xref: /linux/fs/fhandle.c (revision 8bc7c5e525584903ea83332e18a2118ed3b1985e)
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 				  int __user *mnt_id, int fh_flags)
20 {
21 	long retval;
22 	struct file_handle f_handle;
23 	int handle_dwords, handle_bytes;
24 	struct file_handle *handle = NULL;
25 
26 	/*
27 	 * We need to make sure whether the file system support decoding of
28 	 * the file handle if decodeable file handle was requested.
29 	 */
30 	if (!exportfs_can_encode_fh(path->dentry->d_sb->s_export_op, fh_flags))
31 		return -EOPNOTSUPP;
32 
33 	if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle)))
34 		return -EFAULT;
35 
36 	if (f_handle.handle_bytes > MAX_HANDLE_SZ)
37 		return -EINVAL;
38 
39 	handle = kzalloc(struct_size(handle, f_handle, f_handle.handle_bytes),
40 			 GFP_KERNEL);
41 	if (!handle)
42 		return -ENOMEM;
43 
44 	/* convert handle size to multiple of sizeof(u32) */
45 	handle_dwords = f_handle.handle_bytes >> 2;
46 
47 	/* we ask for a non connectable maybe decodeable file handle */
48 	retval = exportfs_encode_fh(path->dentry,
49 				    (struct fid *)handle->f_handle,
50 				    &handle_dwords, fh_flags);
51 	handle->handle_type = retval;
52 	/* convert handle size to bytes */
53 	handle_bytes = handle_dwords * sizeof(u32);
54 	handle->handle_bytes = handle_bytes;
55 	if ((handle->handle_bytes > f_handle.handle_bytes) ||
56 	    (retval == FILEID_INVALID) || (retval < 0)) {
57 		/* As per old exportfs_encode_fh documentation
58 		 * we could return ENOSPC to indicate overflow
59 		 * But file system returned 255 always. So handle
60 		 * both the values
61 		 */
62 		if (retval == FILEID_INVALID || retval == -ENOSPC)
63 			retval = -EOVERFLOW;
64 		/*
65 		 * set the handle size to zero so we copy only
66 		 * non variable part of the file_handle
67 		 */
68 		handle_bytes = 0;
69 	} else
70 		retval = 0;
71 	/* copy the mount id */
72 	if (put_user(real_mount(path->mnt)->mnt_id, mnt_id) ||
73 	    copy_to_user(ufh, handle,
74 			 struct_size(handle, f_handle, handle_bytes)))
75 		retval = -EFAULT;
76 	kfree(handle);
77 	return retval;
78 }
79 
80 /**
81  * sys_name_to_handle_at: convert name to handle
82  * @dfd: directory relative to which name is interpreted if not absolute
83  * @name: name that should be converted to handle.
84  * @handle: resulting file handle
85  * @mnt_id: mount id of the file system containing the file
86  * @flag: flag value to indicate whether to follow symlink or not
87  *        and whether a decodable file handle is required.
88  *
89  * @handle->handle_size indicate the space available to store the
90  * variable part of the file handle in bytes. If there is not
91  * enough space, the field is updated to return the minimum
92  * value required.
93  */
94 SYSCALL_DEFINE5(name_to_handle_at, int, dfd, const char __user *, name,
95 		struct file_handle __user *, handle, int __user *, mnt_id,
96 		int, flag)
97 {
98 	struct path path;
99 	int lookup_flags;
100 	int fh_flags;
101 	int err;
102 
103 	if (flag & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH | AT_HANDLE_FID))
104 		return -EINVAL;
105 
106 	lookup_flags = (flag & AT_SYMLINK_FOLLOW) ? LOOKUP_FOLLOW : 0;
107 	fh_flags = (flag & AT_HANDLE_FID) ? EXPORT_FH_FID : 0;
108 	if (flag & AT_EMPTY_PATH)
109 		lookup_flags |= LOOKUP_EMPTY;
110 	err = user_path_at(dfd, name, lookup_flags, &path);
111 	if (!err) {
112 		err = do_sys_name_to_handle(&path, handle, mnt_id, fh_flags);
113 		path_put(&path);
114 	}
115 	return err;
116 }
117 
118 static int get_path_from_fd(int fd, struct path *root)
119 {
120 	if (fd == AT_FDCWD) {
121 		struct fs_struct *fs = current->fs;
122 		spin_lock(&fs->lock);
123 		*root = fs->pwd;
124 		path_get(root);
125 		spin_unlock(&fs->lock);
126 	} else {
127 		struct fd f = fdget(fd);
128 		if (!f.file)
129 			return -EBADF;
130 		*root = f.file->f_path;
131 		path_get(root);
132 		fdput(f);
133 	}
134 
135 	return 0;
136 }
137 
138 enum handle_to_path_flags {
139 	HANDLE_CHECK_PERMS   = (1 << 0),
140 	HANDLE_CHECK_SUBTREE = (1 << 1),
141 };
142 
143 struct handle_to_path_ctx {
144 	struct path root;
145 	enum handle_to_path_flags flags;
146 	unsigned int fh_flags;
147 };
148 
149 static int vfs_dentry_acceptable(void *context, struct dentry *dentry)
150 {
151 	struct handle_to_path_ctx *ctx = context;
152 	struct user_namespace *user_ns = current_user_ns();
153 	struct dentry *d, *root = ctx->root.dentry;
154 	struct mnt_idmap *idmap = mnt_idmap(ctx->root.mnt);
155 	int retval = 0;
156 
157 	if (!root)
158 		return 1;
159 
160 	/* Old permission model with global CAP_DAC_READ_SEARCH. */
161 	if (!ctx->flags)
162 		return 1;
163 
164 	/*
165 	 * It's racy as we're not taking rename_lock but we're able to ignore
166 	 * permissions and we just need an approximation whether we were able
167 	 * to follow a path to the file.
168 	 *
169 	 * It's also potentially expensive on some filesystems especially if
170 	 * there is a deep path.
171 	 */
172 	d = dget(dentry);
173 	while (d != root && !IS_ROOT(d)) {
174 		struct dentry *parent = dget_parent(d);
175 
176 		/*
177 		 * We know that we have the ability to override DAC permissions
178 		 * as we've verified this earlier via CAP_DAC_READ_SEARCH. But
179 		 * we also need to make sure that there aren't any unmapped
180 		 * inodes in the path that would prevent us from reaching the
181 		 * file.
182 		 */
183 		if (!privileged_wrt_inode_uidgid(user_ns, idmap,
184 						 d_inode(parent))) {
185 			dput(d);
186 			dput(parent);
187 			return retval;
188 		}
189 
190 		dput(d);
191 		d = parent;
192 	}
193 
194 	if (!(ctx->flags & HANDLE_CHECK_SUBTREE) || d == root)
195 		retval = 1;
196 	WARN_ON_ONCE(d != root && d != root->d_sb->s_root);
197 	dput(d);
198 	return retval;
199 }
200 
201 static int do_handle_to_path(struct file_handle *handle, struct path *path,
202 			     struct handle_to_path_ctx *ctx)
203 {
204 	int handle_dwords;
205 	struct vfsmount *mnt = ctx->root.mnt;
206 
207 	/* change the handle size to multiple of sizeof(u32) */
208 	handle_dwords = handle->handle_bytes >> 2;
209 	path->dentry = exportfs_decode_fh_raw(mnt,
210 					  (struct fid *)handle->f_handle,
211 					  handle_dwords, handle->handle_type,
212 					  ctx->fh_flags,
213 					  vfs_dentry_acceptable, ctx);
214 	if (IS_ERR_OR_NULL(path->dentry)) {
215 		if (path->dentry == ERR_PTR(-ENOMEM))
216 			return -ENOMEM;
217 		return -ESTALE;
218 	}
219 	path->mnt = mntget(mnt);
220 	return 0;
221 }
222 
223 /*
224  * Allow relaxed permissions of file handles if the caller has the
225  * ability to mount the filesystem or create a bind-mount of the
226  * provided @mountdirfd.
227  *
228  * In both cases the caller may be able to get an unobstructed way to
229  * the encoded file handle. If the caller is only able to create a
230  * bind-mount we need to verify that there are no locked mounts on top
231  * of it that could prevent us from getting to the encoded file.
232  *
233  * In principle, locked mounts can prevent the caller from mounting the
234  * filesystem but that only applies to procfs and sysfs neither of which
235  * support decoding file handles.
236  */
237 static inline bool may_decode_fh(struct handle_to_path_ctx *ctx,
238 				 unsigned int o_flags)
239 {
240 	struct path *root = &ctx->root;
241 
242 	/*
243 	 * Restrict to O_DIRECTORY to provide a deterministic API that avoids a
244 	 * confusing api in the face of disconnected non-dir dentries.
245 	 *
246 	 * There's only one dentry for each directory inode (VFS rule)...
247 	 */
248 	if (!(o_flags & O_DIRECTORY))
249 		return false;
250 
251 	if (ns_capable(root->mnt->mnt_sb->s_user_ns, CAP_SYS_ADMIN))
252 		ctx->flags = HANDLE_CHECK_PERMS;
253 	else if (is_mounted(root->mnt) &&
254 		 ns_capable(real_mount(root->mnt)->mnt_ns->user_ns,
255 			    CAP_SYS_ADMIN) &&
256 		 !has_locked_children(real_mount(root->mnt), root->dentry))
257 		ctx->flags = HANDLE_CHECK_PERMS | HANDLE_CHECK_SUBTREE;
258 	else
259 		return false;
260 
261 	/* Are we able to override DAC permissions? */
262 	if (!ns_capable(current_user_ns(), CAP_DAC_READ_SEARCH))
263 		return false;
264 
265 	ctx->fh_flags = EXPORT_FH_DIR_ONLY;
266 	return true;
267 }
268 
269 static int handle_to_path(int mountdirfd, struct file_handle __user *ufh,
270 		   struct path *path, unsigned int o_flags)
271 {
272 	int retval = 0;
273 	struct file_handle f_handle;
274 	struct file_handle *handle = NULL;
275 	struct handle_to_path_ctx ctx = {};
276 
277 	retval = get_path_from_fd(mountdirfd, &ctx.root);
278 	if (retval)
279 		goto out_err;
280 
281 	if (!capable(CAP_DAC_READ_SEARCH) && !may_decode_fh(&ctx, o_flags)) {
282 		retval = -EPERM;
283 		goto out_path;
284 	}
285 
286 	if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle))) {
287 		retval = -EFAULT;
288 		goto out_path;
289 	}
290 	if ((f_handle.handle_bytes > MAX_HANDLE_SZ) ||
291 	    (f_handle.handle_bytes == 0)) {
292 		retval = -EINVAL;
293 		goto out_path;
294 	}
295 	handle = kmalloc(struct_size(handle, f_handle, f_handle.handle_bytes),
296 			 GFP_KERNEL);
297 	if (!handle) {
298 		retval = -ENOMEM;
299 		goto out_path;
300 	}
301 	/* copy the full handle */
302 	*handle = f_handle;
303 	if (copy_from_user(&handle->f_handle,
304 			   &ufh->f_handle,
305 			   f_handle.handle_bytes)) {
306 		retval = -EFAULT;
307 		goto out_handle;
308 	}
309 
310 	retval = do_handle_to_path(handle, path, &ctx);
311 
312 out_handle:
313 	kfree(handle);
314 out_path:
315 	path_put(&ctx.root);
316 out_err:
317 	return retval;
318 }
319 
320 static long do_handle_open(int mountdirfd, struct file_handle __user *ufh,
321 			   int open_flag)
322 {
323 	long retval = 0;
324 	struct path path;
325 	struct file *file;
326 	int fd;
327 
328 	retval = handle_to_path(mountdirfd, ufh, &path, open_flag);
329 	if (retval)
330 		return retval;
331 
332 	fd = get_unused_fd_flags(open_flag);
333 	if (fd < 0) {
334 		path_put(&path);
335 		return fd;
336 	}
337 	file = file_open_root(&path, "", open_flag, 0);
338 	if (IS_ERR(file)) {
339 		put_unused_fd(fd);
340 		retval =  PTR_ERR(file);
341 	} else {
342 		retval = fd;
343 		fd_install(fd, file);
344 	}
345 	path_put(&path);
346 	return retval;
347 }
348 
349 /**
350  * sys_open_by_handle_at: Open the file handle
351  * @mountdirfd: directory file descriptor
352  * @handle: file handle to be opened
353  * @flags: open flags.
354  *
355  * @mountdirfd indicate the directory file descriptor
356  * of the mount point. file handle is decoded relative
357  * to the vfsmount pointed by the @mountdirfd. @flags
358  * value is same as the open(2) flags.
359  */
360 SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
361 		struct file_handle __user *, handle,
362 		int, flags)
363 {
364 	long ret;
365 
366 	if (force_o_largefile())
367 		flags |= O_LARGEFILE;
368 
369 	ret = do_handle_open(mountdirfd, handle, flags);
370 	return ret;
371 }
372 
373 #ifdef CONFIG_COMPAT
374 /*
375  * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
376  * doesn't set the O_LARGEFILE flag.
377  */
378 COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
379 			     struct file_handle __user *, handle, int, flags)
380 {
381 	return do_handle_open(mountdirfd, handle, flags);
382 }
383 #endif
384