xref: /openbmc/linux/fs/overlayfs/file.c (revision 5b394b2d) 
1 /*
2  * Copyright (C) 2017 Red Hat, Inc.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of the GNU General Public License version 2 as published by
6  * the Free Software Foundation.
7  */
8 
9 #include <linux/cred.h>
10 #include <linux/file.h>
11 #include <linux/mount.h>
12 #include <linux/xattr.h>
13 #include <linux/uio.h>
14 #include "overlayfs.h"
15 
16 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
17 {
18 	if (realinode != ovl_inode_upper(inode))
19 		return 'l';
20 	if (ovl_has_upperdata(inode))
21 		return 'u';
22 	else
23 		return 'm';
24 }
25 
26 static struct file *ovl_open_realfile(const struct file *file,
27 				      struct inode *realinode)
28 {
29 	struct inode *inode = file_inode(file);
30 	struct file *realfile;
31 	const struct cred *old_cred;
32 
33 	old_cred = ovl_override_creds(inode->i_sb);
34 	realfile = open_with_fake_path(&file->f_path, file->f_flags | O_NOATIME,
35 				       realinode, current_cred());
36 	revert_creds(old_cred);
37 
38 	pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
39 		 file, file, ovl_whatisit(inode, realinode), file->f_flags,
40 		 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
41 
42 	return realfile;
43 }
44 
45 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
46 
47 static int ovl_change_flags(struct file *file, unsigned int flags)
48 {
49 	struct inode *inode = file_inode(file);
50 	int err;
51 
52 	/* No atime modificaton on underlying */
53 	flags |= O_NOATIME;
54 
55 	/* If some flag changed that cannot be changed then something's amiss */
56 	if (WARN_ON((file->f_flags ^ flags) & ~OVL_SETFL_MASK))
57 		return -EIO;
58 
59 	flags &= OVL_SETFL_MASK;
60 
61 	if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
62 		return -EPERM;
63 
64 	if (flags & O_DIRECT) {
65 		if (!file->f_mapping->a_ops ||
66 		    !file->f_mapping->a_ops->direct_IO)
67 			return -EINVAL;
68 	}
69 
70 	if (file->f_op->check_flags) {
71 		err = file->f_op->check_flags(flags);
72 		if (err)
73 			return err;
74 	}
75 
76 	spin_lock(&file->f_lock);
77 	file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
78 	spin_unlock(&file->f_lock);
79 
80 	return 0;
81 }
82 
83 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
84 			       bool allow_meta)
85 {
86 	struct inode *inode = file_inode(file);
87 	struct inode *realinode;
88 
89 	real->flags = 0;
90 	real->file = file->private_data;
91 
92 	if (allow_meta)
93 		realinode = ovl_inode_real(inode);
94 	else
95 		realinode = ovl_inode_realdata(inode);
96 
97 	/* Has it been copied up since we'd opened it? */
98 	if (unlikely(file_inode(real->file) != realinode)) {
99 		real->flags = FDPUT_FPUT;
100 		real->file = ovl_open_realfile(file, realinode);
101 
102 		return PTR_ERR_OR_ZERO(real->file);
103 	}
104 
105 	/* Did the flags change since open? */
106 	if (unlikely((file->f_flags ^ real->file->f_flags) & ~O_NOATIME))
107 		return ovl_change_flags(real->file, file->f_flags);
108 
109 	return 0;
110 }
111 
112 static int ovl_real_fdget(const struct file *file, struct fd *real)
113 {
114 	return ovl_real_fdget_meta(file, real, false);
115 }
116 
117 static int ovl_open(struct inode *inode, struct file *file)
118 {
119 	struct dentry *dentry = file_dentry(file);
120 	struct file *realfile;
121 	int err;
122 
123 	err = ovl_open_maybe_copy_up(dentry, file->f_flags);
124 	if (err)
125 		return err;
126 
127 	/* No longer need these flags, so don't pass them on to underlying fs */
128 	file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
129 
130 	realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
131 	if (IS_ERR(realfile))
132 		return PTR_ERR(realfile);
133 
134 	/* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
135 	file->f_mapping = realfile->f_mapping;
136 
137 	file->private_data = realfile;
138 
139 	return 0;
140 }
141 
142 static int ovl_release(struct inode *inode, struct file *file)
143 {
144 	fput(file->private_data);
145 
146 	return 0;
147 }
148 
149 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
150 {
151 	struct inode *realinode = ovl_inode_real(file_inode(file));
152 
153 	return generic_file_llseek_size(file, offset, whence,
154 					realinode->i_sb->s_maxbytes,
155 					i_size_read(realinode));
156 }
157 
158 static void ovl_file_accessed(struct file *file)
159 {
160 	struct inode *inode, *upperinode;
161 
162 	if (file->f_flags & O_NOATIME)
163 		return;
164 
165 	inode = file_inode(file);
166 	upperinode = ovl_inode_upper(inode);
167 
168 	if (!upperinode)
169 		return;
170 
171 	if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
172 	     !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
173 		inode->i_mtime = upperinode->i_mtime;
174 		inode->i_ctime = upperinode->i_ctime;
175 	}
176 
177 	touch_atime(&file->f_path);
178 }
179 
180 static rwf_t ovl_iocb_to_rwf(struct kiocb *iocb)
181 {
182 	int ifl = iocb->ki_flags;
183 	rwf_t flags = 0;
184 
185 	if (ifl & IOCB_NOWAIT)
186 		flags |= RWF_NOWAIT;
187 	if (ifl & IOCB_HIPRI)
188 		flags |= RWF_HIPRI;
189 	if (ifl & IOCB_DSYNC)
190 		flags |= RWF_DSYNC;
191 	if (ifl & IOCB_SYNC)
192 		flags |= RWF_SYNC;
193 
194 	return flags;
195 }
196 
197 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
198 {
199 	struct file *file = iocb->ki_filp;
200 	struct fd real;
201 	const struct cred *old_cred;
202 	ssize_t ret;
203 
204 	if (!iov_iter_count(iter))
205 		return 0;
206 
207 	ret = ovl_real_fdget(file, &real);
208 	if (ret)
209 		return ret;
210 
211 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
212 	ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
213 			    ovl_iocb_to_rwf(iocb));
214 	revert_creds(old_cred);
215 
216 	ovl_file_accessed(file);
217 
218 	fdput(real);
219 
220 	return ret;
221 }
222 
223 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
224 {
225 	struct file *file = iocb->ki_filp;
226 	struct inode *inode = file_inode(file);
227 	struct fd real;
228 	const struct cred *old_cred;
229 	ssize_t ret;
230 
231 	if (!iov_iter_count(iter))
232 		return 0;
233 
234 	inode_lock(inode);
235 	/* Update mode */
236 	ovl_copyattr(ovl_inode_real(inode), inode);
237 	ret = file_remove_privs(file);
238 	if (ret)
239 		goto out_unlock;
240 
241 	ret = ovl_real_fdget(file, &real);
242 	if (ret)
243 		goto out_unlock;
244 
245 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
246 	ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
247 			     ovl_iocb_to_rwf(iocb));
248 	revert_creds(old_cred);
249 
250 	/* Update size */
251 	ovl_copyattr(ovl_inode_real(inode), inode);
252 
253 	fdput(real);
254 
255 out_unlock:
256 	inode_unlock(inode);
257 
258 	return ret;
259 }
260 
261 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
262 {
263 	struct fd real;
264 	const struct cred *old_cred;
265 	int ret;
266 
267 	ret = ovl_real_fdget_meta(file, &real, !datasync);
268 	if (ret)
269 		return ret;
270 
271 	/* Don't sync lower file for fear of receiving EROFS error */
272 	if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
273 		old_cred = ovl_override_creds(file_inode(file)->i_sb);
274 		ret = vfs_fsync_range(real.file, start, end, datasync);
275 		revert_creds(old_cred);
276 	}
277 
278 	fdput(real);
279 
280 	return ret;
281 }
282 
283 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
284 {
285 	struct file *realfile = file->private_data;
286 	const struct cred *old_cred;
287 	int ret;
288 
289 	if (!realfile->f_op->mmap)
290 		return -ENODEV;
291 
292 	if (WARN_ON(file != vma->vm_file))
293 		return -EIO;
294 
295 	vma->vm_file = get_file(realfile);
296 
297 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
298 	ret = call_mmap(vma->vm_file, vma);
299 	revert_creds(old_cred);
300 
301 	if (ret) {
302 		/* Drop reference count from new vm_file value */
303 		fput(realfile);
304 	} else {
305 		/* Drop reference count from previous vm_file value */
306 		fput(file);
307 	}
308 
309 	ovl_file_accessed(file);
310 
311 	return ret;
312 }
313 
314 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
315 {
316 	struct inode *inode = file_inode(file);
317 	struct fd real;
318 	const struct cred *old_cred;
319 	int ret;
320 
321 	ret = ovl_real_fdget(file, &real);
322 	if (ret)
323 		return ret;
324 
325 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
326 	ret = vfs_fallocate(real.file, mode, offset, len);
327 	revert_creds(old_cred);
328 
329 	/* Update size */
330 	ovl_copyattr(ovl_inode_real(inode), inode);
331 
332 	fdput(real);
333 
334 	return ret;
335 }
336 
337 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
338 			   unsigned long arg)
339 {
340 	struct fd real;
341 	const struct cred *old_cred;
342 	long ret;
343 
344 	ret = ovl_real_fdget(file, &real);
345 	if (ret)
346 		return ret;
347 
348 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
349 	ret = vfs_ioctl(real.file, cmd, arg);
350 	revert_creds(old_cred);
351 
352 	fdput(real);
353 
354 	return ret;
355 }
356 
357 static long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
358 {
359 	long ret;
360 	struct inode *inode = file_inode(file);
361 
362 	switch (cmd) {
363 	case FS_IOC_GETFLAGS:
364 		ret = ovl_real_ioctl(file, cmd, arg);
365 		break;
366 
367 	case FS_IOC_SETFLAGS:
368 		if (!inode_owner_or_capable(inode))
369 			return -EACCES;
370 
371 		ret = mnt_want_write_file(file);
372 		if (ret)
373 			return ret;
374 
375 		ret = ovl_copy_up_with_data(file_dentry(file));
376 		if (!ret) {
377 			ret = ovl_real_ioctl(file, cmd, arg);
378 
379 			inode_lock(inode);
380 			ovl_copyflags(ovl_inode_real(inode), inode);
381 			inode_unlock(inode);
382 		}
383 
384 		mnt_drop_write_file(file);
385 		break;
386 
387 	default:
388 		ret = -ENOTTY;
389 	}
390 
391 	return ret;
392 }
393 
394 static long ovl_compat_ioctl(struct file *file, unsigned int cmd,
395 			     unsigned long arg)
396 {
397 	switch (cmd) {
398 	case FS_IOC32_GETFLAGS:
399 		cmd = FS_IOC_GETFLAGS;
400 		break;
401 
402 	case FS_IOC32_SETFLAGS:
403 		cmd = FS_IOC_SETFLAGS;
404 		break;
405 
406 	default:
407 		return -ENOIOCTLCMD;
408 	}
409 
410 	return ovl_ioctl(file, cmd, arg);
411 }
412 
413 enum ovl_copyop {
414 	OVL_COPY,
415 	OVL_CLONE,
416 	OVL_DEDUPE,
417 };
418 
419 static ssize_t ovl_copyfile(struct file *file_in, loff_t pos_in,
420 			    struct file *file_out, loff_t pos_out,
421 			    u64 len, unsigned int flags, enum ovl_copyop op)
422 {
423 	struct inode *inode_out = file_inode(file_out);
424 	struct fd real_in, real_out;
425 	const struct cred *old_cred;
426 	ssize_t ret;
427 
428 	ret = ovl_real_fdget(file_out, &real_out);
429 	if (ret)
430 		return ret;
431 
432 	ret = ovl_real_fdget(file_in, &real_in);
433 	if (ret) {
434 		fdput(real_out);
435 		return ret;
436 	}
437 
438 	old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
439 	switch (op) {
440 	case OVL_COPY:
441 		ret = vfs_copy_file_range(real_in.file, pos_in,
442 					  real_out.file, pos_out, len, flags);
443 		break;
444 
445 	case OVL_CLONE:
446 		ret = vfs_clone_file_range(real_in.file, pos_in,
447 					   real_out.file, pos_out, len);
448 		break;
449 
450 	case OVL_DEDUPE:
451 		ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
452 						real_out.file, pos_out, len);
453 		break;
454 	}
455 	revert_creds(old_cred);
456 
457 	/* Update size */
458 	ovl_copyattr(ovl_inode_real(inode_out), inode_out);
459 
460 	fdput(real_in);
461 	fdput(real_out);
462 
463 	return ret;
464 }
465 
466 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
467 				   struct file *file_out, loff_t pos_out,
468 				   size_t len, unsigned int flags)
469 {
470 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
471 			    OVL_COPY);
472 }
473 
474 static int ovl_clone_file_range(struct file *file_in, loff_t pos_in,
475 				struct file *file_out, loff_t pos_out, u64 len)
476 {
477 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, 0,
478 			    OVL_CLONE);
479 }
480 
481 static int ovl_dedupe_file_range(struct file *file_in, loff_t pos_in,
482 				 struct file *file_out, loff_t pos_out, u64 len)
483 {
484 	/*
485 	 * Don't copy up because of a dedupe request, this wouldn't make sense
486 	 * most of the time (data would be duplicated instead of deduplicated).
487 	 */
488 	if (!ovl_inode_upper(file_inode(file_in)) ||
489 	    !ovl_inode_upper(file_inode(file_out)))
490 		return -EPERM;
491 
492 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, 0,
493 			    OVL_DEDUPE);
494 }
495 
496 const struct file_operations ovl_file_operations = {
497 	.open		= ovl_open,
498 	.release	= ovl_release,
499 	.llseek		= ovl_llseek,
500 	.read_iter	= ovl_read_iter,
501 	.write_iter	= ovl_write_iter,
502 	.fsync		= ovl_fsync,
503 	.mmap		= ovl_mmap,
504 	.fallocate	= ovl_fallocate,
505 	.unlocked_ioctl	= ovl_ioctl,
506 	.compat_ioctl	= ovl_compat_ioctl,
507 
508 	.copy_file_range	= ovl_copy_file_range,
509 	.clone_file_range	= ovl_clone_file_range,
510 	.dedupe_file_range	= ovl_dedupe_file_range,
511 };
512