Linux Audio

Check our new training course

Loading...
v4.6
 
   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#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(struct path *path, loff_t length)
  69{
  70	struct inode *inode;
  71	long error;
  72
  73	inode = path->dentry->d_inode;
  74
  75	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  76	if (S_ISDIR(inode->i_mode))
  77		return -EISDIR;
  78	if (!S_ISREG(inode->i_mode))
  79		return -EINVAL;
  80
  81	error = mnt_want_write(path->mnt);
  82	if (error)
  83		goto out;
  84
  85	error = inode_permission(inode, MAY_WRITE);
  86	if (error)
  87		goto mnt_drop_write_and_out;
  88
  89	error = -EPERM;
  90	if (IS_APPEND(inode))
  91		goto mnt_drop_write_and_out;
  92
  93	error = get_write_access(inode);
  94	if (error)
  95		goto mnt_drop_write_and_out;
  96
  97	/*
  98	 * Make sure that there are no leases.  get_write_access() protects
  99	 * against the truncate racing with a lease-granting setlease().
 100	 */
 101	error = break_lease(inode, O_WRONLY);
 102	if (error)
 103		goto put_write_and_out;
 104
 105	error = locks_verify_truncate(inode, NULL, length);
 106	if (!error)
 107		error = security_path_truncate(path);
 108	if (!error)
 109		error = do_truncate(path->dentry, length, 0, NULL);
 110
 111put_write_and_out:
 112	put_write_access(inode);
 113mnt_drop_write_and_out:
 114	mnt_drop_write(path->mnt);
 115out:
 116	return error;
 117}
 118EXPORT_SYMBOL_GPL(vfs_truncate);
 119
 120static long do_sys_truncate(const char __user *pathname, loff_t length)
 121{
 122	unsigned int lookup_flags = LOOKUP_FOLLOW;
 123	struct path path;
 124	int error;
 125
 126	if (length < 0)	/* sorry, but loff_t says... */
 127		return -EINVAL;
 128
 129retry:
 130	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 131	if (!error) {
 132		error = vfs_truncate(&path, length);
 133		path_put(&path);
 134	}
 135	if (retry_estale(error, lookup_flags)) {
 136		lookup_flags |= LOOKUP_REVAL;
 137		goto retry;
 138	}
 139	return error;
 140}
 141
 142SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 143{
 144	return do_sys_truncate(path, length);
 145}
 146
 147#ifdef CONFIG_COMPAT
 148COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 149{
 150	return do_sys_truncate(path, length);
 151}
 152#endif
 153
 154static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 155{
 156	struct inode *inode;
 157	struct dentry *dentry;
 158	struct fd f;
 159	int error;
 160
 161	error = -EINVAL;
 162	if (length < 0)
 163		goto out;
 164	error = -EBADF;
 165	f = fdget(fd);
 166	if (!f.file)
 167		goto out;
 168
 169	/* explicitly opened as large or we are on 64-bit box */
 170	if (f.file->f_flags & O_LARGEFILE)
 171		small = 0;
 172
 173	dentry = f.file->f_path.dentry;
 174	inode = dentry->d_inode;
 175	error = -EINVAL;
 176	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 177		goto out_putf;
 178
 179	error = -EINVAL;
 180	/* Cannot ftruncate over 2^31 bytes without large file support */
 181	if (small && length > MAX_NON_LFS)
 182		goto out_putf;
 183
 184	error = -EPERM;
 185	if (IS_APPEND(inode))
 
 186		goto out_putf;
 187
 188	sb_start_write(inode->i_sb);
 189	error = locks_verify_truncate(inode, f.file, length);
 190	if (!error)
 191		error = security_path_truncate(&f.file->f_path);
 192	if (!error)
 193		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 194	sb_end_write(inode->i_sb);
 195out_putf:
 196	fdput(f);
 197out:
 198	return error;
 199}
 200
 201SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 202{
 203	return do_sys_ftruncate(fd, length, 1);
 204}
 205
 206#ifdef CONFIG_COMPAT
 207COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 208{
 209	return do_sys_ftruncate(fd, length, 1);
 210}
 211#endif
 212
 213/* LFS versions of truncate are only needed on 32 bit machines */
 214#if BITS_PER_LONG == 32
 215SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 216{
 217	return do_sys_truncate(path, length);
 218}
 219
 220SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 221{
 222	return do_sys_ftruncate(fd, length, 0);
 223}
 224#endif /* BITS_PER_LONG == 32 */
 225
 226
 227int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 228{
 229	struct inode *inode = file_inode(file);
 230	long ret;
 231
 232	if (offset < 0 || len <= 0)
 233		return -EINVAL;
 234
 235	/* Return error if mode is not supported */
 236	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 237		return -EOPNOTSUPP;
 238
 239	/* Punch hole and zero range are mutually exclusive */
 240	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 241	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 242		return -EOPNOTSUPP;
 243
 244	/* Punch hole must have keep size set */
 245	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 246	    !(mode & FALLOC_FL_KEEP_SIZE))
 247		return -EOPNOTSUPP;
 248
 249	/* Collapse range should only be used exclusively. */
 250	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 251	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 252		return -EINVAL;
 253
 254	/* Insert range should only be used exclusively. */
 255	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 256	    (mode & ~FALLOC_FL_INSERT_RANGE))
 257		return -EINVAL;
 258
 
 
 
 
 
 259	if (!(file->f_mode & FMODE_WRITE))
 260		return -EBADF;
 261
 262	/*
 263	 * We can only allow pure fallocate on append only files
 264	 */
 265	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 266		return -EPERM;
 267
 268	if (IS_IMMUTABLE(inode))
 269		return -EPERM;
 270
 271	/*
 272	 * We cannot allow any fallocate operation on an active swapfile
 273	 */
 274	if (IS_SWAPFILE(inode))
 275		return -ETXTBSY;
 276
 277	/*
 278	 * Revalidate the write permissions, in case security policy has
 279	 * changed since the files were opened.
 280	 */
 281	ret = security_file_permission(file, MAY_WRITE);
 282	if (ret)
 283		return ret;
 284
 285	if (S_ISFIFO(inode->i_mode))
 286		return -ESPIPE;
 287
 288	/*
 289	 * Let individual file system decide if it supports preallocation
 290	 * for directories or not.
 291	 */
 292	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 293		return -ENODEV;
 294
 295	/* Check for wrap through zero too */
 296	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 297		return -EFBIG;
 298
 299	if (!file->f_op->fallocate)
 300		return -EOPNOTSUPP;
 301
 302	sb_start_write(inode->i_sb);
 303	ret = file->f_op->fallocate(file, mode, offset, len);
 304
 305	/*
 306	 * Create inotify and fanotify events.
 307	 *
 308	 * To keep the logic simple always create events if fallocate succeeds.
 309	 * This implies that events are even created if the file size remains
 310	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 311	 */
 312	if (ret == 0)
 313		fsnotify_modify(file);
 314
 315	sb_end_write(inode->i_sb);
 316	return ret;
 317}
 318EXPORT_SYMBOL_GPL(vfs_fallocate);
 319
 320SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 321{
 322	struct fd f = fdget(fd);
 323	int error = -EBADF;
 324
 325	if (f.file) {
 326		error = vfs_fallocate(f.file, mode, offset, len);
 327		fdput(f);
 328	}
 329	return error;
 330}
 331
 
 
 
 
 
 332/*
 333 * access() needs to use the real uid/gid, not the effective uid/gid.
 334 * We do this by temporarily clearing all FS-related capabilities and
 335 * switching the fsuid/fsgid around to the real ones.
 336 */
 337SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 338{
 339	const struct cred *old_cred;
 340	struct cred *override_cred;
 341	struct path path;
 342	struct inode *inode;
 343	int res;
 344	unsigned int lookup_flags = LOOKUP_FOLLOW;
 345
 346	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 347		return -EINVAL;
 348
 349	override_cred = prepare_creds();
 350	if (!override_cred)
 351		return -ENOMEM;
 352
 353	override_cred->fsuid = override_cred->uid;
 354	override_cred->fsgid = override_cred->gid;
 355
 356	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 357		/* Clear the capabilities if we switch to a non-root user */
 358		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 359		if (!uid_eq(override_cred->uid, root_uid))
 360			cap_clear(override_cred->cap_effective);
 361		else
 362			override_cred->cap_effective =
 363				override_cred->cap_permitted;
 364	}
 365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 366	old_cred = override_creds(override_cred);
 367retry:
 368	res = user_path_at(dfd, filename, lookup_flags, &path);
 369	if (res)
 370		goto out;
 371
 372	inode = d_backing_inode(path.dentry);
 373
 374	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 375		/*
 376		 * MAY_EXEC on regular files is denied if the fs is mounted
 377		 * with the "noexec" flag.
 378		 */
 379		res = -EACCES;
 380		if (path_noexec(&path))
 381			goto out_path_release;
 382	}
 383
 384	res = inode_permission(inode, mode | MAY_ACCESS);
 385	/* SuS v2 requires we report a read only fs too */
 386	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 387		goto out_path_release;
 388	/*
 389	 * This is a rare case where using __mnt_is_readonly()
 390	 * is OK without a mnt_want/drop_write() pair.  Since
 391	 * no actual write to the fs is performed here, we do
 392	 * not need to telegraph to that to anyone.
 393	 *
 394	 * By doing this, we accept that this access is
 395	 * inherently racy and know that the fs may change
 396	 * state before we even see this result.
 397	 */
 398	if (__mnt_is_readonly(path.mnt))
 399		res = -EROFS;
 400
 401out_path_release:
 402	path_put(&path);
 403	if (retry_estale(res, lookup_flags)) {
 404		lookup_flags |= LOOKUP_REVAL;
 405		goto retry;
 406	}
 407out:
 408	revert_creds(old_cred);
 409	put_cred(override_cred);
 410	return res;
 411}
 412
 
 
 
 
 
 413SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 414{
 415	return sys_faccessat(AT_FDCWD, filename, mode);
 416}
 417
 418SYSCALL_DEFINE1(chdir, const char __user *, filename)
 419{
 420	struct path path;
 421	int error;
 422	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 423retry:
 424	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 425	if (error)
 426		goto out;
 427
 428	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 429	if (error)
 430		goto dput_and_out;
 431
 432	set_fs_pwd(current->fs, &path);
 433
 434dput_and_out:
 435	path_put(&path);
 436	if (retry_estale(error, lookup_flags)) {
 437		lookup_flags |= LOOKUP_REVAL;
 438		goto retry;
 439	}
 440out:
 441	return error;
 442}
 443
 
 
 
 
 
 444SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 445{
 446	struct fd f = fdget_raw(fd);
 447	struct inode *inode;
 448	int error = -EBADF;
 449
 450	error = -EBADF;
 451	if (!f.file)
 452		goto out;
 453
 454	inode = file_inode(f.file);
 455
 456	error = -ENOTDIR;
 457	if (!S_ISDIR(inode->i_mode))
 458		goto out_putf;
 459
 460	error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 461	if (!error)
 462		set_fs_pwd(current->fs, &f.file->f_path);
 463out_putf:
 464	fdput(f);
 465out:
 466	return error;
 467}
 468
 469SYSCALL_DEFINE1(chroot, const char __user *, filename)
 470{
 471	struct path path;
 472	int error;
 473	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 474retry:
 475	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 476	if (error)
 477		goto out;
 478
 479	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 480	if (error)
 481		goto dput_and_out;
 482
 483	error = -EPERM;
 484	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 485		goto dput_and_out;
 486	error = security_path_chroot(&path);
 487	if (error)
 488		goto dput_and_out;
 489
 490	set_fs_root(current->fs, &path);
 491	error = 0;
 492dput_and_out:
 493	path_put(&path);
 494	if (retry_estale(error, lookup_flags)) {
 495		lookup_flags |= LOOKUP_REVAL;
 496		goto retry;
 497	}
 498out:
 499	return error;
 500}
 501
 502static int chmod_common(struct path *path, umode_t mode)
 
 
 
 
 
 503{
 504	struct inode *inode = path->dentry->d_inode;
 505	struct inode *delegated_inode = NULL;
 506	struct iattr newattrs;
 507	int error;
 508
 509	error = mnt_want_write(path->mnt);
 510	if (error)
 511		return error;
 512retry_deleg:
 513	inode_lock(inode);
 514	error = security_path_chmod(path, mode);
 515	if (error)
 516		goto out_unlock;
 517	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 518	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 519	error = notify_change(path->dentry, &newattrs, &delegated_inode);
 520out_unlock:
 521	inode_unlock(inode);
 522	if (delegated_inode) {
 523		error = break_deleg_wait(&delegated_inode);
 524		if (!error)
 525			goto retry_deleg;
 526	}
 527	mnt_drop_write(path->mnt);
 528	return error;
 529}
 530
 531SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 532{
 533	struct fd f = fdget(fd);
 534	int err = -EBADF;
 535
 536	if (f.file) {
 537		audit_file(f.file);
 538		err = chmod_common(&f.file->f_path, mode);
 539		fdput(f);
 540	}
 541	return err;
 542}
 543
 544SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 
 
 
 
 
 545{
 546	struct path path;
 547	int error;
 548	unsigned int lookup_flags = LOOKUP_FOLLOW;
 549retry:
 550	error = user_path_at(dfd, filename, lookup_flags, &path);
 551	if (!error) {
 552		error = chmod_common(&path, mode);
 553		path_put(&path);
 554		if (retry_estale(error, lookup_flags)) {
 555			lookup_flags |= LOOKUP_REVAL;
 556			goto retry;
 557		}
 558	}
 559	return error;
 560}
 561
 
 
 
 
 
 
 562SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 563{
 564	return sys_fchmodat(AT_FDCWD, filename, mode);
 565}
 566
 567static int chown_common(struct path *path, uid_t user, gid_t group)
 568{
 569	struct inode *inode = path->dentry->d_inode;
 570	struct inode *delegated_inode = NULL;
 571	int error;
 572	struct iattr newattrs;
 573	kuid_t uid;
 574	kgid_t gid;
 575
 576	uid = make_kuid(current_user_ns(), user);
 577	gid = make_kgid(current_user_ns(), group);
 578
 579retry_deleg:
 580	newattrs.ia_valid =  ATTR_CTIME;
 581	if (user != (uid_t) -1) {
 582		if (!uid_valid(uid))
 583			return -EINVAL;
 584		newattrs.ia_valid |= ATTR_UID;
 585		newattrs.ia_uid = uid;
 586	}
 587	if (group != (gid_t) -1) {
 588		if (!gid_valid(gid))
 589			return -EINVAL;
 590		newattrs.ia_valid |= ATTR_GID;
 591		newattrs.ia_gid = gid;
 592	}
 593	if (!S_ISDIR(inode->i_mode))
 594		newattrs.ia_valid |=
 595			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 596	inode_lock(inode);
 597	error = security_path_chown(path, uid, gid);
 598	if (!error)
 599		error = notify_change(path->dentry, &newattrs, &delegated_inode);
 600	inode_unlock(inode);
 601	if (delegated_inode) {
 602		error = break_deleg_wait(&delegated_inode);
 603		if (!error)
 604			goto retry_deleg;
 605	}
 606	return error;
 607}
 608
 609SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 610		gid_t, group, int, flag)
 611{
 612	struct path path;
 613	int error = -EINVAL;
 614	int lookup_flags;
 615
 616	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 617		goto out;
 618
 619	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 620	if (flag & AT_EMPTY_PATH)
 621		lookup_flags |= LOOKUP_EMPTY;
 622retry:
 623	error = user_path_at(dfd, filename, lookup_flags, &path);
 624	if (error)
 625		goto out;
 626	error = mnt_want_write(path.mnt);
 627	if (error)
 628		goto out_release;
 629	error = chown_common(&path, user, group);
 630	mnt_drop_write(path.mnt);
 631out_release:
 632	path_put(&path);
 633	if (retry_estale(error, lookup_flags)) {
 634		lookup_flags |= LOOKUP_REVAL;
 635		goto retry;
 636	}
 637out:
 638	return error;
 639}
 640
 
 
 
 
 
 
 641SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 642{
 643	return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 644}
 645
 646SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 647{
 648	return sys_fchownat(AT_FDCWD, filename, user, group,
 649			    AT_SYMLINK_NOFOLLOW);
 650}
 651
 652SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 653{
 654	struct fd f = fdget(fd);
 655	int error = -EBADF;
 656
 657	if (!f.file)
 658		goto out;
 659
 660	error = mnt_want_write_file(f.file);
 661	if (error)
 662		goto out_fput;
 663	audit_file(f.file);
 664	error = chown_common(&f.file->f_path, user, group);
 665	mnt_drop_write_file(f.file);
 666out_fput:
 667	fdput(f);
 668out:
 669	return error;
 670}
 671
 672int open_check_o_direct(struct file *f)
 673{
 674	/* NB: we're sure to have correct a_ops only after f_op->open */
 675	if (f->f_flags & O_DIRECT) {
 676		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 677			return -EINVAL;
 678	}
 679	return 0;
 680}
 681
 682static int do_dentry_open(struct file *f,
 683			  struct inode *inode,
 684			  int (*open)(struct inode *, struct file *),
 685			  const struct cred *cred)
 686{
 687	static const struct file_operations empty_fops = {};
 688	int error;
 689
 690	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 691				FMODE_PREAD | FMODE_PWRITE;
 692
 693	path_get(&f->f_path);
 694	f->f_inode = inode;
 695	f->f_mapping = inode->i_mapping;
 696
 
 
 
 697	if (unlikely(f->f_flags & O_PATH)) {
 698		f->f_mode = FMODE_PATH;
 699		f->f_op = &empty_fops;
 700		return 0;
 701	}
 702
 
 
 
 
 
 
 703	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 704		error = get_write_access(inode);
 705		if (unlikely(error))
 706			goto cleanup_file;
 707		error = __mnt_want_write(f->f_path.mnt);
 708		if (unlikely(error)) {
 709			put_write_access(inode);
 710			goto cleanup_file;
 711		}
 712		f->f_mode |= FMODE_WRITER;
 713	}
 714
 715	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 716	if (S_ISREG(inode->i_mode))
 717		f->f_mode |= FMODE_ATOMIC_POS;
 718
 719	f->f_op = fops_get(inode->i_fop);
 720	if (unlikely(WARN_ON(!f->f_op))) {
 721		error = -ENODEV;
 722		goto cleanup_all;
 723	}
 724
 725	error = security_file_open(f, cred);
 726	if (error)
 727		goto cleanup_all;
 728
 729	error = break_lease(inode, f->f_flags);
 730	if (error)
 731		goto cleanup_all;
 732
 
 
 733	if (!open)
 734		open = f->f_op->open;
 735	if (open) {
 736		error = open(inode, f);
 737		if (error)
 738			goto cleanup_all;
 739	}
 
 740	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 741		i_readcount_inc(inode);
 742	if ((f->f_mode & FMODE_READ) &&
 743	     likely(f->f_op->read || f->f_op->read_iter))
 744		f->f_mode |= FMODE_CAN_READ;
 745	if ((f->f_mode & FMODE_WRITE) &&
 746	     likely(f->f_op->write || f->f_op->write_iter))
 747		f->f_mode |= FMODE_CAN_WRITE;
 748
 
 749	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 750
 751	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 752
 
 
 
 
 
 
 
 
 
 
 
 
 
 753	return 0;
 754
 755cleanup_all:
 
 
 756	fops_put(f->f_op);
 757	if (f->f_mode & FMODE_WRITER) {
 758		put_write_access(inode);
 759		__mnt_drop_write(f->f_path.mnt);
 760	}
 761cleanup_file:
 762	path_put(&f->f_path);
 763	f->f_path.mnt = NULL;
 764	f->f_path.dentry = NULL;
 765	f->f_inode = NULL;
 766	return error;
 767}
 768
 769/**
 770 * finish_open - finish opening a file
 771 * @file: file pointer
 772 * @dentry: pointer to dentry
 773 * @open: open callback
 774 * @opened: state of open
 775 *
 776 * This can be used to finish opening a file passed to i_op->atomic_open().
 777 *
 778 * If the open callback is set to NULL, then the standard f_op->open()
 779 * filesystem callback is substituted.
 780 *
 781 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 782 * the return value of d_splice_alias(), then the caller needs to perform dput()
 783 * on it after finish_open().
 784 *
 785 * On successful return @file is a fully instantiated open file.  After this, if
 786 * an error occurs in ->atomic_open(), it needs to clean up with fput().
 787 *
 788 * Returns zero on success or -errno if the open failed.
 789 */
 790int finish_open(struct file *file, struct dentry *dentry,
 791		int (*open)(struct inode *, struct file *),
 792		int *opened)
 793{
 794	int error;
 795	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 796
 797	file->f_path.dentry = dentry;
 798	error = do_dentry_open(file, d_backing_inode(dentry), open,
 799			       current_cred());
 800	if (!error)
 801		*opened |= FILE_OPENED;
 802
 803	return error;
 804}
 805EXPORT_SYMBOL(finish_open);
 806
 807/**
 808 * finish_no_open - finish ->atomic_open() without opening the file
 809 *
 810 * @file: file pointer
 811 * @dentry: dentry or NULL (as returned from ->lookup())
 812 *
 813 * This can be used to set the result of a successful lookup in ->atomic_open().
 814 *
 815 * NB: unlike finish_open() this function does consume the dentry reference and
 816 * the caller need not dput() it.
 817 *
 818 * Returns "1" which must be the return value of ->atomic_open() after having
 819 * called this function.
 820 */
 821int finish_no_open(struct file *file, struct dentry *dentry)
 822{
 823	file->f_path.dentry = dentry;
 824	return 1;
 825}
 826EXPORT_SYMBOL(finish_no_open);
 827
 828char *file_path(struct file *filp, char *buf, int buflen)
 829{
 830	return d_path(&filp->f_path, buf, buflen);
 831}
 832EXPORT_SYMBOL(file_path);
 833
 834/**
 835 * vfs_open - open the file at the given path
 836 * @path: path to open
 837 * @file: newly allocated file with f_flag initialized
 838 * @cred: credentials to use
 839 */
 840int vfs_open(const struct path *path, struct file *file,
 841	     const struct cred *cred)
 842{
 843	struct inode *inode = vfs_select_inode(path->dentry, file->f_flags);
 844
 845	if (IS_ERR(inode))
 846		return PTR_ERR(inode);
 847
 848	file->f_path = *path;
 849	return do_dentry_open(file, inode, NULL, cred);
 850}
 851
 852struct file *dentry_open(const struct path *path, int flags,
 853			 const struct cred *cred)
 854{
 855	int error;
 856	struct file *f;
 857
 858	validate_creds(cred);
 859
 860	/* We must always pass in a valid mount pointer. */
 861	BUG_ON(!path->mnt);
 862
 863	f = get_empty_filp();
 864	if (!IS_ERR(f)) {
 865		f->f_flags = flags;
 866		error = vfs_open(path, f, cred);
 867		if (!error) {
 868			/* from now on we need fput() to dispose of f */
 869			error = open_check_o_direct(f);
 870			if (error) {
 871				fput(f);
 872				f = ERR_PTR(error);
 873			}
 874		} else { 
 875			put_filp(f);
 876			f = ERR_PTR(error);
 877		}
 878	}
 879	return f;
 880}
 881EXPORT_SYMBOL(dentry_open);
 882
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 883static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 884{
 885	int lookup_flags = 0;
 886	int acc_mode = ACC_MODE(flags);
 887
 
 
 
 
 
 
 888	if (flags & (O_CREAT | __O_TMPFILE))
 889		op->mode = (mode & S_IALLUGO) | S_IFREG;
 890	else
 891		op->mode = 0;
 892
 893	/* Must never be set by userspace */
 894	flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 895
 896	/*
 897	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 898	 * check for O_DSYNC if the need any syncing at all we enforce it's
 899	 * always set instead of having to deal with possibly weird behaviour
 900	 * for malicious applications setting only __O_SYNC.
 901	 */
 902	if (flags & __O_SYNC)
 903		flags |= O_DSYNC;
 904
 905	if (flags & __O_TMPFILE) {
 906		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 907			return -EINVAL;
 908		if (!(acc_mode & MAY_WRITE))
 909			return -EINVAL;
 910	} else if (flags & O_PATH) {
 911		/*
 912		 * If we have O_PATH in the open flag. Then we
 913		 * cannot have anything other than the below set of flags
 914		 */
 915		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 916		acc_mode = 0;
 917	}
 918
 919	op->open_flag = flags;
 920
 921	/* O_TRUNC implies we need access checks for write permissions */
 922	if (flags & O_TRUNC)
 923		acc_mode |= MAY_WRITE;
 924
 925	/* Allow the LSM permission hook to distinguish append
 926	   access from general write access. */
 927	if (flags & O_APPEND)
 928		acc_mode |= MAY_APPEND;
 929
 930	op->acc_mode = acc_mode;
 931
 932	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 933
 934	if (flags & O_CREAT) {
 935		op->intent |= LOOKUP_CREATE;
 936		if (flags & O_EXCL)
 937			op->intent |= LOOKUP_EXCL;
 938	}
 939
 940	if (flags & O_DIRECTORY)
 941		lookup_flags |= LOOKUP_DIRECTORY;
 942	if (!(flags & O_NOFOLLOW))
 943		lookup_flags |= LOOKUP_FOLLOW;
 944	op->lookup_flags = lookup_flags;
 945	return 0;
 946}
 947
 948/**
 949 * file_open_name - open file and return file pointer
 950 *
 951 * @name:	struct filename containing path to open
 952 * @flags:	open flags as per the open(2) second argument
 953 * @mode:	mode for the new file if O_CREAT is set, else ignored
 954 *
 955 * This is the helper to open a file from kernelspace if you really
 956 * have to.  But in generally you should not do this, so please move
 957 * along, nothing to see here..
 958 */
 959struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 960{
 961	struct open_flags op;
 962	int err = build_open_flags(flags, mode, &op);
 963	return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 964}
 965
 966/**
 967 * filp_open - open file and return file pointer
 968 *
 969 * @filename:	path to open
 970 * @flags:	open flags as per the open(2) second argument
 971 * @mode:	mode for the new file if O_CREAT is set, else ignored
 972 *
 973 * This is the helper to open a file from kernelspace if you really
 974 * have to.  But in generally you should not do this, so please move
 975 * along, nothing to see here..
 976 */
 977struct file *filp_open(const char *filename, int flags, umode_t mode)
 978{
 979	struct filename *name = getname_kernel(filename);
 980	struct file *file = ERR_CAST(name);
 981	
 982	if (!IS_ERR(name)) {
 983		file = file_open_name(name, flags, mode);
 984		putname(name);
 985	}
 986	return file;
 987}
 988EXPORT_SYMBOL(filp_open);
 989
 990struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 991			    const char *filename, int flags, umode_t mode)
 992{
 993	struct open_flags op;
 994	int err = build_open_flags(flags, mode, &op);
 995	if (err)
 996		return ERR_PTR(err);
 997	return do_file_open_root(dentry, mnt, filename, &op);
 998}
 999EXPORT_SYMBOL(file_open_root);
1000
1001long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1002{
1003	struct open_flags op;
1004	int fd = build_open_flags(flags, mode, &op);
1005	struct filename *tmp;
1006
1007	if (fd)
1008		return fd;
1009
1010	tmp = getname(filename);
1011	if (IS_ERR(tmp))
1012		return PTR_ERR(tmp);
1013
1014	fd = get_unused_fd_flags(flags);
1015	if (fd >= 0) {
1016		struct file *f = do_filp_open(dfd, tmp, &op);
1017		if (IS_ERR(f)) {
1018			put_unused_fd(fd);
1019			fd = PTR_ERR(f);
1020		} else {
1021			fsnotify_open(f);
1022			fd_install(fd, f);
1023		}
1024	}
1025	putname(tmp);
1026	return fd;
1027}
1028
1029SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1030{
1031	if (force_o_largefile())
1032		flags |= O_LARGEFILE;
1033
1034	return do_sys_open(AT_FDCWD, filename, flags, mode);
1035}
1036
1037SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1038		umode_t, mode)
1039{
1040	if (force_o_largefile())
1041		flags |= O_LARGEFILE;
1042
1043	return do_sys_open(dfd, filename, flags, mode);
1044}
1045
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1046#ifndef __alpha__
1047
1048/*
1049 * For backward compatibility?  Maybe this should be moved
1050 * into arch/i386 instead?
1051 */
1052SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1053{
1054	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1055}
1056
1057#endif
1058
1059/*
1060 * "id" is the POSIX thread ID. We use the
1061 * files pointer for this..
1062 */
1063int filp_close(struct file *filp, fl_owner_t id)
1064{
1065	int retval = 0;
1066
1067	if (!file_count(filp)) {
1068		printk(KERN_ERR "VFS: Close: file count is 0\n");
1069		return 0;
1070	}
1071
1072	if (filp->f_op->flush)
1073		retval = filp->f_op->flush(filp, id);
1074
1075	if (likely(!(filp->f_mode & FMODE_PATH))) {
1076		dnotify_flush(filp, id);
1077		locks_remove_posix(filp, id);
1078	}
1079	fput(filp);
1080	return retval;
1081}
1082
1083EXPORT_SYMBOL(filp_close);
1084
1085/*
1086 * Careful here! We test whether the file pointer is NULL before
1087 * releasing the fd. This ensures that one clone task can't release
1088 * an fd while another clone is opening it.
1089 */
1090SYSCALL_DEFINE1(close, unsigned int, fd)
1091{
1092	int retval = __close_fd(current->files, fd);
1093
1094	/* can't restart close syscall because file table entry was cleared */
1095	if (unlikely(retval == -ERESTARTSYS ||
1096		     retval == -ERESTARTNOINTR ||
1097		     retval == -ERESTARTNOHAND ||
1098		     retval == -ERESTART_RESTARTBLOCK))
1099		retval = -EINTR;
1100
1101	return retval;
1102}
1103EXPORT_SYMBOL(sys_close);
1104
1105/*
1106 * This routine simulates a hangup on the tty, to arrange that users
1107 * are given clean terminals at login time.
1108 */
1109SYSCALL_DEFINE0(vhangup)
1110{
1111	if (capable(CAP_SYS_TTY_CONFIG)) {
1112		tty_vhangup_self();
1113		return 0;
1114	}
1115	return -EPERM;
1116}
1117
1118/*
1119 * Called when an inode is about to be open.
1120 * We use this to disallow opening large files on 32bit systems if
1121 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1122 * on this flag in sys_open.
1123 */
1124int generic_file_open(struct inode * inode, struct file * filp)
1125{
1126	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1127		return -EOVERFLOW;
1128	return 0;
1129}
1130
1131EXPORT_SYMBOL(generic_file_open);
1132
1133/*
1134 * This is used by subsystems that don't want seekable
1135 * file descriptors. The function is not supposed to ever fail, the only
1136 * reason it returns an 'int' and not 'void' is so that it can be plugged
1137 * directly into file_operations structure.
1138 */
1139int nonseekable_open(struct inode *inode, struct file *filp)
1140{
1141	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1142	return 0;
1143}
1144
1145EXPORT_SYMBOL(nonseekable_open);
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/open.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8#include <linux/string.h>
   9#include <linux/mm.h>
  10#include <linux/file.h>
  11#include <linux/fdtable.h>
  12#include <linux/fsnotify.h>
  13#include <linux/module.h>
  14#include <linux/tty.h>
  15#include <linux/namei.h>
  16#include <linux/backing-dev.h>
  17#include <linux/capability.h>
  18#include <linux/securebits.h>
  19#include <linux/security.h>
  20#include <linux/mount.h>
  21#include <linux/fcntl.h>
  22#include <linux/slab.h>
  23#include <linux/uaccess.h>
  24#include <linux/fs.h>
  25#include <linux/personality.h>
  26#include <linux/pagemap.h>
  27#include <linux/syscalls.h>
  28#include <linux/rcupdate.h>
  29#include <linux/audit.h>
  30#include <linux/falloc.h>
  31#include <linux/fs_struct.h>
  32#include <linux/ima.h>
  33#include <linux/dnotify.h>
  34#include <linux/compat.h>
  35
  36#include "internal.h"
  37
  38int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  39	struct file *filp)
  40{
  41	int ret;
  42	struct iattr newattrs;
  43
  44	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  45	if (length < 0)
  46		return -EINVAL;
  47
  48	newattrs.ia_size = length;
  49	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  50	if (filp) {
  51		newattrs.ia_file = filp;
  52		newattrs.ia_valid |= ATTR_FILE;
  53	}
  54
  55	/* Remove suid, sgid, and file capabilities on truncate too */
  56	ret = dentry_needs_remove_privs(dentry);
  57	if (ret < 0)
  58		return ret;
  59	if (ret)
  60		newattrs.ia_valid |= ret | ATTR_FORCE;
  61
  62	inode_lock(dentry->d_inode);
  63	/* Note any delegations or leases have already been broken: */
  64	ret = notify_change(dentry, &newattrs, NULL);
  65	inode_unlock(dentry->d_inode);
  66	return ret;
  67}
  68
  69long vfs_truncate(const struct path *path, loff_t length)
  70{
  71	struct inode *inode;
  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	error = get_write_access(inode);
  95	if (error)
  96		goto mnt_drop_write_and_out;
  97
  98	/*
  99	 * Make sure that there are no leases.  get_write_access() protects
 100	 * against the truncate racing with a lease-granting setlease().
 101	 */
 102	error = break_lease(inode, O_WRONLY);
 103	if (error)
 104		goto put_write_and_out;
 105
 106	error = locks_verify_truncate(inode, NULL, length);
 107	if (!error)
 108		error = security_path_truncate(path);
 109	if (!error)
 110		error = do_truncate(path->dentry, length, 0, NULL);
 111
 112put_write_and_out:
 113	put_write_access(inode);
 114mnt_drop_write_and_out:
 115	mnt_drop_write(path->mnt);
 116out:
 117	return error;
 118}
 119EXPORT_SYMBOL_GPL(vfs_truncate);
 120
 121long do_sys_truncate(const char __user *pathname, loff_t length)
 122{
 123	unsigned int lookup_flags = LOOKUP_FOLLOW;
 124	struct path path;
 125	int error;
 126
 127	if (length < 0)	/* sorry, but loff_t says... */
 128		return -EINVAL;
 129
 130retry:
 131	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 132	if (!error) {
 133		error = vfs_truncate(&path, length);
 134		path_put(&path);
 135	}
 136	if (retry_estale(error, lookup_flags)) {
 137		lookup_flags |= LOOKUP_REVAL;
 138		goto retry;
 139	}
 140	return error;
 141}
 142
 143SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 144{
 145	return do_sys_truncate(path, length);
 146}
 147
 148#ifdef CONFIG_COMPAT
 149COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 150{
 151	return do_sys_truncate(path, length);
 152}
 153#endif
 154
 155long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 156{
 157	struct inode *inode;
 158	struct dentry *dentry;
 159	struct fd f;
 160	int error;
 161
 162	error = -EINVAL;
 163	if (length < 0)
 164		goto out;
 165	error = -EBADF;
 166	f = fdget(fd);
 167	if (!f.file)
 168		goto out;
 169
 170	/* explicitly opened as large or we are on 64-bit box */
 171	if (f.file->f_flags & O_LARGEFILE)
 172		small = 0;
 173
 174	dentry = f.file->f_path.dentry;
 175	inode = dentry->d_inode;
 176	error = -EINVAL;
 177	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 178		goto out_putf;
 179
 180	error = -EINVAL;
 181	/* Cannot ftruncate over 2^31 bytes without large file support */
 182	if (small && length > MAX_NON_LFS)
 183		goto out_putf;
 184
 185	error = -EPERM;
 186	/* Check IS_APPEND on real upper inode */
 187	if (IS_APPEND(file_inode(f.file)))
 188		goto out_putf;
 189
 190	sb_start_write(inode->i_sb);
 191	error = locks_verify_truncate(inode, f.file, length);
 192	if (!error)
 193		error = security_path_truncate(&f.file->f_path);
 194	if (!error)
 195		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 196	sb_end_write(inode->i_sb);
 197out_putf:
 198	fdput(f);
 199out:
 200	return error;
 201}
 202
 203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 204{
 205	return do_sys_ftruncate(fd, length, 1);
 206}
 207
 208#ifdef CONFIG_COMPAT
 209COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 210{
 211	return do_sys_ftruncate(fd, length, 1);
 212}
 213#endif
 214
 215/* LFS versions of truncate are only needed on 32 bit machines */
 216#if BITS_PER_LONG == 32
 217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 218{
 219	return do_sys_truncate(path, length);
 220}
 221
 222SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 223{
 224	return do_sys_ftruncate(fd, length, 0);
 225}
 226#endif /* BITS_PER_LONG == 32 */
 227
 228
 229int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 230{
 231	struct inode *inode = file_inode(file);
 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_SUPPORTED_MASK)
 239		return -EOPNOTSUPP;
 240
 241	/* Punch hole and zero range are mutually exclusive */
 242	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 243	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 244		return -EOPNOTSUPP;
 245
 246	/* Punch hole must have keep size set */
 247	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 248	    !(mode & FALLOC_FL_KEEP_SIZE))
 249		return -EOPNOTSUPP;
 250
 251	/* Collapse range should only be used exclusively. */
 252	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 253	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 254		return -EINVAL;
 255
 256	/* Insert range should only be used exclusively. */
 257	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 258	    (mode & ~FALLOC_FL_INSERT_RANGE))
 259		return -EINVAL;
 260
 261	/* Unshare range should only be used with allocate mode. */
 262	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 263	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 264		return -EINVAL;
 265
 266	if (!(file->f_mode & FMODE_WRITE))
 267		return -EBADF;
 268
 269	/*
 270	 * We can only allow pure fallocate on append only files
 271	 */
 272	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 273		return -EPERM;
 274
 275	if (IS_IMMUTABLE(inode))
 276		return -EPERM;
 277
 278	/*
 279	 * We cannot allow any fallocate operation on an active swapfile
 280	 */
 281	if (IS_SWAPFILE(inode))
 282		return -ETXTBSY;
 283
 284	/*
 285	 * Revalidate the write permissions, in case security policy has
 286	 * changed since the files were opened.
 287	 */
 288	ret = security_file_permission(file, MAY_WRITE);
 289	if (ret)
 290		return ret;
 291
 292	if (S_ISFIFO(inode->i_mode))
 293		return -ESPIPE;
 294
 295	if (S_ISDIR(inode->i_mode))
 296		return -EISDIR;
 297
 298	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 
 299		return -ENODEV;
 300
 301	/* Check for wrap through zero too */
 302	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 303		return -EFBIG;
 304
 305	if (!file->f_op->fallocate)
 306		return -EOPNOTSUPP;
 307
 308	file_start_write(file);
 309	ret = file->f_op->fallocate(file, mode, offset, len);
 310
 311	/*
 312	 * Create inotify and fanotify events.
 313	 *
 314	 * To keep the logic simple always create events if fallocate succeeds.
 315	 * This implies that events are even created if the file size remains
 316	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 317	 */
 318	if (ret == 0)
 319		fsnotify_modify(file);
 320
 321	file_end_write(file);
 322	return ret;
 323}
 324EXPORT_SYMBOL_GPL(vfs_fallocate);
 325
 326int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 327{
 328	struct fd f = fdget(fd);
 329	int error = -EBADF;
 330
 331	if (f.file) {
 332		error = vfs_fallocate(f.file, mode, offset, len);
 333		fdput(f);
 334	}
 335	return error;
 336}
 337
 338SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 339{
 340	return ksys_fallocate(fd, mode, offset, len);
 341}
 342
 343/*
 344 * access() needs to use the real uid/gid, not the effective uid/gid.
 345 * We do this by temporarily clearing all FS-related capabilities and
 346 * switching the fsuid/fsgid around to the real ones.
 347 */
 348long do_faccessat(int dfd, const char __user *filename, int mode)
 349{
 350	const struct cred *old_cred;
 351	struct cred *override_cred;
 352	struct path path;
 353	struct inode *inode;
 354	int res;
 355	unsigned int lookup_flags = LOOKUP_FOLLOW;
 356
 357	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 358		return -EINVAL;
 359
 360	override_cred = prepare_creds();
 361	if (!override_cred)
 362		return -ENOMEM;
 363
 364	override_cred->fsuid = override_cred->uid;
 365	override_cred->fsgid = override_cred->gid;
 366
 367	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 368		/* Clear the capabilities if we switch to a non-root user */
 369		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 370		if (!uid_eq(override_cred->uid, root_uid))
 371			cap_clear(override_cred->cap_effective);
 372		else
 373			override_cred->cap_effective =
 374				override_cred->cap_permitted;
 375	}
 376
 377	/*
 378	 * The new set of credentials can *only* be used in
 379	 * task-synchronous circumstances, and does not need
 380	 * RCU freeing, unless somebody then takes a separate
 381	 * reference to it.
 382	 *
 383	 * NOTE! This is _only_ true because this credential
 384	 * is used purely for override_creds() that installs
 385	 * it as the subjective cred. Other threads will be
 386	 * accessing ->real_cred, not the subjective cred.
 387	 *
 388	 * If somebody _does_ make a copy of this (using the
 389	 * 'get_current_cred()' function), that will clear the
 390	 * non_rcu field, because now that other user may be
 391	 * expecting RCU freeing. But normal thread-synchronous
 392	 * cred accesses will keep things non-RCY.
 393	 */
 394	override_cred->non_rcu = 1;
 395
 396	old_cred = override_creds(override_cred);
 397retry:
 398	res = user_path_at(dfd, filename, lookup_flags, &path);
 399	if (res)
 400		goto out;
 401
 402	inode = d_backing_inode(path.dentry);
 403
 404	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 405		/*
 406		 * MAY_EXEC on regular files is denied if the fs is mounted
 407		 * with the "noexec" flag.
 408		 */
 409		res = -EACCES;
 410		if (path_noexec(&path))
 411			goto out_path_release;
 412	}
 413
 414	res = inode_permission(inode, mode | MAY_ACCESS);
 415	/* SuS v2 requires we report a read only fs too */
 416	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 417		goto out_path_release;
 418	/*
 419	 * This is a rare case where using __mnt_is_readonly()
 420	 * is OK without a mnt_want/drop_write() pair.  Since
 421	 * no actual write to the fs is performed here, we do
 422	 * not need to telegraph to that to anyone.
 423	 *
 424	 * By doing this, we accept that this access is
 425	 * inherently racy and know that the fs may change
 426	 * state before we even see this result.
 427	 */
 428	if (__mnt_is_readonly(path.mnt))
 429		res = -EROFS;
 430
 431out_path_release:
 432	path_put(&path);
 433	if (retry_estale(res, lookup_flags)) {
 434		lookup_flags |= LOOKUP_REVAL;
 435		goto retry;
 436	}
 437out:
 438	revert_creds(old_cred);
 439	put_cred(override_cred);
 440	return res;
 441}
 442
 443SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 444{
 445	return do_faccessat(dfd, filename, mode);
 446}
 447
 448SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 449{
 450	return do_faccessat(AT_FDCWD, filename, mode);
 451}
 452
 453int ksys_chdir(const char __user *filename)
 454{
 455	struct path path;
 456	int error;
 457	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 458retry:
 459	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 460	if (error)
 461		goto out;
 462
 463	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 464	if (error)
 465		goto dput_and_out;
 466
 467	set_fs_pwd(current->fs, &path);
 468
 469dput_and_out:
 470	path_put(&path);
 471	if (retry_estale(error, lookup_flags)) {
 472		lookup_flags |= LOOKUP_REVAL;
 473		goto retry;
 474	}
 475out:
 476	return error;
 477}
 478
 479SYSCALL_DEFINE1(chdir, const char __user *, filename)
 480{
 481	return ksys_chdir(filename);
 482}
 483
 484SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 485{
 486	struct fd f = fdget_raw(fd);
 487	int error;
 
 488
 489	error = -EBADF;
 490	if (!f.file)
 491		goto out;
 492
 
 
 493	error = -ENOTDIR;
 494	if (!d_can_lookup(f.file->f_path.dentry))
 495		goto out_putf;
 496
 497	error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 498	if (!error)
 499		set_fs_pwd(current->fs, &f.file->f_path);
 500out_putf:
 501	fdput(f);
 502out:
 503	return error;
 504}
 505
 506int ksys_chroot(const char __user *filename)
 507{
 508	struct path path;
 509	int error;
 510	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 511retry:
 512	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 513	if (error)
 514		goto out;
 515
 516	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 517	if (error)
 518		goto dput_and_out;
 519
 520	error = -EPERM;
 521	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 522		goto dput_and_out;
 523	error = security_path_chroot(&path);
 524	if (error)
 525		goto dput_and_out;
 526
 527	set_fs_root(current->fs, &path);
 528	error = 0;
 529dput_and_out:
 530	path_put(&path);
 531	if (retry_estale(error, lookup_flags)) {
 532		lookup_flags |= LOOKUP_REVAL;
 533		goto retry;
 534	}
 535out:
 536	return error;
 537}
 538
 539SYSCALL_DEFINE1(chroot, const char __user *, filename)
 540{
 541	return ksys_chroot(filename);
 542}
 543
 544static int chmod_common(const struct path *path, umode_t mode)
 545{
 546	struct inode *inode = path->dentry->d_inode;
 547	struct inode *delegated_inode = NULL;
 548	struct iattr newattrs;
 549	int error;
 550
 551	error = mnt_want_write(path->mnt);
 552	if (error)
 553		return error;
 554retry_deleg:
 555	inode_lock(inode);
 556	error = security_path_chmod(path, mode);
 557	if (error)
 558		goto out_unlock;
 559	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 560	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 561	error = notify_change(path->dentry, &newattrs, &delegated_inode);
 562out_unlock:
 563	inode_unlock(inode);
 564	if (delegated_inode) {
 565		error = break_deleg_wait(&delegated_inode);
 566		if (!error)
 567			goto retry_deleg;
 568	}
 569	mnt_drop_write(path->mnt);
 570	return error;
 571}
 572
 573int ksys_fchmod(unsigned int fd, umode_t mode)
 574{
 575	struct fd f = fdget(fd);
 576	int err = -EBADF;
 577
 578	if (f.file) {
 579		audit_file(f.file);
 580		err = chmod_common(&f.file->f_path, mode);
 581		fdput(f);
 582	}
 583	return err;
 584}
 585
 586SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 587{
 588	return ksys_fchmod(fd, mode);
 589}
 590
 591int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 592{
 593	struct path path;
 594	int error;
 595	unsigned int lookup_flags = LOOKUP_FOLLOW;
 596retry:
 597	error = user_path_at(dfd, filename, lookup_flags, &path);
 598	if (!error) {
 599		error = chmod_common(&path, mode);
 600		path_put(&path);
 601		if (retry_estale(error, lookup_flags)) {
 602			lookup_flags |= LOOKUP_REVAL;
 603			goto retry;
 604		}
 605	}
 606	return error;
 607}
 608
 609SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 610		umode_t, mode)
 611{
 612	return do_fchmodat(dfd, filename, mode);
 613}
 614
 615SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 616{
 617	return do_fchmodat(AT_FDCWD, filename, mode);
 618}
 619
 620static int chown_common(const struct path *path, uid_t user, gid_t group)
 621{
 622	struct inode *inode = path->dentry->d_inode;
 623	struct inode *delegated_inode = NULL;
 624	int error;
 625	struct iattr newattrs;
 626	kuid_t uid;
 627	kgid_t gid;
 628
 629	uid = make_kuid(current_user_ns(), user);
 630	gid = make_kgid(current_user_ns(), group);
 631
 632retry_deleg:
 633	newattrs.ia_valid =  ATTR_CTIME;
 634	if (user != (uid_t) -1) {
 635		if (!uid_valid(uid))
 636			return -EINVAL;
 637		newattrs.ia_valid |= ATTR_UID;
 638		newattrs.ia_uid = uid;
 639	}
 640	if (group != (gid_t) -1) {
 641		if (!gid_valid(gid))
 642			return -EINVAL;
 643		newattrs.ia_valid |= ATTR_GID;
 644		newattrs.ia_gid = gid;
 645	}
 646	if (!S_ISDIR(inode->i_mode))
 647		newattrs.ia_valid |=
 648			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 649	inode_lock(inode);
 650	error = security_path_chown(path, uid, gid);
 651	if (!error)
 652		error = notify_change(path->dentry, &newattrs, &delegated_inode);
 653	inode_unlock(inode);
 654	if (delegated_inode) {
 655		error = break_deleg_wait(&delegated_inode);
 656		if (!error)
 657			goto retry_deleg;
 658	}
 659	return error;
 660}
 661
 662int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 663		int flag)
 664{
 665	struct path path;
 666	int error = -EINVAL;
 667	int lookup_flags;
 668
 669	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 670		goto out;
 671
 672	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 673	if (flag & AT_EMPTY_PATH)
 674		lookup_flags |= LOOKUP_EMPTY;
 675retry:
 676	error = user_path_at(dfd, filename, lookup_flags, &path);
 677	if (error)
 678		goto out;
 679	error = mnt_want_write(path.mnt);
 680	if (error)
 681		goto out_release;
 682	error = chown_common(&path, user, group);
 683	mnt_drop_write(path.mnt);
 684out_release:
 685	path_put(&path);
 686	if (retry_estale(error, lookup_flags)) {
 687		lookup_flags |= LOOKUP_REVAL;
 688		goto retry;
 689	}
 690out:
 691	return error;
 692}
 693
 694SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 695		gid_t, group, int, flag)
 696{
 697	return do_fchownat(dfd, filename, user, group, flag);
 698}
 699
 700SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 701{
 702	return do_fchownat(AT_FDCWD, filename, user, group, 0);
 703}
 704
 705SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 706{
 707	return do_fchownat(AT_FDCWD, filename, user, group,
 708			   AT_SYMLINK_NOFOLLOW);
 709}
 710
 711int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 712{
 713	struct fd f = fdget(fd);
 714	int error = -EBADF;
 715
 716	if (!f.file)
 717		goto out;
 718
 719	error = mnt_want_write_file(f.file);
 720	if (error)
 721		goto out_fput;
 722	audit_file(f.file);
 723	error = chown_common(&f.file->f_path, user, group);
 724	mnt_drop_write_file(f.file);
 725out_fput:
 726	fdput(f);
 727out:
 728	return error;
 729}
 730
 731SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 732{
 733	return ksys_fchown(fd, user, group);
 
 
 
 
 
 734}
 735
 736static int do_dentry_open(struct file *f,
 737			  struct inode *inode,
 738			  int (*open)(struct inode *, struct file *))
 
 739{
 740	static const struct file_operations empty_fops = {};
 741	int error;
 742
 
 
 
 743	path_get(&f->f_path);
 744	f->f_inode = inode;
 745	f->f_mapping = inode->i_mapping;
 746
 747	/* Ensure that we skip any errors that predate opening of the file */
 748	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 749
 750	if (unlikely(f->f_flags & O_PATH)) {
 751		f->f_mode = FMODE_PATH | FMODE_OPENED;
 752		f->f_op = &empty_fops;
 753		return 0;
 754	}
 755
 756	/* Any file opened for execve()/uselib() has to be a regular file. */
 757	if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 758		error = -EACCES;
 759		goto cleanup_file;
 760	}
 761
 762	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 763		error = get_write_access(inode);
 764		if (unlikely(error))
 765			goto cleanup_file;
 766		error = __mnt_want_write(f->f_path.mnt);
 767		if (unlikely(error)) {
 768			put_write_access(inode);
 769			goto cleanup_file;
 770		}
 771		f->f_mode |= FMODE_WRITER;
 772	}
 773
 774	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 775	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 776		f->f_mode |= FMODE_ATOMIC_POS;
 777
 778	f->f_op = fops_get(inode->i_fop);
 779	if (WARN_ON(!f->f_op)) {
 780		error = -ENODEV;
 781		goto cleanup_all;
 782	}
 783
 784	error = security_file_open(f);
 785	if (error)
 786		goto cleanup_all;
 787
 788	error = break_lease(locks_inode(f), f->f_flags);
 789	if (error)
 790		goto cleanup_all;
 791
 792	/* normally all 3 are set; ->open() can clear them if needed */
 793	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 794	if (!open)
 795		open = f->f_op->open;
 796	if (open) {
 797		error = open(inode, f);
 798		if (error)
 799			goto cleanup_all;
 800	}
 801	f->f_mode |= FMODE_OPENED;
 802	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 803		i_readcount_inc(inode);
 804	if ((f->f_mode & FMODE_READ) &&
 805	     likely(f->f_op->read || f->f_op->read_iter))
 806		f->f_mode |= FMODE_CAN_READ;
 807	if ((f->f_mode & FMODE_WRITE) &&
 808	     likely(f->f_op->write || f->f_op->write_iter))
 809		f->f_mode |= FMODE_CAN_WRITE;
 810
 811	f->f_write_hint = WRITE_LIFE_NOT_SET;
 812	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 813
 814	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 815
 816	/* NB: we're sure to have correct a_ops only after f_op->open */
 817	if (f->f_flags & O_DIRECT) {
 818		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 819			return -EINVAL;
 820	}
 821
 822	/*
 823	 * XXX: Huge page cache doesn't support writing yet. Drop all page
 824	 * cache for this file before processing writes.
 825	 */
 826	if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
 827		truncate_pagecache(inode, 0);
 828
 829	return 0;
 830
 831cleanup_all:
 832	if (WARN_ON_ONCE(error > 0))
 833		error = -EINVAL;
 834	fops_put(f->f_op);
 835	if (f->f_mode & FMODE_WRITER) {
 836		put_write_access(inode);
 837		__mnt_drop_write(f->f_path.mnt);
 838	}
 839cleanup_file:
 840	path_put(&f->f_path);
 841	f->f_path.mnt = NULL;
 842	f->f_path.dentry = NULL;
 843	f->f_inode = NULL;
 844	return error;
 845}
 846
 847/**
 848 * finish_open - finish opening a file
 849 * @file: file pointer
 850 * @dentry: pointer to dentry
 851 * @open: open callback
 852 * @opened: state of open
 853 *
 854 * This can be used to finish opening a file passed to i_op->atomic_open().
 855 *
 856 * If the open callback is set to NULL, then the standard f_op->open()
 857 * filesystem callback is substituted.
 858 *
 859 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 860 * the return value of d_splice_alias(), then the caller needs to perform dput()
 861 * on it after finish_open().
 862 *
 863 * On successful return @file is a fully instantiated open file.  After this, if
 864 * an error occurs in ->atomic_open(), it needs to clean up with fput().
 865 *
 866 * Returns zero on success or -errno if the open failed.
 867 */
 868int finish_open(struct file *file, struct dentry *dentry,
 869		int (*open)(struct inode *, struct file *))
 
 870{
 871	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 
 872
 873	file->f_path.dentry = dentry;
 874	return do_dentry_open(file, d_backing_inode(dentry), open);
 
 
 
 
 
 875}
 876EXPORT_SYMBOL(finish_open);
 877
 878/**
 879 * finish_no_open - finish ->atomic_open() without opening the file
 880 *
 881 * @file: file pointer
 882 * @dentry: dentry or NULL (as returned from ->lookup())
 883 *
 884 * This can be used to set the result of a successful lookup in ->atomic_open().
 885 *
 886 * NB: unlike finish_open() this function does consume the dentry reference and
 887 * the caller need not dput() it.
 888 *
 889 * Returns "0" which must be the return value of ->atomic_open() after having
 890 * called this function.
 891 */
 892int finish_no_open(struct file *file, struct dentry *dentry)
 893{
 894	file->f_path.dentry = dentry;
 895	return 0;
 896}
 897EXPORT_SYMBOL(finish_no_open);
 898
 899char *file_path(struct file *filp, char *buf, int buflen)
 900{
 901	return d_path(&filp->f_path, buf, buflen);
 902}
 903EXPORT_SYMBOL(file_path);
 904
 905/**
 906 * vfs_open - open the file at the given path
 907 * @path: path to open
 908 * @file: newly allocated file with f_flag initialized
 909 * @cred: credentials to use
 910 */
 911int vfs_open(const struct path *path, struct file *file)
 
 912{
 
 
 
 
 
 913	file->f_path = *path;
 914	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
 915}
 916
 917struct file *dentry_open(const struct path *path, int flags,
 918			 const struct cred *cred)
 919{
 920	int error;
 921	struct file *f;
 922
 923	validate_creds(cred);
 924
 925	/* We must always pass in a valid mount pointer. */
 926	BUG_ON(!path->mnt);
 927
 928	f = alloc_empty_file(flags, cred);
 929	if (!IS_ERR(f)) {
 930		error = vfs_open(path, f);
 931		if (error) {
 932			fput(f);
 
 
 
 
 
 
 
 
 933			f = ERR_PTR(error);
 934		}
 935	}
 936	return f;
 937}
 938EXPORT_SYMBOL(dentry_open);
 939
 940struct file *open_with_fake_path(const struct path *path, int flags,
 941				struct inode *inode, const struct cred *cred)
 942{
 943	struct file *f = alloc_empty_file_noaccount(flags, cred);
 944	if (!IS_ERR(f)) {
 945		int error;
 946
 947		f->f_path = *path;
 948		error = do_dentry_open(f, inode, NULL);
 949		if (error) {
 950			fput(f);
 951			f = ERR_PTR(error);
 952		}
 953	}
 954	return f;
 955}
 956EXPORT_SYMBOL(open_with_fake_path);
 957
 958static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 959{
 960	int lookup_flags = 0;
 961	int acc_mode = ACC_MODE(flags);
 962
 963	/*
 964	 * Clear out all open flags we don't know about so that we don't report
 965	 * them in fcntl(F_GETFD) or similar interfaces.
 966	 */
 967	flags &= VALID_OPEN_FLAGS;
 968
 969	if (flags & (O_CREAT | __O_TMPFILE))
 970		op->mode = (mode & S_IALLUGO) | S_IFREG;
 971	else
 972		op->mode = 0;
 973
 974	/* Must never be set by userspace */
 975	flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 976
 977	/*
 978	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 979	 * check for O_DSYNC if the need any syncing at all we enforce it's
 980	 * always set instead of having to deal with possibly weird behaviour
 981	 * for malicious applications setting only __O_SYNC.
 982	 */
 983	if (flags & __O_SYNC)
 984		flags |= O_DSYNC;
 985
 986	if (flags & __O_TMPFILE) {
 987		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 988			return -EINVAL;
 989		if (!(acc_mode & MAY_WRITE))
 990			return -EINVAL;
 991	} else if (flags & O_PATH) {
 992		/*
 993		 * If we have O_PATH in the open flag. Then we
 994		 * cannot have anything other than the below set of flags
 995		 */
 996		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 997		acc_mode = 0;
 998	}
 999
1000	op->open_flag = flags;
1001
1002	/* O_TRUNC implies we need access checks for write permissions */
1003	if (flags & O_TRUNC)
1004		acc_mode |= MAY_WRITE;
1005
1006	/* Allow the LSM permission hook to distinguish append
1007	   access from general write access. */
1008	if (flags & O_APPEND)
1009		acc_mode |= MAY_APPEND;
1010
1011	op->acc_mode = acc_mode;
1012
1013	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1014
1015	if (flags & O_CREAT) {
1016		op->intent |= LOOKUP_CREATE;
1017		if (flags & O_EXCL)
1018			op->intent |= LOOKUP_EXCL;
1019	}
1020
1021	if (flags & O_DIRECTORY)
1022		lookup_flags |= LOOKUP_DIRECTORY;
1023	if (!(flags & O_NOFOLLOW))
1024		lookup_flags |= LOOKUP_FOLLOW;
1025	op->lookup_flags = lookup_flags;
1026	return 0;
1027}
1028
1029/**
1030 * file_open_name - open file and return file pointer
1031 *
1032 * @name:	struct filename containing path to open
1033 * @flags:	open flags as per the open(2) second argument
1034 * @mode:	mode for the new file if O_CREAT is set, else ignored
1035 *
1036 * This is the helper to open a file from kernelspace if you really
1037 * have to.  But in generally you should not do this, so please move
1038 * along, nothing to see here..
1039 */
1040struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1041{
1042	struct open_flags op;
1043	int err = build_open_flags(flags, mode, &op);
1044	return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1045}
1046
1047/**
1048 * filp_open - open file and return file pointer
1049 *
1050 * @filename:	path to open
1051 * @flags:	open flags as per the open(2) second argument
1052 * @mode:	mode for the new file if O_CREAT is set, else ignored
1053 *
1054 * This is the helper to open a file from kernelspace if you really
1055 * have to.  But in generally you should not do this, so please move
1056 * along, nothing to see here..
1057 */
1058struct file *filp_open(const char *filename, int flags, umode_t mode)
1059{
1060	struct filename *name = getname_kernel(filename);
1061	struct file *file = ERR_CAST(name);
1062	
1063	if (!IS_ERR(name)) {
1064		file = file_open_name(name, flags, mode);
1065		putname(name);
1066	}
1067	return file;
1068}
1069EXPORT_SYMBOL(filp_open);
1070
1071struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1072			    const char *filename, int flags, umode_t mode)
1073{
1074	struct open_flags op;
1075	int err = build_open_flags(flags, mode, &op);
1076	if (err)
1077		return ERR_PTR(err);
1078	return do_file_open_root(dentry, mnt, filename, &op);
1079}
1080EXPORT_SYMBOL(file_open_root);
1081
1082long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1083{
1084	struct open_flags op;
1085	int fd = build_open_flags(flags, mode, &op);
1086	struct filename *tmp;
1087
1088	if (fd)
1089		return fd;
1090
1091	tmp = getname(filename);
1092	if (IS_ERR(tmp))
1093		return PTR_ERR(tmp);
1094
1095	fd = get_unused_fd_flags(flags);
1096	if (fd >= 0) {
1097		struct file *f = do_filp_open(dfd, tmp, &op);
1098		if (IS_ERR(f)) {
1099			put_unused_fd(fd);
1100			fd = PTR_ERR(f);
1101		} else {
1102			fsnotify_open(f);
1103			fd_install(fd, f);
1104		}
1105	}
1106	putname(tmp);
1107	return fd;
1108}
1109
1110SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1111{
1112	if (force_o_largefile())
1113		flags |= O_LARGEFILE;
1114
1115	return do_sys_open(AT_FDCWD, filename, flags, mode);
1116}
1117
1118SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1119		umode_t, mode)
1120{
1121	if (force_o_largefile())
1122		flags |= O_LARGEFILE;
1123
1124	return do_sys_open(dfd, filename, flags, mode);
1125}
1126
1127#ifdef CONFIG_COMPAT
1128/*
1129 * Exactly like sys_open(), except that it doesn't set the
1130 * O_LARGEFILE flag.
1131 */
1132COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1133{
1134	return do_sys_open(AT_FDCWD, filename, flags, mode);
1135}
1136
1137/*
1138 * Exactly like sys_openat(), except that it doesn't set the
1139 * O_LARGEFILE flag.
1140 */
1141COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1142{
1143	return do_sys_open(dfd, filename, flags, mode);
1144}
1145#endif
1146
1147#ifndef __alpha__
1148
1149/*
1150 * For backward compatibility?  Maybe this should be moved
1151 * into arch/i386 instead?
1152 */
1153SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1154{
1155	return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1156}
1157
1158#endif
1159
1160/*
1161 * "id" is the POSIX thread ID. We use the
1162 * files pointer for this..
1163 */
1164int filp_close(struct file *filp, fl_owner_t id)
1165{
1166	int retval = 0;
1167
1168	if (!file_count(filp)) {
1169		printk(KERN_ERR "VFS: Close: file count is 0\n");
1170		return 0;
1171	}
1172
1173	if (filp->f_op->flush)
1174		retval = filp->f_op->flush(filp, id);
1175
1176	if (likely(!(filp->f_mode & FMODE_PATH))) {
1177		dnotify_flush(filp, id);
1178		locks_remove_posix(filp, id);
1179	}
1180	fput(filp);
1181	return retval;
1182}
1183
1184EXPORT_SYMBOL(filp_close);
1185
1186/*
1187 * Careful here! We test whether the file pointer is NULL before
1188 * releasing the fd. This ensures that one clone task can't release
1189 * an fd while another clone is opening it.
1190 */
1191SYSCALL_DEFINE1(close, unsigned int, fd)
1192{
1193	int retval = __close_fd(current->files, fd);
1194
1195	/* can't restart close syscall because file table entry was cleared */
1196	if (unlikely(retval == -ERESTARTSYS ||
1197		     retval == -ERESTARTNOINTR ||
1198		     retval == -ERESTARTNOHAND ||
1199		     retval == -ERESTART_RESTARTBLOCK))
1200		retval = -EINTR;
1201
1202	return retval;
1203}
 
1204
1205/*
1206 * This routine simulates a hangup on the tty, to arrange that users
1207 * are given clean terminals at login time.
1208 */
1209SYSCALL_DEFINE0(vhangup)
1210{
1211	if (capable(CAP_SYS_TTY_CONFIG)) {
1212		tty_vhangup_self();
1213		return 0;
1214	}
1215	return -EPERM;
1216}
1217
1218/*
1219 * Called when an inode is about to be open.
1220 * We use this to disallow opening large files on 32bit systems if
1221 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1222 * on this flag in sys_open.
1223 */
1224int generic_file_open(struct inode * inode, struct file * filp)
1225{
1226	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1227		return -EOVERFLOW;
1228	return 0;
1229}
1230
1231EXPORT_SYMBOL(generic_file_open);
1232
1233/*
1234 * This is used by subsystems that don't want seekable
1235 * file descriptors. The function is not supposed to ever fail, the only
1236 * reason it returns an 'int' and not 'void' is so that it can be plugged
1237 * directly into file_operations structure.
1238 */
1239int nonseekable_open(struct inode *inode, struct file *filp)
1240{
1241	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1242	return 0;
1243}
1244
1245EXPORT_SYMBOL(nonseekable_open);
1246
1247/*
1248 * stream_open is used by subsystems that want stream-like file descriptors.
1249 * Such file descriptors are not seekable and don't have notion of position
1250 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1251 * Contrary to file descriptors of other regular files, .read() and .write()
1252 * can run simultaneously.
1253 *
1254 * stream_open never fails and is marked to return int so that it could be
1255 * directly used as file_operations.open .
1256 */
1257int stream_open(struct inode *inode, struct file *filp)
1258{
1259	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1260	filp->f_mode |= FMODE_STREAM;
1261	return 0;
1262}
1263
1264EXPORT_SYMBOL(stream_open);