xref: /linux/fs/fhandle.c (revision b5ca88927e353185b3d9ac4362d33e5aeb25771f)
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 
do_sys_name_to_handle(const struct path * path,struct file_handle __user * ufh,void __user * mnt_id,bool unique_mntid,int fh_flags)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  */
SYSCALL_DEFINE5(name_to_handle_at,int,dfd,const char __user *,name,struct file_handle __user *,handle,void __user *,mnt_id,int,flag)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 
get_path_anchor(int fd,struct path * root)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 		get_fs_pwd(current->fs, root);
184 		return 0;
185 	}
186 
187 	if (fd == FD_PIDFS_ROOT) {
188 		pidfs_get_root(root);
189 		return 0;
190 	}
191 
192 	return -EBADF;
193 }
194 
vfs_dentry_acceptable(void * context,struct dentry * dentry)195 static int vfs_dentry_acceptable(void *context, struct dentry *dentry)
196 {
197 	struct handle_to_path_ctx *ctx = context;
198 	struct user_namespace *user_ns = current_user_ns();
199 	struct dentry *d, *root = ctx->root.dentry;
200 	struct mnt_idmap *idmap = mnt_idmap(ctx->root.mnt);
201 	int retval = 0;
202 
203 	if (!root)
204 		return 1;
205 
206 	/* Old permission model with global CAP_DAC_READ_SEARCH. */
207 	if (!ctx->flags)
208 		return 1;
209 
210 	/*
211 	 * It's racy as we're not taking rename_lock but we're able to ignore
212 	 * permissions and we just need an approximation whether we were able
213 	 * to follow a path to the file.
214 	 *
215 	 * It's also potentially expensive on some filesystems especially if
216 	 * there is a deep path.
217 	 */
218 	d = dget(dentry);
219 	while (d != root && !IS_ROOT(d)) {
220 		struct dentry *parent = dget_parent(d);
221 
222 		/*
223 		 * We know that we have the ability to override DAC permissions
224 		 * as we've verified this earlier via CAP_DAC_READ_SEARCH. But
225 		 * we also need to make sure that there aren't any unmapped
226 		 * inodes in the path that would prevent us from reaching the
227 		 * file.
228 		 */
229 		if (!privileged_wrt_inode_uidgid(user_ns, idmap,
230 						 d_inode(parent))) {
231 			dput(d);
232 			dput(parent);
233 			return retval;
234 		}
235 
236 		dput(d);
237 		d = parent;
238 	}
239 
240 	if (!(ctx->flags & HANDLE_CHECK_SUBTREE) || d == root)
241 		retval = 1;
242 	/*
243 	 * exportfs_decode_fh_raw() does not call acceptable() callback with
244 	 * a disconnected directory dentry, so we should have reached either
245 	 * mount fd directory or sb root.
246 	 */
247 	if (ctx->fh_flags & EXPORT_FH_DIR_ONLY)
248 		WARN_ON_ONCE(d != root && d != root->d_sb->s_root);
249 	dput(d);
250 	return retval;
251 }
252 
do_handle_to_path(struct file_handle * handle,struct path * path,struct handle_to_path_ctx * ctx)253 static int do_handle_to_path(struct file_handle *handle, struct path *path,
254 			     struct handle_to_path_ctx *ctx)
255 {
256 	int handle_dwords;
257 	struct vfsmount *mnt = ctx->root.mnt;
258 	struct dentry *dentry;
259 
260 	/* change the handle size to multiple of sizeof(u32) */
261 	handle_dwords = handle->handle_bytes >> 2;
262 	dentry = exportfs_decode_fh_raw(mnt, (struct fid *)handle->f_handle,
263 					handle_dwords, handle->handle_type,
264 					ctx->fh_flags, vfs_dentry_acceptable,
265 					ctx);
266 	if (IS_ERR_OR_NULL(dentry)) {
267 		if (dentry == ERR_PTR(-ENOMEM))
268 			return -ENOMEM;
269 		return -ESTALE;
270 	}
271 	path->dentry = dentry;
272 	path->mnt = mntget(mnt);
273 	return 0;
274 }
275 
may_decode_fh(struct handle_to_path_ctx * ctx,unsigned int o_flags)276 static inline int may_decode_fh(struct handle_to_path_ctx *ctx,
277 				unsigned int o_flags)
278 {
279 	struct path *root = &ctx->root;
280 
281 	if (capable(CAP_DAC_READ_SEARCH))
282 		return 0;
283 
284 	/*
285 	 * Allow relaxed permissions of file handles if the caller has
286 	 * the ability to mount the filesystem or create a bind-mount of
287 	 * the provided @mountdirfd.
288 	 *
289 	 * In both cases the caller may be able to get an unobstructed
290 	 * way to the encoded file handle. If the caller is only able to
291 	 * create a bind-mount we need to verify that there are no
292 	 * locked mounts on top of it that could prevent us from getting
293 	 * to the encoded file.
294 	 *
295 	 * In principle, locked mounts can prevent the caller from
296 	 * mounting the filesystem but that only applies to procfs and
297 	 * sysfs neither of which support decoding file handles.
298 	 *
299 	 * Restrict to O_DIRECTORY to provide a deterministic API that
300 	 * avoids a confusing api in the face of disconnected non-dir
301 	 * dentries.
302 	 *
303 	 * There's only one dentry for each directory inode (VFS rule)...
304 	 */
305 	if (!(o_flags & O_DIRECTORY))
306 		return -EPERM;
307 
308 	if (ns_capable(root->mnt->mnt_sb->s_user_ns, CAP_SYS_ADMIN))
309 		ctx->flags = HANDLE_CHECK_PERMS;
310 	else if (is_mounted(root->mnt) &&
311 		 ns_capable(real_mount(root->mnt)->mnt_ns->user_ns,
312 			    CAP_SYS_ADMIN) &&
313 		 !has_locked_children(real_mount(root->mnt), root->dentry))
314 		ctx->flags = HANDLE_CHECK_PERMS | HANDLE_CHECK_SUBTREE;
315 	else
316 		return -EPERM;
317 
318 	/* Are we able to override DAC permissions? */
319 	if (!ns_capable(current_user_ns(), CAP_DAC_READ_SEARCH))
320 		return -EPERM;
321 
322 	ctx->fh_flags = EXPORT_FH_DIR_ONLY;
323 	return 0;
324 }
325 
handle_to_path(int mountdirfd,struct file_handle __user * ufh,struct path * path,unsigned int o_flags)326 static int handle_to_path(int mountdirfd, struct file_handle __user *ufh,
327 		   struct path *path, unsigned int o_flags)
328 {
329 	int retval = 0;
330 	struct file_handle f_handle;
331 	struct file_handle *handle __free(kfree) = NULL;
332 	struct handle_to_path_ctx ctx = {};
333 	const struct export_operations *eops;
334 
335 	if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle)))
336 		return -EFAULT;
337 
338 	if ((f_handle.handle_bytes > MAX_HANDLE_SZ) ||
339 	    (f_handle.handle_bytes == 0))
340 		return -EINVAL;
341 
342 	if (f_handle.handle_type < 0 ||
343 	    FILEID_USER_FLAGS(f_handle.handle_type) & ~FILEID_VALID_USER_FLAGS)
344 		return -EINVAL;
345 
346 	retval = get_path_anchor(mountdirfd, &ctx.root);
347 	if (retval)
348 		return retval;
349 
350 	eops = ctx.root.mnt->mnt_sb->s_export_op;
351 	if (eops && eops->permission)
352 		retval = eops->permission(&ctx, o_flags);
353 	else
354 		retval = may_decode_fh(&ctx, o_flags);
355 	if (retval)
356 		goto out_path;
357 
358 	handle = kmalloc(struct_size(handle, f_handle, f_handle.handle_bytes),
359 			 GFP_KERNEL);
360 	if (!handle) {
361 		retval = -ENOMEM;
362 		goto out_path;
363 	}
364 	/* copy the full handle */
365 	*handle = f_handle;
366 	if (copy_from_user(&handle->f_handle,
367 			   &ufh->f_handle,
368 			   f_handle.handle_bytes)) {
369 		retval = -EFAULT;
370 		goto out_path;
371 	}
372 
373 	/*
374 	 * If handle was encoded with AT_HANDLE_CONNECTABLE, verify that we
375 	 * are decoding an fd with connected path, which is accessible from
376 	 * the mount fd path.
377 	 */
378 	if (f_handle.handle_type & FILEID_IS_CONNECTABLE) {
379 		ctx.fh_flags |= EXPORT_FH_CONNECTABLE;
380 		ctx.flags |= HANDLE_CHECK_SUBTREE;
381 	}
382 	if (f_handle.handle_type & FILEID_IS_DIR)
383 		ctx.fh_flags |= EXPORT_FH_DIR_ONLY;
384 	/* Filesystem code should not be exposed to user flags */
385 	handle->handle_type &= ~FILEID_USER_FLAGS_MASK;
386 	retval = do_handle_to_path(handle, path, &ctx);
387 
388 out_path:
389 	path_put(&ctx.root);
390 	return retval;
391 }
392 
do_handle_open(int mountdirfd,struct file_handle __user * ufh,int open_flag)393 static long do_handle_open(int mountdirfd, struct file_handle __user *ufh,
394 			   int open_flag)
395 {
396 	long retval = 0;
397 	struct path path __free(path_put) = {};
398 	struct file *file;
399 	const struct export_operations *eops;
400 
401 	retval = handle_to_path(mountdirfd, ufh, &path, open_flag);
402 	if (retval)
403 		return retval;
404 
405 	CLASS(get_unused_fd, fd)(open_flag);
406 	if (fd < 0)
407 		return fd;
408 
409 	eops = path.mnt->mnt_sb->s_export_op;
410 	if (eops->open)
411 		file = eops->open(&path, open_flag);
412 	else
413 		file = file_open_root(&path, "", open_flag, 0);
414 	if (IS_ERR(file))
415 		return PTR_ERR(file);
416 
417 	fd_install(fd, file);
418 	return take_fd(fd);
419 }
420 
421 /**
422  * sys_open_by_handle_at: Open the file handle
423  * @mountdirfd: directory file descriptor
424  * @handle: file handle to be opened
425  * @flags: open flags.
426  *
427  * @mountdirfd indicate the directory file descriptor
428  * of the mount point. file handle is decoded relative
429  * to the vfsmount pointed by the @mountdirfd. @flags
430  * value is same as the open(2) flags.
431  */
SYSCALL_DEFINE3(open_by_handle_at,int,mountdirfd,struct file_handle __user *,handle,int,flags)432 SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
433 		struct file_handle __user *, handle,
434 		int, flags)
435 {
436 	long ret;
437 
438 	if (force_o_largefile())
439 		flags |= O_LARGEFILE;
440 
441 	ret = do_handle_open(mountdirfd, handle, flags);
442 	return ret;
443 }
444 
445 #ifdef CONFIG_COMPAT
446 /*
447  * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
448  * doesn't set the O_LARGEFILE flag.
449  */
COMPAT_SYSCALL_DEFINE3(open_by_handle_at,int,mountdirfd,struct file_handle __user *,handle,int,flags)450 COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
451 			     struct file_handle __user *, handle, int, flags)
452 {
453 	return do_handle_open(mountdirfd, handle, flags);
454 }
455 #endif
456