xref: /openbmc/linux/fs/d_path.c (revision dd1431e5)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #include <linux/syscalls.h>
3 #include <linux/export.h>
4 #include <linux/uaccess.h>
5 #include <linux/fs_struct.h>
6 #include <linux/fs.h>
7 #include <linux/slab.h>
8 #include <linux/prefetch.h>
9 #include "mount.h"
10 
11 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
12 {
13 	*buflen -= namelen;
14 	if (*buflen < 0)
15 		return -ENAMETOOLONG;
16 	*buffer -= namelen;
17 	memcpy(*buffer, str, namelen);
18 	return 0;
19 }
20 
21 /**
22  * prepend_name - prepend a pathname in front of current buffer pointer
23  * @buffer: buffer pointer
24  * @buflen: allocated length of the buffer
25  * @name:   name string and length qstr structure
26  *
27  * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
28  * make sure that either the old or the new name pointer and length are
29  * fetched. However, there may be mismatch between length and pointer.
30  * The length cannot be trusted, we need to copy it byte-by-byte until
31  * the length is reached or a null byte is found. It also prepends "/" at
32  * the beginning of the name. The sequence number check at the caller will
33  * retry it again when a d_move() does happen. So any garbage in the buffer
34  * due to mismatched pointer and length will be discarded.
35  *
36  * Load acquire is needed to make sure that we see that terminating NUL.
37  */
38 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
39 {
40 	const char *dname = smp_load_acquire(&name->name); /* ^^^ */
41 	u32 dlen = READ_ONCE(name->len);
42 	char *p;
43 
44 	*buflen -= dlen + 1;
45 	if (*buflen < 0)
46 		return -ENAMETOOLONG;
47 	p = *buffer -= dlen + 1;
48 	*p++ = '/';
49 	while (dlen--) {
50 		char c = *dname++;
51 		if (!c)
52 			break;
53 		*p++ = c;
54 	}
55 	return 0;
56 }
57 
58 /**
59  * prepend_path - Prepend path string to a buffer
60  * @path: the dentry/vfsmount to report
61  * @root: root vfsmnt/dentry
62  * @buffer: pointer to the end of the buffer
63  * @buflen: pointer to buffer length
64  *
65  * The function will first try to write out the pathname without taking any
66  * lock other than the RCU read lock to make sure that dentries won't go away.
67  * It only checks the sequence number of the global rename_lock as any change
68  * in the dentry's d_seq will be preceded by changes in the rename_lock
69  * sequence number. If the sequence number had been changed, it will restart
70  * the whole pathname back-tracing sequence again by taking the rename_lock.
71  * In this case, there is no need to take the RCU read lock as the recursive
72  * parent pointer references will keep the dentry chain alive as long as no
73  * rename operation is performed.
74  */
75 static int prepend_path(const struct path *path,
76 			const struct path *root,
77 			char **buffer, int *buflen)
78 {
79 	struct dentry *dentry;
80 	struct vfsmount *vfsmnt;
81 	struct mount *mnt;
82 	int error = 0;
83 	unsigned seq, m_seq = 0;
84 	char *bptr;
85 	int blen;
86 
87 	rcu_read_lock();
88 restart_mnt:
89 	read_seqbegin_or_lock(&mount_lock, &m_seq);
90 	seq = 0;
91 	rcu_read_lock();
92 restart:
93 	bptr = *buffer;
94 	blen = *buflen;
95 	error = 0;
96 	dentry = path->dentry;
97 	vfsmnt = path->mnt;
98 	mnt = real_mount(vfsmnt);
99 	read_seqbegin_or_lock(&rename_lock, &seq);
100 	while (dentry != root->dentry || vfsmnt != root->mnt) {
101 		struct dentry * parent;
102 
103 		if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
104 			struct mount *parent = READ_ONCE(mnt->mnt_parent);
105 			struct mnt_namespace *mnt_ns;
106 
107 			/* Escaped? */
108 			if (dentry != vfsmnt->mnt_root) {
109 				bptr = *buffer;
110 				blen = *buflen;
111 				error = 3;
112 				break;
113 			}
114 			/* Global root? */
115 			if (mnt != parent) {
116 				dentry = READ_ONCE(mnt->mnt_mountpoint);
117 				mnt = parent;
118 				vfsmnt = &mnt->mnt;
119 				continue;
120 			}
121 			mnt_ns = READ_ONCE(mnt->mnt_ns);
122 			/* open-coded is_mounted() to use local mnt_ns */
123 			if (!IS_ERR_OR_NULL(mnt_ns) && !is_anon_ns(mnt_ns))
124 				error = 1;	// absolute root
125 			else
126 				error = 2;	// detached or not attached yet
127 			break;
128 		}
129 		parent = dentry->d_parent;
130 		prefetch(parent);
131 		error = prepend_name(&bptr, &blen, &dentry->d_name);
132 		if (error)
133 			break;
134 
135 		dentry = parent;
136 	}
137 	if (!(seq & 1))
138 		rcu_read_unlock();
139 	if (need_seqretry(&rename_lock, seq)) {
140 		seq = 1;
141 		goto restart;
142 	}
143 	done_seqretry(&rename_lock, seq);
144 
145 	if (!(m_seq & 1))
146 		rcu_read_unlock();
147 	if (need_seqretry(&mount_lock, m_seq)) {
148 		m_seq = 1;
149 		goto restart_mnt;
150 	}
151 	done_seqretry(&mount_lock, m_seq);
152 
153 	if (error >= 0 && bptr == *buffer) {
154 		if (--blen < 0)
155 			error = -ENAMETOOLONG;
156 		else
157 			*--bptr = '/';
158 	}
159 	*buffer = bptr;
160 	*buflen = blen;
161 	return error;
162 }
163 
164 /**
165  * __d_path - return the path of a dentry
166  * @path: the dentry/vfsmount to report
167  * @root: root vfsmnt/dentry
168  * @buf: buffer to return value in
169  * @buflen: buffer length
170  *
171  * Convert a dentry into an ASCII path name.
172  *
173  * Returns a pointer into the buffer or an error code if the
174  * path was too long.
175  *
176  * "buflen" should be positive.
177  *
178  * If the path is not reachable from the supplied root, return %NULL.
179  */
180 char *__d_path(const struct path *path,
181 	       const struct path *root,
182 	       char *buf, int buflen)
183 {
184 	char *res = buf + buflen;
185 	int error;
186 
187 	prepend(&res, &buflen, "\0", 1);
188 	error = prepend_path(path, root, &res, &buflen);
189 
190 	if (error < 0)
191 		return ERR_PTR(error);
192 	if (error > 0)
193 		return NULL;
194 	return res;
195 }
196 
197 char *d_absolute_path(const struct path *path,
198 	       char *buf, int buflen)
199 {
200 	struct path root = {};
201 	char *res = buf + buflen;
202 	int error;
203 
204 	prepend(&res, &buflen, "\0", 1);
205 	error = prepend_path(path, &root, &res, &buflen);
206 
207 	if (error > 1)
208 		error = -EINVAL;
209 	if (error < 0)
210 		return ERR_PTR(error);
211 	return res;
212 }
213 
214 /*
215  * same as __d_path but appends "(deleted)" for unlinked files.
216  */
217 static int path_with_deleted(const struct path *path,
218 			     const struct path *root,
219 			     char **buf, int *buflen)
220 {
221 	prepend(buf, buflen, "\0", 1);
222 	if (d_unlinked(path->dentry)) {
223 		int error = prepend(buf, buflen, " (deleted)", 10);
224 		if (error)
225 			return error;
226 	}
227 
228 	return prepend_path(path, root, buf, buflen);
229 }
230 
231 static int prepend_unreachable(char **buffer, int *buflen)
232 {
233 	return prepend(buffer, buflen, "(unreachable)", 13);
234 }
235 
236 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
237 {
238 	unsigned seq;
239 
240 	do {
241 		seq = read_seqcount_begin(&fs->seq);
242 		*root = fs->root;
243 	} while (read_seqcount_retry(&fs->seq, seq));
244 }
245 
246 /**
247  * d_path - return the path of a dentry
248  * @path: path to report
249  * @buf: buffer to return value in
250  * @buflen: buffer length
251  *
252  * Convert a dentry into an ASCII path name. If the entry has been deleted
253  * the string " (deleted)" is appended. Note that this is ambiguous.
254  *
255  * Returns a pointer into the buffer or an error code if the path was
256  * too long. Note: Callers should use the returned pointer, not the passed
257  * in buffer, to use the name! The implementation often starts at an offset
258  * into the buffer, and may leave 0 bytes at the start.
259  *
260  * "buflen" should be positive.
261  */
262 char *d_path(const struct path *path, char *buf, int buflen)
263 {
264 	char *res = buf + buflen;
265 	struct path root;
266 	int error;
267 
268 	/*
269 	 * We have various synthetic filesystems that never get mounted.  On
270 	 * these filesystems dentries are never used for lookup purposes, and
271 	 * thus don't need to be hashed.  They also don't need a name until a
272 	 * user wants to identify the object in /proc/pid/fd/.  The little hack
273 	 * below allows us to generate a name for these objects on demand:
274 	 *
275 	 * Some pseudo inodes are mountable.  When they are mounted
276 	 * path->dentry == path->mnt->mnt_root.  In that case don't call d_dname
277 	 * and instead have d_path return the mounted path.
278 	 */
279 	if (path->dentry->d_op && path->dentry->d_op->d_dname &&
280 	    (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
281 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
282 
283 	rcu_read_lock();
284 	get_fs_root_rcu(current->fs, &root);
285 	error = path_with_deleted(path, &root, &res, &buflen);
286 	rcu_read_unlock();
287 
288 	if (error < 0)
289 		res = ERR_PTR(error);
290 	return res;
291 }
292 EXPORT_SYMBOL(d_path);
293 
294 /*
295  * Helper function for dentry_operations.d_dname() members
296  */
297 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
298 			const char *fmt, ...)
299 {
300 	va_list args;
301 	char temp[64];
302 	int sz;
303 
304 	va_start(args, fmt);
305 	sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
306 	va_end(args);
307 
308 	if (sz > sizeof(temp) || sz > buflen)
309 		return ERR_PTR(-ENAMETOOLONG);
310 
311 	buffer += buflen - sz;
312 	return memcpy(buffer, temp, sz);
313 }
314 
315 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
316 {
317 	char *end = buffer + buflen;
318 	/* these dentries are never renamed, so d_lock is not needed */
319 	if (prepend(&end, &buflen, " (deleted)", 11) ||
320 	    prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
321 	    prepend(&end, &buflen, "/", 1))
322 		end = ERR_PTR(-ENAMETOOLONG);
323 	return end;
324 }
325 
326 /*
327  * Write full pathname from the root of the filesystem into the buffer.
328  */
329 static char *__dentry_path(const struct dentry *d, char *buf, int buflen)
330 {
331 	const struct dentry *dentry;
332 	char *end, *retval;
333 	int len, seq = 0;
334 	int error = 0;
335 
336 	if (buflen < 2)
337 		goto Elong;
338 
339 	rcu_read_lock();
340 restart:
341 	dentry = d;
342 	end = buf + buflen;
343 	len = buflen;
344 	prepend(&end, &len, "\0", 1);
345 	/* Get '/' right */
346 	retval = end-1;
347 	*retval = '/';
348 	read_seqbegin_or_lock(&rename_lock, &seq);
349 	while (!IS_ROOT(dentry)) {
350 		const struct dentry *parent = dentry->d_parent;
351 
352 		prefetch(parent);
353 		error = prepend_name(&end, &len, &dentry->d_name);
354 		if (error)
355 			break;
356 
357 		retval = end;
358 		dentry = parent;
359 	}
360 	if (!(seq & 1))
361 		rcu_read_unlock();
362 	if (need_seqretry(&rename_lock, seq)) {
363 		seq = 1;
364 		goto restart;
365 	}
366 	done_seqretry(&rename_lock, seq);
367 	if (error)
368 		goto Elong;
369 	return retval;
370 Elong:
371 	return ERR_PTR(-ENAMETOOLONG);
372 }
373 
374 char *dentry_path_raw(const struct dentry *dentry, char *buf, int buflen)
375 {
376 	return __dentry_path(dentry, buf, buflen);
377 }
378 EXPORT_SYMBOL(dentry_path_raw);
379 
380 char *dentry_path(const struct dentry *dentry, char *buf, int buflen)
381 {
382 	char *p = NULL;
383 	char *retval;
384 
385 	if (d_unlinked(dentry)) {
386 		p = buf + buflen;
387 		if (prepend(&p, &buflen, "//deleted", 10) != 0)
388 			goto Elong;
389 		buflen++;
390 	}
391 	retval = __dentry_path(dentry, buf, buflen);
392 	if (!IS_ERR(retval) && p)
393 		*p = '/';	/* restore '/' overriden with '\0' */
394 	return retval;
395 Elong:
396 	return ERR_PTR(-ENAMETOOLONG);
397 }
398 
399 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
400 				    struct path *pwd)
401 {
402 	unsigned seq;
403 
404 	do {
405 		seq = read_seqcount_begin(&fs->seq);
406 		*root = fs->root;
407 		*pwd = fs->pwd;
408 	} while (read_seqcount_retry(&fs->seq, seq));
409 }
410 
411 /*
412  * NOTE! The user-level library version returns a
413  * character pointer. The kernel system call just
414  * returns the length of the buffer filled (which
415  * includes the ending '\0' character), or a negative
416  * error value. So libc would do something like
417  *
418  *	char *getcwd(char * buf, size_t size)
419  *	{
420  *		int retval;
421  *
422  *		retval = sys_getcwd(buf, size);
423  *		if (retval >= 0)
424  *			return buf;
425  *		errno = -retval;
426  *		return NULL;
427  *	}
428  */
429 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
430 {
431 	int error;
432 	struct path pwd, root;
433 	char *page = __getname();
434 
435 	if (!page)
436 		return -ENOMEM;
437 
438 	rcu_read_lock();
439 	get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
440 
441 	error = -ENOENT;
442 	if (!d_unlinked(pwd.dentry)) {
443 		unsigned long len;
444 		char *cwd = page + PATH_MAX;
445 		int buflen = PATH_MAX;
446 
447 		prepend(&cwd, &buflen, "\0", 1);
448 		error = prepend_path(&pwd, &root, &cwd, &buflen);
449 		rcu_read_unlock();
450 
451 		if (error < 0)
452 			goto out;
453 
454 		/* Unreachable from current root */
455 		if (error > 0) {
456 			error = prepend_unreachable(&cwd, &buflen);
457 			if (error)
458 				goto out;
459 		}
460 
461 		error = -ERANGE;
462 		len = PATH_MAX + page - cwd;
463 		if (len <= size) {
464 			error = len;
465 			if (copy_to_user(buf, cwd, len))
466 				error = -EFAULT;
467 		}
468 	} else {
469 		rcu_read_unlock();
470 	}
471 
472 out:
473 	__putname(page);
474 	return error;
475 }
476