xref: /linux/fs/fhandle.c (revision 67fcec2919e4ed31ab845eb456ad7d6f1e85505c)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/syscalls.h>
3 #include <linux/slab.h>
4 #include <linux/fs.h>
5 #include <linux/file.h>
6 #include <linux/mount.h>
7 #include <linux/namei.h>
8 #include <linux/exportfs.h>
9 #include <linux/fs_struct.h>
10 #include <linux/fsnotify.h>
11 #include <linux/personality.h>
12 #include <linux/uaccess.h>
13 #include <linux/compat.h>
14 #include "internal.h"
15 #include "mount.h"
16 
17 static long do_sys_name_to_handle(const struct path *path,
18 				  struct file_handle __user *ufh,
19 				  void __user *mnt_id, bool unique_mntid,
20 				  int fh_flags)
21 {
22 	long retval;
23 	struct file_handle f_handle;
24 	int handle_dwords, handle_bytes;
25 	struct file_handle *handle = NULL;
26 
27 	/*
28 	 * We need to make sure whether the file system support decoding of
29 	 * the file handle if decodeable file handle was requested.
30 	 */
31 	if (!exportfs_can_encode_fh(path->dentry->d_sb->s_export_op, fh_flags))
32 		return -EOPNOTSUPP;
33 
34 	/*
35 	 * A request to encode a connectable handle for a disconnected dentry
36 	 * is unexpected since AT_EMPTY_PATH is not allowed.
37 	 */
38 	if (fh_flags & EXPORT_FH_CONNECTABLE &&
39 	    WARN_ON(path->dentry->d_flags & DCACHE_DISCONNECTED))
40 		return -EINVAL;
41 
42 	if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle)))
43 		return -EFAULT;
44 
45 	if (f_handle.handle_bytes > MAX_HANDLE_SZ)
46 		return -EINVAL;
47 
48 	handle = kzalloc(struct_size(handle, f_handle, f_handle.handle_bytes),
49 			 GFP_KERNEL);
50 	if (!handle)
51 		return -ENOMEM;
52 
53 	/* convert handle size to multiple of sizeof(u32) */
54 	handle_dwords = f_handle.handle_bytes >> 2;
55 
56 	/* Encode a possibly decodeable/connectable file handle */
57 	retval = exportfs_encode_fh(path->dentry,
58 				    (struct fid *)handle->f_handle,
59 				    &handle_dwords, fh_flags);
60 	handle->handle_type = retval;
61 	/* convert handle size to bytes */
62 	handle_bytes = handle_dwords * sizeof(u32);
63 	handle->handle_bytes = handle_bytes;
64 	if ((handle->handle_bytes > f_handle.handle_bytes) ||
65 	    (retval == FILEID_INVALID) || (retval < 0)) {
66 		/* As per old exportfs_encode_fh documentation
67 		 * we could return ENOSPC to indicate overflow
68 		 * But file system returned 255 always. So handle
69 		 * both the values
70 		 */
71 		if (retval == FILEID_INVALID || retval == -ENOSPC)
72 			retval = -EOVERFLOW;
73 		/*
74 		 * set the handle size to zero so we copy only
75 		 * non variable part of the file_handle
76 		 */
77 		handle_bytes = 0;
78 	} else {
79 		/*
80 		 * When asked to encode a connectable file handle, encode this
81 		 * property in the file handle itself, so that we later know
82 		 * how to decode it.
83 		 * For sanity, also encode in the file handle if the encoded
84 		 * object is a directory and verify this during decode, because
85 		 * decoding directory file handles is quite different than
86 		 * decoding connectable non-directory file handles.
87 		 */
88 		if (fh_flags & EXPORT_FH_CONNECTABLE) {
89 			handle->handle_type |= FILEID_IS_CONNECTABLE;
90 			if (d_is_dir(path->dentry))
91 				handle->handle_type |= FILEID_IS_DIR;
92 		}
93 		retval = 0;
94 	}
95 	/* copy the mount id */
96 	if (unique_mntid) {
97 		if (put_user(real_mount(path->mnt)->mnt_id_unique,
98 			     (u64 __user *) mnt_id))
99 			retval = -EFAULT;
100 	} else {
101 		if (put_user(real_mount(path->mnt)->mnt_id,
102 			     (int __user *) mnt_id))
103 			retval = -EFAULT;
104 	}
105 	/* copy the handle */
106 	if (retval != -EFAULT &&
107 		copy_to_user(ufh, handle,
108 			     struct_size(handle, f_handle, handle_bytes)))
109 		retval = -EFAULT;
110 	kfree(handle);
111 	return retval;
112 }
113 
114 /**
115  * sys_name_to_handle_at: convert name to handle
116  * @dfd: directory relative to which name is interpreted if not absolute
117  * @name: name that should be converted to handle.
118  * @handle: resulting file handle
119  * @mnt_id: mount id of the file system containing the file
120  *          (u64 if AT_HANDLE_MNT_ID_UNIQUE, otherwise int)
121  * @flag: flag value to indicate whether to follow symlink or not
122  *        and whether a decodable file handle is required.
123  *
124  * @handle->handle_size indicate the space available to store the
125  * variable part of the file handle in bytes. If there is not
126  * enough space, the field is updated to return the minimum
127  * value required.
128  */
129 SYSCALL_DEFINE5(name_to_handle_at, int, dfd, const char __user *, name,
130 		struct file_handle __user *, handle, void __user *, mnt_id,
131 		int, flag)
132 {
133 	struct path path;
134 	int lookup_flags;
135 	int fh_flags = 0;
136 	int err;
137 
138 	if (flag & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH | AT_HANDLE_FID |
139 		     AT_HANDLE_MNT_ID_UNIQUE | AT_HANDLE_CONNECTABLE))
140 		return -EINVAL;
141 
142 	/*
143 	 * AT_HANDLE_FID means there is no intention to decode file handle
144 	 * AT_HANDLE_CONNECTABLE means there is an intention to decode a
145 	 * connected fd (with known path), so these flags are conflicting.
146 	 * AT_EMPTY_PATH could be used along with a dfd that refers to a
147 	 * disconnected non-directory, which cannot be used to encode a
148 	 * connectable file handle, because its parent is unknown.
149 	 */
150 	if (flag & AT_HANDLE_CONNECTABLE &&
151 	    flag & (AT_HANDLE_FID | AT_EMPTY_PATH))
152 		return -EINVAL;
153 	else if (flag & AT_HANDLE_FID)
154 		fh_flags |= EXPORT_FH_FID;
155 	else if (flag & AT_HANDLE_CONNECTABLE)
156 		fh_flags |= EXPORT_FH_CONNECTABLE;
157 
158 	lookup_flags = (flag & AT_SYMLINK_FOLLOW) ? LOOKUP_FOLLOW : 0;
159 	if (flag & AT_EMPTY_PATH)
160 		lookup_flags |= LOOKUP_EMPTY;
161 	err = user_path_at(dfd, name, lookup_flags, &path);
162 	if (!err) {
163 		err = do_sys_name_to_handle(&path, handle, mnt_id,
164 					    flag & AT_HANDLE_MNT_ID_UNIQUE,
165 					    fh_flags);
166 		path_put(&path);
167 	}
168 	return err;
169 }
170 
171 static int get_path_anchor(int fd, struct path *root)
172 {
173 	if (fd >= 0) {
174 		CLASS(fd, f)(fd);
175 		if (fd_empty(f))
176 			return -EBADF;
177 		*root = fd_file(f)->f_path;
178 		path_get(root);
179 		return 0;
180 	}
181 
182 	if (fd == AT_FDCWD) {
183 		struct fs_struct *fs = current->fs;
184 		spin_lock(&fs->lock);
185 		*root = fs->pwd;
186 		path_get(root);
187 		spin_unlock(&fs->lock);
188 		return 0;
189 	}
190 
191 	return -EBADF;
192 }
193 
194 static int vfs_dentry_acceptable(void *context, struct dentry *dentry)
195 {
196 	struct handle_to_path_ctx *ctx = context;
197 	struct user_namespace *user_ns = current_user_ns();
198 	struct dentry *d, *root = ctx->root.dentry;
199 	struct mnt_idmap *idmap = mnt_idmap(ctx->root.mnt);
200 	int retval = 0;
201 
202 	if (!root)
203 		return 1;
204 
205 	/* Old permission model with global CAP_DAC_READ_SEARCH. */
206 	if (!ctx->flags)
207 		return 1;
208 
209 	/*
210 	 * It's racy as we're not taking rename_lock but we're able to ignore
211 	 * permissions and we just need an approximation whether we were able
212 	 * to follow a path to the file.
213 	 *
214 	 * It's also potentially expensive on some filesystems especially if
215 	 * there is a deep path.
216 	 */
217 	d = dget(dentry);
218 	while (d != root && !IS_ROOT(d)) {
219 		struct dentry *parent = dget_parent(d);
220 
221 		/*
222 		 * We know that we have the ability to override DAC permissions
223 		 * as we've verified this earlier via CAP_DAC_READ_SEARCH. But
224 		 * we also need to make sure that there aren't any unmapped
225 		 * inodes in the path that would prevent us from reaching the
226 		 * file.
227 		 */
228 		if (!privileged_wrt_inode_uidgid(user_ns, idmap,
229 						 d_inode(parent))) {
230 			dput(d);
231 			dput(parent);
232 			return retval;
233 		}
234 
235 		dput(d);
236 		d = parent;
237 	}
238 
239 	if (!(ctx->flags & HANDLE_CHECK_SUBTREE) || d == root)
240 		retval = 1;
241 	/*
242 	 * exportfs_decode_fh_raw() does not call acceptable() callback with
243 	 * a disconnected directory dentry, so we should have reached either
244 	 * mount fd directory or sb root.
245 	 */
246 	if (ctx->fh_flags & EXPORT_FH_DIR_ONLY)
247 		WARN_ON_ONCE(d != root && d != root->d_sb->s_root);
248 	dput(d);
249 	return retval;
250 }
251 
252 static int do_handle_to_path(struct file_handle *handle, struct path *path,
253 			     struct handle_to_path_ctx *ctx)
254 {
255 	int handle_dwords;
256 	struct vfsmount *mnt = ctx->root.mnt;
257 	struct dentry *dentry;
258 
259 	/* change the handle size to multiple of sizeof(u32) */
260 	handle_dwords = handle->handle_bytes >> 2;
261 	dentry = exportfs_decode_fh_raw(mnt, (struct fid *)handle->f_handle,
262 					handle_dwords, handle->handle_type,
263 					ctx->fh_flags, vfs_dentry_acceptable,
264 					ctx);
265 	if (IS_ERR_OR_NULL(dentry)) {
266 		if (dentry == ERR_PTR(-ENOMEM))
267 			return -ENOMEM;
268 		return -ESTALE;
269 	}
270 	path->dentry = dentry;
271 	path->mnt = mntget(mnt);
272 	return 0;
273 }
274 
275 static inline int may_decode_fh(struct handle_to_path_ctx *ctx,
276 				unsigned int o_flags)
277 {
278 	struct path *root = &ctx->root;
279 
280 	if (capable(CAP_DAC_READ_SEARCH))
281 		return 0;
282 
283 	/*
284 	 * Allow relaxed permissions of file handles if the caller has
285 	 * the ability to mount the filesystem or create a bind-mount of
286 	 * the provided @mountdirfd.
287 	 *
288 	 * In both cases the caller may be able to get an unobstructed
289 	 * way to the encoded file handle. If the caller is only able to
290 	 * create a bind-mount we need to verify that there are no
291 	 * locked mounts on top of it that could prevent us from getting
292 	 * to the encoded file.
293 	 *
294 	 * In principle, locked mounts can prevent the caller from
295 	 * mounting the filesystem but that only applies to procfs and
296 	 * sysfs neither of which support decoding file handles.
297 	 *
298 	 * Restrict to O_DIRECTORY to provide a deterministic API that
299 	 * avoids a confusing api in the face of disconnected non-dir
300 	 * dentries.
301 	 *
302 	 * There's only one dentry for each directory inode (VFS rule)...
303 	 */
304 	if (!(o_flags & O_DIRECTORY))
305 		return -EPERM;
306 
307 	if (ns_capable(root->mnt->mnt_sb->s_user_ns, CAP_SYS_ADMIN))
308 		ctx->flags = HANDLE_CHECK_PERMS;
309 	else if (is_mounted(root->mnt) &&
310 		 ns_capable(real_mount(root->mnt)->mnt_ns->user_ns,
311 			    CAP_SYS_ADMIN) &&
312 		 !has_locked_children(real_mount(root->mnt), root->dentry))
313 		ctx->flags = HANDLE_CHECK_PERMS | HANDLE_CHECK_SUBTREE;
314 	else
315 		return -EPERM;
316 
317 	/* Are we able to override DAC permissions? */
318 	if (!ns_capable(current_user_ns(), CAP_DAC_READ_SEARCH))
319 		return -EPERM;
320 
321 	ctx->fh_flags = EXPORT_FH_DIR_ONLY;
322 	return 0;
323 }
324 
325 static int handle_to_path(int mountdirfd, struct file_handle __user *ufh,
326 		   struct path *path, unsigned int o_flags)
327 {
328 	int retval = 0;
329 	struct file_handle f_handle;
330 	struct file_handle *handle __free(kfree) = NULL;
331 	struct handle_to_path_ctx ctx = {};
332 	const struct export_operations *eops;
333 
334 	if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle)))
335 		return -EFAULT;
336 
337 	if ((f_handle.handle_bytes > MAX_HANDLE_SZ) ||
338 	    (f_handle.handle_bytes == 0))
339 		return -EINVAL;
340 
341 	if (f_handle.handle_type < 0 ||
342 	    FILEID_USER_FLAGS(f_handle.handle_type) & ~FILEID_VALID_USER_FLAGS)
343 		return -EINVAL;
344 
345 	retval = get_path_anchor(mountdirfd, &ctx.root);
346 	if (retval)
347 		return retval;
348 
349 	eops = ctx.root.mnt->mnt_sb->s_export_op;
350 	if (eops && eops->permission)
351 		retval = eops->permission(&ctx, o_flags);
352 	else
353 		retval = may_decode_fh(&ctx, o_flags);
354 	if (retval)
355 		goto out_path;
356 
357 	handle = kmalloc(struct_size(handle, f_handle, f_handle.handle_bytes),
358 			 GFP_KERNEL);
359 	if (!handle) {
360 		retval = -ENOMEM;
361 		goto out_path;
362 	}
363 	/* copy the full handle */
364 	*handle = f_handle;
365 	if (copy_from_user(&handle->f_handle,
366 			   &ufh->f_handle,
367 			   f_handle.handle_bytes)) {
368 		retval = -EFAULT;
369 		goto out_path;
370 	}
371 
372 	/*
373 	 * If handle was encoded with AT_HANDLE_CONNECTABLE, verify that we
374 	 * are decoding an fd with connected path, which is accessible from
375 	 * the mount fd path.
376 	 */
377 	if (f_handle.handle_type & FILEID_IS_CONNECTABLE) {
378 		ctx.fh_flags |= EXPORT_FH_CONNECTABLE;
379 		ctx.flags |= HANDLE_CHECK_SUBTREE;
380 	}
381 	if (f_handle.handle_type & FILEID_IS_DIR)
382 		ctx.fh_flags |= EXPORT_FH_DIR_ONLY;
383 	/* Filesystem code should not be exposed to user flags */
384 	handle->handle_type &= ~FILEID_USER_FLAGS_MASK;
385 	retval = do_handle_to_path(handle, path, &ctx);
386 
387 out_path:
388 	path_put(&ctx.root);
389 	return retval;
390 }
391 
392 static long do_handle_open(int mountdirfd, struct file_handle __user *ufh,
393 			   int open_flag)
394 {
395 	long retval = 0;
396 	struct path path __free(path_put) = {};
397 	struct file *file;
398 	const struct export_operations *eops;
399 
400 	retval = handle_to_path(mountdirfd, ufh, &path, open_flag);
401 	if (retval)
402 		return retval;
403 
404 	CLASS(get_unused_fd, fd)(O_CLOEXEC);
405 	if (fd < 0)
406 		return fd;
407 
408 	eops = path.mnt->mnt_sb->s_export_op;
409 	if (eops->open)
410 		file = eops->open(&path, open_flag);
411 	else
412 		file = file_open_root(&path, "", open_flag, 0);
413 	if (IS_ERR(file))
414 		return PTR_ERR(file);
415 
416 	fd_install(fd, file);
417 	return take_fd(fd);
418 }
419 
420 /**
421  * sys_open_by_handle_at: Open the file handle
422  * @mountdirfd: directory file descriptor
423  * @handle: file handle to be opened
424  * @flags: open flags.
425  *
426  * @mountdirfd indicate the directory file descriptor
427  * of the mount point. file handle is decoded relative
428  * to the vfsmount pointed by the @mountdirfd. @flags
429  * value is same as the open(2) flags.
430  */
431 SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
432 		struct file_handle __user *, handle,
433 		int, flags)
434 {
435 	long ret;
436 
437 	if (force_o_largefile())
438 		flags |= O_LARGEFILE;
439 
440 	ret = do_handle_open(mountdirfd, handle, flags);
441 	return ret;
442 }
443 
444 #ifdef CONFIG_COMPAT
445 /*
446  * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
447  * doesn't set the O_LARGEFILE flag.
448  */
449 COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
450 			     struct file_handle __user *, handle, int, flags)
451 {
452 	return do_handle_open(mountdirfd, handle, flags);
453 }
454 #endif
455