xref: /openbmc/linux/fs/open.c (revision 8a0a6c9af053fb93c0edf4581518c77fd131803e)
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  #include <linux/mnt_idmapping.h>
36  #include <linux/filelock.h>
37  
38  #include "internal.h"
39  
40  int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
41  		loff_t length, unsigned int time_attrs, struct file *filp)
42  {
43  	int ret;
44  	struct iattr newattrs;
45  
46  	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
47  	if (length < 0)
48  		return -EINVAL;
49  
50  	newattrs.ia_size = length;
51  	newattrs.ia_valid = ATTR_SIZE | time_attrs;
52  	if (filp) {
53  		newattrs.ia_file = filp;
54  		newattrs.ia_valid |= ATTR_FILE;
55  	}
56  
57  	/* Remove suid, sgid, and file capabilities on truncate too */
58  	ret = dentry_needs_remove_privs(idmap, dentry);
59  	if (ret < 0)
60  		return ret;
61  	if (ret)
62  		newattrs.ia_valid |= ret | ATTR_FORCE;
63  
64  	inode_lock(dentry->d_inode);
65  	/* Note any delegations or leases have already been broken: */
66  	ret = notify_change(idmap, dentry, &newattrs, NULL);
67  	inode_unlock(dentry->d_inode);
68  	return ret;
69  }
70  
71  long vfs_truncate(const struct path *path, loff_t length)
72  {
73  	struct mnt_idmap *idmap;
74  	struct inode *inode;
75  	long error;
76  
77  	inode = path->dentry->d_inode;
78  
79  	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80  	if (S_ISDIR(inode->i_mode))
81  		return -EISDIR;
82  	if (!S_ISREG(inode->i_mode))
83  		return -EINVAL;
84  
85  	error = mnt_want_write(path->mnt);
86  	if (error)
87  		goto out;
88  
89  	idmap = mnt_idmap(path->mnt);
90  	error = inode_permission(idmap, inode, MAY_WRITE);
91  	if (error)
92  		goto mnt_drop_write_and_out;
93  
94  	error = -EPERM;
95  	if (IS_APPEND(inode))
96  		goto mnt_drop_write_and_out;
97  
98  	error = get_write_access(inode);
99  	if (error)
100  		goto mnt_drop_write_and_out;
101  
102  	/*
103  	 * Make sure that there are no leases.  get_write_access() protects
104  	 * against the truncate racing with a lease-granting setlease().
105  	 */
106  	error = break_lease(inode, O_WRONLY);
107  	if (error)
108  		goto put_write_and_out;
109  
110  	error = security_path_truncate(path);
111  	if (!error)
112  		error = do_truncate(idmap, path->dentry, length, 0, NULL);
113  
114  put_write_and_out:
115  	put_write_access(inode);
116  mnt_drop_write_and_out:
117  	mnt_drop_write(path->mnt);
118  out:
119  	return error;
120  }
121  EXPORT_SYMBOL_GPL(vfs_truncate);
122  
123  long do_sys_truncate(const char __user *pathname, loff_t length)
124  {
125  	unsigned int lookup_flags = LOOKUP_FOLLOW;
126  	struct path path;
127  	int error;
128  
129  	if (length < 0)	/* sorry, but loff_t says... */
130  		return -EINVAL;
131  
132  retry:
133  	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
134  	if (!error) {
135  		error = vfs_truncate(&path, length);
136  		path_put(&path);
137  	}
138  	if (retry_estale(error, lookup_flags)) {
139  		lookup_flags |= LOOKUP_REVAL;
140  		goto retry;
141  	}
142  	return error;
143  }
144  
145  SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
146  {
147  	return do_sys_truncate(path, length);
148  }
149  
150  #ifdef CONFIG_COMPAT
151  COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
152  {
153  	return do_sys_truncate(path, length);
154  }
155  #endif
156  
157  long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
158  {
159  	struct inode *inode;
160  	struct dentry *dentry;
161  	struct fd f;
162  	int error;
163  
164  	error = -EINVAL;
165  	if (length < 0)
166  		goto out;
167  	error = -EBADF;
168  	f = fdget(fd);
169  	if (!f.file)
170  		goto out;
171  
172  	/* explicitly opened as large or we are on 64-bit box */
173  	if (f.file->f_flags & O_LARGEFILE)
174  		small = 0;
175  
176  	dentry = f.file->f_path.dentry;
177  	inode = dentry->d_inode;
178  	error = -EINVAL;
179  	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
180  		goto out_putf;
181  
182  	error = -EINVAL;
183  	/* Cannot ftruncate over 2^31 bytes without large file support */
184  	if (small && length > MAX_NON_LFS)
185  		goto out_putf;
186  
187  	error = -EPERM;
188  	/* Check IS_APPEND on real upper inode */
189  	if (IS_APPEND(file_inode(f.file)))
190  		goto out_putf;
191  	sb_start_write(inode->i_sb);
192  	error = security_file_truncate(f.file);
193  	if (!error)
194  		error = do_truncate(file_mnt_idmap(f.file), dentry, length,
195  				    ATTR_MTIME | ATTR_CTIME, f.file);
196  	sb_end_write(inode->i_sb);
197  out_putf:
198  	fdput(f);
199  out:
200  	return error;
201  }
202  
203  SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204  {
205  	return do_sys_ftruncate(fd, length, 1);
206  }
207  
208  #ifdef CONFIG_COMPAT
209  COMPAT_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
217  SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218  {
219  	return do_sys_truncate(path, length);
220  }
221  
222  SYSCALL_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  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
229  COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
230  		       compat_arg_u64_dual(length))
231  {
232  	return ksys_truncate(pathname, compat_arg_u64_glue(length));
233  }
234  #endif
235  
236  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
237  COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
238  		       compat_arg_u64_dual(length))
239  {
240  	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
241  }
242  #endif
243  
244  int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
245  {
246  	struct inode *inode = file_inode(file);
247  	long ret;
248  
249  	if (offset < 0 || len <= 0)
250  		return -EINVAL;
251  
252  	/* Return error if mode is not supported */
253  	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
254  		return -EOPNOTSUPP;
255  
256  	/* Punch hole and zero range are mutually exclusive */
257  	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
258  	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
259  		return -EOPNOTSUPP;
260  
261  	/* Punch hole must have keep size set */
262  	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
263  	    !(mode & FALLOC_FL_KEEP_SIZE))
264  		return -EOPNOTSUPP;
265  
266  	/* Collapse range should only be used exclusively. */
267  	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
268  	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
269  		return -EINVAL;
270  
271  	/* Insert range should only be used exclusively. */
272  	if ((mode & FALLOC_FL_INSERT_RANGE) &&
273  	    (mode & ~FALLOC_FL_INSERT_RANGE))
274  		return -EINVAL;
275  
276  	/* Unshare range should only be used with allocate mode. */
277  	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
278  	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
279  		return -EINVAL;
280  
281  	if (!(file->f_mode & FMODE_WRITE))
282  		return -EBADF;
283  
284  	/*
285  	 * We can only allow pure fallocate on append only files
286  	 */
287  	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
288  		return -EPERM;
289  
290  	if (IS_IMMUTABLE(inode))
291  		return -EPERM;
292  
293  	/*
294  	 * We cannot allow any fallocate operation on an active swapfile
295  	 */
296  	if (IS_SWAPFILE(inode))
297  		return -ETXTBSY;
298  
299  	/*
300  	 * Revalidate the write permissions, in case security policy has
301  	 * changed since the files were opened.
302  	 */
303  	ret = security_file_permission(file, MAY_WRITE);
304  	if (ret)
305  		return ret;
306  
307  	if (S_ISFIFO(inode->i_mode))
308  		return -ESPIPE;
309  
310  	if (S_ISDIR(inode->i_mode))
311  		return -EISDIR;
312  
313  	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
314  		return -ENODEV;
315  
316  	/* Check for wrap through zero too */
317  	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
318  		return -EFBIG;
319  
320  	if (!file->f_op->fallocate)
321  		return -EOPNOTSUPP;
322  
323  	file_start_write(file);
324  	ret = file->f_op->fallocate(file, mode, offset, len);
325  
326  	/*
327  	 * Create inotify and fanotify events.
328  	 *
329  	 * To keep the logic simple always create events if fallocate succeeds.
330  	 * This implies that events are even created if the file size remains
331  	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
332  	 */
333  	if (ret == 0)
334  		fsnotify_modify(file);
335  
336  	file_end_write(file);
337  	return ret;
338  }
339  EXPORT_SYMBOL_GPL(vfs_fallocate);
340  
341  int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
342  {
343  	struct fd f = fdget(fd);
344  	int error = -EBADF;
345  
346  	if (f.file) {
347  		error = vfs_fallocate(f.file, mode, offset, len);
348  		fdput(f);
349  	}
350  	return error;
351  }
352  
353  SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
354  {
355  	return ksys_fallocate(fd, mode, offset, len);
356  }
357  
358  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
359  COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
360  		       compat_arg_u64_dual(len))
361  {
362  	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
363  			      compat_arg_u64_glue(len));
364  }
365  #endif
366  
367  /*
368   * access() needs to use the real uid/gid, not the effective uid/gid.
369   * We do this by temporarily clearing all FS-related capabilities and
370   * switching the fsuid/fsgid around to the real ones.
371   *
372   * Creating new credentials is expensive, so we try to skip doing it,
373   * which we can if the result would match what we already got.
374   */
375  static bool access_need_override_creds(int flags)
376  {
377  	const struct cred *cred;
378  
379  	if (flags & AT_EACCESS)
380  		return false;
381  
382  	cred = current_cred();
383  	if (!uid_eq(cred->fsuid, cred->uid) ||
384  	    !gid_eq(cred->fsgid, cred->gid))
385  		return true;
386  
387  	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
388  		kuid_t root_uid = make_kuid(cred->user_ns, 0);
389  		if (!uid_eq(cred->uid, root_uid)) {
390  			if (!cap_isclear(cred->cap_effective))
391  				return true;
392  		} else {
393  			if (!cap_isidentical(cred->cap_effective,
394  			    cred->cap_permitted))
395  				return true;
396  		}
397  	}
398  
399  	return false;
400  }
401  
402  static const struct cred *access_override_creds(void)
403  {
404  	const struct cred *old_cred;
405  	struct cred *override_cred;
406  
407  	override_cred = prepare_creds();
408  	if (!override_cred)
409  		return NULL;
410  
411  	/*
412  	 * XXX access_need_override_creds performs checks in hopes of skipping
413  	 * this work. Make sure it stays in sync if making any changes in this
414  	 * routine.
415  	 */
416  
417  	override_cred->fsuid = override_cred->uid;
418  	override_cred->fsgid = override_cred->gid;
419  
420  	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
421  		/* Clear the capabilities if we switch to a non-root user */
422  		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
423  		if (!uid_eq(override_cred->uid, root_uid))
424  			cap_clear(override_cred->cap_effective);
425  		else
426  			override_cred->cap_effective =
427  				override_cred->cap_permitted;
428  	}
429  
430  	/*
431  	 * The new set of credentials can *only* be used in
432  	 * task-synchronous circumstances, and does not need
433  	 * RCU freeing, unless somebody then takes a separate
434  	 * reference to it.
435  	 *
436  	 * NOTE! This is _only_ true because this credential
437  	 * is used purely for override_creds() that installs
438  	 * it as the subjective cred. Other threads will be
439  	 * accessing ->real_cred, not the subjective cred.
440  	 *
441  	 * If somebody _does_ make a copy of this (using the
442  	 * 'get_current_cred()' function), that will clear the
443  	 * non_rcu field, because now that other user may be
444  	 * expecting RCU freeing. But normal thread-synchronous
445  	 * cred accesses will keep things non-RCY.
446  	 */
447  	override_cred->non_rcu = 1;
448  
449  	old_cred = override_creds(override_cred);
450  
451  	/* override_cred() gets its own ref */
452  	put_cred(override_cred);
453  
454  	return old_cred;
455  }
456  
457  static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
458  {
459  	struct path path;
460  	struct inode *inode;
461  	int res;
462  	unsigned int lookup_flags = LOOKUP_FOLLOW;
463  	const struct cred *old_cred = NULL;
464  
465  	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
466  		return -EINVAL;
467  
468  	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
469  		return -EINVAL;
470  
471  	if (flags & AT_SYMLINK_NOFOLLOW)
472  		lookup_flags &= ~LOOKUP_FOLLOW;
473  	if (flags & AT_EMPTY_PATH)
474  		lookup_flags |= LOOKUP_EMPTY;
475  
476  	if (access_need_override_creds(flags)) {
477  		old_cred = access_override_creds();
478  		if (!old_cred)
479  			return -ENOMEM;
480  	}
481  
482  retry:
483  	res = user_path_at(dfd, filename, lookup_flags, &path);
484  	if (res)
485  		goto out;
486  
487  	inode = d_backing_inode(path.dentry);
488  
489  	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
490  		/*
491  		 * MAY_EXEC on regular files is denied if the fs is mounted
492  		 * with the "noexec" flag.
493  		 */
494  		res = -EACCES;
495  		if (path_noexec(&path))
496  			goto out_path_release;
497  	}
498  
499  	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
500  	/* SuS v2 requires we report a read only fs too */
501  	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
502  		goto out_path_release;
503  	/*
504  	 * This is a rare case where using __mnt_is_readonly()
505  	 * is OK without a mnt_want/drop_write() pair.  Since
506  	 * no actual write to the fs is performed here, we do
507  	 * not need to telegraph to that to anyone.
508  	 *
509  	 * By doing this, we accept that this access is
510  	 * inherently racy and know that the fs may change
511  	 * state before we even see this result.
512  	 */
513  	if (__mnt_is_readonly(path.mnt))
514  		res = -EROFS;
515  
516  out_path_release:
517  	path_put(&path);
518  	if (retry_estale(res, lookup_flags)) {
519  		lookup_flags |= LOOKUP_REVAL;
520  		goto retry;
521  	}
522  out:
523  	if (old_cred)
524  		revert_creds(old_cred);
525  
526  	return res;
527  }
528  
529  SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
530  {
531  	return do_faccessat(dfd, filename, mode, 0);
532  }
533  
534  SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
535  		int, flags)
536  {
537  	return do_faccessat(dfd, filename, mode, flags);
538  }
539  
540  SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
541  {
542  	return do_faccessat(AT_FDCWD, filename, mode, 0);
543  }
544  
545  SYSCALL_DEFINE1(chdir, const char __user *, filename)
546  {
547  	struct path path;
548  	int error;
549  	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
550  retry:
551  	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
552  	if (error)
553  		goto out;
554  
555  	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
556  	if (error)
557  		goto dput_and_out;
558  
559  	set_fs_pwd(current->fs, &path);
560  
561  dput_and_out:
562  	path_put(&path);
563  	if (retry_estale(error, lookup_flags)) {
564  		lookup_flags |= LOOKUP_REVAL;
565  		goto retry;
566  	}
567  out:
568  	return error;
569  }
570  
571  SYSCALL_DEFINE1(fchdir, unsigned int, fd)
572  {
573  	struct fd f = fdget_raw(fd);
574  	int error;
575  
576  	error = -EBADF;
577  	if (!f.file)
578  		goto out;
579  
580  	error = -ENOTDIR;
581  	if (!d_can_lookup(f.file->f_path.dentry))
582  		goto out_putf;
583  
584  	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
585  	if (!error)
586  		set_fs_pwd(current->fs, &f.file->f_path);
587  out_putf:
588  	fdput(f);
589  out:
590  	return error;
591  }
592  
593  SYSCALL_DEFINE1(chroot, const char __user *, filename)
594  {
595  	struct path path;
596  	int error;
597  	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
598  retry:
599  	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
600  	if (error)
601  		goto out;
602  
603  	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
604  	if (error)
605  		goto dput_and_out;
606  
607  	error = -EPERM;
608  	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
609  		goto dput_and_out;
610  	error = security_path_chroot(&path);
611  	if (error)
612  		goto dput_and_out;
613  
614  	set_fs_root(current->fs, &path);
615  	error = 0;
616  dput_and_out:
617  	path_put(&path);
618  	if (retry_estale(error, lookup_flags)) {
619  		lookup_flags |= LOOKUP_REVAL;
620  		goto retry;
621  	}
622  out:
623  	return error;
624  }
625  
626  int chmod_common(const struct path *path, umode_t mode)
627  {
628  	struct inode *inode = path->dentry->d_inode;
629  	struct inode *delegated_inode = NULL;
630  	struct iattr newattrs;
631  	int error;
632  
633  	error = mnt_want_write(path->mnt);
634  	if (error)
635  		return error;
636  retry_deleg:
637  	inode_lock(inode);
638  	error = security_path_chmod(path, mode);
639  	if (error)
640  		goto out_unlock;
641  	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
642  	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
643  	error = notify_change(mnt_idmap(path->mnt), path->dentry,
644  			      &newattrs, &delegated_inode);
645  out_unlock:
646  	inode_unlock(inode);
647  	if (delegated_inode) {
648  		error = break_deleg_wait(&delegated_inode);
649  		if (!error)
650  			goto retry_deleg;
651  	}
652  	mnt_drop_write(path->mnt);
653  	return error;
654  }
655  
656  int vfs_fchmod(struct file *file, umode_t mode)
657  {
658  	audit_file(file);
659  	return chmod_common(&file->f_path, mode);
660  }
661  
662  SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
663  {
664  	struct fd f = fdget(fd);
665  	int err = -EBADF;
666  
667  	if (f.file) {
668  		err = vfs_fchmod(f.file, mode);
669  		fdput(f);
670  	}
671  	return err;
672  }
673  
674  static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
675  {
676  	struct path path;
677  	int error;
678  	unsigned int lookup_flags = LOOKUP_FOLLOW;
679  retry:
680  	error = user_path_at(dfd, filename, lookup_flags, &path);
681  	if (!error) {
682  		error = chmod_common(&path, mode);
683  		path_put(&path);
684  		if (retry_estale(error, lookup_flags)) {
685  			lookup_flags |= LOOKUP_REVAL;
686  			goto retry;
687  		}
688  	}
689  	return error;
690  }
691  
692  SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
693  		umode_t, mode)
694  {
695  	return do_fchmodat(dfd, filename, mode);
696  }
697  
698  SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
699  {
700  	return do_fchmodat(AT_FDCWD, filename, mode);
701  }
702  
703  /**
704   * setattr_vfsuid - check and set ia_fsuid attribute
705   * @kuid: new inode owner
706   *
707   * Check whether @kuid is valid and if so generate and set vfsuid_t in
708   * ia_vfsuid.
709   *
710   * Return: true if @kuid is valid, false if not.
711   */
712  static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
713  {
714  	if (!uid_valid(kuid))
715  		return false;
716  	attr->ia_valid |= ATTR_UID;
717  	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
718  	return true;
719  }
720  
721  /**
722   * setattr_vfsgid - check and set ia_fsgid attribute
723   * @kgid: new inode owner
724   *
725   * Check whether @kgid is valid and if so generate and set vfsgid_t in
726   * ia_vfsgid.
727   *
728   * Return: true if @kgid is valid, false if not.
729   */
730  static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
731  {
732  	if (!gid_valid(kgid))
733  		return false;
734  	attr->ia_valid |= ATTR_GID;
735  	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
736  	return true;
737  }
738  
739  int chown_common(const struct path *path, uid_t user, gid_t group)
740  {
741  	struct mnt_idmap *idmap;
742  	struct user_namespace *fs_userns;
743  	struct inode *inode = path->dentry->d_inode;
744  	struct inode *delegated_inode = NULL;
745  	int error;
746  	struct iattr newattrs;
747  	kuid_t uid;
748  	kgid_t gid;
749  
750  	uid = make_kuid(current_user_ns(), user);
751  	gid = make_kgid(current_user_ns(), group);
752  
753  	idmap = mnt_idmap(path->mnt);
754  	fs_userns = i_user_ns(inode);
755  
756  retry_deleg:
757  	newattrs.ia_vfsuid = INVALID_VFSUID;
758  	newattrs.ia_vfsgid = INVALID_VFSGID;
759  	newattrs.ia_valid =  ATTR_CTIME;
760  	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
761  		return -EINVAL;
762  	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
763  		return -EINVAL;
764  	inode_lock(inode);
765  	if (!S_ISDIR(inode->i_mode))
766  		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
767  				     setattr_should_drop_sgid(idmap, inode);
768  	/* Continue to send actual fs values, not the mount values. */
769  	error = security_path_chown(
770  		path,
771  		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
772  		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
773  	if (!error)
774  		error = notify_change(idmap, path->dentry, &newattrs,
775  				      &delegated_inode);
776  	inode_unlock(inode);
777  	if (delegated_inode) {
778  		error = break_deleg_wait(&delegated_inode);
779  		if (!error)
780  			goto retry_deleg;
781  	}
782  	return error;
783  }
784  
785  int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
786  		int flag)
787  {
788  	struct path path;
789  	int error = -EINVAL;
790  	int lookup_flags;
791  
792  	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
793  		goto out;
794  
795  	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
796  	if (flag & AT_EMPTY_PATH)
797  		lookup_flags |= LOOKUP_EMPTY;
798  retry:
799  	error = user_path_at(dfd, filename, lookup_flags, &path);
800  	if (error)
801  		goto out;
802  	error = mnt_want_write(path.mnt);
803  	if (error)
804  		goto out_release;
805  	error = chown_common(&path, user, group);
806  	mnt_drop_write(path.mnt);
807  out_release:
808  	path_put(&path);
809  	if (retry_estale(error, lookup_flags)) {
810  		lookup_flags |= LOOKUP_REVAL;
811  		goto retry;
812  	}
813  out:
814  	return error;
815  }
816  
817  SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
818  		gid_t, group, int, flag)
819  {
820  	return do_fchownat(dfd, filename, user, group, flag);
821  }
822  
823  SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
824  {
825  	return do_fchownat(AT_FDCWD, filename, user, group, 0);
826  }
827  
828  SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
829  {
830  	return do_fchownat(AT_FDCWD, filename, user, group,
831  			   AT_SYMLINK_NOFOLLOW);
832  }
833  
834  int vfs_fchown(struct file *file, uid_t user, gid_t group)
835  {
836  	int error;
837  
838  	error = mnt_want_write_file(file);
839  	if (error)
840  		return error;
841  	audit_file(file);
842  	error = chown_common(&file->f_path, user, group);
843  	mnt_drop_write_file(file);
844  	return error;
845  }
846  
847  int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
848  {
849  	struct fd f = fdget(fd);
850  	int error = -EBADF;
851  
852  	if (f.file) {
853  		error = vfs_fchown(f.file, user, group);
854  		fdput(f);
855  	}
856  	return error;
857  }
858  
859  SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
860  {
861  	return ksys_fchown(fd, user, group);
862  }
863  
864  static int do_dentry_open(struct file *f,
865  			  struct inode *inode,
866  			  int (*open)(struct inode *, struct file *))
867  {
868  	static const struct file_operations empty_fops = {};
869  	int error;
870  
871  	path_get(&f->f_path);
872  	f->f_inode = inode;
873  	f->f_mapping = inode->i_mapping;
874  	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
875  	f->f_sb_err = file_sample_sb_err(f);
876  
877  	if (unlikely(f->f_flags & O_PATH)) {
878  		f->f_mode = FMODE_PATH | FMODE_OPENED;
879  		f->f_op = &empty_fops;
880  		return 0;
881  	}
882  
883  	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
884  		i_readcount_inc(inode);
885  	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
886  		error = get_write_access(inode);
887  		if (unlikely(error))
888  			goto cleanup_file;
889  		error = __mnt_want_write(f->f_path.mnt);
890  		if (unlikely(error)) {
891  			put_write_access(inode);
892  			goto cleanup_file;
893  		}
894  		f->f_mode |= FMODE_WRITER;
895  	}
896  
897  	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
898  	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
899  		f->f_mode |= FMODE_ATOMIC_POS;
900  
901  	f->f_op = fops_get(inode->i_fop);
902  	if (WARN_ON(!f->f_op)) {
903  		error = -ENODEV;
904  		goto cleanup_all;
905  	}
906  
907  	error = security_file_open(f);
908  	if (error)
909  		goto cleanup_all;
910  
911  	error = break_lease(file_inode(f), f->f_flags);
912  	if (error)
913  		goto cleanup_all;
914  
915  	/* normally all 3 are set; ->open() can clear them if needed */
916  	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
917  	if (!open)
918  		open = f->f_op->open;
919  	if (open) {
920  		error = open(inode, f);
921  		if (error)
922  			goto cleanup_all;
923  	}
924  	f->f_mode |= FMODE_OPENED;
925  	if ((f->f_mode & FMODE_READ) &&
926  	     likely(f->f_op->read || f->f_op->read_iter))
927  		f->f_mode |= FMODE_CAN_READ;
928  	if ((f->f_mode & FMODE_WRITE) &&
929  	     likely(f->f_op->write || f->f_op->write_iter))
930  		f->f_mode |= FMODE_CAN_WRITE;
931  	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
932  		f->f_mode &= ~FMODE_LSEEK;
933  	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
934  		f->f_mode |= FMODE_CAN_ODIRECT;
935  
936  	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
937  	f->f_iocb_flags = iocb_flags(f);
938  
939  	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
940  
941  	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
942  		return -EINVAL;
943  
944  	/*
945  	 * XXX: Huge page cache doesn't support writing yet. Drop all page
946  	 * cache for this file before processing writes.
947  	 */
948  	if (f->f_mode & FMODE_WRITE) {
949  		/*
950  		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
951  		 * is up to date and the update to i_writecount by
952  		 * get_write_access() is visible. Ensures subsequent insertion
953  		 * of THPs into the page cache will fail.
954  		 */
955  		smp_mb();
956  		if (filemap_nr_thps(inode->i_mapping)) {
957  			struct address_space *mapping = inode->i_mapping;
958  
959  			filemap_invalidate_lock(inode->i_mapping);
960  			/*
961  			 * unmap_mapping_range just need to be called once
962  			 * here, because the private pages is not need to be
963  			 * unmapped mapping (e.g. data segment of dynamic
964  			 * shared libraries here).
965  			 */
966  			unmap_mapping_range(mapping, 0, 0, 0);
967  			truncate_inode_pages(mapping, 0);
968  			filemap_invalidate_unlock(inode->i_mapping);
969  		}
970  	}
971  
972  	return 0;
973  
974  cleanup_all:
975  	if (WARN_ON_ONCE(error > 0))
976  		error = -EINVAL;
977  	fops_put(f->f_op);
978  	put_file_access(f);
979  cleanup_file:
980  	path_put(&f->f_path);
981  	f->f_path.mnt = NULL;
982  	f->f_path.dentry = NULL;
983  	f->f_inode = NULL;
984  	return error;
985  }
986  
987  /**
988   * finish_open - finish opening a file
989   * @file: file pointer
990   * @dentry: pointer to dentry
991   * @open: open callback
992   * @opened: state of open
993   *
994   * This can be used to finish opening a file passed to i_op->atomic_open().
995   *
996   * If the open callback is set to NULL, then the standard f_op->open()
997   * filesystem callback is substituted.
998   *
999   * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1000   * the return value of d_splice_alias(), then the caller needs to perform dput()
1001   * on it after finish_open().
1002   *
1003   * Returns zero on success or -errno if the open failed.
1004   */
1005  int finish_open(struct file *file, struct dentry *dentry,
1006  		int (*open)(struct inode *, struct file *))
1007  {
1008  	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1009  
1010  	file->f_path.dentry = dentry;
1011  	return do_dentry_open(file, d_backing_inode(dentry), open);
1012  }
1013  EXPORT_SYMBOL(finish_open);
1014  
1015  /**
1016   * finish_no_open - finish ->atomic_open() without opening the file
1017   *
1018   * @file: file pointer
1019   * @dentry: dentry or NULL (as returned from ->lookup())
1020   *
1021   * This can be used to set the result of a successful lookup in ->atomic_open().
1022   *
1023   * NB: unlike finish_open() this function does consume the dentry reference and
1024   * the caller need not dput() it.
1025   *
1026   * Returns "0" which must be the return value of ->atomic_open() after having
1027   * called this function.
1028   */
1029  int finish_no_open(struct file *file, struct dentry *dentry)
1030  {
1031  	file->f_path.dentry = dentry;
1032  	return 0;
1033  }
1034  EXPORT_SYMBOL(finish_no_open);
1035  
1036  char *file_path(struct file *filp, char *buf, int buflen)
1037  {
1038  	return d_path(&filp->f_path, buf, buflen);
1039  }
1040  EXPORT_SYMBOL(file_path);
1041  
1042  /**
1043   * vfs_open - open the file at the given path
1044   * @path: path to open
1045   * @file: newly allocated file with f_flag initialized
1046   * @cred: credentials to use
1047   */
1048  int vfs_open(const struct path *path, struct file *file)
1049  {
1050  	file->f_path = *path;
1051  	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1052  }
1053  
1054  struct file *dentry_open(const struct path *path, int flags,
1055  			 const struct cred *cred)
1056  {
1057  	int error;
1058  	struct file *f;
1059  
1060  	validate_creds(cred);
1061  
1062  	/* We must always pass in a valid mount pointer. */
1063  	BUG_ON(!path->mnt);
1064  
1065  	f = alloc_empty_file(flags, cred);
1066  	if (!IS_ERR(f)) {
1067  		error = vfs_open(path, f);
1068  		if (error) {
1069  			fput(f);
1070  			f = ERR_PTR(error);
1071  		}
1072  	}
1073  	return f;
1074  }
1075  EXPORT_SYMBOL(dentry_open);
1076  
1077  /**
1078   * dentry_create - Create and open a file
1079   * @path: path to create
1080   * @flags: O_ flags
1081   * @mode: mode bits for new file
1082   * @cred: credentials to use
1083   *
1084   * Caller must hold the parent directory's lock, and have prepared
1085   * a negative dentry, placed in @path->dentry, for the new file.
1086   *
1087   * Caller sets @path->mnt to the vfsmount of the filesystem where
1088   * the new file is to be created. The parent directory and the
1089   * negative dentry must reside on the same filesystem instance.
1090   *
1091   * On success, returns a "struct file *". Otherwise a ERR_PTR
1092   * is returned.
1093   */
1094  struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1095  			   const struct cred *cred)
1096  {
1097  	struct file *f;
1098  	int error;
1099  
1100  	validate_creds(cred);
1101  	f = alloc_empty_file(flags, cred);
1102  	if (IS_ERR(f))
1103  		return f;
1104  
1105  	error = vfs_create(mnt_idmap(path->mnt),
1106  			   d_inode(path->dentry->d_parent),
1107  			   path->dentry, mode, true);
1108  	if (!error)
1109  		error = vfs_open(path, f);
1110  
1111  	if (unlikely(error)) {
1112  		fput(f);
1113  		return ERR_PTR(error);
1114  	}
1115  	return f;
1116  }
1117  EXPORT_SYMBOL(dentry_create);
1118  
1119  struct file *open_with_fake_path(const struct path *path, int flags,
1120  				struct inode *inode, const struct cred *cred)
1121  {
1122  	struct file *f = alloc_empty_file_noaccount(flags, cred);
1123  	if (!IS_ERR(f)) {
1124  		int error;
1125  
1126  		f->f_path = *path;
1127  		error = do_dentry_open(f, inode, NULL);
1128  		if (error) {
1129  			fput(f);
1130  			f = ERR_PTR(error);
1131  		}
1132  	}
1133  	return f;
1134  }
1135  EXPORT_SYMBOL(open_with_fake_path);
1136  
1137  #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1138  #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1139  
1140  inline struct open_how build_open_how(int flags, umode_t mode)
1141  {
1142  	struct open_how how = {
1143  		.flags = flags & VALID_OPEN_FLAGS,
1144  		.mode = mode & S_IALLUGO,
1145  	};
1146  
1147  	/* O_PATH beats everything else. */
1148  	if (how.flags & O_PATH)
1149  		how.flags &= O_PATH_FLAGS;
1150  	/* Modes should only be set for create-like flags. */
1151  	if (!WILL_CREATE(how.flags))
1152  		how.mode = 0;
1153  	return how;
1154  }
1155  
1156  inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1157  {
1158  	u64 flags = how->flags;
1159  	u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1160  	int lookup_flags = 0;
1161  	int acc_mode = ACC_MODE(flags);
1162  
1163  	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1164  			 "struct open_flags doesn't yet handle flags > 32 bits");
1165  
1166  	/*
1167  	 * Strip flags that either shouldn't be set by userspace like
1168  	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1169  	 * open_flags like O_CLOEXEC.
1170  	 */
1171  	flags &= ~strip;
1172  
1173  	/*
1174  	 * Older syscalls implicitly clear all of the invalid flags or argument
1175  	 * values before calling build_open_flags(), but openat2(2) checks all
1176  	 * of its arguments.
1177  	 */
1178  	if (flags & ~VALID_OPEN_FLAGS)
1179  		return -EINVAL;
1180  	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1181  		return -EINVAL;
1182  
1183  	/* Scoping flags are mutually exclusive. */
1184  	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1185  		return -EINVAL;
1186  
1187  	/* Deal with the mode. */
1188  	if (WILL_CREATE(flags)) {
1189  		if (how->mode & ~S_IALLUGO)
1190  			return -EINVAL;
1191  		op->mode = how->mode | S_IFREG;
1192  	} else {
1193  		if (how->mode != 0)
1194  			return -EINVAL;
1195  		op->mode = 0;
1196  	}
1197  
1198  	/*
1199  	 * In order to ensure programs get explicit errors when trying to use
1200  	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1201  	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1202  	 * have to require userspace to explicitly set it.
1203  	 */
1204  	if (flags & __O_TMPFILE) {
1205  		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1206  			return -EINVAL;
1207  		if (!(acc_mode & MAY_WRITE))
1208  			return -EINVAL;
1209  	}
1210  	if (flags & O_PATH) {
1211  		/* O_PATH only permits certain other flags to be set. */
1212  		if (flags & ~O_PATH_FLAGS)
1213  			return -EINVAL;
1214  		acc_mode = 0;
1215  	}
1216  
1217  	/*
1218  	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1219  	 * check for O_DSYNC if the need any syncing at all we enforce it's
1220  	 * always set instead of having to deal with possibly weird behaviour
1221  	 * for malicious applications setting only __O_SYNC.
1222  	 */
1223  	if (flags & __O_SYNC)
1224  		flags |= O_DSYNC;
1225  
1226  	op->open_flag = flags;
1227  
1228  	/* O_TRUNC implies we need access checks for write permissions */
1229  	if (flags & O_TRUNC)
1230  		acc_mode |= MAY_WRITE;
1231  
1232  	/* Allow the LSM permission hook to distinguish append
1233  	   access from general write access. */
1234  	if (flags & O_APPEND)
1235  		acc_mode |= MAY_APPEND;
1236  
1237  	op->acc_mode = acc_mode;
1238  
1239  	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1240  
1241  	if (flags & O_CREAT) {
1242  		op->intent |= LOOKUP_CREATE;
1243  		if (flags & O_EXCL) {
1244  			op->intent |= LOOKUP_EXCL;
1245  			flags |= O_NOFOLLOW;
1246  		}
1247  	}
1248  
1249  	if (flags & O_DIRECTORY)
1250  		lookup_flags |= LOOKUP_DIRECTORY;
1251  	if (!(flags & O_NOFOLLOW))
1252  		lookup_flags |= LOOKUP_FOLLOW;
1253  
1254  	if (how->resolve & RESOLVE_NO_XDEV)
1255  		lookup_flags |= LOOKUP_NO_XDEV;
1256  	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1257  		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1258  	if (how->resolve & RESOLVE_NO_SYMLINKS)
1259  		lookup_flags |= LOOKUP_NO_SYMLINKS;
1260  	if (how->resolve & RESOLVE_BENEATH)
1261  		lookup_flags |= LOOKUP_BENEATH;
1262  	if (how->resolve & RESOLVE_IN_ROOT)
1263  		lookup_flags |= LOOKUP_IN_ROOT;
1264  	if (how->resolve & RESOLVE_CACHED) {
1265  		/* Don't bother even trying for create/truncate/tmpfile open */
1266  		if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1267  			return -EAGAIN;
1268  		lookup_flags |= LOOKUP_CACHED;
1269  	}
1270  
1271  	op->lookup_flags = lookup_flags;
1272  	return 0;
1273  }
1274  
1275  /**
1276   * file_open_name - open file and return file pointer
1277   *
1278   * @name:	struct filename containing path to open
1279   * @flags:	open flags as per the open(2) second argument
1280   * @mode:	mode for the new file if O_CREAT is set, else ignored
1281   *
1282   * This is the helper to open a file from kernelspace if you really
1283   * have to.  But in generally you should not do this, so please move
1284   * along, nothing to see here..
1285   */
1286  struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1287  {
1288  	struct open_flags op;
1289  	struct open_how how = build_open_how(flags, mode);
1290  	int err = build_open_flags(&how, &op);
1291  	if (err)
1292  		return ERR_PTR(err);
1293  	return do_filp_open(AT_FDCWD, name, &op);
1294  }
1295  
1296  /**
1297   * filp_open - open file and return file pointer
1298   *
1299   * @filename:	path to open
1300   * @flags:	open flags as per the open(2) second argument
1301   * @mode:	mode for the new file if O_CREAT is set, else ignored
1302   *
1303   * This is the helper to open a file from kernelspace if you really
1304   * have to.  But in generally you should not do this, so please move
1305   * along, nothing to see here..
1306   */
1307  struct file *filp_open(const char *filename, int flags, umode_t mode)
1308  {
1309  	struct filename *name = getname_kernel(filename);
1310  	struct file *file = ERR_CAST(name);
1311  
1312  	if (!IS_ERR(name)) {
1313  		file = file_open_name(name, flags, mode);
1314  		putname(name);
1315  	}
1316  	return file;
1317  }
1318  EXPORT_SYMBOL(filp_open);
1319  
1320  struct file *file_open_root(const struct path *root,
1321  			    const char *filename, int flags, umode_t mode)
1322  {
1323  	struct open_flags op;
1324  	struct open_how how = build_open_how(flags, mode);
1325  	int err = build_open_flags(&how, &op);
1326  	if (err)
1327  		return ERR_PTR(err);
1328  	return do_file_open_root(root, filename, &op);
1329  }
1330  EXPORT_SYMBOL(file_open_root);
1331  
1332  static long do_sys_openat2(int dfd, const char __user *filename,
1333  			   struct open_how *how)
1334  {
1335  	struct open_flags op;
1336  	int fd = build_open_flags(how, &op);
1337  	struct filename *tmp;
1338  
1339  	if (fd)
1340  		return fd;
1341  
1342  	tmp = getname(filename);
1343  	if (IS_ERR(tmp))
1344  		return PTR_ERR(tmp);
1345  
1346  	fd = get_unused_fd_flags(how->flags);
1347  	if (fd >= 0) {
1348  		struct file *f = do_filp_open(dfd, tmp, &op);
1349  		if (IS_ERR(f)) {
1350  			put_unused_fd(fd);
1351  			fd = PTR_ERR(f);
1352  		} else {
1353  			fsnotify_open(f);
1354  			fd_install(fd, f);
1355  		}
1356  	}
1357  	putname(tmp);
1358  	return fd;
1359  }
1360  
1361  long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1362  {
1363  	struct open_how how = build_open_how(flags, mode);
1364  	return do_sys_openat2(dfd, filename, &how);
1365  }
1366  
1367  
1368  SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1369  {
1370  	if (force_o_largefile())
1371  		flags |= O_LARGEFILE;
1372  	return do_sys_open(AT_FDCWD, filename, flags, mode);
1373  }
1374  
1375  SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1376  		umode_t, mode)
1377  {
1378  	if (force_o_largefile())
1379  		flags |= O_LARGEFILE;
1380  	return do_sys_open(dfd, filename, flags, mode);
1381  }
1382  
1383  SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1384  		struct open_how __user *, how, size_t, usize)
1385  {
1386  	int err;
1387  	struct open_how tmp;
1388  
1389  	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1390  	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1391  
1392  	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1393  		return -EINVAL;
1394  
1395  	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1396  	if (err)
1397  		return err;
1398  
1399  	audit_openat2_how(&tmp);
1400  
1401  	/* O_LARGEFILE is only allowed for non-O_PATH. */
1402  	if (!(tmp.flags & O_PATH) && force_o_largefile())
1403  		tmp.flags |= O_LARGEFILE;
1404  
1405  	return do_sys_openat2(dfd, filename, &tmp);
1406  }
1407  
1408  #ifdef CONFIG_COMPAT
1409  /*
1410   * Exactly like sys_open(), except that it doesn't set the
1411   * O_LARGEFILE flag.
1412   */
1413  COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1414  {
1415  	return do_sys_open(AT_FDCWD, filename, flags, mode);
1416  }
1417  
1418  /*
1419   * Exactly like sys_openat(), except that it doesn't set the
1420   * O_LARGEFILE flag.
1421   */
1422  COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1423  {
1424  	return do_sys_open(dfd, filename, flags, mode);
1425  }
1426  #endif
1427  
1428  #ifndef __alpha__
1429  
1430  /*
1431   * For backward compatibility?  Maybe this should be moved
1432   * into arch/i386 instead?
1433   */
1434  SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1435  {
1436  	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1437  
1438  	if (force_o_largefile())
1439  		flags |= O_LARGEFILE;
1440  	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1441  }
1442  #endif
1443  
1444  /*
1445   * "id" is the POSIX thread ID. We use the
1446   * files pointer for this..
1447   */
1448  int filp_close(struct file *filp, fl_owner_t id)
1449  {
1450  	int retval = 0;
1451  
1452  	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1453  			"VFS: Close: file count is 0 (f_op=%ps)",
1454  			filp->f_op)) {
1455  		return 0;
1456  	}
1457  
1458  	if (filp->f_op->flush)
1459  		retval = filp->f_op->flush(filp, id);
1460  
1461  	if (likely(!(filp->f_mode & FMODE_PATH))) {
1462  		dnotify_flush(filp, id);
1463  		locks_remove_posix(filp, id);
1464  	}
1465  	fput(filp);
1466  	return retval;
1467  }
1468  
1469  EXPORT_SYMBOL(filp_close);
1470  
1471  /*
1472   * Careful here! We test whether the file pointer is NULL before
1473   * releasing the fd. This ensures that one clone task can't release
1474   * an fd while another clone is opening it.
1475   */
1476  SYSCALL_DEFINE1(close, unsigned int, fd)
1477  {
1478  	int retval = close_fd(fd);
1479  
1480  	/* can't restart close syscall because file table entry was cleared */
1481  	if (unlikely(retval == -ERESTARTSYS ||
1482  		     retval == -ERESTARTNOINTR ||
1483  		     retval == -ERESTARTNOHAND ||
1484  		     retval == -ERESTART_RESTARTBLOCK))
1485  		retval = -EINTR;
1486  
1487  	return retval;
1488  }
1489  
1490  /**
1491   * close_range() - Close all file descriptors in a given range.
1492   *
1493   * @fd:     starting file descriptor to close
1494   * @max_fd: last file descriptor to close
1495   * @flags:  reserved for future extensions
1496   *
1497   * This closes a range of file descriptors. All file descriptors
1498   * from @fd up to and including @max_fd are closed.
1499   * Currently, errors to close a given file descriptor are ignored.
1500   */
1501  SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1502  		unsigned int, flags)
1503  {
1504  	return __close_range(fd, max_fd, flags);
1505  }
1506  
1507  /*
1508   * This routine simulates a hangup on the tty, to arrange that users
1509   * are given clean terminals at login time.
1510   */
1511  SYSCALL_DEFINE0(vhangup)
1512  {
1513  	if (capable(CAP_SYS_TTY_CONFIG)) {
1514  		tty_vhangup_self();
1515  		return 0;
1516  	}
1517  	return -EPERM;
1518  }
1519  
1520  /*
1521   * Called when an inode is about to be open.
1522   * We use this to disallow opening large files on 32bit systems if
1523   * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1524   * on this flag in sys_open.
1525   */
1526  int generic_file_open(struct inode * inode, struct file * filp)
1527  {
1528  	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1529  		return -EOVERFLOW;
1530  	return 0;
1531  }
1532  
1533  EXPORT_SYMBOL(generic_file_open);
1534  
1535  /*
1536   * This is used by subsystems that don't want seekable
1537   * file descriptors. The function is not supposed to ever fail, the only
1538   * reason it returns an 'int' and not 'void' is so that it can be plugged
1539   * directly into file_operations structure.
1540   */
1541  int nonseekable_open(struct inode *inode, struct file *filp)
1542  {
1543  	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1544  	return 0;
1545  }
1546  
1547  EXPORT_SYMBOL(nonseekable_open);
1548  
1549  /*
1550   * stream_open is used by subsystems that want stream-like file descriptors.
1551   * Such file descriptors are not seekable and don't have notion of position
1552   * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1553   * Contrary to file descriptors of other regular files, .read() and .write()
1554   * can run simultaneously.
1555   *
1556   * stream_open never fails and is marked to return int so that it could be
1557   * directly used as file_operations.open .
1558   */
1559  int stream_open(struct inode *inode, struct file *filp)
1560  {
1561  	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1562  	filp->f_mode |= FMODE_STREAM;
1563  	return 0;
1564  }
1565  
1566  EXPORT_SYMBOL(stream_open);
1567