xref: /openbmc/linux/fs/d_path.c (revision 0ef9d78b)
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 #include "internal.h"
11 
12 struct prepend_buffer {
13 	char *buf;
14 	int len;
15 };
16 #define DECLARE_BUFFER(__name, __buf, __len) \
17 	struct prepend_buffer __name = {.buf = __buf + __len, .len = __len}
18 
19 static char *extract_string(struct prepend_buffer *p)
20 {
21 	if (likely(p->len >= 0))
22 		return p->buf;
23 	return ERR_PTR(-ENAMETOOLONG);
24 }
25 
26 static bool prepend_char(struct prepend_buffer *p, unsigned char c)
27 {
28 	if (likely(p->len > 0)) {
29 		p->len--;
30 		*--p->buf = c;
31 		return true;
32 	}
33 	p->len = -1;
34 	return false;
35 }
36 
37 /*
38  * The source of the prepend data can be an optimistic load
39  * of a dentry name and length. And because we don't hold any
40  * locks, the length and the pointer to the name may not be
41  * in sync if a concurrent rename happens, and the kernel
42  * copy might fault as a result.
43  *
44  * The end result will correct itself when we check the
45  * rename sequence count, but we need to be able to handle
46  * the fault gracefully.
47  */
48 static bool prepend_copy(void *dst, const void *src, int len)
49 {
50 	if (unlikely(copy_from_kernel_nofault(dst, src, len))) {
51 		memset(dst, 'x', len);
52 		return false;
53 	}
54 	return true;
55 }
56 
57 static bool prepend(struct prepend_buffer *p, const char *str, int namelen)
58 {
59 	// Already overflowed?
60 	if (p->len < 0)
61 		return false;
62 
63 	// Will overflow?
64 	if (p->len < namelen) {
65 		// Fill as much as possible from the end of the name
66 		str += namelen - p->len;
67 		p->buf -= p->len;
68 		prepend_copy(p->buf, str, p->len);
69 		p->len = -1;
70 		return false;
71 	}
72 
73 	// Fits fully
74 	p->len -= namelen;
75 	p->buf -= namelen;
76 	return prepend_copy(p->buf, str, namelen);
77 }
78 
79 /**
80  * prepend_name - prepend a pathname in front of current buffer pointer
81  * @p: prepend buffer which contains buffer pointer and allocated length
82  * @name: name string and length qstr structure
83  *
84  * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
85  * make sure that either the old or the new name pointer and length are
86  * fetched. However, there may be mismatch between length and pointer.
87  * But since the length cannot be trusted, we need to copy the name very
88  * carefully when doing the prepend_copy(). It also prepends "/" at
89  * the beginning of the name. The sequence number check at the caller will
90  * retry it again when a d_move() does happen. So any garbage in the buffer
91  * due to mismatched pointer and length will be discarded.
92  *
93  * Load acquire is needed to make sure that we see the new name data even
94  * if we might get the length wrong.
95  */
96 static bool prepend_name(struct prepend_buffer *p, const struct qstr *name)
97 {
98 	const char *dname = smp_load_acquire(&name->name); /* ^^^ */
99 	u32 dlen = READ_ONCE(name->len);
100 
101 	return prepend(p, dname, dlen) && prepend_char(p, '/');
102 }
103 
104 static int __prepend_path(const struct dentry *dentry, const struct mount *mnt,
105 			  const struct path *root, struct prepend_buffer *p)
106 {
107 	while (dentry != root->dentry || &mnt->mnt != root->mnt) {
108 		const struct dentry *parent = READ_ONCE(dentry->d_parent);
109 
110 		if (dentry == mnt->mnt.mnt_root) {
111 			struct mount *m = READ_ONCE(mnt->mnt_parent);
112 			struct mnt_namespace *mnt_ns;
113 
114 			if (likely(mnt != m)) {
115 				dentry = READ_ONCE(mnt->mnt_mountpoint);
116 				mnt = m;
117 				continue;
118 			}
119 			/* Global root */
120 			mnt_ns = READ_ONCE(mnt->mnt_ns);
121 			/* open-coded is_mounted() to use local mnt_ns */
122 			if (!IS_ERR_OR_NULL(mnt_ns) && !is_anon_ns(mnt_ns))
123 				return 1;	// absolute root
124 			else
125 				return 2;	// detached or not attached yet
126 		}
127 
128 		if (unlikely(dentry == parent))
129 			/* Escaped? */
130 			return 3;
131 
132 		prefetch(parent);
133 		if (!prepend_name(p, &dentry->d_name))
134 			break;
135 		dentry = parent;
136 	}
137 	return 0;
138 }
139 
140 /**
141  * prepend_path - Prepend path string to a buffer
142  * @path: the dentry/vfsmount to report
143  * @root: root vfsmnt/dentry
144  * @p: prepend buffer which contains buffer pointer and allocated length
145  *
146  * The function will first try to write out the pathname without taking any
147  * lock other than the RCU read lock to make sure that dentries won't go away.
148  * It only checks the sequence number of the global rename_lock as any change
149  * in the dentry's d_seq will be preceded by changes in the rename_lock
150  * sequence number. If the sequence number had been changed, it will restart
151  * the whole pathname back-tracing sequence again by taking the rename_lock.
152  * In this case, there is no need to take the RCU read lock as the recursive
153  * parent pointer references will keep the dentry chain alive as long as no
154  * rename operation is performed.
155  */
156 static int prepend_path(const struct path *path,
157 			const struct path *root,
158 			struct prepend_buffer *p)
159 {
160 	unsigned seq, m_seq = 0;
161 	struct prepend_buffer b;
162 	int error;
163 
164 	rcu_read_lock();
165 restart_mnt:
166 	read_seqbegin_or_lock(&mount_lock, &m_seq);
167 	seq = 0;
168 	rcu_read_lock();
169 restart:
170 	b = *p;
171 	read_seqbegin_or_lock(&rename_lock, &seq);
172 	error = __prepend_path(path->dentry, real_mount(path->mnt), root, &b);
173 	if (!(seq & 1))
174 		rcu_read_unlock();
175 	if (need_seqretry(&rename_lock, seq)) {
176 		seq = 1;
177 		goto restart;
178 	}
179 	done_seqretry(&rename_lock, seq);
180 
181 	if (!(m_seq & 1))
182 		rcu_read_unlock();
183 	if (need_seqretry(&mount_lock, m_seq)) {
184 		m_seq = 1;
185 		goto restart_mnt;
186 	}
187 	done_seqretry(&mount_lock, m_seq);
188 
189 	if (unlikely(error == 3))
190 		b = *p;
191 
192 	if (b.len == p->len)
193 		prepend_char(&b, '/');
194 
195 	*p = b;
196 	return error;
197 }
198 
199 /**
200  * __d_path - return the path of a dentry
201  * @path: the dentry/vfsmount to report
202  * @root: root vfsmnt/dentry
203  * @buf: buffer to return value in
204  * @buflen: buffer length
205  *
206  * Convert a dentry into an ASCII path name.
207  *
208  * Returns a pointer into the buffer or an error code if the
209  * path was too long.
210  *
211  * "buflen" should be positive.
212  *
213  * If the path is not reachable from the supplied root, return %NULL.
214  */
215 char *__d_path(const struct path *path,
216 	       const struct path *root,
217 	       char *buf, int buflen)
218 {
219 	DECLARE_BUFFER(b, buf, buflen);
220 
221 	prepend_char(&b, 0);
222 	if (unlikely(prepend_path(path, root, &b) > 0))
223 		return NULL;
224 	return extract_string(&b);
225 }
226 
227 char *d_absolute_path(const struct path *path,
228 	       char *buf, int buflen)
229 {
230 	struct path root = {};
231 	DECLARE_BUFFER(b, buf, buflen);
232 
233 	prepend_char(&b, 0);
234 	if (unlikely(prepend_path(path, &root, &b) > 1))
235 		return ERR_PTR(-EINVAL);
236 	return extract_string(&b);
237 }
238 
239 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
240 {
241 	unsigned seq;
242 
243 	do {
244 		seq = read_seqcount_begin(&fs->seq);
245 		*root = fs->root;
246 	} while (read_seqcount_retry(&fs->seq, seq));
247 }
248 
249 /**
250  * d_path - return the path of a dentry
251  * @path: path to report
252  * @buf: buffer to return value in
253  * @buflen: buffer length
254  *
255  * Convert a dentry into an ASCII path name. If the entry has been deleted
256  * the string " (deleted)" is appended. Note that this is ambiguous.
257  *
258  * Returns a pointer into the buffer or an error code if the path was
259  * too long. Note: Callers should use the returned pointer, not the passed
260  * in buffer, to use the name! The implementation often starts at an offset
261  * into the buffer, and may leave 0 bytes at the start.
262  *
263  * "buflen" should be positive.
264  */
265 char *d_path(const struct path *path, char *buf, int buflen)
266 {
267 	DECLARE_BUFFER(b, buf, buflen);
268 	struct path root;
269 
270 	/*
271 	 * We have various synthetic filesystems that never get mounted.  On
272 	 * these filesystems dentries are never used for lookup purposes, and
273 	 * thus don't need to be hashed.  They also don't need a name until a
274 	 * user wants to identify the object in /proc/pid/fd/.  The little hack
275 	 * below allows us to generate a name for these objects on demand:
276 	 *
277 	 * Some pseudo inodes are mountable.  When they are mounted
278 	 * path->dentry == path->mnt->mnt_root.  In that case don't call d_dname
279 	 * and instead have d_path return the mounted path.
280 	 */
281 	if (path->dentry->d_op && path->dentry->d_op->d_dname &&
282 	    (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
283 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
284 
285 	rcu_read_lock();
286 	get_fs_root_rcu(current->fs, &root);
287 	if (unlikely(d_unlinked(path->dentry)))
288 		prepend(&b, " (deleted)", 11);
289 	else
290 		prepend_char(&b, 0);
291 	prepend_path(path, &root, &b);
292 	rcu_read_unlock();
293 
294 	return extract_string(&b);
295 }
296 EXPORT_SYMBOL(d_path);
297 
298 /*
299  * Helper function for dentry_operations.d_dname() members
300  */
301 char *dynamic_dname(char *buffer, int buflen, const char *fmt, ...)
302 {
303 	va_list args;
304 	char temp[64];
305 	int sz;
306 
307 	va_start(args, fmt);
308 	sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
309 	va_end(args);
310 
311 	if (sz > sizeof(temp) || sz > buflen)
312 		return ERR_PTR(-ENAMETOOLONG);
313 
314 	buffer += buflen - sz;
315 	return memcpy(buffer, temp, sz);
316 }
317 
318 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
319 {
320 	DECLARE_BUFFER(b, buffer, buflen);
321 	/* these dentries are never renamed, so d_lock is not needed */
322 	prepend(&b, " (deleted)", 11);
323 	prepend(&b, dentry->d_name.name, dentry->d_name.len);
324 	prepend_char(&b, '/');
325 	return extract_string(&b);
326 }
327 
328 /*
329  * Write full pathname from the root of the filesystem into the buffer.
330  */
331 static char *__dentry_path(const struct dentry *d, struct prepend_buffer *p)
332 {
333 	const struct dentry *dentry;
334 	struct prepend_buffer b;
335 	int seq = 0;
336 
337 	rcu_read_lock();
338 restart:
339 	dentry = d;
340 	b = *p;
341 	read_seqbegin_or_lock(&rename_lock, &seq);
342 	while (!IS_ROOT(dentry)) {
343 		const struct dentry *parent = dentry->d_parent;
344 
345 		prefetch(parent);
346 		if (!prepend_name(&b, &dentry->d_name))
347 			break;
348 		dentry = parent;
349 	}
350 	if (!(seq & 1))
351 		rcu_read_unlock();
352 	if (need_seqretry(&rename_lock, seq)) {
353 		seq = 1;
354 		goto restart;
355 	}
356 	done_seqretry(&rename_lock, seq);
357 	if (b.len == p->len)
358 		prepend_char(&b, '/');
359 	return extract_string(&b);
360 }
361 
362 char *dentry_path_raw(const struct dentry *dentry, char *buf, int buflen)
363 {
364 	DECLARE_BUFFER(b, buf, buflen);
365 
366 	prepend_char(&b, 0);
367 	return __dentry_path(dentry, &b);
368 }
369 EXPORT_SYMBOL(dentry_path_raw);
370 
371 char *dentry_path(const struct dentry *dentry, char *buf, int buflen)
372 {
373 	DECLARE_BUFFER(b, buf, buflen);
374 
375 	if (unlikely(d_unlinked(dentry)))
376 		prepend(&b, "//deleted", 10);
377 	else
378 		prepend_char(&b, 0);
379 	return __dentry_path(dentry, &b);
380 }
381 
382 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
383 				    struct path *pwd)
384 {
385 	unsigned seq;
386 
387 	do {
388 		seq = read_seqcount_begin(&fs->seq);
389 		*root = fs->root;
390 		*pwd = fs->pwd;
391 	} while (read_seqcount_retry(&fs->seq, seq));
392 }
393 
394 /*
395  * NOTE! The user-level library version returns a
396  * character pointer. The kernel system call just
397  * returns the length of the buffer filled (which
398  * includes the ending '\0' character), or a negative
399  * error value. So libc would do something like
400  *
401  *	char *getcwd(char * buf, size_t size)
402  *	{
403  *		int retval;
404  *
405  *		retval = sys_getcwd(buf, size);
406  *		if (retval >= 0)
407  *			return buf;
408  *		errno = -retval;
409  *		return NULL;
410  *	}
411  */
412 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
413 {
414 	int error;
415 	struct path pwd, root;
416 	char *page = __getname();
417 
418 	if (!page)
419 		return -ENOMEM;
420 
421 	rcu_read_lock();
422 	get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
423 
424 	if (unlikely(d_unlinked(pwd.dentry))) {
425 		rcu_read_unlock();
426 		error = -ENOENT;
427 	} else {
428 		unsigned len;
429 		DECLARE_BUFFER(b, page, PATH_MAX);
430 
431 		prepend_char(&b, 0);
432 		if (unlikely(prepend_path(&pwd, &root, &b) > 0))
433 			prepend(&b, "(unreachable)", 13);
434 		rcu_read_unlock();
435 
436 		len = PATH_MAX - b.len;
437 		if (unlikely(len > PATH_MAX))
438 			error = -ENAMETOOLONG;
439 		else if (unlikely(len > size))
440 			error = -ERANGE;
441 		else if (copy_to_user(buf, b.buf, len))
442 			error = -EFAULT;
443 		else
444 			error = len;
445 	}
446 	__putname(page);
447 	return error;
448 }
449