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
extract_string(struct prepend_buffer * p)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
prepend_char(struct prepend_buffer * p,unsigned char c)25 static bool prepend_char(struct prepend_buffer *p, unsigned char c)
26 {
27 if (likely(p->len > 0)) {
28 p->len--;
29 *--p->buf = c;
30 return true;
31 }
32 p->len = -1;
33 return false;
34 }
35
36 /*
37 * The source of the prepend data can be an optimistoc load
38 * of a dentry name and length. And because we don't hold any
39 * locks, the length and the pointer to the name may not be
40 * in sync if a concurrent rename happens, and the kernel
41 * copy might fault as a result.
42 *
43 * The end result will correct itself when we check the
44 * rename sequence count, but we need to be able to handle
45 * the fault gracefully.
46 */
prepend_copy(void * dst,const void * src,int len)47 static bool prepend_copy(void *dst, const void *src, int len)
48 {
49 if (unlikely(copy_from_kernel_nofault(dst, src, len))) {
50 memset(dst, 'x', len);
51 return false;
52 }
53 return true;
54 }
55
prepend(struct prepend_buffer * p,const char * str,int namelen)56 static bool prepend(struct prepend_buffer *p, const char *str, int namelen)
57 {
58 // Already overflowed?
59 if (p->len < 0)
60 return false;
61
62 // Will overflow?
63 if (p->len < namelen) {
64 // Fill as much as possible from the end of the name
65 str += namelen - p->len;
66 p->buf -= p->len;
67 prepend_copy(p->buf, str, p->len);
68 p->len = -1;
69 return false;
70 }
71
72 // Fits fully
73 p->len -= namelen;
74 p->buf -= namelen;
75 return prepend_copy(p->buf, str, namelen);
76 }
77
78 /**
79 * prepend_name - prepend a pathname in front of current buffer pointer
80 * @p: prepend buffer which contains buffer pointer and allocated length
81 * @name: name string and length qstr structure
82 *
83 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
84 * make sure that either the old or the new name pointer and length are
85 * fetched. However, there may be mismatch between length and pointer.
86 * But since the length cannot be trusted, we need to copy the name very
87 * carefully when doing the prepend_copy(). It also prepends "/" at
88 * the beginning of the name. The sequence number check at the caller will
89 * retry it again when a d_move() does happen. So any garbage in the buffer
90 * due to mismatched pointer and length will be discarded.
91 *
92 * Load acquire is needed to make sure that we see the new name data even
93 * if we might get the length wrong.
94 */
prepend_name(struct prepend_buffer * p,const struct qstr * name)95 static bool prepend_name(struct prepend_buffer *p, const struct qstr *name)
96 {
97 const char *dname = smp_load_acquire(&name->name); /* ^^^ */
98 u32 dlen = READ_ONCE(name->len);
99
100 return prepend(p, dname, dlen) && prepend_char(p, '/');
101 }
102
__prepend_path(const struct dentry * dentry,const struct mount * mnt,const struct path * root,struct prepend_buffer * p)103 static int __prepend_path(const struct dentry *dentry, const struct mount *mnt,
104 const struct path *root, struct prepend_buffer *p)
105 {
106 while (dentry != root->dentry || &mnt->mnt != root->mnt) {
107 const struct dentry *parent = READ_ONCE(dentry->d_parent);
108
109 if (dentry == mnt->mnt.mnt_root) {
110 struct mount *m = READ_ONCE(mnt->mnt_parent);
111 struct mnt_namespace *mnt_ns;
112
113 if (likely(mnt != m)) {
114 dentry = READ_ONCE(mnt->mnt_mountpoint);
115 mnt = m;
116 continue;
117 }
118 /* Global root */
119 mnt_ns = READ_ONCE(mnt->mnt_ns);
120 /* open-coded is_mounted() to use local mnt_ns */
121 if (!IS_ERR_OR_NULL(mnt_ns) && !is_anon_ns(mnt_ns))
122 return 1; // absolute root
123 else
124 return 2; // detached or not attached yet
125 }
126
127 if (unlikely(dentry == parent))
128 /* Escaped? */
129 return 3;
130
131 prefetch(parent);
132 if (!prepend_name(p, &dentry->d_name))
133 break;
134 dentry = parent;
135 }
136 return 0;
137 }
138
139 /**
140 * prepend_path - Prepend path string to a buffer
141 * @path: the dentry/vfsmount to report
142 * @root: root vfsmnt/dentry
143 * @p: prepend buffer which contains buffer pointer and allocated length
144 *
145 * The function will first try to write out the pathname without taking any
146 * lock other than the RCU read lock to make sure that dentries won't go away.
147 * It only checks the sequence number of the global rename_lock as any change
148 * in the dentry's d_seq will be preceded by changes in the rename_lock
149 * sequence number. If the sequence number had been changed, it will restart
150 * the whole pathname back-tracing sequence again by taking the rename_lock.
151 * In this case, there is no need to take the RCU read lock as the recursive
152 * parent pointer references will keep the dentry chain alive as long as no
153 * rename operation is performed.
154 */
prepend_path(const struct path * path,const struct path * root,struct prepend_buffer * p)155 static int prepend_path(const struct path *path,
156 const struct path *root,
157 struct prepend_buffer *p)
158 {
159 unsigned seq, m_seq = 0;
160 struct prepend_buffer b;
161 int error;
162
163 rcu_read_lock();
164 restart_mnt:
165 read_seqbegin_or_lock(&mount_lock, &m_seq);
166 seq = 0;
167 rcu_read_lock();
168 restart:
169 b = *p;
170 read_seqbegin_or_lock(&rename_lock, &seq);
171 error = __prepend_path(path->dentry, real_mount(path->mnt), root, &b);
172 if (!(seq & 1))
173 rcu_read_unlock();
174 if (need_seqretry(&rename_lock, seq)) {
175 seq = 1;
176 goto restart;
177 }
178 done_seqretry(&rename_lock, seq);
179
180 if (!(m_seq & 1))
181 rcu_read_unlock();
182 if (need_seqretry(&mount_lock, m_seq)) {
183 m_seq = 1;
184 goto restart_mnt;
185 }
186 done_seqretry(&mount_lock, m_seq);
187
188 if (unlikely(error == 3))
189 b = *p;
190
191 if (b.len == p->len)
192 prepend_char(&b, '/');
193
194 *p = b;
195 return error;
196 }
197
198 /**
199 * __d_path - return the path of a dentry
200 * @path: the dentry/vfsmount to report
201 * @root: root vfsmnt/dentry
202 * @buf: buffer to return value in
203 * @buflen: buffer length
204 *
205 * Convert a dentry into an ASCII path name.
206 *
207 * Returns a pointer into the buffer or an error code if the
208 * path was too long.
209 *
210 * "buflen" should be positive.
211 *
212 * If the path is not reachable from the supplied root, return %NULL.
213 */
__d_path(const struct path * path,const struct path * root,char * buf,int buflen)214 char *__d_path(const struct path *path,
215 const struct path *root,
216 char *buf, int buflen)
217 {
218 DECLARE_BUFFER(b, buf, buflen);
219
220 prepend_char(&b, 0);
221 if (unlikely(prepend_path(path, root, &b) > 0))
222 return NULL;
223 return extract_string(&b);
224 }
225
d_absolute_path(const struct path * path,char * buf,int buflen)226 char *d_absolute_path(const struct path *path,
227 char *buf, int buflen)
228 {
229 struct path root = {};
230 DECLARE_BUFFER(b, buf, buflen);
231
232 prepend_char(&b, 0);
233 if (unlikely(prepend_path(path, &root, &b) > 1))
234 return ERR_PTR(-EINVAL);
235 return extract_string(&b);
236 }
237
get_fs_root_rcu(struct fs_struct * fs,struct path * root)238 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
239 {
240 unsigned seq;
241
242 do {
243 seq = read_seqcount_begin(&fs->seq);
244 *root = fs->root;
245 } while (read_seqcount_retry(&fs->seq, seq));
246 }
247
248 /**
249 * d_path - return the path of a dentry
250 * @path: path to report
251 * @buf: buffer to return value in
252 * @buflen: buffer length
253 *
254 * Convert a dentry into an ASCII path name. If the entry has been deleted
255 * the string " (deleted)" is appended. Note that this is ambiguous.
256 *
257 * Returns a pointer into the buffer or an error code if the path was
258 * too long. Note: Callers should use the returned pointer, not the passed
259 * in buffer, to use the name! The implementation often starts at an offset
260 * into the buffer, and may leave 0 bytes at the start.
261 *
262 * "buflen" should be positive.
263 */
d_path(const struct path * path,char * buf,int buflen)264 char *d_path(const struct path *path, char *buf, int buflen)
265 {
266 DECLARE_BUFFER(b, buf, buflen);
267 struct path root;
268
269 /*
270 * We have various synthetic filesystems that never get mounted. On
271 * these filesystems dentries are never used for lookup purposes, and
272 * thus don't need to be hashed. They also don't need a name until a
273 * user wants to identify the object in /proc/pid/fd/. The little hack
274 * below allows us to generate a name for these objects on demand:
275 *
276 * Some pseudo inodes are mountable. When they are mounted
277 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
278 * and instead have d_path return the mounted path.
279 */
280 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
281 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
282 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
283
284 rcu_read_lock();
285 get_fs_root_rcu(current->fs, &root);
286 if (unlikely(d_unlinked(path->dentry)))
287 prepend(&b, " (deleted)", 11);
288 else
289 prepend_char(&b, 0);
290 prepend_path(path, &root, &b);
291 rcu_read_unlock();
292
293 return extract_string(&b);
294 }
295 EXPORT_SYMBOL(d_path);
296
297 /*
298 * Helper function for dentry_operations.d_dname() members
299 */
dynamic_dname(struct dentry * dentry,char * buffer,int buflen,const char * fmt,...)300 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
301 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
simple_dname(struct dentry * dentry,char * buffer,int buflen)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 */
__dentry_path(const struct dentry * d,struct prepend_buffer * p)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
dentry_path_raw(const struct dentry * dentry,char * buf,int buflen)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
dentry_path(const struct dentry * dentry,char * buf,int buflen)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
get_fs_root_and_pwd_rcu(struct fs_struct * fs,struct path * root,struct path * pwd)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 */
SYSCALL_DEFINE2(getcwd,char __user *,buf,unsigned long,size)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