Linux Audio

Check our new training course

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);
v3.1
   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 <asm/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
  34#include "internal.h"
  35
  36int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  37	struct file *filp)
  38{
  39	int ret;
  40	struct iattr newattrs;
  41
  42	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  43	if (length < 0)
  44		return -EINVAL;
  45
  46	newattrs.ia_size = length;
  47	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  48	if (filp) {
  49		newattrs.ia_file = filp;
  50		newattrs.ia_valid |= ATTR_FILE;
  51	}
  52
  53	/* Remove suid/sgid on truncate too */
  54	ret = should_remove_suid(dentry);
 
 
  55	if (ret)
  56		newattrs.ia_valid |= ret | ATTR_FORCE;
  57
  58	mutex_lock(&dentry->d_inode->i_mutex);
  59	ret = notify_change(dentry, &newattrs);
  60	mutex_unlock(&dentry->d_inode->i_mutex);
 
  61	return ret;
  62}
  63
  64static long do_sys_truncate(const char __user *pathname, loff_t length)
  65{
  66	struct path path;
  67	struct inode *inode;
  68	int error;
  69
  70	error = -EINVAL;
  71	if (length < 0)	/* sorry, but loff_t says... */
  72		goto out;
  73
  74	error = user_path(pathname, &path);
  75	if (error)
  76		goto out;
  77	inode = path.dentry->d_inode;
  78
  79	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  80	error = -EISDIR;
  81	if (S_ISDIR(inode->i_mode))
  82		goto dput_and_out;
  83
  84	error = -EINVAL;
  85	if (!S_ISREG(inode->i_mode))
  86		goto dput_and_out;
  87
  88	error = mnt_want_write(path.mnt);
  89	if (error)
  90		goto dput_and_out;
  91
  92	error = inode_permission(inode, MAY_WRITE);
  93	if (error)
  94		goto mnt_drop_write_and_out;
  95
  96	error = -EPERM;
  97	if (IS_APPEND(inode))
  98		goto mnt_drop_write_and_out;
  99
 100	error = get_write_access(inode);
 
 
 
 
 
 
 
 
 
 
 101	if (error)
 102		goto mnt_drop_write_and_out;
 103
 104	/*
 105	 * Make sure that there are no leases.  get_write_access() protects
 106	 * against the truncate racing with a lease-granting setlease().
 107	 */
 108	error = break_lease(inode, O_WRONLY);
 109	if (error)
 110		goto put_write_and_out;
 111
 112	error = locks_verify_truncate(inode, NULL, length);
 113	if (!error)
 114		error = security_path_truncate(&path);
 115	if (!error)
 116		error = do_truncate(path.dentry, length, 0, NULL);
 117
 118put_write_and_out:
 119	put_write_access(inode);
 120mnt_drop_write_and_out:
 121	mnt_drop_write(path.mnt);
 122dput_and_out:
 123	path_put(&path);
 124out:
 125	return error;
 126}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 127
 128SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 129{
 130	return do_sys_truncate(path, length);
 131}
 132
 133static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 
 134{
 135	struct inode * inode;
 
 
 
 
 
 
 136	struct dentry *dentry;
 137	struct file * file;
 138	int error;
 139
 140	error = -EINVAL;
 141	if (length < 0)
 142		goto out;
 143	error = -EBADF;
 144	file = fget(fd);
 145	if (!file)
 146		goto out;
 147
 148	/* explicitly opened as large or we are on 64-bit box */
 149	if (file->f_flags & O_LARGEFILE)
 150		small = 0;
 151
 152	dentry = file->f_path.dentry;
 153	inode = dentry->d_inode;
 154	error = -EINVAL;
 155	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 156		goto out_putf;
 157
 158	error = -EINVAL;
 159	/* Cannot ftruncate over 2^31 bytes without large file support */
 160	if (small && length > MAX_NON_LFS)
 161		goto out_putf;
 162
 163	error = -EPERM;
 164	if (IS_APPEND(inode))
 
 165		goto out_putf;
 166
 167	error = locks_verify_truncate(inode, file, length);
 
 168	if (!error)
 169		error = security_path_truncate(&file->f_path);
 170	if (!error)
 171		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 
 172out_putf:
 173	fput(file);
 174out:
 175	return error;
 176}
 177
 178SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 179{
 180	long ret = do_sys_ftruncate(fd, length, 1);
 181	/* avoid REGPARM breakage on x86: */
 182	asmlinkage_protect(2, ret, fd, length);
 183	return ret;
 
 
 
 184}
 
 185
 186/* LFS versions of truncate are only needed on 32 bit machines */
 187#if BITS_PER_LONG == 32
 188SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 189{
 190	return do_sys_truncate(path, length);
 191}
 192#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 193asmlinkage long SyS_truncate64(long path, loff_t length)
 194{
 195	return SYSC_truncate64((const char __user *) path, length);
 196}
 197SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 198#endif
 199
 200SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 201{
 202	long ret = do_sys_ftruncate(fd, length, 0);
 203	/* avoid REGPARM breakage on x86: */
 204	asmlinkage_protect(2, ret, fd, length);
 205	return ret;
 206}
 207#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 208asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 209{
 210	return SYSC_ftruncate64((unsigned int) fd, length);
 211}
 212SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 213#endif
 214#endif /* BITS_PER_LONG == 32 */
 215
 216
 217int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 218{
 219	struct inode *inode = file->f_path.dentry->d_inode;
 220	long ret;
 221
 222	if (offset < 0 || len <= 0)
 223		return -EINVAL;
 224
 225	/* Return error if mode is not supported */
 226	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 
 
 
 
 
 227		return -EOPNOTSUPP;
 228
 229	/* Punch hole must have keep size set */
 230	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 231	    !(mode & FALLOC_FL_KEEP_SIZE))
 232		return -EOPNOTSUPP;
 233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 234	if (!(file->f_mode & FMODE_WRITE))
 235		return -EBADF;
 236
 237	/* It's not possible punch hole on append only file */
 238	if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
 
 
 239		return -EPERM;
 240
 241	if (IS_IMMUTABLE(inode))
 242		return -EPERM;
 243
 244	/*
 
 
 
 
 
 
 245	 * Revalidate the write permissions, in case security policy has
 246	 * changed since the files were opened.
 247	 */
 248	ret = security_file_permission(file, MAY_WRITE);
 249	if (ret)
 250		return ret;
 251
 252	if (S_ISFIFO(inode->i_mode))
 253		return -ESPIPE;
 254
 255	/*
 256	 * Let individual file system decide if it supports preallocation
 257	 * for directories or not.
 258	 */
 259	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 260		return -ENODEV;
 261
 262	/* Check for wrap through zero too */
 263	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 264		return -EFBIG;
 265
 266	if (!file->f_op->fallocate)
 267		return -EOPNOTSUPP;
 268
 269	return file->f_op->fallocate(file, mode, offset, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270}
 
 271
 272SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 273{
 274	struct file *file;
 275	int error = -EBADF;
 276
 277	file = fget(fd);
 278	if (file) {
 279		error = do_fallocate(file, mode, offset, len);
 280		fput(file);
 281	}
 282
 283	return error;
 284}
 285
 286#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 287asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 288{
 289	return SYSC_fallocate((int)fd, (int)mode, offset, len);
 290}
 291SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 292#endif
 293
 294/*
 295 * access() needs to use the real uid/gid, not the effective uid/gid.
 296 * We do this by temporarily clearing all FS-related capabilities and
 297 * switching the fsuid/fsgid around to the real ones.
 298 */
 299SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 300{
 301	const struct cred *old_cred;
 302	struct cred *override_cred;
 303	struct path path;
 304	struct inode *inode;
 305	int res;
 
 306
 307	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 308		return -EINVAL;
 309
 310	override_cred = prepare_creds();
 311	if (!override_cred)
 312		return -ENOMEM;
 313
 314	override_cred->fsuid = override_cred->uid;
 315	override_cred->fsgid = override_cred->gid;
 316
 317	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 318		/* Clear the capabilities if we switch to a non-root user */
 319		if (override_cred->uid)
 
 320			cap_clear(override_cred->cap_effective);
 321		else
 322			override_cred->cap_effective =
 323				override_cred->cap_permitted;
 324	}
 325
 326	old_cred = override_creds(override_cred);
 327
 328	res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 329	if (res)
 330		goto out;
 331
 332	inode = path.dentry->d_inode;
 333
 334	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 335		/*
 336		 * MAY_EXEC on regular files is denied if the fs is mounted
 337		 * with the "noexec" flag.
 338		 */
 339		res = -EACCES;
 340		if (path.mnt->mnt_flags & MNT_NOEXEC)
 341			goto out_path_release;
 342	}
 343
 344	res = inode_permission(inode, mode | MAY_ACCESS);
 345	/* SuS v2 requires we report a read only fs too */
 346	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 347		goto out_path_release;
 348	/*
 349	 * This is a rare case where using __mnt_is_readonly()
 350	 * is OK without a mnt_want/drop_write() pair.  Since
 351	 * no actual write to the fs is performed here, we do
 352	 * not need to telegraph to that to anyone.
 353	 *
 354	 * By doing this, we accept that this access is
 355	 * inherently racy and know that the fs may change
 356	 * state before we even see this result.
 357	 */
 358	if (__mnt_is_readonly(path.mnt))
 359		res = -EROFS;
 360
 361out_path_release:
 362	path_put(&path);
 
 
 
 
 363out:
 364	revert_creds(old_cred);
 365	put_cred(override_cred);
 366	return res;
 367}
 368
 
 
 
 
 
 369SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 370{
 371	return sys_faccessat(AT_FDCWD, filename, mode);
 372}
 373
 374SYSCALL_DEFINE1(chdir, const char __user *, filename)
 375{
 376	struct path path;
 377	int error;
 378
 379	error = user_path_dir(filename, &path);
 
 380	if (error)
 381		goto out;
 382
 383	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 384	if (error)
 385		goto dput_and_out;
 386
 387	set_fs_pwd(current->fs, &path);
 388
 389dput_and_out:
 390	path_put(&path);
 
 
 
 
 391out:
 392	return error;
 393}
 394
 
 
 
 
 
 395SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 396{
 397	struct file *file;
 398	struct inode *inode;
 399	int error;
 400
 401	error = -EBADF;
 402	file = fget(fd);
 403	if (!file)
 404		goto out;
 405
 406	inode = file->f_path.dentry->d_inode;
 407
 408	error = -ENOTDIR;
 409	if (!S_ISDIR(inode->i_mode))
 410		goto out_putf;
 411
 412	error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 413	if (!error)
 414		set_fs_pwd(current->fs, &file->f_path);
 415out_putf:
 416	fput(file);
 417out:
 418	return error;
 419}
 420
 421SYSCALL_DEFINE1(chroot, const char __user *, filename)
 422{
 423	struct path path;
 424	int error;
 425
 426	error = user_path_dir(filename, &path);
 
 427	if (error)
 428		goto out;
 429
 430	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 431	if (error)
 432		goto dput_and_out;
 433
 434	error = -EPERM;
 435	if (!capable(CAP_SYS_CHROOT))
 436		goto dput_and_out;
 437	error = security_path_chroot(&path);
 438	if (error)
 439		goto dput_and_out;
 440
 441	set_fs_root(current->fs, &path);
 442	error = 0;
 443dput_and_out:
 444	path_put(&path);
 
 
 
 
 445out:
 446	return error;
 447}
 448
 449static int chmod_common(struct path *path, umode_t mode)
 
 
 
 
 
 450{
 451	struct inode *inode = path->dentry->d_inode;
 
 452	struct iattr newattrs;
 453	int error;
 454
 455	error = mnt_want_write(path->mnt);
 456	if (error)
 457		return error;
 458	mutex_lock(&inode->i_mutex);
 459	error = security_path_chmod(path->dentry, path->mnt, mode);
 
 460	if (error)
 461		goto out_unlock;
 462	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 463	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 464	error = notify_change(path->dentry, &newattrs);
 465out_unlock:
 466	mutex_unlock(&inode->i_mutex);
 
 
 
 
 
 467	mnt_drop_write(path->mnt);
 468	return error;
 469}
 470
 471SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 472{
 473	struct file * file;
 474	int err = -EBADF;
 475
 476	file = fget(fd);
 477	if (file) {
 478		audit_inode(NULL, file->f_path.dentry);
 479		err = chmod_common(&file->f_path, mode);
 480		fput(file);
 481	}
 482	return err;
 483}
 484
 485SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 
 
 
 
 
 486{
 487	struct path path;
 488	int error;
 489
 490	error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 
 491	if (!error) {
 492		error = chmod_common(&path, mode);
 493		path_put(&path);
 
 
 
 
 494	}
 495	return error;
 496}
 497
 498SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 
 
 
 
 
 
 499{
 500	return sys_fchmodat(AT_FDCWD, filename, mode);
 501}
 502
 503static int chown_common(struct path *path, uid_t user, gid_t group)
 504{
 505	struct inode *inode = path->dentry->d_inode;
 
 506	int error;
 507	struct iattr newattrs;
 
 
 508
 
 
 
 
 509	newattrs.ia_valid =  ATTR_CTIME;
 510	if (user != (uid_t) -1) {
 
 
 511		newattrs.ia_valid |= ATTR_UID;
 512		newattrs.ia_uid = user;
 513	}
 514	if (group != (gid_t) -1) {
 
 
 515		newattrs.ia_valid |= ATTR_GID;
 516		newattrs.ia_gid = group;
 517	}
 518	if (!S_ISDIR(inode->i_mode))
 519		newattrs.ia_valid |=
 520			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 521	mutex_lock(&inode->i_mutex);
 522	error = security_path_chown(path, user, group);
 523	if (!error)
 524		error = notify_change(path->dentry, &newattrs);
 525	mutex_unlock(&inode->i_mutex);
 526
 527	return error;
 528}
 529
 530SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 531{
 532	struct path path;
 533	int error;
 534
 535	error = user_path(filename, &path);
 536	if (error)
 537		goto out;
 538	error = mnt_want_write(path.mnt);
 539	if (error)
 540		goto out_release;
 541	error = chown_common(&path, user, group);
 542	mnt_drop_write(path.mnt);
 543out_release:
 544	path_put(&path);
 545out:
 546	return error;
 547}
 548
 549SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 550		gid_t, group, int, flag)
 551{
 552	struct path path;
 553	int error = -EINVAL;
 554	int lookup_flags;
 555
 556	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 557		goto out;
 558
 559	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 560	if (flag & AT_EMPTY_PATH)
 561		lookup_flags |= LOOKUP_EMPTY;
 
 562	error = user_path_at(dfd, filename, lookup_flags, &path);
 563	if (error)
 564		goto out;
 565	error = mnt_want_write(path.mnt);
 566	if (error)
 567		goto out_release;
 568	error = chown_common(&path, user, group);
 569	mnt_drop_write(path.mnt);
 570out_release:
 571	path_put(&path);
 
 
 
 
 572out:
 573	return error;
 574}
 575
 
 
 
 
 
 
 
 
 
 
 
 576SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 577{
 578	struct path path;
 579	int error;
 580
 581	error = user_lpath(filename, &path);
 582	if (error)
 583		goto out;
 584	error = mnt_want_write(path.mnt);
 585	if (error)
 586		goto out_release;
 587	error = chown_common(&path, user, group);
 588	mnt_drop_write(path.mnt);
 589out_release:
 590	path_put(&path);
 591out:
 592	return error;
 593}
 594
 595SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 596{
 597	struct file * file;
 598	int error = -EBADF;
 599	struct dentry * dentry;
 600
 601	file = fget(fd);
 602	if (!file)
 603		goto out;
 604
 605	error = mnt_want_write_file(file);
 606	if (error)
 607		goto out_fput;
 608	dentry = file->f_path.dentry;
 609	audit_inode(NULL, dentry);
 610	error = chown_common(&file->f_path, user, group);
 611	mnt_drop_write(file->f_path.mnt);
 612out_fput:
 613	fput(file);
 614out:
 615	return error;
 616}
 617
 618/*
 619 * You have to be very careful that these write
 620 * counts get cleaned up in error cases and
 621 * upon __fput().  This should probably never
 622 * be called outside of __dentry_open().
 623 */
 624static inline int __get_file_write_access(struct inode *inode,
 625					  struct vfsmount *mnt)
 626{
 627	int error;
 628	error = get_write_access(inode);
 629	if (error)
 630		return error;
 631	/*
 632	 * Do not take mount writer counts on
 633	 * special files since no writes to
 634	 * the mount itself will occur.
 635	 */
 636	if (!special_file(inode->i_mode)) {
 637		/*
 638		 * Balanced in __fput()
 639		 */
 640		error = mnt_want_write(mnt);
 641		if (error)
 642			put_write_access(inode);
 643	}
 644	return error;
 645}
 646
 647static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 648					struct file *f,
 649					int (*open)(struct inode *, struct file *),
 650					const struct cred *cred)
 651{
 652	static const struct file_operations empty_fops = {};
 653	struct inode *inode;
 654	int error;
 655
 656	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 657				FMODE_PREAD | FMODE_PWRITE;
 658
 659	if (unlikely(f->f_flags & O_PATH))
 
 
 
 
 
 
 
 660		f->f_mode = FMODE_PATH;
 
 
 
 661
 662	inode = dentry->d_inode;
 663	if (f->f_mode & FMODE_WRITE) {
 664		error = __get_file_write_access(inode, mnt);
 665		if (error)
 
 
 
 666			goto cleanup_file;
 667		if (!special_file(inode->i_mode))
 668			file_take_write(f);
 669	}
 670
 671	f->f_mapping = inode->i_mapping;
 672	f->f_path.dentry = dentry;
 673	f->f_path.mnt = mnt;
 674	f->f_pos = 0;
 675	file_sb_list_add(f, inode->i_sb);
 676
 677	if (unlikely(f->f_mode & FMODE_PATH)) {
 678		f->f_op = &empty_fops;
 679		return f;
 
 680	}
 681
 682	f->f_op = fops_get(inode->i_fop);
 
 
 683
 684	error = security_dentry_open(f, cred);
 685	if (error)
 686		goto cleanup_all;
 687
 688	if (!open && f->f_op)
 689		open = f->f_op->open;
 690	if (open) {
 691		error = open(inode, f);
 692		if (error)
 693			goto cleanup_all;
 694	}
 695	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 696		i_readcount_inc(inode);
 
 
 
 
 
 
 697
 
 698	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 699
 700	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 701
 702	/* NB: we're sure to have correct a_ops only after f_op->open */
 703	if (f->f_flags & O_DIRECT) {
 704		if (!f->f_mapping->a_ops ||
 705		    ((!f->f_mapping->a_ops->direct_IO) &&
 706		    (!f->f_mapping->a_ops->get_xip_mem))) {
 707			fput(f);
 708			f = ERR_PTR(-EINVAL);
 709		}
 710	}
 711
 712	return f;
 713
 714cleanup_all:
 715	fops_put(f->f_op);
 716	if (f->f_mode & FMODE_WRITE) {
 717		put_write_access(inode);
 718		if (!special_file(inode->i_mode)) {
 719			/*
 720			 * We don't consider this a real
 721			 * mnt_want/drop_write() pair
 722			 * because it all happenend right
 723			 * here, so just reset the state.
 724			 */
 725			file_reset_write(f);
 726			mnt_drop_write(mnt);
 727		}
 728	}
 729	file_sb_list_del(f);
 
 
 730	f->f_path.dentry = NULL;
 731	f->f_path.mnt = NULL;
 732cleanup_file:
 733	put_filp(f);
 734	dput(dentry);
 735	mntput(mnt);
 736	return ERR_PTR(error);
 737}
 738
 739/**
 740 * lookup_instantiate_filp - instantiates the open intent filp
 741 * @nd: pointer to nameidata
 742 * @dentry: pointer to dentry
 743 * @open: open callback
 
 
 
 744 *
 745 * Helper for filesystems that want to use lookup open intents and pass back
 746 * a fully instantiated struct file to the caller.
 747 * This function is meant to be called from within a filesystem's
 748 * lookup method.
 749 * Beware of calling it for non-regular files! Those ->open methods might block
 750 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 751 * leading to a deadlock, as nobody can open that fifo anymore, because
 752 * another process to open fifo will block on locked parent when doing lookup).
 753 * Note that in case of error, nd->intent.open.file is destroyed, but the
 754 * path information remains valid.
 755 * If the open callback is set to NULL, then the standard f_op->open()
 756 * filesystem callback is substituted.
 
 
 
 
 
 
 
 
 
 757 */
 758struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 759		int (*open)(struct inode *, struct file *))
 
 760{
 761	const struct cred *cred = current_cred();
 
 762
 763	if (IS_ERR(nd->intent.open.file))
 764		goto out;
 765	if (IS_ERR(dentry))
 766		goto out_err;
 767	nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 768					     nd->intent.open.file,
 769					     open, cred);
 770out:
 771	return nd->intent.open.file;
 772out_err:
 773	release_open_intent(nd);
 774	nd->intent.open.file = ERR_CAST(dentry);
 775	goto out;
 776}
 777EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 778
 779/**
 780 * nameidata_to_filp - convert a nameidata to an open filp.
 781 * @nd: pointer to nameidata
 782 * @flags: open flags
 
 
 
 783 *
 784 * Note that this function destroys the original nameidata
 
 
 
 
 785 */
 786struct file *nameidata_to_filp(struct nameidata *nd)
 787{
 788	const struct cred *cred = current_cred();
 789	struct file *filp;
 
 
 790
 791	/* Pick up the filp from the open intent */
 792	filp = nd->intent.open.file;
 793	nd->intent.open.file = NULL;
 794
 795	/* Has the filesystem initialised the file for us? */
 796	if (filp->f_path.dentry == NULL) {
 797		path_get(&nd->path);
 798		filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
 799				     NULL, cred);
 800	}
 801	return filp;
 802}
 
 803
 804/*
 805 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 806 * error.
 
 
 807 */
 808struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 
 
 
 
 
 
 
 
 
 
 
 
 809			 const struct cred *cred)
 810{
 811	int error;
 812	struct file *f;
 813
 814	validate_creds(cred);
 815
 816	/* We must always pass in a valid mount pointer. */
 817	BUG_ON(!mnt);
 818
 819	error = -ENFILE;
 820	f = get_empty_filp();
 821	if (f == NULL) {
 822		dput(dentry);
 823		mntput(mnt);
 824		return ERR_PTR(error);
 
 
 
 
 
 
 
 
 
 
 825	}
 826
 827	f->f_flags = flags;
 828	return __dentry_open(dentry, mnt, f, NULL, cred);
 829}
 830EXPORT_SYMBOL(dentry_open);
 831
 832static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 833{
 834	struct fdtable *fdt = files_fdtable(files);
 835	__FD_CLR(fd, fdt->open_fds);
 836	if (fd < files->next_fd)
 837		files->next_fd = fd;
 838}
 839
 840void put_unused_fd(unsigned int fd)
 841{
 842	struct files_struct *files = current->files;
 843	spin_lock(&files->file_lock);
 844	__put_unused_fd(files, fd);
 845	spin_unlock(&files->file_lock);
 846}
 847
 848EXPORT_SYMBOL(put_unused_fd);
 849
 850/*
 851 * Install a file pointer in the fd array.
 852 *
 853 * The VFS is full of places where we drop the files lock between
 854 * setting the open_fds bitmap and installing the file in the file
 855 * array.  At any such point, we are vulnerable to a dup2() race
 856 * installing a file in the array before us.  We need to detect this and
 857 * fput() the struct file we are about to overwrite in this case.
 858 *
 859 * It should never happen - if we allow dup2() do it, _really_ bad things
 860 * will follow.
 861 */
 862
 863void fd_install(unsigned int fd, struct file *file)
 864{
 865	struct files_struct *files = current->files;
 866	struct fdtable *fdt;
 867	spin_lock(&files->file_lock);
 868	fdt = files_fdtable(files);
 869	BUG_ON(fdt->fd[fd] != NULL);
 870	rcu_assign_pointer(fdt->fd[fd], file);
 871	spin_unlock(&files->file_lock);
 872}
 873
 874EXPORT_SYMBOL(fd_install);
 875
 876static inline int build_open_flags(int flags, int mode, struct open_flags *op)
 877{
 878	int lookup_flags = 0;
 879	int acc_mode;
 880
 881	if (!(flags & O_CREAT))
 882		mode = 0;
 883	op->mode = mode;
 
 884
 885	/* Must never be set by userspace */
 886	flags &= ~FMODE_NONOTIFY;
 887
 888	/*
 889	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 890	 * check for O_DSYNC if the need any syncing at all we enforce it's
 891	 * always set instead of having to deal with possibly weird behaviour
 892	 * for malicious applications setting only __O_SYNC.
 893	 */
 894	if (flags & __O_SYNC)
 895		flags |= O_DSYNC;
 896
 897	/*
 898	 * If we have O_PATH in the open flag. Then we
 899	 * cannot have anything other than the below set of flags
 900	 */
 901	if (flags & O_PATH) {
 
 
 
 
 
 902		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 903		acc_mode = 0;
 904	} else {
 905		acc_mode = MAY_OPEN | ACC_MODE(flags);
 906	}
 907
 908	op->open_flag = flags;
 909
 910	/* O_TRUNC implies we need access checks for write permissions */
 911	if (flags & O_TRUNC)
 912		acc_mode |= MAY_WRITE;
 913
 914	/* Allow the LSM permission hook to distinguish append
 915	   access from general write access. */
 916	if (flags & O_APPEND)
 917		acc_mode |= MAY_APPEND;
 918
 919	op->acc_mode = acc_mode;
 920
 921	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 922
 923	if (flags & O_CREAT) {
 924		op->intent |= LOOKUP_CREATE;
 925		if (flags & O_EXCL)
 926			op->intent |= LOOKUP_EXCL;
 927	}
 928
 929	if (flags & O_DIRECTORY)
 930		lookup_flags |= LOOKUP_DIRECTORY;
 931	if (!(flags & O_NOFOLLOW))
 932		lookup_flags |= LOOKUP_FOLLOW;
 933	return lookup_flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 934}
 935
 936/**
 937 * filp_open - open file and return file pointer
 938 *
 939 * @filename:	path to open
 940 * @flags:	open flags as per the open(2) second argument
 941 * @mode:	mode for the new file if O_CREAT is set, else ignored
 942 *
 943 * This is the helper to open a file from kernelspace if you really
 944 * have to.  But in generally you should not do this, so please move
 945 * along, nothing to see here..
 946 */
 947struct file *filp_open(const char *filename, int flags, int mode)
 948{
 949	struct open_flags op;
 950	int lookup = build_open_flags(flags, mode, &op);
 951	return do_filp_open(AT_FDCWD, filename, &op, lookup);
 
 
 
 
 
 952}
 953EXPORT_SYMBOL(filp_open);
 954
 955struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 956			    const char *filename, int flags)
 957{
 958	struct open_flags op;
 959	int lookup = build_open_flags(flags, 0, &op);
 960	if (flags & O_CREAT)
 961		return ERR_PTR(-EINVAL);
 962	if (!filename && (flags & O_DIRECTORY))
 963		if (!dentry->d_inode->i_op->lookup)
 964			return ERR_PTR(-ENOTDIR);
 965	return do_file_open_root(dentry, mnt, filename, &op, lookup);
 966}
 967EXPORT_SYMBOL(file_open_root);
 968
 969long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 970{
 971	struct open_flags op;
 972	int lookup = build_open_flags(flags, mode, &op);
 973	char *tmp = getname(filename);
 974	int fd = PTR_ERR(tmp);
 975
 976	if (!IS_ERR(tmp)) {
 977		fd = get_unused_fd_flags(flags);
 978		if (fd >= 0) {
 979			struct file *f = do_filp_open(dfd, tmp, &op, lookup);
 980			if (IS_ERR(f)) {
 981				put_unused_fd(fd);
 982				fd = PTR_ERR(f);
 983			} else {
 984				fsnotify_open(f);
 985				fd_install(fd, f);
 986			}
 
 
 
 
 987		}
 988		putname(tmp);
 989	}
 
 990	return fd;
 991}
 992
 993SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
 994{
 995	long ret;
 996
 997	if (force_o_largefile())
 998		flags |= O_LARGEFILE;
 999
1000	ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1001	/* avoid REGPARM breakage on x86: */
1002	asmlinkage_protect(3, ret, filename, flags, mode);
1003	return ret;
1004}
1005
1006SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1007		int, mode)
1008{
1009	long ret;
1010
1011	if (force_o_largefile())
1012		flags |= O_LARGEFILE;
1013
1014	ret = do_sys_open(dfd, filename, flags, mode);
1015	/* avoid REGPARM breakage on x86: */
1016	asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1017	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1018}
 
1019
1020#ifndef __alpha__
1021
1022/*
1023 * For backward compatibility?  Maybe this should be moved
1024 * into arch/i386 instead?
1025 */
1026SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1027{
1028	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1029}
1030
1031#endif
1032
1033/*
1034 * "id" is the POSIX thread ID. We use the
1035 * files pointer for this..
1036 */
1037int filp_close(struct file *filp, fl_owner_t id)
1038{
1039	int retval = 0;
1040
1041	if (!file_count(filp)) {
1042		printk(KERN_ERR "VFS: Close: file count is 0\n");
1043		return 0;
1044	}
1045
1046	if (filp->f_op && filp->f_op->flush)
1047		retval = filp->f_op->flush(filp, id);
1048
1049	if (likely(!(filp->f_mode & FMODE_PATH))) {
1050		dnotify_flush(filp, id);
1051		locks_remove_posix(filp, id);
1052	}
1053	fput(filp);
1054	return retval;
1055}
1056
1057EXPORT_SYMBOL(filp_close);
1058
1059/*
1060 * Careful here! We test whether the file pointer is NULL before
1061 * releasing the fd. This ensures that one clone task can't release
1062 * an fd while another clone is opening it.
1063 */
1064SYSCALL_DEFINE1(close, unsigned int, fd)
1065{
1066	struct file * filp;
1067	struct files_struct *files = current->files;
1068	struct fdtable *fdt;
1069	int retval;
1070
1071	spin_lock(&files->file_lock);
1072	fdt = files_fdtable(files);
1073	if (fd >= fdt->max_fds)
1074		goto out_unlock;
1075	filp = fdt->fd[fd];
1076	if (!filp)
1077		goto out_unlock;
1078	rcu_assign_pointer(fdt->fd[fd], NULL);
1079	FD_CLR(fd, fdt->close_on_exec);
1080	__put_unused_fd(files, fd);
1081	spin_unlock(&files->file_lock);
1082	retval = filp_close(filp, files);
1083
1084	/* can't restart close syscall because file table entry was cleared */
1085	if (unlikely(retval == -ERESTARTSYS ||
1086		     retval == -ERESTARTNOINTR ||
1087		     retval == -ERESTARTNOHAND ||
1088		     retval == -ERESTART_RESTARTBLOCK))
1089		retval = -EINTR;
1090
1091	return retval;
1092
1093out_unlock:
1094	spin_unlock(&files->file_lock);
1095	return -EBADF;
1096}
1097EXPORT_SYMBOL(sys_close);
1098
1099/*
1100 * This routine simulates a hangup on the tty, to arrange that users
1101 * are given clean terminals at login time.
1102 */
1103SYSCALL_DEFINE0(vhangup)
1104{
1105	if (capable(CAP_SYS_TTY_CONFIG)) {
1106		tty_vhangup_self();
1107		return 0;
1108	}
1109	return -EPERM;
1110}
1111
1112/*
1113 * Called when an inode is about to be open.
1114 * We use this to disallow opening large files on 32bit systems if
1115 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1116 * on this flag in sys_open.
1117 */
1118int generic_file_open(struct inode * inode, struct file * filp)
1119{
1120	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1121		return -EOVERFLOW;
1122	return 0;
1123}
1124
1125EXPORT_SYMBOL(generic_file_open);
1126
1127/*
1128 * This is used by subsystems that don't want seekable
1129 * file descriptors. The function is not supposed to ever fail, the only
1130 * reason it returns an 'int' and not 'void' is so that it can be plugged
1131 * directly into file_operations structure.
1132 */
1133int nonseekable_open(struct inode *inode, struct file *filp)
1134{
1135	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1136	return 0;
1137}
1138
1139EXPORT_SYMBOL(nonseekable_open);