GNU Linux-libre 4.19.286-gnu1
[releases.git] / fs / ext4 / dir.c
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 "ext4.h"
30 #include "xattr.h"
31
32 static int ext4_dx_readdir(struct file *, struct dir_context *);
33
34 /**
35  * Check if the given dir-inode refers to an htree-indexed directory
36  * (or a directory which could potentially get converted to use htree
37  * indexing).
38  *
39  * Return 1 if it is a dx dir, 0 if not
40  */
41 static int is_dx_dir(struct inode *inode)
42 {
43         struct super_block *sb = inode->i_sb;
44
45         if (ext4_has_feature_dir_index(inode->i_sb) &&
46             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
47              ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
48              ext4_has_inline_data(inode)))
49                 return 1;
50
51         return 0;
52 }
53
54 /*
55  * Return 0 if the directory entry is OK, and 1 if there is a problem
56  *
57  * Note: this is the opposite of what ext2 and ext3 historically returned...
58  *
59  * bh passed here can be an inode block or a dir data block, depending
60  * on the inode inline data flag.
61  */
62 int __ext4_check_dir_entry(const char *function, unsigned int line,
63                            struct inode *dir, struct file *filp,
64                            struct ext4_dir_entry_2 *de,
65                            struct buffer_head *bh, char *buf, int size,
66                            unsigned int offset)
67 {
68         const char *error_msg = NULL;
69         const int rlen = ext4_rec_len_from_disk(de->rec_len,
70                                                 dir->i_sb->s_blocksize);
71
72         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
73                 error_msg = "rec_len is smaller than minimal";
74         else if (unlikely(rlen % 4 != 0))
75                 error_msg = "rec_len % 4 != 0";
76         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
77                 error_msg = "rec_len is too small for name_len";
78         else if (unlikely(((char *) de - buf) + rlen > size))
79                 error_msg = "directory entry overrun";
80         else if (unlikely(((char *) de - buf) + rlen >
81                           size - EXT4_DIR_REC_LEN(1) &&
82                           ((char *) de - buf) + rlen != size)) {
83                 error_msg = "directory entry too close to block end";
84         }
85         else if (unlikely(le32_to_cpu(de->inode) >
86                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
87                 error_msg = "inode out of bounds";
88         else
89                 return 0;
90
91         if (filp)
92                 ext4_error_file(filp, function, line, bh->b_blocknr,
93                                 "bad entry in directory: %s - offset=%u, "
94                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
95                                 error_msg, offset, le32_to_cpu(de->inode),
96                                 rlen, de->name_len, size);
97         else
98                 ext4_error_inode(dir, function, line, bh->b_blocknr,
99                                 "bad entry in directory: %s - offset=%u, "
100                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
101                                  error_msg, offset, le32_to_cpu(de->inode),
102                                  rlen, de->name_len, size);
103
104         return 1;
105 }
106
107 static int ext4_readdir(struct file *file, struct dir_context *ctx)
108 {
109         unsigned int offset;
110         int i;
111         struct ext4_dir_entry_2 *de;
112         int err;
113         struct inode *inode = file_inode(file);
114         struct super_block *sb = inode->i_sb;
115         struct buffer_head *bh = NULL;
116         struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
117
118         if (ext4_encrypted_inode(inode)) {
119                 err = fscrypt_get_encryption_info(inode);
120                 if (err && err != -ENOKEY)
121                         return err;
122         }
123
124         if (is_dx_dir(inode)) {
125                 err = ext4_dx_readdir(file, ctx);
126                 if (err != ERR_BAD_DX_DIR) {
127                         return err;
128                 }
129                 /* Can we just clear INDEX flag to ignore htree information? */
130                 if (!ext4_has_metadata_csum(sb)) {
131                         /*
132                          * We don't set the inode dirty flag since it's not
133                          * critical that it gets flushed back to the disk.
134                          */
135                         ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
136                 }
137         }
138
139         if (ext4_has_inline_data(inode)) {
140                 int has_inline_data = 1;
141                 err = ext4_read_inline_dir(file, ctx,
142                                            &has_inline_data);
143                 if (has_inline_data)
144                         return err;
145         }
146
147         if (ext4_encrypted_inode(inode)) {
148                 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
149                 if (err < 0)
150                         return err;
151         }
152
153         while (ctx->pos < inode->i_size) {
154                 struct ext4_map_blocks map;
155
156                 if (fatal_signal_pending(current)) {
157                         err = -ERESTARTSYS;
158                         goto errout;
159                 }
160                 cond_resched();
161                 offset = ctx->pos & (sb->s_blocksize - 1);
162                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
163                 map.m_len = 1;
164                 err = ext4_map_blocks(NULL, inode, &map, 0);
165                 if (err == 0) {
166                         /* m_len should never be zero but let's avoid
167                          * an infinite loop if it somehow is */
168                         if (map.m_len == 0)
169                                 map.m_len = 1;
170                         ctx->pos += map.m_len * sb->s_blocksize;
171                         continue;
172                 }
173                 if (err > 0) {
174                         pgoff_t index = map.m_pblk >>
175                                         (PAGE_SHIFT - inode->i_blkbits);
176                         if (!ra_has_index(&file->f_ra, index))
177                                 page_cache_sync_readahead(
178                                         sb->s_bdev->bd_inode->i_mapping,
179                                         &file->f_ra, file,
180                                         index, 1);
181                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
182                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
183                         if (IS_ERR(bh)) {
184                                 err = PTR_ERR(bh);
185                                 bh = NULL;
186                                 goto errout;
187                         }
188                 }
189
190                 if (!bh) {
191                         /* corrupt size?  Maybe no more blocks to read */
192                         if (ctx->pos > inode->i_blocks << 9)
193                                 break;
194                         ctx->pos += sb->s_blocksize - offset;
195                         continue;
196                 }
197
198                 /* Check the checksum */
199                 if (!buffer_verified(bh) &&
200                     !ext4_dirent_csum_verify(inode,
201                                 (struct ext4_dir_entry *)bh->b_data)) {
202                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
203                                         "at offset %llu",
204                                         (unsigned long long)ctx->pos);
205                         ctx->pos += sb->s_blocksize - offset;
206                         brelse(bh);
207                         bh = NULL;
208                         continue;
209                 }
210                 set_buffer_verified(bh);
211
212                 /* If the dir block has changed since the last call to
213                  * readdir(2), then we might be pointing to an invalid
214                  * dirent right now.  Scan from the start of the block
215                  * to make sure. */
216                 if (!inode_eq_iversion(inode, file->f_version)) {
217                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
218                                 de = (struct ext4_dir_entry_2 *)
219                                         (bh->b_data + i);
220                                 /* It's too expensive to do a full
221                                  * dirent test each time round this
222                                  * loop, but we do have to test at
223                                  * least that it is non-zero.  A
224                                  * failure will be detected in the
225                                  * dirent test below. */
226                                 if (ext4_rec_len_from_disk(de->rec_len,
227                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
228                                         break;
229                                 i += ext4_rec_len_from_disk(de->rec_len,
230                                                             sb->s_blocksize);
231                         }
232                         offset = i;
233                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
234                                 | offset;
235                         file->f_version = inode_query_iversion(inode);
236                 }
237
238                 while (ctx->pos < inode->i_size
239                        && offset < sb->s_blocksize) {
240                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
241                         if (ext4_check_dir_entry(inode, file, de, bh,
242                                                  bh->b_data, bh->b_size,
243                                                  offset)) {
244                                 /*
245                                  * On error, skip to the next block
246                                  */
247                                 ctx->pos = (ctx->pos |
248                                                 (sb->s_blocksize - 1)) + 1;
249                                 break;
250                         }
251                         offset += ext4_rec_len_from_disk(de->rec_len,
252                                         sb->s_blocksize);
253                         if (le32_to_cpu(de->inode)) {
254                                 if (!ext4_encrypted_inode(inode)) {
255                                         if (!dir_emit(ctx, de->name,
256                                             de->name_len,
257                                             le32_to_cpu(de->inode),
258                                             get_dtype(sb, de->file_type)))
259                                                 goto done;
260                                 } else {
261                                         int save_len = fstr.len;
262                                         struct fscrypt_str de_name =
263                                                         FSTR_INIT(de->name,
264                                                                 de->name_len);
265
266                                         /* Directory is encrypted */
267                                         err = fscrypt_fname_disk_to_usr(inode,
268                                                 0, 0, &de_name, &fstr);
269                                         de_name = fstr;
270                                         fstr.len = save_len;
271                                         if (err)
272                                                 goto errout;
273                                         if (!dir_emit(ctx,
274                                             de_name.name, de_name.len,
275                                             le32_to_cpu(de->inode),
276                                             get_dtype(sb, de->file_type)))
277                                                 goto done;
278                                 }
279                         }
280                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
281                                                 sb->s_blocksize);
282                 }
283                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
284                         goto done;
285                 brelse(bh);
286                 bh = NULL;
287                 offset = 0;
288         }
289 done:
290         err = 0;
291 errout:
292 #ifdef CONFIG_EXT4_FS_ENCRYPTION
293         fscrypt_fname_free_buffer(&fstr);
294 #endif
295         brelse(bh);
296         return err;
297 }
298
299 static inline int is_32bit_api(void)
300 {
301 #ifdef CONFIG_COMPAT
302         return in_compat_syscall();
303 #else
304         return (BITS_PER_LONG == 32);
305 #endif
306 }
307
308 /*
309  * These functions convert from the major/minor hash to an f_pos
310  * value for dx directories
311  *
312  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
313  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
314  * directly on both 32-bit and 64-bit nodes, under such case, neither
315  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
316  */
317 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
318 {
319         if ((filp->f_mode & FMODE_32BITHASH) ||
320             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
321                 return major >> 1;
322         else
323                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
324 }
325
326 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
327 {
328         if ((filp->f_mode & FMODE_32BITHASH) ||
329             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
330                 return (pos << 1) & 0xffffffff;
331         else
332                 return ((pos >> 32) << 1) & 0xffffffff;
333 }
334
335 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
336 {
337         if ((filp->f_mode & FMODE_32BITHASH) ||
338             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
339                 return 0;
340         else
341                 return pos & 0xffffffff;
342 }
343
344 /*
345  * Return 32- or 64-bit end-of-file for dx directories
346  */
347 static inline loff_t ext4_get_htree_eof(struct file *filp)
348 {
349         if ((filp->f_mode & FMODE_32BITHASH) ||
350             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
351                 return EXT4_HTREE_EOF_32BIT;
352         else
353                 return EXT4_HTREE_EOF_64BIT;
354 }
355
356
357 /*
358  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
359  * directories, where the "offset" is in terms of the filename hash
360  * value instead of the byte offset.
361  *
362  * Because we may return a 64-bit hash that is well beyond offset limits,
363  * we need to pass the max hash as the maximum allowable offset in
364  * the htree directory case.
365  *
366  * For non-htree, ext4_llseek already chooses the proper max offset.
367  */
368 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
369 {
370         struct inode *inode = file->f_mapping->host;
371         int dx_dir = is_dx_dir(inode);
372         loff_t ret, htree_max = ext4_get_htree_eof(file);
373
374         if (likely(dx_dir))
375                 ret = generic_file_llseek_size(file, offset, whence,
376                                                     htree_max, htree_max);
377         else
378                 ret = ext4_llseek(file, offset, whence);
379         file->f_version = inode_peek_iversion(inode) - 1;
380         return ret;
381 }
382
383 /*
384  * This structure holds the nodes of the red-black tree used to store
385  * the directory entry in hash order.
386  */
387 struct fname {
388         __u32           hash;
389         __u32           minor_hash;
390         struct rb_node  rb_hash;
391         struct fname    *next;
392         __u32           inode;
393         __u8            name_len;
394         __u8            file_type;
395         char            name[0];
396 };
397
398 /*
399  * This functoin implements a non-recursive way of freeing all of the
400  * nodes in the red-black tree.
401  */
402 static void free_rb_tree_fname(struct rb_root *root)
403 {
404         struct fname *fname, *next;
405
406         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
407                 while (fname) {
408                         struct fname *old = fname;
409                         fname = fname->next;
410                         kfree(old);
411                 }
412
413         *root = RB_ROOT;
414 }
415
416
417 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
418                                                            loff_t pos)
419 {
420         struct dir_private_info *p;
421
422         p = kzalloc(sizeof(*p), GFP_KERNEL);
423         if (!p)
424                 return NULL;
425         p->curr_hash = pos2maj_hash(filp, pos);
426         p->curr_minor_hash = pos2min_hash(filp, pos);
427         return p;
428 }
429
430 void ext4_htree_free_dir_info(struct dir_private_info *p)
431 {
432         free_rb_tree_fname(&p->root);
433         kfree(p);
434 }
435
436 /*
437  * Given a directory entry, enter it into the fname rb tree.
438  *
439  * When filename encryption is enabled, the dirent will hold the
440  * encrypted filename, while the htree will hold decrypted filename.
441  * The decrypted filename is passed in via ent_name.  parameter.
442  */
443 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
444                              __u32 minor_hash,
445                             struct ext4_dir_entry_2 *dirent,
446                             struct fscrypt_str *ent_name)
447 {
448         struct rb_node **p, *parent = NULL;
449         struct fname *fname, *new_fn;
450         struct dir_private_info *info;
451         int len;
452
453         info = dir_file->private_data;
454         p = &info->root.rb_node;
455
456         /* Create and allocate the fname structure */
457         len = sizeof(struct fname) + ent_name->len + 1;
458         new_fn = kzalloc(len, GFP_KERNEL);
459         if (!new_fn)
460                 return -ENOMEM;
461         new_fn->hash = hash;
462         new_fn->minor_hash = minor_hash;
463         new_fn->inode = le32_to_cpu(dirent->inode);
464         new_fn->name_len = ent_name->len;
465         new_fn->file_type = dirent->file_type;
466         memcpy(new_fn->name, ent_name->name, ent_name->len);
467         new_fn->name[ent_name->len] = 0;
468
469         while (*p) {
470                 parent = *p;
471                 fname = rb_entry(parent, struct fname, rb_hash);
472
473                 /*
474                  * If the hash and minor hash match up, then we put
475                  * them on a linked list.  This rarely happens...
476                  */
477                 if ((new_fn->hash == fname->hash) &&
478                     (new_fn->minor_hash == fname->minor_hash)) {
479                         new_fn->next = fname->next;
480                         fname->next = new_fn;
481                         return 0;
482                 }
483
484                 if (new_fn->hash < fname->hash)
485                         p = &(*p)->rb_left;
486                 else if (new_fn->hash > fname->hash)
487                         p = &(*p)->rb_right;
488                 else if (new_fn->minor_hash < fname->minor_hash)
489                         p = &(*p)->rb_left;
490                 else /* if (new_fn->minor_hash > fname->minor_hash) */
491                         p = &(*p)->rb_right;
492         }
493
494         rb_link_node(&new_fn->rb_hash, parent, p);
495         rb_insert_color(&new_fn->rb_hash, &info->root);
496         return 0;
497 }
498
499
500
501 /*
502  * This is a helper function for ext4_dx_readdir.  It calls filldir
503  * for all entres on the fname linked list.  (Normally there is only
504  * one entry on the linked list, unless there are 62 bit hash collisions.)
505  */
506 static int call_filldir(struct file *file, struct dir_context *ctx,
507                         struct fname *fname)
508 {
509         struct dir_private_info *info = file->private_data;
510         struct inode *inode = file_inode(file);
511         struct super_block *sb = inode->i_sb;
512
513         if (!fname) {
514                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
515                          "called with null fname?!?", __func__, __LINE__,
516                          inode->i_ino, current->comm);
517                 return 0;
518         }
519         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
520         while (fname) {
521                 if (!dir_emit(ctx, fname->name,
522                                 fname->name_len,
523                                 fname->inode,
524                                 get_dtype(sb, fname->file_type))) {
525                         info->extra_fname = fname;
526                         return 1;
527                 }
528                 fname = fname->next;
529         }
530         return 0;
531 }
532
533 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
534 {
535         struct dir_private_info *info = file->private_data;
536         struct inode *inode = file_inode(file);
537         struct fname *fname;
538         int ret = 0;
539
540         if (!info) {
541                 info = ext4_htree_create_dir_info(file, ctx->pos);
542                 if (!info)
543                         return -ENOMEM;
544                 file->private_data = info;
545         }
546
547         if (ctx->pos == ext4_get_htree_eof(file))
548                 return 0;       /* EOF */
549
550         /* Some one has messed with f_pos; reset the world */
551         if (info->last_pos != ctx->pos) {
552                 free_rb_tree_fname(&info->root);
553                 info->curr_node = NULL;
554                 info->extra_fname = NULL;
555                 info->curr_hash = pos2maj_hash(file, ctx->pos);
556                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
557         }
558
559         /*
560          * If there are any leftover names on the hash collision
561          * chain, return them first.
562          */
563         if (info->extra_fname) {
564                 if (call_filldir(file, ctx, info->extra_fname))
565                         goto finished;
566                 info->extra_fname = NULL;
567                 goto next_node;
568         } else if (!info->curr_node)
569                 info->curr_node = rb_first(&info->root);
570
571         while (1) {
572                 /*
573                  * Fill the rbtree if we have no more entries,
574                  * or the inode has changed since we last read in the
575                  * cached entries.
576                  */
577                 if ((!info->curr_node) ||
578                     !inode_eq_iversion(inode, file->f_version)) {
579                         info->curr_node = NULL;
580                         free_rb_tree_fname(&info->root);
581                         file->f_version = inode_query_iversion(inode);
582                         ret = ext4_htree_fill_tree(file, info->curr_hash,
583                                                    info->curr_minor_hash,
584                                                    &info->next_hash);
585                         if (ret < 0)
586                                 goto finished;
587                         if (ret == 0) {
588                                 ctx->pos = ext4_get_htree_eof(file);
589                                 break;
590                         }
591                         info->curr_node = rb_first(&info->root);
592                 }
593
594                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
595                 info->curr_hash = fname->hash;
596                 info->curr_minor_hash = fname->minor_hash;
597                 if (call_filldir(file, ctx, fname))
598                         break;
599         next_node:
600                 info->curr_node = rb_next(info->curr_node);
601                 if (info->curr_node) {
602                         fname = rb_entry(info->curr_node, struct fname,
603                                          rb_hash);
604                         info->curr_hash = fname->hash;
605                         info->curr_minor_hash = fname->minor_hash;
606                 } else {
607                         if (info->next_hash == ~0) {
608                                 ctx->pos = ext4_get_htree_eof(file);
609                                 break;
610                         }
611                         info->curr_hash = info->next_hash;
612                         info->curr_minor_hash = 0;
613                 }
614         }
615 finished:
616         info->last_pos = ctx->pos;
617         return ret < 0 ? ret : 0;
618 }
619
620 static int ext4_dir_open(struct inode * inode, struct file * filp)
621 {
622         if (ext4_encrypted_inode(inode))
623                 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
624         return 0;
625 }
626
627 static int ext4_release_dir(struct inode *inode, struct file *filp)
628 {
629         if (filp->private_data)
630                 ext4_htree_free_dir_info(filp->private_data);
631
632         return 0;
633 }
634
635 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
636                       int buf_size)
637 {
638         struct ext4_dir_entry_2 *de;
639         int rlen;
640         unsigned int offset = 0;
641         char *top;
642
643         de = (struct ext4_dir_entry_2 *)buf;
644         top = buf + buf_size;
645         while ((char *) de < top) {
646                 if (ext4_check_dir_entry(dir, NULL, de, bh,
647                                          buf, buf_size, offset))
648                         return -EFSCORRUPTED;
649                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
650                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
651                 offset += rlen;
652         }
653         if ((char *) de > top)
654                 return -EFSCORRUPTED;
655
656         return 0;
657 }
658
659 const struct file_operations ext4_dir_operations = {
660         .llseek         = ext4_dir_llseek,
661         .read           = generic_read_dir,
662         .iterate_shared = ext4_readdir,
663         .unlocked_ioctl = ext4_ioctl,
664 #ifdef CONFIG_COMPAT
665         .compat_ioctl   = ext4_compat_ioctl,
666 #endif
667         .fsync          = ext4_sync_file,
668         .open           = ext4_dir_open,
669         .release        = ext4_release_dir,
670 };