ext4: Optimize ext4 DIO overwrites
[linux/fpc-iii.git] / fs / d_path.c
blob0f1fc1743302f329eb1b6ce7066237410ed6cc07
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"
11 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
13 *buflen -= namelen;
14 if (*buflen < 0)
15 return -ENAMETOOLONG;
16 *buffer -= namelen;
17 memcpy(*buffer, str, namelen);
18 return 0;
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
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.
36 * Load acquire is needed to make sure that we see that terminating NUL.
38 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
40 const char *dname = smp_load_acquire(&name->name); /* ^^^ */
41 u32 dlen = READ_ONCE(name->len);
42 char *p;
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;
55 return 0;
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
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.
75 static int prepend_path(const struct path *path,
76 const struct path *root,
77 char **buffer, int *buflen)
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;
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;
103 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
104 struct mount *parent = READ_ONCE(mnt->mnt_parent);
105 /* Escaped? */
106 if (dentry != vfsmnt->mnt_root) {
107 bptr = *buffer;
108 blen = *buflen;
109 error = 3;
110 break;
112 /* Global root? */
113 if (mnt != parent) {
114 dentry = READ_ONCE(mnt->mnt_mountpoint);
115 mnt = parent;
116 vfsmnt = &mnt->mnt;
117 continue;
119 if (is_mounted(vfsmnt) && !is_anon_ns(mnt->mnt_ns))
120 error = 1; // absolute root
121 else
122 error = 2; // detached or not attached yet
123 break;
125 parent = dentry->d_parent;
126 prefetch(parent);
127 error = prepend_name(&bptr, &blen, &dentry->d_name);
128 if (error)
129 break;
131 dentry = parent;
133 if (!(seq & 1))
134 rcu_read_unlock();
135 if (need_seqretry(&rename_lock, seq)) {
136 seq = 1;
137 goto restart;
139 done_seqretry(&rename_lock, seq);
141 if (!(m_seq & 1))
142 rcu_read_unlock();
143 if (need_seqretry(&mount_lock, m_seq)) {
144 m_seq = 1;
145 goto restart_mnt;
147 done_seqretry(&mount_lock, m_seq);
149 if (error >= 0 && bptr == *buffer) {
150 if (--blen < 0)
151 error = -ENAMETOOLONG;
152 else
153 *--bptr = '/';
155 *buffer = bptr;
156 *buflen = blen;
157 return error;
161 * __d_path - return the path of a dentry
162 * @path: the dentry/vfsmount to report
163 * @root: root vfsmnt/dentry
164 * @buf: buffer to return value in
165 * @buflen: buffer length
167 * Convert a dentry into an ASCII path name.
169 * Returns a pointer into the buffer or an error code if the
170 * path was too long.
172 * "buflen" should be positive.
174 * If the path is not reachable from the supplied root, return %NULL.
176 char *__d_path(const struct path *path,
177 const struct path *root,
178 char *buf, int buflen)
180 char *res = buf + buflen;
181 int error;
183 prepend(&res, &buflen, "\0", 1);
184 error = prepend_path(path, root, &res, &buflen);
186 if (error < 0)
187 return ERR_PTR(error);
188 if (error > 0)
189 return NULL;
190 return res;
193 char *d_absolute_path(const struct path *path,
194 char *buf, int buflen)
196 struct path root = {};
197 char *res = buf + buflen;
198 int error;
200 prepend(&res, &buflen, "\0", 1);
201 error = prepend_path(path, &root, &res, &buflen);
203 if (error > 1)
204 error = -EINVAL;
205 if (error < 0)
206 return ERR_PTR(error);
207 return res;
211 * same as __d_path but appends "(deleted)" for unlinked files.
213 static int path_with_deleted(const struct path *path,
214 const struct path *root,
215 char **buf, int *buflen)
217 prepend(buf, buflen, "\0", 1);
218 if (d_unlinked(path->dentry)) {
219 int error = prepend(buf, buflen, " (deleted)", 10);
220 if (error)
221 return error;
224 return prepend_path(path, root, buf, buflen);
227 static int prepend_unreachable(char **buffer, int *buflen)
229 return prepend(buffer, buflen, "(unreachable)", 13);
232 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
234 unsigned seq;
236 do {
237 seq = read_seqcount_begin(&fs->seq);
238 *root = fs->root;
239 } while (read_seqcount_retry(&fs->seq, seq));
243 * d_path - return the path of a dentry
244 * @path: path to report
245 * @buf: buffer to return value in
246 * @buflen: buffer length
248 * Convert a dentry into an ASCII path name. If the entry has been deleted
249 * the string " (deleted)" is appended. Note that this is ambiguous.
251 * Returns a pointer into the buffer or an error code if the path was
252 * too long. Note: Callers should use the returned pointer, not the passed
253 * in buffer, to use the name! The implementation often starts at an offset
254 * into the buffer, and may leave 0 bytes at the start.
256 * "buflen" should be positive.
258 char *d_path(const struct path *path, char *buf, int buflen)
260 char *res = buf + buflen;
261 struct path root;
262 int error;
265 * We have various synthetic filesystems that never get mounted. On
266 * these filesystems dentries are never used for lookup purposes, and
267 * thus don't need to be hashed. They also don't need a name until a
268 * user wants to identify the object in /proc/pid/fd/. The little hack
269 * below allows us to generate a name for these objects on demand:
271 * Some pseudo inodes are mountable. When they are mounted
272 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
273 * and instead have d_path return the mounted path.
275 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
276 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
277 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
279 rcu_read_lock();
280 get_fs_root_rcu(current->fs, &root);
281 error = path_with_deleted(path, &root, &res, &buflen);
282 rcu_read_unlock();
284 if (error < 0)
285 res = ERR_PTR(error);
286 return res;
288 EXPORT_SYMBOL(d_path);
291 * Helper function for dentry_operations.d_dname() members
293 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
294 const char *fmt, ...)
296 va_list args;
297 char temp[64];
298 int sz;
300 va_start(args, fmt);
301 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
302 va_end(args);
304 if (sz > sizeof(temp) || sz > buflen)
305 return ERR_PTR(-ENAMETOOLONG);
307 buffer += buflen - sz;
308 return memcpy(buffer, temp, sz);
311 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
313 char *end = buffer + buflen;
314 /* these dentries are never renamed, so d_lock is not needed */
315 if (prepend(&end, &buflen, " (deleted)", 11) ||
316 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
317 prepend(&end, &buflen, "/", 1))
318 end = ERR_PTR(-ENAMETOOLONG);
319 return end;
323 * Write full pathname from the root of the filesystem into the buffer.
325 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
327 struct dentry *dentry;
328 char *end, *retval;
329 int len, seq = 0;
330 int error = 0;
332 if (buflen < 2)
333 goto Elong;
335 rcu_read_lock();
336 restart:
337 dentry = d;
338 end = buf + buflen;
339 len = buflen;
340 prepend(&end, &len, "\0", 1);
341 /* Get '/' right */
342 retval = end-1;
343 *retval = '/';
344 read_seqbegin_or_lock(&rename_lock, &seq);
345 while (!IS_ROOT(dentry)) {
346 struct dentry *parent = dentry->d_parent;
348 prefetch(parent);
349 error = prepend_name(&end, &len, &dentry->d_name);
350 if (error)
351 break;
353 retval = end;
354 dentry = parent;
356 if (!(seq & 1))
357 rcu_read_unlock();
358 if (need_seqretry(&rename_lock, seq)) {
359 seq = 1;
360 goto restart;
362 done_seqretry(&rename_lock, seq);
363 if (error)
364 goto Elong;
365 return retval;
366 Elong:
367 return ERR_PTR(-ENAMETOOLONG);
370 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
372 return __dentry_path(dentry, buf, buflen);
374 EXPORT_SYMBOL(dentry_path_raw);
376 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
378 char *p = NULL;
379 char *retval;
381 if (d_unlinked(dentry)) {
382 p = buf + buflen;
383 if (prepend(&p, &buflen, "//deleted", 10) != 0)
384 goto Elong;
385 buflen++;
387 retval = __dentry_path(dentry, buf, buflen);
388 if (!IS_ERR(retval) && p)
389 *p = '/'; /* restore '/' overriden with '\0' */
390 return retval;
391 Elong:
392 return ERR_PTR(-ENAMETOOLONG);
395 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
396 struct path *pwd)
398 unsigned seq;
400 do {
401 seq = read_seqcount_begin(&fs->seq);
402 *root = fs->root;
403 *pwd = fs->pwd;
404 } while (read_seqcount_retry(&fs->seq, seq));
408 * NOTE! The user-level library version returns a
409 * character pointer. The kernel system call just
410 * returns the length of the buffer filled (which
411 * includes the ending '\0' character), or a negative
412 * error value. So libc would do something like
414 * char *getcwd(char * buf, size_t size)
416 * int retval;
418 * retval = sys_getcwd(buf, size);
419 * if (retval >= 0)
420 * return buf;
421 * errno = -retval;
422 * return NULL;
425 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
427 int error;
428 struct path pwd, root;
429 char *page = __getname();
431 if (!page)
432 return -ENOMEM;
434 rcu_read_lock();
435 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
437 error = -ENOENT;
438 if (!d_unlinked(pwd.dentry)) {
439 unsigned long len;
440 char *cwd = page + PATH_MAX;
441 int buflen = PATH_MAX;
443 prepend(&cwd, &buflen, "\0", 1);
444 error = prepend_path(&pwd, &root, &cwd, &buflen);
445 rcu_read_unlock();
447 if (error < 0)
448 goto out;
450 /* Unreachable from current root */
451 if (error > 0) {
452 error = prepend_unreachable(&cwd, &buflen);
453 if (error)
454 goto out;
457 error = -ERANGE;
458 len = PATH_MAX + page - cwd;
459 if (len <= size) {
460 error = len;
461 if (copy_to_user(buf, cwd, len))
462 error = -EFAULT;
464 } else {
465 rcu_read_unlock();
468 out:
469 __putname(page);
470 return error;