
 140	}
 141	return error;
 142}
 143
 144SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 145{
 146	return do_sys_truncate(path, length);
 147}
 148
 149#ifdef CONFIG_COMPAT
 150COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 151{
 152	return do_sys_truncate(path, length);
 153}
 154#endif
 155
 156long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 157{
 158	struct inode *inode;
 159	struct dentry *dentry;
 160	struct fd f;
 161	int error;
 162
 163	error = -EINVAL;
 164	if (length < 0)
 165		goto out;
 166	error = -EBADF;
 167	f = fdget(fd);
 168	if (!f.file)
 169		goto out;
 170
 171	/* explicitly opened as large or we are on 64-bit box */
 172	if (f.file->f_flags & O_LARGEFILE)
 173		small = 0;
 174
 175	dentry = f.file->f_path.dentry;
 176	inode = dentry->d_inode;
 177	error = -EINVAL;
 178	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 179		goto out_putf;
 180
 181	error = -EINVAL;
 182	/* Cannot ftruncate over 2^31 bytes without large file support */
 183	if (small && length > MAX_NON_LFS)
 184		goto out_putf;
 185
 186	error = -EPERM;
 187	/* Check IS_APPEND on real upper inode */
 188	if (IS_APPEND(file_inode(f.file)))
 189		goto out_putf;
 190	sb_start_write(inode->i_sb);
 191	error = security_file_truncate(f.file);
 192	if (!error)
 193		error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
 194				    ATTR_MTIME | ATTR_CTIME, f.file);
 195	sb_end_write(inode->i_sb);
 196out_putf:
 197	fdput(f);
 198out:
 199	return error;
 200}
 201
 202SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 203{
 204	return do_sys_ftruncate(fd, length, 1);
 205}
 206
 207#ifdef CONFIG_COMPAT
 208COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 209{
 210	return do_sys_ftruncate(fd, length, 1);
 211}
 212#endif
 213
 214/* LFS versions of truncate are only needed on 32 bit machines */
 215#if BITS_PER_LONG == 32
 216SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 217{
 218	return do_sys_truncate(path, length);
 219}
 220
 221SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 222{
 223	return do_sys_ftruncate(fd, length, 0);
 224}
 225#endif /* BITS_PER_LONG == 32 */
 226
 227#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
 228COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
 229		       compat_arg_u64_dual(length))
 230{
 231	return ksys_truncate(pathname, compat_arg_u64_glue(length));
 232}
 233#endif
 234
 235#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
 236COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
 237		       compat_arg_u64_dual(length))
 238{
 239	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
 240}
 241#endif
 242
 243int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 244{
 245	struct inode *inode = file_inode(file);
 246	long ret;
 247
 248	if (offset < 0 || len <= 0)
 249		return -EINVAL;
 250
 251	/* Return error if mode is not supported */
 252	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 253		return -EOPNOTSUPP;
 254
 255	/* Punch hole and zero range are mutually exclusive */
 256	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 257	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 258		return -EOPNOTSUPP;
 259
 260	/* Punch hole must have keep size set */
 261	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 262	    !(mode & FALLOC_FL_KEEP_SIZE))
 263		return -EOPNOTSUPP;
 264
 265	/* Collapse range should only be used exclusively. */
 266	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 267	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 268		return -EINVAL;
 269
 270	/* Insert range should only be used exclusively. */
 271	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 272	    (mode & ~FALLOC_FL_INSERT_RANGE))
 273		return -EINVAL;
 274
 275	/* Unshare range should only be used with allocate mode. */
 276	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 277	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 278		return -EINVAL;
 279
 280	if (!(file->f_mode & FMODE_WRITE))
 281		return -EBADF;
 282
 283	/*
 284	 * We can only allow pure fallocate on append only files
 285	 */
 286	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 287		return -EPERM;
 288
 289	if (IS_IMMUTABLE(inode))
 290		return -EPERM;
 291
 292	/*
 293	 * We cannot allow any fallocate operation on an active swapfile
 294	 */
 295	if (IS_SWAPFILE(inode))
 296		return -ETXTBSY;
 297
 298	/*
 299	 * Revalidate the write permissions, in case security policy has
 300	 * changed since the files were opened.
 301	 */
 302	ret = security_file_permission(file, MAY_WRITE);
 303	if (ret)
 304		return ret;
 305
 
 
 
 
 306	if (S_ISFIFO(inode->i_mode))
 307		return -ESPIPE;
 308
 309	if (S_ISDIR(inode->i_mode))
 310		return -EISDIR;
 311
 312	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 313		return -ENODEV;
 314
 315	/* Check for wrap through zero too */
 316	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 317		return -EFBIG;
 318
 319	if (!file->f_op->fallocate)
 320		return -EOPNOTSUPP;
 321
 322	file_start_write(file);
 323	ret = file->f_op->fallocate(file, mode, offset, len);
 324
 325	/*
 326	 * Create inotify and fanotify events.
 327	 *
 328	 * To keep the logic simple always create events if fallocate succeeds.
 329	 * This implies that events are even created if the file size remains
 330	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 331	 */
 332	if (ret == 0)
 333		fsnotify_modify(file);
 334
 335	file_end_write(file);
 336	return ret;
 337}
 338EXPORT_SYMBOL_GPL(vfs_fallocate);
 339
 340int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 341{
 342	struct fd f = fdget(fd);
 343	int error = -EBADF;
 344
 345	if (f.file) {
 346		error = vfs_fallocate(f.file, mode, offset, len);
 347		fdput(f);
 348	}
 349	return error;
 350}
 351
 352SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 353{
 354	return ksys_fallocate(fd, mode, offset, len);
 355}
 356
 357#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
 358COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
 359		       compat_arg_u64_dual(len))
 360{
 361	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
 362			      compat_arg_u64_glue(len));
 363}
 364#endif
 365
 366/*
 367 * access() needs to use the real uid/gid, not the effective uid/gid.
 368 * We do this by temporarily clearing all FS-related capabilities and
 369 * switching the fsuid/fsgid around to the real ones.
 
 
 
 370 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 371static const struct cred *access_override_creds(void)
 372{
 373	const struct cred *old_cred;
 374	struct cred *override_cred;
 375
 376	override_cred = prepare_creds();
 377	if (!override_cred)
 378		return NULL;
 379
 
 
 
 
 
 
 380	override_cred->fsuid = override_cred->uid;
 381	override_cred->fsgid = override_cred->gid;
 382
 383	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 384		/* Clear the capabilities if we switch to a non-root user */
 385		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 386		if (!uid_eq(override_cred->uid, root_uid))
 387			cap_clear(override_cred->cap_effective);
 388		else
 389			override_cred->cap_effective =
 390				override_cred->cap_permitted;
 391	}
 392
 393	/*
 394	 * The new set of credentials can *only* be used in
 395	 * task-synchronous circumstances, and does not need
 396	 * RCU freeing, unless somebody then takes a separate
 397	 * reference to it.
 398	 *
 399	 * NOTE! This is _only_ true because this credential
 400	 * is used purely for override_creds() that installs
 401	 * it as the subjective cred. Other threads will be
 402	 * accessing ->real_cred, not the subjective cred.
 403	 *
 404	 * If somebody _does_ make a copy of this (using the
 405	 * 'get_current_cred()' function), that will clear the
 406	 * non_rcu field, because now that other user may be
 407	 * expecting RCU freeing. But normal thread-synchronous
 408	 * cred accesses will keep things non-RCY.
 
 409	 */
 410	override_cred->non_rcu = 1;
 411
 412	old_cred = override_creds(override_cred);
 413
 414	/* override_cred() gets its own ref */
 415	put_cred(override_cred);
 416
 417	return old_cred;
 418}
 419
 420static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 421{
 422	struct path path;
 423	struct inode *inode;
 424	int res;
 425	unsigned int lookup_flags = LOOKUP_FOLLOW;
 426	const struct cred *old_cred = NULL;
 427
 428	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 429		return -EINVAL;
 430
 431	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 432		return -EINVAL;
 433
 434	if (flags & AT_SYMLINK_NOFOLLOW)
 435		lookup_flags &= ~LOOKUP_FOLLOW;
 436	if (flags & AT_EMPTY_PATH)
 437		lookup_flags |= LOOKUP_EMPTY;
 438
 439	if (!(flags & AT_EACCESS)) {
 440		old_cred = access_override_creds();
 441		if (!old_cred)
 442			return -ENOMEM;
 443	}
 444
 445retry:
 446	res = user_path_at(dfd, filename, lookup_flags, &path);
 447	if (res)
 448		goto out;
 449
 450	inode = d_backing_inode(path.dentry);
 451
 452	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 453		/*
 454		 * MAY_EXEC on regular files is denied if the fs is mounted
 455		 * with the "noexec" flag.
 456		 */
 457		res = -EACCES;
 458		if (path_noexec(&path))
 459			goto out_path_release;
 460	}
 461
 462	res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
 463	/* SuS v2 requires we report a read only fs too */
 464	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 465		goto out_path_release;
 466	/*
 467	 * This is a rare case where using __mnt_is_readonly()
 468	 * is OK without a mnt_want/drop_write() pair.  Since
 469	 * no actual write to the fs is performed here, we do
 470	 * not need to telegraph to that to anyone.
 471	 *
 472	 * By doing this, we accept that this access is
 473	 * inherently racy and know that the fs may change
 474	 * state before we even see this result.
 475	 */
 476	if (__mnt_is_readonly(path.mnt))
 477		res = -EROFS;
 478
 479out_path_release:
 480	path_put(&path);
 481	if (retry_estale(res, lookup_flags)) {
 482		lookup_flags |= LOOKUP_REVAL;
 483		goto retry;
 484	}
 485out:
 486	if (old_cred)
 487		revert_creds(old_cred);
 488
 489	return res;
 490}
 491
 492SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 493{
 494	return do_faccessat(dfd, filename, mode, 0);
 495}
 496
 497SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 498		int, flags)
 499{
 500	return do_faccessat(dfd, filename, mode, flags);
 501}
 502
 503SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 504{
 505	return do_faccessat(AT_FDCWD, filename, mode, 0);
 506}
 507
 508SYSCALL_DEFINE1(chdir, const char __user *, filename)
 509{
 510	struct path path;
 511	int error;
 512	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 513retry:
 514	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 515	if (error)
 516		goto out;
 517
 518	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 519	if (error)
 520		goto dput_and_out;
 521
 522	set_fs_pwd(current->fs, &path);
 523
 524dput_and_out:
 525	path_put(&path);
 526	if (retry_estale(error, lookup_flags)) {
 527		lookup_flags |= LOOKUP_REVAL;
 528		goto retry;
 529	}
 530out:
 531	return error;
 532}
 533
 534SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 535{
 536	struct fd f = fdget_raw(fd);
 537	int error;
 538
 539	error = -EBADF;
 540	if (!f.file)
 541		goto out;
 542
 543	error = -ENOTDIR;
 544	if (!d_can_lookup(f.file->f_path.dentry))
 545		goto out_putf;
 546
 547	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
 548	if (!error)
 549		set_fs_pwd(current->fs, &f.file->f_path);
 550out_putf:
 551	fdput(f);
 552out:
 553	return error;
 554}
 555
 556SYSCALL_DEFINE1(chroot, const char __user *, filename)
 557{
 558	struct path path;
 559	int error;
 560	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 561retry:
 562	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 563	if (error)
 564		goto out;
 565
 566	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 567	if (error)
 568		goto dput_and_out;
 569
 570	error = -EPERM;
 571	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 572		goto dput_and_out;
 573	error = security_path_chroot(&path);
 574	if (error)
 575		goto dput_and_out;
 576
 577	set_fs_root(current->fs, &path);
 578	error = 0;
 579dput_and_out:
 580	path_put(&path);
 581	if (retry_estale(error, lookup_flags)) {
 582		lookup_flags |= LOOKUP_REVAL;
 583		goto retry;
 584	}
 585out:
 586	return error;
 587}
 588
 589int chmod_common(const struct path *path, umode_t mode)
 590{
 591	struct inode *inode = path->dentry->d_inode;
 592	struct inode *delegated_inode = NULL;
 593	struct iattr newattrs;
 594	int error;
 595
 596	error = mnt_want_write(path->mnt);
 597	if (error)
 598		return error;
 599retry_deleg:
 600	inode_lock(inode);
 601	error = security_path_chmod(path, mode);
 602	if (error)
 603		goto out_unlock;
 604	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 605	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 606	error = notify_change(mnt_user_ns(path->mnt), path->dentry,
 607			      &newattrs, &delegated_inode);
 608out_unlock:
 609	inode_unlock(inode);
 610	if (delegated_inode) {
 611		error = break_deleg_wait(&delegated_inode);
 612		if (!error)
 613			goto retry_deleg;
 614	}
 615	mnt_drop_write(path->mnt);
 616	return error;
 617}
 618
 619int vfs_fchmod(struct file *file, umode_t mode)
 620{
 621	audit_file(file);
 622	return chmod_common(&file->f_path, mode);
 623}
 624
 625SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 626{
 627	struct fd f = fdget(fd);
 628	int err = -EBADF;
 629
 630	if (f.file) {
 631		err = vfs_fchmod(f.file, mode);
 632		fdput(f);
 633	}
 634	return err;
 635}
 636
 637static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 
 638{
 639	struct path path;
 640	int error;
 641	unsigned int lookup_flags = LOOKUP_FOLLOW;
 
 
 
 
 
 
 
 
 642retry:
 643	error = user_path_at(dfd, filename, lookup_flags, &path);
 644	if (!error) {
 645		error = chmod_common(&path, mode);
 646		path_put(&path);
 647		if (retry_estale(error, lookup_flags)) {
 648			lookup_flags |= LOOKUP_REVAL;
 649			goto retry;
 650		}
 651	}
 652	return error;
 653}
 654
 
 
 
 
 
 
 655SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 656		umode_t, mode)
 657{
 658	return do_fchmodat(dfd, filename, mode);
 659}
 660
 661SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 662{
 663	return do_fchmodat(AT_FDCWD, filename, mode);
 664}
 665
 666/**
 667 * setattr_vfsuid - check and set ia_fsuid attribute
 668 * @kuid: new inode owner
 669 *
 670 * Check whether @kuid is valid and if so generate and set vfsuid_t in
 671 * ia_vfsuid.
 672 *
 673 * Return: true if @kuid is valid, false if not.
 674 */
 675static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
 676{
 677	if (!uid_valid(kuid))
 678		return false;
 679	attr->ia_valid |= ATTR_UID;
 680	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
 681	return true;
 682}
 683
 684/**
 685 * setattr_vfsgid - check and set ia_fsgid attribute
 686 * @kgid: new inode owner
 687 *
 688 * Check whether @kgid is valid and if so generate and set vfsgid_t in
 689 * ia_vfsgid.
 690 *
 691 * Return: true if @kgid is valid, false if not.
 692 */
 693static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
 694{
 695	if (!gid_valid(kgid))
 696		return false;
 697	attr->ia_valid |= ATTR_GID;
 698	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
 699	return true;
 700}
 701
 702int chown_common(const struct path *path, uid_t user, gid_t group)
 703{
 704	struct user_namespace *mnt_userns, *fs_userns;
 
 705	struct inode *inode = path->dentry->d_inode;
 706	struct inode *delegated_inode = NULL;
 707	int error;
 708	struct iattr newattrs;
 709	kuid_t uid;
 710	kgid_t gid;
 711
 712	uid = make_kuid(current_user_ns(), user);
 713	gid = make_kgid(current_user_ns(), group);
 714
 715	mnt_userns = mnt_user_ns(path->mnt);
 716	fs_userns = i_user_ns(inode);
 717
 718retry_deleg:
 719	newattrs.ia_vfsuid = INVALID_VFSUID;
 720	newattrs.ia_vfsgid = INVALID_VFSGID;
 721	newattrs.ia_valid =  ATTR_CTIME;
 722	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 723		return -EINVAL;
 724	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 725		return -EINVAL;
 726	inode_lock(inode);
 727	if (!S_ISDIR(inode->i_mode))
 728		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
 729				     setattr_should_drop_sgid(mnt_userns, inode);
 730	/* Continue to send actual fs values, not the mount values. */
 731	error = security_path_chown(
 732		path,
 733		from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid),
 734		from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid));
 735	if (!error)
 736		error = notify_change(mnt_userns, path->dentry, &newattrs,
 737				      &delegated_inode);
 738	inode_unlock(inode);
 739	if (delegated_inode) {
 740		error = break_deleg_wait(&delegated_inode);
 741		if (!error)
 742			goto retry_deleg;
 743	}
 744	return error;
 745}
 746
 747int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 748		int flag)
 749{
 750	struct path path;
 751	int error = -EINVAL;
 752	int lookup_flags;
 753
 754	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 755		goto out;
 756
 757	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 758	if (flag & AT_EMPTY_PATH)
 759		lookup_flags |= LOOKUP_EMPTY;
 760retry:
 761	error = user_path_at(dfd, filename, lookup_flags, &path);
 762	if (error)
 763		goto out;
 764	error = mnt_want_write(path.mnt);
 765	if (error)
 766		goto out_release;
 767	error = chown_common(&path, user, group);
 768	mnt_drop_write(path.mnt);
 769out_release:
 770	path_put(&path);
 771	if (retry_estale(error, lookup_flags)) {
 772		lookup_flags |= LOOKUP_REVAL;
 773		goto retry;
 774	}
 775out:
 776	return error;
 777}
 778
 779SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 780		gid_t, group, int, flag)
 781{
 782	return do_fchownat(dfd, filename, user, group, flag);
 783}
 784
 785SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 786{
 787	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 788}
 789
 790SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 791{
 792	return do_fchownat(AT_FDCWD, filename, user, group,
 793			   AT_SYMLINK_NOFOLLOW);
 794}
 795
 796int vfs_fchown(struct file *file, uid_t user, gid_t group)
 797{
 798	int error;
 799
 800	error = mnt_want_write_file(file);
 801	if (error)
 802		return error;
 803	audit_file(file);
 804	error = chown_common(&file->f_path, user, group);
 805	mnt_drop_write_file(file);
 806	return error;
 807}
 808
 809int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 810{
 811	struct fd f = fdget(fd);
 812	int error = -EBADF;
 813
 814	if (f.file) {
 815		error = vfs_fchown(f.file, user, group);
 816		fdput(f);
 817	}
 818	return error;
 819}
 820
 821SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 822{
 823	return ksys_fchown(fd, user, group);
 824}
 825
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 826static int do_dentry_open(struct file *f,
 827			  struct inode *inode,
 828			  int (*open)(struct inode *, struct file *))
 829{
 830	static const struct file_operations empty_fops = {};
 831	int error;
 832
 833	path_get(&f->f_path);
 834	f->f_inode = inode;
 835	f->f_mapping = inode->i_mapping;
 836	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 837	f->f_sb_err = file_sample_sb_err(f);
 838
 839	if (unlikely(f->f_flags & O_PATH)) {
 840		f->f_mode = FMODE_PATH | FMODE_OPENED;
 841		f->f_op = &empty_fops;
 842		return 0;
 843	}
 844
 845	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
 846		i_readcount_inc(inode);
 847	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 848		error = get_write_access(inode);
 849		if (unlikely(error))
 850			goto cleanup_file;
 851		error = __mnt_want_write(f->f_path.mnt);
 852		if (unlikely(error)) {
 853			put_write_access(inode);
 854			goto cleanup_file;
 855		}
 856		f->f_mode |= FMODE_WRITER;
 857	}
 858
 859	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 860	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 861		f->f_mode |= FMODE_ATOMIC_POS;
 862
 863	f->f_op = fops_get(inode->i_fop);
 864	if (WARN_ON(!f->f_op)) {
 865		error = -ENODEV;
 866		goto cleanup_all;
 867	}
 868
 869	error = security_file_open(f);
 870	if (error)
 871		goto cleanup_all;
 872
 873	error = break_lease(locks_inode(f), f->f_flags);
 874	if (error)
 875		goto cleanup_all;
 876
 877	/* normally all 3 are set; ->open() can clear them if needed */
 878	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 879	if (!open)
 880		open = f->f_op->open;
 881	if (open) {
 882		error = open(inode, f);
 883		if (error)
 884			goto cleanup_all;
 885	}
 886	f->f_mode |= FMODE_OPENED;
 887	if ((f->f_mode & FMODE_READ) &&
 888	     likely(f->f_op->read || f->f_op->read_iter))
 889		f->f_mode |= FMODE_CAN_READ;
 890	if ((f->f_mode & FMODE_WRITE) &&
 891	     likely(f->f_op->write || f->f_op->write_iter))
 892		f->f_mode |= FMODE_CAN_WRITE;
 893	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 894		f->f_mode &= ~FMODE_LSEEK;
 895	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 896		f->f_mode |= FMODE_CAN_ODIRECT;
 897
 898	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 899	f->f_iocb_flags = iocb_flags(f);
 900
 901	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 902
 903	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 904		return -EINVAL;
 905
 906	/*
 907	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 908	 * cache for this file before processing writes.
 909	 */
 910	if (f->f_mode & FMODE_WRITE) {
 911		/*
 912		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
 913		 * is up to date and the update to i_writecount by
 914		 * get_write_access() is visible. Ensures subsequent insertion
 915		 * of THPs into the page cache will fail.
 916		 */
 917		smp_mb();
 918		if (filemap_nr_thps(inode->i_mapping)) {
 919			struct address_space *mapping = inode->i_mapping;
 920
 921			filemap_invalidate_lock(inode->i_mapping);
 922			/*
 923			 * unmap_mapping_range just need to be called once
 924			 * here, because the private pages is not need to be
 925			 * unmapped mapping (e.g. data segment of dynamic
 926			 * shared libraries here).
 927			 */
 928			unmap_mapping_range(mapping, 0, 0, 0);
 929			truncate_inode_pages(mapping, 0);
 930			filemap_invalidate_unlock(inode->i_mapping);
 931		}
 932	}
 933
 
 
 
 
 
 934	return 0;
 935
 936cleanup_all:
 937	if (WARN_ON_ONCE(error > 0))
 938		error = -EINVAL;
 939	fops_put(f->f_op);
 940	put_file_access(f);
 941cleanup_file:
 942	path_put(&f->f_path);
 943	f->f_path.mnt = NULL;
 944	f->f_path.dentry = NULL;
 945	f->f_inode = NULL;
 946	return error;
 947}
 948
 949/**
 950 * finish_open - finish opening a file
 951 * @file: file pointer
 952 * @dentry: pointer to dentry
 953 * @open: open callback
 954 * @opened: state of open
 955 *
 956 * This can be used to finish opening a file passed to i_op->atomic_open().
 957 *
 958 * If the open callback is set to NULL, then the standard f_op->open()
 959 * filesystem callback is substituted.
 960 *
 961 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 962 * the return value of d_splice_alias(), then the caller needs to perform dput()
 963 * on it after finish_open().
 964 *
 965 * Returns zero on success or -errno if the open failed.
 966 */
 967int finish_open(struct file *file, struct dentry *dentry,
 968		int (*open)(struct inode *, struct file *))
 969{
 970	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 971
 972	file->f_path.dentry = dentry;
 973	return do_dentry_open(file, d_backing_inode(dentry), open);
 974}
 975EXPORT_SYMBOL(finish_open);
 976
 977/**
 978 * finish_no_open - finish ->atomic_open() without opening the file
 979 *
 980 * @file: file pointer
 981 * @dentry: dentry or NULL (as returned from ->lookup())
 982 *
 983 * This can be used to set the result of a successful lookup in ->atomic_open().
 984 *
 985 * NB: unlike finish_open() this function does consume the dentry reference and
 986 * the caller need not dput() it.
 987 *
 988 * Returns "0" which must be the return value of ->atomic_open() after having
 989 * called this function.
 990 */
 991int finish_no_open(struct file *file, struct dentry *dentry)
 992{
 993	file->f_path.dentry = dentry;
 994	return 0;
 995}
 996EXPORT_SYMBOL(finish_no_open);
 997
 998char *file_path(struct file *filp, char *buf, int buflen)
 999{
1000	return d_path(&filp->f_path, buf, buflen);
1001}
1002EXPORT_SYMBOL(file_path);
1003
1004/**
1005 * vfs_open - open the file at the given path
1006 * @path: path to open
1007 * @file: newly allocated file with f_flag initialized
1008 * @cred: credentials to use
1009 */
1010int vfs_open(const struct path *path, struct file *file)
1011{
1012	file->f_path = *path;
1013	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1014}
1015
1016struct file *dentry_open(const struct path *path, int flags,
1017			 const struct cred *cred)
1018{
1019	int error;
1020	struct file *f;
1021
1022	validate_creds(cred);
1023
1024	/* We must always pass in a valid mount pointer. */
1025	BUG_ON(!path->mnt);
1026
1027	f = alloc_empty_file(flags, cred);
1028	if (!IS_ERR(f)) {
1029		error = vfs_open(path, f);
1030		if (error) {
1031			fput(f);
1032			f = ERR_PTR(error);
1033		}
1034	}
1035	return f;
1036}
1037EXPORT_SYMBOL(dentry_open);
1038
1039/**
1040 * dentry_create - Create and open a file
1041 * @path: path to create
1042 * @flags: O_ flags
1043 * @mode: mode bits for new file
1044 * @cred: credentials to use
1045 *
1046 * Caller must hold the parent directory's lock, and have prepared
1047 * a negative dentry, placed in @path->dentry, for the new file.
1048 *
1049 * Caller sets @path->mnt to the vfsmount of the filesystem where
1050 * the new file is to be created. The parent directory and the
1051 * negative dentry must reside on the same filesystem instance.
1052 *
1053 * On success, returns a "struct file *". Otherwise a ERR_PTR
1054 * is returned.
1055 */
1056struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1057			   const struct cred *cred)
1058{
1059	struct file *f;
1060	int error;
1061
1062	validate_creds(cred);
1063	f = alloc_empty_file(flags, cred);
1064	if (IS_ERR(f))
1065		return f;
1066
1067	error = vfs_create(mnt_user_ns(path->mnt),
1068			   d_inode(path->dentry->d_parent),
1069			   path->dentry, mode, true);
1070	if (!error)
1071		error = vfs_open(path, f);
1072
1073	if (unlikely(error)) {
1074		fput(f);
1075		return ERR_PTR(error);
1076	}
1077	return f;
1078}
1079EXPORT_SYMBOL(dentry_create);
1080
1081struct file *open_with_fake_path(const struct path *path, int flags,
 
 
 
 
 
 
 
 
 
 
 
 
 
1082				struct inode *inode, const struct cred *cred)
1083{
1084	struct file *f = alloc_empty_file_noaccount(flags, cred);
1085	if (!IS_ERR(f)) {
1086		int error;
1087
1088		f->f_path = *path;
1089		error = do_dentry_open(f, inode, NULL);
1090		if (error) {
1091			fput(f);
1092			f = ERR_PTR(error);
1093		}
 
 
 
1094	}
1095	return f;
1096}
1097EXPORT_SYMBOL(open_with_fake_path);
1098
1099#define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1100#define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1101
1102inline struct open_how build_open_how(int flags, umode_t mode)
1103{
1104	struct open_how how = {
1105		.flags = flags & VALID_OPEN_FLAGS,
1106		.mode = mode & S_IALLUGO,
1107	};
1108
1109	/* O_PATH beats everything else. */
1110	if (how.flags & O_PATH)
1111		how.flags &= O_PATH_FLAGS;
1112	/* Modes should only be set for create-like flags. */
1113	if (!WILL_CREATE(how.flags))
1114		how.mode = 0;
1115	return how;
1116}
1117
1118inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1119{
1120	u64 flags = how->flags;
1121	u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1122	int lookup_flags = 0;
1123	int acc_mode = ACC_MODE(flags);
1124
1125	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1126			 "struct open_flags doesn't yet handle flags > 32 bits");
1127
1128	/*
1129	 * Strip flags that either shouldn't be set by userspace like
1130	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1131	 * open_flags like O_CLOEXEC.
1132	 */
1133	flags &= ~strip;
1134
1135	/*
1136	 * Older syscalls implicitly clear all of the invalid flags or argument
1137	 * values before calling build_open_flags(), but openat2(2) checks all
1138	 * of its arguments.
1139	 */
1140	if (flags & ~VALID_OPEN_FLAGS)
1141		return -EINVAL;
1142	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1143		return -EINVAL;
1144
1145	/* Scoping flags are mutually exclusive. */
1146	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1147		return -EINVAL;
1148
1149	/* Deal with the mode. */
1150	if (WILL_CREATE(flags)) {
1151		if (how->mode & ~S_IALLUGO)
1152			return -EINVAL;
1153		op->mode = how->mode | S_IFREG;
1154	} else {
1155		if (how->mode != 0)
1156			return -EINVAL;
1157		op->mode = 0;
1158	}
1159
1160	/*
1161	 * In order to ensure programs get explicit errors when trying to use
1162	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1163	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1164	 * have to require userspace to explicitly set it.
1165	 */
 
 
 
 
1166	if (flags & __O_TMPFILE) {
1167		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 
 
 
 
 
1168			return -EINVAL;
1169		if (!(acc_mode & MAY_WRITE))
1170			return -EINVAL;
1171	}
1172	if (flags & O_PATH) {
1173		/* O_PATH only permits certain other flags to be set. */
1174		if (flags & ~O_PATH_FLAGS)
1175			return -EINVAL;
1176		acc_mode = 0;
1177	}
1178
1179	/*
1180	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1181	 * check for O_DSYNC if the need any syncing at all we enforce it's
1182	 * always set instead of having to deal with possibly weird behaviour
1183	 * for malicious applications setting only __O_SYNC.
1184	 */
1185	if (flags & __O_SYNC)
1186		flags |= O_DSYNC;
1187
1188	op->open_flag = flags;
1189
1190	/* O_TRUNC implies we need access checks for write permissions */
1191	if (flags & O_TRUNC)
1192		acc_mode |= MAY_WRITE;
1193
1194	/* Allow the LSM permission hook to distinguish append
1195	   access from general write access. */
1196	if (flags & O_APPEND)
1197		acc_mode |= MAY_APPEND;
1198
1199	op->acc_mode = acc_mode;
1200
1201	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1202
1203	if (flags & O_CREAT) {
1204		op->intent |= LOOKUP_CREATE;
1205		if (flags & O_EXCL) {
1206			op->intent |= LOOKUP_EXCL;
1207			flags |= O_NOFOLLOW;
1208		}
1209	}
1210
1211	if (flags & O_DIRECTORY)
1212		lookup_flags |= LOOKUP_DIRECTORY;
1213	if (!(flags & O_NOFOLLOW))
1214		lookup_flags |= LOOKUP_FOLLOW;
1215
1216	if (how->resolve & RESOLVE_NO_XDEV)
1217		lookup_flags |= LOOKUP_NO_XDEV;
1218	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1219		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1220	if (how->resolve & RESOLVE_NO_SYMLINKS)
1221		lookup_flags |= LOOKUP_NO_SYMLINKS;
1222	if (how->resolve & RESOLVE_BENEATH)
1223		lookup_flags |= LOOKUP_BENEATH;
1224	if (how->resolve & RESOLVE_IN_ROOT)
1225		lookup_flags |= LOOKUP_IN_ROOT;
1226	if (how->resolve & RESOLVE_CACHED) {
1227		/* Don't bother even trying for create/truncate/tmpfile open */
1228		if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1229			return -EAGAIN;
1230		lookup_flags |= LOOKUP_CACHED;
1231	}
1232
1233	op->lookup_flags = lookup_flags;
1234	return 0;
1235}
1236
1237/**
1238 * file_open_name - open file and return file pointer
1239 *
1240 * @name:	struct filename containing path to open
1241 * @flags:	open flags as per the open(2) second argument
1242 * @mode:	mode for the new file if O_CREAT is set, else ignored
1243 *
1244 * This is the helper to open a file from kernelspace if you really
1245 * have to.  But in generally you should not do this, so please move
1246 * along, nothing to see here..
1247 */
1248struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1249{
1250	struct open_flags op;
1251	struct open_how how = build_open_how(flags, mode);
1252	int err = build_open_flags(&how, &op);
1253	if (err)
1254		return ERR_PTR(err);
1255	return do_filp_open(AT_FDCWD, name, &op);
1256}
1257
1258/**
1259 * filp_open - open file and return file pointer
1260 *
1261 * @filename:	path to open
1262 * @flags:	open flags as per the open(2) second argument
1263 * @mode:	mode for the new file if O_CREAT is set, else ignored
1264 *
1265 * This is the helper to open a file from kernelspace if you really
1266 * have to.  But in generally you should not do this, so please move
1267 * along, nothing to see here..
1268 */
1269struct file *filp_open(const char *filename, int flags, umode_t mode)
1270{
1271	struct filename *name = getname_kernel(filename);
1272	struct file *file = ERR_CAST(name);
1273	
1274	if (!IS_ERR(name)) {
1275		file = file_open_name(name, flags, mode);
1276		putname(name);
1277	}
1278	return file;
1279}
1280EXPORT_SYMBOL(filp_open);
1281
1282struct file *file_open_root(const struct path *root,
1283			    const char *filename, int flags, umode_t mode)
1284{
1285	struct open_flags op;
1286	struct open_how how = build_open_how(flags, mode);
1287	int err = build_open_flags(&how, &op);
1288	if (err)
1289		return ERR_PTR(err);
1290	return do_file_open_root(root, filename, &op);
1291}
1292EXPORT_SYMBOL(file_open_root);
1293
1294static long do_sys_openat2(int dfd, const char __user *filename,
1295			   struct open_how *how)
1296{
1297	struct open_flags op;
1298	int fd = build_open_flags(how, &op);
1299	struct filename *tmp;
1300
1301	if (fd)
1302		return fd;
1303
1304	tmp = getname(filename);
1305	if (IS_ERR(tmp))
1306		return PTR_ERR(tmp);
1307
1308	fd = get_unused_fd_flags(how->flags);
1309	if (fd >= 0) {
1310		struct file *f = do_filp_open(dfd, tmp, &op);
1311		if (IS_ERR(f)) {
1312			put_unused_fd(fd);
1313			fd = PTR_ERR(f);
1314		} else {
1315			fsnotify_open(f);
1316			fd_install(fd, f);
1317		}
1318	}
1319	putname(tmp);
1320	return fd;
1321}
1322
1323long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1324{
1325	struct open_how how = build_open_how(flags, mode);
1326	return do_sys_openat2(dfd, filename, &how);
1327}
1328
1329
1330SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1331{
1332	if (force_o_largefile())
1333		flags |= O_LARGEFILE;
1334	return do_sys_open(AT_FDCWD, filename, flags, mode);
1335}
1336
1337SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1338		umode_t, mode)
1339{
1340	if (force_o_largefile())
1341		flags |= O_LARGEFILE;
1342	return do_sys_open(dfd, filename, flags, mode);
1343}
1344
1345SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1346		struct open_how __user *, how, size_t, usize)
1347{
1348	int err;
1349	struct open_how tmp;
1350
1351	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1352	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1353
1354	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1355		return -EINVAL;
1356
1357	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1358	if (err)
1359		return err;
1360
1361	audit_openat2_how(&tmp);
1362
1363	/* O_LARGEFILE is only allowed for non-O_PATH. */
1364	if (!(tmp.flags & O_PATH) && force_o_largefile())
1365		tmp.flags |= O_LARGEFILE;
1366
1367	return do_sys_openat2(dfd, filename, &tmp);
1368}
1369
1370#ifdef CONFIG_COMPAT
1371/*
1372 * Exactly like sys_open(), except that it doesn't set the
1373 * O_LARGEFILE flag.
1374 */
1375COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1376{
1377	return do_sys_open(AT_FDCWD, filename, flags, mode);
1378}
1379
1380/*
1381 * Exactly like sys_openat(), except that it doesn't set the
1382 * O_LARGEFILE flag.
1383 */
1384COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1385{
1386	return do_sys_open(dfd, filename, flags, mode);
1387}
1388#endif
1389
1390#ifndef __alpha__
1391
1392/*
1393 * For backward compatibility?  Maybe this should be moved
1394 * into arch/i386 instead?
1395 */
1396SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1397{
1398	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1399
1400	if (force_o_largefile())
1401		flags |= O_LARGEFILE;
1402	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1403}
1404#endif
1405
1406/*
1407 * "id" is the POSIX thread ID. We use the
1408 * files pointer for this..
1409 */
1410int filp_close(struct file *filp, fl_owner_t id)
1411{
1412	int retval = 0;
1413
1414	if (!file_count(filp)) {
1415		printk(KERN_ERR "VFS: Close: file count is 0\n");
 
1416		return 0;
1417	}
1418
1419	if (filp->f_op->flush)
1420		retval = filp->f_op->flush(filp, id);
1421
1422	if (likely(!(filp->f_mode & FMODE_PATH))) {
1423		dnotify_flush(filp, id);
1424		locks_remove_posix(filp, id);
1425	}
1426	fput(filp);
1427	return retval;
1428}
1429
 
 
 
 
 
 
 
 
 
1430EXPORT_SYMBOL(filp_close);
1431
1432/*
1433 * Careful here! We test whether the file pointer is NULL before
1434 * releasing the fd. This ensures that one clone task can't release
1435 * an fd while another clone is opening it.
1436 */
1437SYSCALL_DEFINE1(close, unsigned int, fd)
1438{
1439	int retval = close_fd(fd);
 
 
 
 
 
 
 
 
 
 
 
 
 
1440
1441	/* can't restart close syscall because file table entry was cleared */
1442	if (unlikely(retval == -ERESTARTSYS ||
1443		     retval == -ERESTARTNOINTR ||
1444		     retval == -ERESTARTNOHAND ||
1445		     retval == -ERESTART_RESTARTBLOCK))
1446		retval = -EINTR;
1447
1448	return retval;
1449}
1450
1451/**
1452 * close_range() - Close all file descriptors in a given range.
1453 *
1454 * @fd:     starting file descriptor to close
1455 * @max_fd: last file descriptor to close
1456 * @flags:  reserved for future extensions
1457 *
1458 * This closes a range of file descriptors. All file descriptors
1459 * from @fd up to and including @max_fd are closed.
1460 * Currently, errors to close a given file descriptor are ignored.
1461 */
1462SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1463		unsigned int, flags)
1464{
1465	return __close_range(fd, max_fd, flags);
1466}
1467
1468/*
1469 * This routine simulates a hangup on the tty, to arrange that users
1470 * are given clean terminals at login time.
1471 */
1472SYSCALL_DEFINE0(vhangup)
1473{
1474	if (capable(CAP_SYS_TTY_CONFIG)) {
1475		tty_vhangup_self();
1476		return 0;
1477	}
1478	return -EPERM;
1479}
1480
1481/*
1482 * Called when an inode is about to be open.
1483 * We use this to disallow opening large files on 32bit systems if
1484 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1485 * on this flag in sys_open.
1486 */
1487int generic_file_open(struct inode * inode, struct file * filp)
1488{
1489	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1490		return -EOVERFLOW;
1491	return 0;
1492}
1493
1494EXPORT_SYMBOL(generic_file_open);
1495
1496/*
1497 * This is used by subsystems that don't want seekable
1498 * file descriptors. The function is not supposed to ever fail, the only
1499 * reason it returns an 'int' and not 'void' is so that it can be plugged
1500 * directly into file_operations structure.
1501 */
1502int nonseekable_open(struct inode *inode, struct file *filp)
1503{
1504	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1505	return 0;
1506}
1507
1508EXPORT_SYMBOL(nonseekable_open);
1509
1510/*
1511 * stream_open is used by subsystems that want stream-like file descriptors.
1512 * Such file descriptors are not seekable and don't have notion of position
1513 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1514 * Contrary to file descriptors of other regular files, .read() and .write()
1515 * can run simultaneously.
1516 *
1517 * stream_open never fails and is marked to return int so that it could be
1518 * directly used as file_operations.open .
1519 */
1520int stream_open(struct inode *inode, struct file *filp)
1521{
1522	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1523	filp->f_mode |= FMODE_STREAM;
1524	return 0;
1525}
1526
1527EXPORT_SYMBOL(stream_open);
v6.8
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/open.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8#include <linux/string.h>
   9#include <linux/mm.h>
  10#include <linux/file.h>
  11#include <linux/fdtable.h>
  12#include <linux/fsnotify.h>
  13#include <linux/module.h>
  14#include <linux/tty.h>
  15#include <linux/namei.h>
  16#include <linux/backing-dev.h>
  17#include <linux/capability.h>
  18#include <linux/securebits.h>
  19#include <linux/security.h>
  20#include <linux/mount.h>
  21#include <linux/fcntl.h>
  22#include <linux/slab.h>
  23#include <linux/uaccess.h>
  24#include <linux/fs.h>
  25#include <linux/personality.h>
  26#include <linux/pagemap.h>
  27#include <linux/syscalls.h>
  28#include <linux/rcupdate.h>
  29#include <linux/audit.h>
  30#include <linux/falloc.h>
  31#include <linux/fs_struct.h>
  32#include <linux/ima.h>
  33#include <linux/dnotify.h>
  34#include <linux/compat.h>
  35#include <linux/mnt_idmapping.h>
  36#include <linux/filelock.h>
  37
  38#include "internal.h"
  39
  40int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
  41		loff_t length, unsigned int time_attrs, struct file *filp)
  42{
  43	int ret;
  44	struct iattr newattrs;
  45
  46	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  47	if (length < 0)
  48		return -EINVAL;
  49
  50	newattrs.ia_size = length;
  51	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  52	if (filp) {
  53		newattrs.ia_file = filp;
  54		newattrs.ia_valid |= ATTR_FILE;
  55	}
  56
  57	/* Remove suid, sgid, and file capabilities on truncate too */
  58	ret = dentry_needs_remove_privs(idmap, dentry);
  59	if (ret < 0)
  60		return ret;
  61	if (ret)
  62		newattrs.ia_valid |= ret | ATTR_FORCE;
  63
  64	inode_lock(dentry->d_inode);
  65	/* Note any delegations or leases have already been broken: */
  66	ret = notify_change(idmap, dentry, &newattrs, NULL);
  67	inode_unlock(dentry->d_inode);
  68	return ret;
  69}
  70
  71long vfs_truncate(const struct path *path, loff_t length)
  72{
  73	struct mnt_idmap *idmap;
  74	struct inode *inode;
  75	long error;
  76
  77	inode = path->dentry->d_inode;
  78
  79	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80	if (S_ISDIR(inode->i_mode))
  81		return -EISDIR;
  82	if (!S_ISREG(inode->i_mode))
  83		return -EINVAL;
  84
  85	error = mnt_want_write(path->mnt);
  86	if (error)
  87		goto out;
  88
  89	idmap = mnt_idmap(path->mnt);
  90	error = inode_permission(idmap, inode, MAY_WRITE);
  91	if (error)
  92		goto mnt_drop_write_and_out;
  93
  94	error = -EPERM;
  95	if (IS_APPEND(inode))
  96		goto mnt_drop_write_and_out;
  97
  98	error = get_write_access(inode);
  99	if (error)
 100		goto mnt_drop_write_and_out;
 101
 102	/*
 103	 * Make sure that there are no leases.  get_write_access() protects
 104	 * against the truncate racing with a lease-granting setlease().
 105	 */
 106	error = break_lease(inode, O_WRONLY);
 107	if (error)
 108		goto put_write_and_out;
 109
 110	error = security_path_truncate(path);
 111	if (!error)
 112		error = do_truncate(idmap, path->dentry, length, 0, NULL);
 113
 114put_write_and_out:
 115	put_write_access(inode);
 116mnt_drop_write_and_out:
 117	mnt_drop_write(path->mnt);
 118out:
 119	return error;
 120}
 121EXPORT_SYMBOL_GPL(vfs_truncate);
 122
 123long do_sys_truncate(const char __user *pathname, loff_t length)
 124{
 125	unsigned int lookup_flags = LOOKUP_FOLLOW;
 126	struct path path;
 127	int error;
 128
 129	if (length < 0)	/* sorry, but loff_t says... */
 130		return -EINVAL;
 131
 132retry:
 133	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 134	if (!error) {
 135		error = vfs_truncate(&path, length);
 136		path_put(&path);
 137	}
 138	if (retry_estale(error, lookup_flags)) {
 139		lookup_flags |= LOOKUP_REVAL;
 140		goto retry;
 141	}
 142	return error;
 143}
 144
 145SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 146{
 147	return do_sys_truncate(path, length);
 148}
 149
 150#ifdef CONFIG_COMPAT
 151COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 152{
 153	return do_sys_truncate(path, length);
 154}
 155#endif
 156
 157long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 158{
 159	struct inode *inode;
 160	struct dentry *dentry;
 161	struct fd f;
 162	int error;
 163
 164	error = -EINVAL;
 165	if (length < 0)
 166		goto out;
 167	error = -EBADF;
 168	f = fdget(fd);
 169	if (!f.file)
 170		goto out;
 171
 172	/* explicitly opened as large or we are on 64-bit box */
 173	if (f.file->f_flags & O_LARGEFILE)
 174		small = 0;
 175
 176	dentry = f.file->f_path.dentry;
 177	inode = dentry->d_inode;
 178	error = -EINVAL;
 179	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 180		goto out_putf;
 181
 182	error = -EINVAL;
 183	/* Cannot ftruncate over 2^31 bytes without large file support */
 184	if (small && length > MAX_NON_LFS)
 185		goto out_putf;
 186
 187	error = -EPERM;
 188	/* Check IS_APPEND on real upper inode */
 189	if (IS_APPEND(file_inode(f.file)))
 190		goto out_putf;
 191	sb_start_write(inode->i_sb);
 192	error = security_file_truncate(f.file);
 193	if (!error)
 194		error = do_truncate(file_mnt_idmap(f.file), dentry, length,
 195				    ATTR_MTIME | ATTR_CTIME, f.file);
 196	sb_end_write(inode->i_sb);
 197out_putf:
 198	fdput(f);
 199out:
 200	return error;
 201}
 202
 203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 204{
 205	return do_sys_ftruncate(fd, length, 1);
 206}
 207
 208#ifdef CONFIG_COMPAT
 209COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 210{
 211	return do_sys_ftruncate(fd, length, 1);
 212}
 213#endif
 214
 215/* LFS versions of truncate are only needed on 32 bit machines */
 216#if BITS_PER_LONG == 32
 217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 218{
 219	return do_sys_truncate(path, length);
 220}
 221
 222SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 223{
 224	return do_sys_ftruncate(fd, length, 0);
 225}
 226#endif /* BITS_PER_LONG == 32 */
 227
 228#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
 229COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
 230		       compat_arg_u64_dual(length))
 231{
 232	return ksys_truncate(pathname, compat_arg_u64_glue(length));
 233}
 234#endif
 235
 236#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
 237COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
 238		       compat_arg_u64_dual(length))
 239{
 240	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
 241}
 242#endif
 243
 244int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 245{
 246	struct inode *inode = file_inode(file);
 247	long ret;
 248
 249	if (offset < 0 || len <= 0)
 250		return -EINVAL;
 251
 252	/* Return error if mode is not supported */
 253	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 254		return -EOPNOTSUPP;
 255
 256	/* Punch hole and zero range are mutually exclusive */
 257	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 258	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 259		return -EOPNOTSUPP;
 260
 261	/* Punch hole must have keep size set */
 262	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 263	    !(mode & FALLOC_FL_KEEP_SIZE))
 264		return -EOPNOTSUPP;
 265
 266	/* Collapse range should only be used exclusively. */
 267	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 268	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 269		return -EINVAL;
 270
 271	/* Insert range should only be used exclusively. */
 272	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 273	    (mode & ~FALLOC_FL_INSERT_RANGE))
 274		return -EINVAL;
 275
 276	/* Unshare range should only be used with allocate mode. */
 277	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 278	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 279		return -EINVAL;
 280
 281	if (!(file->f_mode & FMODE_WRITE))
 282		return -EBADF;
 283
 284	/*
 285	 * We can only allow pure fallocate on append only files
 286	 */
 287	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 288		return -EPERM;
 289
 290	if (IS_IMMUTABLE(inode))
 291		return -EPERM;
 292
 293	/*
 294	 * We cannot allow any fallocate operation on an active swapfile
 295	 */
 296	if (IS_SWAPFILE(inode))
 297		return -ETXTBSY;
 298
 299	/*
 300	 * Revalidate the write permissions, in case security policy has
 301	 * changed since the files were opened.
 302	 */
 303	ret = security_file_permission(file, MAY_WRITE);
 304	if (ret)
 305		return ret;
 306
 307	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
 308	if (ret)
 309		return ret;
 310
 311	if (S_ISFIFO(inode->i_mode))
 312		return -ESPIPE;
 313
 314	if (S_ISDIR(inode->i_mode))
 315		return -EISDIR;
 316
 317	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 318		return -ENODEV;
 319
 320	/* Check for wrap through zero too */
 321	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 322		return -EFBIG;
 323
 324	if (!file->f_op->fallocate)
 325		return -EOPNOTSUPP;
 326
 327	file_start_write(file);
 328	ret = file->f_op->fallocate(file, mode, offset, len);
 329
 330	/*
 331	 * Create inotify and fanotify events.
 332	 *
 333	 * To keep the logic simple always create events if fallocate succeeds.
 334	 * This implies that events are even created if the file size remains
 335	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 336	 */
 337	if (ret == 0)
 338		fsnotify_modify(file);
 339
 340	file_end_write(file);
 341	return ret;
 342}
 343EXPORT_SYMBOL_GPL(vfs_fallocate);
 344
 345int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 346{
 347	struct fd f = fdget(fd);
 348	int error = -EBADF;
 349
 350	if (f.file) {
 351		error = vfs_fallocate(f.file, mode, offset, len);
 352		fdput(f);
 353	}
 354	return error;
 355}
 356
 357SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 358{
 359	return ksys_fallocate(fd, mode, offset, len);
 360}
 361
 362#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
 363COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
 364		       compat_arg_u64_dual(len))
 365{
 366	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
 367			      compat_arg_u64_glue(len));
 368}
 369#endif
 370
 371/*
 372 * access() needs to use the real uid/gid, not the effective uid/gid.
 373 * We do this by temporarily clearing all FS-related capabilities and
 374 * switching the fsuid/fsgid around to the real ones.
 375 *
 376 * Creating new credentials is expensive, so we try to skip doing it,
 377 * which we can if the result would match what we already got.
 378 */
 379static bool access_need_override_creds(int flags)
 380{
 381	const struct cred *cred;
 382
 383	if (flags & AT_EACCESS)
 384		return false;
 385
 386	cred = current_cred();
 387	if (!uid_eq(cred->fsuid, cred->uid) ||
 388	    !gid_eq(cred->fsgid, cred->gid))
 389		return true;
 390
 391	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 392		kuid_t root_uid = make_kuid(cred->user_ns, 0);
 393		if (!uid_eq(cred->uid, root_uid)) {
 394			if (!cap_isclear(cred->cap_effective))
 395				return true;
 396		} else {
 397			if (!cap_isidentical(cred->cap_effective,
 398			    cred->cap_permitted))
 399				return true;
 400		}
 401	}
 402
 403	return false;
 404}
 405
 406static const struct cred *access_override_creds(void)
 407{
 408	const struct cred *old_cred;
 409	struct cred *override_cred;
 410
 411	override_cred = prepare_creds();
 412	if (!override_cred)
 413		return NULL;
 414
 415	/*
 416	 * XXX access_need_override_creds performs checks in hopes of skipping
 417	 * this work. Make sure it stays in sync if making any changes in this
 418	 * routine.
 419	 */
 420
 421	override_cred->fsuid = override_cred->uid;
 422	override_cred->fsgid = override_cred->gid;
 423
 424	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 425		/* Clear the capabilities if we switch to a non-root user */
 426		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 427		if (!uid_eq(override_cred->uid, root_uid))
 428			cap_clear(override_cred->cap_effective);
 429		else
 430			override_cred->cap_effective =
 431				override_cred->cap_permitted;
 432	}
 433
 434	/*
 435	 * The new set of credentials can *only* be used in
 436	 * task-synchronous circumstances, and does not need
 437	 * RCU freeing, unless somebody then takes a separate
 438	 * reference to it.
 439	 *
 440	 * NOTE! This is _only_ true because this credential
 441	 * is used purely for override_creds() that installs
 442	 * it as the subjective cred. Other threads will be
 443	 * accessing ->real_cred, not the subjective cred.
 444	 *
 445	 * If somebody _does_ make a copy of this (using the
 446	 * 'get_current_cred()' function), that will clear the
 447	 * non_rcu field, because now that other user may be
 448	 * expecting RCU freeing. But normal thread-synchronous
 449	 * cred accesses will keep things non-racy to avoid RCU
 450	 * freeing.
 451	 */
 452	override_cred->non_rcu = 1;
 453
 454	old_cred = override_creds(override_cred);
 455
 456	/* override_cred() gets its own ref */
 457	put_cred(override_cred);
 458
 459	return old_cred;
 460}
 461
 462static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 463{
 464	struct path path;
 465	struct inode *inode;
 466	int res;
 467	unsigned int lookup_flags = LOOKUP_FOLLOW;
 468	const struct cred *old_cred = NULL;
 469
 470	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 471		return -EINVAL;
 472
 473	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 474		return -EINVAL;
 475
 476	if (flags & AT_SYMLINK_NOFOLLOW)
 477		lookup_flags &= ~LOOKUP_FOLLOW;
 478	if (flags & AT_EMPTY_PATH)
 479		lookup_flags |= LOOKUP_EMPTY;
 480
 481	if (access_need_override_creds(flags)) {
 482		old_cred = access_override_creds();
 483		if (!old_cred)
 484			return -ENOMEM;
 485	}
 486
 487retry:
 488	res = user_path_at(dfd, filename, lookup_flags, &path);
 489	if (res)
 490		goto out;
 491
 492	inode = d_backing_inode(path.dentry);
 493
 494	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 495		/*
 496		 * MAY_EXEC on regular files is denied if the fs is mounted
 497		 * with the "noexec" flag.
 498		 */
 499		res = -EACCES;
 500		if (path_noexec(&path))
 501			goto out_path_release;
 502	}
 503
 504	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
 505	/* SuS v2 requires we report a read only fs too */
 506	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 507		goto out_path_release;
 508	/*
 509	 * This is a rare case where using __mnt_is_readonly()
 510	 * is OK without a mnt_want/drop_write() pair.  Since
 511	 * no actual write to the fs is performed here, we do
 512	 * not need to telegraph to that to anyone.
 513	 *
 514	 * By doing this, we accept that this access is
 515	 * inherently racy and know that the fs may change
 516	 * state before we even see this result.
 517	 */
 518	if (__mnt_is_readonly(path.mnt))
 519		res = -EROFS;
 520
 521out_path_release:
 522	path_put(&path);
 523	if (retry_estale(res, lookup_flags)) {
 524		lookup_flags |= LOOKUP_REVAL;
 525		goto retry;
 526	}
 527out:
 528	if (old_cred)
 529		revert_creds(old_cred);
 530
 531	return res;
 532}
 533
 534SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 535{
 536	return do_faccessat(dfd, filename, mode, 0);
 537}
 538
 539SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 540		int, flags)
 541{
 542	return do_faccessat(dfd, filename, mode, flags);
 543}
 544
 545SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 546{
 547	return do_faccessat(AT_FDCWD, filename, mode, 0);
 548}
 549
 550SYSCALL_DEFINE1(chdir, const char __user *, filename)
 551{
 552	struct path path;
 553	int error;
 554	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 555retry:
 556	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 557	if (error)
 558		goto out;
 559
 560	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 561	if (error)
 562		goto dput_and_out;
 563
 564	set_fs_pwd(current->fs, &path);
 565
 566dput_and_out:
 567	path_put(&path);
 568	if (retry_estale(error, lookup_flags)) {
 569		lookup_flags |= LOOKUP_REVAL;
 570		goto retry;
 571	}
 572out:
 573	return error;
 574}
 575
 576SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 577{
 578	struct fd f = fdget_raw(fd);
 579	int error;
 580
 581	error = -EBADF;
 582	if (!f.file)
 583		goto out;
 584
 585	error = -ENOTDIR;
 586	if (!d_can_lookup(f.file->f_path.dentry))
 587		goto out_putf;
 588
 589	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
 590	if (!error)
 591		set_fs_pwd(current->fs, &f.file->f_path);
 592out_putf:
 593	fdput(f);
 594out:
 595	return error;
 596}
 597
 598SYSCALL_DEFINE1(chroot, const char __user *, filename)
 599{
 600	struct path path;
 601	int error;
 602	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 603retry:
 604	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 605	if (error)
 606		goto out;
 607
 608	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 609	if (error)
 610		goto dput_and_out;
 611
 612	error = -EPERM;
 613	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 614		goto dput_and_out;
 615	error = security_path_chroot(&path);
 616	if (error)
 617		goto dput_and_out;
 618
 619	set_fs_root(current->fs, &path);
 620	error = 0;
 621dput_and_out:
 622	path_put(&path);
 623	if (retry_estale(error, lookup_flags)) {
 624		lookup_flags |= LOOKUP_REVAL;
 625		goto retry;
 626	}
 627out:
 628	return error;
 629}
 630
 631int chmod_common(const struct path *path, umode_t mode)
 632{
 633	struct inode *inode = path->dentry->d_inode;
 634	struct inode *delegated_inode = NULL;
 635	struct iattr newattrs;
 636	int error;
 637
 638	error = mnt_want_write(path->mnt);
 639	if (error)
 640		return error;
 641retry_deleg:
 642	inode_lock(inode);
 643	error = security_path_chmod(path, mode);
 644	if (error)
 645		goto out_unlock;
 646	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 647	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 648	error = notify_change(mnt_idmap(path->mnt), path->dentry,
 649			      &newattrs, &delegated_inode);
 650out_unlock:
 651	inode_unlock(inode);
 652	if (delegated_inode) {
 653		error = break_deleg_wait(&delegated_inode);
 654		if (!error)
 655			goto retry_deleg;
 656	}
 657	mnt_drop_write(path->mnt);
 658	return error;
 659}
 660
 661int vfs_fchmod(struct file *file, umode_t mode)
 662{
 663	audit_file(file);
 664	return chmod_common(&file->f_path, mode);
 665}
 666
 667SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 668{
 669	struct fd f = fdget(fd);
 670	int err = -EBADF;
 671
 672	if (f.file) {
 673		err = vfs_fchmod(f.file, mode);
 674		fdput(f);
 675	}
 676	return err;
 677}
 678
 679static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
 680		       unsigned int flags)
 681{
 682	struct path path;
 683	int error;
 684	unsigned int lookup_flags;
 685
 686	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
 687		return -EINVAL;
 688
 689	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 690	if (flags & AT_EMPTY_PATH)
 691		lookup_flags |= LOOKUP_EMPTY;
 692
 693retry:
 694	error = user_path_at(dfd, filename, lookup_flags, &path);
 695	if (!error) {
 696		error = chmod_common(&path, mode);
 697		path_put(&path);
 698		if (retry_estale(error, lookup_flags)) {
 699			lookup_flags |= LOOKUP_REVAL;
 700			goto retry;
 701		}
 702	}
 703	return error;
 704}
 705
 706SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
 707		umode_t, mode, unsigned int, flags)
 708{
 709	return do_fchmodat(dfd, filename, mode, flags);
 710}
 711
 712SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 713		umode_t, mode)
 714{
 715	return do_fchmodat(dfd, filename, mode, 0);
 716}
 717
 718SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 719{
 720	return do_fchmodat(AT_FDCWD, filename, mode, 0);
 721}
 722
 723/*
 
 
 
 724 * Check whether @kuid is valid and if so generate and set vfsuid_t in
 725 * ia_vfsuid.
 726 *
 727 * Return: true if @kuid is valid, false if not.
 728 */
 729static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
 730{
 731	if (!uid_valid(kuid))
 732		return false;
 733	attr->ia_valid |= ATTR_UID;
 734	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
 735	return true;
 736}
 737
 738/*
 
 
 
 739 * Check whether @kgid is valid and if so generate and set vfsgid_t in
 740 * ia_vfsgid.
 741 *
 742 * Return: true if @kgid is valid, false if not.
 743 */
 744static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
 745{
 746	if (!gid_valid(kgid))
 747		return false;
 748	attr->ia_valid |= ATTR_GID;
 749	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
 750	return true;
 751}
 752
 753int chown_common(const struct path *path, uid_t user, gid_t group)
 754{
 755	struct mnt_idmap *idmap;
 756	struct user_namespace *fs_userns;
 757	struct inode *inode = path->dentry->d_inode;
 758	struct inode *delegated_inode = NULL;
 759	int error;
 760	struct iattr newattrs;
 761	kuid_t uid;
 762	kgid_t gid;
 763
 764	uid = make_kuid(current_user_ns(), user);
 765	gid = make_kgid(current_user_ns(), group);
 766
 767	idmap = mnt_idmap(path->mnt);
 768	fs_userns = i_user_ns(inode);
 769
 770retry_deleg:
 771	newattrs.ia_vfsuid = INVALID_VFSUID;
 772	newattrs.ia_vfsgid = INVALID_VFSGID;
 773	newattrs.ia_valid =  ATTR_CTIME;
 774	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 775		return -EINVAL;
 776	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 777		return -EINVAL;
 778	inode_lock(inode);
 779	if (!S_ISDIR(inode->i_mode))
 780		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
 781				     setattr_should_drop_sgid(idmap, inode);
 782	/* Continue to send actual fs values, not the mount values. */
 783	error = security_path_chown(
 784		path,
 785		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
 786		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
 787	if (!error)
 788		error = notify_change(idmap, path->dentry, &newattrs,
 789				      &delegated_inode);
 790	inode_unlock(inode);
 791	if (delegated_inode) {
 792		error = break_deleg_wait(&delegated_inode);
 793		if (!error)
 794			goto retry_deleg;
 795	}
 796	return error;
 797}
 798
 799int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 800		int flag)
 801{
 802	struct path path;
 803	int error = -EINVAL;
 804	int lookup_flags;
 805
 806	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 807		goto out;
 808
 809	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 810	if (flag & AT_EMPTY_PATH)
 811		lookup_flags |= LOOKUP_EMPTY;
 812retry:
 813	error = user_path_at(dfd, filename, lookup_flags, &path);
 814	if (error)
 815		goto out;
 816	error = mnt_want_write(path.mnt);
 817	if (error)
 818		goto out_release;
 819	error = chown_common(&path, user, group);
 820	mnt_drop_write(path.mnt);
 821out_release:
 822	path_put(&path);
 823	if (retry_estale(error, lookup_flags)) {
 824		lookup_flags |= LOOKUP_REVAL;
 825		goto retry;
 826	}
 827out:
 828	return error;
 829}
 830
 831SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 832		gid_t, group, int, flag)
 833{
 834	return do_fchownat(dfd, filename, user, group, flag);
 835}
 836
 837SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 838{
 839	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 840}
 841
 842SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 843{
 844	return do_fchownat(AT_FDCWD, filename, user, group,
 845			   AT_SYMLINK_NOFOLLOW);
 846}
 847
 848int vfs_fchown(struct file *file, uid_t user, gid_t group)
 849{
 850	int error;
 851
 852	error = mnt_want_write_file(file);
 853	if (error)
 854		return error;
 855	audit_file(file);
 856	error = chown_common(&file->f_path, user, group);
 857	mnt_drop_write_file(file);
 858	return error;
 859}
 860
 861int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 862{
 863	struct fd f = fdget(fd);
 864	int error = -EBADF;
 865
 866	if (f.file) {
 867		error = vfs_fchown(f.file, user, group);
 868		fdput(f);
 869	}
 870	return error;
 871}
 872
 873SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 874{
 875	return ksys_fchown(fd, user, group);
 876}
 877
 878static inline int file_get_write_access(struct file *f)
 879{
 880	int error;
 881
 882	error = get_write_access(f->f_inode);
 883	if (unlikely(error))
 884		return error;
 885	error = mnt_get_write_access(f->f_path.mnt);
 886	if (unlikely(error))
 887		goto cleanup_inode;
 888	if (unlikely(f->f_mode & FMODE_BACKING)) {
 889		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
 890		if (unlikely(error))
 891			goto cleanup_mnt;
 892	}
 893	return 0;
 894
 895cleanup_mnt:
 896	mnt_put_write_access(f->f_path.mnt);
 897cleanup_inode:
 898	put_write_access(f->f_inode);
 899	return error;
 900}
 901
 902static int do_dentry_open(struct file *f,
 903			  struct inode *inode,
 904			  int (*open)(struct inode *, struct file *))
 905{
 906	static const struct file_operations empty_fops = {};
 907	int error;
 908
 909	path_get(&f->f_path);
 910	f->f_inode = inode;
 911	f->f_mapping = inode->i_mapping;
 912	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 913	f->f_sb_err = file_sample_sb_err(f);
 914
 915	if (unlikely(f->f_flags & O_PATH)) {
 916		f->f_mode = FMODE_PATH | FMODE_OPENED;
 917		f->f_op = &empty_fops;
 918		return 0;
 919	}
 920
 921	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
 922		i_readcount_inc(inode);
 923	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 924		error = file_get_write_access(f);
 925		if (unlikely(error))
 926			goto cleanup_file;
 
 
 
 
 
 927		f->f_mode |= FMODE_WRITER;
 928	}
 929
 930	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 931	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 932		f->f_mode |= FMODE_ATOMIC_POS;
 933
 934	f->f_op = fops_get(inode->i_fop);
 935	if (WARN_ON(!f->f_op)) {
 936		error = -ENODEV;
 937		goto cleanup_all;
 938	}
 939
 940	error = security_file_open(f);
 941	if (error)
 942		goto cleanup_all;
 943
 944	error = break_lease(file_inode(f), f->f_flags);
 945	if (error)
 946		goto cleanup_all;
 947
 948	/* normally all 3 are set; ->open() can clear them if needed */
 949	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 950	if (!open)
 951		open = f->f_op->open;
 952	if (open) {
 953		error = open(inode, f);
 954		if (error)
 955			goto cleanup_all;
 956	}
 957	f->f_mode |= FMODE_OPENED;
 958	if ((f->f_mode & FMODE_READ) &&
 959	     likely(f->f_op->read || f->f_op->read_iter))
 960		f->f_mode |= FMODE_CAN_READ;
 961	if ((f->f_mode & FMODE_WRITE) &&
 962	     likely(f->f_op->write || f->f_op->write_iter))
 963		f->f_mode |= FMODE_CAN_WRITE;
 964	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 965		f->f_mode &= ~FMODE_LSEEK;
 966	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 967		f->f_mode |= FMODE_CAN_ODIRECT;
 968
 969	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 970	f->f_iocb_flags = iocb_flags(f);
 971
 972	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 973
 974	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 975		return -EINVAL;
 976
 977	/*
 978	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 979	 * cache for this file before processing writes.
 980	 */
 981	if (f->f_mode & FMODE_WRITE) {
 982		/*
 983		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
 984		 * is up to date and the update to i_writecount by
 985		 * get_write_access() is visible. Ensures subsequent insertion
 986		 * of THPs into the page cache will fail.
 987		 */
 988		smp_mb();
 989		if (filemap_nr_thps(inode->i_mapping)) {
 990			struct address_space *mapping = inode->i_mapping;
 991
 992			filemap_invalidate_lock(inode->i_mapping);
 993			/*
 994			 * unmap_mapping_range just need to be called once
 995			 * here, because the private pages is not need to be
 996			 * unmapped mapping (e.g. data segment of dynamic
 997			 * shared libraries here).
 998			 */
 999			unmap_mapping_range(mapping, 0, 0, 0);
1000			truncate_inode_pages(mapping, 0);
1001			filemap_invalidate_unlock(inode->i_mapping);
1002		}
1003	}
1004
1005	/*
1006	 * Once we return a file with FMODE_OPENED, __fput() will call
1007	 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1008	 */
1009	fsnotify_open(f);
1010	return 0;
1011
1012cleanup_all:
1013	if (WARN_ON_ONCE(error > 0))
1014		error = -EINVAL;
1015	fops_put(f->f_op);
1016	put_file_access(f);
1017cleanup_file:
1018	path_put(&f->f_path);
1019	f->f_path.mnt = NULL;
1020	f->f_path.dentry = NULL;
1021	f->f_inode = NULL;
1022	return error;
1023}
1024
1025/**
1026 * finish_open - finish opening a file
1027 * @file: file pointer
1028 * @dentry: pointer to dentry
1029 * @open: open callback
 
1030 *
1031 * This can be used to finish opening a file passed to i_op->atomic_open().
1032 *
1033 * If the open callback is set to NULL, then the standard f_op->open()
1034 * filesystem callback is substituted.
1035 *
1036 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1037 * the return value of d_splice_alias(), then the caller needs to perform dput()
1038 * on it after finish_open().
1039 *
1040 * Returns zero on success or -errno if the open failed.
1041 */
1042int finish_open(struct file *file, struct dentry *dentry,
1043		int (*open)(struct inode *, struct file *))
1044{
1045	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1046
1047	file->f_path.dentry = dentry;
1048	return do_dentry_open(file, d_backing_inode(dentry), open);
1049}
1050EXPORT_SYMBOL(finish_open);
1051
1052/**
1053 * finish_no_open - finish ->atomic_open() without opening the file
1054 *
1055 * @file: file pointer
1056 * @dentry: dentry or NULL (as returned from ->lookup())
1057 *
1058 * This can be used to set the result of a successful lookup in ->atomic_open().
1059 *
1060 * NB: unlike finish_open() this function does consume the dentry reference and
1061 * the caller need not dput() it.
1062 *
1063 * Returns "0" which must be the return value of ->atomic_open() after having
1064 * called this function.
1065 */
1066int finish_no_open(struct file *file, struct dentry *dentry)
1067{
1068	file->f_path.dentry = dentry;
1069	return 0;
1070}
1071EXPORT_SYMBOL(finish_no_open);
1072
1073char *file_path(struct file *filp, char *buf, int buflen)
1074{
1075	return d_path(&filp->f_path, buf, buflen);
1076}
1077EXPORT_SYMBOL(file_path);
1078
1079/**
1080 * vfs_open - open the file at the given path
1081 * @path: path to open
1082 * @file: newly allocated file with f_flag initialized
 
1083 */
1084int vfs_open(const struct path *path, struct file *file)
1085{
1086	file->f_path = *path;
1087	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1088}
1089
1090struct file *dentry_open(const struct path *path, int flags,
1091			 const struct cred *cred)
1092{
1093	int error;
1094	struct file *f;
1095
 
 
1096	/* We must always pass in a valid mount pointer. */
1097	BUG_ON(!path->mnt);
1098
1099	f = alloc_empty_file(flags, cred);
1100	if (!IS_ERR(f)) {
1101		error = vfs_open(path, f);
1102		if (error) {
1103			fput(f);
1104			f = ERR_PTR(error);
1105		}
1106	}
1107	return f;
1108}
1109EXPORT_SYMBOL(dentry_open);
1110
1111/**
1112 * dentry_create - Create and open a file
1113 * @path: path to create
1114 * @flags: O_ flags
1115 * @mode: mode bits for new file
1116 * @cred: credentials to use
1117 *
1118 * Caller must hold the parent directory's lock, and have prepared
1119 * a negative dentry, placed in @path->dentry, for the new file.
1120 *
1121 * Caller sets @path->mnt to the vfsmount of the filesystem where
1122 * the new file is to be created. The parent directory and the
1123 * negative dentry must reside on the same filesystem instance.
1124 *
1125 * On success, returns a "struct file *". Otherwise a ERR_PTR
1126 * is returned.
1127 */
1128struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1129			   const struct cred *cred)
1130{
1131	struct file *f;
1132	int error;
1133
 
1134	f = alloc_empty_file(flags, cred);
1135	if (IS_ERR(f))
1136		return f;
1137
1138	error = vfs_create(mnt_idmap(path->mnt),
1139			   d_inode(path->dentry->d_parent),
1140			   path->dentry, mode, true);
1141	if (!error)
1142		error = vfs_open(path, f);
1143
1144	if (unlikely(error)) {
1145		fput(f);
1146		return ERR_PTR(error);
1147	}
1148	return f;
1149}
1150EXPORT_SYMBOL(dentry_create);
1151
1152/**
1153 * kernel_file_open - open a file for kernel internal use
1154 * @path:	path of the file to open
1155 * @flags:	open flags
1156 * @inode:	the inode
1157 * @cred:	credentials for open
1158 *
1159 * Open a file for use by in-kernel consumers. The file is not accounted
1160 * against nr_files and must not be installed into the file descriptor
1161 * table.
1162 *
1163 * Return: Opened file on success, an error pointer on failure.
1164 */
1165struct file *kernel_file_open(const struct path *path, int flags,
1166				struct inode *inode, const struct cred *cred)
1167{
1168	struct file *f;
1169	int error;
 
1170
1171	f = alloc_empty_file_noaccount(flags, cred);
1172	if (IS_ERR(f))
1173		return f;
1174
1175	f->f_path = *path;
1176	error = do_dentry_open(f, inode, NULL);
1177	if (error) {
1178		fput(f);
1179		f = ERR_PTR(error);
1180	}
1181	return f;
1182}
1183EXPORT_SYMBOL_GPL(kernel_file_open);
1184
1185#define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1186#define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1187
1188inline struct open_how build_open_how(int flags, umode_t mode)
1189{
1190	struct open_how how = {
1191		.flags = flags & VALID_OPEN_FLAGS,
1192		.mode = mode & S_IALLUGO,
1193	};
1194
1195	/* O_PATH beats everything else. */
1196	if (how.flags & O_PATH)
1197		how.flags &= O_PATH_FLAGS;
1198	/* Modes should only be set for create-like flags. */
1199	if (!WILL_CREATE(how.flags))
1200		how.mode = 0;
1201	return how;
1202}
1203
1204inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1205{
1206	u64 flags = how->flags;
1207	u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1208	int lookup_flags = 0;
1209	int acc_mode = ACC_MODE(flags);
1210
1211	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1212			 "struct open_flags doesn't yet handle flags > 32 bits");
1213
1214	/*
1215	 * Strip flags that either shouldn't be set by userspace like
1216	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1217	 * open_flags like O_CLOEXEC.
1218	 */
1219	flags &= ~strip;
1220
1221	/*
1222	 * Older syscalls implicitly clear all of the invalid flags or argument
1223	 * values before calling build_open_flags(), but openat2(2) checks all
1224	 * of its arguments.
1225	 */
1226	if (flags & ~VALID_OPEN_FLAGS)
1227		return -EINVAL;
1228	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1229		return -EINVAL;
1230
1231	/* Scoping flags are mutually exclusive. */
1232	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1233		return -EINVAL;
1234
1235	/* Deal with the mode. */
1236	if (WILL_CREATE(flags)) {
1237		if (how->mode & ~S_IALLUGO)
1238			return -EINVAL;
1239		op->mode = how->mode | S_IFREG;
1240	} else {
1241		if (how->mode != 0)
1242			return -EINVAL;
1243		op->mode = 0;
1244	}
1245
1246	/*
1247	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1248	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1249	 * O_TMPFILE below which requires O_DIRECTORY being raised.
 
1250	 */
1251	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1252		return -EINVAL;
1253
1254	/* Now handle the creative implementation of O_TMPFILE. */
1255	if (flags & __O_TMPFILE) {
1256		/*
1257		 * In order to ensure programs get explicit errors when trying
1258		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1259		 * is raised alongside __O_TMPFILE.
1260		 */
1261		if (!(flags & O_DIRECTORY))
1262			return -EINVAL;
1263		if (!(acc_mode & MAY_WRITE))
1264			return -EINVAL;
1265	}
1266	if (flags & O_PATH) {
1267		/* O_PATH only permits certain other flags to be set. */
1268		if (flags & ~O_PATH_FLAGS)
1269			return -EINVAL;
1270		acc_mode = 0;
1271	}
1272
1273	/*
1274	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1275	 * check for O_DSYNC if the need any syncing at all we enforce it's
1276	 * always set instead of having to deal with possibly weird behaviour
1277	 * for malicious applications setting only __O_SYNC.
1278	 */
1279	if (flags & __O_SYNC)
1280		flags |= O_DSYNC;
1281
1282	op->open_flag = flags;
1283
1284	/* O_TRUNC implies we need access checks for write permissions */
1285	if (flags & O_TRUNC)
1286		acc_mode |= MAY_WRITE;
1287
1288	/* Allow the LSM permission hook to distinguish append
1289	   access from general write access. */
1290	if (flags & O_APPEND)
1291		acc_mode |= MAY_APPEND;
1292
1293	op->acc_mode = acc_mode;
1294
1295	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1296
1297	if (flags & O_CREAT) {
1298		op->intent |= LOOKUP_CREATE;
1299		if (flags & O_EXCL) {
1300			op->intent |= LOOKUP_EXCL;
1301			flags |= O_NOFOLLOW;
1302		}
1303	}
1304
1305	if (flags & O_DIRECTORY)
1306		lookup_flags |= LOOKUP_DIRECTORY;
1307	if (!(flags & O_NOFOLLOW))
1308		lookup_flags |= LOOKUP_FOLLOW;
1309
1310	if (how->resolve & RESOLVE_NO_XDEV)
1311		lookup_flags |= LOOKUP_NO_XDEV;
1312	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1313		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1314	if (how->resolve & RESOLVE_NO_SYMLINKS)
1315		lookup_flags |= LOOKUP_NO_SYMLINKS;
1316	if (how->resolve & RESOLVE_BENEATH)
1317		lookup_flags |= LOOKUP_BENEATH;
1318	if (how->resolve & RESOLVE_IN_ROOT)
1319		lookup_flags |= LOOKUP_IN_ROOT;
1320	if (how->resolve & RESOLVE_CACHED) {
1321		/* Don't bother even trying for create/truncate/tmpfile open */
1322		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1323			return -EAGAIN;
1324		lookup_flags |= LOOKUP_CACHED;
1325	}
1326
1327	op->lookup_flags = lookup_flags;
1328	return 0;
1329}
1330
1331/**
1332 * file_open_name - open file and return file pointer
1333 *
1334 * @name:	struct filename containing path to open
1335 * @flags:	open flags as per the open(2) second argument
1336 * @mode:	mode for the new file if O_CREAT is set, else ignored
1337 *
1338 * This is the helper to open a file from kernelspace if you really
1339 * have to.  But in generally you should not do this, so please move
1340 * along, nothing to see here..
1341 */
1342struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1343{
1344	struct open_flags op;
1345	struct open_how how = build_open_how(flags, mode);
1346	int err = build_open_flags(&how, &op);
1347	if (err)
1348		return ERR_PTR(err);
1349	return do_filp_open(AT_FDCWD, name, &op);
1350}
1351
1352/**
1353 * filp_open - open file and return file pointer
1354 *
1355 * @filename:	path to open
1356 * @flags:	open flags as per the open(2) second argument
1357 * @mode:	mode for the new file if O_CREAT is set, else ignored
1358 *
1359 * This is the helper to open a file from kernelspace if you really
1360 * have to.  But in generally you should not do this, so please move
1361 * along, nothing to see here..
1362 */
1363struct file *filp_open(const char *filename, int flags, umode_t mode)
1364{
1365	struct filename *name = getname_kernel(filename);
1366	struct file *file = ERR_CAST(name);
1367	
1368	if (!IS_ERR(name)) {
1369		file = file_open_name(name, flags, mode);
1370		putname(name);
1371	}
1372	return file;
1373}
1374EXPORT_SYMBOL(filp_open);
1375
1376struct file *file_open_root(const struct path *root,
1377			    const char *filename, int flags, umode_t mode)
1378{
1379	struct open_flags op;
1380	struct open_how how = build_open_how(flags, mode);
1381	int err = build_open_flags(&how, &op);
1382	if (err)
1383		return ERR_PTR(err);
1384	return do_file_open_root(root, filename, &op);
1385}
1386EXPORT_SYMBOL(file_open_root);
1387
1388static long do_sys_openat2(int dfd, const char __user *filename,
1389			   struct open_how *how)
1390{
1391	struct open_flags op;
1392	int fd = build_open_flags(how, &op);
1393	struct filename *tmp;
1394
1395	if (fd)
1396		return fd;
1397
1398	tmp = getname(filename);
1399	if (IS_ERR(tmp))
1400		return PTR_ERR(tmp);
1401
1402	fd = get_unused_fd_flags(how->flags);
1403	if (fd >= 0) {
1404		struct file *f = do_filp_open(dfd, tmp, &op);
1405		if (IS_ERR(f)) {
1406			put_unused_fd(fd);
1407			fd = PTR_ERR(f);
1408		} else {
 
1409			fd_install(fd, f);
1410		}
1411	}
1412	putname(tmp);
1413	return fd;
1414}
1415
1416long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1417{
1418	struct open_how how = build_open_how(flags, mode);
1419	return do_sys_openat2(dfd, filename, &how);
1420}
1421
1422
1423SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1424{
1425	if (force_o_largefile())
1426		flags |= O_LARGEFILE;
1427	return do_sys_open(AT_FDCWD, filename, flags, mode);
1428}
1429
1430SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1431		umode_t, mode)
1432{
1433	if (force_o_largefile())
1434		flags |= O_LARGEFILE;
1435	return do_sys_open(dfd, filename, flags, mode);
1436}
1437
1438SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1439		struct open_how __user *, how, size_t, usize)
1440{
1441	int err;
1442	struct open_how tmp;
1443
1444	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1445	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1446
1447	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1448		return -EINVAL;
1449
1450	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1451	if (err)
1452		return err;
1453
1454	audit_openat2_how(&tmp);
1455
1456	/* O_LARGEFILE is only allowed for non-O_PATH. */
1457	if (!(tmp.flags & O_PATH) && force_o_largefile())
1458		tmp.flags |= O_LARGEFILE;
1459
1460	return do_sys_openat2(dfd, filename, &tmp);
1461}
1462
1463#ifdef CONFIG_COMPAT
1464/*
1465 * Exactly like sys_open(), except that it doesn't set the
1466 * O_LARGEFILE flag.
1467 */
1468COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1469{
1470	return do_sys_open(AT_FDCWD, filename, flags, mode);
1471}
1472
1473/*
1474 * Exactly like sys_openat(), except that it doesn't set the
1475 * O_LARGEFILE flag.
1476 */
1477COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1478{
1479	return do_sys_open(dfd, filename, flags, mode);
1480}
1481#endif
1482
1483#ifndef __alpha__
1484
1485/*
1486 * For backward compatibility?  Maybe this should be moved
1487 * into arch/i386 instead?
1488 */
1489SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1490{
1491	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1492
1493	if (force_o_largefile())
1494		flags |= O_LARGEFILE;
1495	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1496}
1497#endif
1498
1499/*
1500 * "id" is the POSIX thread ID. We use the
1501 * files pointer for this..
1502 */
1503static int filp_flush(struct file *filp, fl_owner_t id)
1504{
1505	int retval = 0;
1506
1507	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1508			"VFS: Close: file count is 0 (f_op=%ps)",
1509			filp->f_op)) {
1510		return 0;
1511	}
1512
1513	if (filp->f_op->flush)
1514		retval = filp->f_op->flush(filp, id);
1515
1516	if (likely(!(filp->f_mode & FMODE_PATH))) {
1517		dnotify_flush(filp, id);
1518		locks_remove_posix(filp, id);
1519	}
 
1520	return retval;
1521}
1522
1523int filp_close(struct file *filp, fl_owner_t id)
1524{
1525	int retval;
1526
1527	retval = filp_flush(filp, id);
1528	fput(filp);
1529
1530	return retval;
1531}
1532EXPORT_SYMBOL(filp_close);
1533
1534/*
1535 * Careful here! We test whether the file pointer is NULL before
1536 * releasing the fd. This ensures that one clone task can't release
1537 * an fd while another clone is opening it.
1538 */
1539SYSCALL_DEFINE1(close, unsigned int, fd)
1540{
1541	int retval;
1542	struct file *file;
1543
1544	file = file_close_fd(fd);
1545	if (!file)
1546		return -EBADF;
1547
1548	retval = filp_flush(file, current->files);
1549
1550	/*
1551	 * We're returning to user space. Don't bother
1552	 * with any delayed fput() cases.
1553	 */
1554	__fput_sync(file);
1555
1556	/* can't restart close syscall because file table entry was cleared */
1557	if (unlikely(retval == -ERESTARTSYS ||
1558		     retval == -ERESTARTNOINTR ||
1559		     retval == -ERESTARTNOHAND ||
1560		     retval == -ERESTART_RESTARTBLOCK))
1561		retval = -EINTR;
1562
1563	return retval;
1564}
1565
1566/**
1567 * sys_close_range() - Close all file descriptors in a given range.
1568 *
1569 * @fd:     starting file descriptor to close
1570 * @max_fd: last file descriptor to close
1571 * @flags:  reserved for future extensions
1572 *
1573 * This closes a range of file descriptors. All file descriptors
1574 * from @fd up to and including @max_fd are closed.
1575 * Currently, errors to close a given file descriptor are ignored.
1576 */
1577SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1578		unsigned int, flags)
1579{
1580	return __close_range(fd, max_fd, flags);
1581}
1582
1583/*
1584 * This routine simulates a hangup on the tty, to arrange that users
1585 * are given clean terminals at login time.
1586 */
1587SYSCALL_DEFINE0(vhangup)
1588{
1589	if (capable(CAP_SYS_TTY_CONFIG)) {
1590		tty_vhangup_self();
1591		return 0;
1592	}
1593	return -EPERM;
1594}
1595
1596/*
1597 * Called when an inode is about to be open.
1598 * We use this to disallow opening large files on 32bit systems if
1599 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1600 * on this flag in sys_open.
1601 */
1602int generic_file_open(struct inode * inode, struct file * filp)
1603{
1604	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1605		return -EOVERFLOW;
1606	return 0;
1607}
1608
1609EXPORT_SYMBOL(generic_file_open);
1610
1611/*
1612 * This is used by subsystems that don't want seekable
1613 * file descriptors. The function is not supposed to ever fail, the only
1614 * reason it returns an 'int' and not 'void' is so that it can be plugged
1615 * directly into file_operations structure.
1616 */
1617int nonseekable_open(struct inode *inode, struct file *filp)
1618{
1619	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1620	return 0;
1621}
1622
1623EXPORT_SYMBOL(nonseekable_open);
1624
1625/*
1626 * stream_open is used by subsystems that want stream-like file descriptors.
1627 * Such file descriptors are not seekable and don't have notion of position
1628 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1629 * Contrary to file descriptors of other regular files, .read() and .write()
1630 * can run simultaneously.
1631 *
1632 * stream_open never fails and is marked to return int so that it could be
1633 * directly used as file_operations.open .
1634 */
1635int stream_open(struct inode *inode, struct file *filp)
1636{
1637	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1638	filp->f_mode |= FMODE_STREAM;
1639	return 0;
1640}
1641
1642EXPORT_SYMBOL(stream_open);