Linux Audio

Check our new training course

Loading...
v4.10.11
   1/*
   2 *  linux/fs/open.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/string.h>
   8#include <linux/mm.h>
   9#include <linux/file.h>
  10#include <linux/fdtable.h>
  11#include <linux/fsnotify.h>
  12#include <linux/module.h>
  13#include <linux/tty.h>
  14#include <linux/namei.h>
  15#include <linux/backing-dev.h>
  16#include <linux/capability.h>
  17#include <linux/securebits.h>
  18#include <linux/security.h>
  19#include <linux/mount.h>
  20#include <linux/fcntl.h>
  21#include <linux/slab.h>
  22#include <linux/uaccess.h>
  23#include <linux/fs.h>
  24#include <linux/personality.h>
  25#include <linux/pagemap.h>
  26#include <linux/syscalls.h>
  27#include <linux/rcupdate.h>
  28#include <linux/audit.h>
  29#include <linux/falloc.h>
  30#include <linux/fs_struct.h>
  31#include <linux/ima.h>
  32#include <linux/dnotify.h>
  33#include <linux/compat.h>
  34
  35#include "internal.h"
  36
  37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38	struct file *filp)
  39{
  40	int ret;
  41	struct iattr newattrs;
  42
  43	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44	if (length < 0)
  45		return -EINVAL;
  46
  47	newattrs.ia_size = length;
  48	newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49	if (filp) {
  50		newattrs.ia_file = filp;
  51		newattrs.ia_valid |= ATTR_FILE;
  52	}
  53
  54	/* Remove suid, sgid, and file capabilities on truncate too */
  55	ret = dentry_needs_remove_privs(dentry);
  56	if (ret < 0)
  57		return ret;
  58	if (ret)
  59		newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61	inode_lock(dentry->d_inode);
  62	/* Note any delegations or leases have already been broken: */
  63	ret = notify_change(dentry, &newattrs, NULL);
  64	inode_unlock(dentry->d_inode);
  65	return ret;
  66}
  67
  68long vfs_truncate(const struct path *path, loff_t length)
  69{
  70	struct inode *inode;
  71	struct dentry *upperdentry;
  72	long error;
  73
  74	inode = path->dentry->d_inode;
  75
  76	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77	if (S_ISDIR(inode->i_mode))
  78		return -EISDIR;
  79	if (!S_ISREG(inode->i_mode))
  80		return -EINVAL;
  81
  82	error = mnt_want_write(path->mnt);
  83	if (error)
  84		goto out;
  85
  86	error = inode_permission(inode, MAY_WRITE);
  87	if (error)
  88		goto mnt_drop_write_and_out;
  89
  90	error = -EPERM;
  91	if (IS_APPEND(inode))
  92		goto mnt_drop_write_and_out;
  93
  94	/*
  95	 * If this is an overlayfs then do as if opening the file so we get
  96	 * write access on the upper inode, not on the overlay inode.  For
  97	 * non-overlay filesystems d_real() is an identity function.
  98	 */
  99	upperdentry = d_real(path->dentry, NULL, O_WRONLY);
 100	error = PTR_ERR(upperdentry);
 101	if (IS_ERR(upperdentry))
 102		goto mnt_drop_write_and_out;
 103
 104	error = get_write_access(upperdentry->d_inode);
 105	if (error)
 106		goto mnt_drop_write_and_out;
 107
 108	/*
 109	 * Make sure that there are no leases.  get_write_access() protects
 110	 * against the truncate racing with a lease-granting setlease().
 111	 */
 112	error = break_lease(inode, O_WRONLY);
 113	if (error)
 114		goto put_write_and_out;
 115
 116	error = locks_verify_truncate(inode, NULL, length);
 117	if (!error)
 118		error = security_path_truncate(path);
 119	if (!error)
 120		error = do_truncate(path->dentry, length, 0, NULL);
 121
 122put_write_and_out:
 123	put_write_access(upperdentry->d_inode);
 124mnt_drop_write_and_out:
 125	mnt_drop_write(path->mnt);
 126out:
 127	return error;
 128}
 129EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131static long do_sys_truncate(const char __user *pathname, loff_t length)
 132{
 133	unsigned int lookup_flags = LOOKUP_FOLLOW;
 134	struct path path;
 135	int error;
 136
 137	if (length < 0)	/* sorry, but loff_t says... */
 138		return -EINVAL;
 139
 140retry:
 141	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142	if (!error) {
 143		error = vfs_truncate(&path, length);
 144		path_put(&path);
 145	}
 146	if (retry_estale(error, lookup_flags)) {
 147		lookup_flags |= LOOKUP_REVAL;
 148		goto retry;
 149	}
 150	return error;
 151}
 152
 153SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154{
 155	return do_sys_truncate(path, length);
 156}
 157
 158#ifdef CONFIG_COMPAT
 159COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160{
 161	return do_sys_truncate(path, length);
 162}
 163#endif
 164
 165static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166{
 167	struct inode *inode;
 168	struct dentry *dentry;
 169	struct fd f;
 170	int error;
 171
 172	error = -EINVAL;
 173	if (length < 0)
 174		goto out;
 175	error = -EBADF;
 176	f = fdget(fd);
 177	if (!f.file)
 178		goto out;
 179
 180	/* explicitly opened as large or we are on 64-bit box */
 181	if (f.file->f_flags & O_LARGEFILE)
 182		small = 0;
 183
 184	dentry = f.file->f_path.dentry;
 185	inode = dentry->d_inode;
 186	error = -EINVAL;
 187	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188		goto out_putf;
 189
 190	error = -EINVAL;
 191	/* Cannot ftruncate over 2^31 bytes without large file support */
 192	if (small && length > MAX_NON_LFS)
 193		goto out_putf;
 194
 195	error = -EPERM;
 196	if (IS_APPEND(inode))
 197		goto out_putf;
 198
 199	sb_start_write(inode->i_sb);
 200	error = locks_verify_truncate(inode, f.file, length);
 201	if (!error)
 202		error = security_path_truncate(&f.file->f_path);
 203	if (!error)
 204		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 205	sb_end_write(inode->i_sb);
 206out_putf:
 207	fdput(f);
 208out:
 209	return error;
 210}
 211
 212SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 213{
 214	return do_sys_ftruncate(fd, length, 1);
 215}
 216
 217#ifdef CONFIG_COMPAT
 218COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 219{
 220	return do_sys_ftruncate(fd, length, 1);
 221}
 222#endif
 223
 224/* LFS versions of truncate are only needed on 32 bit machines */
 225#if BITS_PER_LONG == 32
 226SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 227{
 228	return do_sys_truncate(path, length);
 229}
 230
 231SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 232{
 233	return do_sys_ftruncate(fd, length, 0);
 234}
 235#endif /* BITS_PER_LONG == 32 */
 236
 237
 238int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 239{
 240	struct inode *inode = file_inode(file);
 241	long ret;
 242
 243	if (offset < 0 || len <= 0)
 244		return -EINVAL;
 245
 246	/* Return error if mode is not supported */
 247	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 248		return -EOPNOTSUPP;
 249
 250	/* Punch hole and zero range are mutually exclusive */
 251	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 252	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 253		return -EOPNOTSUPP;
 254
 255	/* Punch hole must have keep size set */
 256	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 257	    !(mode & FALLOC_FL_KEEP_SIZE))
 258		return -EOPNOTSUPP;
 259
 260	/* Collapse range should only be used exclusively. */
 261	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 262	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 263		return -EINVAL;
 264
 265	/* Insert range should only be used exclusively. */
 266	if ((mode & FALLOC_FL_INSERT_RANGE) &&
 267	    (mode & ~FALLOC_FL_INSERT_RANGE))
 268		return -EINVAL;
 269
 270	/* Unshare range should only be used with allocate mode. */
 271	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 272	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 273		return -EINVAL;
 274
 275	if (!(file->f_mode & FMODE_WRITE))
 276		return -EBADF;
 277
 278	/*
 279	 * We can only allow pure fallocate on append only files
 280	 */
 281	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 282		return -EPERM;
 283
 284	if (IS_IMMUTABLE(inode))
 285		return -EPERM;
 286
 287	/*
 288	 * We cannot allow any fallocate operation on an active swapfile
 289	 */
 290	if (IS_SWAPFILE(inode))
 291		return -ETXTBSY;
 292
 293	/*
 294	 * Revalidate the write permissions, in case security policy has
 295	 * changed since the files were opened.
 296	 */
 297	ret = security_file_permission(file, MAY_WRITE);
 298	if (ret)
 299		return ret;
 300
 301	if (S_ISFIFO(inode->i_mode))
 302		return -ESPIPE;
 303
 304	/*
 305	 * Let individual file system decide if it supports preallocation
 306	 * for directories or not.
 307	 */
 308	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
 309	    !S_ISBLK(inode->i_mode))
 310		return -ENODEV;
 311
 312	/* Check for wrap through zero too */
 313	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 314		return -EFBIG;
 315
 316	if (!file->f_op->fallocate)
 317		return -EOPNOTSUPP;
 318
 319	sb_start_write(inode->i_sb);
 320	ret = file->f_op->fallocate(file, mode, offset, len);
 321
 322	/*
 323	 * Create inotify and fanotify events.
 324	 *
 325	 * To keep the logic simple always create events if fallocate succeeds.
 326	 * This implies that events are even created if the file size remains
 327	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 328	 */
 329	if (ret == 0)
 330		fsnotify_modify(file);
 331
 332	sb_end_write(inode->i_sb);
 333	return ret;
 334}
 335EXPORT_SYMBOL_GPL(vfs_fallocate);
 336
 337SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 338{
 339	struct fd f = fdget(fd);
 340	int error = -EBADF;
 341
 342	if (f.file) {
 343		error = vfs_fallocate(f.file, mode, offset, len);
 344		fdput(f);
 345	}
 346	return error;
 347}
 348
 349/*
 350 * access() needs to use the real uid/gid, not the effective uid/gid.
 351 * We do this by temporarily clearing all FS-related capabilities and
 352 * switching the fsuid/fsgid around to the real ones.
 353 */
 354SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 355{
 356	const struct cred *old_cred;
 357	struct cred *override_cred;
 358	struct path path;
 359	struct inode *inode;
 360	int res;
 361	unsigned int lookup_flags = LOOKUP_FOLLOW;
 362
 363	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 364		return -EINVAL;
 365
 366	override_cred = prepare_creds();
 367	if (!override_cred)
 368		return -ENOMEM;
 369
 370	override_cred->fsuid = override_cred->uid;
 371	override_cred->fsgid = override_cred->gid;
 372
 373	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 374		/* Clear the capabilities if we switch to a non-root user */
 375		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 376		if (!uid_eq(override_cred->uid, root_uid))
 377			cap_clear(override_cred->cap_effective);
 378		else
 379			override_cred->cap_effective =
 380				override_cred->cap_permitted;
 381	}
 382
 383	old_cred = override_creds(override_cred);
 384retry:
 385	res = user_path_at(dfd, filename, lookup_flags, &path);
 386	if (res)
 387		goto out;
 388
 389	inode = d_backing_inode(path.dentry);
 390
 391	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 392		/*
 393		 * MAY_EXEC on regular files is denied if the fs is mounted
 394		 * with the "noexec" flag.
 395		 */
 396		res = -EACCES;
 397		if (path_noexec(&path))
 398			goto out_path_release;
 399	}
 400
 401	res = inode_permission(inode, mode | MAY_ACCESS);
 402	/* SuS v2 requires we report a read only fs too */
 403	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 404		goto out_path_release;
 405	/*
 406	 * This is a rare case where using __mnt_is_readonly()
 407	 * is OK without a mnt_want/drop_write() pair.  Since
 408	 * no actual write to the fs is performed here, we do
 409	 * not need to telegraph to that to anyone.
 410	 *
 411	 * By doing this, we accept that this access is
 412	 * inherently racy and know that the fs may change
 413	 * state before we even see this result.
 414	 */
 415	if (__mnt_is_readonly(path.mnt))
 416		res = -EROFS;
 417
 418out_path_release:
 419	path_put(&path);
 420	if (retry_estale(res, lookup_flags)) {
 421		lookup_flags |= LOOKUP_REVAL;
 422		goto retry;
 423	}
 424out:
 425	revert_creds(old_cred);
 426	put_cred(override_cred);
 427	return res;
 428}
 429
 430SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 431{
 432	return sys_faccessat(AT_FDCWD, filename, mode);
 433}
 434
 435SYSCALL_DEFINE1(chdir, const char __user *, filename)
 436{
 437	struct path path;
 438	int error;
 439	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 440retry:
 441	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 442	if (error)
 443		goto out;
 444
 445	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 446	if (error)
 447		goto dput_and_out;
 448
 449	set_fs_pwd(current->fs, &path);
 450
 451dput_and_out:
 452	path_put(&path);
 453	if (retry_estale(error, lookup_flags)) {
 454		lookup_flags |= LOOKUP_REVAL;
 455		goto retry;
 456	}
 457out:
 458	return error;
 459}
 460
 461SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 462{
 463	struct fd f = fdget_raw(fd);
 464	struct inode *inode;
 465	int error = -EBADF;
 466
 467	error = -EBADF;
 468	if (!f.file)
 469		goto out;
 470
 471	inode = file_inode(f.file);
 472
 473	error = -ENOTDIR;
 474	if (!S_ISDIR(inode->i_mode))
 475		goto out_putf;
 476
 477	error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 478	if (!error)
 479		set_fs_pwd(current->fs, &f.file->f_path);
 480out_putf:
 481	fdput(f);
 482out:
 483	return error;
 484}
 485
 486SYSCALL_DEFINE1(chroot, const char __user *, filename)
 487{
 488	struct path path;
 489	int error;
 490	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 491retry:
 492	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 493	if (error)
 494		goto out;
 495
 496	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 497	if (error)
 498		goto dput_and_out;
 499
 500	error = -EPERM;
 501	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 502		goto dput_and_out;
 503	error = security_path_chroot(&path);
 504	if (error)
 505		goto dput_and_out;
 506
 507	set_fs_root(current->fs, &path);
 508	error = 0;
 509dput_and_out:
 510	path_put(&path);
 511	if (retry_estale(error, lookup_flags)) {
 512		lookup_flags |= LOOKUP_REVAL;
 513		goto retry;
 514	}
 515out:
 516	return error;
 517}
 518
 519static int chmod_common(const struct path *path, umode_t mode)
 520{
 521	struct inode *inode = path->dentry->d_inode;
 522	struct inode *delegated_inode = NULL;
 523	struct iattr newattrs;
 524	int error;
 525
 526	error = mnt_want_write(path->mnt);
 527	if (error)
 528		return error;
 529retry_deleg:
 530	inode_lock(inode);
 531	error = security_path_chmod(path, mode);
 532	if (error)
 533		goto out_unlock;
 534	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 535	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 536	error = notify_change(path->dentry, &newattrs, &delegated_inode);
 537out_unlock:
 538	inode_unlock(inode);
 539	if (delegated_inode) {
 540		error = break_deleg_wait(&delegated_inode);
 541		if (!error)
 542			goto retry_deleg;
 543	}
 544	mnt_drop_write(path->mnt);
 545	return error;
 546}
 547
 548SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 549{
 550	struct fd f = fdget(fd);
 551	int err = -EBADF;
 552
 553	if (f.file) {
 554		audit_file(f.file);
 555		err = chmod_common(&f.file->f_path, mode);
 556		fdput(f);
 557	}
 558	return err;
 559}
 560
 561SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 562{
 563	struct path path;
 564	int error;
 565	unsigned int lookup_flags = LOOKUP_FOLLOW;
 566retry:
 567	error = user_path_at(dfd, filename, lookup_flags, &path);
 568	if (!error) {
 569		error = chmod_common(&path, mode);
 570		path_put(&path);
 571		if (retry_estale(error, lookup_flags)) {
 572			lookup_flags |= LOOKUP_REVAL;
 573			goto retry;
 574		}
 575	}
 576	return error;
 577}
 578
 579SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 580{
 581	return sys_fchmodat(AT_FDCWD, filename, mode);
 582}
 583
 584static int chown_common(const struct path *path, uid_t user, gid_t group)
 585{
 586	struct inode *inode = path->dentry->d_inode;
 587	struct inode *delegated_inode = NULL;
 588	int error;
 589	struct iattr newattrs;
 590	kuid_t uid;
 591	kgid_t gid;
 592
 593	uid = make_kuid(current_user_ns(), user);
 594	gid = make_kgid(current_user_ns(), group);
 595
 596retry_deleg:
 597	newattrs.ia_valid =  ATTR_CTIME;
 598	if (user != (uid_t) -1) {
 599		if (!uid_valid(uid))
 600			return -EINVAL;
 601		newattrs.ia_valid |= ATTR_UID;
 602		newattrs.ia_uid = uid;
 603	}
 604	if (group != (gid_t) -1) {
 605		if (!gid_valid(gid))
 606			return -EINVAL;
 607		newattrs.ia_valid |= ATTR_GID;
 608		newattrs.ia_gid = gid;
 609	}
 610	if (!S_ISDIR(inode->i_mode))
 611		newattrs.ia_valid |=
 612			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 613	inode_lock(inode);
 614	error = security_path_chown(path, uid, gid);
 615	if (!error)
 616		error = notify_change(path->dentry, &newattrs, &delegated_inode);
 617	inode_unlock(inode);
 618	if (delegated_inode) {
 619		error = break_deleg_wait(&delegated_inode);
 620		if (!error)
 621			goto retry_deleg;
 622	}
 623	return error;
 624}
 625
 626SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 627		gid_t, group, int, flag)
 628{
 629	struct path path;
 630	int error = -EINVAL;
 631	int lookup_flags;
 632
 633	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 634		goto out;
 635
 636	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 637	if (flag & AT_EMPTY_PATH)
 638		lookup_flags |= LOOKUP_EMPTY;
 639retry:
 640	error = user_path_at(dfd, filename, lookup_flags, &path);
 641	if (error)
 642		goto out;
 643	error = mnt_want_write(path.mnt);
 644	if (error)
 645		goto out_release;
 646	error = chown_common(&path, user, group);
 647	mnt_drop_write(path.mnt);
 648out_release:
 649	path_put(&path);
 650	if (retry_estale(error, lookup_flags)) {
 651		lookup_flags |= LOOKUP_REVAL;
 652		goto retry;
 653	}
 654out:
 655	return error;
 656}
 657
 658SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 659{
 660	return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 661}
 662
 663SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 664{
 665	return sys_fchownat(AT_FDCWD, filename, user, group,
 666			    AT_SYMLINK_NOFOLLOW);
 667}
 668
 669SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 670{
 671	struct fd f = fdget(fd);
 672	int error = -EBADF;
 673
 674	if (!f.file)
 675		goto out;
 676
 677	error = mnt_want_write_file(f.file);
 678	if (error)
 679		goto out_fput;
 680	audit_file(f.file);
 681	error = chown_common(&f.file->f_path, user, group);
 682	mnt_drop_write_file(f.file);
 683out_fput:
 684	fdput(f);
 685out:
 686	return error;
 687}
 688
 689int open_check_o_direct(struct file *f)
 690{
 691	/* NB: we're sure to have correct a_ops only after f_op->open */
 692	if (f->f_flags & O_DIRECT) {
 693		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 694			return -EINVAL;
 695	}
 696	return 0;
 697}
 698
 699static int do_dentry_open(struct file *f,
 700			  struct inode *inode,
 701			  int (*open)(struct inode *, struct file *),
 702			  const struct cred *cred)
 703{
 704	static const struct file_operations empty_fops = {};
 705	int error;
 706
 707	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 708				FMODE_PREAD | FMODE_PWRITE;
 709
 710	path_get(&f->f_path);
 711	f->f_inode = inode;
 712	f->f_mapping = inode->i_mapping;
 713
 714	if (unlikely(f->f_flags & O_PATH)) {
 715		f->f_mode = FMODE_PATH;
 716		f->f_op = &empty_fops;
 717		return 0;
 718	}
 719
 720	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 721		error = get_write_access(inode);
 722		if (unlikely(error))
 723			goto cleanup_file;
 724		error = __mnt_want_write(f->f_path.mnt);
 725		if (unlikely(error)) {
 726			put_write_access(inode);
 727			goto cleanup_file;
 728		}
 729		f->f_mode |= FMODE_WRITER;
 730	}
 731
 732	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 733	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 734		f->f_mode |= FMODE_ATOMIC_POS;
 735
 736	f->f_op = fops_get(inode->i_fop);
 737	if (unlikely(WARN_ON(!f->f_op))) {
 738		error = -ENODEV;
 739		goto cleanup_all;
 740	}
 741
 742	error = security_file_open(f, cred);
 743	if (error)
 744		goto cleanup_all;
 745
 746	error = break_lease(locks_inode(f), f->f_flags);
 747	if (error)
 748		goto cleanup_all;
 749
 750	if (!open)
 751		open = f->f_op->open;
 752	if (open) {
 753		error = open(inode, f);
 754		if (error)
 755			goto cleanup_all;
 756	}
 757	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 758		i_readcount_inc(inode);
 759	if ((f->f_mode & FMODE_READ) &&
 760	     likely(f->f_op->read || f->f_op->read_iter))
 761		f->f_mode |= FMODE_CAN_READ;
 762	if ((f->f_mode & FMODE_WRITE) &&
 763	     likely(f->f_op->write || f->f_op->write_iter))
 764		f->f_mode |= FMODE_CAN_WRITE;
 765
 766	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 767
 768	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 769
 770	return 0;
 771
 772cleanup_all:
 773	fops_put(f->f_op);
 774	if (f->f_mode & FMODE_WRITER) {
 775		put_write_access(inode);
 776		__mnt_drop_write(f->f_path.mnt);
 777	}
 778cleanup_file:
 779	path_put(&f->f_path);
 780	f->f_path.mnt = NULL;
 781	f->f_path.dentry = NULL;
 782	f->f_inode = NULL;
 783	return error;
 784}
 785
 786/**
 787 * finish_open - finish opening a file
 788 * @file: file pointer
 789 * @dentry: pointer to dentry
 790 * @open: open callback
 791 * @opened: state of open
 792 *
 793 * This can be used to finish opening a file passed to i_op->atomic_open().
 794 *
 795 * If the open callback is set to NULL, then the standard f_op->open()
 796 * filesystem callback is substituted.
 797 *
 798 * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 799 * the return value of d_splice_alias(), then the caller needs to perform dput()
 800 * on it after finish_open().
 801 *
 802 * On successful return @file is a fully instantiated open file.  After this, if
 803 * an error occurs in ->atomic_open(), it needs to clean up with fput().
 804 *
 805 * Returns zero on success or -errno if the open failed.
 806 */
 807int finish_open(struct file *file, struct dentry *dentry,
 808		int (*open)(struct inode *, struct file *),
 809		int *opened)
 810{
 811	int error;
 812	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 813
 814	file->f_path.dentry = dentry;
 815	error = do_dentry_open(file, d_backing_inode(dentry), open,
 816			       current_cred());
 817	if (!error)
 818		*opened |= FILE_OPENED;
 819
 820	return error;
 821}
 822EXPORT_SYMBOL(finish_open);
 823
 824/**
 825 * finish_no_open - finish ->atomic_open() without opening the file
 826 *
 827 * @file: file pointer
 828 * @dentry: dentry or NULL (as returned from ->lookup())
 829 *
 830 * This can be used to set the result of a successful lookup in ->atomic_open().
 831 *
 832 * NB: unlike finish_open() this function does consume the dentry reference and
 833 * the caller need not dput() it.
 834 *
 835 * Returns "1" which must be the return value of ->atomic_open() after having
 836 * called this function.
 837 */
 838int finish_no_open(struct file *file, struct dentry *dentry)
 839{
 840	file->f_path.dentry = dentry;
 841	return 1;
 842}
 843EXPORT_SYMBOL(finish_no_open);
 844
 845char *file_path(struct file *filp, char *buf, int buflen)
 846{
 847	return d_path(&filp->f_path, buf, buflen);
 848}
 849EXPORT_SYMBOL(file_path);
 850
 851/**
 852 * vfs_open - open the file at the given path
 853 * @path: path to open
 854 * @file: newly allocated file with f_flag initialized
 855 * @cred: credentials to use
 856 */
 857int vfs_open(const struct path *path, struct file *file,
 858	     const struct cred *cred)
 859{
 860	struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 861
 862	if (IS_ERR(dentry))
 863		return PTR_ERR(dentry);
 864
 865	file->f_path = *path;
 866	return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 867}
 868
 869struct file *dentry_open(const struct path *path, int flags,
 870			 const struct cred *cred)
 871{
 872	int error;
 873	struct file *f;
 874
 875	validate_creds(cred);
 876
 877	/* We must always pass in a valid mount pointer. */
 878	BUG_ON(!path->mnt);
 879
 880	f = get_empty_filp();
 881	if (!IS_ERR(f)) {
 882		f->f_flags = flags;
 883		error = vfs_open(path, f, cred);
 884		if (!error) {
 885			/* from now on we need fput() to dispose of f */
 886			error = open_check_o_direct(f);
 887			if (error) {
 888				fput(f);
 889				f = ERR_PTR(error);
 890			}
 891		} else { 
 892			put_filp(f);
 893			f = ERR_PTR(error);
 894		}
 895	}
 896	return f;
 897}
 898EXPORT_SYMBOL(dentry_open);
 899
 900static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 901{
 902	int lookup_flags = 0;
 903	int acc_mode = ACC_MODE(flags);
 904
 905	if (flags & (O_CREAT | __O_TMPFILE))
 906		op->mode = (mode & S_IALLUGO) | S_IFREG;
 907	else
 908		op->mode = 0;
 909
 910	/* Must never be set by userspace */
 911	flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 912
 913	/*
 914	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 915	 * check for O_DSYNC if the need any syncing at all we enforce it's
 916	 * always set instead of having to deal with possibly weird behaviour
 917	 * for malicious applications setting only __O_SYNC.
 918	 */
 919	if (flags & __O_SYNC)
 920		flags |= O_DSYNC;
 921
 922	if (flags & __O_TMPFILE) {
 923		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 924			return -EINVAL;
 925		if (!(acc_mode & MAY_WRITE))
 926			return -EINVAL;
 927	} else if (flags & O_PATH) {
 928		/*
 929		 * If we have O_PATH in the open flag. Then we
 930		 * cannot have anything other than the below set of flags
 931		 */
 932		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 933		acc_mode = 0;
 934	}
 935
 936	op->open_flag = flags;
 937
 938	/* O_TRUNC implies we need access checks for write permissions */
 939	if (flags & O_TRUNC)
 940		acc_mode |= MAY_WRITE;
 941
 942	/* Allow the LSM permission hook to distinguish append
 943	   access from general write access. */
 944	if (flags & O_APPEND)
 945		acc_mode |= MAY_APPEND;
 946
 947	op->acc_mode = acc_mode;
 948
 949	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 950
 951	if (flags & O_CREAT) {
 952		op->intent |= LOOKUP_CREATE;
 953		if (flags & O_EXCL)
 954			op->intent |= LOOKUP_EXCL;
 955	}
 956
 957	if (flags & O_DIRECTORY)
 958		lookup_flags |= LOOKUP_DIRECTORY;
 959	if (!(flags & O_NOFOLLOW))
 960		lookup_flags |= LOOKUP_FOLLOW;
 961	op->lookup_flags = lookup_flags;
 962	return 0;
 963}
 964
 965/**
 966 * file_open_name - open file and return file pointer
 967 *
 968 * @name:	struct filename containing path to open
 969 * @flags:	open flags as per the open(2) second argument
 970 * @mode:	mode for the new file if O_CREAT is set, else ignored
 971 *
 972 * This is the helper to open a file from kernelspace if you really
 973 * have to.  But in generally you should not do this, so please move
 974 * along, nothing to see here..
 975 */
 976struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 977{
 978	struct open_flags op;
 979	int err = build_open_flags(flags, mode, &op);
 980	return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 981}
 982
 983/**
 984 * filp_open - open file and return file pointer
 985 *
 986 * @filename:	path to open
 987 * @flags:	open flags as per the open(2) second argument
 988 * @mode:	mode for the new file if O_CREAT is set, else ignored
 989 *
 990 * This is the helper to open a file from kernelspace if you really
 991 * have to.  But in generally you should not do this, so please move
 992 * along, nothing to see here..
 993 */
 994struct file *filp_open(const char *filename, int flags, umode_t mode)
 995{
 996	struct filename *name = getname_kernel(filename);
 997	struct file *file = ERR_CAST(name);
 998	
 999	if (!IS_ERR(name)) {
1000		file = file_open_name(name, flags, mode);
1001		putname(name);
1002	}
1003	return file;
1004}
1005EXPORT_SYMBOL(filp_open);
1006
1007struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1008			    const char *filename, int flags, umode_t mode)
1009{
1010	struct open_flags op;
1011	int err = build_open_flags(flags, mode, &op);
1012	if (err)
1013		return ERR_PTR(err);
1014	return do_file_open_root(dentry, mnt, filename, &op);
1015}
1016EXPORT_SYMBOL(file_open_root);
1017
1018struct file *filp_clone_open(struct file *oldfile)
1019{
1020	struct file *file;
1021	int retval;
1022
1023	file = get_empty_filp();
1024	if (IS_ERR(file))
1025		return file;
1026
1027	file->f_flags = oldfile->f_flags;
1028	retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1029	if (retval) {
1030		put_filp(file);
1031		return ERR_PTR(retval);
1032	}
1033
1034	return file;
1035}
1036EXPORT_SYMBOL(filp_clone_open);
1037
1038long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1039{
1040	struct open_flags op;
1041	int fd = build_open_flags(flags, mode, &op);
1042	struct filename *tmp;
1043
1044	if (fd)
1045		return fd;
1046
1047	tmp = getname(filename);
1048	if (IS_ERR(tmp))
1049		return PTR_ERR(tmp);
1050
1051	fd = get_unused_fd_flags(flags);
1052	if (fd >= 0) {
1053		struct file *f = do_filp_open(dfd, tmp, &op);
1054		if (IS_ERR(f)) {
1055			put_unused_fd(fd);
1056			fd = PTR_ERR(f);
1057		} else {
1058			fsnotify_open(f);
1059			fd_install(fd, f);
1060		}
1061	}
1062	putname(tmp);
1063	return fd;
1064}
1065
1066SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1067{
1068	if (force_o_largefile())
1069		flags |= O_LARGEFILE;
1070
1071	return do_sys_open(AT_FDCWD, filename, flags, mode);
1072}
1073
1074SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1075		umode_t, mode)
1076{
1077	if (force_o_largefile())
1078		flags |= O_LARGEFILE;
1079
1080	return do_sys_open(dfd, filename, flags, mode);
1081}
1082
1083#ifndef __alpha__
1084
1085/*
1086 * For backward compatibility?  Maybe this should be moved
1087 * into arch/i386 instead?
1088 */
1089SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1090{
1091	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1092}
1093
1094#endif
1095
1096/*
1097 * "id" is the POSIX thread ID. We use the
1098 * files pointer for this..
1099 */
1100int filp_close(struct file *filp, fl_owner_t id)
1101{
1102	int retval = 0;
1103
1104	if (!file_count(filp)) {
1105		printk(KERN_ERR "VFS: Close: file count is 0\n");
1106		return 0;
1107	}
1108
1109	if (filp->f_op->flush)
1110		retval = filp->f_op->flush(filp, id);
1111
1112	if (likely(!(filp->f_mode & FMODE_PATH))) {
1113		dnotify_flush(filp, id);
1114		locks_remove_posix(filp, id);
1115	}
1116	fput(filp);
1117	return retval;
1118}
1119
1120EXPORT_SYMBOL(filp_close);
1121
1122/*
1123 * Careful here! We test whether the file pointer is NULL before
1124 * releasing the fd. This ensures that one clone task can't release
1125 * an fd while another clone is opening it.
1126 */
1127SYSCALL_DEFINE1(close, unsigned int, fd)
1128{
1129	int retval = __close_fd(current->files, fd);
1130
1131	/* can't restart close syscall because file table entry was cleared */
1132	if (unlikely(retval == -ERESTARTSYS ||
1133		     retval == -ERESTARTNOINTR ||
1134		     retval == -ERESTARTNOHAND ||
1135		     retval == -ERESTART_RESTARTBLOCK))
1136		retval = -EINTR;
1137
1138	return retval;
1139}
1140EXPORT_SYMBOL(sys_close);
1141
1142/*
1143 * This routine simulates a hangup on the tty, to arrange that users
1144 * are given clean terminals at login time.
1145 */
1146SYSCALL_DEFINE0(vhangup)
1147{
1148	if (capable(CAP_SYS_TTY_CONFIG)) {
1149		tty_vhangup_self();
1150		return 0;
1151	}
1152	return -EPERM;
1153}
1154
1155/*
1156 * Called when an inode is about to be open.
1157 * We use this to disallow opening large files on 32bit systems if
1158 * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1159 * on this flag in sys_open.
1160 */
1161int generic_file_open(struct inode * inode, struct file * filp)
1162{
1163	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1164		return -EOVERFLOW;
1165	return 0;
1166}
1167
1168EXPORT_SYMBOL(generic_file_open);
1169
1170/*
1171 * This is used by subsystems that don't want seekable
1172 * file descriptors. The function is not supposed to ever fail, the only
1173 * reason it returns an 'int' and not 'void' is so that it can be plugged
1174 * directly into file_operations structure.
1175 */
1176int nonseekable_open(struct inode *inode, struct file *filp)
1177{
1178	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1179	return 0;
1180}
1181
1182EXPORT_SYMBOL(nonseekable_open);
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);