Linux Audio

Check our new training course

Buildroot integration, development and maintenance

Need a Buildroot system for your embedded project?
Loading...
v4.17
 
   1/*
   2 *  linux/fs/open.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/string.h>
   8#include <linux/mm.h>
   9#include <linux/file.h>
  10#include <linux/fdtable.h>
  11#include <linux/fsnotify.h>
  12#include <linux/module.h>
  13#include <linux/tty.h>
  14#include <linux/namei.h>
  15#include <linux/backing-dev.h>
  16#include <linux/capability.h>
  17#include <linux/securebits.h>
  18#include <linux/security.h>
  19#include <linux/mount.h>
  20#include <linux/fcntl.h>
  21#include <linux/slab.h>
  22#include <linux/uaccess.h>
  23#include <linux/fs.h>
  24#include <linux/personality.h>
  25#include <linux/pagemap.h>
  26#include <linux/syscalls.h>
  27#include <linux/rcupdate.h>
  28#include <linux/audit.h>
  29#include <linux/falloc.h>
  30#include <linux/fs_struct.h>
  31#include <linux/ima.h>
  32#include <linux/dnotify.h>
  33#include <linux/compat.h>
 
 
  34
  35#include "internal.h"
  36
  37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38	struct file *filp)
  39{
  40	int ret;
  41	struct iattr newattrs;
  42
  43	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44	if (length < 0)
  45		return -EINVAL;
  46
  47	newattrs.ia_size = length;
  48	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49	if (filp) {
  50		newattrs.ia_file = filp;
  51		newattrs.ia_valid |= ATTR_FILE;
  52	}
  53
  54	/* Remove suid, sgid, and file capabilities on truncate too */
  55	ret = dentry_needs_remove_privs(dentry);
  56	if (ret < 0)
  57		return ret;
  58	if (ret)
  59		newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61	inode_lock(dentry->d_inode);
  62	/* Note any delegations or leases have already been broken: */
  63	ret = notify_change(dentry, &newattrs, NULL);
  64	inode_unlock(dentry->d_inode);
  65	return ret;
  66}
  67
  68long vfs_truncate(const struct path *path, loff_t length)
  69{
 
  70	struct inode *inode;
  71	struct dentry *upperdentry;
  72	long error;
  73
  74	inode = path->dentry->d_inode;
  75
  76	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77	if (S_ISDIR(inode->i_mode))
  78		return -EISDIR;
  79	if (!S_ISREG(inode->i_mode))
  80		return -EINVAL;
  81
  82	error = mnt_want_write(path->mnt);
  83	if (error)
  84		goto out;
  85
  86	error = inode_permission(inode, MAY_WRITE);
 
  87	if (error)
  88		goto mnt_drop_write_and_out;
  89
  90	error = -EPERM;
  91	if (IS_APPEND(inode))
  92		goto mnt_drop_write_and_out;
  93
  94	/*
  95	 * If this is an overlayfs then do as if opening the file so we get
  96	 * write access on the upper inode, not on the overlay inode.  For
  97	 * non-overlay filesystems d_real() is an identity function.
  98	 */
  99	upperdentry = d_real(path->dentry, NULL, O_WRONLY, 0);
 100	error = PTR_ERR(upperdentry);
 101	if (IS_ERR(upperdentry))
 102		goto mnt_drop_write_and_out;
 103
 104	error = get_write_access(upperdentry->d_inode);
 105	if (error)
 106		goto mnt_drop_write_and_out;
 107
 108	/*
 109	 * Make sure that there are no leases.  get_write_access() protects
 110	 * against the truncate racing with a lease-granting setlease().
 111	 */
 112	error = break_lease(inode, O_WRONLY);
 113	if (error)
 114		goto put_write_and_out;
 115
 116	error = locks_verify_truncate(inode, NULL, length);
 117	if (!error)
 118		error = security_path_truncate(path);
 119	if (!error)
 120		error = do_truncate(path->dentry, length, 0, NULL);
 121
 122put_write_and_out:
 123	put_write_access(upperdentry->d_inode);
 124mnt_drop_write_and_out:
 125	mnt_drop_write(path->mnt);
 126out:
 127	return error;
 128}
 129EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131long do_sys_truncate(const char __user *pathname, loff_t length)
 132{
 133	unsigned int lookup_flags = LOOKUP_FOLLOW;
 134	struct path path;
 135	int error;
 136
 137	if (length < 0)	/* sorry, but loff_t says... */
 138		return -EINVAL;
 139
 140retry:
 141	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142	if (!error) {
 143		error = vfs_truncate(&path, length);
 144		path_put(&path);
 145	}
 146	if (retry_estale(error, lookup_flags)) {
 147		lookup_flags |= LOOKUP_REVAL;
 148		goto retry;
 149	}
 150	return error;
 151}
 152
 153SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154{
 155	return do_sys_truncate(path, length);
 156}
 157
 158#ifdef CONFIG_COMPAT
 159COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160{
 161	return do_sys_truncate(path, length);
 162}
 163#endif
 164
 165long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166{
 167	struct inode *inode;
 168	struct dentry *dentry;
 169	struct fd f;
 170	int error;
 171
 172	error = -EINVAL;
 173	if (length < 0)
 174		goto out;
 175	error = -EBADF;
 176	f = fdget(fd);
 177	if (!f.file)
 178		goto out;
 179
 180	/* explicitly opened as large or we are on 64-bit box */
 181	if (f.file->f_flags & O_LARGEFILE)
 182		small = 0;
 183
 184	dentry = f.file->f_path.dentry;
 185	inode = dentry->d_inode;
 186	error = -EINVAL;
 187	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188		goto out_putf;
 189
 190	error = -EINVAL;
 191	/* Cannot ftruncate over 2^31 bytes without large file support */
 192	if (small && length > MAX_NON_LFS)
 193		goto out_putf;
 194
 195	error = -EPERM;
 196	/* Check IS_APPEND on real upper inode */
 197	if (IS_APPEND(file_inode(f.file)))
 198		goto out_putf;
 199
 200	sb_start_write(inode->i_sb);
 201	error = locks_verify_truncate(inode, f.file, length);
 202	if (!error)
 203		error = security_path_truncate(&f.file->f_path);
 204	if (!error)
 205		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 
 206	sb_end_write(inode->i_sb);
 207out_putf:
 208	fdput(f);
 209out:
 210	return error;
 211}
 212
 213SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 
 
 
 
 
 
 
 
 
 
 
 214{
 215	return do_sys_ftruncate(fd, length, 1);
 216}
 217
 218#ifdef CONFIG_COMPAT
 219COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 220{
 221	return do_sys_ftruncate(fd, length, 1);
 222}
 223#endif
 224
 225/* LFS versions of truncate are only needed on 32 bit machines */
 226#if BITS_PER_LONG == 32
 227SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 228{
 229	return do_sys_truncate(path, length);
 230}
 231
 232SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 233{
 234	return do_sys_ftruncate(fd, length, 0);
 235}
 236#endif /* BITS_PER_LONG == 32 */
 237
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 238
 239int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 240{
 241	struct inode *inode = file_inode(file);
 242	long ret;
 
 243
 244	if (offset < 0 || len <= 0)
 245		return -EINVAL;
 246
 247	/* Return error if mode is not supported */
 248	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 249		return -EOPNOTSUPP;
 250
 251	/* Punch hole and zero range are mutually exclusive */
 252	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 253	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 254		return -EOPNOTSUPP;
 255
 256	/* Punch hole must have keep size set */
 257	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 258	    !(mode & FALLOC_FL_KEEP_SIZE))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 259		return -EOPNOTSUPP;
 260
 261	/* Collapse range should only be used exclusively. */
 262	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 263	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 264		return -EINVAL;
 265
 266	/* Insert range should only be used exclusively. */
 267	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 268	    (mode & ~FALLOC_FL_INSERT_RANGE))
 269		return -EINVAL;
 270
 271	/* Unshare range should only be used with allocate mode. */
 272	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 273	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 274		return -EINVAL;
 275
 276	if (!(file->f_mode & FMODE_WRITE))
 277		return -EBADF;
 278
 279	/*
 280	 * We can only allow pure fallocate on append only files
 281	 */
 282	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 283		return -EPERM;
 284
 285	if (IS_IMMUTABLE(inode))
 286		return -EPERM;
 287
 288	/*
 289	 * We cannot allow any fallocate operation on an active swapfile
 290	 */
 291	if (IS_SWAPFILE(inode))
 292		return -ETXTBSY;
 293
 294	/*
 295	 * Revalidate the write permissions, in case security policy has
 296	 * changed since the files were opened.
 297	 */
 298	ret = security_file_permission(file, MAY_WRITE);
 299	if (ret)
 300		return ret;
 301
 
 
 
 
 302	if (S_ISFIFO(inode->i_mode))
 303		return -ESPIPE;
 304
 305	if (S_ISDIR(inode->i_mode))
 306		return -EISDIR;
 307
 308	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 309		return -ENODEV;
 310
 311	/* Check for wrap through zero too */
 312	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 
 
 
 313		return -EFBIG;
 314
 315	if (!file->f_op->fallocate)
 316		return -EOPNOTSUPP;
 317
 318	file_start_write(file);
 319	ret = file->f_op->fallocate(file, mode, offset, len);
 320
 321	/*
 322	 * Create inotify and fanotify events.
 323	 *
 324	 * To keep the logic simple always create events if fallocate succeeds.
 325	 * This implies that events are even created if the file size remains
 326	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 327	 */
 328	if (ret == 0)
 329		fsnotify_modify(file);
 330
 331	file_end_write(file);
 332	return ret;
 333}
 334EXPORT_SYMBOL_GPL(vfs_fallocate);
 335
 336int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 337{
 338	struct fd f = fdget(fd);
 339	int error = -EBADF;
 340
 341	if (f.file) {
 342		error = vfs_fallocate(f.file, mode, offset, len);
 343		fdput(f);
 344	}
 345	return error;
 346}
 347
 348SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 349{
 350	return ksys_fallocate(fd, mode, offset, len);
 351}
 352
 
 
 
 
 
 
 
 
 
 353/*
 354 * access() needs to use the real uid/gid, not the effective uid/gid.
 355 * We do this by temporarily clearing all FS-related capabilities and
 356 * switching the fsuid/fsgid around to the real ones.
 
 
 
 357 */
 358long do_faccessat(int dfd, const char __user *filename, int mode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 359{
 360	const struct cred *old_cred;
 361	struct cred *override_cred;
 362	struct path path;
 363	struct inode *inode;
 364	int res;
 365	unsigned int lookup_flags = LOOKUP_FOLLOW;
 366
 367	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 368		return -EINVAL;
 369
 370	override_cred = prepare_creds();
 371	if (!override_cred)
 372		return -ENOMEM;
 
 
 
 
 
 
 373
 374	override_cred->fsuid = override_cred->uid;
 375	override_cred->fsgid = override_cred->gid;
 376
 377	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 378		/* Clear the capabilities if we switch to a non-root user */
 379		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 380		if (!uid_eq(override_cred->uid, root_uid))
 381			cap_clear(override_cred->cap_effective);
 382		else
 383			override_cred->cap_effective =
 384				override_cred->cap_permitted;
 385	}
 386
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 387	old_cred = override_creds(override_cred);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 388retry:
 389	res = user_path_at(dfd, filename, lookup_flags, &path);
 390	if (res)
 391		goto out;
 392
 393	inode = d_backing_inode(path.dentry);
 394
 395	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 396		/*
 397		 * MAY_EXEC on regular files is denied if the fs is mounted
 398		 * with the "noexec" flag.
 399		 */
 400		res = -EACCES;
 401		if (path_noexec(&path))
 402			goto out_path_release;
 403	}
 404
 405	res = inode_permission(inode, mode | MAY_ACCESS);
 406	/* SuS v2 requires we report a read only fs too */
 407	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 408		goto out_path_release;
 409	/*
 410	 * This is a rare case where using __mnt_is_readonly()
 411	 * is OK without a mnt_want/drop_write() pair.  Since
 412	 * no actual write to the fs is performed here, we do
 413	 * not need to telegraph to that to anyone.
 414	 *
 415	 * By doing this, we accept that this access is
 416	 * inherently racy and know that the fs may change
 417	 * state before we even see this result.
 418	 */
 419	if (__mnt_is_readonly(path.mnt))
 420		res = -EROFS;
 421
 422out_path_release:
 423	path_put(&path);
 424	if (retry_estale(res, lookup_flags)) {
 425		lookup_flags |= LOOKUP_REVAL;
 426		goto retry;
 427	}
 428out:
 429	revert_creds(old_cred);
 430	put_cred(override_cred);
 
 431	return res;
 432}
 433
 434SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 435{
 436	return do_faccessat(dfd, filename, mode);
 
 
 
 
 
 
 437}
 438
 439SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 440{
 441	return do_faccessat(AT_FDCWD, filename, mode);
 442}
 443
 444int ksys_chdir(const char __user *filename)
 445{
 446	struct path path;
 447	int error;
 448	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 449retry:
 450	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 451	if (error)
 452		goto out;
 453
 454	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 455	if (error)
 456		goto dput_and_out;
 457
 458	set_fs_pwd(current->fs, &path);
 459
 460dput_and_out:
 461	path_put(&path);
 462	if (retry_estale(error, lookup_flags)) {
 463		lookup_flags |= LOOKUP_REVAL;
 464		goto retry;
 465	}
 466out:
 467	return error;
 468}
 469
 470SYSCALL_DEFINE1(chdir, const char __user *, filename)
 471{
 472	return ksys_chdir(filename);
 473}
 474
 475SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 476{
 477	struct fd f = fdget_raw(fd);
 478	int error;
 479
 480	error = -EBADF;
 481	if (!f.file)
 482		goto out;
 483
 484	error = -ENOTDIR;
 485	if (!d_can_lookup(f.file->f_path.dentry))
 486		goto out_putf;
 487
 488	error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 489	if (!error)
 490		set_fs_pwd(current->fs, &f.file->f_path);
 491out_putf:
 492	fdput(f);
 493out:
 494	return error;
 495}
 496
 497int ksys_chroot(const char __user *filename)
 498{
 499	struct path path;
 500	int error;
 501	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 502retry:
 503	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 504	if (error)
 505		goto out;
 506
 507	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 508	if (error)
 509		goto dput_and_out;
 510
 511	error = -EPERM;
 512	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 513		goto dput_and_out;
 514	error = security_path_chroot(&path);
 515	if (error)
 516		goto dput_and_out;
 517
 518	set_fs_root(current->fs, &path);
 519	error = 0;
 520dput_and_out:
 521	path_put(&path);
 522	if (retry_estale(error, lookup_flags)) {
 523		lookup_flags |= LOOKUP_REVAL;
 524		goto retry;
 525	}
 526out:
 527	return error;
 528}
 529
 530SYSCALL_DEFINE1(chroot, const char __user *, filename)
 531{
 532	return ksys_chroot(filename);
 533}
 534
 535static int chmod_common(const struct path *path, umode_t mode)
 536{
 537	struct inode *inode = path->dentry->d_inode;
 538	struct inode *delegated_inode = NULL;
 539	struct iattr newattrs;
 540	int error;
 541
 542	error = mnt_want_write(path->mnt);
 543	if (error)
 544		return error;
 545retry_deleg:
 546	inode_lock(inode);
 547	error = security_path_chmod(path, mode);
 548	if (error)
 549		goto out_unlock;
 550	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 551	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 552	error = notify_change(path->dentry, &newattrs, &delegated_inode);
 
 553out_unlock:
 554	inode_unlock(inode);
 555	if (delegated_inode) {
 556		error = break_deleg_wait(&delegated_inode);
 557		if (!error)
 558			goto retry_deleg;
 559	}
 560	mnt_drop_write(path->mnt);
 561	return error;
 562}
 563
 564int ksys_fchmod(unsigned int fd, umode_t mode)
 565{
 566	struct fd f = fdget(fd);
 567	int err = -EBADF;
 568
 569	if (f.file) {
 570		audit_file(f.file);
 571		err = chmod_common(&f.file->f_path, mode);
 572		fdput(f);
 573	}
 574	return err;
 575}
 576
 577SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 578{
 579	return ksys_fchmod(fd, mode);
 
 
 
 
 
 580}
 581
 582int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 
 583{
 584	struct path path;
 585	int error;
 586	unsigned int lookup_flags = LOOKUP_FOLLOW;
 
 
 
 
 
 
 
 
 587retry:
 588	error = user_path_at(dfd, filename, lookup_flags, &path);
 589	if (!error) {
 590		error = chmod_common(&path, mode);
 591		path_put(&path);
 592		if (retry_estale(error, lookup_flags)) {
 593			lookup_flags |= LOOKUP_REVAL;
 594			goto retry;
 595		}
 596	}
 597	return error;
 598}
 599
 
 
 
 
 
 
 600SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 601		umode_t, mode)
 602{
 603	return do_fchmodat(dfd, filename, mode);
 604}
 605
 606SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 607{
 608	return do_fchmodat(AT_FDCWD, filename, mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 609}
 610
 611static int chown_common(const struct path *path, uid_t user, gid_t group)
 
 
 
 
 
 
 612{
 
 
 
 
 
 
 
 
 
 
 
 613	struct inode *inode = path->dentry->d_inode;
 614	struct inode *delegated_inode = NULL;
 615	int error;
 616	struct iattr newattrs;
 617	kuid_t uid;
 618	kgid_t gid;
 619
 620	uid = make_kuid(current_user_ns(), user);
 621	gid = make_kgid(current_user_ns(), group);
 622
 
 
 
 623retry_deleg:
 
 
 624	newattrs.ia_valid =  ATTR_CTIME;
 625	if (user != (uid_t) -1) {
 626		if (!uid_valid(uid))
 627			return -EINVAL;
 628		newattrs.ia_valid |= ATTR_UID;
 629		newattrs.ia_uid = uid;
 630	}
 631	if (group != (gid_t) -1) {
 632		if (!gid_valid(gid))
 633			return -EINVAL;
 634		newattrs.ia_valid |= ATTR_GID;
 635		newattrs.ia_gid = gid;
 636	}
 637	if (!S_ISDIR(inode->i_mode))
 638		newattrs.ia_valid |=
 639			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 640	inode_lock(inode);
 641	error = security_path_chown(path, uid, gid);
 
 
 
 
 
 
 
 642	if (!error)
 643		error = notify_change(path->dentry, &newattrs, &delegated_inode);
 
 644	inode_unlock(inode);
 645	if (delegated_inode) {
 646		error = break_deleg_wait(&delegated_inode);
 647		if (!error)
 648			goto retry_deleg;
 649	}
 650	return error;
 651}
 652
 653int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 654		int flag)
 655{
 656	struct path path;
 657	int error = -EINVAL;
 658	int lookup_flags;
 659
 660	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 661		goto out;
 662
 663	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 664	if (flag & AT_EMPTY_PATH)
 665		lookup_flags |= LOOKUP_EMPTY;
 666retry:
 667	error = user_path_at(dfd, filename, lookup_flags, &path);
 668	if (error)
 669		goto out;
 670	error = mnt_want_write(path.mnt);
 671	if (error)
 672		goto out_release;
 673	error = chown_common(&path, user, group);
 674	mnt_drop_write(path.mnt);
 675out_release:
 676	path_put(&path);
 677	if (retry_estale(error, lookup_flags)) {
 678		lookup_flags |= LOOKUP_REVAL;
 679		goto retry;
 680	}
 681out:
 682	return error;
 683}
 684
 685SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 686		gid_t, group, int, flag)
 687{
 688	return do_fchownat(dfd, filename, user, group, flag);
 689}
 690
 691SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 692{
 693	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 694}
 695
 696SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 697{
 698	return do_fchownat(AT_FDCWD, filename, user, group,
 699			   AT_SYMLINK_NOFOLLOW);
 700}
 701
 702int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 703{
 704	struct fd f = fdget(fd);
 705	int error = -EBADF;
 706
 707	if (!f.file)
 708		goto out;
 709
 710	error = mnt_want_write_file_path(f.file);
 711	if (error)
 712		goto out_fput;
 713	audit_file(f.file);
 714	error = chown_common(&f.file->f_path, user, group);
 715	mnt_drop_write_file_path(f.file);
 716out_fput:
 717	fdput(f);
 718out:
 719	return error;
 720}
 721
 
 
 
 
 
 
 
 
 
 
 722SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 723{
 724	return ksys_fchown(fd, user, group);
 725}
 726
 727int open_check_o_direct(struct file *f)
 728{
 729	/* NB: we're sure to have correct a_ops only after f_op->open */
 730	if (f->f_flags & O_DIRECT) {
 731		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 732			return -EINVAL;
 
 
 
 
 
 
 
 
 733	}
 734	return 0;
 
 
 
 
 
 
 735}
 736
 737static int do_dentry_open(struct file *f,
 738			  struct inode *inode,
 739			  int (*open)(struct inode *, struct file *),
 740			  const struct cred *cred)
 741{
 742	static const struct file_operations empty_fops = {};
 
 743	int error;
 744
 745	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 746				FMODE_PREAD | FMODE_PWRITE;
 747
 748	path_get(&f->f_path);
 749	f->f_inode = inode;
 750	f->f_mapping = inode->i_mapping;
 751
 752	/* Ensure that we skip any errors that predate opening of the file */
 753	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 
 754
 755	if (unlikely(f->f_flags & O_PATH)) {
 756		f->f_mode = FMODE_PATH;
 757		f->f_op = &empty_fops;
 758		return 0;
 759	}
 760
 761	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 762		error = get_write_access(inode);
 
 
 763		if (unlikely(error))
 764			goto cleanup_file;
 765		error = __mnt_want_write(f->f_path.mnt);
 766		if (unlikely(error)) {
 767			put_write_access(inode);
 768			goto cleanup_file;
 769		}
 770		f->f_mode |= FMODE_WRITER;
 771	}
 772
 773	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 774	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 775		f->f_mode |= FMODE_ATOMIC_POS;
 776
 777	f->f_op = fops_get(inode->i_fop);
 778	if (unlikely(WARN_ON(!f->f_op))) {
 779		error = -ENODEV;
 780		goto cleanup_all;
 781	}
 782
 783	error = security_file_open(f, cred);
 
 
 
 
 784	if (error)
 785		goto cleanup_all;
 786
 787	error = break_lease(locks_inode(f), f->f_flags);
 788	if (error)
 789		goto cleanup_all;
 790
 
 
 791	if (!open)
 792		open = f->f_op->open;
 793	if (open) {
 794		error = open(inode, f);
 795		if (error)
 796			goto cleanup_all;
 797	}
 798	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 799		i_readcount_inc(inode);
 800	if ((f->f_mode & FMODE_READ) &&
 801	     likely(f->f_op->read || f->f_op->read_iter))
 802		f->f_mode |= FMODE_CAN_READ;
 803	if ((f->f_mode & FMODE_WRITE) &&
 804	     likely(f->f_op->write || f->f_op->write_iter))
 805		f->f_mode |= FMODE_CAN_WRITE;
 
 
 
 
 806
 807	f->f_write_hint = WRITE_LIFE_NOT_SET;
 808	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 
 809
 810	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 811
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 812	return 0;
 813
 814cleanup_all:
 
 
 815	fops_put(f->f_op);
 816	if (f->f_mode & FMODE_WRITER) {
 817		put_write_access(inode);
 818		__mnt_drop_write(f->f_path.mnt);
 819	}
 820cleanup_file:
 821	path_put(&f->f_path);
 822	f->f_path.mnt = NULL;
 823	f->f_path.dentry = NULL;
 824	f->f_inode = NULL;
 825	return error;
 826}
 827
 828/**
 829 * finish_open - finish opening a file
 830 * @file: file pointer
 831 * @dentry: pointer to dentry
 832 * @open: open callback
 833 * @opened: state of open
 834 *
 835 * This can be used to finish opening a file passed to i_op->atomic_open().
 836 *
 837 * If the open callback is set to NULL, then the standard f_op->open()
 838 * filesystem callback is substituted.
 839 *
 840 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 841 * the return value of d_splice_alias(), then the caller needs to perform dput()
 842 * on it after finish_open().
 843 *
 844 * On successful return @file is a fully instantiated open file.  After this, if
 845 * an error occurs in ->atomic_open(), it needs to clean up with fput().
 846 *
 847 * Returns zero on success or -errno if the open failed.
 848 */
 849int finish_open(struct file *file, struct dentry *dentry,
 850		int (*open)(struct inode *, struct file *),
 851		int *opened)
 852{
 853	int error;
 854	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 855
 856	file->f_path.dentry = dentry;
 857	error = do_dentry_open(file, d_backing_inode(dentry), open,
 858			       current_cred());
 859	if (!error)
 860		*opened |= FILE_OPENED;
 861
 862	return error;
 863}
 864EXPORT_SYMBOL(finish_open);
 865
 866/**
 867 * finish_no_open - finish ->atomic_open() without opening the file
 868 *
 869 * @file: file pointer
 870 * @dentry: dentry or NULL (as returned from ->lookup())
 871 *
 872 * This can be used to set the result of a successful lookup in ->atomic_open().
 873 *
 874 * NB: unlike finish_open() this function does consume the dentry reference and
 875 * the caller need not dput() it.
 876 *
 877 * Returns "1" which must be the return value of ->atomic_open() after having
 878 * called this function.
 879 */
 880int finish_no_open(struct file *file, struct dentry *dentry)
 881{
 882	file->f_path.dentry = dentry;
 883	return 1;
 884}
 885EXPORT_SYMBOL(finish_no_open);
 886
 887char *file_path(struct file *filp, char *buf, int buflen)
 888{
 889	return d_path(&filp->f_path, buf, buflen);
 890}
 891EXPORT_SYMBOL(file_path);
 892
 893/**
 894 * vfs_open - open the file at the given path
 895 * @path: path to open
 896 * @file: newly allocated file with f_flag initialized
 897 * @cred: credentials to use
 898 */
 899int vfs_open(const struct path *path, struct file *file,
 900	     const struct cred *cred)
 901{
 902	struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 903
 904	if (IS_ERR(dentry))
 905		return PTR_ERR(dentry);
 906
 907	file->f_path = *path;
 908	return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 
 
 
 
 
 
 
 
 
 909}
 910
 911struct file *dentry_open(const struct path *path, int flags,
 912			 const struct cred *cred)
 913{
 914	int error;
 915	struct file *f;
 916
 917	validate_creds(cred);
 918
 919	/* We must always pass in a valid mount pointer. */
 920	BUG_ON(!path->mnt);
 921
 922	f = get_empty_filp();
 923	if (!IS_ERR(f)) {
 924		f->f_flags = flags;
 925		error = vfs_open(path, f, cred);
 926		if (!error) {
 927			/* from now on we need fput() to dispose of f */
 928			error = open_check_o_direct(f);
 929			if (error) {
 930				fput(f);
 931				f = ERR_PTR(error);
 932			}
 933		} else { 
 934			put_filp(f);
 935			f = ERR_PTR(error);
 936		}
 937	}
 938	return f;
 939}
 940EXPORT_SYMBOL(dentry_open);
 941
 942static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 943{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 944	int lookup_flags = 0;
 945	int acc_mode = ACC_MODE(flags);
 946
 
 
 
 947	/*
 948	 * Clear out all open flags we don't know about so that we don't report
 949	 * them in fcntl(F_GETFD) or similar interfaces.
 
 950	 */
 951	flags &= VALID_OPEN_FLAGS;
 952
 953	if (flags & (O_CREAT | __O_TMPFILE))
 954		op->mode = (mode & S_IALLUGO) | S_IFREG;
 955	else
 956		op->mode = 0;
 
 
 
 
 
 
 
 
 
 957
 958	/* Must never be set by userspace */
 959	flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 
 
 
 
 
 
 
 
 960
 961	/*
 962	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 963	 * check for O_DSYNC if the need any syncing at all we enforce it's
 964	 * always set instead of having to deal with possibly weird behaviour
 965	 * for malicious applications setting only __O_SYNC.
 966	 */
 967	if (flags & __O_SYNC)
 968		flags |= O_DSYNC;
 969
 
 970	if (flags & __O_TMPFILE) {
 971		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 
 
 
 
 
 972			return -EINVAL;
 973		if (!(acc_mode & MAY_WRITE))
 974			return -EINVAL;
 975	} else if (flags & O_PATH) {
 976		/*
 977		 * If we have O_PATH in the open flag. Then we
 978		 * cannot have anything other than the below set of flags
 979		 */
 980		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 981		acc_mode = 0;
 982	}
 983
 
 
 
 
 
 
 
 
 
 984	op->open_flag = flags;
 985
 986	/* O_TRUNC implies we need access checks for write permissions */
 987	if (flags & O_TRUNC)
 988		acc_mode |= MAY_WRITE;
 989
 990	/* Allow the LSM permission hook to distinguish append
 991	   access from general write access. */
 992	if (flags & O_APPEND)
 993		acc_mode |= MAY_APPEND;
 994
 995	op->acc_mode = acc_mode;
 996
 997	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 998
 999	if (flags & O_CREAT) {
1000		op->intent |= LOOKUP_CREATE;
1001		if (flags & O_EXCL)
1002			op->intent |= LOOKUP_EXCL;
 
 
1003	}
1004
1005	if (flags & O_DIRECTORY)
1006		lookup_flags |= LOOKUP_DIRECTORY;
1007	if (!(flags & O_NOFOLLOW))
1008		lookup_flags |= LOOKUP_FOLLOW;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1009	op->lookup_flags = lookup_flags;
1010	return 0;
1011}
1012
1013/**
1014 * file_open_name - open file and return file pointer
1015 *
1016 * @name:	struct filename containing path to open
1017 * @flags:	open flags as per the open(2) second argument
1018 * @mode:	mode for the new file if O_CREAT is set, else ignored
1019 *
1020 * This is the helper to open a file from kernelspace if you really
1021 * have to.  But in generally you should not do this, so please move
1022 * along, nothing to see here..
1023 */
1024struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1025{
1026	struct open_flags op;
1027	int err = build_open_flags(flags, mode, &op);
1028	return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 
 
 
1029}
1030
1031/**
1032 * filp_open - open file and return file pointer
1033 *
1034 * @filename:	path to open
1035 * @flags:	open flags as per the open(2) second argument
1036 * @mode:	mode for the new file if O_CREAT is set, else ignored
1037 *
1038 * This is the helper to open a file from kernelspace if you really
1039 * have to.  But in generally you should not do this, so please move
1040 * along, nothing to see here..
1041 */
1042struct file *filp_open(const char *filename, int flags, umode_t mode)
1043{
1044	struct filename *name = getname_kernel(filename);
1045	struct file *file = ERR_CAST(name);
1046	
1047	if (!IS_ERR(name)) {
1048		file = file_open_name(name, flags, mode);
1049		putname(name);
1050	}
1051	return file;
1052}
1053EXPORT_SYMBOL(filp_open);
1054
1055struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1056			    const char *filename, int flags, umode_t mode)
1057{
1058	struct open_flags op;
1059	int err = build_open_flags(flags, mode, &op);
 
1060	if (err)
1061		return ERR_PTR(err);
1062	return do_file_open_root(dentry, mnt, filename, &op);
1063}
1064EXPORT_SYMBOL(file_open_root);
1065
1066struct file *filp_clone_open(struct file *oldfile)
1067{
1068	struct file *file;
1069	int retval;
1070
1071	file = get_empty_filp();
1072	if (IS_ERR(file))
1073		return file;
1074
1075	file->f_flags = oldfile->f_flags;
1076	retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1077	if (retval) {
1078		put_filp(file);
1079		return ERR_PTR(retval);
1080	}
1081
1082	return file;
1083}
1084EXPORT_SYMBOL(filp_clone_open);
1085
1086long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1087{
1088	struct open_flags op;
1089	int fd = build_open_flags(flags, mode, &op);
1090	struct filename *tmp;
1091
1092	if (fd)
1093		return fd;
1094
1095	tmp = getname(filename);
1096	if (IS_ERR(tmp))
1097		return PTR_ERR(tmp);
1098
1099	fd = get_unused_fd_flags(flags);
1100	if (fd >= 0) {
1101		struct file *f = do_filp_open(dfd, tmp, &op);
1102		if (IS_ERR(f)) {
1103			put_unused_fd(fd);
1104			fd = PTR_ERR(f);
1105		} else {
1106			fsnotify_open(f);
1107			fd_install(fd, f);
1108		}
1109	}
1110	putname(tmp);
1111	return fd;
1112}
1113
 
 
 
 
 
 
 
1114SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1115{
1116	if (force_o_largefile())
1117		flags |= O_LARGEFILE;
1118
1119	return do_sys_open(AT_FDCWD, filename, flags, mode);
1120}
1121
1122SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1123		umode_t, mode)
1124{
1125	if (force_o_largefile())
1126		flags |= O_LARGEFILE;
1127
1128	return do_sys_open(dfd, filename, flags, mode);
1129}
1130
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1131#ifdef CONFIG_COMPAT
1132/*
1133 * Exactly like sys_open(), except that it doesn't set the
1134 * O_LARGEFILE flag.
1135 */
1136COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1137{
1138	return do_sys_open(AT_FDCWD, filename, flags, mode);
1139}
1140
1141/*
1142 * Exactly like sys_openat(), except that it doesn't set the
1143 * O_LARGEFILE flag.
1144 */
1145COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1146{
1147	return do_sys_open(dfd, filename, flags, mode);
1148}
1149#endif
1150
1151#ifndef __alpha__
1152
1153/*
1154 * For backward compatibility?  Maybe this should be moved
1155 * into arch/i386 instead?
1156 */
1157SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1158{
1159	return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1160}
1161
 
 
 
 
1162#endif
1163
1164/*
1165 * "id" is the POSIX thread ID. We use the
1166 * files pointer for this..
1167 */
1168int filp_close(struct file *filp, fl_owner_t id)
1169{
1170	int retval = 0;
1171
1172	if (!file_count(filp)) {
1173		printk(KERN_ERR "VFS: Close: file count is 0\n");
 
1174		return 0;
1175	}
1176
1177	if (filp->f_op->flush)
1178		retval = filp->f_op->flush(filp, id);
1179
1180	if (likely(!(filp->f_mode & FMODE_PATH))) {
1181		dnotify_flush(filp, id);
1182		locks_remove_posix(filp, id);
1183	}
1184	fput(filp);
1185	return retval;
1186}
1187
 
 
 
 
 
 
 
 
 
1188EXPORT_SYMBOL(filp_close);
1189
1190/*
1191 * Careful here! We test whether the file pointer is NULL before
1192 * releasing the fd. This ensures that one clone task can't release
1193 * an fd while another clone is opening it.
1194 */
1195SYSCALL_DEFINE1(close, unsigned int, fd)
1196{
1197	int retval = __close_fd(current->files, fd);
 
 
 
 
 
 
 
 
 
 
 
 
 
1198
1199	/* can't restart close syscall because file table entry was cleared */
1200	if (unlikely(retval == -ERESTARTSYS ||
1201		     retval == -ERESTARTNOINTR ||
1202		     retval == -ERESTARTNOHAND ||
1203		     retval == -ERESTART_RESTARTBLOCK))
1204		retval = -EINTR;
1205
1206	return retval;
1207}
1208
1209/*
1210 * This routine simulates a hangup on the tty, to arrange that users
1211 * are given clean terminals at login time.
1212 */
1213SYSCALL_DEFINE0(vhangup)
1214{
1215	if (capable(CAP_SYS_TTY_CONFIG)) {
1216		tty_vhangup_self();
1217		return 0;
1218	}
1219	return -EPERM;
1220}
1221
1222/*
1223 * Called when an inode is about to be open.
1224 * We use this to disallow opening large files on 32bit systems if
1225 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1226 * on this flag in sys_open.
1227 */
1228int generic_file_open(struct inode * inode, struct file * filp)
1229{
1230	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1231		return -EOVERFLOW;
1232	return 0;
1233}
1234
1235EXPORT_SYMBOL(generic_file_open);
1236
1237/*
1238 * This is used by subsystems that don't want seekable
1239 * file descriptors. The function is not supposed to ever fail, the only
1240 * reason it returns an 'int' and not 'void' is so that it can be plugged
1241 * directly into file_operations structure.
1242 */
1243int nonseekable_open(struct inode *inode, struct file *filp)
1244{
1245	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1246	return 0;
1247}
1248
1249EXPORT_SYMBOL(nonseekable_open);
v6.13.7
   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/dnotify.h>
  33#include <linux/compat.h>
  34#include <linux/mnt_idmapping.h>
  35#include <linux/filelock.h>
  36
  37#include "internal.h"
  38
  39int do_truncate(struct mnt_idmap *idmap, 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(idmap, 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(idmap, 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 mnt_idmap *idmap;
  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	idmap = mnt_idmap(path->mnt);
  89	error = inode_permission(idmap, 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(idmap, 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_ftruncate(struct file *file, loff_t length, int small)
 157{
 158	struct inode *inode;
 159	struct dentry *dentry;
 
 160	int error;
 161
 
 
 
 
 
 
 
 
 162	/* explicitly opened as large or we are on 64-bit box */
 163	if (file->f_flags & O_LARGEFILE)
 164		small = 0;
 165
 166	dentry = file->f_path.dentry;
 167	inode = dentry->d_inode;
 168	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 169		return -EINVAL;
 
 170
 
 171	/* Cannot ftruncate over 2^31 bytes without large file support */
 172	if (small && length > MAX_NON_LFS)
 173		return -EINVAL;
 174
 
 175	/* Check IS_APPEND on real upper inode */
 176	if (IS_APPEND(file_inode(file)))
 177		return -EPERM;
 
 178	sb_start_write(inode->i_sb);
 179	error = security_file_truncate(file);
 
 
 180	if (!error)
 181		error = do_truncate(file_mnt_idmap(file), dentry, length,
 182				    ATTR_MTIME | ATTR_CTIME, file);
 183	sb_end_write(inode->i_sb);
 184
 
 
 185	return error;
 186}
 187
 188long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 189{
 190	if (length < 0)
 191		return -EINVAL;
 192	CLASS(fd, f)(fd);
 193	if (fd_empty(f))
 194		return -EBADF;
 195
 196	return do_ftruncate(fd_file(f), length, small);
 197}
 198
 199SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
 200{
 201	return do_sys_ftruncate(fd, length, 1);
 202}
 203
 204#ifdef CONFIG_COMPAT
 205COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
 206{
 207	return do_sys_ftruncate(fd, length, 1);
 208}
 209#endif
 210
 211/* LFS versions of truncate are only needed on 32 bit machines */
 212#if BITS_PER_LONG == 32
 213SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 214{
 215	return do_sys_truncate(path, length);
 216}
 217
 218SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 219{
 220	return do_sys_ftruncate(fd, length, 0);
 221}
 222#endif /* BITS_PER_LONG == 32 */
 223
 224#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
 225COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
 226		       compat_arg_u64_dual(length))
 227{
 228	return ksys_truncate(pathname, compat_arg_u64_glue(length));
 229}
 230#endif
 231
 232#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
 233COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
 234		       compat_arg_u64_dual(length))
 235{
 236	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
 237}
 238#endif
 239
 240int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 241{
 242	struct inode *inode = file_inode(file);
 243	long ret;
 244	loff_t sum;
 245
 246	if (offset < 0 || len <= 0)
 247		return -EINVAL;
 248
 249	if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
 
 250		return -EOPNOTSUPP;
 251
 252	/*
 253	 * Modes are exclusive, even if that is not obvious from the encoding
 254	 * as bit masks and the mix with the flag in the same namespace.
 255	 *
 256	 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
 257	 * encoded as no bit set.
 258	 */
 259	switch (mode & FALLOC_FL_MODE_MASK) {
 260	case FALLOC_FL_ALLOCATE_RANGE:
 261	case FALLOC_FL_UNSHARE_RANGE:
 262	case FALLOC_FL_ZERO_RANGE:
 263		break;
 264	case FALLOC_FL_PUNCH_HOLE:
 265		if (!(mode & FALLOC_FL_KEEP_SIZE))
 266			return -EOPNOTSUPP;
 267		break;
 268	case FALLOC_FL_COLLAPSE_RANGE:
 269	case FALLOC_FL_INSERT_RANGE:
 270		if (mode & FALLOC_FL_KEEP_SIZE)
 271			return -EOPNOTSUPP;
 272		break;
 273	default:
 274		return -EOPNOTSUPP;
 275	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 276
 277	if (!(file->f_mode & FMODE_WRITE))
 278		return -EBADF;
 279
 280	/*
 281	 * On append-only files only space preallocation is supported.
 282	 */
 283	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 284		return -EPERM;
 285
 286	if (IS_IMMUTABLE(inode))
 287		return -EPERM;
 288
 289	/*
 290	 * We cannot allow any fallocate operation on an active swapfile
 291	 */
 292	if (IS_SWAPFILE(inode))
 293		return -ETXTBSY;
 294
 295	/*
 296	 * Revalidate the write permissions, in case security policy has
 297	 * changed since the files were opened.
 298	 */
 299	ret = security_file_permission(file, MAY_WRITE);
 300	if (ret)
 301		return ret;
 302
 303	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
 304	if (ret)
 305		return ret;
 306
 307	if (S_ISFIFO(inode->i_mode))
 308		return -ESPIPE;
 309
 310	if (S_ISDIR(inode->i_mode))
 311		return -EISDIR;
 312
 313	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 314		return -ENODEV;
 315
 316	/* Check for wraparound */
 317	if (check_add_overflow(offset, len, &sum))
 318		return -EFBIG;
 319
 320	if (sum > inode->i_sb->s_maxbytes)
 321		return -EFBIG;
 322
 323	if (!file->f_op->fallocate)
 324		return -EOPNOTSUPP;
 325
 326	file_start_write(file);
 327	ret = file->f_op->fallocate(file, mode, offset, len);
 328
 329	/*
 330	 * Create inotify and fanotify events.
 331	 *
 332	 * To keep the logic simple always create events if fallocate succeeds.
 333	 * This implies that events are even created if the file size remains
 334	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 335	 */
 336	if (ret == 0)
 337		fsnotify_modify(file);
 338
 339	file_end_write(file);
 340	return ret;
 341}
 342EXPORT_SYMBOL_GPL(vfs_fallocate);
 343
 344int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 345{
 346	CLASS(fd, f)(fd);
 
 347
 348	if (fd_empty(f))
 349		return -EBADF;
 350
 351	return vfs_fallocate(fd_file(f), mode, offset, len);
 
 352}
 353
 354SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 355{
 356	return ksys_fallocate(fd, mode, offset, len);
 357}
 358
 359#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
 360COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
 361		       compat_arg_u64_dual(len))
 362{
 363	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
 364			      compat_arg_u64_glue(len));
 365}
 366#endif
 367
 368/*
 369 * access() needs to use the real uid/gid, not the effective uid/gid.
 370 * We do this by temporarily clearing all FS-related capabilities and
 371 * switching the fsuid/fsgid around to the real ones.
 372 *
 373 * Creating new credentials is expensive, so we try to skip doing it,
 374 * which we can if the result would match what we already got.
 375 */
 376static bool access_need_override_creds(int flags)
 377{
 378	const struct cred *cred;
 379
 380	if (flags & AT_EACCESS)
 381		return false;
 382
 383	cred = current_cred();
 384	if (!uid_eq(cred->fsuid, cred->uid) ||
 385	    !gid_eq(cred->fsgid, cred->gid))
 386		return true;
 387
 388	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 389		kuid_t root_uid = make_kuid(cred->user_ns, 0);
 390		if (!uid_eq(cred->uid, root_uid)) {
 391			if (!cap_isclear(cred->cap_effective))
 392				return true;
 393		} else {
 394			if (!cap_isidentical(cred->cap_effective,
 395			    cred->cap_permitted))
 396				return true;
 397		}
 398	}
 399
 400	return false;
 401}
 402
 403static const struct cred *access_override_creds(void)
 404{
 405	const struct cred *old_cred;
 406	struct cred *override_cred;
 
 
 
 
 
 
 
 407
 408	override_cred = prepare_creds();
 409	if (!override_cred)
 410		return NULL;
 411
 412	/*
 413	 * XXX access_need_override_creds performs checks in hopes of skipping
 414	 * this work. Make sure it stays in sync if making any changes in this
 415	 * routine.
 416	 */
 417
 418	override_cred->fsuid = override_cred->uid;
 419	override_cred->fsgid = override_cred->gid;
 420
 421	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 422		/* Clear the capabilities if we switch to a non-root user */
 423		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 424		if (!uid_eq(override_cred->uid, root_uid))
 425			cap_clear(override_cred->cap_effective);
 426		else
 427			override_cred->cap_effective =
 428				override_cred->cap_permitted;
 429	}
 430
 431	/*
 432	 * The new set of credentials can *only* be used in
 433	 * task-synchronous circumstances, and does not need
 434	 * RCU freeing, unless somebody then takes a separate
 435	 * reference to it.
 436	 *
 437	 * NOTE! This is _only_ true because this credential
 438	 * is used purely for override_creds() that installs
 439	 * it as the subjective cred. Other threads will be
 440	 * accessing ->real_cred, not the subjective cred.
 441	 *
 442	 * If somebody _does_ make a copy of this (using the
 443	 * 'get_current_cred()' function), that will clear the
 444	 * non_rcu field, because now that other user may be
 445	 * expecting RCU freeing. But normal thread-synchronous
 446	 * cred accesses will keep things non-racy to avoid RCU
 447	 * freeing.
 448	 */
 449	override_cred->non_rcu = 1;
 450
 451	old_cred = override_creds(override_cred);
 452
 453	/* override_cred() gets its own ref */
 454	put_cred(override_cred);
 455
 456	return old_cred;
 457}
 458
 459static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 460{
 461	struct path path;
 462	struct inode *inode;
 463	int res;
 464	unsigned int lookup_flags = LOOKUP_FOLLOW;
 465	const struct cred *old_cred = NULL;
 466
 467	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 468		return -EINVAL;
 469
 470	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 471		return -EINVAL;
 472
 473	if (flags & AT_SYMLINK_NOFOLLOW)
 474		lookup_flags &= ~LOOKUP_FOLLOW;
 475	if (flags & AT_EMPTY_PATH)
 476		lookup_flags |= LOOKUP_EMPTY;
 477
 478	if (access_need_override_creds(flags)) {
 479		old_cred = access_override_creds();
 480		if (!old_cred)
 481			return -ENOMEM;
 482	}
 483
 484retry:
 485	res = user_path_at(dfd, filename, lookup_flags, &path);
 486	if (res)
 487		goto out;
 488
 489	inode = d_backing_inode(path.dentry);
 490
 491	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 492		/*
 493		 * MAY_EXEC on regular files is denied if the fs is mounted
 494		 * with the "noexec" flag.
 495		 */
 496		res = -EACCES;
 497		if (path_noexec(&path))
 498			goto out_path_release;
 499	}
 500
 501	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
 502	/* SuS v2 requires we report a read only fs too */
 503	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 504		goto out_path_release;
 505	/*
 506	 * This is a rare case where using __mnt_is_readonly()
 507	 * is OK without a mnt_want/drop_write() pair.  Since
 508	 * no actual write to the fs is performed here, we do
 509	 * not need to telegraph to that to anyone.
 510	 *
 511	 * By doing this, we accept that this access is
 512	 * inherently racy and know that the fs may change
 513	 * state before we even see this result.
 514	 */
 515	if (__mnt_is_readonly(path.mnt))
 516		res = -EROFS;
 517
 518out_path_release:
 519	path_put(&path);
 520	if (retry_estale(res, lookup_flags)) {
 521		lookup_flags |= LOOKUP_REVAL;
 522		goto retry;
 523	}
 524out:
 525	if (old_cred)
 526		revert_creds(old_cred);
 527
 528	return res;
 529}
 530
 531SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 532{
 533	return do_faccessat(dfd, filename, mode, 0);
 534}
 535
 536SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 537		int, flags)
 538{
 539	return do_faccessat(dfd, filename, mode, flags);
 540}
 541
 542SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 543{
 544	return do_faccessat(AT_FDCWD, filename, mode, 0);
 545}
 546
 547SYSCALL_DEFINE1(chdir, const char __user *, filename)
 548{
 549	struct path path;
 550	int error;
 551	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 552retry:
 553	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 554	if (error)
 555		goto out;
 556
 557	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 558	if (error)
 559		goto dput_and_out;
 560
 561	set_fs_pwd(current->fs, &path);
 562
 563dput_and_out:
 564	path_put(&path);
 565	if (retry_estale(error, lookup_flags)) {
 566		lookup_flags |= LOOKUP_REVAL;
 567		goto retry;
 568	}
 569out:
 570	return error;
 571}
 572
 
 
 
 
 
 573SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 574{
 575	CLASS(fd_raw, f)(fd);
 576	int error;
 577
 578	if (fd_empty(f))
 579		return -EBADF;
 
 580
 581	if (!d_can_lookup(fd_file(f)->f_path.dentry))
 582		return -ENOTDIR;
 
 583
 584	error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
 585	if (!error)
 586		set_fs_pwd(current->fs, &fd_file(f)->f_path);
 
 
 
 587	return error;
 588}
 589
 590SYSCALL_DEFINE1(chroot, const char __user *, filename)
 591{
 592	struct path path;
 593	int error;
 594	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 595retry:
 596	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 597	if (error)
 598		goto out;
 599
 600	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 601	if (error)
 602		goto dput_and_out;
 603
 604	error = -EPERM;
 605	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 606		goto dput_and_out;
 607	error = security_path_chroot(&path);
 608	if (error)
 609		goto dput_and_out;
 610
 611	set_fs_root(current->fs, &path);
 612	error = 0;
 613dput_and_out:
 614	path_put(&path);
 615	if (retry_estale(error, lookup_flags)) {
 616		lookup_flags |= LOOKUP_REVAL;
 617		goto retry;
 618	}
 619out:
 620	return error;
 621}
 622
 623int chmod_common(const struct path *path, umode_t mode)
 
 
 
 
 
 624{
 625	struct inode *inode = path->dentry->d_inode;
 626	struct inode *delegated_inode = NULL;
 627	struct iattr newattrs;
 628	int error;
 629
 630	error = mnt_want_write(path->mnt);
 631	if (error)
 632		return error;
 633retry_deleg:
 634	inode_lock(inode);
 635	error = security_path_chmod(path, mode);
 636	if (error)
 637		goto out_unlock;
 638	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 639	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 640	error = notify_change(mnt_idmap(path->mnt), path->dentry,
 641			      &newattrs, &delegated_inode);
 642out_unlock:
 643	inode_unlock(inode);
 644	if (delegated_inode) {
 645		error = break_deleg_wait(&delegated_inode);
 646		if (!error)
 647			goto retry_deleg;
 648	}
 649	mnt_drop_write(path->mnt);
 650	return error;
 651}
 652
 653int vfs_fchmod(struct file *file, umode_t mode)
 654{
 655	audit_file(file);
 656	return chmod_common(&file->f_path, mode);
 
 
 
 
 
 
 
 657}
 658
 659SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 660{
 661	CLASS(fd, f)(fd);
 662
 663	if (fd_empty(f))
 664		return -EBADF;
 665
 666	return vfs_fchmod(fd_file(f), mode);
 667}
 668
 669static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
 670		       unsigned int flags)
 671{
 672	struct path path;
 673	int error;
 674	unsigned int lookup_flags;
 675
 676	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
 677		return -EINVAL;
 678
 679	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 680	if (flags & AT_EMPTY_PATH)
 681		lookup_flags |= LOOKUP_EMPTY;
 682
 683retry:
 684	error = user_path_at(dfd, filename, lookup_flags, &path);
 685	if (!error) {
 686		error = chmod_common(&path, mode);
 687		path_put(&path);
 688		if (retry_estale(error, lookup_flags)) {
 689			lookup_flags |= LOOKUP_REVAL;
 690			goto retry;
 691		}
 692	}
 693	return error;
 694}
 695
 696SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
 697		umode_t, mode, unsigned int, flags)
 698{
 699	return do_fchmodat(dfd, filename, mode, flags);
 700}
 701
 702SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 703		umode_t, mode)
 704{
 705	return do_fchmodat(dfd, filename, mode, 0);
 706}
 707
 708SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 709{
 710	return do_fchmodat(AT_FDCWD, filename, mode, 0);
 711}
 712
 713/*
 714 * Check whether @kuid is valid and if so generate and set vfsuid_t in
 715 * ia_vfsuid.
 716 *
 717 * Return: true if @kuid is valid, false if not.
 718 */
 719static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
 720{
 721	if (!uid_valid(kuid))
 722		return false;
 723	attr->ia_valid |= ATTR_UID;
 724	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
 725	return true;
 726}
 727
 728/*
 729 * Check whether @kgid is valid and if so generate and set vfsgid_t in
 730 * ia_vfsgid.
 731 *
 732 * Return: true if @kgid is valid, false if not.
 733 */
 734static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
 735{
 736	if (!gid_valid(kgid))
 737		return false;
 738	attr->ia_valid |= ATTR_GID;
 739	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
 740	return true;
 741}
 742
 743int chown_common(const struct path *path, uid_t user, gid_t group)
 744{
 745	struct mnt_idmap *idmap;
 746	struct user_namespace *fs_userns;
 747	struct inode *inode = path->dentry->d_inode;
 748	struct inode *delegated_inode = NULL;
 749	int error;
 750	struct iattr newattrs;
 751	kuid_t uid;
 752	kgid_t gid;
 753
 754	uid = make_kuid(current_user_ns(), user);
 755	gid = make_kgid(current_user_ns(), group);
 756
 757	idmap = mnt_idmap(path->mnt);
 758	fs_userns = i_user_ns(inode);
 759
 760retry_deleg:
 761	newattrs.ia_vfsuid = INVALID_VFSUID;
 762	newattrs.ia_vfsgid = INVALID_VFSGID;
 763	newattrs.ia_valid =  ATTR_CTIME;
 764	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 765		return -EINVAL;
 766	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 767		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 768	inode_lock(inode);
 769	if (!S_ISDIR(inode->i_mode))
 770		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
 771				     setattr_should_drop_sgid(idmap, inode);
 772	/* Continue to send actual fs values, not the mount values. */
 773	error = security_path_chown(
 774		path,
 775		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
 776		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
 777	if (!error)
 778		error = notify_change(idmap, path->dentry, &newattrs,
 779				      &delegated_inode);
 780	inode_unlock(inode);
 781	if (delegated_inode) {
 782		error = break_deleg_wait(&delegated_inode);
 783		if (!error)
 784			goto retry_deleg;
 785	}
 786	return error;
 787}
 788
 789int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 790		int flag)
 791{
 792	struct path path;
 793	int error = -EINVAL;
 794	int lookup_flags;
 795
 796	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 797		goto out;
 798
 799	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 800	if (flag & AT_EMPTY_PATH)
 801		lookup_flags |= LOOKUP_EMPTY;
 802retry:
 803	error = user_path_at(dfd, filename, lookup_flags, &path);
 804	if (error)
 805		goto out;
 806	error = mnt_want_write(path.mnt);
 807	if (error)
 808		goto out_release;
 809	error = chown_common(&path, user, group);
 810	mnt_drop_write(path.mnt);
 811out_release:
 812	path_put(&path);
 813	if (retry_estale(error, lookup_flags)) {
 814		lookup_flags |= LOOKUP_REVAL;
 815		goto retry;
 816	}
 817out:
 818	return error;
 819}
 820
 821SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 822		gid_t, group, int, flag)
 823{
 824	return do_fchownat(dfd, filename, user, group, flag);
 825}
 826
 827SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 828{
 829	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 830}
 831
 832SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 833{
 834	return do_fchownat(AT_FDCWD, filename, user, group,
 835			   AT_SYMLINK_NOFOLLOW);
 836}
 837
 838int vfs_fchown(struct file *file, uid_t user, gid_t group)
 839{
 840	int error;
 
 
 
 
 841
 842	error = mnt_want_write_file(file);
 843	if (error)
 844		return error;
 845	audit_file(file);
 846	error = chown_common(&file->f_path, user, group);
 847	mnt_drop_write_file(file);
 
 
 
 848	return error;
 849}
 850
 851int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 852{
 853	CLASS(fd, f)(fd);
 854
 855	if (fd_empty(f))
 856		return -EBADF;
 857
 858	return vfs_fchown(fd_file(f), user, group);
 859}
 860
 861SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 862{
 863	return ksys_fchown(fd, user, group);
 864}
 865
 866static inline int file_get_write_access(struct file *f)
 867{
 868	int error;
 869
 870	error = get_write_access(f->f_inode);
 871	if (unlikely(error))
 872		return error;
 873	error = mnt_get_write_access(f->f_path.mnt);
 874	if (unlikely(error))
 875		goto cleanup_inode;
 876	if (unlikely(f->f_mode & FMODE_BACKING)) {
 877		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
 878		if (unlikely(error))
 879			goto cleanup_mnt;
 880	}
 881	return 0;
 882
 883cleanup_mnt:
 884	mnt_put_write_access(f->f_path.mnt);
 885cleanup_inode:
 886	put_write_access(f->f_inode);
 887	return error;
 888}
 889
 890static int do_dentry_open(struct file *f,
 891			  int (*open)(struct inode *, struct file *))
 
 
 892{
 893	static const struct file_operations empty_fops = {};
 894	struct inode *inode = f->f_path.dentry->d_inode;
 895	int error;
 896
 
 
 
 897	path_get(&f->f_path);
 898	f->f_inode = inode;
 899	f->f_mapping = inode->i_mapping;
 
 
 900	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 901	f->f_sb_err = file_sample_sb_err(f);
 902
 903	if (unlikely(f->f_flags & O_PATH)) {
 904		f->f_mode = FMODE_PATH | FMODE_OPENED;
 905		f->f_op = &empty_fops;
 906		return 0;
 907	}
 908
 909	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
 910		i_readcount_inc(inode);
 911	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 912		error = file_get_write_access(f);
 913		if (unlikely(error))
 914			goto cleanup_file;
 
 
 
 
 
 915		f->f_mode |= FMODE_WRITER;
 916	}
 917
 918	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 919	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 920		f->f_mode |= FMODE_ATOMIC_POS;
 921
 922	f->f_op = fops_get(inode->i_fop);
 923	if (WARN_ON(!f->f_op)) {
 924		error = -ENODEV;
 925		goto cleanup_all;
 926	}
 927
 928	error = security_file_open(f);
 929	if (error)
 930		goto cleanup_all;
 931
 932	error = fsnotify_open_perm(f);
 933	if (error)
 934		goto cleanup_all;
 935
 936	error = break_lease(file_inode(f), f->f_flags);
 937	if (error)
 938		goto cleanup_all;
 939
 940	/* normally all 3 are set; ->open() can clear them if needed */
 941	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 942	if (!open)
 943		open = f->f_op->open;
 944	if (open) {
 945		error = open(inode, f);
 946		if (error)
 947			goto cleanup_all;
 948	}
 949	f->f_mode |= FMODE_OPENED;
 
 950	if ((f->f_mode & FMODE_READ) &&
 951	     likely(f->f_op->read || f->f_op->read_iter))
 952		f->f_mode |= FMODE_CAN_READ;
 953	if ((f->f_mode & FMODE_WRITE) &&
 954	     likely(f->f_op->write || f->f_op->write_iter))
 955		f->f_mode |= FMODE_CAN_WRITE;
 956	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 957		f->f_mode &= ~FMODE_LSEEK;
 958	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 959		f->f_mode |= FMODE_CAN_ODIRECT;
 960
 
 961	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 962	f->f_iocb_flags = iocb_flags(f);
 963
 964	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 965
 966	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 967		return -EINVAL;
 968
 969	/*
 970	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 971	 * cache for this file before processing writes.
 972	 */
 973	if (f->f_mode & FMODE_WRITE) {
 974		/*
 975		 * Depends on full fence from get_write_access() to synchronize
 976		 * against collapse_file() regarding i_writecount and nr_thps
 977		 * updates. Ensures subsequent insertion of THPs into the page
 978		 * cache will fail.
 979		 */
 980		if (filemap_nr_thps(inode->i_mapping)) {
 981			struct address_space *mapping = inode->i_mapping;
 982
 983			filemap_invalidate_lock(inode->i_mapping);
 984			/*
 985			 * unmap_mapping_range just need to be called once
 986			 * here, because the private pages is not need to be
 987			 * unmapped mapping (e.g. data segment of dynamic
 988			 * shared libraries here).
 989			 */
 990			unmap_mapping_range(mapping, 0, 0, 0);
 991			truncate_inode_pages(mapping, 0);
 992			filemap_invalidate_unlock(inode->i_mapping);
 993		}
 994	}
 995
 996	return 0;
 997
 998cleanup_all:
 999	if (WARN_ON_ONCE(error > 0))
1000		error = -EINVAL;
1001	fops_put(f->f_op);
1002	put_file_access(f);
 
 
 
1003cleanup_file:
1004	path_put(&f->f_path);
1005	f->f_path.mnt = NULL;
1006	f->f_path.dentry = NULL;
1007	f->f_inode = NULL;
1008	return error;
1009}
1010
1011/**
1012 * finish_open - finish opening a file
1013 * @file: file pointer
1014 * @dentry: pointer to dentry
1015 * @open: open callback
 
1016 *
1017 * This can be used to finish opening a file passed to i_op->atomic_open().
1018 *
1019 * If the open callback is set to NULL, then the standard f_op->open()
1020 * filesystem callback is substituted.
1021 *
1022 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1023 * the return value of d_splice_alias(), then the caller needs to perform dput()
1024 * on it after finish_open().
1025 *
 
 
 
1026 * Returns zero on success or -errno if the open failed.
1027 */
1028int finish_open(struct file *file, struct dentry *dentry,
1029		int (*open)(struct inode *, struct file *))
 
1030{
1031	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 
1032
1033	file->f_path.dentry = dentry;
1034	return do_dentry_open(file, open);
 
 
 
 
 
1035}
1036EXPORT_SYMBOL(finish_open);
1037
1038/**
1039 * finish_no_open - finish ->atomic_open() without opening the file
1040 *
1041 * @file: file pointer
1042 * @dentry: dentry or NULL (as returned from ->lookup())
1043 *
1044 * This can be used to set the result of a successful lookup in ->atomic_open().
1045 *
1046 * NB: unlike finish_open() this function does consume the dentry reference and
1047 * the caller need not dput() it.
1048 *
1049 * Returns "0" which must be the return value of ->atomic_open() after having
1050 * called this function.
1051 */
1052int finish_no_open(struct file *file, struct dentry *dentry)
1053{
1054	file->f_path.dentry = dentry;
1055	return 0;
1056}
1057EXPORT_SYMBOL(finish_no_open);
1058
1059char *file_path(struct file *filp, char *buf, int buflen)
1060{
1061	return d_path(&filp->f_path, buf, buflen);
1062}
1063EXPORT_SYMBOL(file_path);
1064
1065/**
1066 * vfs_open - open the file at the given path
1067 * @path: path to open
1068 * @file: newly allocated file with f_flag initialized
 
1069 */
1070int vfs_open(const struct path *path, struct file *file)
 
1071{
1072	int ret;
 
 
 
1073
1074	file->f_path = *path;
1075	ret = do_dentry_open(file, NULL);
1076	if (!ret) {
1077		/*
1078		 * Once we return a file with FMODE_OPENED, __fput() will call
1079		 * fsnotify_close(), so we need fsnotify_open() here for
1080		 * symmetry.
1081		 */
1082		fsnotify_open(file);
1083	}
1084	return ret;
1085}
1086
1087struct file *dentry_open(const struct path *path, int flags,
1088			 const struct cred *cred)
1089{
1090	int error;
1091	struct file *f;
1092
 
 
1093	/* We must always pass in a valid mount pointer. */
1094	BUG_ON(!path->mnt);
1095
1096	f = alloc_empty_file(flags, cred);
1097	if (!IS_ERR(f)) {
1098		error = vfs_open(path, f);
1099		if (error) {
1100			fput(f);
 
 
 
 
 
 
 
 
1101			f = ERR_PTR(error);
1102		}
1103	}
1104	return f;
1105}
1106EXPORT_SYMBOL(dentry_open);
1107
1108/**
1109 * dentry_create - Create and open a file
1110 * @path: path to create
1111 * @flags: O_ flags
1112 * @mode: mode bits for new file
1113 * @cred: credentials to use
1114 *
1115 * Caller must hold the parent directory's lock, and have prepared
1116 * a negative dentry, placed in @path->dentry, for the new file.
1117 *
1118 * Caller sets @path->mnt to the vfsmount of the filesystem where
1119 * the new file is to be created. The parent directory and the
1120 * negative dentry must reside on the same filesystem instance.
1121 *
1122 * On success, returns a "struct file *". Otherwise a ERR_PTR
1123 * is returned.
1124 */
1125struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1126			   const struct cred *cred)
1127{
1128	struct file *f;
1129	int error;
1130
1131	f = alloc_empty_file(flags, cred);
1132	if (IS_ERR(f))
1133		return f;
1134
1135	error = vfs_create(mnt_idmap(path->mnt),
1136			   d_inode(path->dentry->d_parent),
1137			   path->dentry, mode, true);
1138	if (!error)
1139		error = vfs_open(path, f);
1140
1141	if (unlikely(error)) {
1142		fput(f);
1143		return ERR_PTR(error);
1144	}
1145	return f;
1146}
1147EXPORT_SYMBOL(dentry_create);
1148
1149/**
1150 * kernel_file_open - open a file for kernel internal use
1151 * @path:	path of the file to open
1152 * @flags:	open flags
1153 * @cred:	credentials for open
1154 *
1155 * Open a file for use by in-kernel consumers. The file is not accounted
1156 * against nr_files and must not be installed into the file descriptor
1157 * table.
1158 *
1159 * Return: Opened file on success, an error pointer on failure.
1160 */
1161struct file *kernel_file_open(const struct path *path, int flags,
1162				const struct cred *cred)
1163{
1164	struct file *f;
1165	int error;
1166
1167	f = alloc_empty_file_noaccount(flags, cred);
1168	if (IS_ERR(f))
1169		return f;
1170
1171	f->f_path = *path;
1172	error = do_dentry_open(f, NULL);
1173	if (error) {
1174		fput(f);
1175		return ERR_PTR(error);
1176	}
1177
1178	fsnotify_open(f);
1179	return f;
1180}
1181EXPORT_SYMBOL_GPL(kernel_file_open);
1182
1183#define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1184#define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1185
1186inline struct open_how build_open_how(int flags, umode_t mode)
1187{
1188	struct open_how how = {
1189		.flags = flags & VALID_OPEN_FLAGS,
1190		.mode = mode & S_IALLUGO,
1191	};
1192
1193	/* O_PATH beats everything else. */
1194	if (how.flags & O_PATH)
1195		how.flags &= O_PATH_FLAGS;
1196	/* Modes should only be set for create-like flags. */
1197	if (!WILL_CREATE(how.flags))
1198		how.mode = 0;
1199	return how;
1200}
1201
1202inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1203{
1204	u64 flags = how->flags;
1205	u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1206	int lookup_flags = 0;
1207	int acc_mode = ACC_MODE(flags);
1208
1209	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1210			 "struct open_flags doesn't yet handle flags > 32 bits");
1211
1212	/*
1213	 * Strip flags that either shouldn't be set by userspace like
1214	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1215	 * open_flags like O_CLOEXEC.
1216	 */
1217	flags &= ~strip;
1218
1219	/*
1220	 * Older syscalls implicitly clear all of the invalid flags or argument
1221	 * values before calling build_open_flags(), but openat2(2) checks all
1222	 * of its arguments.
1223	 */
1224	if (flags & ~VALID_OPEN_FLAGS)
1225		return -EINVAL;
1226	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1227		return -EINVAL;
1228
1229	/* Scoping flags are mutually exclusive. */
1230	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1231		return -EINVAL;
1232
1233	/* Deal with the mode. */
1234	if (WILL_CREATE(flags)) {
1235		if (how->mode & ~S_IALLUGO)
1236			return -EINVAL;
1237		op->mode = how->mode | S_IFREG;
1238	} else {
1239		if (how->mode != 0)
1240			return -EINVAL;
1241		op->mode = 0;
1242	}
1243
1244	/*
1245	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1246	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1247	 * O_TMPFILE below which requires O_DIRECTORY being raised.
 
1248	 */
1249	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1250		return -EINVAL;
1251
1252	/* Now handle the creative implementation of O_TMPFILE. */
1253	if (flags & __O_TMPFILE) {
1254		/*
1255		 * In order to ensure programs get explicit errors when trying
1256		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1257		 * is raised alongside __O_TMPFILE.
1258		 */
1259		if (!(flags & O_DIRECTORY))
1260			return -EINVAL;
1261		if (!(acc_mode & MAY_WRITE))
1262			return -EINVAL;
1263	}
1264	if (flags & O_PATH) {
1265		/* O_PATH only permits certain other flags to be set. */
1266		if (flags & ~O_PATH_FLAGS)
1267			return -EINVAL;
 
1268		acc_mode = 0;
1269	}
1270
1271	/*
1272	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1273	 * check for O_DSYNC if the need any syncing at all we enforce it's
1274	 * always set instead of having to deal with possibly weird behaviour
1275	 * for malicious applications setting only __O_SYNC.
1276	 */
1277	if (flags & __O_SYNC)
1278		flags |= O_DSYNC;
1279
1280	op->open_flag = flags;
1281
1282	/* O_TRUNC implies we need access checks for write permissions */
1283	if (flags & O_TRUNC)
1284		acc_mode |= MAY_WRITE;
1285
1286	/* Allow the LSM permission hook to distinguish append
1287	   access from general write access. */
1288	if (flags & O_APPEND)
1289		acc_mode |= MAY_APPEND;
1290
1291	op->acc_mode = acc_mode;
1292
1293	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1294
1295	if (flags & O_CREAT) {
1296		op->intent |= LOOKUP_CREATE;
1297		if (flags & O_EXCL) {
1298			op->intent |= LOOKUP_EXCL;
1299			flags |= O_NOFOLLOW;
1300		}
1301	}
1302
1303	if (flags & O_DIRECTORY)
1304		lookup_flags |= LOOKUP_DIRECTORY;
1305	if (!(flags & O_NOFOLLOW))
1306		lookup_flags |= LOOKUP_FOLLOW;
1307
1308	if (how->resolve & RESOLVE_NO_XDEV)
1309		lookup_flags |= LOOKUP_NO_XDEV;
1310	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1311		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1312	if (how->resolve & RESOLVE_NO_SYMLINKS)
1313		lookup_flags |= LOOKUP_NO_SYMLINKS;
1314	if (how->resolve & RESOLVE_BENEATH)
1315		lookup_flags |= LOOKUP_BENEATH;
1316	if (how->resolve & RESOLVE_IN_ROOT)
1317		lookup_flags |= LOOKUP_IN_ROOT;
1318	if (how->resolve & RESOLVE_CACHED) {
1319		/* Don't bother even trying for create/truncate/tmpfile open */
1320		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1321			return -EAGAIN;
1322		lookup_flags |= LOOKUP_CACHED;
1323	}
1324
1325	op->lookup_flags = lookup_flags;
1326	return 0;
1327}
1328
1329/**
1330 * file_open_name - open file and return file pointer
1331 *
1332 * @name:	struct filename containing path to open
1333 * @flags:	open flags as per the open(2) second argument
1334 * @mode:	mode for the new file if O_CREAT is set, else ignored
1335 *
1336 * This is the helper to open a file from kernelspace if you really
1337 * have to.  But in generally you should not do this, so please move
1338 * along, nothing to see here..
1339 */
1340struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1341{
1342	struct open_flags op;
1343	struct open_how how = build_open_how(flags, mode);
1344	int err = build_open_flags(&how, &op);
1345	if (err)
1346		return ERR_PTR(err);
1347	return do_filp_open(AT_FDCWD, name, &op);
1348}
1349
1350/**
1351 * filp_open - open file and return file pointer
1352 *
1353 * @filename:	path to open
1354 * @flags:	open flags as per the open(2) second argument
1355 * @mode:	mode for the new file if O_CREAT is set, else ignored
1356 *
1357 * This is the helper to open a file from kernelspace if you really
1358 * have to.  But in generally you should not do this, so please move
1359 * along, nothing to see here..
1360 */
1361struct file *filp_open(const char *filename, int flags, umode_t mode)
1362{
1363	struct filename *name = getname_kernel(filename);
1364	struct file *file = ERR_CAST(name);
1365
1366	if (!IS_ERR(name)) {
1367		file = file_open_name(name, flags, mode);
1368		putname(name);
1369	}
1370	return file;
1371}
1372EXPORT_SYMBOL(filp_open);
1373
1374struct file *file_open_root(const struct path *root,
1375			    const char *filename, int flags, umode_t mode)
1376{
1377	struct open_flags op;
1378	struct open_how how = build_open_how(flags, mode);
1379	int err = build_open_flags(&how, &op);
1380	if (err)
1381		return ERR_PTR(err);
1382	return do_file_open_root(root, filename, &op);
1383}
1384EXPORT_SYMBOL(file_open_root);
1385
1386static long do_sys_openat2(int dfd, const char __user *filename,
1387			   struct open_how *how)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1388{
1389	struct open_flags op;
1390	int fd = build_open_flags(how, &op);
1391	struct filename *tmp;
1392
1393	if (fd)
1394		return fd;
1395
1396	tmp = getname(filename);
1397	if (IS_ERR(tmp))
1398		return PTR_ERR(tmp);
1399
1400	fd = get_unused_fd_flags(how->flags);
1401	if (fd >= 0) {
1402		struct file *f = do_filp_open(dfd, tmp, &op);
1403		if (IS_ERR(f)) {
1404			put_unused_fd(fd);
1405			fd = PTR_ERR(f);
1406		} else {
 
1407			fd_install(fd, f);
1408		}
1409	}
1410	putname(tmp);
1411	return fd;
1412}
1413
1414long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1415{
1416	struct open_how how = build_open_how(flags, mode);
1417	return do_sys_openat2(dfd, filename, &how);
1418}
1419
1420
1421SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1422{
1423	if (force_o_largefile())
1424		flags |= O_LARGEFILE;
 
1425	return do_sys_open(AT_FDCWD, filename, flags, mode);
1426}
1427
1428SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1429		umode_t, mode)
1430{
1431	if (force_o_largefile())
1432		flags |= O_LARGEFILE;
 
1433	return do_sys_open(dfd, filename, flags, mode);
1434}
1435
1436SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1437		struct open_how __user *, how, size_t, usize)
1438{
1439	int err;
1440	struct open_how tmp;
1441
1442	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1443	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1444
1445	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1446		return -EINVAL;
1447	if (unlikely(usize > PAGE_SIZE))
1448		return -E2BIG;
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 * This routine simulates a hangup on the tty, to arrange that users
1568 * are given clean terminals at login time.
1569 */
1570SYSCALL_DEFINE0(vhangup)
1571{
1572	if (capable(CAP_SYS_TTY_CONFIG)) {
1573		tty_vhangup_self();
1574		return 0;
1575	}
1576	return -EPERM;
1577}
1578
1579/*
1580 * Called when an inode is about to be open.
1581 * We use this to disallow opening large files on 32bit systems if
1582 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1583 * on this flag in sys_open.
1584 */
1585int generic_file_open(struct inode * inode, struct file * filp)
1586{
1587	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1588		return -EOVERFLOW;
1589	return 0;
1590}
1591
1592EXPORT_SYMBOL(generic_file_open);
1593
1594/*
1595 * This is used by subsystems that don't want seekable
1596 * file descriptors. The function is not supposed to ever fail, the only
1597 * reason it returns an 'int' and not 'void' is so that it can be plugged
1598 * directly into file_operations structure.
1599 */
1600int nonseekable_open(struct inode *inode, struct file *filp)
1601{
1602	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1603	return 0;
1604}
1605
1606EXPORT_SYMBOL(nonseekable_open);
1607
1608/*
1609 * stream_open is used by subsystems that want stream-like file descriptors.
1610 * Such file descriptors are not seekable and don't have notion of position
1611 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1612 * Contrary to file descriptors of other regular files, .read() and .write()
1613 * can run simultaneously.
1614 *
1615 * stream_open never fails and is marked to return int so that it could be
1616 * directly used as file_operations.open .
1617 */
1618int stream_open(struct inode *inode, struct file *filp)
1619{
1620	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1621	filp->f_mode |= FMODE_STREAM;
1622	return 0;
1623}
1624
1625EXPORT_SYMBOL(stream_open);