1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/dir.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/dir.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * ext4 directory handling functions
17 *
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
20 *
21 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
22 *
23 */
24
25#include <linux/fs.h>
26#include <linux/buffer_head.h>
27#include <linux/slab.h>
28#include <linux/iversion.h>
29#include <linux/unicode.h>
30#include "ext4.h"
31#include "xattr.h"
32
33static int ext4_dx_readdir(struct file *, struct dir_context *);
34
35/**
36 * is_dx_dir() - check if a directory is using htree indexing
37 * @inode: directory inode
38 *
39 * Check if the given dir-inode refers to an htree-indexed directory
40 * (or a directory which could potentially get converted to use htree
41 * indexing).
42 *
43 * Return 1 if it is a dx dir, 0 if not
44 */
45static int is_dx_dir(struct inode *inode)
46{
47 struct super_block *sb = inode->i_sb;
48
49 if (ext4_has_feature_dir_index(sb: inode->i_sb) &&
50 ((ext4_test_inode_flag(inode, bit: EXT4_INODE_INDEX)) ||
51 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
52 ext4_has_inline_data(inode)))
53 return 1;
54
55 return 0;
56}
57
58static bool is_fake_dir_entry(struct ext4_dir_entry_2 *de)
59{
60 /* Check if . or .. , or skip if namelen is 0 */
61 if ((de->name_len > 0) && (de->name_len <= 2) && (de->name[0] == '.') &&
62 (de->name[1] == '.' || de->name[1] == '\0'))
63 return true;
64 /* Check if this is a csum entry */
65 if (de->file_type == EXT4_FT_DIR_CSUM)
66 return true;
67 return false;
68}
69
70/*
71 * Return 0 if the directory entry is OK, and 1 if there is a problem
72 *
73 * Note: this is the opposite of what ext2 and ext3 historically returned...
74 *
75 * bh passed here can be an inode block or a dir data block, depending
76 * on the inode inline data flag.
77 */
78int __ext4_check_dir_entry(const char *function, unsigned int line,
79 struct inode *dir, struct file *filp,
80 struct ext4_dir_entry_2 *de,
81 struct buffer_head *bh, char *buf, int size,
82 unsigned int offset)
83{
84 const char *error_msg = NULL;
85 const int rlen = ext4_rec_len_from_disk(dlen: de->rec_len,
86 blocksize: dir->i_sb->s_blocksize);
87 const int next_offset = ((char *) de - buf) + rlen;
88 bool fake = is_fake_dir_entry(de);
89 bool has_csum = ext4_has_feature_metadata_csum(sb: dir->i_sb);
90
91 if (unlikely(rlen < ext4_dir_rec_len(1, fake ? NULL : dir)))
92 error_msg = "rec_len is smaller than minimal";
93 else if (unlikely(rlen % 4 != 0))
94 error_msg = "rec_len % 4 != 0";
95 else if (unlikely(rlen < ext4_dir_rec_len(de->name_len,
96 fake ? NULL : dir)))
97 error_msg = "rec_len is too small for name_len";
98 else if (unlikely(next_offset > size))
99 error_msg = "directory entry overrun";
100 else if (unlikely(next_offset > size - ext4_dir_rec_len(1,
101 has_csum ? NULL : dir) &&
102 next_offset != size))
103 error_msg = "directory entry too close to block end";
104 else if (unlikely(le32_to_cpu(de->inode) >
105 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
106 error_msg = "inode out of bounds";
107 else if (unlikely(next_offset == size && de->name_len == 1 &&
108 de->name[0] == '.'))
109 error_msg = "'.' directory cannot be the last in data block";
110 else
111 return 0;
112
113 if (filp)
114 ext4_error_file(filp, function, line, bh->b_blocknr,
115 "bad entry in directory: %s - offset=%u, "
116 "inode=%u, rec_len=%d, size=%d fake=%d",
117 error_msg, offset, le32_to_cpu(de->inode),
118 rlen, size, fake);
119 else
120 ext4_error_inode(dir, function, line, bh->b_blocknr,
121 "bad entry in directory: %s - offset=%u, "
122 "inode=%u, rec_len=%d, size=%d fake=%d",
123 error_msg, offset, le32_to_cpu(de->inode),
124 rlen, size, fake);
125
126 return 1;
127}
128
129static int ext4_readdir(struct file *file, struct dir_context *ctx)
130{
131 unsigned int offset;
132 int i;
133 struct ext4_dir_entry_2 *de;
134 int err;
135 struct inode *inode = file_inode(f: file);
136 struct super_block *sb = inode->i_sb;
137 struct buffer_head *bh = NULL;
138 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
139 struct dir_private_info *info = file->private_data;
140
141 err = fscrypt_prepare_readdir(dir: inode);
142 if (err)
143 return err;
144
145 if (is_dx_dir(inode)) {
146 err = ext4_dx_readdir(file, ctx);
147 if (err != ERR_BAD_DX_DIR)
148 return err;
149
150 /* Can we just clear INDEX flag to ignore htree information? */
151 if (!ext4_has_feature_metadata_csum(sb)) {
152 /*
153 * We don't set the inode dirty flag since it's not
154 * critical that it gets flushed back to the disk.
155 */
156 ext4_clear_inode_flag(inode, bit: EXT4_INODE_INDEX);
157 }
158 }
159
160 if (ext4_has_inline_data(inode)) {
161 int has_inline_data = 1;
162 err = ext4_read_inline_dir(filp: file, ctx,
163 has_inline_data: &has_inline_data);
164 if (has_inline_data)
165 return err;
166 }
167
168 if (IS_ENCRYPTED(inode)) {
169 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, crypto_str: &fstr);
170 if (err < 0)
171 return err;
172 }
173
174 while (ctx->pos < inode->i_size) {
175 struct ext4_map_blocks map;
176
177 if (fatal_signal_pending(current)) {
178 err = -ERESTARTSYS;
179 goto errout;
180 }
181 cond_resched();
182 offset = ctx->pos & (sb->s_blocksize - 1);
183 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
184 map.m_len = 1;
185 err = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
186 if (err == 0) {
187 /* m_len should never be zero but let's avoid
188 * an infinite loop if it somehow is */
189 if (map.m_len == 0)
190 map.m_len = 1;
191 ctx->pos += map.m_len * sb->s_blocksize;
192 continue;
193 }
194 if (err > 0) {
195 pgoff_t index = map.m_pblk >>
196 (PAGE_SHIFT - inode->i_blkbits);
197 if (!ra_has_index(ra: &file->f_ra, index))
198 page_cache_sync_readahead(
199 mapping: sb->s_bdev->bd_mapping,
200 ra: &file->f_ra, file,
201 index, req_count: 1);
202 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
203 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
204 if (IS_ERR(ptr: bh)) {
205 err = PTR_ERR(ptr: bh);
206 bh = NULL;
207 goto errout;
208 }
209 }
210
211 if (!bh) {
212 /* corrupt size? Maybe no more blocks to read */
213 if (ctx->pos > inode->i_blocks << 9)
214 break;
215 ctx->pos += sb->s_blocksize - offset;
216 continue;
217 }
218
219 /* Check the checksum */
220 if (!buffer_verified(bh) &&
221 !ext4_dirblock_csum_verify(inode, bh)) {
222 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
223 "at offset %llu",
224 (unsigned long long)ctx->pos);
225 ctx->pos += sb->s_blocksize - offset;
226 brelse(bh);
227 bh = NULL;
228 continue;
229 }
230 set_buffer_verified(bh);
231
232 /* If the dir block has changed since the last call to
233 * readdir(2), then we might be pointing to an invalid
234 * dirent right now. Scan from the start of the block
235 * to make sure. */
236 if (!inode_eq_iversion(inode, old: info->cookie)) {
237 for (i = 0; i < sb->s_blocksize && i < offset; ) {
238 de = (struct ext4_dir_entry_2 *)
239 (bh->b_data + i);
240 /* It's too expensive to do a full
241 * dirent test each time round this
242 * loop, but we do have to test at
243 * least that it is non-zero. A
244 * failure will be detected in the
245 * dirent test below. */
246 if (ext4_rec_len_from_disk(dlen: de->rec_len,
247 blocksize: sb->s_blocksize) < ext4_dir_rec_len(name_len: 1,
248 dir: inode))
249 break;
250 i += ext4_rec_len_from_disk(dlen: de->rec_len,
251 blocksize: sb->s_blocksize);
252 }
253 offset = i;
254 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
255 | offset;
256 info->cookie = inode_query_iversion(inode);
257 }
258
259 while (ctx->pos < inode->i_size
260 && offset < sb->s_blocksize) {
261 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
262 if (ext4_check_dir_entry(inode, file, de, bh,
263 bh->b_data, bh->b_size,
264 offset)) {
265 /*
266 * On error, skip to the next block
267 */
268 ctx->pos = (ctx->pos |
269 (sb->s_blocksize - 1)) + 1;
270 break;
271 }
272 offset += ext4_rec_len_from_disk(dlen: de->rec_len,
273 blocksize: sb->s_blocksize);
274 if (le32_to_cpu(de->inode)) {
275 if (!IS_ENCRYPTED(inode)) {
276 if (!dir_emit(ctx, name: de->name,
277 namelen: de->name_len,
278 le32_to_cpu(de->inode),
279 type: get_dtype(sb, filetype: de->file_type)))
280 goto done;
281 } else {
282 int save_len = fstr.len;
283 struct fscrypt_str de_name =
284 FSTR_INIT(de->name,
285 de->name_len);
286 u32 hash;
287 u32 minor_hash;
288
289 if (IS_CASEFOLDED(inode)) {
290 hash = EXT4_DIRENT_HASH(de);
291 minor_hash = EXT4_DIRENT_MINOR_HASH(de);
292 } else {
293 hash = 0;
294 minor_hash = 0;
295 }
296
297 /* Directory is encrypted */
298 err = fscrypt_fname_disk_to_usr(inode,
299 hash, minor_hash, iname: &de_name, oname: &fstr);
300 de_name = fstr;
301 fstr.len = save_len;
302 if (err)
303 goto errout;
304 if (!dir_emit(ctx,
305 name: de_name.name, namelen: de_name.len,
306 le32_to_cpu(de->inode),
307 type: get_dtype(sb, filetype: de->file_type)))
308 goto done;
309 }
310 }
311 ctx->pos += ext4_rec_len_from_disk(dlen: de->rec_len,
312 blocksize: sb->s_blocksize);
313 }
314 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
315 goto done;
316 brelse(bh);
317 bh = NULL;
318 }
319done:
320 err = 0;
321errout:
322 fscrypt_fname_free_buffer(crypto_str: &fstr);
323 brelse(bh);
324 return err;
325}
326
327static inline int is_32bit_api(void)
328{
329#ifdef CONFIG_COMPAT
330 return in_compat_syscall();
331#else
332 return (BITS_PER_LONG == 32);
333#endif
334}
335
336/*
337 * These functions convert from the major/minor hash to an f_pos
338 * value for dx directories
339 *
340 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
341 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
342 * directly on both 32-bit and 64-bit nodes, under such case, neither
343 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
344 */
345static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
346{
347 if ((filp->f_mode & FMODE_32BITHASH) ||
348 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
349 return major >> 1;
350 else
351 return ((__u64)(major >> 1) << 32) | (__u64)minor;
352}
353
354static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
355{
356 if ((filp->f_mode & FMODE_32BITHASH) ||
357 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
358 return (pos << 1) & 0xffffffff;
359 else
360 return ((pos >> 32) << 1) & 0xffffffff;
361}
362
363static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
364{
365 if ((filp->f_mode & FMODE_32BITHASH) ||
366 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
367 return 0;
368 else
369 return pos & 0xffffffff;
370}
371
372/*
373 * Return 32- or 64-bit end-of-file for dx directories
374 */
375static inline loff_t ext4_get_htree_eof(struct file *filp)
376{
377 if ((filp->f_mode & FMODE_32BITHASH) ||
378 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
379 return EXT4_HTREE_EOF_32BIT;
380 else
381 return EXT4_HTREE_EOF_64BIT;
382}
383
384
385/*
386 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
387 * directories, where the "offset" is in terms of the filename hash
388 * value instead of the byte offset.
389 *
390 * Because we may return a 64-bit hash that is well beyond offset limits,
391 * we need to pass the max hash as the maximum allowable offset in
392 * the htree directory case.
393 *
394 * For non-htree, ext4_llseek already chooses the proper max offset.
395 */
396static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
397{
398 struct inode *inode = file->f_mapping->host;
399 struct dir_private_info *info = file->private_data;
400 int dx_dir = is_dx_dir(inode);
401 loff_t ret, htree_max = ext4_get_htree_eof(filp: file);
402
403 if (likely(dx_dir))
404 ret = generic_file_llseek_size(file, offset, whence,
405 maxsize: htree_max, eof: htree_max);
406 else
407 ret = ext4_llseek(file, offset, origin: whence);
408 info->cookie = inode_peek_iversion(inode) - 1;
409 return ret;
410}
411
412/*
413 * This structure holds the nodes of the red-black tree used to store
414 * the directory entry in hash order.
415 */
416struct fname {
417 __u32 hash;
418 __u32 minor_hash;
419 struct rb_node rb_hash;
420 struct fname *next;
421 __u32 inode;
422 __u8 name_len;
423 __u8 file_type;
424 char name[] __counted_by(name_len);
425};
426
427/*
428 * This function implements a non-recursive way of freeing all of the
429 * nodes in the red-black tree.
430 */
431static void free_rb_tree_fname(struct rb_root *root)
432{
433 struct fname *fname, *next;
434
435 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
436 while (fname) {
437 struct fname *old = fname;
438 fname = fname->next;
439 kfree(objp: old);
440 }
441
442 *root = RB_ROOT;
443}
444
445static void ext4_htree_init_dir_info(struct file *filp, loff_t pos)
446{
447 struct dir_private_info *p = filp->private_data;
448
449 if (is_dx_dir(inode: file_inode(f: filp)) && !p->initialized) {
450 p->curr_hash = pos2maj_hash(filp, pos);
451 p->curr_minor_hash = pos2min_hash(filp, pos);
452 p->initialized = true;
453 }
454}
455
456void ext4_htree_free_dir_info(struct dir_private_info *p)
457{
458 free_rb_tree_fname(root: &p->root);
459 kfree(objp: p);
460}
461
462/*
463 * Given a directory entry, enter it into the fname rb tree.
464 *
465 * When filename encryption is enabled, the dirent will hold the
466 * encrypted filename, while the htree will hold decrypted filename.
467 * The decrypted filename is passed in via ent_name. parameter.
468 */
469int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
470 __u32 minor_hash,
471 struct ext4_dir_entry_2 *dirent,
472 struct fscrypt_str *ent_name)
473{
474 struct rb_node **p, *parent = NULL;
475 struct fname *fname, *new_fn;
476 struct dir_private_info *info;
477
478 info = dir_file->private_data;
479 p = &info->root.rb_node;
480
481 /* Create and allocate the fname structure */
482 new_fn = kzalloc(struct_size(new_fn, name, ent_name->len + 1),
483 GFP_KERNEL);
484 if (!new_fn)
485 return -ENOMEM;
486 new_fn->hash = hash;
487 new_fn->minor_hash = minor_hash;
488 new_fn->inode = le32_to_cpu(dirent->inode);
489 new_fn->name_len = ent_name->len;
490 new_fn->file_type = dirent->file_type;
491 memcpy(to: new_fn->name, from: ent_name->name, len: ent_name->len);
492
493 while (*p) {
494 parent = *p;
495 fname = rb_entry(parent, struct fname, rb_hash);
496
497 /*
498 * If the hash and minor hash match up, then we put
499 * them on a linked list. This rarely happens...
500 */
501 if ((new_fn->hash == fname->hash) &&
502 (new_fn->minor_hash == fname->minor_hash)) {
503 new_fn->next = fname->next;
504 fname->next = new_fn;
505 return 0;
506 }
507
508 if (new_fn->hash < fname->hash)
509 p = &(*p)->rb_left;
510 else if (new_fn->hash > fname->hash)
511 p = &(*p)->rb_right;
512 else if (new_fn->minor_hash < fname->minor_hash)
513 p = &(*p)->rb_left;
514 else /* if (new_fn->minor_hash > fname->minor_hash) */
515 p = &(*p)->rb_right;
516 }
517
518 rb_link_node(node: &new_fn->rb_hash, parent, rb_link: p);
519 rb_insert_color(&new_fn->rb_hash, &info->root);
520 return 0;
521}
522
523
524
525/*
526 * This is a helper function for ext4_dx_readdir. It calls filldir
527 * for all entries on the fname linked list. (Normally there is only
528 * one entry on the linked list, unless there are 62 bit hash collisions.)
529 */
530static int call_filldir(struct file *file, struct dir_context *ctx,
531 struct fname *fname)
532{
533 struct dir_private_info *info = file->private_data;
534 struct inode *inode = file_inode(f: file);
535 struct super_block *sb = inode->i_sb;
536
537 if (!fname) {
538 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
539 "called with null fname?!?", __func__, __LINE__,
540 inode->i_ino, current->comm);
541 return 0;
542 }
543 ctx->pos = hash2pos(filp: file, major: fname->hash, minor: fname->minor_hash);
544 while (fname) {
545 if (!dir_emit(ctx, name: fname->name,
546 namelen: fname->name_len,
547 ino: fname->inode,
548 type: get_dtype(sb, filetype: fname->file_type))) {
549 info->extra_fname = fname;
550 return 1;
551 }
552 fname = fname->next;
553 }
554 return 0;
555}
556
557static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
558{
559 struct dir_private_info *info = file->private_data;
560 struct inode *inode = file_inode(f: file);
561 struct fname *fname;
562 int ret = 0;
563
564 ext4_htree_init_dir_info(filp: file, pos: ctx->pos);
565
566 if (ctx->pos == ext4_get_htree_eof(filp: file))
567 return 0; /* EOF */
568
569 /* Some one has messed with f_pos; reset the world */
570 if (info->last_pos != ctx->pos) {
571 free_rb_tree_fname(root: &info->root);
572 info->curr_node = NULL;
573 info->extra_fname = NULL;
574 info->curr_hash = pos2maj_hash(filp: file, pos: ctx->pos);
575 info->curr_minor_hash = pos2min_hash(filp: file, pos: ctx->pos);
576 }
577
578 /*
579 * If there are any leftover names on the hash collision
580 * chain, return them first.
581 */
582 if (info->extra_fname) {
583 if (call_filldir(file, ctx, fname: info->extra_fname))
584 goto finished;
585 info->extra_fname = NULL;
586 goto next_node;
587 } else if (!info->curr_node)
588 info->curr_node = rb_first(&info->root);
589
590 while (1) {
591 /*
592 * Fill the rbtree if we have no more entries,
593 * or the inode has changed since we last read in the
594 * cached entries.
595 */
596 if ((!info->curr_node) ||
597 !inode_eq_iversion(inode, old: info->cookie)) {
598 info->curr_node = NULL;
599 free_rb_tree_fname(root: &info->root);
600 info->cookie = inode_query_iversion(inode);
601 ret = ext4_htree_fill_tree(dir_file: file, start_hash: info->curr_hash,
602 start_minor_hash: info->curr_minor_hash,
603 next_hash: &info->next_hash);
604 if (ret < 0)
605 goto finished;
606 if (ret == 0) {
607 ctx->pos = ext4_get_htree_eof(filp: file);
608 break;
609 }
610 info->curr_node = rb_first(&info->root);
611 }
612
613 fname = rb_entry(info->curr_node, struct fname, rb_hash);
614 info->curr_hash = fname->hash;
615 info->curr_minor_hash = fname->minor_hash;
616 if (call_filldir(file, ctx, fname))
617 break;
618 next_node:
619 info->curr_node = rb_next(info->curr_node);
620 if (info->curr_node) {
621 fname = rb_entry(info->curr_node, struct fname,
622 rb_hash);
623 info->curr_hash = fname->hash;
624 info->curr_minor_hash = fname->minor_hash;
625 } else {
626 if (info->next_hash == ~0) {
627 ctx->pos = ext4_get_htree_eof(filp: file);
628 break;
629 }
630 info->curr_hash = info->next_hash;
631 info->curr_minor_hash = 0;
632 }
633 }
634finished:
635 info->last_pos = ctx->pos;
636 return ret < 0 ? ret : 0;
637}
638
639static int ext4_release_dir(struct inode *inode, struct file *filp)
640{
641 if (filp->private_data)
642 ext4_htree_free_dir_info(p: filp->private_data);
643
644 return 0;
645}
646
647int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
648 int buf_size)
649{
650 struct ext4_dir_entry_2 *de;
651 int rlen;
652 unsigned int offset = 0;
653 char *top;
654
655 de = buf;
656 top = buf + buf_size;
657 while ((char *) de < top) {
658 if (ext4_check_dir_entry(dir, NULL, de, bh,
659 buf, buf_size, offset))
660 return -EFSCORRUPTED;
661 rlen = ext4_rec_len_from_disk(dlen: de->rec_len, blocksize: buf_size);
662 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
663 offset += rlen;
664 }
665 if ((char *) de > top)
666 return -EFSCORRUPTED;
667
668 return 0;
669}
670
671static int ext4_dir_open(struct inode *inode, struct file *file)
672{
673 struct dir_private_info *info;
674
675 info = kzalloc(sizeof(*info), GFP_KERNEL);
676 if (!info)
677 return -ENOMEM;
678 file->private_data = info;
679 return 0;
680}
681
682const struct file_operations ext4_dir_operations = {
683 .open = ext4_dir_open,
684 .llseek = ext4_dir_llseek,
685 .read = generic_read_dir,
686 .iterate_shared = ext4_readdir,
687 .unlocked_ioctl = ext4_ioctl,
688#ifdef CONFIG_COMPAT
689 .compat_ioctl = ext4_compat_ioctl,
690#endif
691 .fsync = ext4_sync_file,
692 .release = ext4_release_dir,
693};
694