xref: /openbmc/linux/fs/overlayfs/namei.c (revision 239480ab)
1 /*
2  * Copyright (C) 2011 Novell Inc.
3  * Copyright (C) 2016 Red Hat, Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published by
7  * the Free Software Foundation.
8  */
9 
10 #include <linux/fs.h>
11 #include <linux/cred.h>
12 #include <linux/namei.h>
13 #include <linux/xattr.h>
14 #include <linux/ratelimit.h>
15 #include <linux/mount.h>
16 #include <linux/exportfs.h>
17 #include "overlayfs.h"
18 #include "ovl_entry.h"
19 
20 struct ovl_lookup_data {
21 	struct qstr name;
22 	bool is_dir;
23 	bool opaque;
24 	bool stop;
25 	bool last;
26 	char *redirect;
27 };
28 
29 static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
30 			      size_t prelen, const char *post)
31 {
32 	int res;
33 	char *s, *next, *buf = NULL;
34 
35 	res = vfs_getxattr(dentry, OVL_XATTR_REDIRECT, NULL, 0);
36 	if (res < 0) {
37 		if (res == -ENODATA || res == -EOPNOTSUPP)
38 			return 0;
39 		goto fail;
40 	}
41 	buf = kzalloc(prelen + res + strlen(post) + 1, GFP_TEMPORARY);
42 	if (!buf)
43 		return -ENOMEM;
44 
45 	if (res == 0)
46 		goto invalid;
47 
48 	res = vfs_getxattr(dentry, OVL_XATTR_REDIRECT, buf, res);
49 	if (res < 0)
50 		goto fail;
51 	if (res == 0)
52 		goto invalid;
53 	if (buf[0] == '/') {
54 		for (s = buf; *s++ == '/'; s = next) {
55 			next = strchrnul(s, '/');
56 			if (s == next)
57 				goto invalid;
58 		}
59 	} else {
60 		if (strchr(buf, '/') != NULL)
61 			goto invalid;
62 
63 		memmove(buf + prelen, buf, res);
64 		memcpy(buf, d->name.name, prelen);
65 	}
66 
67 	strcat(buf, post);
68 	kfree(d->redirect);
69 	d->redirect = buf;
70 	d->name.name = d->redirect;
71 	d->name.len = strlen(d->redirect);
72 
73 	return 0;
74 
75 err_free:
76 	kfree(buf);
77 	return 0;
78 fail:
79 	pr_warn_ratelimited("overlayfs: failed to get redirect (%i)\n", res);
80 	goto err_free;
81 invalid:
82 	pr_warn_ratelimited("overlayfs: invalid redirect (%s)\n", buf);
83 	goto err_free;
84 }
85 
86 static int ovl_acceptable(void *ctx, struct dentry *dentry)
87 {
88 	return 1;
89 }
90 
91 static struct dentry *ovl_get_origin(struct dentry *dentry,
92 				     struct vfsmount *mnt)
93 {
94 	int res;
95 	struct ovl_fh *fh = NULL;
96 	struct dentry *origin = NULL;
97 	int bytes;
98 
99 	res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
100 	if (res < 0) {
101 		if (res == -ENODATA || res == -EOPNOTSUPP)
102 			return NULL;
103 		goto fail;
104 	}
105 	/* Zero size value means "copied up but origin unknown" */
106 	if (res == 0)
107 		return NULL;
108 
109 	fh  = kzalloc(res, GFP_TEMPORARY);
110 	if (!fh)
111 		return ERR_PTR(-ENOMEM);
112 
113 	res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, fh, res);
114 	if (res < 0)
115 		goto fail;
116 
117 	if (res < sizeof(struct ovl_fh) || res < fh->len)
118 		goto invalid;
119 
120 	if (fh->magic != OVL_FH_MAGIC)
121 		goto invalid;
122 
123 	/* Treat larger version and unknown flags as "origin unknown" */
124 	if (fh->version > OVL_FH_VERSION || fh->flags & ~OVL_FH_FLAG_ALL)
125 		goto out;
126 
127 	/* Treat endianness mismatch as "origin unknown" */
128 	if (!(fh->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
129 	    (fh->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
130 		goto out;
131 
132 	bytes = (fh->len - offsetof(struct ovl_fh, fid));
133 
134 	/*
135 	 * Make sure that the stored uuid matches the uuid of the lower
136 	 * layer where file handle will be decoded.
137 	 */
138 	if (uuid_be_cmp(fh->uuid, *(uuid_be *) &mnt->mnt_sb->s_uuid))
139 		goto out;
140 
141 	origin = exportfs_decode_fh(mnt, (struct fid *)fh->fid,
142 				    bytes >> 2, (int)fh->type,
143 				    ovl_acceptable, NULL);
144 	if (IS_ERR(origin)) {
145 		/* Treat stale file handle as "origin unknown" */
146 		if (origin == ERR_PTR(-ESTALE))
147 			origin = NULL;
148 		goto out;
149 	}
150 
151 	if (ovl_dentry_weird(origin) ||
152 	    ((d_inode(origin)->i_mode ^ d_inode(dentry)->i_mode) & S_IFMT)) {
153 		dput(origin);
154 		origin = NULL;
155 		goto invalid;
156 	}
157 
158 out:
159 	kfree(fh);
160 	return origin;
161 
162 fail:
163 	pr_warn_ratelimited("overlayfs: failed to get origin (%i)\n", res);
164 	goto out;
165 invalid:
166 	pr_warn_ratelimited("overlayfs: invalid origin (%*phN)\n", res, fh);
167 	goto out;
168 }
169 
170 static bool ovl_is_opaquedir(struct dentry *dentry)
171 {
172 	int res;
173 	char val;
174 
175 	if (!d_is_dir(dentry))
176 		return false;
177 
178 	res = vfs_getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
179 	if (res == 1 && val == 'y')
180 		return true;
181 
182 	return false;
183 }
184 
185 static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
186 			     const char *name, unsigned int namelen,
187 			     size_t prelen, const char *post,
188 			     struct dentry **ret)
189 {
190 	struct dentry *this;
191 	int err;
192 
193 	this = lookup_one_len_unlocked(name, base, namelen);
194 	if (IS_ERR(this)) {
195 		err = PTR_ERR(this);
196 		this = NULL;
197 		if (err == -ENOENT || err == -ENAMETOOLONG)
198 			goto out;
199 		goto out_err;
200 	}
201 	if (!this->d_inode)
202 		goto put_and_out;
203 
204 	if (ovl_dentry_weird(this)) {
205 		/* Don't support traversing automounts and other weirdness */
206 		err = -EREMOTE;
207 		goto out_err;
208 	}
209 	if (ovl_is_whiteout(this)) {
210 		d->stop = d->opaque = true;
211 		goto put_and_out;
212 	}
213 	if (!d_can_lookup(this)) {
214 		d->stop = true;
215 		if (d->is_dir)
216 			goto put_and_out;
217 		goto out;
218 	}
219 	d->is_dir = true;
220 	if (!d->last && ovl_is_opaquedir(this)) {
221 		d->stop = d->opaque = true;
222 		goto out;
223 	}
224 	err = ovl_check_redirect(this, d, prelen, post);
225 	if (err)
226 		goto out_err;
227 out:
228 	*ret = this;
229 	return 0;
230 
231 put_and_out:
232 	dput(this);
233 	this = NULL;
234 	goto out;
235 
236 out_err:
237 	dput(this);
238 	return err;
239 }
240 
241 static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
242 			    struct dentry **ret)
243 {
244 	/* Counting down from the end, since the prefix can change */
245 	size_t rem = d->name.len - 1;
246 	struct dentry *dentry = NULL;
247 	int err;
248 
249 	if (d->name.name[0] != '/')
250 		return ovl_lookup_single(base, d, d->name.name, d->name.len,
251 					 0, "", ret);
252 
253 	while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
254 		const char *s = d->name.name + d->name.len - rem;
255 		const char *next = strchrnul(s, '/');
256 		size_t thislen = next - s;
257 		bool end = !next[0];
258 
259 		/* Verify we did not go off the rails */
260 		if (WARN_ON(s[-1] != '/'))
261 			return -EIO;
262 
263 		err = ovl_lookup_single(base, d, s, thislen,
264 					d->name.len - rem, next, &base);
265 		dput(dentry);
266 		if (err)
267 			return err;
268 		dentry = base;
269 		if (end)
270 			break;
271 
272 		rem -= thislen + 1;
273 
274 		if (WARN_ON(rem >= d->name.len))
275 			return -EIO;
276 	}
277 	*ret = dentry;
278 	return 0;
279 }
280 
281 
282 static int ovl_check_origin(struct dentry *dentry, struct dentry *upperdentry,
283 			    struct path **stackp, unsigned int *ctrp)
284 {
285 	struct super_block *same_sb = ovl_same_sb(dentry->d_sb);
286 	struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
287 	struct vfsmount *mnt;
288 	struct dentry *origin;
289 
290 	if (!same_sb || !roe->numlower)
291 		return 0;
292 
293        /*
294 	* Since all layers are on the same fs, we use the first layer for
295 	* decoding the file handle.  We may get a disconnected dentry,
296 	* which is fine, because we only need to hold the origin inode in
297 	* cache and use its inode number.  We may even get a connected dentry,
298 	* that is not under the first layer's root.  That is also fine for
299 	* using it's inode number - it's the same as if we held a reference
300 	* to a dentry in first layer that was moved under us.
301 	*/
302 	mnt = roe->lowerstack[0].mnt;
303 
304 	origin = ovl_get_origin(upperdentry, mnt);
305 	if (IS_ERR_OR_NULL(origin))
306 		return PTR_ERR(origin);
307 
308 	BUG_ON(*stackp || *ctrp);
309 	*stackp = kmalloc(sizeof(struct path), GFP_TEMPORARY);
310 	if (!*stackp) {
311 		dput(origin);
312 		return -ENOMEM;
313 	}
314 	**stackp = (struct path) { .dentry = origin, .mnt = mnt };
315 	*ctrp = 1;
316 
317 	return 0;
318 }
319 
320 /*
321  * Returns next layer in stack starting from top.
322  * Returns -1 if this is the last layer.
323  */
324 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
325 {
326 	struct ovl_entry *oe = dentry->d_fsdata;
327 
328 	BUG_ON(idx < 0);
329 	if (idx == 0) {
330 		ovl_path_upper(dentry, path);
331 		if (path->dentry)
332 			return oe->numlower ? 1 : -1;
333 		idx++;
334 	}
335 	BUG_ON(idx > oe->numlower);
336 	*path = oe->lowerstack[idx - 1];
337 
338 	return (idx < oe->numlower) ? idx + 1 : -1;
339 }
340 
341 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
342 			  unsigned int flags)
343 {
344 	struct ovl_entry *oe;
345 	const struct cred *old_cred;
346 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
347 	struct ovl_entry *poe = dentry->d_parent->d_fsdata;
348 	struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
349 	struct path *stack = NULL;
350 	struct dentry *upperdir, *upperdentry = NULL;
351 	unsigned int ctr = 0;
352 	struct inode *inode = NULL;
353 	bool upperopaque = false;
354 	char *upperredirect = NULL;
355 	struct dentry *this;
356 	unsigned int i;
357 	int err;
358 	struct ovl_lookup_data d = {
359 		.name = dentry->d_name,
360 		.is_dir = false,
361 		.opaque = false,
362 		.stop = false,
363 		.last = !poe->numlower,
364 		.redirect = NULL,
365 	};
366 
367 	if (dentry->d_name.len > ofs->namelen)
368 		return ERR_PTR(-ENAMETOOLONG);
369 
370 	old_cred = ovl_override_creds(dentry->d_sb);
371 	upperdir = ovl_upperdentry_dereference(poe);
372 	if (upperdir) {
373 		err = ovl_lookup_layer(upperdir, &d, &upperdentry);
374 		if (err)
375 			goto out;
376 
377 		if (upperdentry && unlikely(ovl_dentry_remote(upperdentry))) {
378 			dput(upperdentry);
379 			err = -EREMOTE;
380 			goto out;
381 		}
382 		if (upperdentry && !d.is_dir) {
383 			BUG_ON(!d.stop || d.redirect);
384 			err = ovl_check_origin(dentry, upperdentry,
385 					       &stack, &ctr);
386 			if (err)
387 				goto out;
388 		}
389 
390 		if (d.redirect) {
391 			upperredirect = kstrdup(d.redirect, GFP_KERNEL);
392 			if (!upperredirect)
393 				goto out_put_upper;
394 			if (d.redirect[0] == '/')
395 				poe = roe;
396 		}
397 		upperopaque = d.opaque;
398 	}
399 
400 	if (!d.stop && poe->numlower) {
401 		err = -ENOMEM;
402 		stack = kcalloc(ofs->numlower, sizeof(struct path),
403 				GFP_TEMPORARY);
404 		if (!stack)
405 			goto out_put_upper;
406 	}
407 
408 	for (i = 0; !d.stop && i < poe->numlower; i++) {
409 		struct path lowerpath = poe->lowerstack[i];
410 
411 		d.last = i == poe->numlower - 1;
412 		err = ovl_lookup_layer(lowerpath.dentry, &d, &this);
413 		if (err)
414 			goto out_put;
415 
416 		if (!this)
417 			continue;
418 
419 		stack[ctr].dentry = this;
420 		stack[ctr].mnt = lowerpath.mnt;
421 		ctr++;
422 
423 		if (d.stop)
424 			break;
425 
426 		if (d.redirect && d.redirect[0] == '/' && poe != roe) {
427 			poe = roe;
428 
429 			/* Find the current layer on the root dentry */
430 			for (i = 0; i < poe->numlower; i++)
431 				if (poe->lowerstack[i].mnt == lowerpath.mnt)
432 					break;
433 			if (WARN_ON(i == poe->numlower))
434 				break;
435 		}
436 	}
437 
438 	oe = ovl_alloc_entry(ctr);
439 	err = -ENOMEM;
440 	if (!oe)
441 		goto out_put;
442 
443 	if (upperdentry || ctr) {
444 		struct dentry *realdentry;
445 		struct inode *realinode;
446 
447 		realdentry = upperdentry ? upperdentry : stack[0].dentry;
448 		realinode = d_inode(realdentry);
449 
450 		err = -ENOMEM;
451 		if (upperdentry && !d_is_dir(upperdentry)) {
452 			inode = ovl_get_inode(dentry->d_sb, realinode);
453 		} else {
454 			inode = ovl_new_inode(dentry->d_sb, realinode->i_mode,
455 					      realinode->i_rdev);
456 			if (inode)
457 				ovl_inode_init(inode, realinode, !!upperdentry);
458 		}
459 		if (!inode)
460 			goto out_free_oe;
461 		ovl_copyattr(realdentry->d_inode, inode);
462 	}
463 
464 	revert_creds(old_cred);
465 	oe->opaque = upperopaque;
466 	oe->redirect = upperredirect;
467 	oe->__upperdentry = upperdentry;
468 	memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
469 	kfree(stack);
470 	kfree(d.redirect);
471 	dentry->d_fsdata = oe;
472 	d_add(dentry, inode);
473 
474 	return NULL;
475 
476 out_free_oe:
477 	kfree(oe);
478 out_put:
479 	for (i = 0; i < ctr; i++)
480 		dput(stack[i].dentry);
481 	kfree(stack);
482 out_put_upper:
483 	dput(upperdentry);
484 	kfree(upperredirect);
485 out:
486 	kfree(d.redirect);
487 	revert_creds(old_cred);
488 	return ERR_PTR(err);
489 }
490 
491 bool ovl_lower_positive(struct dentry *dentry)
492 {
493 	struct ovl_entry *oe = dentry->d_fsdata;
494 	struct ovl_entry *poe = dentry->d_parent->d_fsdata;
495 	const struct qstr *name = &dentry->d_name;
496 	unsigned int i;
497 	bool positive = false;
498 	bool done = false;
499 
500 	/*
501 	 * If dentry is negative, then lower is positive iff this is a
502 	 * whiteout.
503 	 */
504 	if (!dentry->d_inode)
505 		return oe->opaque;
506 
507 	/* Negative upper -> positive lower */
508 	if (!oe->__upperdentry)
509 		return true;
510 
511 	/* Positive upper -> have to look up lower to see whether it exists */
512 	for (i = 0; !done && !positive && i < poe->numlower; i++) {
513 		struct dentry *this;
514 		struct dentry *lowerdir = poe->lowerstack[i].dentry;
515 
516 		this = lookup_one_len_unlocked(name->name, lowerdir,
517 					       name->len);
518 		if (IS_ERR(this)) {
519 			switch (PTR_ERR(this)) {
520 			case -ENOENT:
521 			case -ENAMETOOLONG:
522 				break;
523 
524 			default:
525 				/*
526 				 * Assume something is there, we just couldn't
527 				 * access it.
528 				 */
529 				positive = true;
530 				break;
531 			}
532 		} else {
533 			if (this->d_inode) {
534 				positive = !ovl_is_whiteout(this);
535 				done = true;
536 			}
537 			dput(this);
538 		}
539 	}
540 
541 	return positive;
542 }
543