xref: /linux/fs/d_path.c (revision ad08ae58)
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 struct prepend_buffer {
12 	char *buf;
13 	int len;
14 };
15 #define DECLARE_BUFFER(__name, __buf, __len) \
16 	struct prepend_buffer __name = {.buf = __buf + __len, .len = __len}
17 
18 static char *extract_string(struct prepend_buffer *p)
19 {
20 	if (likely(p->len >= 0))
21 		return p->buf;
22 	return ERR_PTR(-ENAMETOOLONG);
23 }
24 
25 static void prepend(struct prepend_buffer *p, const char *str, int namelen)
26 {
27 	p->len -= namelen;
28 	if (likely(p->len >= 0)) {
29 		p->buf -= namelen;
30 		memcpy(p->buf, str, namelen);
31 	}
32 }
33 
34 /**
35  * prepend_name - prepend a pathname in front of current buffer pointer
36  * @buffer: buffer pointer
37  * @buflen: allocated length of the buffer
38  * @name:   name string and length qstr structure
39  *
40  * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
41  * make sure that either the old or the new name pointer and length are
42  * fetched. However, there may be mismatch between length and pointer.
43  * The length cannot be trusted, we need to copy it byte-by-byte until
44  * the length is reached or a null byte is found. It also prepends "/" at
45  * the beginning of the name. The sequence number check at the caller will
46  * retry it again when a d_move() does happen. So any garbage in the buffer
47  * due to mismatched pointer and length will be discarded.
48  *
49  * Load acquire is needed to make sure that we see that terminating NUL.
50  */
51 static bool prepend_name(struct prepend_buffer *p, const struct qstr *name)
52 {
53 	const char *dname = smp_load_acquire(&name->name); /* ^^^ */
54 	u32 dlen = READ_ONCE(name->len);
55 	char *s;
56 
57 	p->len -= dlen + 1;
58 	if (unlikely(p->len < 0))
59 		return false;
60 	s = p->buf -= dlen + 1;
61 	*s++ = '/';
62 	while (dlen--) {
63 		char c = *dname++;
64 		if (!c)
65 			break;
66 		*s++ = c;
67 	}
68 	return true;
69 }
70 
71 /**
72  * prepend_path - Prepend path string to a buffer
73  * @path: the dentry/vfsmount to report
74  * @root: root vfsmnt/dentry
75  * @buffer: pointer to the end of the buffer
76  * @buflen: pointer to buffer length
77  *
78  * The function will first try to write out the pathname without taking any
79  * lock other than the RCU read lock to make sure that dentries won't go away.
80  * It only checks the sequence number of the global rename_lock as any change
81  * in the dentry's d_seq will be preceded by changes in the rename_lock
82  * sequence number. If the sequence number had been changed, it will restart
83  * the whole pathname back-tracing sequence again by taking the rename_lock.
84  * In this case, there is no need to take the RCU read lock as the recursive
85  * parent pointer references will keep the dentry chain alive as long as no
86  * rename operation is performed.
87  */
88 static int prepend_path(const struct path *path,
89 			const struct path *root,
90 			struct prepend_buffer *p)
91 {
92 	struct dentry *dentry;
93 	struct vfsmount *vfsmnt;
94 	struct mount *mnt;
95 	int error = 0;
96 	unsigned seq, m_seq = 0;
97 	struct prepend_buffer b;
98 
99 	rcu_read_lock();
100 restart_mnt:
101 	read_seqbegin_or_lock(&mount_lock, &m_seq);
102 	seq = 0;
103 	rcu_read_lock();
104 restart:
105 	b = *p;
106 	error = 0;
107 	dentry = path->dentry;
108 	vfsmnt = path->mnt;
109 	mnt = real_mount(vfsmnt);
110 	read_seqbegin_or_lock(&rename_lock, &seq);
111 	while (dentry != root->dentry || vfsmnt != root->mnt) {
112 		struct dentry * parent;
113 
114 		if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
115 			struct mount *parent = READ_ONCE(mnt->mnt_parent);
116 			struct mnt_namespace *mnt_ns;
117 
118 			/* Escaped? */
119 			if (dentry != vfsmnt->mnt_root) {
120 				b = *p;
121 				error = 3;
122 				break;
123 			}
124 			/* Global root? */
125 			if (mnt != parent) {
126 				dentry = READ_ONCE(mnt->mnt_mountpoint);
127 				mnt = parent;
128 				vfsmnt = &mnt->mnt;
129 				continue;
130 			}
131 			mnt_ns = READ_ONCE(mnt->mnt_ns);
132 			/* open-coded is_mounted() to use local mnt_ns */
133 			if (!IS_ERR_OR_NULL(mnt_ns) && !is_anon_ns(mnt_ns))
134 				error = 1;	// absolute root
135 			else
136 				error = 2;	// detached or not attached yet
137 			break;
138 		}
139 		parent = dentry->d_parent;
140 		prefetch(parent);
141 		if (!prepend_name(&b, &dentry->d_name))
142 			break;
143 
144 		dentry = parent;
145 	}
146 	if (!(seq & 1))
147 		rcu_read_unlock();
148 	if (need_seqretry(&rename_lock, seq)) {
149 		seq = 1;
150 		goto restart;
151 	}
152 	done_seqretry(&rename_lock, seq);
153 
154 	if (!(m_seq & 1))
155 		rcu_read_unlock();
156 	if (need_seqretry(&mount_lock, m_seq)) {
157 		m_seq = 1;
158 		goto restart_mnt;
159 	}
160 	done_seqretry(&mount_lock, m_seq);
161 
162 	if (b.len == p->len)
163 		prepend(&b, "/", 1);
164 
165 	*p = b;
166 	return error;
167 }
168 
169 /**
170  * __d_path - return the path of a dentry
171  * @path: the dentry/vfsmount to report
172  * @root: root vfsmnt/dentry
173  * @buf: buffer to return value in
174  * @buflen: buffer length
175  *
176  * Convert a dentry into an ASCII path name.
177  *
178  * Returns a pointer into the buffer or an error code if the
179  * path was too long.
180  *
181  * "buflen" should be positive.
182  *
183  * If the path is not reachable from the supplied root, return %NULL.
184  */
185 char *__d_path(const struct path *path,
186 	       const struct path *root,
187 	       char *buf, int buflen)
188 {
189 	DECLARE_BUFFER(b, buf, buflen);
190 
191 	prepend(&b, "", 1);
192 	if (prepend_path(path, root, &b) > 0)
193 		return NULL;
194 	return extract_string(&b);
195 }
196 
197 char *d_absolute_path(const struct path *path,
198 	       char *buf, int buflen)
199 {
200 	struct path root = {};
201 	DECLARE_BUFFER(b, buf, buflen);
202 
203 	prepend(&b, "", 1);
204 	if (prepend_path(path, &root, &b) > 1)
205 		return ERR_PTR(-EINVAL);
206 	return extract_string(&b);
207 }
208 
209 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
210 {
211 	unsigned seq;
212 
213 	do {
214 		seq = read_seqcount_begin(&fs->seq);
215 		*root = fs->root;
216 	} while (read_seqcount_retry(&fs->seq, seq));
217 }
218 
219 /**
220  * d_path - return the path of a dentry
221  * @path: path to report
222  * @buf: buffer to return value in
223  * @buflen: buffer length
224  *
225  * Convert a dentry into an ASCII path name. If the entry has been deleted
226  * the string " (deleted)" is appended. Note that this is ambiguous.
227  *
228  * Returns a pointer into the buffer or an error code if the path was
229  * too long. Note: Callers should use the returned pointer, not the passed
230  * in buffer, to use the name! The implementation often starts at an offset
231  * into the buffer, and may leave 0 bytes at the start.
232  *
233  * "buflen" should be positive.
234  */
235 char *d_path(const struct path *path, char *buf, int buflen)
236 {
237 	DECLARE_BUFFER(b, buf, buflen);
238 	struct path root;
239 
240 	/*
241 	 * We have various synthetic filesystems that never get mounted.  On
242 	 * these filesystems dentries are never used for lookup purposes, and
243 	 * thus don't need to be hashed.  They also don't need a name until a
244 	 * user wants to identify the object in /proc/pid/fd/.  The little hack
245 	 * below allows us to generate a name for these objects on demand:
246 	 *
247 	 * Some pseudo inodes are mountable.  When they are mounted
248 	 * path->dentry == path->mnt->mnt_root.  In that case don't call d_dname
249 	 * and instead have d_path return the mounted path.
250 	 */
251 	if (path->dentry->d_op && path->dentry->d_op->d_dname &&
252 	    (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
253 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
254 
255 	rcu_read_lock();
256 	get_fs_root_rcu(current->fs, &root);
257 	if (unlikely(d_unlinked(path->dentry)))
258 		prepend(&b, " (deleted)", 11);
259 	else
260 		prepend(&b, "", 1);
261 	prepend_path(path, &root, &b);
262 	rcu_read_unlock();
263 
264 	return extract_string(&b);
265 }
266 EXPORT_SYMBOL(d_path);
267 
268 /*
269  * Helper function for dentry_operations.d_dname() members
270  */
271 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
272 			const char *fmt, ...)
273 {
274 	va_list args;
275 	char temp[64];
276 	int sz;
277 
278 	va_start(args, fmt);
279 	sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
280 	va_end(args);
281 
282 	if (sz > sizeof(temp) || sz > buflen)
283 		return ERR_PTR(-ENAMETOOLONG);
284 
285 	buffer += buflen - sz;
286 	return memcpy(buffer, temp, sz);
287 }
288 
289 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
290 {
291 	DECLARE_BUFFER(b, buffer, buflen);
292 	/* these dentries are never renamed, so d_lock is not needed */
293 	prepend(&b, " (deleted)", 11);
294 	prepend(&b, dentry->d_name.name, dentry->d_name.len);
295 	prepend(&b, "/", 1);
296 	return extract_string(&b);
297 }
298 
299 /*
300  * Write full pathname from the root of the filesystem into the buffer.
301  */
302 static char *__dentry_path(const struct dentry *d, struct prepend_buffer *p)
303 {
304 	const struct dentry *dentry;
305 	struct prepend_buffer b;
306 	int seq = 0;
307 
308 	rcu_read_lock();
309 restart:
310 	dentry = d;
311 	b = *p;
312 	read_seqbegin_or_lock(&rename_lock, &seq);
313 	while (!IS_ROOT(dentry)) {
314 		const struct dentry *parent = dentry->d_parent;
315 
316 		prefetch(parent);
317 		if (!prepend_name(&b, &dentry->d_name))
318 			break;
319 		dentry = parent;
320 	}
321 	if (!(seq & 1))
322 		rcu_read_unlock();
323 	if (need_seqretry(&rename_lock, seq)) {
324 		seq = 1;
325 		goto restart;
326 	}
327 	done_seqretry(&rename_lock, seq);
328 	if (b.len == p->len)
329 		prepend(&b, "/", 1);
330 	return extract_string(&b);
331 }
332 
333 char *dentry_path_raw(const struct dentry *dentry, char *buf, int buflen)
334 {
335 	DECLARE_BUFFER(b, buf, buflen);
336 
337 	prepend(&b, "", 1);
338 	return __dentry_path(dentry, &b);
339 }
340 EXPORT_SYMBOL(dentry_path_raw);
341 
342 char *dentry_path(const struct dentry *dentry, char *buf, int buflen)
343 {
344 	DECLARE_BUFFER(b, buf, buflen);
345 
346 	if (unlikely(d_unlinked(dentry)))
347 		prepend(&b, "//deleted", 10);
348 	else
349 		prepend(&b, "", 1);
350 	return __dentry_path(dentry, &b);
351 }
352 
353 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
354 				    struct path *pwd)
355 {
356 	unsigned seq;
357 
358 	do {
359 		seq = read_seqcount_begin(&fs->seq);
360 		*root = fs->root;
361 		*pwd = fs->pwd;
362 	} while (read_seqcount_retry(&fs->seq, seq));
363 }
364 
365 /*
366  * NOTE! The user-level library version returns a
367  * character pointer. The kernel system call just
368  * returns the length of the buffer filled (which
369  * includes the ending '\0' character), or a negative
370  * error value. So libc would do something like
371  *
372  *	char *getcwd(char * buf, size_t size)
373  *	{
374  *		int retval;
375  *
376  *		retval = sys_getcwd(buf, size);
377  *		if (retval >= 0)
378  *			return buf;
379  *		errno = -retval;
380  *		return NULL;
381  *	}
382  */
383 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
384 {
385 	int error;
386 	struct path pwd, root;
387 	char *page = __getname();
388 
389 	if (!page)
390 		return -ENOMEM;
391 
392 	rcu_read_lock();
393 	get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
394 
395 	error = -ENOENT;
396 	if (!d_unlinked(pwd.dentry)) {
397 		unsigned long len;
398 		DECLARE_BUFFER(b, page, PATH_MAX);
399 
400 		prepend(&b, "", 1);
401 		if (prepend_path(&pwd, &root, &b) > 0)
402 			prepend(&b, "(unreachable)", 13);
403 		rcu_read_unlock();
404 
405 		if (b.len < 0) {
406 			error = -ENAMETOOLONG;
407 			goto out;
408 		}
409 
410 		error = -ERANGE;
411 		len = PATH_MAX - b.len;
412 		if (len <= size) {
413 			error = len;
414 			if (copy_to_user(buf, b.buf, len))
415 				error = -EFAULT;
416 		}
417 	} else {
418 		rcu_read_unlock();
419 	}
420 
421 out:
422 	__putname(page);
423 	return error;
424 }
425