1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/readdir.c
4 *
5 * Copyright (C) 1995 Linus Torvalds
6 */
7
8 #include <linux/stddef.h>
9 #include <linux/kernel.h>
10 #include <linux/export.h>
11 #include <linux/time.h>
12 #include <linux/mm.h>
13 #include <linux/errno.h>
14 #include <linux/stat.h>
15 #include <linux/file.h>
16 #include <linux/fs.h>
17 #include <linux/fsnotify.h>
18 #include <linux/dirent.h>
19 #include <linux/security.h>
20 #include <linux/syscalls.h>
21 #include <linux/unistd.h>
22 #include <linux/compat.h>
23 #include <linux/uaccess.h>
24
25 /*
26 * Some filesystems were never converted to '->iterate_shared()'
27 * and their directory iterators want the inode lock held for
28 * writing. This wrapper allows for converting from the shared
29 * semantics to the exclusive inode use.
30 */
wrap_directory_iterator(struct file * file,struct dir_context * ctx,int (* iter)(struct file *,struct dir_context *))31 int wrap_directory_iterator(struct file *file,
32 struct dir_context *ctx,
33 int (*iter)(struct file *, struct dir_context *))
34 {
35 struct inode *inode = file_inode(file);
36 int ret;
37
38 /*
39 * We'd love to have an 'inode_upgrade_trylock()' operation,
40 * see the comment in mmap_upgrade_trylock() in mm/memory.c.
41 *
42 * But considering this is for "filesystems that never got
43 * converted", it really doesn't matter.
44 *
45 * Also note that since we have to return with the lock held
46 * for reading, we can't use the "killable()" locking here,
47 * since we do need to get the lock even if we're dying.
48 *
49 * We could do the write part killably and then get the read
50 * lock unconditionally if it mattered, but see above on why
51 * this does the very simplistic conversion.
52 */
53 up_read(&inode->i_rwsem);
54 down_write(&inode->i_rwsem);
55
56 /*
57 * Since we dropped the inode lock, we should do the
58 * DEADDIR test again. See 'iterate_dir()' below.
59 *
60 * Note that we don't need to re-do the f_pos games,
61 * since the file must be locked wrt f_pos anyway.
62 */
63 ret = -ENOENT;
64 if (!IS_DEADDIR(inode))
65 ret = iter(file, ctx);
66
67 downgrade_write(&inode->i_rwsem);
68 return ret;
69 }
70 EXPORT_SYMBOL(wrap_directory_iterator);
71
72 /*
73 * Note the "unsafe_put_user()" semantics: we goto a
74 * label for errors.
75 */
76 #define unsafe_copy_dirent_name(_dst, _src, _len, label) do { \
77 char __user *dst = (_dst); \
78 const char *src = (_src); \
79 size_t len = (_len); \
80 unsafe_put_user(0, dst+len, label); \
81 unsafe_copy_to_user(dst, src, len, label); \
82 } while (0)
83
84
iterate_dir(struct file * file,struct dir_context * ctx)85 int iterate_dir(struct file *file, struct dir_context *ctx)
86 {
87 struct inode *inode = file_inode(file);
88 int res = -ENOTDIR;
89
90 if (!file->f_op->iterate_shared)
91 goto out;
92
93 res = security_file_permission(file, MAY_READ);
94 if (res)
95 goto out;
96
97 res = fsnotify_file_perm(file, MAY_READ);
98 if (res)
99 goto out;
100
101 res = down_read_killable(&inode->i_rwsem);
102 if (res)
103 goto out;
104
105 res = -ENOENT;
106 if (!IS_DEADDIR(inode)) {
107 ctx->pos = file->f_pos;
108 res = file->f_op->iterate_shared(file, ctx);
109 file->f_pos = ctx->pos;
110 fsnotify_access(file);
111 file_accessed(file);
112 }
113 inode_unlock_shared(inode);
114 out:
115 return res;
116 }
117 EXPORT_SYMBOL(iterate_dir);
118
119 /*
120 * POSIX says that a dirent name cannot contain NULL or a '/'.
121 *
122 * It's not 100% clear what we should really do in this case.
123 * The filesystem is clearly corrupted, but returning a hard
124 * error means that you now don't see any of the other names
125 * either, so that isn't a perfect alternative.
126 *
127 * And if you return an error, what error do you use? Several
128 * filesystems seem to have decided on EUCLEAN being the error
129 * code for EFSCORRUPTED, and that may be the error to use. Or
130 * just EIO, which is perhaps more obvious to users.
131 *
132 * In order to see the other file names in the directory, the
133 * caller might want to make this a "soft" error: skip the
134 * entry, and return the error at the end instead.
135 *
136 * Note that this should likely do a "memchr(name, 0, len)"
137 * check too, since that would be filesystem corruption as
138 * well. However, that case can't actually confuse user space,
139 * which has to do a strlen() on the name anyway to find the
140 * filename length, and the above "soft error" worry means
141 * that it's probably better left alone until we have that
142 * issue clarified.
143 *
144 * Note the PATH_MAX check - it's arbitrary but the real
145 * kernel limit on a possible path component, not NAME_MAX,
146 * which is the technical standard limit.
147 */
verify_dirent_name(const char * name,int len)148 static int verify_dirent_name(const char *name, int len)
149 {
150 if (len <= 0 || len >= PATH_MAX)
151 return -EIO;
152 if (memchr(name, '/', len))
153 return -EIO;
154 return 0;
155 }
156
157 /*
158 * Traditional linux readdir() handling..
159 *
160 * "count=1" is a special case, meaning that the buffer is one
161 * dirent-structure in size and that the code can't handle more
162 * anyway. Thus the special "fillonedir()" function for that
163 * case (the low-level handlers don't need to care about this).
164 */
165
166 #ifdef __ARCH_WANT_OLD_READDIR
167
168 struct old_linux_dirent {
169 unsigned long d_ino;
170 unsigned long d_offset;
171 unsigned short d_namlen;
172 char d_name[];
173 };
174
175 struct readdir_callback {
176 struct dir_context ctx;
177 struct old_linux_dirent __user * dirent;
178 int result;
179 };
180
fillonedir(struct dir_context * ctx,const char * name,int namlen,loff_t offset,u64 ino,unsigned int d_type)181 static bool fillonedir(struct dir_context *ctx, const char *name, int namlen,
182 loff_t offset, u64 ino, unsigned int d_type)
183 {
184 struct readdir_callback *buf =
185 container_of(ctx, struct readdir_callback, ctx);
186 struct old_linux_dirent __user * dirent;
187 unsigned long d_ino;
188
189 if (buf->result)
190 return false;
191 buf->result = verify_dirent_name(name, namlen);
192 if (buf->result)
193 return false;
194 d_ino = ino;
195 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
196 buf->result = -EOVERFLOW;
197 return false;
198 }
199 buf->result++;
200 dirent = buf->dirent;
201 if (!user_write_access_begin(dirent,
202 (unsigned long)(dirent->d_name + namlen + 1) -
203 (unsigned long)dirent))
204 goto efault;
205 unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
206 unsafe_put_user(offset, &dirent->d_offset, efault_end);
207 unsafe_put_user(namlen, &dirent->d_namlen, efault_end);
208 unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
209 user_write_access_end();
210 return true;
211 efault_end:
212 user_write_access_end();
213 efault:
214 buf->result = -EFAULT;
215 return false;
216 }
217
SYSCALL_DEFINE3(old_readdir,unsigned int,fd,struct old_linux_dirent __user *,dirent,unsigned int,count)218 SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
219 struct old_linux_dirent __user *, dirent, unsigned int, count)
220 {
221 int error;
222 struct fd f = fdget_pos(fd);
223 struct readdir_callback buf = {
224 .ctx.actor = fillonedir,
225 .dirent = dirent
226 };
227
228 if (!fd_file(f))
229 return -EBADF;
230
231 error = iterate_dir(fd_file(f), &buf.ctx);
232 if (buf.result)
233 error = buf.result;
234
235 fdput_pos(f);
236 return error;
237 }
238
239 #endif /* __ARCH_WANT_OLD_READDIR */
240
241 /*
242 * New, all-improved, singing, dancing, iBCS2-compliant getdents()
243 * interface.
244 */
245 struct linux_dirent {
246 unsigned long d_ino;
247 unsigned long d_off;
248 unsigned short d_reclen;
249 char d_name[];
250 };
251
252 struct getdents_callback {
253 struct dir_context ctx;
254 struct linux_dirent __user * current_dir;
255 int prev_reclen;
256 int count;
257 int error;
258 };
259
filldir(struct dir_context * ctx,const char * name,int namlen,loff_t offset,u64 ino,unsigned int d_type)260 static bool filldir(struct dir_context *ctx, const char *name, int namlen,
261 loff_t offset, u64 ino, unsigned int d_type)
262 {
263 struct linux_dirent __user *dirent, *prev;
264 struct getdents_callback *buf =
265 container_of(ctx, struct getdents_callback, ctx);
266 unsigned long d_ino;
267 int reclen = ALIGN(offsetof(struct linux_dirent, d_name) + namlen + 2,
268 sizeof(long));
269 int prev_reclen;
270
271 buf->error = verify_dirent_name(name, namlen);
272 if (unlikely(buf->error))
273 return false;
274 buf->error = -EINVAL; /* only used if we fail.. */
275 if (reclen > buf->count)
276 return false;
277 d_ino = ino;
278 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
279 buf->error = -EOVERFLOW;
280 return false;
281 }
282 prev_reclen = buf->prev_reclen;
283 if (prev_reclen && signal_pending(current))
284 return false;
285 dirent = buf->current_dir;
286 prev = (void __user *) dirent - prev_reclen;
287 if (!user_write_access_begin(prev, reclen + prev_reclen))
288 goto efault;
289
290 /* This might be 'dirent->d_off', but if so it will get overwritten */
291 unsafe_put_user(offset, &prev->d_off, efault_end);
292 unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
293 unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
294 unsafe_put_user(d_type, (char __user *) dirent + reclen - 1, efault_end);
295 unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
296 user_write_access_end();
297
298 buf->current_dir = (void __user *)dirent + reclen;
299 buf->prev_reclen = reclen;
300 buf->count -= reclen;
301 return true;
302 efault_end:
303 user_write_access_end();
304 efault:
305 buf->error = -EFAULT;
306 return false;
307 }
308
SYSCALL_DEFINE3(getdents,unsigned int,fd,struct linux_dirent __user *,dirent,unsigned int,count)309 SYSCALL_DEFINE3(getdents, unsigned int, fd,
310 struct linux_dirent __user *, dirent, unsigned int, count)
311 {
312 struct fd f;
313 struct getdents_callback buf = {
314 .ctx.actor = filldir,
315 .count = count,
316 .current_dir = dirent
317 };
318 int error;
319
320 f = fdget_pos(fd);
321 if (!fd_file(f))
322 return -EBADF;
323
324 error = iterate_dir(fd_file(f), &buf.ctx);
325 if (error >= 0)
326 error = buf.error;
327 if (buf.prev_reclen) {
328 struct linux_dirent __user * lastdirent;
329 lastdirent = (void __user *)buf.current_dir - buf.prev_reclen;
330
331 if (put_user(buf.ctx.pos, &lastdirent->d_off))
332 error = -EFAULT;
333 else
334 error = count - buf.count;
335 }
336 fdput_pos(f);
337 return error;
338 }
339
340 struct getdents_callback64 {
341 struct dir_context ctx;
342 struct linux_dirent64 __user * current_dir;
343 int prev_reclen;
344 int count;
345 int error;
346 };
347
filldir64(struct dir_context * ctx,const char * name,int namlen,loff_t offset,u64 ino,unsigned int d_type)348 static bool filldir64(struct dir_context *ctx, const char *name, int namlen,
349 loff_t offset, u64 ino, unsigned int d_type)
350 {
351 struct linux_dirent64 __user *dirent, *prev;
352 struct getdents_callback64 *buf =
353 container_of(ctx, struct getdents_callback64, ctx);
354 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
355 sizeof(u64));
356 int prev_reclen;
357
358 buf->error = verify_dirent_name(name, namlen);
359 if (unlikely(buf->error))
360 return false;
361 buf->error = -EINVAL; /* only used if we fail.. */
362 if (reclen > buf->count)
363 return false;
364 prev_reclen = buf->prev_reclen;
365 if (prev_reclen && signal_pending(current))
366 return false;
367 dirent = buf->current_dir;
368 prev = (void __user *)dirent - prev_reclen;
369 if (!user_write_access_begin(prev, reclen + prev_reclen))
370 goto efault;
371
372 /* This might be 'dirent->d_off', but if so it will get overwritten */
373 unsafe_put_user(offset, &prev->d_off, efault_end);
374 unsafe_put_user(ino, &dirent->d_ino, efault_end);
375 unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
376 unsafe_put_user(d_type, &dirent->d_type, efault_end);
377 unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
378 user_write_access_end();
379
380 buf->prev_reclen = reclen;
381 buf->current_dir = (void __user *)dirent + reclen;
382 buf->count -= reclen;
383 return true;
384
385 efault_end:
386 user_write_access_end();
387 efault:
388 buf->error = -EFAULT;
389 return false;
390 }
391
SYSCALL_DEFINE3(getdents64,unsigned int,fd,struct linux_dirent64 __user *,dirent,unsigned int,count)392 SYSCALL_DEFINE3(getdents64, unsigned int, fd,
393 struct linux_dirent64 __user *, dirent, unsigned int, count)
394 {
395 struct fd f;
396 struct getdents_callback64 buf = {
397 .ctx.actor = filldir64,
398 .count = count,
399 .current_dir = dirent
400 };
401 int error;
402
403 f = fdget_pos(fd);
404 if (!fd_file(f))
405 return -EBADF;
406
407 error = iterate_dir(fd_file(f), &buf.ctx);
408 if (error >= 0)
409 error = buf.error;
410 if (buf.prev_reclen) {
411 struct linux_dirent64 __user * lastdirent;
412 typeof(lastdirent->d_off) d_off = buf.ctx.pos;
413
414 lastdirent = (void __user *) buf.current_dir - buf.prev_reclen;
415 if (put_user(d_off, &lastdirent->d_off))
416 error = -EFAULT;
417 else
418 error = count - buf.count;
419 }
420 fdput_pos(f);
421 return error;
422 }
423
424 #ifdef CONFIG_COMPAT
425 struct compat_old_linux_dirent {
426 compat_ulong_t d_ino;
427 compat_ulong_t d_offset;
428 unsigned short d_namlen;
429 char d_name[];
430 };
431
432 struct compat_readdir_callback {
433 struct dir_context ctx;
434 struct compat_old_linux_dirent __user *dirent;
435 int result;
436 };
437
compat_fillonedir(struct dir_context * ctx,const char * name,int namlen,loff_t offset,u64 ino,unsigned int d_type)438 static bool compat_fillonedir(struct dir_context *ctx, const char *name,
439 int namlen, loff_t offset, u64 ino,
440 unsigned int d_type)
441 {
442 struct compat_readdir_callback *buf =
443 container_of(ctx, struct compat_readdir_callback, ctx);
444 struct compat_old_linux_dirent __user *dirent;
445 compat_ulong_t d_ino;
446
447 if (buf->result)
448 return false;
449 buf->result = verify_dirent_name(name, namlen);
450 if (buf->result)
451 return false;
452 d_ino = ino;
453 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
454 buf->result = -EOVERFLOW;
455 return false;
456 }
457 buf->result++;
458 dirent = buf->dirent;
459 if (!user_write_access_begin(dirent,
460 (unsigned long)(dirent->d_name + namlen + 1) -
461 (unsigned long)dirent))
462 goto efault;
463 unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
464 unsafe_put_user(offset, &dirent->d_offset, efault_end);
465 unsafe_put_user(namlen, &dirent->d_namlen, efault_end);
466 unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
467 user_write_access_end();
468 return true;
469 efault_end:
470 user_write_access_end();
471 efault:
472 buf->result = -EFAULT;
473 return false;
474 }
475
COMPAT_SYSCALL_DEFINE3(old_readdir,unsigned int,fd,struct compat_old_linux_dirent __user *,dirent,unsigned int,count)476 COMPAT_SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
477 struct compat_old_linux_dirent __user *, dirent, unsigned int, count)
478 {
479 int error;
480 struct fd f = fdget_pos(fd);
481 struct compat_readdir_callback buf = {
482 .ctx.actor = compat_fillonedir,
483 .dirent = dirent
484 };
485
486 if (!fd_file(f))
487 return -EBADF;
488
489 error = iterate_dir(fd_file(f), &buf.ctx);
490 if (buf.result)
491 error = buf.result;
492
493 fdput_pos(f);
494 return error;
495 }
496
497 struct compat_linux_dirent {
498 compat_ulong_t d_ino;
499 compat_ulong_t d_off;
500 unsigned short d_reclen;
501 char d_name[];
502 };
503
504 struct compat_getdents_callback {
505 struct dir_context ctx;
506 struct compat_linux_dirent __user *current_dir;
507 int prev_reclen;
508 int count;
509 int error;
510 };
511
compat_filldir(struct dir_context * ctx,const char * name,int namlen,loff_t offset,u64 ino,unsigned int d_type)512 static bool compat_filldir(struct dir_context *ctx, const char *name, int namlen,
513 loff_t offset, u64 ino, unsigned int d_type)
514 {
515 struct compat_linux_dirent __user *dirent, *prev;
516 struct compat_getdents_callback *buf =
517 container_of(ctx, struct compat_getdents_callback, ctx);
518 compat_ulong_t d_ino;
519 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
520 namlen + 2, sizeof(compat_long_t));
521 int prev_reclen;
522
523 buf->error = verify_dirent_name(name, namlen);
524 if (unlikely(buf->error))
525 return false;
526 buf->error = -EINVAL; /* only used if we fail.. */
527 if (reclen > buf->count)
528 return false;
529 d_ino = ino;
530 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
531 buf->error = -EOVERFLOW;
532 return false;
533 }
534 prev_reclen = buf->prev_reclen;
535 if (prev_reclen && signal_pending(current))
536 return false;
537 dirent = buf->current_dir;
538 prev = (void __user *) dirent - prev_reclen;
539 if (!user_write_access_begin(prev, reclen + prev_reclen))
540 goto efault;
541
542 unsafe_put_user(offset, &prev->d_off, efault_end);
543 unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
544 unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
545 unsafe_put_user(d_type, (char __user *) dirent + reclen - 1, efault_end);
546 unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
547 user_write_access_end();
548
549 buf->prev_reclen = reclen;
550 buf->current_dir = (void __user *)dirent + reclen;
551 buf->count -= reclen;
552 return true;
553 efault_end:
554 user_write_access_end();
555 efault:
556 buf->error = -EFAULT;
557 return false;
558 }
559
COMPAT_SYSCALL_DEFINE3(getdents,unsigned int,fd,struct compat_linux_dirent __user *,dirent,unsigned int,count)560 COMPAT_SYSCALL_DEFINE3(getdents, unsigned int, fd,
561 struct compat_linux_dirent __user *, dirent, unsigned int, count)
562 {
563 struct fd f;
564 struct compat_getdents_callback buf = {
565 .ctx.actor = compat_filldir,
566 .current_dir = dirent,
567 .count = count
568 };
569 int error;
570
571 f = fdget_pos(fd);
572 if (!fd_file(f))
573 return -EBADF;
574
575 error = iterate_dir(fd_file(f), &buf.ctx);
576 if (error >= 0)
577 error = buf.error;
578 if (buf.prev_reclen) {
579 struct compat_linux_dirent __user * lastdirent;
580 lastdirent = (void __user *)buf.current_dir - buf.prev_reclen;
581
582 if (put_user(buf.ctx.pos, &lastdirent->d_off))
583 error = -EFAULT;
584 else
585 error = count - buf.count;
586 }
587 fdput_pos(f);
588 return error;
589 }
590 #endif
591