readdir.c

  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 */
 31int 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}
 70EXPORT_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
 85int 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);
114out:
115	return res;
116}
117EXPORT_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 */
148static 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
168struct old_linux_dirent {
169	unsigned long	d_ino;
170	unsigned long	d_offset;
171	unsigned short	d_namlen;
172	char		d_name[];
173};
174
175struct readdir_callback {
176	struct dir_context ctx;
177	struct old_linux_dirent __user * dirent;
178	int result;
179};
180
181static 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;
211efault_end:
212	user_write_access_end();
213efault:
214	buf->result = -EFAULT;
215	return false;
216}
217
218SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
219		struct old_linux_dirent __user *, dirent, unsigned int, count)
220{
221	int error;
222	CLASS(fd_pos, f)(fd);
223	struct readdir_callback buf = {
224		.ctx.actor = fillonedir,
225		.ctx.count = 1, /* Hint to fs: just one entry. */
226		.dirent = dirent
227	};
228
229	if (fd_empty(f))
230		return -EBADF;
231
232	error = iterate_dir(fd_file(f), &buf.ctx);
233	if (buf.result)
234		error = buf.result;
235
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 */
245struct linux_dirent {
246	unsigned long	d_ino;
247	unsigned long	d_off;
248	unsigned short	d_reclen;
249	char		d_name[];
250};
251
252struct getdents_callback {
253	struct dir_context ctx;
254	struct linux_dirent __user * current_dir;
255	int prev_reclen;
256	int error;
257};
258
259static bool filldir(struct dir_context *ctx, const char *name, int namlen,
260		   loff_t offset, u64 ino, unsigned int d_type)
261{
262	struct linux_dirent __user *dirent, *prev;
263	struct getdents_callback *buf =
264		container_of(ctx, struct getdents_callback, ctx);
265	unsigned long d_ino;
266	int reclen = ALIGN(offsetof(struct linux_dirent, d_name) + namlen + 2,
267		sizeof(long));
268	int prev_reclen;
269	unsigned int flags = d_type;
270
271	BUILD_BUG_ON(FILLDIR_FLAG_NOINTR & S_DT_MASK);
272	d_type &= S_DT_MASK;
273
274	buf->error = verify_dirent_name(name, namlen);
275	if (unlikely(buf->error))
276		return false;
277	buf->error = -EINVAL;	/* only used if we fail.. */
278	if (reclen > ctx->count)
279		return false;
280	d_ino = ino;
281	if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
282		buf->error = -EOVERFLOW;
283		return false;
284	}
285	prev_reclen = buf->prev_reclen;
286	if (!(flags & FILLDIR_FLAG_NOINTR) && prev_reclen && signal_pending(current))
287		return false;
288	dirent = buf->current_dir;
289	prev = (void __user *) dirent - prev_reclen;
290	if (!user_write_access_begin(prev, reclen + prev_reclen))
291		goto efault;
292
293	/* This might be 'dirent->d_off', but if so it will get overwritten */
294	unsafe_put_user(offset, &prev->d_off, efault_end);
295	unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
296	unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
297	unsafe_put_user(d_type, (char __user *) dirent + reclen - 1, efault_end);
298	unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
299	user_write_access_end();
300
301	buf->current_dir = (void __user *)dirent + reclen;
302	buf->prev_reclen = reclen;
303	ctx->count -= reclen;
304	return true;
305efault_end:
306	user_write_access_end();
307efault:
308	buf->error = -EFAULT;
309	return false;
310}
311
312SYSCALL_DEFINE3(getdents, unsigned int, fd,
313		struct linux_dirent __user *, dirent, unsigned int, count)
314{
315	CLASS(fd_pos, f)(fd);
316	struct getdents_callback buf = {
317		.ctx.actor = filldir,
318		.ctx.count = count,
319		.ctx.dt_flags_mask = FILLDIR_FLAG_NOINTR,
320		.current_dir = dirent
321	};
322	int error;
323
324	if (fd_empty(f))
325		return -EBADF;
326
327	error = iterate_dir(fd_file(f), &buf.ctx);
328	if (error >= 0)
329		error = buf.error;
330	if (buf.prev_reclen) {
331		struct linux_dirent __user * lastdirent;
332		lastdirent = (void __user *)buf.current_dir - buf.prev_reclen;
333
334		if (put_user(buf.ctx.pos, &lastdirent->d_off))
335			error = -EFAULT;
336		else
337			error = count - buf.ctx.count;
338	}
339	return error;
340}
341
342struct getdents_callback64 {
343	struct dir_context ctx;
344	struct linux_dirent64 __user * current_dir;
345	int prev_reclen;
346	int error;
347};
348
349static bool filldir64(struct dir_context *ctx, const char *name, int namlen,
350		     loff_t offset, u64 ino, unsigned int d_type)
351{
352	struct linux_dirent64 __user *dirent, *prev;
353	struct getdents_callback64 *buf =
354		container_of(ctx, struct getdents_callback64, ctx);
355	int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
356		sizeof(u64));
357	int prev_reclen;
358	unsigned int flags = d_type;
359
360	BUILD_BUG_ON(FILLDIR_FLAG_NOINTR & S_DT_MASK);
361	d_type &= S_DT_MASK;
362
363	buf->error = verify_dirent_name(name, namlen);
364	if (unlikely(buf->error))
365		return false;
366	buf->error = -EINVAL;	/* only used if we fail.. */
367	if (reclen > ctx->count)
368		return false;
369	prev_reclen = buf->prev_reclen;
370	if (!(flags & FILLDIR_FLAG_NOINTR) && prev_reclen && signal_pending(current))
371		return false;
372	dirent = buf->current_dir;
373	prev = (void __user *)dirent - prev_reclen;
374	if (!user_write_access_begin(prev, reclen + prev_reclen))
375		goto efault;
376
377	/* This might be 'dirent->d_off', but if so it will get overwritten */
378	unsafe_put_user(offset, &prev->d_off, efault_end);
379	unsafe_put_user(ino, &dirent->d_ino, efault_end);
380	unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
381	unsafe_put_user(d_type, &dirent->d_type, efault_end);
382	unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
383	user_write_access_end();
384
385	buf->prev_reclen = reclen;
386	buf->current_dir = (void __user *)dirent + reclen;
387	ctx->count -= reclen;
388	return true;
389
390efault_end:
391	user_write_access_end();
392efault:
393	buf->error = -EFAULT;
394	return false;
395}
396
397SYSCALL_DEFINE3(getdents64, unsigned int, fd,
398		struct linux_dirent64 __user *, dirent, unsigned int, count)
399{
400	CLASS(fd_pos, f)(fd);
401	struct getdents_callback64 buf = {
402		.ctx.actor = filldir64,
403		.ctx.count = count,
404		.ctx.dt_flags_mask = FILLDIR_FLAG_NOINTR,
405		.current_dir = dirent
406	};
407	int error;
408
409	if (fd_empty(f))
410		return -EBADF;
411
412	error = iterate_dir(fd_file(f), &buf.ctx);
413	if (error >= 0)
414		error = buf.error;
415	if (buf.prev_reclen) {
416		struct linux_dirent64 __user * lastdirent;
417		typeof(lastdirent->d_off) d_off = buf.ctx.pos;
418
419		lastdirent = (void __user *) buf.current_dir - buf.prev_reclen;
420		if (put_user(d_off, &lastdirent->d_off))
421			error = -EFAULT;
422		else
423			error = count - buf.ctx.count;
424	}
425	return error;
426}
427
428#ifdef CONFIG_COMPAT
429struct compat_old_linux_dirent {
430	compat_ulong_t	d_ino;
431	compat_ulong_t	d_offset;
432	unsigned short	d_namlen;
433	char		d_name[];
434};
435
436struct compat_readdir_callback {
437	struct dir_context ctx;
438	struct compat_old_linux_dirent __user *dirent;
439	int result;
440};
441
442static bool compat_fillonedir(struct dir_context *ctx, const char *name,
443			     int namlen, loff_t offset, u64 ino,
444			     unsigned int d_type)
445{
446	struct compat_readdir_callback *buf =
447		container_of(ctx, struct compat_readdir_callback, ctx);
448	struct compat_old_linux_dirent __user *dirent;
449	compat_ulong_t d_ino;
450
451	if (buf->result)
452		return false;
453	buf->result = verify_dirent_name(name, namlen);
454	if (buf->result)
455		return false;
456	d_ino = ino;
457	if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
458		buf->result = -EOVERFLOW;
459		return false;
460	}
461	buf->result++;
462	dirent = buf->dirent;
463	if (!user_write_access_begin(dirent,
464			(unsigned long)(dirent->d_name + namlen + 1) -
465				(unsigned long)dirent))
466		goto efault;
467	unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
468	unsafe_put_user(offset, &dirent->d_offset, efault_end);
469	unsafe_put_user(namlen, &dirent->d_namlen, efault_end);
470	unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
471	user_write_access_end();
472	return true;
473efault_end:
474	user_write_access_end();
475efault:
476	buf->result = -EFAULT;
477	return false;
478}
479
480COMPAT_SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
481		struct compat_old_linux_dirent __user *, dirent, unsigned int, count)
482{
483	int error;
484	CLASS(fd_pos, f)(fd);
485	struct compat_readdir_callback buf = {
486		.ctx.actor = compat_fillonedir,
487		.ctx.count = 1, /* Hint to fs: just one entry. */
488		.dirent = dirent
489	};
490
491	if (fd_empty(f))
492		return -EBADF;
493
494	error = iterate_dir(fd_file(f), &buf.ctx);
495	if (buf.result)
496		error = buf.result;
497
498	return error;
499}
500
501struct compat_linux_dirent {
502	compat_ulong_t	d_ino;
503	compat_ulong_t	d_off;
504	unsigned short	d_reclen;
505	char		d_name[];
506};
507
508struct compat_getdents_callback {
509	struct dir_context ctx;
510	struct compat_linux_dirent __user *current_dir;
511	int prev_reclen;
512	int error;
513};
514
515static bool compat_filldir(struct dir_context *ctx, const char *name, int namlen,
516		loff_t offset, u64 ino, unsigned int d_type)
517{
518	struct compat_linux_dirent __user *dirent, *prev;
519	struct compat_getdents_callback *buf =
520		container_of(ctx, struct compat_getdents_callback, ctx);
521	compat_ulong_t d_ino;
522	int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
523		namlen + 2, sizeof(compat_long_t));
524	int prev_reclen;
525	unsigned int flags = d_type;
526
527	BUILD_BUG_ON(FILLDIR_FLAG_NOINTR & S_DT_MASK);
528	d_type &= S_DT_MASK;
529
530	buf->error = verify_dirent_name(name, namlen);
531	if (unlikely(buf->error))
532		return false;
533	buf->error = -EINVAL;	/* only used if we fail.. */
534	if (reclen > ctx->count)
535		return false;
536	d_ino = ino;
537	if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
538		buf->error = -EOVERFLOW;
539		return false;
540	}
541	prev_reclen = buf->prev_reclen;
542	if (!(flags & FILLDIR_FLAG_NOINTR) && prev_reclen && signal_pending(current))
543		return false;
544	dirent = buf->current_dir;
545	prev = (void __user *) dirent - prev_reclen;
546	if (!user_write_access_begin(prev, reclen + prev_reclen))
547		goto efault;
548
549	unsafe_put_user(offset, &prev->d_off, efault_end);
550	unsafe_put_user(d_ino, &dirent->d_ino, efault_end);
551	unsafe_put_user(reclen, &dirent->d_reclen, efault_end);
552	unsafe_put_user(d_type, (char __user *) dirent + reclen - 1, efault_end);
553	unsafe_copy_dirent_name(dirent->d_name, name, namlen, efault_end);
554	user_write_access_end();
555
556	buf->prev_reclen = reclen;
557	buf->current_dir = (void __user *)dirent + reclen;
558	ctx->count -= reclen;
559	return true;
560efault_end:
561	user_write_access_end();
562efault:
563	buf->error = -EFAULT;
564	return false;
565}
566
567COMPAT_SYSCALL_DEFINE3(getdents, unsigned int, fd,
568		struct compat_linux_dirent __user *, dirent, unsigned int, count)
569{
570	CLASS(fd_pos, f)(fd);
571	struct compat_getdents_callback buf = {
572		.ctx.actor = compat_filldir,
573		.ctx.count = count,
574		.ctx.dt_flags_mask = FILLDIR_FLAG_NOINTR,
575		.current_dir = dirent,
576	};
577	int error;
578
579	if (fd_empty(f))
580		return -EBADF;
581
582	error = iterate_dir(fd_file(f), &buf.ctx);
583	if (error >= 0)
584		error = buf.error;
585	if (buf.prev_reclen) {
586		struct compat_linux_dirent __user * lastdirent;
587		lastdirent = (void __user *)buf.current_dir - buf.prev_reclen;
588
589		if (put_user(buf.ctx.pos, &lastdirent->d_off))
590			error = -EFAULT;
591		else
592			error = count - buf.ctx.count;
593	}
594	return error;
595}
596#endif