Merge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm
[sfrench/cifs-2.6.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 "ext4.h"
29 #include "xattr.h"
30
31 static int ext4_dx_readdir(struct file *, struct dir_context *);
32
33 /**
34  * Check if the given dir-inode refers to an htree-indexed directory
35  * (or a directory which could potentially get converted to use htree
36  * indexing).
37  *
38  * Return 1 if it is a dx dir, 0 if not
39  */
40 static int is_dx_dir(struct inode *inode)
41 {
42         struct super_block *sb = inode->i_sb;
43
44         if (ext4_has_feature_dir_index(inode->i_sb) &&
45             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
46              ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
47              ext4_has_inline_data(inode)))
48                 return 1;
49
50         return 0;
51 }
52
53 /*
54  * Return 0 if the directory entry is OK, and 1 if there is a problem
55  *
56  * Note: this is the opposite of what ext2 and ext3 historically returned...
57  *
58  * bh passed here can be an inode block or a dir data block, depending
59  * on the inode inline data flag.
60  */
61 int __ext4_check_dir_entry(const char *function, unsigned int line,
62                            struct inode *dir, struct file *filp,
63                            struct ext4_dir_entry_2 *de,
64                            struct buffer_head *bh, char *buf, int size,
65                            unsigned int offset)
66 {
67         const char *error_msg = NULL;
68         const int rlen = ext4_rec_len_from_disk(de->rec_len,
69                                                 dir->i_sb->s_blocksize);
70
71         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
72                 error_msg = "rec_len is smaller than minimal";
73         else if (unlikely(rlen % 4 != 0))
74                 error_msg = "rec_len % 4 != 0";
75         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
76                 error_msg = "rec_len is too small for name_len";
77         else if (unlikely(((char *) de - buf) + rlen > size))
78                 error_msg = "directory entry across range";
79         else if (unlikely(le32_to_cpu(de->inode) >
80                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
81                 error_msg = "inode out of bounds";
82         else
83                 return 0;
84
85         if (filp)
86                 ext4_error_file(filp, function, line, bh->b_blocknr,
87                                 "bad entry in directory: %s - offset=%u(%u), "
88                                 "inode=%u, rec_len=%d, name_len=%d",
89                                 error_msg, (unsigned) (offset % size),
90                                 offset, le32_to_cpu(de->inode),
91                                 rlen, de->name_len);
92         else
93                 ext4_error_inode(dir, function, line, bh->b_blocknr,
94                                 "bad entry in directory: %s - offset=%u(%u), "
95                                 "inode=%u, rec_len=%d, name_len=%d",
96                                 error_msg, (unsigned) (offset % size),
97                                 offset, le32_to_cpu(de->inode),
98                                 rlen, de->name_len);
99
100         return 1;
101 }
102
103 static int ext4_readdir(struct file *file, struct dir_context *ctx)
104 {
105         unsigned int offset;
106         int i;
107         struct ext4_dir_entry_2 *de;
108         int err;
109         struct inode *inode = file_inode(file);
110         struct super_block *sb = inode->i_sb;
111         struct buffer_head *bh = NULL;
112         int dir_has_error = 0;
113         struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
114
115         if (ext4_encrypted_inode(inode)) {
116                 err = fscrypt_get_encryption_info(inode);
117                 if (err && err != -ENOKEY)
118                         return err;
119         }
120
121         if (is_dx_dir(inode)) {
122                 err = ext4_dx_readdir(file, ctx);
123                 if (err != ERR_BAD_DX_DIR) {
124                         return err;
125                 }
126                 /*
127                  * We don't set the inode dirty flag since it's not
128                  * critical that it get flushed back to the disk.
129                  */
130                 ext4_clear_inode_flag(file_inode(file),
131                                       EXT4_INODE_INDEX);
132         }
133
134         if (ext4_has_inline_data(inode)) {
135                 int has_inline_data = 1;
136                 err = ext4_read_inline_dir(file, ctx,
137                                            &has_inline_data);
138                 if (has_inline_data)
139                         return err;
140         }
141
142         if (ext4_encrypted_inode(inode)) {
143                 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
144                 if (err < 0)
145                         return err;
146         }
147
148         offset = ctx->pos & (sb->s_blocksize - 1);
149
150         while (ctx->pos < inode->i_size) {
151                 struct ext4_map_blocks map;
152
153                 if (fatal_signal_pending(current)) {
154                         err = -ERESTARTSYS;
155                         goto errout;
156                 }
157                 cond_resched();
158                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
159                 map.m_len = 1;
160                 err = ext4_map_blocks(NULL, inode, &map, 0);
161                 if (err > 0) {
162                         pgoff_t index = map.m_pblk >>
163                                         (PAGE_SHIFT - inode->i_blkbits);
164                         if (!ra_has_index(&file->f_ra, index))
165                                 page_cache_sync_readahead(
166                                         sb->s_bdev->bd_inode->i_mapping,
167                                         &file->f_ra, file,
168                                         index, 1);
169                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
170                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
171                         if (IS_ERR(bh)) {
172                                 err = PTR_ERR(bh);
173                                 bh = NULL;
174                                 goto errout;
175                         }
176                 }
177
178                 if (!bh) {
179                         if (!dir_has_error) {
180                                 EXT4_ERROR_FILE(file, 0,
181                                                 "directory contains a "
182                                                 "hole at offset %llu",
183                                            (unsigned long long) ctx->pos);
184                                 dir_has_error = 1;
185                         }
186                         /* corrupt size?  Maybe no more blocks to read */
187                         if (ctx->pos > inode->i_blocks << 9)
188                                 break;
189                         ctx->pos += sb->s_blocksize - offset;
190                         continue;
191                 }
192
193                 /* Check the checksum */
194                 if (!buffer_verified(bh) &&
195                     !ext4_dirent_csum_verify(inode,
196                                 (struct ext4_dir_entry *)bh->b_data)) {
197                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
198                                         "at offset %llu",
199                                         (unsigned long long)ctx->pos);
200                         ctx->pos += sb->s_blocksize - offset;
201                         brelse(bh);
202                         bh = NULL;
203                         continue;
204                 }
205                 set_buffer_verified(bh);
206
207                 /* If the dir block has changed since the last call to
208                  * readdir(2), then we might be pointing to an invalid
209                  * dirent right now.  Scan from the start of the block
210                  * to make sure. */
211                 if (file->f_version != inode->i_version) {
212                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
213                                 de = (struct ext4_dir_entry_2 *)
214                                         (bh->b_data + i);
215                                 /* It's too expensive to do a full
216                                  * dirent test each time round this
217                                  * loop, but we do have to test at
218                                  * least that it is non-zero.  A
219                                  * failure will be detected in the
220                                  * dirent test below. */
221                                 if (ext4_rec_len_from_disk(de->rec_len,
222                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
223                                         break;
224                                 i += ext4_rec_len_from_disk(de->rec_len,
225                                                             sb->s_blocksize);
226                         }
227                         offset = i;
228                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
229                                 | offset;
230                         file->f_version = inode->i_version;
231                 }
232
233                 while (ctx->pos < inode->i_size
234                        && offset < sb->s_blocksize) {
235                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
236                         if (ext4_check_dir_entry(inode, file, de, bh,
237                                                  bh->b_data, bh->b_size,
238                                                  offset)) {
239                                 /*
240                                  * On error, skip to the next block
241                                  */
242                                 ctx->pos = (ctx->pos |
243                                                 (sb->s_blocksize - 1)) + 1;
244                                 break;
245                         }
246                         offset += ext4_rec_len_from_disk(de->rec_len,
247                                         sb->s_blocksize);
248                         if (le32_to_cpu(de->inode)) {
249                                 if (!ext4_encrypted_inode(inode)) {
250                                         if (!dir_emit(ctx, de->name,
251                                             de->name_len,
252                                             le32_to_cpu(de->inode),
253                                             get_dtype(sb, de->file_type)))
254                                                 goto done;
255                                 } else {
256                                         int save_len = fstr.len;
257                                         struct fscrypt_str de_name =
258                                                         FSTR_INIT(de->name,
259                                                                 de->name_len);
260
261                                         /* Directory is encrypted */
262                                         err = fscrypt_fname_disk_to_usr(inode,
263                                                 0, 0, &de_name, &fstr);
264                                         de_name = fstr;
265                                         fstr.len = save_len;
266                                         if (err)
267                                                 goto errout;
268                                         if (!dir_emit(ctx,
269                                             de_name.name, de_name.len,
270                                             le32_to_cpu(de->inode),
271                                             get_dtype(sb, de->file_type)))
272                                                 goto done;
273                                 }
274                         }
275                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
276                                                 sb->s_blocksize);
277                 }
278                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
279                         goto done;
280                 brelse(bh);
281                 bh = NULL;
282                 offset = 0;
283         }
284 done:
285         err = 0;
286 errout:
287 #ifdef CONFIG_EXT4_FS_ENCRYPTION
288         fscrypt_fname_free_buffer(&fstr);
289 #endif
290         brelse(bh);
291         return err;
292 }
293
294 static inline int is_32bit_api(void)
295 {
296 #ifdef CONFIG_COMPAT
297         return in_compat_syscall();
298 #else
299         return (BITS_PER_LONG == 32);
300 #endif
301 }
302
303 /*
304  * These functions convert from the major/minor hash to an f_pos
305  * value for dx directories
306  *
307  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
308  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
309  * directly on both 32-bit and 64-bit nodes, under such case, neither
310  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
311  */
312 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
313 {
314         if ((filp->f_mode & FMODE_32BITHASH) ||
315             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
316                 return major >> 1;
317         else
318                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
319 }
320
321 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
322 {
323         if ((filp->f_mode & FMODE_32BITHASH) ||
324             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
325                 return (pos << 1) & 0xffffffff;
326         else
327                 return ((pos >> 32) << 1) & 0xffffffff;
328 }
329
330 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
331 {
332         if ((filp->f_mode & FMODE_32BITHASH) ||
333             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
334                 return 0;
335         else
336                 return pos & 0xffffffff;
337 }
338
339 /*
340  * Return 32- or 64-bit end-of-file for dx directories
341  */
342 static inline loff_t ext4_get_htree_eof(struct file *filp)
343 {
344         if ((filp->f_mode & FMODE_32BITHASH) ||
345             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
346                 return EXT4_HTREE_EOF_32BIT;
347         else
348                 return EXT4_HTREE_EOF_64BIT;
349 }
350
351
352 /*
353  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
354  * directories, where the "offset" is in terms of the filename hash
355  * value instead of the byte offset.
356  *
357  * Because we may return a 64-bit hash that is well beyond offset limits,
358  * we need to pass the max hash as the maximum allowable offset in
359  * the htree directory case.
360  *
361  * For non-htree, ext4_llseek already chooses the proper max offset.
362  */
363 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
364 {
365         struct inode *inode = file->f_mapping->host;
366         int dx_dir = is_dx_dir(inode);
367         loff_t htree_max = ext4_get_htree_eof(file);
368
369         if (likely(dx_dir))
370                 return generic_file_llseek_size(file, offset, whence,
371                                                     htree_max, htree_max);
372         else
373                 return ext4_llseek(file, offset, whence);
374 }
375
376 /*
377  * This structure holds the nodes of the red-black tree used to store
378  * the directory entry in hash order.
379  */
380 struct fname {
381         __u32           hash;
382         __u32           minor_hash;
383         struct rb_node  rb_hash;
384         struct fname    *next;
385         __u32           inode;
386         __u8            name_len;
387         __u8            file_type;
388         char            name[0];
389 };
390
391 /*
392  * This functoin implements a non-recursive way of freeing all of the
393  * nodes in the red-black tree.
394  */
395 static void free_rb_tree_fname(struct rb_root *root)
396 {
397         struct fname *fname, *next;
398
399         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
400                 while (fname) {
401                         struct fname *old = fname;
402                         fname = fname->next;
403                         kfree(old);
404                 }
405
406         *root = RB_ROOT;
407 }
408
409
410 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
411                                                            loff_t pos)
412 {
413         struct dir_private_info *p;
414
415         p = kzalloc(sizeof(*p), GFP_KERNEL);
416         if (!p)
417                 return NULL;
418         p->curr_hash = pos2maj_hash(filp, pos);
419         p->curr_minor_hash = pos2min_hash(filp, pos);
420         return p;
421 }
422
423 void ext4_htree_free_dir_info(struct dir_private_info *p)
424 {
425         free_rb_tree_fname(&p->root);
426         kfree(p);
427 }
428
429 /*
430  * Given a directory entry, enter it into the fname rb tree.
431  *
432  * When filename encryption is enabled, the dirent will hold the
433  * encrypted filename, while the htree will hold decrypted filename.
434  * The decrypted filename is passed in via ent_name.  parameter.
435  */
436 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
437                              __u32 minor_hash,
438                             struct ext4_dir_entry_2 *dirent,
439                             struct fscrypt_str *ent_name)
440 {
441         struct rb_node **p, *parent = NULL;
442         struct fname *fname, *new_fn;
443         struct dir_private_info *info;
444         int len;
445
446         info = dir_file->private_data;
447         p = &info->root.rb_node;
448
449         /* Create and allocate the fname structure */
450         len = sizeof(struct fname) + ent_name->len + 1;
451         new_fn = kzalloc(len, GFP_KERNEL);
452         if (!new_fn)
453                 return -ENOMEM;
454         new_fn->hash = hash;
455         new_fn->minor_hash = minor_hash;
456         new_fn->inode = le32_to_cpu(dirent->inode);
457         new_fn->name_len = ent_name->len;
458         new_fn->file_type = dirent->file_type;
459         memcpy(new_fn->name, ent_name->name, ent_name->len);
460         new_fn->name[ent_name->len] = 0;
461
462         while (*p) {
463                 parent = *p;
464                 fname = rb_entry(parent, struct fname, rb_hash);
465
466                 /*
467                  * If the hash and minor hash match up, then we put
468                  * them on a linked list.  This rarely happens...
469                  */
470                 if ((new_fn->hash == fname->hash) &&
471                     (new_fn->minor_hash == fname->minor_hash)) {
472                         new_fn->next = fname->next;
473                         fname->next = new_fn;
474                         return 0;
475                 }
476
477                 if (new_fn->hash < fname->hash)
478                         p = &(*p)->rb_left;
479                 else if (new_fn->hash > fname->hash)
480                         p = &(*p)->rb_right;
481                 else if (new_fn->minor_hash < fname->minor_hash)
482                         p = &(*p)->rb_left;
483                 else /* if (new_fn->minor_hash > fname->minor_hash) */
484                         p = &(*p)->rb_right;
485         }
486
487         rb_link_node(&new_fn->rb_hash, parent, p);
488         rb_insert_color(&new_fn->rb_hash, &info->root);
489         return 0;
490 }
491
492
493
494 /*
495  * This is a helper function for ext4_dx_readdir.  It calls filldir
496  * for all entres on the fname linked list.  (Normally there is only
497  * one entry on the linked list, unless there are 62 bit hash collisions.)
498  */
499 static int call_filldir(struct file *file, struct dir_context *ctx,
500                         struct fname *fname)
501 {
502         struct dir_private_info *info = file->private_data;
503         struct inode *inode = file_inode(file);
504         struct super_block *sb = inode->i_sb;
505
506         if (!fname) {
507                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
508                          "called with null fname?!?", __func__, __LINE__,
509                          inode->i_ino, current->comm);
510                 return 0;
511         }
512         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
513         while (fname) {
514                 if (!dir_emit(ctx, fname->name,
515                                 fname->name_len,
516                                 fname->inode,
517                                 get_dtype(sb, fname->file_type))) {
518                         info->extra_fname = fname;
519                         return 1;
520                 }
521                 fname = fname->next;
522         }
523         return 0;
524 }
525
526 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
527 {
528         struct dir_private_info *info = file->private_data;
529         struct inode *inode = file_inode(file);
530         struct fname *fname;
531         int     ret;
532
533         if (!info) {
534                 info = ext4_htree_create_dir_info(file, ctx->pos);
535                 if (!info)
536                         return -ENOMEM;
537                 file->private_data = info;
538         }
539
540         if (ctx->pos == ext4_get_htree_eof(file))
541                 return 0;       /* EOF */
542
543         /* Some one has messed with f_pos; reset the world */
544         if (info->last_pos != ctx->pos) {
545                 free_rb_tree_fname(&info->root);
546                 info->curr_node = NULL;
547                 info->extra_fname = NULL;
548                 info->curr_hash = pos2maj_hash(file, ctx->pos);
549                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
550         }
551
552         /*
553          * If there are any leftover names on the hash collision
554          * chain, return them first.
555          */
556         if (info->extra_fname) {
557                 if (call_filldir(file, ctx, info->extra_fname))
558                         goto finished;
559                 info->extra_fname = NULL;
560                 goto next_node;
561         } else if (!info->curr_node)
562                 info->curr_node = rb_first(&info->root);
563
564         while (1) {
565                 /*
566                  * Fill the rbtree if we have no more entries,
567                  * or the inode has changed since we last read in the
568                  * cached entries.
569                  */
570                 if ((!info->curr_node) ||
571                     (file->f_version != inode->i_version)) {
572                         info->curr_node = NULL;
573                         free_rb_tree_fname(&info->root);
574                         file->f_version = inode->i_version;
575                         ret = ext4_htree_fill_tree(file, info->curr_hash,
576                                                    info->curr_minor_hash,
577                                                    &info->next_hash);
578                         if (ret < 0)
579                                 return ret;
580                         if (ret == 0) {
581                                 ctx->pos = ext4_get_htree_eof(file);
582                                 break;
583                         }
584                         info->curr_node = rb_first(&info->root);
585                 }
586
587                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
588                 info->curr_hash = fname->hash;
589                 info->curr_minor_hash = fname->minor_hash;
590                 if (call_filldir(file, ctx, fname))
591                         break;
592         next_node:
593                 info->curr_node = rb_next(info->curr_node);
594                 if (info->curr_node) {
595                         fname = rb_entry(info->curr_node, struct fname,
596                                          rb_hash);
597                         info->curr_hash = fname->hash;
598                         info->curr_minor_hash = fname->minor_hash;
599                 } else {
600                         if (info->next_hash == ~0) {
601                                 ctx->pos = ext4_get_htree_eof(file);
602                                 break;
603                         }
604                         info->curr_hash = info->next_hash;
605                         info->curr_minor_hash = 0;
606                 }
607         }
608 finished:
609         info->last_pos = ctx->pos;
610         return 0;
611 }
612
613 static int ext4_dir_open(struct inode * inode, struct file * filp)
614 {
615         if (ext4_encrypted_inode(inode))
616                 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
617         return 0;
618 }
619
620 static int ext4_release_dir(struct inode *inode, struct file *filp)
621 {
622         if (filp->private_data)
623                 ext4_htree_free_dir_info(filp->private_data);
624
625         return 0;
626 }
627
628 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
629                       int buf_size)
630 {
631         struct ext4_dir_entry_2 *de;
632         int rlen;
633         unsigned int offset = 0;
634         char *top;
635
636         de = (struct ext4_dir_entry_2 *)buf;
637         top = buf + buf_size;
638         while ((char *) de < top) {
639                 if (ext4_check_dir_entry(dir, NULL, de, bh,
640                                          buf, buf_size, offset))
641                         return -EFSCORRUPTED;
642                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
643                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
644                 offset += rlen;
645         }
646         if ((char *) de > top)
647                 return -EFSCORRUPTED;
648
649         return 0;
650 }
651
652 const struct file_operations ext4_dir_operations = {
653         .llseek         = ext4_dir_llseek,
654         .read           = generic_read_dir,
655         .iterate_shared = ext4_readdir,
656         .unlocked_ioctl = ext4_ioctl,
657 #ifdef CONFIG_COMPAT
658         .compat_ioctl   = ext4_compat_ioctl,
659 #endif
660         .fsync          = ext4_sync_file,
661         .open           = ext4_dir_open,
662         .release        = ext4_release_dir,
663 };