/fs/open.c

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