Merge tags 'spi-nor/for-5.11' and 'nand/for-5.11' into mtd/next
[linux-2.6-microblaze.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 <linux/unicode.h>
30 #include "ext4.h"
31 #include "xattr.h"
32
33 static 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  */
45 static int is_dx_dir(struct inode *inode)
46 {
47         struct super_block *sb = inode->i_sb;
48
49         if (ext4_has_feature_dir_index(inode->i_sb) &&
50             ((ext4_test_inode_flag(inode, 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
58 /*
59  * Return 0 if the directory entry is OK, and 1 if there is a problem
60  *
61  * Note: this is the opposite of what ext2 and ext3 historically returned...
62  *
63  * bh passed here can be an inode block or a dir data block, depending
64  * on the inode inline data flag.
65  */
66 int __ext4_check_dir_entry(const char *function, unsigned int line,
67                            struct inode *dir, struct file *filp,
68                            struct ext4_dir_entry_2 *de,
69                            struct buffer_head *bh, char *buf, int size,
70                            unsigned int offset)
71 {
72         const char *error_msg = NULL;
73         const int rlen = ext4_rec_len_from_disk(de->rec_len,
74                                                 dir->i_sb->s_blocksize);
75         const int next_offset = ((char *) de - buf) + rlen;
76
77         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
78                 error_msg = "rec_len is smaller than minimal";
79         else if (unlikely(rlen % 4 != 0))
80                 error_msg = "rec_len % 4 != 0";
81         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
82                 error_msg = "rec_len is too small for name_len";
83         else if (unlikely(next_offset > size))
84                 error_msg = "directory entry overrun";
85         else if (unlikely(next_offset > size - EXT4_DIR_REC_LEN(1) &&
86                           next_offset != size))
87                 error_msg = "directory entry too close to block end";
88         else if (unlikely(le32_to_cpu(de->inode) >
89                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
90                 error_msg = "inode out of bounds";
91         else
92                 return 0;
93
94         if (filp)
95                 ext4_error_file(filp, function, line, bh->b_blocknr,
96                                 "bad entry in directory: %s - offset=%u, "
97                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
98                                 error_msg, offset, le32_to_cpu(de->inode),
99                                 rlen, de->name_len, size);
100         else
101                 ext4_error_inode(dir, function, line, bh->b_blocknr,
102                                 "bad entry in directory: %s - offset=%u, "
103                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
104                                  error_msg, offset, le32_to_cpu(de->inode),
105                                  rlen, de->name_len, size);
106
107         return 1;
108 }
109
110 static int ext4_readdir(struct file *file, struct dir_context *ctx)
111 {
112         unsigned int offset;
113         int i;
114         struct ext4_dir_entry_2 *de;
115         int err;
116         struct inode *inode = file_inode(file);
117         struct super_block *sb = inode->i_sb;
118         struct buffer_head *bh = NULL;
119         struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
120
121         if (IS_ENCRYPTED(inode)) {
122                 err = fscrypt_get_encryption_info(inode);
123                 if (err)
124                         return err;
125         }
126
127         if (is_dx_dir(inode)) {
128                 err = ext4_dx_readdir(file, ctx);
129                 if (err != ERR_BAD_DX_DIR) {
130                         return err;
131                 }
132                 /* Can we just clear INDEX flag to ignore htree information? */
133                 if (!ext4_has_metadata_csum(sb)) {
134                         /*
135                          * We don't set the inode dirty flag since it's not
136                          * critical that it gets flushed back to the disk.
137                          */
138                         ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
139                 }
140         }
141
142         if (ext4_has_inline_data(inode)) {
143                 int has_inline_data = 1;
144                 err = ext4_read_inline_dir(file, ctx,
145                                            &has_inline_data);
146                 if (has_inline_data)
147                         return err;
148         }
149
150         if (IS_ENCRYPTED(inode)) {
151                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr);
152                 if (err < 0)
153                         return err;
154         }
155
156         while (ctx->pos < inode->i_size) {
157                 struct ext4_map_blocks map;
158
159                 if (fatal_signal_pending(current)) {
160                         err = -ERESTARTSYS;
161                         goto errout;
162                 }
163                 cond_resched();
164                 offset = ctx->pos & (sb->s_blocksize - 1);
165                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
166                 map.m_len = 1;
167                 err = ext4_map_blocks(NULL, inode, &map, 0);
168                 if (err == 0) {
169                         /* m_len should never be zero but let's avoid
170                          * an infinite loop if it somehow is */
171                         if (map.m_len == 0)
172                                 map.m_len = 1;
173                         ctx->pos += map.m_len * sb->s_blocksize;
174                         continue;
175                 }
176                 if (err > 0) {
177                         pgoff_t index = map.m_pblk >>
178                                         (PAGE_SHIFT - inode->i_blkbits);
179                         if (!ra_has_index(&file->f_ra, index))
180                                 page_cache_sync_readahead(
181                                         sb->s_bdev->bd_inode->i_mapping,
182                                         &file->f_ra, file,
183                                         index, 1);
184                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
185                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
186                         if (IS_ERR(bh)) {
187                                 err = PTR_ERR(bh);
188                                 bh = NULL;
189                                 goto errout;
190                         }
191                 }
192
193                 if (!bh) {
194                         /* corrupt size?  Maybe no more blocks to read */
195                         if (ctx->pos > inode->i_blocks << 9)
196                                 break;
197                         ctx->pos += sb->s_blocksize - offset;
198                         continue;
199                 }
200
201                 /* Check the checksum */
202                 if (!buffer_verified(bh) &&
203                     !ext4_dirblock_csum_verify(inode, bh)) {
204                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
205                                         "at offset %llu",
206                                         (unsigned long long)ctx->pos);
207                         ctx->pos += sb->s_blocksize - offset;
208                         brelse(bh);
209                         bh = NULL;
210                         continue;
211                 }
212                 set_buffer_verified(bh);
213
214                 /* If the dir block has changed since the last call to
215                  * readdir(2), then we might be pointing to an invalid
216                  * dirent right now.  Scan from the start of the block
217                  * to make sure. */
218                 if (!inode_eq_iversion(inode, file->f_version)) {
219                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
220                                 de = (struct ext4_dir_entry_2 *)
221                                         (bh->b_data + i);
222                                 /* It's too expensive to do a full
223                                  * dirent test each time round this
224                                  * loop, but we do have to test at
225                                  * least that it is non-zero.  A
226                                  * failure will be detected in the
227                                  * dirent test below. */
228                                 if (ext4_rec_len_from_disk(de->rec_len,
229                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
230                                         break;
231                                 i += ext4_rec_len_from_disk(de->rec_len,
232                                                             sb->s_blocksize);
233                         }
234                         offset = i;
235                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
236                                 | offset;
237                         file->f_version = inode_query_iversion(inode);
238                 }
239
240                 while (ctx->pos < inode->i_size
241                        && offset < sb->s_blocksize) {
242                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
243                         if (ext4_check_dir_entry(inode, file, de, bh,
244                                                  bh->b_data, bh->b_size,
245                                                  offset)) {
246                                 /*
247                                  * On error, skip to the next block
248                                  */
249                                 ctx->pos = (ctx->pos |
250                                                 (sb->s_blocksize - 1)) + 1;
251                                 break;
252                         }
253                         offset += ext4_rec_len_from_disk(de->rec_len,
254                                         sb->s_blocksize);
255                         if (le32_to_cpu(de->inode)) {
256                                 if (!IS_ENCRYPTED(inode)) {
257                                         if (!dir_emit(ctx, de->name,
258                                             de->name_len,
259                                             le32_to_cpu(de->inode),
260                                             get_dtype(sb, de->file_type)))
261                                                 goto done;
262                                 } else {
263                                         int save_len = fstr.len;
264                                         struct fscrypt_str de_name =
265                                                         FSTR_INIT(de->name,
266                                                                 de->name_len);
267
268                                         /* Directory is encrypted */
269                                         err = fscrypt_fname_disk_to_usr(inode,
270                                                 0, 0, &de_name, &fstr);
271                                         de_name = fstr;
272                                         fstr.len = save_len;
273                                         if (err)
274                                                 goto errout;
275                                         if (!dir_emit(ctx,
276                                             de_name.name, de_name.len,
277                                             le32_to_cpu(de->inode),
278                                             get_dtype(sb, de->file_type)))
279                                                 goto done;
280                                 }
281                         }
282                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
283                                                 sb->s_blocksize);
284                 }
285                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
286                         goto done;
287                 brelse(bh);
288                 bh = NULL;
289                 offset = 0;
290         }
291 done:
292         err = 0;
293 errout:
294         fscrypt_fname_free_buffer(&fstr);
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[];
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
468         while (*p) {
469                 parent = *p;
470                 fname = rb_entry(parent, struct fname, rb_hash);
471
472                 /*
473                  * If the hash and minor hash match up, then we put
474                  * them on a linked list.  This rarely happens...
475                  */
476                 if ((new_fn->hash == fname->hash) &&
477                     (new_fn->minor_hash == fname->minor_hash)) {
478                         new_fn->next = fname->next;
479                         fname->next = new_fn;
480                         return 0;
481                 }
482
483                 if (new_fn->hash < fname->hash)
484                         p = &(*p)->rb_left;
485                 else if (new_fn->hash > fname->hash)
486                         p = &(*p)->rb_right;
487                 else if (new_fn->minor_hash < fname->minor_hash)
488                         p = &(*p)->rb_left;
489                 else /* if (new_fn->minor_hash > fname->minor_hash) */
490                         p = &(*p)->rb_right;
491         }
492
493         rb_link_node(&new_fn->rb_hash, parent, p);
494         rb_insert_color(&new_fn->rb_hash, &info->root);
495         return 0;
496 }
497
498
499
500 /*
501  * This is a helper function for ext4_dx_readdir.  It calls filldir
502  * for all entres on the fname linked list.  (Normally there is only
503  * one entry on the linked list, unless there are 62 bit hash collisions.)
504  */
505 static int call_filldir(struct file *file, struct dir_context *ctx,
506                         struct fname *fname)
507 {
508         struct dir_private_info *info = file->private_data;
509         struct inode *inode = file_inode(file);
510         struct super_block *sb = inode->i_sb;
511
512         if (!fname) {
513                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
514                          "called with null fname?!?", __func__, __LINE__,
515                          inode->i_ino, current->comm);
516                 return 0;
517         }
518         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
519         while (fname) {
520                 if (!dir_emit(ctx, fname->name,
521                                 fname->name_len,
522                                 fname->inode,
523                                 get_dtype(sb, fname->file_type))) {
524                         info->extra_fname = fname;
525                         return 1;
526                 }
527                 fname = fname->next;
528         }
529         return 0;
530 }
531
532 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
533 {
534         struct dir_private_info *info = file->private_data;
535         struct inode *inode = file_inode(file);
536         struct fname *fname;
537         int     ret;
538
539         if (!info) {
540                 info = ext4_htree_create_dir_info(file, ctx->pos);
541                 if (!info)
542                         return -ENOMEM;
543                 file->private_data = info;
544         }
545
546         if (ctx->pos == ext4_get_htree_eof(file))
547                 return 0;       /* EOF */
548
549         /* Some one has messed with f_pos; reset the world */
550         if (info->last_pos != ctx->pos) {
551                 free_rb_tree_fname(&info->root);
552                 info->curr_node = NULL;
553                 info->extra_fname = NULL;
554                 info->curr_hash = pos2maj_hash(file, ctx->pos);
555                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
556         }
557
558         /*
559          * If there are any leftover names on the hash collision
560          * chain, return them first.
561          */
562         if (info->extra_fname) {
563                 if (call_filldir(file, ctx, info->extra_fname))
564                         goto finished;
565                 info->extra_fname = NULL;
566                 goto next_node;
567         } else if (!info->curr_node)
568                 info->curr_node = rb_first(&info->root);
569
570         while (1) {
571                 /*
572                  * Fill the rbtree if we have no more entries,
573                  * or the inode has changed since we last read in the
574                  * cached entries.
575                  */
576                 if ((!info->curr_node) ||
577                     !inode_eq_iversion(inode, file->f_version)) {
578                         info->curr_node = NULL;
579                         free_rb_tree_fname(&info->root);
580                         file->f_version = inode_query_iversion(inode);
581                         ret = ext4_htree_fill_tree(file, info->curr_hash,
582                                                    info->curr_minor_hash,
583                                                    &info->next_hash);
584                         if (ret < 0)
585                                 return ret;
586                         if (ret == 0) {
587                                 ctx->pos = ext4_get_htree_eof(file);
588                                 break;
589                         }
590                         info->curr_node = rb_first(&info->root);
591                 }
592
593                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
594                 info->curr_hash = fname->hash;
595                 info->curr_minor_hash = fname->minor_hash;
596                 if (call_filldir(file, ctx, fname))
597                         break;
598         next_node:
599                 info->curr_node = rb_next(info->curr_node);
600                 if (info->curr_node) {
601                         fname = rb_entry(info->curr_node, struct fname,
602                                          rb_hash);
603                         info->curr_hash = fname->hash;
604                         info->curr_minor_hash = fname->minor_hash;
605                 } else {
606                         if (info->next_hash == ~0) {
607                                 ctx->pos = ext4_get_htree_eof(file);
608                                 break;
609                         }
610                         info->curr_hash = info->next_hash;
611                         info->curr_minor_hash = 0;
612                 }
613         }
614 finished:
615         info->last_pos = ctx->pos;
616         return 0;
617 }
618
619 static int ext4_dir_open(struct inode * inode, struct file * filp)
620 {
621         if (IS_ENCRYPTED(inode))
622                 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
623         return 0;
624 }
625
626 static int ext4_release_dir(struct inode *inode, struct file *filp)
627 {
628         if (filp->private_data)
629                 ext4_htree_free_dir_info(filp->private_data);
630
631         return 0;
632 }
633
634 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
635                       int buf_size)
636 {
637         struct ext4_dir_entry_2 *de;
638         int rlen;
639         unsigned int offset = 0;
640         char *top;
641
642         de = (struct ext4_dir_entry_2 *)buf;
643         top = buf + buf_size;
644         while ((char *) de < top) {
645                 if (ext4_check_dir_entry(dir, NULL, de, bh,
646                                          buf, buf_size, offset))
647                         return -EFSCORRUPTED;
648                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
649                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
650                 offset += rlen;
651         }
652         if ((char *) de > top)
653                 return -EFSCORRUPTED;
654
655         return 0;
656 }
657
658 const struct file_operations ext4_dir_operations = {
659         .llseek         = ext4_dir_llseek,
660         .read           = generic_read_dir,
661         .iterate_shared = ext4_readdir,
662         .unlocked_ioctl = ext4_ioctl,
663 #ifdef CONFIG_COMPAT
664         .compat_ioctl   = ext4_compat_ioctl,
665 #endif
666         .fsync          = ext4_sync_file,
667         .open           = ext4_dir_open,
668         .release        = ext4_release_dir,
669 };
670
671 #ifdef CONFIG_UNICODE
672 const struct dentry_operations ext4_dentry_ops = {
673         .d_hash = generic_ci_d_hash,
674         .d_compare = generic_ci_d_compare,
675 };
676 #endif