xref: /linux/fs/open.c (revision 415f135ace7fd824cde083184a922e39156055b5)
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/dnotify.h>
33  #include <linux/compat.h>
34  #include <linux/mnt_idmapping.h>
35  #include <linux/filelock.h>
36  
37  #include "internal.h"
38  
39  int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
40  		loff_t length, unsigned int time_attrs, struct file *filp)
41  {
42  	int ret;
43  	struct iattr newattrs;
44  
45  	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46  	if (length < 0)
47  		return -EINVAL;
48  
49  	newattrs.ia_size = length;
50  	newattrs.ia_valid = ATTR_SIZE | time_attrs;
51  	if (filp) {
52  		newattrs.ia_file = filp;
53  		newattrs.ia_valid |= ATTR_FILE;
54  	}
55  
56  	/* Remove suid, sgid, and file capabilities on truncate too */
57  	ret = dentry_needs_remove_privs(idmap, dentry);
58  	if (ret < 0)
59  		return ret;
60  	if (ret)
61  		newattrs.ia_valid |= ret | ATTR_FORCE;
62  
63  	inode_lock(dentry->d_inode);
64  	/* Note any delegations or leases have already been broken: */
65  	ret = notify_change(idmap, dentry, &newattrs, NULL);
66  	inode_unlock(dentry->d_inode);
67  	return ret;
68  }
69  
70  long vfs_truncate(const struct path *path, loff_t length)
71  {
72  	struct mnt_idmap *idmap;
73  	struct inode *inode;
74  	long error;
75  
76  	inode = path->dentry->d_inode;
77  
78  	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79  	if (S_ISDIR(inode->i_mode))
80  		return -EISDIR;
81  	if (!S_ISREG(inode->i_mode))
82  		return -EINVAL;
83  
84  	idmap = mnt_idmap(path->mnt);
85  	error = inode_permission(idmap, inode, MAY_WRITE);
86  	if (error)
87  		return error;
88  
89  	error = fsnotify_truncate_perm(path, length);
90  	if (error)
91  		return error;
92  
93  	error = mnt_want_write(path->mnt);
94  	if (error)
95  		return error;
96  
97  	error = -EPERM;
98  	if (IS_APPEND(inode))
99  		goto mnt_drop_write_and_out;
100  
101  	error = get_write_access(inode);
102  	if (error)
103  		goto mnt_drop_write_and_out;
104  
105  	/*
106  	 * Make sure that there are no leases.  get_write_access() protects
107  	 * against the truncate racing with a lease-granting setlease().
108  	 */
109  	error = break_lease(inode, O_WRONLY);
110  	if (error)
111  		goto put_write_and_out;
112  
113  	error = security_path_truncate(path);
114  	if (!error)
115  		error = do_truncate(idmap, path->dentry, length, 0, NULL);
116  
117  put_write_and_out:
118  	put_write_access(inode);
119  mnt_drop_write_and_out:
120  	mnt_drop_write(path->mnt);
121  
122  	return error;
123  }
124  EXPORT_SYMBOL_GPL(vfs_truncate);
125  
126  long do_sys_truncate(const char __user *pathname, loff_t length)
127  {
128  	unsigned int lookup_flags = LOOKUP_FOLLOW;
129  	struct path path;
130  	int error;
131  
132  	if (length < 0)	/* sorry, but loff_t says... */
133  		return -EINVAL;
134  
135  retry:
136  	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
137  	if (!error) {
138  		error = vfs_truncate(&path, length);
139  		path_put(&path);
140  	}
141  	if (retry_estale(error, lookup_flags)) {
142  		lookup_flags |= LOOKUP_REVAL;
143  		goto retry;
144  	}
145  	return error;
146  }
147  
148  SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
149  {
150  	return do_sys_truncate(path, length);
151  }
152  
153  #ifdef CONFIG_COMPAT
154  COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
155  {
156  	return do_sys_truncate(path, length);
157  }
158  #endif
159  
160  long do_ftruncate(struct file *file, loff_t length, int small)
161  {
162  	struct inode *inode;
163  	struct dentry *dentry;
164  	int error;
165  
166  	/* explicitly opened as large or we are on 64-bit box */
167  	if (file->f_flags & O_LARGEFILE)
168  		small = 0;
169  
170  	dentry = file->f_path.dentry;
171  	inode = dentry->d_inode;
172  	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
173  		return -EINVAL;
174  
175  	/* Cannot ftruncate over 2^31 bytes without large file support */
176  	if (small && length > MAX_NON_LFS)
177  		return -EINVAL;
178  
179  	/* Check IS_APPEND on real upper inode */
180  	if (IS_APPEND(file_inode(file)))
181  		return -EPERM;
182  
183  	error = security_file_truncate(file);
184  	if (error)
185  		return error;
186  
187  	error = fsnotify_truncate_perm(&file->f_path, length);
188  	if (error)
189  		return error;
190  
191  	sb_start_write(inode->i_sb);
192  	error = do_truncate(file_mnt_idmap(file), dentry, length,
193  			    ATTR_MTIME | ATTR_CTIME, file);
194  	sb_end_write(inode->i_sb);
195  
196  	return error;
197  }
198  
199  long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
200  {
201  	if (length < 0)
202  		return -EINVAL;
203  	CLASS(fd, f)(fd);
204  	if (fd_empty(f))
205  		return -EBADF;
206  
207  	return do_ftruncate(fd_file(f), length, small);
208  }
209  
210  SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
211  {
212  	return do_sys_ftruncate(fd, length, 1);
213  }
214  
215  #ifdef CONFIG_COMPAT
216  COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
217  {
218  	return do_sys_ftruncate(fd, length, 1);
219  }
220  #endif
221  
222  /* LFS versions of truncate are only needed on 32 bit machines */
223  #if BITS_PER_LONG == 32
224  SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
225  {
226  	return do_sys_truncate(path, length);
227  }
228  
229  SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
230  {
231  	return do_sys_ftruncate(fd, length, 0);
232  }
233  #endif /* BITS_PER_LONG == 32 */
234  
235  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
236  COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
237  		       compat_arg_u64_dual(length))
238  {
239  	return ksys_truncate(pathname, compat_arg_u64_glue(length));
240  }
241  #endif
242  
243  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
244  COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
245  		       compat_arg_u64_dual(length))
246  {
247  	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
248  }
249  #endif
250  
251  int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
252  {
253  	struct inode *inode = file_inode(file);
254  	long ret;
255  	loff_t sum;
256  
257  	if (offset < 0 || len <= 0)
258  		return -EINVAL;
259  
260  	if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
261  		return -EOPNOTSUPP;
262  
263  	/*
264  	 * Modes are exclusive, even if that is not obvious from the encoding
265  	 * as bit masks and the mix with the flag in the same namespace.
266  	 *
267  	 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
268  	 * encoded as no bit set.
269  	 */
270  	switch (mode & FALLOC_FL_MODE_MASK) {
271  	case FALLOC_FL_ALLOCATE_RANGE:
272  	case FALLOC_FL_UNSHARE_RANGE:
273  	case FALLOC_FL_ZERO_RANGE:
274  		break;
275  	case FALLOC_FL_PUNCH_HOLE:
276  		if (!(mode & FALLOC_FL_KEEP_SIZE))
277  			return -EOPNOTSUPP;
278  		break;
279  	case FALLOC_FL_COLLAPSE_RANGE:
280  	case FALLOC_FL_INSERT_RANGE:
281  		if (mode & FALLOC_FL_KEEP_SIZE)
282  			return -EOPNOTSUPP;
283  		break;
284  	default:
285  		return -EOPNOTSUPP;
286  	}
287  
288  	if (!(file->f_mode & FMODE_WRITE))
289  		return -EBADF;
290  
291  	/*
292  	 * On append-only files only space preallocation is supported.
293  	 */
294  	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
295  		return -EPERM;
296  
297  	if (IS_IMMUTABLE(inode))
298  		return -EPERM;
299  
300  	/*
301  	 * We cannot allow any fallocate operation on an active swapfile
302  	 */
303  	if (IS_SWAPFILE(inode))
304  		return -ETXTBSY;
305  
306  	/*
307  	 * Revalidate the write permissions, in case security policy has
308  	 * changed since the files were opened.
309  	 */
310  	ret = security_file_permission(file, MAY_WRITE);
311  	if (ret)
312  		return ret;
313  
314  	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
315  	if (ret)
316  		return ret;
317  
318  	if (S_ISFIFO(inode->i_mode))
319  		return -ESPIPE;
320  
321  	if (S_ISDIR(inode->i_mode))
322  		return -EISDIR;
323  
324  	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
325  		return -ENODEV;
326  
327  	/* Check for wraparound */
328  	if (check_add_overflow(offset, len, &sum))
329  		return -EFBIG;
330  
331  	if (sum > inode->i_sb->s_maxbytes)
332  		return -EFBIG;
333  
334  	if (!file->f_op->fallocate)
335  		return -EOPNOTSUPP;
336  
337  	file_start_write(file);
338  	ret = file->f_op->fallocate(file, mode, offset, len);
339  
340  	/*
341  	 * Create inotify and fanotify events.
342  	 *
343  	 * To keep the logic simple always create events if fallocate succeeds.
344  	 * This implies that events are even created if the file size remains
345  	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
346  	 */
347  	if (ret == 0)
348  		fsnotify_modify(file);
349  
350  	file_end_write(file);
351  	return ret;
352  }
353  EXPORT_SYMBOL_GPL(vfs_fallocate);
354  
355  int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
356  {
357  	CLASS(fd, f)(fd);
358  
359  	if (fd_empty(f))
360  		return -EBADF;
361  
362  	return vfs_fallocate(fd_file(f), mode, offset, len);
363  }
364  
365  SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
366  {
367  	return ksys_fallocate(fd, mode, offset, len);
368  }
369  
370  #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
371  COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
372  		       compat_arg_u64_dual(len))
373  {
374  	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
375  			      compat_arg_u64_glue(len));
376  }
377  #endif
378  
379  /*
380   * access() needs to use the real uid/gid, not the effective uid/gid.
381   * We do this by temporarily clearing all FS-related capabilities and
382   * switching the fsuid/fsgid around to the real ones.
383   *
384   * Creating new credentials is expensive, so we try to skip doing it,
385   * which we can if the result would match what we already got.
386   */
387  static bool access_need_override_creds(int flags)
388  {
389  	const struct cred *cred;
390  
391  	if (flags & AT_EACCESS)
392  		return false;
393  
394  	cred = current_cred();
395  	if (!uid_eq(cred->fsuid, cred->uid) ||
396  	    !gid_eq(cred->fsgid, cred->gid))
397  		return true;
398  
399  	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
400  		kuid_t root_uid = make_kuid(cred->user_ns, 0);
401  		if (!uid_eq(cred->uid, root_uid)) {
402  			if (!cap_isclear(cred->cap_effective))
403  				return true;
404  		} else {
405  			if (!cap_isidentical(cred->cap_effective,
406  			    cred->cap_permitted))
407  				return true;
408  		}
409  	}
410  
411  	return false;
412  }
413  
414  static const struct cred *access_override_creds(void)
415  {
416  	struct cred *override_cred;
417  
418  	override_cred = prepare_creds();
419  	if (!override_cred)
420  		return NULL;
421  
422  	/*
423  	 * XXX access_need_override_creds performs checks in hopes of skipping
424  	 * this work. Make sure it stays in sync if making any changes in this
425  	 * routine.
426  	 */
427  
428  	override_cred->fsuid = override_cred->uid;
429  	override_cred->fsgid = override_cred->gid;
430  
431  	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
432  		/* Clear the capabilities if we switch to a non-root user */
433  		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
434  		if (!uid_eq(override_cred->uid, root_uid))
435  			cap_clear(override_cred->cap_effective);
436  		else
437  			override_cred->cap_effective =
438  				override_cred->cap_permitted;
439  	}
440  
441  	/*
442  	 * The new set of credentials can *only* be used in
443  	 * task-synchronous circumstances, and does not need
444  	 * RCU freeing, unless somebody then takes a separate
445  	 * reference to it.
446  	 *
447  	 * NOTE! This is _only_ true because this credential
448  	 * is used purely for override_creds() that installs
449  	 * it as the subjective cred. Other threads will be
450  	 * accessing ->real_cred, not the subjective cred.
451  	 *
452  	 * If somebody _does_ make a copy of this (using the
453  	 * 'get_current_cred()' function), that will clear the
454  	 * non_rcu field, because now that other user may be
455  	 * expecting RCU freeing. But normal thread-synchronous
456  	 * cred accesses will keep things non-racy to avoid RCU
457  	 * freeing.
458  	 */
459  	override_cred->non_rcu = 1;
460  	return override_creds(override_cred);
461  }
462  
463  static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
464  {
465  	struct path path;
466  	struct inode *inode;
467  	int res;
468  	unsigned int lookup_flags = LOOKUP_FOLLOW;
469  	const struct cred *old_cred = NULL;
470  
471  	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
472  		return -EINVAL;
473  
474  	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
475  		return -EINVAL;
476  
477  	if (flags & AT_SYMLINK_NOFOLLOW)
478  		lookup_flags &= ~LOOKUP_FOLLOW;
479  	if (flags & AT_EMPTY_PATH)
480  		lookup_flags |= LOOKUP_EMPTY;
481  
482  	if (access_need_override_creds(flags)) {
483  		old_cred = access_override_creds();
484  		if (!old_cred)
485  			return -ENOMEM;
486  	}
487  
488  retry:
489  	res = user_path_at(dfd, filename, lookup_flags, &path);
490  	if (res)
491  		goto out;
492  
493  	inode = d_backing_inode(path.dentry);
494  
495  	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
496  		/*
497  		 * MAY_EXEC on regular files is denied if the fs is mounted
498  		 * with the "noexec" flag.
499  		 */
500  		res = -EACCES;
501  		if (path_noexec(&path))
502  			goto out_path_release;
503  	}
504  
505  	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
506  	/* SuS v2 requires we report a read only fs too */
507  	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
508  		goto out_path_release;
509  	/*
510  	 * This is a rare case where using __mnt_is_readonly()
511  	 * is OK without a mnt_want/drop_write() pair.  Since
512  	 * no actual write to the fs is performed here, we do
513  	 * not need to telegraph to that to anyone.
514  	 *
515  	 * By doing this, we accept that this access is
516  	 * inherently racy and know that the fs may change
517  	 * state before we even see this result.
518  	 */
519  	if (__mnt_is_readonly(path.mnt))
520  		res = -EROFS;
521  
522  out_path_release:
523  	path_put(&path);
524  	if (retry_estale(res, lookup_flags)) {
525  		lookup_flags |= LOOKUP_REVAL;
526  		goto retry;
527  	}
528  out:
529  	if (old_cred)
530  		put_cred(revert_creds(old_cred));
531  
532  	return res;
533  }
534  
535  SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
536  {
537  	return do_faccessat(dfd, filename, mode, 0);
538  }
539  
540  SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
541  		int, flags)
542  {
543  	return do_faccessat(dfd, filename, mode, flags);
544  }
545  
546  SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
547  {
548  	return do_faccessat(AT_FDCWD, filename, mode, 0);
549  }
550  
551  SYSCALL_DEFINE1(chdir, const char __user *, filename)
552  {
553  	struct path path;
554  	int error;
555  	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
556  retry:
557  	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
558  	if (error)
559  		goto out;
560  
561  	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
562  	if (error)
563  		goto dput_and_out;
564  
565  	set_fs_pwd(current->fs, &path);
566  
567  dput_and_out:
568  	path_put(&path);
569  	if (retry_estale(error, lookup_flags)) {
570  		lookup_flags |= LOOKUP_REVAL;
571  		goto retry;
572  	}
573  out:
574  	return error;
575  }
576  
577  SYSCALL_DEFINE1(fchdir, unsigned int, fd)
578  {
579  	CLASS(fd_raw, f)(fd);
580  	int error;
581  
582  	if (fd_empty(f))
583  		return -EBADF;
584  
585  	if (!d_can_lookup(fd_file(f)->f_path.dentry))
586  		return -ENOTDIR;
587  
588  	error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
589  	if (!error)
590  		set_fs_pwd(current->fs, &fd_file(f)->f_path);
591  	return error;
592  }
593  
594  SYSCALL_DEFINE1(chroot, const char __user *, filename)
595  {
596  	struct path path;
597  	int error;
598  	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
599  retry:
600  	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
601  	if (error)
602  		goto out;
603  
604  	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
605  	if (error)
606  		goto dput_and_out;
607  
608  	error = -EPERM;
609  	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
610  		goto dput_and_out;
611  	error = security_path_chroot(&path);
612  	if (error)
613  		goto dput_and_out;
614  
615  	set_fs_root(current->fs, &path);
616  	error = 0;
617  dput_and_out:
618  	path_put(&path);
619  	if (retry_estale(error, lookup_flags)) {
620  		lookup_flags |= LOOKUP_REVAL;
621  		goto retry;
622  	}
623  out:
624  	return error;
625  }
626  
627  int chmod_common(const struct path *path, umode_t mode)
628  {
629  	struct inode *inode = path->dentry->d_inode;
630  	struct inode *delegated_inode = NULL;
631  	struct iattr newattrs;
632  	int error;
633  
634  	error = mnt_want_write(path->mnt);
635  	if (error)
636  		return error;
637  retry_deleg:
638  	inode_lock(inode);
639  	error = security_path_chmod(path, mode);
640  	if (error)
641  		goto out_unlock;
642  	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
643  	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
644  	error = notify_change(mnt_idmap(path->mnt), path->dentry,
645  			      &newattrs, &delegated_inode);
646  out_unlock:
647  	inode_unlock(inode);
648  	if (delegated_inode) {
649  		error = break_deleg_wait(&delegated_inode);
650  		if (!error)
651  			goto retry_deleg;
652  	}
653  	mnt_drop_write(path->mnt);
654  	return error;
655  }
656  
657  int vfs_fchmod(struct file *file, umode_t mode)
658  {
659  	audit_file(file);
660  	return chmod_common(&file->f_path, mode);
661  }
662  
663  SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
664  {
665  	CLASS(fd, f)(fd);
666  
667  	if (fd_empty(f))
668  		return -EBADF;
669  
670  	return vfs_fchmod(fd_file(f), mode);
671  }
672  
673  static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
674  		       unsigned int flags)
675  {
676  	struct path path;
677  	int error;
678  	unsigned int lookup_flags;
679  
680  	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
681  		return -EINVAL;
682  
683  	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
684  	if (flags & AT_EMPTY_PATH)
685  		lookup_flags |= LOOKUP_EMPTY;
686  
687  retry:
688  	error = user_path_at(dfd, filename, lookup_flags, &path);
689  	if (!error) {
690  		error = chmod_common(&path, mode);
691  		path_put(&path);
692  		if (retry_estale(error, lookup_flags)) {
693  			lookup_flags |= LOOKUP_REVAL;
694  			goto retry;
695  		}
696  	}
697  	return error;
698  }
699  
700  SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
701  		umode_t, mode, unsigned int, flags)
702  {
703  	return do_fchmodat(dfd, filename, mode, flags);
704  }
705  
706  SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
707  		umode_t, mode)
708  {
709  	return do_fchmodat(dfd, filename, mode, 0);
710  }
711  
712  SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
713  {
714  	return do_fchmodat(AT_FDCWD, filename, mode, 0);
715  }
716  
717  /*
718   * Check whether @kuid is valid and if so generate and set vfsuid_t in
719   * ia_vfsuid.
720   *
721   * Return: true if @kuid is valid, false if not.
722   */
723  static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
724  {
725  	if (!uid_valid(kuid))
726  		return false;
727  	attr->ia_valid |= ATTR_UID;
728  	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
729  	return true;
730  }
731  
732  /*
733   * Check whether @kgid is valid and if so generate and set vfsgid_t in
734   * ia_vfsgid.
735   *
736   * Return: true if @kgid is valid, false if not.
737   */
738  static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
739  {
740  	if (!gid_valid(kgid))
741  		return false;
742  	attr->ia_valid |= ATTR_GID;
743  	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
744  	return true;
745  }
746  
747  int chown_common(const struct path *path, uid_t user, gid_t group)
748  {
749  	struct mnt_idmap *idmap;
750  	struct user_namespace *fs_userns;
751  	struct inode *inode = path->dentry->d_inode;
752  	struct inode *delegated_inode = NULL;
753  	int error;
754  	struct iattr newattrs;
755  	kuid_t uid;
756  	kgid_t gid;
757  
758  	uid = make_kuid(current_user_ns(), user);
759  	gid = make_kgid(current_user_ns(), group);
760  
761  	idmap = mnt_idmap(path->mnt);
762  	fs_userns = i_user_ns(inode);
763  
764  retry_deleg:
765  	newattrs.ia_vfsuid = INVALID_VFSUID;
766  	newattrs.ia_vfsgid = INVALID_VFSGID;
767  	newattrs.ia_valid =  ATTR_CTIME;
768  	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
769  		return -EINVAL;
770  	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
771  		return -EINVAL;
772  	inode_lock(inode);
773  	if (!S_ISDIR(inode->i_mode))
774  		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
775  				     setattr_should_drop_sgid(idmap, inode);
776  	/* Continue to send actual fs values, not the mount values. */
777  	error = security_path_chown(
778  		path,
779  		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
780  		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
781  	if (!error)
782  		error = notify_change(idmap, path->dentry, &newattrs,
783  				      &delegated_inode);
784  	inode_unlock(inode);
785  	if (delegated_inode) {
786  		error = break_deleg_wait(&delegated_inode);
787  		if (!error)
788  			goto retry_deleg;
789  	}
790  	return error;
791  }
792  
793  int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
794  		int flag)
795  {
796  	struct path path;
797  	int error = -EINVAL;
798  	int lookup_flags;
799  
800  	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
801  		goto out;
802  
803  	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
804  	if (flag & AT_EMPTY_PATH)
805  		lookup_flags |= LOOKUP_EMPTY;
806  retry:
807  	error = user_path_at(dfd, filename, lookup_flags, &path);
808  	if (error)
809  		goto out;
810  	error = mnt_want_write(path.mnt);
811  	if (error)
812  		goto out_release;
813  	error = chown_common(&path, user, group);
814  	mnt_drop_write(path.mnt);
815  out_release:
816  	path_put(&path);
817  	if (retry_estale(error, lookup_flags)) {
818  		lookup_flags |= LOOKUP_REVAL;
819  		goto retry;
820  	}
821  out:
822  	return error;
823  }
824  
825  SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
826  		gid_t, group, int, flag)
827  {
828  	return do_fchownat(dfd, filename, user, group, flag);
829  }
830  
831  SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
832  {
833  	return do_fchownat(AT_FDCWD, filename, user, group, 0);
834  }
835  
836  SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
837  {
838  	return do_fchownat(AT_FDCWD, filename, user, group,
839  			   AT_SYMLINK_NOFOLLOW);
840  }
841  
842  int vfs_fchown(struct file *file, uid_t user, gid_t group)
843  {
844  	int error;
845  
846  	error = mnt_want_write_file(file);
847  	if (error)
848  		return error;
849  	audit_file(file);
850  	error = chown_common(&file->f_path, user, group);
851  	mnt_drop_write_file(file);
852  	return error;
853  }
854  
855  int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
856  {
857  	CLASS(fd, f)(fd);
858  
859  	if (fd_empty(f))
860  		return -EBADF;
861  
862  	return vfs_fchown(fd_file(f), user, group);
863  }
864  
865  SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
866  {
867  	return ksys_fchown(fd, user, group);
868  }
869  
870  static inline int file_get_write_access(struct file *f)
871  {
872  	int error;
873  
874  	error = get_write_access(f->f_inode);
875  	if (unlikely(error))
876  		return error;
877  	error = mnt_get_write_access(f->f_path.mnt);
878  	if (unlikely(error))
879  		goto cleanup_inode;
880  	if (unlikely(f->f_mode & FMODE_BACKING)) {
881  		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
882  		if (unlikely(error))
883  			goto cleanup_mnt;
884  	}
885  	return 0;
886  
887  cleanup_mnt:
888  	mnt_put_write_access(f->f_path.mnt);
889  cleanup_inode:
890  	put_write_access(f->f_inode);
891  	return error;
892  }
893  
894  static int do_dentry_open(struct file *f,
895  			  int (*open)(struct inode *, struct file *))
896  {
897  	static const struct file_operations empty_fops = {};
898  	struct inode *inode = f->f_path.dentry->d_inode;
899  	int error;
900  
901  	path_get(&f->f_path);
902  	f->f_inode = inode;
903  	f->f_mapping = inode->i_mapping;
904  	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
905  	f->f_sb_err = file_sample_sb_err(f);
906  
907  	if (unlikely(f->f_flags & O_PATH)) {
908  		f->f_mode = FMODE_PATH | FMODE_OPENED;
909  		file_set_fsnotify_mode(f, FMODE_NONOTIFY);
910  		f->f_op = &empty_fops;
911  		return 0;
912  	}
913  
914  	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
915  		i_readcount_inc(inode);
916  	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
917  		error = file_get_write_access(f);
918  		if (unlikely(error))
919  			goto cleanup_file;
920  		f->f_mode |= FMODE_WRITER;
921  	}
922  
923  	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
924  	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
925  		f->f_mode |= FMODE_ATOMIC_POS;
926  
927  	f->f_op = fops_get(inode->i_fop);
928  	if (WARN_ON(!f->f_op)) {
929  		error = -ENODEV;
930  		goto cleanup_all;
931  	}
932  
933  	error = security_file_open(f);
934  	if (error)
935  		goto cleanup_all;
936  
937  	/*
938  	 * Set FMODE_NONOTIFY_* bits according to existing permission watches.
939  	 * If FMODE_NONOTIFY mode was already set for an fanotify fd or for a
940  	 * pseudo file, this call will not change the mode.
941  	 */
942  	file_set_fsnotify_mode_from_watchers(f);
943  	error = fsnotify_open_perm(f);
944  	if (error)
945  		goto cleanup_all;
946  
947  	error = break_lease(file_inode(f), f->f_flags);
948  	if (error)
949  		goto cleanup_all;
950  
951  	/* normally all 3 are set; ->open() can clear them if needed */
952  	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
953  	if (!open)
954  		open = f->f_op->open;
955  	if (open) {
956  		error = open(inode, f);
957  		if (error)
958  			goto cleanup_all;
959  	}
960  	f->f_mode |= FMODE_OPENED;
961  	if ((f->f_mode & FMODE_READ) &&
962  	     likely(f->f_op->read || f->f_op->read_iter))
963  		f->f_mode |= FMODE_CAN_READ;
964  	if ((f->f_mode & FMODE_WRITE) &&
965  	     likely(f->f_op->write || f->f_op->write_iter))
966  		f->f_mode |= FMODE_CAN_WRITE;
967  	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
968  		f->f_mode &= ~FMODE_LSEEK;
969  	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
970  		f->f_mode |= FMODE_CAN_ODIRECT;
971  
972  	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
973  	f->f_iocb_flags = iocb_flags(f);
974  
975  	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
976  
977  	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
978  		return -EINVAL;
979  
980  	/*
981  	 * XXX: Huge page cache doesn't support writing yet. Drop all page
982  	 * cache for this file before processing writes.
983  	 */
984  	if (f->f_mode & FMODE_WRITE) {
985  		/*
986  		 * Depends on full fence from get_write_access() to synchronize
987  		 * against collapse_file() regarding i_writecount and nr_thps
988  		 * updates. Ensures subsequent insertion of THPs into the page
989  		 * cache will fail.
990  		 */
991  		if (filemap_nr_thps(inode->i_mapping)) {
992  			struct address_space *mapping = inode->i_mapping;
993  
994  			filemap_invalidate_lock(inode->i_mapping);
995  			/*
996  			 * unmap_mapping_range just need to be called once
997  			 * here, because the private pages is not need to be
998  			 * unmapped mapping (e.g. data segment of dynamic
999  			 * shared libraries here).
1000  			 */
1001  			unmap_mapping_range(mapping, 0, 0, 0);
1002  			truncate_inode_pages(mapping, 0);
1003  			filemap_invalidate_unlock(inode->i_mapping);
1004  		}
1005  	}
1006  
1007  	return 0;
1008  
1009  cleanup_all:
1010  	if (WARN_ON_ONCE(error > 0))
1011  		error = -EINVAL;
1012  	fops_put(f->f_op);
1013  	put_file_access(f);
1014  cleanup_file:
1015  	path_put(&f->f_path);
1016  	f->f_path.mnt = NULL;
1017  	f->f_path.dentry = NULL;
1018  	f->f_inode = NULL;
1019  	return error;
1020  }
1021  
1022  /**
1023   * finish_open - finish opening a file
1024   * @file: file pointer
1025   * @dentry: pointer to dentry
1026   * @open: open callback
1027   *
1028   * This can be used to finish opening a file passed to i_op->atomic_open().
1029   *
1030   * If the open callback is set to NULL, then the standard f_op->open()
1031   * filesystem callback is substituted.
1032   *
1033   * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1034   * the return value of d_splice_alias(), then the caller needs to perform dput()
1035   * on it after finish_open().
1036   *
1037   * Returns zero on success or -errno if the open failed.
1038   */
1039  int finish_open(struct file *file, struct dentry *dentry,
1040  		int (*open)(struct inode *, struct file *))
1041  {
1042  	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1043  
1044  	file->f_path.dentry = dentry;
1045  	return do_dentry_open(file, open);
1046  }
1047  EXPORT_SYMBOL(finish_open);
1048  
1049  /**
1050   * finish_no_open - finish ->atomic_open() without opening the file
1051   *
1052   * @file: file pointer
1053   * @dentry: dentry or NULL (as returned from ->lookup())
1054   *
1055   * This can be used to set the result of a successful lookup in ->atomic_open().
1056   *
1057   * NB: unlike finish_open() this function does consume the dentry reference and
1058   * the caller need not dput() it.
1059   *
1060   * Returns "0" which must be the return value of ->atomic_open() after having
1061   * called this function.
1062   */
1063  int finish_no_open(struct file *file, struct dentry *dentry)
1064  {
1065  	file->f_path.dentry = dentry;
1066  	return 0;
1067  }
1068  EXPORT_SYMBOL(finish_no_open);
1069  
1070  char *file_path(struct file *filp, char *buf, int buflen)
1071  {
1072  	return d_path(&filp->f_path, buf, buflen);
1073  }
1074  EXPORT_SYMBOL(file_path);
1075  
1076  /**
1077   * vfs_open - open the file at the given path
1078   * @path: path to open
1079   * @file: newly allocated file with f_flag initialized
1080   */
1081  int vfs_open(const struct path *path, struct file *file)
1082  {
1083  	int ret;
1084  
1085  	file->f_path = *path;
1086  	ret = do_dentry_open(file, NULL);
1087  	if (!ret) {
1088  		/*
1089  		 * Once we return a file with FMODE_OPENED, __fput() will call
1090  		 * fsnotify_close(), so we need fsnotify_open() here for
1091  		 * symmetry.
1092  		 */
1093  		fsnotify_open(file);
1094  	}
1095  	return ret;
1096  }
1097  
1098  struct file *dentry_open(const struct path *path, int flags,
1099  			 const struct cred *cred)
1100  {
1101  	int error;
1102  	struct file *f;
1103  
1104  	/* We must always pass in a valid mount pointer. */
1105  	BUG_ON(!path->mnt);
1106  
1107  	f = alloc_empty_file(flags, cred);
1108  	if (!IS_ERR(f)) {
1109  		error = vfs_open(path, f);
1110  		if (error) {
1111  			fput(f);
1112  			f = ERR_PTR(error);
1113  		}
1114  	}
1115  	return f;
1116  }
1117  EXPORT_SYMBOL(dentry_open);
1118  
1119  struct file *dentry_open_nonotify(const struct path *path, int flags,
1120  				  const struct cred *cred)
1121  {
1122  	struct file *f = alloc_empty_file(flags, cred);
1123  	if (!IS_ERR(f)) {
1124  		int error;
1125  
1126  		file_set_fsnotify_mode(f, FMODE_NONOTIFY);
1127  		error = vfs_open(path, f);
1128  		if (error) {
1129  			fput(f);
1130  			f = ERR_PTR(error);
1131  		}
1132  	}
1133  	return f;
1134  }
1135  
1136  /**
1137   * dentry_create - Create and open a file
1138   * @path: path to create
1139   * @flags: O_ flags
1140   * @mode: mode bits for new file
1141   * @cred: credentials to use
1142   *
1143   * Caller must hold the parent directory's lock, and have prepared
1144   * a negative dentry, placed in @path->dentry, for the new file.
1145   *
1146   * Caller sets @path->mnt to the vfsmount of the filesystem where
1147   * the new file is to be created. The parent directory and the
1148   * negative dentry must reside on the same filesystem instance.
1149   *
1150   * On success, returns a "struct file *". Otherwise a ERR_PTR
1151   * is returned.
1152   */
1153  struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1154  			   const struct cred *cred)
1155  {
1156  	struct file *f;
1157  	int error;
1158  
1159  	f = alloc_empty_file(flags, cred);
1160  	if (IS_ERR(f))
1161  		return f;
1162  
1163  	error = vfs_create(mnt_idmap(path->mnt),
1164  			   d_inode(path->dentry->d_parent),
1165  			   path->dentry, mode, true);
1166  	if (!error)
1167  		error = vfs_open(path, f);
1168  
1169  	if (unlikely(error)) {
1170  		fput(f);
1171  		return ERR_PTR(error);
1172  	}
1173  	return f;
1174  }
1175  EXPORT_SYMBOL(dentry_create);
1176  
1177  /**
1178   * kernel_file_open - open a file for kernel internal use
1179   * @path:	path of the file to open
1180   * @flags:	open flags
1181   * @cred:	credentials for open
1182   *
1183   * Open a file for use by in-kernel consumers. The file is not accounted
1184   * against nr_files and must not be installed into the file descriptor
1185   * table.
1186   *
1187   * Return: Opened file on success, an error pointer on failure.
1188   */
1189  struct file *kernel_file_open(const struct path *path, int flags,
1190  				const struct cred *cred)
1191  {
1192  	struct file *f;
1193  	int error;
1194  
1195  	f = alloc_empty_file_noaccount(flags, cred);
1196  	if (IS_ERR(f))
1197  		return f;
1198  
1199  	f->f_path = *path;
1200  	error = do_dentry_open(f, NULL);
1201  	if (error) {
1202  		fput(f);
1203  		return ERR_PTR(error);
1204  	}
1205  
1206  	fsnotify_open(f);
1207  	return f;
1208  }
1209  EXPORT_SYMBOL_GPL(kernel_file_open);
1210  
1211  #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1212  #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1213  
1214  inline struct open_how build_open_how(int flags, umode_t mode)
1215  {
1216  	struct open_how how = {
1217  		.flags = flags & VALID_OPEN_FLAGS,
1218  		.mode = mode & S_IALLUGO,
1219  	};
1220  
1221  	/* O_PATH beats everything else. */
1222  	if (how.flags & O_PATH)
1223  		how.flags &= O_PATH_FLAGS;
1224  	/* Modes should only be set for create-like flags. */
1225  	if (!WILL_CREATE(how.flags))
1226  		how.mode = 0;
1227  	return how;
1228  }
1229  
1230  inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1231  {
1232  	u64 flags = how->flags;
1233  	u64 strip = O_CLOEXEC;
1234  	int lookup_flags = 0;
1235  	int acc_mode = ACC_MODE(flags);
1236  
1237  	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1238  			 "struct open_flags doesn't yet handle flags > 32 bits");
1239  
1240  	/*
1241  	 * Strip flags that aren't relevant in determining struct open_flags.
1242  	 */
1243  	flags &= ~strip;
1244  
1245  	/*
1246  	 * Older syscalls implicitly clear all of the invalid flags or argument
1247  	 * values before calling build_open_flags(), but openat2(2) checks all
1248  	 * of its arguments.
1249  	 */
1250  	if (flags & ~VALID_OPEN_FLAGS)
1251  		return -EINVAL;
1252  	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1253  		return -EINVAL;
1254  
1255  	/* Scoping flags are mutually exclusive. */
1256  	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1257  		return -EINVAL;
1258  
1259  	/* Deal with the mode. */
1260  	if (WILL_CREATE(flags)) {
1261  		if (how->mode & ~S_IALLUGO)
1262  			return -EINVAL;
1263  		op->mode = how->mode | S_IFREG;
1264  	} else {
1265  		if (how->mode != 0)
1266  			return -EINVAL;
1267  		op->mode = 0;
1268  	}
1269  
1270  	/*
1271  	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1272  	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1273  	 * O_TMPFILE below which requires O_DIRECTORY being raised.
1274  	 */
1275  	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1276  		return -EINVAL;
1277  
1278  	/* Now handle the creative implementation of O_TMPFILE. */
1279  	if (flags & __O_TMPFILE) {
1280  		/*
1281  		 * In order to ensure programs get explicit errors when trying
1282  		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1283  		 * is raised alongside __O_TMPFILE.
1284  		 */
1285  		if (!(flags & O_DIRECTORY))
1286  			return -EINVAL;
1287  		if (!(acc_mode & MAY_WRITE))
1288  			return -EINVAL;
1289  	}
1290  	if (flags & O_PATH) {
1291  		/* O_PATH only permits certain other flags to be set. */
1292  		if (flags & ~O_PATH_FLAGS)
1293  			return -EINVAL;
1294  		acc_mode = 0;
1295  	}
1296  
1297  	/*
1298  	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1299  	 * check for O_DSYNC if the need any syncing at all we enforce it's
1300  	 * always set instead of having to deal with possibly weird behaviour
1301  	 * for malicious applications setting only __O_SYNC.
1302  	 */
1303  	if (flags & __O_SYNC)
1304  		flags |= O_DSYNC;
1305  
1306  	op->open_flag = flags;
1307  
1308  	/* O_TRUNC implies we need access checks for write permissions */
1309  	if (flags & O_TRUNC)
1310  		acc_mode |= MAY_WRITE;
1311  
1312  	/* Allow the LSM permission hook to distinguish append
1313  	   access from general write access. */
1314  	if (flags & O_APPEND)
1315  		acc_mode |= MAY_APPEND;
1316  
1317  	op->acc_mode = acc_mode;
1318  
1319  	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1320  
1321  	if (flags & O_CREAT) {
1322  		op->intent |= LOOKUP_CREATE;
1323  		if (flags & O_EXCL) {
1324  			op->intent |= LOOKUP_EXCL;
1325  			flags |= O_NOFOLLOW;
1326  		}
1327  	}
1328  
1329  	if (flags & O_DIRECTORY)
1330  		lookup_flags |= LOOKUP_DIRECTORY;
1331  	if (!(flags & O_NOFOLLOW))
1332  		lookup_flags |= LOOKUP_FOLLOW;
1333  
1334  	if (how->resolve & RESOLVE_NO_XDEV)
1335  		lookup_flags |= LOOKUP_NO_XDEV;
1336  	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1337  		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1338  	if (how->resolve & RESOLVE_NO_SYMLINKS)
1339  		lookup_flags |= LOOKUP_NO_SYMLINKS;
1340  	if (how->resolve & RESOLVE_BENEATH)
1341  		lookup_flags |= LOOKUP_BENEATH;
1342  	if (how->resolve & RESOLVE_IN_ROOT)
1343  		lookup_flags |= LOOKUP_IN_ROOT;
1344  	if (how->resolve & RESOLVE_CACHED) {
1345  		/* Don't bother even trying for create/truncate/tmpfile open */
1346  		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1347  			return -EAGAIN;
1348  		lookup_flags |= LOOKUP_CACHED;
1349  	}
1350  
1351  	op->lookup_flags = lookup_flags;
1352  	return 0;
1353  }
1354  
1355  /**
1356   * file_open_name - open file and return file pointer
1357   *
1358   * @name:	struct filename containing path to open
1359   * @flags:	open flags as per the open(2) second argument
1360   * @mode:	mode for the new file if O_CREAT is set, else ignored
1361   *
1362   * This is the helper to open a file from kernelspace if you really
1363   * have to.  But in generally you should not do this, so please move
1364   * along, nothing to see here..
1365   */
1366  struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1367  {
1368  	struct open_flags op;
1369  	struct open_how how = build_open_how(flags, mode);
1370  	int err = build_open_flags(&how, &op);
1371  	if (err)
1372  		return ERR_PTR(err);
1373  	return do_filp_open(AT_FDCWD, name, &op);
1374  }
1375  
1376  /**
1377   * filp_open - open file and return file pointer
1378   *
1379   * @filename:	path to open
1380   * @flags:	open flags as per the open(2) second argument
1381   * @mode:	mode for the new file if O_CREAT is set, else ignored
1382   *
1383   * This is the helper to open a file from kernelspace if you really
1384   * have to.  But in generally you should not do this, so please move
1385   * along, nothing to see here..
1386   */
1387  struct file *filp_open(const char *filename, int flags, umode_t mode)
1388  {
1389  	struct filename *name = getname_kernel(filename);
1390  	struct file *file = ERR_CAST(name);
1391  
1392  	if (!IS_ERR(name)) {
1393  		file = file_open_name(name, flags, mode);
1394  		putname(name);
1395  	}
1396  	return file;
1397  }
1398  EXPORT_SYMBOL(filp_open);
1399  
1400  struct file *file_open_root(const struct path *root,
1401  			    const char *filename, int flags, umode_t mode)
1402  {
1403  	struct open_flags op;
1404  	struct open_how how = build_open_how(flags, mode);
1405  	int err = build_open_flags(&how, &op);
1406  	if (err)
1407  		return ERR_PTR(err);
1408  	return do_file_open_root(root, filename, &op);
1409  }
1410  EXPORT_SYMBOL(file_open_root);
1411  
1412  static long do_sys_openat2(int dfd, const char __user *filename,
1413  			   struct open_how *how)
1414  {
1415  	struct open_flags op;
1416  	int fd = build_open_flags(how, &op);
1417  	struct filename *tmp;
1418  
1419  	if (fd)
1420  		return fd;
1421  
1422  	tmp = getname(filename);
1423  	if (IS_ERR(tmp))
1424  		return PTR_ERR(tmp);
1425  
1426  	fd = get_unused_fd_flags(how->flags);
1427  	if (fd >= 0) {
1428  		struct file *f = do_filp_open(dfd, tmp, &op);
1429  		if (IS_ERR(f)) {
1430  			put_unused_fd(fd);
1431  			fd = PTR_ERR(f);
1432  		} else {
1433  			fd_install(fd, f);
1434  		}
1435  	}
1436  	putname(tmp);
1437  	return fd;
1438  }
1439  
1440  long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1441  {
1442  	struct open_how how = build_open_how(flags, mode);
1443  	return do_sys_openat2(dfd, filename, &how);
1444  }
1445  
1446  
1447  SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1448  {
1449  	if (force_o_largefile())
1450  		flags |= O_LARGEFILE;
1451  	return do_sys_open(AT_FDCWD, filename, flags, mode);
1452  }
1453  
1454  SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1455  		umode_t, mode)
1456  {
1457  	if (force_o_largefile())
1458  		flags |= O_LARGEFILE;
1459  	return do_sys_open(dfd, filename, flags, mode);
1460  }
1461  
1462  SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1463  		struct open_how __user *, how, size_t, usize)
1464  {
1465  	int err;
1466  	struct open_how tmp;
1467  
1468  	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1469  	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1470  
1471  	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1472  		return -EINVAL;
1473  	if (unlikely(usize > PAGE_SIZE))
1474  		return -E2BIG;
1475  
1476  	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1477  	if (err)
1478  		return err;
1479  
1480  	audit_openat2_how(&tmp);
1481  
1482  	/* O_LARGEFILE is only allowed for non-O_PATH. */
1483  	if (!(tmp.flags & O_PATH) && force_o_largefile())
1484  		tmp.flags |= O_LARGEFILE;
1485  
1486  	return do_sys_openat2(dfd, filename, &tmp);
1487  }
1488  
1489  #ifdef CONFIG_COMPAT
1490  /*
1491   * Exactly like sys_open(), except that it doesn't set the
1492   * O_LARGEFILE flag.
1493   */
1494  COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1495  {
1496  	return do_sys_open(AT_FDCWD, filename, flags, mode);
1497  }
1498  
1499  /*
1500   * Exactly like sys_openat(), except that it doesn't set the
1501   * O_LARGEFILE flag.
1502   */
1503  COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1504  {
1505  	return do_sys_open(dfd, filename, flags, mode);
1506  }
1507  #endif
1508  
1509  #ifndef __alpha__
1510  
1511  /*
1512   * For backward compatibility?  Maybe this should be moved
1513   * into arch/i386 instead?
1514   */
1515  SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1516  {
1517  	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1518  
1519  	if (force_o_largefile())
1520  		flags |= O_LARGEFILE;
1521  	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1522  }
1523  #endif
1524  
1525  /*
1526   * "id" is the POSIX thread ID. We use the
1527   * files pointer for this..
1528   */
1529  static int filp_flush(struct file *filp, fl_owner_t id)
1530  {
1531  	int retval = 0;
1532  
1533  	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0, filp,
1534  			"VFS: Close: file count is 0 (f_op=%ps)",
1535  			filp->f_op)) {
1536  		return 0;
1537  	}
1538  
1539  	if (filp->f_op->flush)
1540  		retval = filp->f_op->flush(filp, id);
1541  
1542  	if (likely(!(filp->f_mode & FMODE_PATH))) {
1543  		dnotify_flush(filp, id);
1544  		locks_remove_posix(filp, id);
1545  	}
1546  	return retval;
1547  }
1548  
1549  int filp_close(struct file *filp, fl_owner_t id)
1550  {
1551  	int retval;
1552  
1553  	retval = filp_flush(filp, id);
1554  	fput(filp);
1555  
1556  	return retval;
1557  }
1558  EXPORT_SYMBOL(filp_close);
1559  
1560  /*
1561   * Careful here! We test whether the file pointer is NULL before
1562   * releasing the fd. This ensures that one clone task can't release
1563   * an fd while another clone is opening it.
1564   */
1565  SYSCALL_DEFINE1(close, unsigned int, fd)
1566  {
1567  	int retval;
1568  	struct file *file;
1569  
1570  	file = file_close_fd(fd);
1571  	if (!file)
1572  		return -EBADF;
1573  
1574  	retval = filp_flush(file, current->files);
1575  
1576  	/*
1577  	 * We're returning to user space. Don't bother
1578  	 * with any delayed fput() cases.
1579  	 */
1580  	__fput_sync(file);
1581  
1582  	/* can't restart close syscall because file table entry was cleared */
1583  	if (unlikely(retval == -ERESTARTSYS ||
1584  		     retval == -ERESTARTNOINTR ||
1585  		     retval == -ERESTARTNOHAND ||
1586  		     retval == -ERESTART_RESTARTBLOCK))
1587  		retval = -EINTR;
1588  
1589  	return retval;
1590  }
1591  
1592  /*
1593   * This routine simulates a hangup on the tty, to arrange that users
1594   * are given clean terminals at login time.
1595   */
1596  SYSCALL_DEFINE0(vhangup)
1597  {
1598  	if (capable(CAP_SYS_TTY_CONFIG)) {
1599  		tty_vhangup_self();
1600  		return 0;
1601  	}
1602  	return -EPERM;
1603  }
1604  
1605  /*
1606   * Called when an inode is about to be open.
1607   * We use this to disallow opening large files on 32bit systems if
1608   * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1609   * on this flag in sys_open.
1610   */
1611  int generic_file_open(struct inode * inode, struct file * filp)
1612  {
1613  	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1614  		return -EOVERFLOW;
1615  	return 0;
1616  }
1617  
1618  EXPORT_SYMBOL(generic_file_open);
1619  
1620  /*
1621   * This is used by subsystems that don't want seekable
1622   * file descriptors. The function is not supposed to ever fail, the only
1623   * reason it returns an 'int' and not 'void' is so that it can be plugged
1624   * directly into file_operations structure.
1625   */
1626  int nonseekable_open(struct inode *inode, struct file *filp)
1627  {
1628  	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1629  	return 0;
1630  }
1631  
1632  EXPORT_SYMBOL(nonseekable_open);
1633  
1634  /*
1635   * stream_open is used by subsystems that want stream-like file descriptors.
1636   * Such file descriptors are not seekable and don't have notion of position
1637   * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1638   * Contrary to file descriptors of other regular files, .read() and .write()
1639   * can run simultaneously.
1640   *
1641   * stream_open never fails and is marked to return int so that it could be
1642   * directly used as file_operations.open .
1643   */
1644  int stream_open(struct inode *inode, struct file *filp)
1645  {
1646  	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1647  	filp->f_mode |= FMODE_STREAM;
1648  	return 0;
1649  }
1650  
1651  EXPORT_SYMBOL(stream_open);
1652