xref: /openbmc/linux/fs/overlayfs/file.c (revision d4295e12) 
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 	file->private_data = realfile;
135 
136 	return 0;
137 }
138 
139 static int ovl_release(struct inode *inode, struct file *file)
140 {
141 	fput(file->private_data);
142 
143 	return 0;
144 }
145 
146 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
147 {
148 	struct inode *realinode = ovl_inode_real(file_inode(file));
149 
150 	return generic_file_llseek_size(file, offset, whence,
151 					realinode->i_sb->s_maxbytes,
152 					i_size_read(realinode));
153 }
154 
155 static void ovl_file_accessed(struct file *file)
156 {
157 	struct inode *inode, *upperinode;
158 
159 	if (file->f_flags & O_NOATIME)
160 		return;
161 
162 	inode = file_inode(file);
163 	upperinode = ovl_inode_upper(inode);
164 
165 	if (!upperinode)
166 		return;
167 
168 	if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
169 	     !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
170 		inode->i_mtime = upperinode->i_mtime;
171 		inode->i_ctime = upperinode->i_ctime;
172 	}
173 
174 	touch_atime(&file->f_path);
175 }
176 
177 static rwf_t ovl_iocb_to_rwf(struct kiocb *iocb)
178 {
179 	int ifl = iocb->ki_flags;
180 	rwf_t flags = 0;
181 
182 	if (ifl & IOCB_NOWAIT)
183 		flags |= RWF_NOWAIT;
184 	if (ifl & IOCB_HIPRI)
185 		flags |= RWF_HIPRI;
186 	if (ifl & IOCB_DSYNC)
187 		flags |= RWF_DSYNC;
188 	if (ifl & IOCB_SYNC)
189 		flags |= RWF_SYNC;
190 
191 	return flags;
192 }
193 
194 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
195 {
196 	struct file *file = iocb->ki_filp;
197 	struct fd real;
198 	const struct cred *old_cred;
199 	ssize_t ret;
200 
201 	if (!iov_iter_count(iter))
202 		return 0;
203 
204 	ret = ovl_real_fdget(file, &real);
205 	if (ret)
206 		return ret;
207 
208 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
209 	ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
210 			    ovl_iocb_to_rwf(iocb));
211 	revert_creds(old_cred);
212 
213 	ovl_file_accessed(file);
214 
215 	fdput(real);
216 
217 	return ret;
218 }
219 
220 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
221 {
222 	struct file *file = iocb->ki_filp;
223 	struct inode *inode = file_inode(file);
224 	struct fd real;
225 	const struct cred *old_cred;
226 	ssize_t ret;
227 
228 	if (!iov_iter_count(iter))
229 		return 0;
230 
231 	inode_lock(inode);
232 	/* Update mode */
233 	ovl_copyattr(ovl_inode_real(inode), inode);
234 	ret = file_remove_privs(file);
235 	if (ret)
236 		goto out_unlock;
237 
238 	ret = ovl_real_fdget(file, &real);
239 	if (ret)
240 		goto out_unlock;
241 
242 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
243 	file_start_write(real.file);
244 	ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
245 			     ovl_iocb_to_rwf(iocb));
246 	file_end_write(real.file);
247 	revert_creds(old_cred);
248 
249 	/* Update size */
250 	ovl_copyattr(ovl_inode_real(inode), inode);
251 
252 	fdput(real);
253 
254 out_unlock:
255 	inode_unlock(inode);
256 
257 	return ret;
258 }
259 
260 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
261 {
262 	struct fd real;
263 	const struct cred *old_cred;
264 	int ret;
265 
266 	ret = ovl_real_fdget_meta(file, &real, !datasync);
267 	if (ret)
268 		return ret;
269 
270 	/* Don't sync lower file for fear of receiving EROFS error */
271 	if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
272 		old_cred = ovl_override_creds(file_inode(file)->i_sb);
273 		ret = vfs_fsync_range(real.file, start, end, datasync);
274 		revert_creds(old_cred);
275 	}
276 
277 	fdput(real);
278 
279 	return ret;
280 }
281 
282 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
283 {
284 	struct file *realfile = file->private_data;
285 	const struct cred *old_cred;
286 	int ret;
287 
288 	if (!realfile->f_op->mmap)
289 		return -ENODEV;
290 
291 	if (WARN_ON(file != vma->vm_file))
292 		return -EIO;
293 
294 	vma->vm_file = get_file(realfile);
295 
296 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
297 	ret = call_mmap(vma->vm_file, vma);
298 	revert_creds(old_cred);
299 
300 	if (ret) {
301 		/* Drop reference count from new vm_file value */
302 		fput(realfile);
303 	} else {
304 		/* Drop reference count from previous vm_file value */
305 		fput(file);
306 	}
307 
308 	ovl_file_accessed(file);
309 
310 	return ret;
311 }
312 
313 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
314 {
315 	struct inode *inode = file_inode(file);
316 	struct fd real;
317 	const struct cred *old_cred;
318 	int ret;
319 
320 	ret = ovl_real_fdget(file, &real);
321 	if (ret)
322 		return ret;
323 
324 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
325 	ret = vfs_fallocate(real.file, mode, offset, len);
326 	revert_creds(old_cred);
327 
328 	/* Update size */
329 	ovl_copyattr(ovl_inode_real(inode), inode);
330 
331 	fdput(real);
332 
333 	return ret;
334 }
335 
336 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
337 {
338 	struct fd real;
339 	const struct cred *old_cred;
340 	int ret;
341 
342 	ret = ovl_real_fdget(file, &real);
343 	if (ret)
344 		return ret;
345 
346 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
347 	ret = vfs_fadvise(real.file, offset, len, advice);
348 	revert_creds(old_cred);
349 
350 	fdput(real);
351 
352 	return ret;
353 }
354 
355 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
356 			   unsigned long arg)
357 {
358 	struct fd real;
359 	const struct cred *old_cred;
360 	long ret;
361 
362 	ret = ovl_real_fdget(file, &real);
363 	if (ret)
364 		return ret;
365 
366 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
367 	ret = vfs_ioctl(real.file, cmd, arg);
368 	revert_creds(old_cred);
369 
370 	fdput(real);
371 
372 	return ret;
373 }
374 
375 static long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
376 {
377 	long ret;
378 	struct inode *inode = file_inode(file);
379 
380 	switch (cmd) {
381 	case FS_IOC_GETFLAGS:
382 		ret = ovl_real_ioctl(file, cmd, arg);
383 		break;
384 
385 	case FS_IOC_SETFLAGS:
386 		if (!inode_owner_or_capable(inode))
387 			return -EACCES;
388 
389 		ret = mnt_want_write_file(file);
390 		if (ret)
391 			return ret;
392 
393 		ret = ovl_copy_up_with_data(file_dentry(file));
394 		if (!ret) {
395 			ret = ovl_real_ioctl(file, cmd, arg);
396 
397 			inode_lock(inode);
398 			ovl_copyflags(ovl_inode_real(inode), inode);
399 			inode_unlock(inode);
400 		}
401 
402 		mnt_drop_write_file(file);
403 		break;
404 
405 	default:
406 		ret = -ENOTTY;
407 	}
408 
409 	return ret;
410 }
411 
412 static long ovl_compat_ioctl(struct file *file, unsigned int cmd,
413 			     unsigned long arg)
414 {
415 	switch (cmd) {
416 	case FS_IOC32_GETFLAGS:
417 		cmd = FS_IOC_GETFLAGS;
418 		break;
419 
420 	case FS_IOC32_SETFLAGS:
421 		cmd = FS_IOC_SETFLAGS;
422 		break;
423 
424 	default:
425 		return -ENOIOCTLCMD;
426 	}
427 
428 	return ovl_ioctl(file, cmd, arg);
429 }
430 
431 enum ovl_copyop {
432 	OVL_COPY,
433 	OVL_CLONE,
434 	OVL_DEDUPE,
435 };
436 
437 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
438 			    struct file *file_out, loff_t pos_out,
439 			    loff_t len, unsigned int flags, enum ovl_copyop op)
440 {
441 	struct inode *inode_out = file_inode(file_out);
442 	struct fd real_in, real_out;
443 	const struct cred *old_cred;
444 	loff_t ret;
445 
446 	ret = ovl_real_fdget(file_out, &real_out);
447 	if (ret)
448 		return ret;
449 
450 	ret = ovl_real_fdget(file_in, &real_in);
451 	if (ret) {
452 		fdput(real_out);
453 		return ret;
454 	}
455 
456 	old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
457 	switch (op) {
458 	case OVL_COPY:
459 		ret = vfs_copy_file_range(real_in.file, pos_in,
460 					  real_out.file, pos_out, len, flags);
461 		break;
462 
463 	case OVL_CLONE:
464 		ret = vfs_clone_file_range(real_in.file, pos_in,
465 					   real_out.file, pos_out, len, flags);
466 		break;
467 
468 	case OVL_DEDUPE:
469 		ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
470 						real_out.file, pos_out, len,
471 						flags);
472 		break;
473 	}
474 	revert_creds(old_cred);
475 
476 	/* Update size */
477 	ovl_copyattr(ovl_inode_real(inode_out), inode_out);
478 
479 	fdput(real_in);
480 	fdput(real_out);
481 
482 	return ret;
483 }
484 
485 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
486 				   struct file *file_out, loff_t pos_out,
487 				   size_t len, unsigned int flags)
488 {
489 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
490 			    OVL_COPY);
491 }
492 
493 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
494 				   struct file *file_out, loff_t pos_out,
495 				   loff_t len, unsigned int remap_flags)
496 {
497 	enum ovl_copyop op;
498 
499 	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
500 		return -EINVAL;
501 
502 	if (remap_flags & REMAP_FILE_DEDUP)
503 		op = OVL_DEDUPE;
504 	else
505 		op = OVL_CLONE;
506 
507 	/*
508 	 * Don't copy up because of a dedupe request, this wouldn't make sense
509 	 * most of the time (data would be duplicated instead of deduplicated).
510 	 */
511 	if (op == OVL_DEDUPE &&
512 	    (!ovl_inode_upper(file_inode(file_in)) ||
513 	     !ovl_inode_upper(file_inode(file_out))))
514 		return -EPERM;
515 
516 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
517 			    remap_flags, op);
518 }
519 
520 const struct file_operations ovl_file_operations = {
521 	.open		= ovl_open,
522 	.release	= ovl_release,
523 	.llseek		= ovl_llseek,
524 	.read_iter	= ovl_read_iter,
525 	.write_iter	= ovl_write_iter,
526 	.fsync		= ovl_fsync,
527 	.mmap		= ovl_mmap,
528 	.fallocate	= ovl_fallocate,
529 	.fadvise	= ovl_fadvise,
530 	.unlocked_ioctl	= ovl_ioctl,
531 	.compat_ioctl	= ovl_compat_ioctl,
532 
533 	.copy_file_range	= ovl_copy_file_range,
534 	.remap_file_range	= ovl_remap_file_range,
535 };
536