projects/linux·blob master · refs

710 lines · 18.6 KiB

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 ([email protected])
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 ([email protected]), 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/filelock.h>
28#include <linux/slab.h>
29#include <linux/iversion.h>
30#include <linux/unicode.h>
31#include "ext4.h"
32#include "xattr.h"
33
34static int ext4_dx_readdir(struct file *, struct dir_context *);
35
36/**
37 * is_dx_dir() - check if a directory is using htree indexing
38 * @inode: directory inode
39 *
40 * Check if the given dir-inode refers to an htree-indexed directory
41 * (or a directory which could potentially get converted to use htree
42 * indexing).
43 *
44 * Return 1 if it is a dx dir, 0 if not
45 */
46static int is_dx_dir(struct inode *inode)
47{
48 struct super_block *sb = inode->i_sb;
49
50 if (ext4_has_feature_dir_index(inode->i_sb) &&
51 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
52 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
53 ext4_has_inline_data(inode)))
54 return 1;
55
56 return 0;
57}
58
59static bool is_fake_dir_entry(struct ext4_dir_entry_2 *de)
60{
61 /* Check if . or .. , or skip if namelen is 0 */
62 if ((de->name_len > 0) && (de->name_len <= 2) && (de->name[0] == '.') &&
63 (de->name[1] == '.' || de->name[1] == '\0'))
64 return true;
65 /* Check if this is a csum entry */
66 if (de->file_type == EXT4_FT_DIR_CSUM)
67 return true;
68 return false;
69}
70
71/*
72 * Return 0 if the directory entry is OK, and 1 if there is a problem
73 *
74 * Note: this is the opposite of what ext2 and ext3 historically returned...
75 *
76 * bh passed here can be an inode block or a dir data block, depending
77 * on the inode inline data flag.
78 */
79int __ext4_check_dir_entry(const char *function, unsigned int line,
80 struct inode *dir, struct file *filp,
81 struct ext4_dir_entry_2 *de,
82 struct buffer_head *bh, char *buf, int size,
83 unsigned int offset)
84{
85 const char *error_msg = NULL;
86 const int rlen = ext4_rec_len_from_disk(de->rec_len,
87 dir->i_sb->s_blocksize);
88 const int next_offset = ((char *) de - buf) + rlen;
89 bool fake = is_fake_dir_entry(de);
90 bool has_csum = ext4_has_feature_metadata_csum(dir->i_sb);
91
92 if (unlikely(rlen < ext4_dir_rec_len(1, fake ? NULL : dir)))
93 error_msg = "rec_len is smaller than minimal";
94 else if (unlikely(rlen % 4 != 0))
95 error_msg = "rec_len % 4 != 0";
96 else if (unlikely(rlen < ext4_dir_rec_len(de->name_len,
97 fake ? NULL : dir)))
98 error_msg = "rec_len is too small for name_len";
99 else if (unlikely(next_offset > size))
100 error_msg = "directory entry overrun";
101 else if (unlikely(next_offset > size - ext4_dir_rec_len(1,
102 has_csum ? NULL : dir) &&
103 next_offset != size))
104 error_msg = "directory entry too close to block end";
105 else if (unlikely(le32_to_cpu(de->inode) >
106 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
107 error_msg = "inode out of bounds";
108 else if (unlikely(next_offset == size && de->name_len == 1 &&
109 de->name[0] == '.'))
110 error_msg = "'.' directory cannot be the last in data block";
111 else
112 return 0;
113
114 if (filp)
115 ext4_error_file(filp, function, line, bh->b_blocknr,
116 "bad entry in directory: %s - offset=%u, "
117 "inode=%u, rec_len=%d, size=%d fake=%d",
118 error_msg, offset, le32_to_cpu(de->inode),
119 rlen, size, fake);
120 else
121 ext4_error_inode(dir, function, line, bh->b_blocknr,
122 "bad entry in directory: %s - offset=%u, "
123 "inode=%u, rec_len=%d, size=%d fake=%d",
124 error_msg, offset, le32_to_cpu(de->inode),
125 rlen, size, fake);
126
127 return 1;
128}
129
130static int ext4_readdir(struct file *file, struct dir_context *ctx)
131{
132 unsigned int offset;
133 int i;
134 struct ext4_dir_entry_2 *de;
135 int err;
136 struct inode *inode = file_inode(file);
137 struct super_block *sb = inode->i_sb;
138 struct buffer_head *bh = NULL;
139 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
140 struct dir_private_info *info = file->private_data;
141 bool has_csum = ext4_has_feature_metadata_csum(sb);
142
143 err = fscrypt_prepare_readdir(inode);
144 if (err)
145 return err;
146
147 if (is_dx_dir(inode)) {
148 err = ext4_dx_readdir(file, ctx);
149 if (err != ERR_BAD_DX_DIR)
150 return err;
151
152 /* Can we just clear INDEX flag to ignore htree information? */
153 if (!has_csum) {
154 /*
155 * We don't set the inode dirty flag since it's not
156 * critical that it gets flushed back to the disk.
157 */
158 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
159 }
160 }
161
162 if (ext4_has_inline_data(inode)) {
163 int has_inline_data = 1;
164 err = ext4_read_inline_dir(file, ctx,
165 &has_inline_data);
166 if (has_inline_data)
167 return err;
168 }
169
170 if (IS_ENCRYPTED(inode)) {
171 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr);
172 if (err < 0)
173 return err;
174 }
175
176 while (ctx->pos < inode->i_size) {
177 struct ext4_map_blocks map;
178
179 if (fatal_signal_pending(current)) {
180 err = -ERESTARTSYS;
181 goto errout;
182 }
183 cond_resched();
184 offset = ctx->pos & (sb->s_blocksize - 1);
185 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
186 map.m_len = 1;
187 err = ext4_map_blocks(NULL, inode, &map, 0);
188 if (err == 0) {
189 /* m_len should never be zero but let's avoid
190 * an infinite loop if it somehow is */
191 if (map.m_len == 0)
192 map.m_len = 1;
193 ctx->pos += map.m_len * sb->s_blocksize;
194 continue;
195 }
196 if (err > 0) {
197 pgoff_t index = map.m_pblk << inode->i_blkbits >>
198 PAGE_SHIFT;
199 if (!ra_has_index(&file->f_ra, index))
200 page_cache_sync_readahead(
201 sb->s_bdev->bd_mapping,
202 &file->f_ra, file, index,
203 1 << EXT4_SB(sb)->s_min_folio_order);
204 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
205 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
206 if (IS_ERR(bh)) {
207 err = PTR_ERR(bh);
208 bh = NULL;
209 goto errout;
210 }
211 }
212
213 if (!bh) {
214 /* corrupt size? Maybe no more blocks to read */
215 if (ctx->pos > inode->i_blocks << 9)
216 break;
217 ctx->pos += sb->s_blocksize - offset;
218 continue;
219 }
220
221 /* Check the checksum */
222 if (!buffer_verified(bh) &&
223 !ext4_dirblock_csum_verify(inode, bh)) {
224 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
225 "at offset %llu",
226 (unsigned long long)ctx->pos);
227 ctx->pos += sb->s_blocksize - offset;
228 brelse(bh);
229 bh = NULL;
230 continue;
231 }
232 set_buffer_verified(bh);
233
234 /* If the dir block has changed since the last call to
235 * readdir(2), then we might be pointing to an invalid
236 * dirent right now. Scan from the start of the block
237 * to make sure. */
238 if (!inode_eq_iversion(inode, info->cookie)) {
239 for (i = 0;
240 i <= sb->s_blocksize -
241 ext4_dir_rec_len(1, has_csum ? NULL : inode) &&
242 i < offset;) {
243 de = (struct ext4_dir_entry_2 *)
244 (bh->b_data + i);
245 /* It's too expensive to do a full
246 * dirent test each time round this
247 * loop, but we do have to test at
248 * least that it is non-zero. A
249 * failure will be detected in the
250 * dirent test below. */
251 if (ext4_rec_len_from_disk(de->rec_len,
252 sb->s_blocksize) < ext4_dir_rec_len(1,
253 inode))
254 break;
255 i += ext4_rec_len_from_disk(de->rec_len,
256 sb->s_blocksize);
257 }
258 offset = i;
259 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
260 | offset;
261 info->cookie = inode_query_iversion(inode);
262 }
263
264 if (unlikely(offset < sb->s_blocksize &&
265 offset > sb->s_blocksize -
266 ext4_dir_rec_len(1, has_csum ? NULL : inode))) {
267 EXT4_ERROR_FILE(file, bh->b_blocknr,
268 "bad entry in directory: %s - offset=%u, size=%lu",
269 "directory entry too close to block end",
270 offset, sb->s_blocksize);
271 ctx->pos = round_up(ctx->pos, sb->s_blocksize);
272 goto next_block;
273 }
274
275 while (ctx->pos < inode->i_size
276 && offset < sb->s_blocksize) {
277 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
278 if (ext4_check_dir_entry(inode, file, de, bh,
279 bh->b_data, bh->b_size,
280 offset)) {
281 /*
282 * On error, skip to the next block
283 */
284 ctx->pos = (ctx->pos |
285 (sb->s_blocksize - 1)) + 1;
286 break;
287 }
288 offset += ext4_rec_len_from_disk(de->rec_len,
289 sb->s_blocksize);
290 if (le32_to_cpu(de->inode)) {
291 if (!IS_ENCRYPTED(inode)) {
292 if (!dir_emit(ctx, de->name,
293 de->name_len,
294 le32_to_cpu(de->inode),
295 get_dtype(sb, de->file_type)))
296 goto done;
297 } else {
298 int save_len = fstr.len;
299 struct fscrypt_str de_name =
300 FSTR_INIT(de->name,
301 de->name_len);
302 u32 hash;
303 u32 minor_hash;
304
305 if (IS_CASEFOLDED(inode)) {
306 hash = EXT4_DIRENT_HASH(de);
307 minor_hash = EXT4_DIRENT_MINOR_HASH(de);
308 } else {
309 hash = 0;
310 minor_hash = 0;
311 }
312
313 /* Directory is encrypted */
314 err = fscrypt_fname_disk_to_usr(inode,
315 hash, minor_hash, &de_name, &fstr);
316 de_name = fstr;
317 fstr.len = save_len;
318 if (err)
319 goto errout;
320 if (!dir_emit(ctx,
321 de_name.name, de_name.len,
322 le32_to_cpu(de->inode),
323 get_dtype(sb, de->file_type)))
324 goto done;
325 }
326 }
327 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
328 sb->s_blocksize);
329 }
330next_block:
331 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
332 goto done;
333 brelse(bh);
334 bh = NULL;
335 }
336done:
337 err = 0;
338errout:
339 fscrypt_fname_free_buffer(&fstr);
340 brelse(bh);
341 return err;
342}
343
344static inline int is_32bit_api(void)
345{
346#ifdef CONFIG_COMPAT
347 return in_compat_syscall();
348#else
349 return (BITS_PER_LONG == 32);
350#endif
351}
352
353/*
354 * These functions convert from the major/minor hash to an f_pos
355 * value for dx directories
356 *
357 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
358 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
359 * directly on both 32-bit and 64-bit nodes, under such case, neither
360 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
361 */
362static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
363{
364 if ((filp->f_mode & FMODE_32BITHASH) ||
365 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
366 return major >> 1;
367 else
368 return ((__u64)(major >> 1) << 32) | (__u64)minor;
369}
370
371static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
372{
373 if ((filp->f_mode & FMODE_32BITHASH) ||
374 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
375 return (pos << 1) & 0xffffffff;
376 else
377 return ((pos >> 32) << 1) & 0xffffffff;
378}
379
380static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
381{
382 if ((filp->f_mode & FMODE_32BITHASH) ||
383 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
384 return 0;
385 else
386 return pos & 0xffffffff;
387}
388
389/*
390 * Return 32- or 64-bit end-of-file for dx directories
391 */
392static inline loff_t ext4_get_htree_eof(struct file *filp)
393{
394 if ((filp->f_mode & FMODE_32BITHASH) ||
395 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
396 return EXT4_HTREE_EOF_32BIT;
397 else
398 return EXT4_HTREE_EOF_64BIT;
399}
400
401
402/*
403 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
404 * directories, where the "offset" is in terms of the filename hash
405 * value instead of the byte offset.
406 *
407 * Because we may return a 64-bit hash that is well beyond offset limits,
408 * we need to pass the max hash as the maximum allowable offset in
409 * the htree directory case.
410 *
411 * For non-htree, ext4_llseek already chooses the proper max offset.
412 */
413static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
414{
415 struct inode *inode = file->f_mapping->host;
416 struct dir_private_info *info = file->private_data;
417 int dx_dir = is_dx_dir(inode);
418 loff_t ret, htree_max = ext4_get_htree_eof(file);
419
420 if (likely(dx_dir))
421 ret = generic_file_llseek_size(file, offset, whence,
422 htree_max, htree_max);
423 else
424 ret = ext4_llseek(file, offset, whence);
425 info->cookie = inode_peek_iversion(inode) - 1;
426 return ret;
427}
428
429/*
430 * This structure holds the nodes of the red-black tree used to store
431 * the directory entry in hash order.
432 */
433struct fname {
434 __u32 hash;
435 __u32 minor_hash;
436 struct rb_node rb_hash;
437 struct fname *next;
438 __u32 inode;
439 __u8 name_len;
440 __u8 file_type;
441 char name[] __counted_by(name_len);
442};
443
444/*
445 * This function implements a non-recursive way of freeing all of the
446 * nodes in the red-black tree.
447 */
448static void free_rb_tree_fname(struct rb_root *root)
449{
450 struct fname *fname, *next;
451
452 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
453 while (fname) {
454 struct fname *old = fname;
455 fname = fname->next;
456 kfree(old);
457 }
458
459 *root = RB_ROOT;
460}
461
462static void ext4_htree_init_dir_info(struct file *filp, loff_t pos)
463{
464 struct dir_private_info *p = filp->private_data;
465
466 if (is_dx_dir(file_inode(filp)) && !p->initialized) {
467 p->curr_hash = pos2maj_hash(filp, pos);
468 p->curr_minor_hash = pos2min_hash(filp, pos);
469 p->initialized = true;
470 }
471}
472
473void ext4_htree_free_dir_info(struct dir_private_info *p)
474{
475 free_rb_tree_fname(&p->root);
476 kfree(p);
477}
478
479/*
480 * Given a directory entry, enter it into the fname rb tree.
481 *
482 * When filename encryption is enabled, the dirent will hold the
483 * encrypted filename, while the htree will hold decrypted filename.
484 * The decrypted filename is passed in via ent_name. parameter.
485 */
486int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
487 __u32 minor_hash,
488 struct ext4_dir_entry_2 *dirent,
489 struct fscrypt_str *ent_name)
490{
491 struct rb_node **p, *parent = NULL;
492 struct fname *fname, *new_fn;
493 struct dir_private_info *info;
494
495 info = dir_file->private_data;
496 p = &info->root.rb_node;
497
498 /* Create and allocate the fname structure */
499 new_fn = kzalloc_flex(*new_fn, name, ent_name->len + 1);
500 if (!new_fn)
501 return -ENOMEM;
502 new_fn->hash = hash;
503 new_fn->minor_hash = minor_hash;
504 new_fn->inode = le32_to_cpu(dirent->inode);
505 new_fn->name_len = ent_name->len;
506 new_fn->file_type = dirent->file_type;
507 memcpy(new_fn->name, ent_name->name, ent_name->len);
508
509 while (*p) {
510 parent = *p;
511 fname = rb_entry(parent, struct fname, rb_hash);
512
513 /*
514 * If the hash and minor hash match up, then we put
515 * them on a linked list. This rarely happens...
516 */
517 if ((new_fn->hash == fname->hash) &&
518 (new_fn->minor_hash == fname->minor_hash)) {
519 new_fn->next = fname->next;
520 fname->next = new_fn;
521 return 0;
522 }
523
524 if (new_fn->hash < fname->hash)
525 p = &(*p)->rb_left;
526 else if (new_fn->hash > fname->hash)
527 p = &(*p)->rb_right;
528 else if (new_fn->minor_hash < fname->minor_hash)
529 p = &(*p)->rb_left;
530 else /* if (new_fn->minor_hash > fname->minor_hash) */
531 p = &(*p)->rb_right;
532 }
533
534 rb_link_node(&new_fn->rb_hash, parent, p);
535 rb_insert_color(&new_fn->rb_hash, &info->root);
536 return 0;
537}
538
539
540
541/*
542 * This is a helper function for ext4_dx_readdir. It calls filldir
543 * for all entries on the fname linked list. (Normally there is only
544 * one entry on the linked list, unless there are 62 bit hash collisions.)
545 */
546static int call_filldir(struct file *file, struct dir_context *ctx,
547 struct fname *fname)
548{
549 struct dir_private_info *info = file->private_data;
550 struct inode *inode = file_inode(file);
551 struct super_block *sb = inode->i_sb;
552
553 if (!fname) {
554 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%llu: comm %s: "
555 "called with null fname?!?", __func__, __LINE__,
556 inode->i_ino, current->comm);
557 return 0;
558 }
559 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
560 while (fname) {
561 if (!dir_emit(ctx, fname->name,
562 fname->name_len,
563 fname->inode,
564 get_dtype(sb, fname->file_type))) {
565 info->extra_fname = fname;
566 return 1;
567 }
568 fname = fname->next;
569 }
570 return 0;
571}
572
573static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
574{
575 struct dir_private_info *info = file->private_data;
576 struct inode *inode = file_inode(file);
577 struct fname *fname;
578 int ret = 0;
579
580 ext4_htree_init_dir_info(file, ctx->pos);
581
582 if (ctx->pos == ext4_get_htree_eof(file))
583 return 0; /* EOF */
584
585 /* Some one has messed with f_pos; reset the world */
586 if (info->last_pos != ctx->pos) {
587 free_rb_tree_fname(&info->root);
588 info->curr_node = NULL;
589 info->extra_fname = NULL;
590 info->curr_hash = pos2maj_hash(file, ctx->pos);
591 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
592 }
593
594 /*
595 * If there are any leftover names on the hash collision
596 * chain, return them first.
597 */
598 if (info->extra_fname) {
599 if (call_filldir(file, ctx, info->extra_fname))
600 goto finished;
601 info->extra_fname = NULL;
602 goto next_node;
603 } else if (!info->curr_node)
604 info->curr_node = rb_first(&info->root);
605
606 while (1) {
607 /*
608 * Fill the rbtree if we have no more entries,
609 * or the inode has changed since we last read in the
610 * cached entries.
611 */
612 if ((!info->curr_node) ||
613 !inode_eq_iversion(inode, info->cookie)) {
614 info->curr_node = NULL;
615 free_rb_tree_fname(&info->root);
616 info->cookie = inode_query_iversion(inode);
617 ret = ext4_htree_fill_tree(file, info->curr_hash,
618 info->curr_minor_hash,
619 &info->next_hash);
620 if (ret < 0)
621 goto finished;
622 if (ret == 0) {
623 ctx->pos = ext4_get_htree_eof(file);
624 break;
625 }
626 info->curr_node = rb_first(&info->root);
627 }
628
629 fname = rb_entry(info->curr_node, struct fname, rb_hash);
630 info->curr_hash = fname->hash;
631 info->curr_minor_hash = fname->minor_hash;
632 if (call_filldir(file, ctx, fname))
633 break;
634 next_node:
635 info->curr_node = rb_next(info->curr_node);
636 if (info->curr_node) {
637 fname = rb_entry(info->curr_node, struct fname,
638 rb_hash);
639 info->curr_hash = fname->hash;
640 info->curr_minor_hash = fname->minor_hash;
641 } else {
642 if (info->next_hash == ~0) {
643 ctx->pos = ext4_get_htree_eof(file);
644 break;
645 }
646 info->curr_hash = info->next_hash;
647 info->curr_minor_hash = 0;
648 }
649 }
650finished:
651 info->last_pos = ctx->pos;
652 return ret < 0 ? ret : 0;
653}
654
655static int ext4_release_dir(struct inode *inode, struct file *filp)
656{
657 if (filp->private_data)
658 ext4_htree_free_dir_info(filp->private_data);
659
660 return 0;
661}
662
663int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
664 int buf_size)
665{
666 struct ext4_dir_entry_2 *de;
667 int rlen;
668 unsigned int offset = 0;
669 char *top;
670
671 de = buf;
672 top = buf + buf_size;
673 while ((char *) de < top) {
674 if (ext4_check_dir_entry(dir, NULL, de, bh,
675 buf, buf_size, offset))
676 return -EFSCORRUPTED;
677 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
678 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
679 offset += rlen;
680 }
681 if ((char *) de > top)
682 return -EFSCORRUPTED;
683
684 return 0;
685}
686
687static int ext4_dir_open(struct inode *inode, struct file *file)
688{
689 struct dir_private_info *info;
690
691 info = kzalloc_obj(*info);
692 if (!info)
693 return -ENOMEM;
694 file->private_data = info;
695 return 0;
696}
697
698const struct file_operations ext4_dir_operations = {
699 .open = ext4_dir_open,
700 .llseek = ext4_dir_llseek,
701 .read = generic_read_dir,
702 .iterate_shared = ext4_readdir,
703 .unlocked_ioctl = ext4_ioctl,
704#ifdef CONFIG_COMPAT
705 .compat_ioctl = ext4_compat_ioctl,
706#endif
707 .fsync = ext4_sync_file,
708 .release = ext4_release_dir,
709 .setlease = generic_setlease,
710};