ext4: improve error handling from ext4_dirhash()
[linux-block.git] / fs / ext4 / namei.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/namei.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/namei.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Big-endian to little-endian byte-swapping/bitmaps by
17  *        David S. Miller (davem@caip.rutgers.edu), 1995
18  *  Directory entry file type support and forward compatibility hooks
19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20  *  Hash Tree Directory indexing (c)
21  *      Daniel Phillips, 2001
22  *  Hash Tree Directory indexing porting
23  *      Christopher Li, 2002
24  *  Hash Tree Directory indexing cleanup
25  *      Theodore Ts'o, 2002
26  */
27
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
41
42 #include "xattr.h"
43 #include "acl.h"
44
45 #include <trace/events/ext4.h>
46 /*
47  * define how far ahead to read directories while searching them.
48  */
49 #define NAMEI_RA_CHUNKS  2
50 #define NAMEI_RA_BLOCKS  4
51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
53 static struct buffer_head *ext4_append(handle_t *handle,
54                                         struct inode *inode,
55                                         ext4_lblk_t *block)
56 {
57         struct ext4_map_blocks map;
58         struct buffer_head *bh;
59         int err;
60
61         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62                      ((inode->i_size >> 10) >=
63                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64                 return ERR_PTR(-ENOSPC);
65
66         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67         map.m_lblk = *block;
68         map.m_len = 1;
69
70         /*
71          * We're appending new directory block. Make sure the block is not
72          * allocated yet, otherwise we will end up corrupting the
73          * directory.
74          */
75         err = ext4_map_blocks(NULL, inode, &map, 0);
76         if (err < 0)
77                 return ERR_PTR(err);
78         if (err) {
79                 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80                 return ERR_PTR(-EFSCORRUPTED);
81         }
82
83         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84         if (IS_ERR(bh))
85                 return bh;
86         inode->i_size += inode->i_sb->s_blocksize;
87         EXT4_I(inode)->i_disksize = inode->i_size;
88         err = ext4_mark_inode_dirty(handle, inode);
89         if (err)
90                 goto out;
91         BUFFER_TRACE(bh, "get_write_access");
92         err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93                                             EXT4_JTR_NONE);
94         if (err)
95                 goto out;
96         return bh;
97
98 out:
99         brelse(bh);
100         ext4_std_error(inode->i_sb, err);
101         return ERR_PTR(err);
102 }
103
104 static int ext4_dx_csum_verify(struct inode *inode,
105                                struct ext4_dir_entry *dirent);
106
107 /*
108  * Hints to ext4_read_dirblock regarding whether we expect a directory
109  * block being read to be an index block, or a block containing
110  * directory entries (and if the latter, whether it was found via a
111  * logical block in an htree index block).  This is used to control
112  * what sort of sanity checkinig ext4_read_dirblock() will do on the
113  * directory block read from the storage device.  EITHER will means
114  * the caller doesn't know what kind of directory block will be read,
115  * so no specific verification will be done.
116  */
117 typedef enum {
118         EITHER, INDEX, DIRENT, DIRENT_HTREE
119 } dirblock_type_t;
120
121 #define ext4_read_dirblock(inode, block, type) \
122         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123
124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125                                                 ext4_lblk_t block,
126                                                 dirblock_type_t type,
127                                                 const char *func,
128                                                 unsigned int line)
129 {
130         struct buffer_head *bh;
131         struct ext4_dir_entry *dirent;
132         int is_dx_block = 0;
133
134         if (block >= inode->i_size >> inode->i_blkbits) {
135                 ext4_error_inode(inode, func, line, block,
136                        "Attempting to read directory block (%u) that is past i_size (%llu)",
137                        block, inode->i_size);
138                 return ERR_PTR(-EFSCORRUPTED);
139         }
140
141         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142                 bh = ERR_PTR(-EIO);
143         else
144                 bh = ext4_bread(NULL, inode, block, 0);
145         if (IS_ERR(bh)) {
146                 __ext4_warning(inode->i_sb, func, line,
147                                "inode #%lu: lblock %lu: comm %s: "
148                                "error %ld reading directory block",
149                                inode->i_ino, (unsigned long)block,
150                                current->comm, PTR_ERR(bh));
151
152                 return bh;
153         }
154         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
155                 ext4_error_inode(inode, func, line, block,
156                                  "Directory hole found for htree %s block",
157                                  (type == INDEX) ? "index" : "leaf");
158                 return ERR_PTR(-EFSCORRUPTED);
159         }
160         if (!bh)
161                 return NULL;
162         dirent = (struct ext4_dir_entry *) bh->b_data;
163         /* Determine whether or not we have an index block */
164         if (is_dx(inode)) {
165                 if (block == 0)
166                         is_dx_block = 1;
167                 else if (ext4_rec_len_from_disk(dirent->rec_len,
168                                                 inode->i_sb->s_blocksize) ==
169                          inode->i_sb->s_blocksize)
170                         is_dx_block = 1;
171         }
172         if (!is_dx_block && type == INDEX) {
173                 ext4_error_inode(inode, func, line, block,
174                        "directory leaf block found instead of index block");
175                 brelse(bh);
176                 return ERR_PTR(-EFSCORRUPTED);
177         }
178         if (!ext4_has_metadata_csum(inode->i_sb) ||
179             buffer_verified(bh))
180                 return bh;
181
182         /*
183          * An empty leaf block can get mistaken for a index block; for
184          * this reason, we can only check the index checksum when the
185          * caller is sure it should be an index block.
186          */
187         if (is_dx_block && type == INDEX) {
188                 if (ext4_dx_csum_verify(inode, dirent) &&
189                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
190                         set_buffer_verified(bh);
191                 else {
192                         ext4_error_inode_err(inode, func, line, block,
193                                              EFSBADCRC,
194                                              "Directory index failed checksum");
195                         brelse(bh);
196                         return ERR_PTR(-EFSBADCRC);
197                 }
198         }
199         if (!is_dx_block) {
200                 if (ext4_dirblock_csum_verify(inode, bh) &&
201                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
202                         set_buffer_verified(bh);
203                 else {
204                         ext4_error_inode_err(inode, func, line, block,
205                                              EFSBADCRC,
206                                              "Directory block failed checksum");
207                         brelse(bh);
208                         return ERR_PTR(-EFSBADCRC);
209                 }
210         }
211         return bh;
212 }
213
214 #ifdef DX_DEBUG
215 #define dxtrace(command) command
216 #else
217 #define dxtrace(command)
218 #endif
219
220 struct fake_dirent
221 {
222         __le32 inode;
223         __le16 rec_len;
224         u8 name_len;
225         u8 file_type;
226 };
227
228 struct dx_countlimit
229 {
230         __le16 limit;
231         __le16 count;
232 };
233
234 struct dx_entry
235 {
236         __le32 hash;
237         __le32 block;
238 };
239
240 /*
241  * dx_root_info is laid out so that if it should somehow get overlaid by a
242  * dirent the two low bits of the hash version will be zero.  Therefore, the
243  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
244  */
245
246 struct dx_root
247 {
248         struct fake_dirent dot;
249         char dot_name[4];
250         struct fake_dirent dotdot;
251         char dotdot_name[4];
252         struct dx_root_info
253         {
254                 __le32 reserved_zero;
255                 u8 hash_version;
256                 u8 info_length; /* 8 */
257                 u8 indirect_levels;
258                 u8 unused_flags;
259         }
260         info;
261         struct dx_entry entries[];
262 };
263
264 struct dx_node
265 {
266         struct fake_dirent fake;
267         struct dx_entry entries[];
268 };
269
270
271 struct dx_frame
272 {
273         struct buffer_head *bh;
274         struct dx_entry *entries;
275         struct dx_entry *at;
276 };
277
278 struct dx_map_entry
279 {
280         u32 hash;
281         u16 offs;
282         u16 size;
283 };
284
285 /*
286  * This goes at the end of each htree block.
287  */
288 struct dx_tail {
289         u32 dt_reserved;
290         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
291 };
292
293 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
294 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
295 static inline unsigned dx_get_hash(struct dx_entry *entry);
296 static void dx_set_hash(struct dx_entry *entry, unsigned value);
297 static unsigned dx_get_count(struct dx_entry *entries);
298 static unsigned dx_get_limit(struct dx_entry *entries);
299 static void dx_set_count(struct dx_entry *entries, unsigned value);
300 static void dx_set_limit(struct dx_entry *entries, unsigned value);
301 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
302 static unsigned dx_node_limit(struct inode *dir);
303 static struct dx_frame *dx_probe(struct ext4_filename *fname,
304                                  struct inode *dir,
305                                  struct dx_hash_info *hinfo,
306                                  struct dx_frame *frame);
307 static void dx_release(struct dx_frame *frames);
308 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
309                        struct dx_hash_info *hinfo,
310                        struct dx_map_entry *map_tail);
311 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
312 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
313                                         char *to, struct dx_map_entry *offsets,
314                                         int count, unsigned int blocksize);
315 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
316                                                 unsigned int blocksize);
317 static void dx_insert_block(struct dx_frame *frame,
318                                         u32 hash, ext4_lblk_t block);
319 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
320                                  struct dx_frame *frame,
321                                  struct dx_frame *frames,
322                                  __u32 *start_hash);
323 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
324                 struct ext4_filename *fname,
325                 struct ext4_dir_entry_2 **res_dir);
326 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
327                              struct inode *dir, struct inode *inode);
328
329 /* checksumming functions */
330 void ext4_initialize_dirent_tail(struct buffer_head *bh,
331                                  unsigned int blocksize)
332 {
333         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
334
335         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
336         t->det_rec_len = ext4_rec_len_to_disk(
337                         sizeof(struct ext4_dir_entry_tail), blocksize);
338         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
339 }
340
341 /* Walk through a dirent block to find a checksum "dirent" at the tail */
342 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
343                                                    struct buffer_head *bh)
344 {
345         struct ext4_dir_entry_tail *t;
346
347 #ifdef PARANOID
348         struct ext4_dir_entry *d, *top;
349
350         d = (struct ext4_dir_entry *)bh->b_data;
351         top = (struct ext4_dir_entry *)(bh->b_data +
352                 (EXT4_BLOCK_SIZE(inode->i_sb) -
353                  sizeof(struct ext4_dir_entry_tail)));
354         while (d < top && d->rec_len)
355                 d = (struct ext4_dir_entry *)(((void *)d) +
356                     le16_to_cpu(d->rec_len));
357
358         if (d != top)
359                 return NULL;
360
361         t = (struct ext4_dir_entry_tail *)d;
362 #else
363         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
364 #endif
365
366         if (t->det_reserved_zero1 ||
367             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
368             t->det_reserved_zero2 ||
369             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
370                 return NULL;
371
372         return t;
373 }
374
375 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
376 {
377         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
378         struct ext4_inode_info *ei = EXT4_I(inode);
379         __u32 csum;
380
381         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
382         return cpu_to_le32(csum);
383 }
384
385 #define warn_no_space_for_csum(inode)                                   \
386         __warn_no_space_for_csum((inode), __func__, __LINE__)
387
388 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
389                                      unsigned int line)
390 {
391         __ext4_warning_inode(inode, func, line,
392                 "No space for directory leaf checksum. Please run e2fsck -D.");
393 }
394
395 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
396 {
397         struct ext4_dir_entry_tail *t;
398
399         if (!ext4_has_metadata_csum(inode->i_sb))
400                 return 1;
401
402         t = get_dirent_tail(inode, bh);
403         if (!t) {
404                 warn_no_space_for_csum(inode);
405                 return 0;
406         }
407
408         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
409                                                   (char *)t - bh->b_data))
410                 return 0;
411
412         return 1;
413 }
414
415 static void ext4_dirblock_csum_set(struct inode *inode,
416                                  struct buffer_head *bh)
417 {
418         struct ext4_dir_entry_tail *t;
419
420         if (!ext4_has_metadata_csum(inode->i_sb))
421                 return;
422
423         t = get_dirent_tail(inode, bh);
424         if (!t) {
425                 warn_no_space_for_csum(inode);
426                 return;
427         }
428
429         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
430                                              (char *)t - bh->b_data);
431 }
432
433 int ext4_handle_dirty_dirblock(handle_t *handle,
434                                struct inode *inode,
435                                struct buffer_head *bh)
436 {
437         ext4_dirblock_csum_set(inode, bh);
438         return ext4_handle_dirty_metadata(handle, inode, bh);
439 }
440
441 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
442                                                struct ext4_dir_entry *dirent,
443                                                int *offset)
444 {
445         struct ext4_dir_entry *dp;
446         struct dx_root_info *root;
447         int count_offset;
448
449         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
450                 count_offset = 8;
451         else if (le16_to_cpu(dirent->rec_len) == 12) {
452                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
453                 if (le16_to_cpu(dp->rec_len) !=
454                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
455                         return NULL;
456                 root = (struct dx_root_info *)(((void *)dp + 12));
457                 if (root->reserved_zero ||
458                     root->info_length != sizeof(struct dx_root_info))
459                         return NULL;
460                 count_offset = 32;
461         } else
462                 return NULL;
463
464         if (offset)
465                 *offset = count_offset;
466         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
467 }
468
469 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
470                            int count_offset, int count, struct dx_tail *t)
471 {
472         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
473         struct ext4_inode_info *ei = EXT4_I(inode);
474         __u32 csum;
475         int size;
476         __u32 dummy_csum = 0;
477         int offset = offsetof(struct dx_tail, dt_checksum);
478
479         size = count_offset + (count * sizeof(struct dx_entry));
480         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
481         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
482         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
483
484         return cpu_to_le32(csum);
485 }
486
487 static int ext4_dx_csum_verify(struct inode *inode,
488                                struct ext4_dir_entry *dirent)
489 {
490         struct dx_countlimit *c;
491         struct dx_tail *t;
492         int count_offset, limit, count;
493
494         if (!ext4_has_metadata_csum(inode->i_sb))
495                 return 1;
496
497         c = get_dx_countlimit(inode, dirent, &count_offset);
498         if (!c) {
499                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
500                 return 0;
501         }
502         limit = le16_to_cpu(c->limit);
503         count = le16_to_cpu(c->count);
504         if (count_offset + (limit * sizeof(struct dx_entry)) >
505             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
506                 warn_no_space_for_csum(inode);
507                 return 0;
508         }
509         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
510
511         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
512                                             count, t))
513                 return 0;
514         return 1;
515 }
516
517 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
518 {
519         struct dx_countlimit *c;
520         struct dx_tail *t;
521         int count_offset, limit, count;
522
523         if (!ext4_has_metadata_csum(inode->i_sb))
524                 return;
525
526         c = get_dx_countlimit(inode, dirent, &count_offset);
527         if (!c) {
528                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
529                 return;
530         }
531         limit = le16_to_cpu(c->limit);
532         count = le16_to_cpu(c->count);
533         if (count_offset + (limit * sizeof(struct dx_entry)) >
534             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
535                 warn_no_space_for_csum(inode);
536                 return;
537         }
538         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
539
540         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
541 }
542
543 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
544                                             struct inode *inode,
545                                             struct buffer_head *bh)
546 {
547         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
548         return ext4_handle_dirty_metadata(handle, inode, bh);
549 }
550
551 /*
552  * p is at least 6 bytes before the end of page
553  */
554 static inline struct ext4_dir_entry_2 *
555 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
556 {
557         return (struct ext4_dir_entry_2 *)((char *)p +
558                 ext4_rec_len_from_disk(p->rec_len, blocksize));
559 }
560
561 /*
562  * Future: use high four bits of block for coalesce-on-delete flags
563  * Mask them off for now.
564  */
565
566 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
567 {
568         return le32_to_cpu(entry->block) & 0x0fffffff;
569 }
570
571 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
572 {
573         entry->block = cpu_to_le32(value);
574 }
575
576 static inline unsigned dx_get_hash(struct dx_entry *entry)
577 {
578         return le32_to_cpu(entry->hash);
579 }
580
581 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
582 {
583         entry->hash = cpu_to_le32(value);
584 }
585
586 static inline unsigned dx_get_count(struct dx_entry *entries)
587 {
588         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
589 }
590
591 static inline unsigned dx_get_limit(struct dx_entry *entries)
592 {
593         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
594 }
595
596 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
597 {
598         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
599 }
600
601 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
602 {
603         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
604 }
605
606 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
607 {
608         unsigned int entry_space = dir->i_sb->s_blocksize -
609                         ext4_dir_rec_len(1, NULL) -
610                         ext4_dir_rec_len(2, NULL) - infosize;
611
612         if (ext4_has_metadata_csum(dir->i_sb))
613                 entry_space -= sizeof(struct dx_tail);
614         return entry_space / sizeof(struct dx_entry);
615 }
616
617 static inline unsigned dx_node_limit(struct inode *dir)
618 {
619         unsigned int entry_space = dir->i_sb->s_blocksize -
620                         ext4_dir_rec_len(0, dir);
621
622         if (ext4_has_metadata_csum(dir->i_sb))
623                 entry_space -= sizeof(struct dx_tail);
624         return entry_space / sizeof(struct dx_entry);
625 }
626
627 /*
628  * Debug
629  */
630 #ifdef DX_DEBUG
631 static void dx_show_index(char * label, struct dx_entry *entries)
632 {
633         int i, n = dx_get_count (entries);
634         printk(KERN_DEBUG "%s index", label);
635         for (i = 0; i < n; i++) {
636                 printk(KERN_CONT " %x->%lu",
637                        i ? dx_get_hash(entries + i) : 0,
638                        (unsigned long)dx_get_block(entries + i));
639         }
640         printk(KERN_CONT "\n");
641 }
642
643 struct stats
644 {
645         unsigned names;
646         unsigned space;
647         unsigned bcount;
648 };
649
650 static struct stats dx_show_leaf(struct inode *dir,
651                                 struct dx_hash_info *hinfo,
652                                 struct ext4_dir_entry_2 *de,
653                                 int size, int show_names)
654 {
655         unsigned names = 0, space = 0;
656         char *base = (char *) de;
657         struct dx_hash_info h = *hinfo;
658
659         printk("names: ");
660         while ((char *) de < base + size)
661         {
662                 if (de->inode)
663                 {
664                         if (show_names)
665                         {
666 #ifdef CONFIG_FS_ENCRYPTION
667                                 int len;
668                                 char *name;
669                                 struct fscrypt_str fname_crypto_str =
670                                         FSTR_INIT(NULL, 0);
671                                 int res = 0;
672
673                                 name  = de->name;
674                                 len = de->name_len;
675                                 if (!IS_ENCRYPTED(dir)) {
676                                         /* Directory is not encrypted */
677                                         (void) ext4fs_dirhash(dir, de->name,
678                                                 de->name_len, &h);
679                                         printk("%*.s:(U)%x.%u ", len,
680                                                name, h.hash,
681                                                (unsigned) ((char *) de
682                                                            - base));
683                                 } else {
684                                         struct fscrypt_str de_name =
685                                                 FSTR_INIT(name, len);
686
687                                         /* Directory is encrypted */
688                                         res = fscrypt_fname_alloc_buffer(
689                                                 len, &fname_crypto_str);
690                                         if (res)
691                                                 printk(KERN_WARNING "Error "
692                                                         "allocating crypto "
693                                                         "buffer--skipping "
694                                                         "crypto\n");
695                                         res = fscrypt_fname_disk_to_usr(dir,
696                                                 0, 0, &de_name,
697                                                 &fname_crypto_str);
698                                         if (res) {
699                                                 printk(KERN_WARNING "Error "
700                                                         "converting filename "
701                                                         "from disk to usr"
702                                                         "\n");
703                                                 name = "??";
704                                                 len = 2;
705                                         } else {
706                                                 name = fname_crypto_str.name;
707                                                 len = fname_crypto_str.len;
708                                         }
709                                         if (IS_CASEFOLDED(dir))
710                                                 h.hash = EXT4_DIRENT_HASH(de);
711                                         else
712                                                 (void) ext4fs_dirhash(dir,
713                                                         de->name,
714                                                         de->name_len, &h);
715                                         printk("%*.s:(E)%x.%u ", len, name,
716                                                h.hash, (unsigned) ((char *) de
717                                                                    - base));
718                                         fscrypt_fname_free_buffer(
719                                                         &fname_crypto_str);
720                                 }
721 #else
722                                 int len = de->name_len;
723                                 char *name = de->name;
724                                 (void) ext4fs_dirhash(dir, de->name,
725                                                       de->name_len, &h);
726                                 printk("%*.s:%x.%u ", len, name, h.hash,
727                                        (unsigned) ((char *) de - base));
728 #endif
729                         }
730                         space += ext4_dir_rec_len(de->name_len, dir);
731                         names++;
732                 }
733                 de = ext4_next_entry(de, size);
734         }
735         printk(KERN_CONT "(%i)\n", names);
736         return (struct stats) { names, space, 1 };
737 }
738
739 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
740                              struct dx_entry *entries, int levels)
741 {
742         unsigned blocksize = dir->i_sb->s_blocksize;
743         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
744         unsigned bcount = 0;
745         struct buffer_head *bh;
746         printk("%i indexed blocks...\n", count);
747         for (i = 0; i < count; i++, entries++)
748         {
749                 ext4_lblk_t block = dx_get_block(entries);
750                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
751                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
752                 struct stats stats;
753                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
754                 bh = ext4_bread(NULL,dir, block, 0);
755                 if (!bh || IS_ERR(bh))
756                         continue;
757                 stats = levels?
758                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
759                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
760                         bh->b_data, blocksize, 0);
761                 names += stats.names;
762                 space += stats.space;
763                 bcount += stats.bcount;
764                 brelse(bh);
765         }
766         if (bcount)
767                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
768                        levels ? "" : "   ", names, space/bcount,
769                        (space/bcount)*100/blocksize);
770         return (struct stats) { names, space, bcount};
771 }
772
773 /*
774  * Linear search cross check
775  */
776 static inline void htree_rep_invariant_check(struct dx_entry *at,
777                                              struct dx_entry *target,
778                                              u32 hash, unsigned int n)
779 {
780         while (n--) {
781                 dxtrace(printk(KERN_CONT ","));
782                 if (dx_get_hash(++at) > hash) {
783                         at--;
784                         break;
785                 }
786         }
787         ASSERT(at == target - 1);
788 }
789 #else /* DX_DEBUG */
790 static inline void htree_rep_invariant_check(struct dx_entry *at,
791                                              struct dx_entry *target,
792                                              u32 hash, unsigned int n)
793 {
794 }
795 #endif /* DX_DEBUG */
796
797 /*
798  * Probe for a directory leaf block to search.
799  *
800  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
801  * error in the directory index, and the caller should fall back to
802  * searching the directory normally.  The callers of dx_probe **MUST**
803  * check for this error code, and make sure it never gets reflected
804  * back to userspace.
805  */
806 static struct dx_frame *
807 dx_probe(struct ext4_filename *fname, struct inode *dir,
808          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
809 {
810         unsigned count, indirect, level, i;
811         struct dx_entry *at, *entries, *p, *q, *m;
812         struct dx_root *root;
813         struct dx_frame *frame = frame_in;
814         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
815         u32 hash;
816         ext4_lblk_t block;
817         ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
818
819         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
820         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
821         if (IS_ERR(frame->bh))
822                 return (struct dx_frame *) frame->bh;
823
824         root = (struct dx_root *) frame->bh->b_data;
825         if (root->info.hash_version != DX_HASH_TEA &&
826             root->info.hash_version != DX_HASH_HALF_MD4 &&
827             root->info.hash_version != DX_HASH_LEGACY &&
828             root->info.hash_version != DX_HASH_SIPHASH) {
829                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
830                                    root->info.hash_version);
831                 goto fail;
832         }
833         if (ext4_hash_in_dirent(dir)) {
834                 if (root->info.hash_version != DX_HASH_SIPHASH) {
835                         ext4_warning_inode(dir,
836                                 "Hash in dirent, but hash is not SIPHASH");
837                         goto fail;
838                 }
839         } else {
840                 if (root->info.hash_version == DX_HASH_SIPHASH) {
841                         ext4_warning_inode(dir,
842                                 "Hash code is SIPHASH, but hash not in dirent");
843                         goto fail;
844                 }
845         }
846         if (fname)
847                 hinfo = &fname->hinfo;
848         hinfo->hash_version = root->info.hash_version;
849         if (hinfo->hash_version <= DX_HASH_TEA)
850                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
851         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
852         /* hash is already computed for encrypted casefolded directory */
853         if (fname && fname_name(fname) &&
854             !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
855                 int ret = ext4fs_dirhash(dir, fname_name(fname),
856                                          fname_len(fname), hinfo);
857                 if (ret < 0) {
858                         ret_err = ERR_PTR(ret);
859                         goto fail;
860                 }
861         }
862         hash = hinfo->hash;
863
864         if (root->info.unused_flags & 1) {
865                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
866                                    root->info.unused_flags);
867                 goto fail;
868         }
869
870         indirect = root->info.indirect_levels;
871         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
872                 ext4_warning(dir->i_sb,
873                              "Directory (ino: %lu) htree depth %#06x exceed"
874                              "supported value", dir->i_ino,
875                              ext4_dir_htree_level(dir->i_sb));
876                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
877                         ext4_warning(dir->i_sb, "Enable large directory "
878                                                 "feature to access it");
879                 }
880                 goto fail;
881         }
882
883         entries = (struct dx_entry *)(((char *)&root->info) +
884                                       root->info.info_length);
885
886         if (dx_get_limit(entries) != dx_root_limit(dir,
887                                                    root->info.info_length)) {
888                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
889                                    dx_get_limit(entries),
890                                    dx_root_limit(dir, root->info.info_length));
891                 goto fail;
892         }
893
894         dxtrace(printk("Look up %x", hash));
895         level = 0;
896         blocks[0] = 0;
897         while (1) {
898                 count = dx_get_count(entries);
899                 if (!count || count > dx_get_limit(entries)) {
900                         ext4_warning_inode(dir,
901                                            "dx entry: count %u beyond limit %u",
902                                            count, dx_get_limit(entries));
903                         goto fail;
904                 }
905
906                 p = entries + 1;
907                 q = entries + count - 1;
908                 while (p <= q) {
909                         m = p + (q - p) / 2;
910                         dxtrace(printk(KERN_CONT "."));
911                         if (dx_get_hash(m) > hash)
912                                 q = m - 1;
913                         else
914                                 p = m + 1;
915                 }
916
917                 htree_rep_invariant_check(entries, p, hash, count - 1);
918
919                 at = p - 1;
920                 dxtrace(printk(KERN_CONT " %x->%u\n",
921                                at == entries ? 0 : dx_get_hash(at),
922                                dx_get_block(at)));
923                 frame->entries = entries;
924                 frame->at = at;
925
926                 block = dx_get_block(at);
927                 for (i = 0; i <= level; i++) {
928                         if (blocks[i] == block) {
929                                 ext4_warning_inode(dir,
930                                         "dx entry: tree cycle block %u points back to block %u",
931                                         blocks[level], block);
932                                 goto fail;
933                         }
934                 }
935                 if (++level > indirect)
936                         return frame;
937                 blocks[level] = block;
938                 frame++;
939                 frame->bh = ext4_read_dirblock(dir, block, INDEX);
940                 if (IS_ERR(frame->bh)) {
941                         ret_err = (struct dx_frame *) frame->bh;
942                         frame->bh = NULL;
943                         goto fail;
944                 }
945
946                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
947
948                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
949                         ext4_warning_inode(dir,
950                                 "dx entry: limit %u != node limit %u",
951                                 dx_get_limit(entries), dx_node_limit(dir));
952                         goto fail;
953                 }
954         }
955 fail:
956         while (frame >= frame_in) {
957                 brelse(frame->bh);
958                 frame--;
959         }
960
961         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
962                 ext4_warning_inode(dir,
963                         "Corrupt directory, running e2fsck is recommended");
964         return ret_err;
965 }
966
967 static void dx_release(struct dx_frame *frames)
968 {
969         struct dx_root_info *info;
970         int i;
971         unsigned int indirect_levels;
972
973         if (frames[0].bh == NULL)
974                 return;
975
976         info = &((struct dx_root *)frames[0].bh->b_data)->info;
977         /* save local copy, "info" may be freed after brelse() */
978         indirect_levels = info->indirect_levels;
979         for (i = 0; i <= indirect_levels; i++) {
980                 if (frames[i].bh == NULL)
981                         break;
982                 brelse(frames[i].bh);
983                 frames[i].bh = NULL;
984         }
985 }
986
987 /*
988  * This function increments the frame pointer to search the next leaf
989  * block, and reads in the necessary intervening nodes if the search
990  * should be necessary.  Whether or not the search is necessary is
991  * controlled by the hash parameter.  If the hash value is even, then
992  * the search is only continued if the next block starts with that
993  * hash value.  This is used if we are searching for a specific file.
994  *
995  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
996  *
997  * This function returns 1 if the caller should continue to search,
998  * or 0 if it should not.  If there is an error reading one of the
999  * index blocks, it will a negative error code.
1000  *
1001  * If start_hash is non-null, it will be filled in with the starting
1002  * hash of the next page.
1003  */
1004 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1005                                  struct dx_frame *frame,
1006                                  struct dx_frame *frames,
1007                                  __u32 *start_hash)
1008 {
1009         struct dx_frame *p;
1010         struct buffer_head *bh;
1011         int num_frames = 0;
1012         __u32 bhash;
1013
1014         p = frame;
1015         /*
1016          * Find the next leaf page by incrementing the frame pointer.
1017          * If we run out of entries in the interior node, loop around and
1018          * increment pointer in the parent node.  When we break out of
1019          * this loop, num_frames indicates the number of interior
1020          * nodes need to be read.
1021          */
1022         while (1) {
1023                 if (++(p->at) < p->entries + dx_get_count(p->entries))
1024                         break;
1025                 if (p == frames)
1026                         return 0;
1027                 num_frames++;
1028                 p--;
1029         }
1030
1031         /*
1032          * If the hash is 1, then continue only if the next page has a
1033          * continuation hash of any value.  This is used for readdir
1034          * handling.  Otherwise, check to see if the hash matches the
1035          * desired continuation hash.  If it doesn't, return since
1036          * there's no point to read in the successive index pages.
1037          */
1038         bhash = dx_get_hash(p->at);
1039         if (start_hash)
1040                 *start_hash = bhash;
1041         if ((hash & 1) == 0) {
1042                 if ((bhash & ~1) != hash)
1043                         return 0;
1044         }
1045         /*
1046          * If the hash is HASH_NB_ALWAYS, we always go to the next
1047          * block so no check is necessary
1048          */
1049         while (num_frames--) {
1050                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1051                 if (IS_ERR(bh))
1052                         return PTR_ERR(bh);
1053                 p++;
1054                 brelse(p->bh);
1055                 p->bh = bh;
1056                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1057         }
1058         return 1;
1059 }
1060
1061
1062 /*
1063  * This function fills a red-black tree with information from a
1064  * directory block.  It returns the number directory entries loaded
1065  * into the tree.  If there is an error it is returned in err.
1066  */
1067 static int htree_dirblock_to_tree(struct file *dir_file,
1068                                   struct inode *dir, ext4_lblk_t block,
1069                                   struct dx_hash_info *hinfo,
1070                                   __u32 start_hash, __u32 start_minor_hash)
1071 {
1072         struct buffer_head *bh;
1073         struct ext4_dir_entry_2 *de, *top;
1074         int err = 0, count = 0;
1075         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1076         int csum = ext4_has_metadata_csum(dir->i_sb);
1077
1078         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1079                                                         (unsigned long)block));
1080         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1081         if (IS_ERR(bh))
1082                 return PTR_ERR(bh);
1083
1084         de = (struct ext4_dir_entry_2 *) bh->b_data;
1085         /* csum entries are not larger in the casefolded encrypted case */
1086         top = (struct ext4_dir_entry_2 *) ((char *) de +
1087                                            dir->i_sb->s_blocksize -
1088                                            ext4_dir_rec_len(0,
1089                                                            csum ? NULL : dir));
1090         /* Check if the directory is encrypted */
1091         if (IS_ENCRYPTED(dir)) {
1092                 err = fscrypt_prepare_readdir(dir);
1093                 if (err < 0) {
1094                         brelse(bh);
1095                         return err;
1096                 }
1097                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1098                                                  &fname_crypto_str);
1099                 if (err < 0) {
1100                         brelse(bh);
1101                         return err;
1102                 }
1103         }
1104
1105         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1106                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1107                                 bh->b_data, bh->b_size,
1108                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1109                                          + ((char *)de - bh->b_data))) {
1110                         /* silently ignore the rest of the block */
1111                         break;
1112                 }
1113                 if (ext4_hash_in_dirent(dir)) {
1114                         if (de->name_len && de->inode) {
1115                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1116                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1117                         } else {
1118                                 hinfo->hash = 0;
1119                                 hinfo->minor_hash = 0;
1120                         }
1121                 } else {
1122                         err = ext4fs_dirhash(dir, de->name,
1123                                              de->name_len, hinfo);
1124                         if (err < 0) {
1125                                 count = err;
1126                                 goto errout;
1127                         }
1128                 }
1129                 if ((hinfo->hash < start_hash) ||
1130                     ((hinfo->hash == start_hash) &&
1131                      (hinfo->minor_hash < start_minor_hash)))
1132                         continue;
1133                 if (de->inode == 0)
1134                         continue;
1135                 if (!IS_ENCRYPTED(dir)) {
1136                         tmp_str.name = de->name;
1137                         tmp_str.len = de->name_len;
1138                         err = ext4_htree_store_dirent(dir_file,
1139                                    hinfo->hash, hinfo->minor_hash, de,
1140                                    &tmp_str);
1141                 } else {
1142                         int save_len = fname_crypto_str.len;
1143                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1144                                                                 de->name_len);
1145
1146                         /* Directory is encrypted */
1147                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1148                                         hinfo->minor_hash, &de_name,
1149                                         &fname_crypto_str);
1150                         if (err) {
1151                                 count = err;
1152                                 goto errout;
1153                         }
1154                         err = ext4_htree_store_dirent(dir_file,
1155                                    hinfo->hash, hinfo->minor_hash, de,
1156                                         &fname_crypto_str);
1157                         fname_crypto_str.len = save_len;
1158                 }
1159                 if (err != 0) {
1160                         count = err;
1161                         goto errout;
1162                 }
1163                 count++;
1164         }
1165 errout:
1166         brelse(bh);
1167         fscrypt_fname_free_buffer(&fname_crypto_str);
1168         return count;
1169 }
1170
1171
1172 /*
1173  * This function fills a red-black tree with information from a
1174  * directory.  We start scanning the directory in hash order, starting
1175  * at start_hash and start_minor_hash.
1176  *
1177  * This function returns the number of entries inserted into the tree,
1178  * or a negative error code.
1179  */
1180 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1181                          __u32 start_minor_hash, __u32 *next_hash)
1182 {
1183         struct dx_hash_info hinfo;
1184         struct ext4_dir_entry_2 *de;
1185         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1186         struct inode *dir;
1187         ext4_lblk_t block;
1188         int count = 0;
1189         int ret, err;
1190         __u32 hashval;
1191         struct fscrypt_str tmp_str;
1192
1193         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1194                        start_hash, start_minor_hash));
1195         dir = file_inode(dir_file);
1196         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1197                 if (ext4_hash_in_dirent(dir))
1198                         hinfo.hash_version = DX_HASH_SIPHASH;
1199                 else
1200                         hinfo.hash_version =
1201                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1202                 if (hinfo.hash_version <= DX_HASH_TEA)
1203                         hinfo.hash_version +=
1204                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1205                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1206                 if (ext4_has_inline_data(dir)) {
1207                         int has_inline_data = 1;
1208                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1209                                                        &hinfo, start_hash,
1210                                                        start_minor_hash,
1211                                                        &has_inline_data);
1212                         if (has_inline_data) {
1213                                 *next_hash = ~0;
1214                                 return count;
1215                         }
1216                 }
1217                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1218                                                start_hash, start_minor_hash);
1219                 *next_hash = ~0;
1220                 return count;
1221         }
1222         hinfo.hash = start_hash;
1223         hinfo.minor_hash = 0;
1224         frame = dx_probe(NULL, dir, &hinfo, frames);
1225         if (IS_ERR(frame))
1226                 return PTR_ERR(frame);
1227
1228         /* Add '.' and '..' from the htree header */
1229         if (!start_hash && !start_minor_hash) {
1230                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1231                 tmp_str.name = de->name;
1232                 tmp_str.len = de->name_len;
1233                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1234                                               de, &tmp_str);
1235                 if (err != 0)
1236                         goto errout;
1237                 count++;
1238         }
1239         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1240                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1241                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1242                 tmp_str.name = de->name;
1243                 tmp_str.len = de->name_len;
1244                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1245                                               de, &tmp_str);
1246                 if (err != 0)
1247                         goto errout;
1248                 count++;
1249         }
1250
1251         while (1) {
1252                 if (fatal_signal_pending(current)) {
1253                         err = -ERESTARTSYS;
1254                         goto errout;
1255                 }
1256                 cond_resched();
1257                 block = dx_get_block(frame->at);
1258                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1259                                              start_hash, start_minor_hash);
1260                 if (ret < 0) {
1261                         err = ret;
1262                         goto errout;
1263                 }
1264                 count += ret;
1265                 hashval = ~0;
1266                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1267                                             frame, frames, &hashval);
1268                 *next_hash = hashval;
1269                 if (ret < 0) {
1270                         err = ret;
1271                         goto errout;
1272                 }
1273                 /*
1274                  * Stop if:  (a) there are no more entries, or
1275                  * (b) we have inserted at least one entry and the
1276                  * next hash value is not a continuation
1277                  */
1278                 if ((ret == 0) ||
1279                     (count && ((hashval & 1) == 0)))
1280                         break;
1281         }
1282         dx_release(frames);
1283         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1284                        "next hash: %x\n", count, *next_hash));
1285         return count;
1286 errout:
1287         dx_release(frames);
1288         return (err);
1289 }
1290
1291 static inline int search_dirblock(struct buffer_head *bh,
1292                                   struct inode *dir,
1293                                   struct ext4_filename *fname,
1294                                   unsigned int offset,
1295                                   struct ext4_dir_entry_2 **res_dir)
1296 {
1297         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1298                                fname, offset, res_dir);
1299 }
1300
1301 /*
1302  * Directory block splitting, compacting
1303  */
1304
1305 /*
1306  * Create map of hash values, offsets, and sizes, stored at end of block.
1307  * Returns number of entries mapped.
1308  */
1309 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1310                        struct dx_hash_info *hinfo,
1311                        struct dx_map_entry *map_tail)
1312 {
1313         int count = 0;
1314         struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1315         unsigned int buflen = bh->b_size;
1316         char *base = bh->b_data;
1317         struct dx_hash_info h = *hinfo;
1318
1319         if (ext4_has_metadata_csum(dir->i_sb))
1320                 buflen -= sizeof(struct ext4_dir_entry_tail);
1321
1322         while ((char *) de < base + buflen) {
1323                 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1324                                          ((char *)de) - base))
1325                         return -EFSCORRUPTED;
1326                 if (de->name_len && de->inode) {
1327                         if (ext4_hash_in_dirent(dir))
1328                                 h.hash = EXT4_DIRENT_HASH(de);
1329                         else {
1330                                 int err = ext4fs_dirhash(dir, de->name,
1331                                                      de->name_len, &h);
1332                                 if (err < 0)
1333                                         return err;
1334                         }
1335                         map_tail--;
1336                         map_tail->hash = h.hash;
1337                         map_tail->offs = ((char *) de - base)>>2;
1338                         map_tail->size = le16_to_cpu(de->rec_len);
1339                         count++;
1340                         cond_resched();
1341                 }
1342                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1343         }
1344         return count;
1345 }
1346
1347 /* Sort map by hash value */
1348 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1349 {
1350         struct dx_map_entry *p, *q, *top = map + count - 1;
1351         int more;
1352         /* Combsort until bubble sort doesn't suck */
1353         while (count > 2) {
1354                 count = count*10/13;
1355                 if (count - 9 < 2) /* 9, 10 -> 11 */
1356                         count = 11;
1357                 for (p = top, q = p - count; q >= map; p--, q--)
1358                         if (p->hash < q->hash)
1359                                 swap(*p, *q);
1360         }
1361         /* Garden variety bubble sort */
1362         do {
1363                 more = 0;
1364                 q = top;
1365                 while (q-- > map) {
1366                         if (q[1].hash >= q[0].hash)
1367                                 continue;
1368                         swap(*(q+1), *q);
1369                         more = 1;
1370                 }
1371         } while(more);
1372 }
1373
1374 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1375 {
1376         struct dx_entry *entries = frame->entries;
1377         struct dx_entry *old = frame->at, *new = old + 1;
1378         int count = dx_get_count(entries);
1379
1380         ASSERT(count < dx_get_limit(entries));
1381         ASSERT(old < entries + count);
1382         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1383         dx_set_hash(new, hash);
1384         dx_set_block(new, block);
1385         dx_set_count(entries, count + 1);
1386 }
1387
1388 #if IS_ENABLED(CONFIG_UNICODE)
1389 /*
1390  * Test whether a case-insensitive directory entry matches the filename
1391  * being searched for.  If quick is set, assume the name being looked up
1392  * is already in the casefolded form.
1393  *
1394  * Returns: 0 if the directory entry matches, more than 0 if it
1395  * doesn't match or less than zero on error.
1396  */
1397 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1398                            u8 *de_name, size_t de_name_len, bool quick)
1399 {
1400         const struct super_block *sb = parent->i_sb;
1401         const struct unicode_map *um = sb->s_encoding;
1402         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1403         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1404         int ret;
1405
1406         if (IS_ENCRYPTED(parent)) {
1407                 const struct fscrypt_str encrypted_name =
1408                                 FSTR_INIT(de_name, de_name_len);
1409
1410                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1411                 if (!decrypted_name.name)
1412                         return -ENOMEM;
1413                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1414                                                 &decrypted_name);
1415                 if (ret < 0)
1416                         goto out;
1417                 entry.name = decrypted_name.name;
1418                 entry.len = decrypted_name.len;
1419         }
1420
1421         if (quick)
1422                 ret = utf8_strncasecmp_folded(um, name, &entry);
1423         else
1424                 ret = utf8_strncasecmp(um, name, &entry);
1425         if (ret < 0) {
1426                 /* Handle invalid character sequence as either an error
1427                  * or as an opaque byte sequence.
1428                  */
1429                 if (sb_has_strict_encoding(sb))
1430                         ret = -EINVAL;
1431                 else if (name->len != entry.len)
1432                         ret = 1;
1433                 else
1434                         ret = !!memcmp(name->name, entry.name, entry.len);
1435         }
1436 out:
1437         kfree(decrypted_name.name);
1438         return ret;
1439 }
1440
1441 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1442                                   struct ext4_filename *name)
1443 {
1444         struct fscrypt_str *cf_name = &name->cf_name;
1445         struct dx_hash_info *hinfo = &name->hinfo;
1446         int len;
1447
1448         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1449             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1450                 cf_name->name = NULL;
1451                 return 0;
1452         }
1453
1454         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1455         if (!cf_name->name)
1456                 return -ENOMEM;
1457
1458         len = utf8_casefold(dir->i_sb->s_encoding,
1459                             iname, cf_name->name,
1460                             EXT4_NAME_LEN);
1461         if (len <= 0) {
1462                 kfree(cf_name->name);
1463                 cf_name->name = NULL;
1464         }
1465         cf_name->len = (unsigned) len;
1466         if (!IS_ENCRYPTED(dir))
1467                 return 0;
1468
1469         hinfo->hash_version = DX_HASH_SIPHASH;
1470         hinfo->seed = NULL;
1471         if (cf_name->name)
1472                 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1473         else
1474                 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1475 }
1476 #endif
1477
1478 /*
1479  * Test whether a directory entry matches the filename being searched for.
1480  *
1481  * Return: %true if the directory entry matches, otherwise %false.
1482  */
1483 static bool ext4_match(struct inode *parent,
1484                               const struct ext4_filename *fname,
1485                               struct ext4_dir_entry_2 *de)
1486 {
1487         struct fscrypt_name f;
1488
1489         if (!de->inode)
1490                 return false;
1491
1492         f.usr_fname = fname->usr_fname;
1493         f.disk_name = fname->disk_name;
1494 #ifdef CONFIG_FS_ENCRYPTION
1495         f.crypto_buf = fname->crypto_buf;
1496 #endif
1497
1498 #if IS_ENABLED(CONFIG_UNICODE)
1499         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1500             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1501                 if (fname->cf_name.name) {
1502                         struct qstr cf = {.name = fname->cf_name.name,
1503                                           .len = fname->cf_name.len};
1504                         if (IS_ENCRYPTED(parent)) {
1505                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1506                                         fname->hinfo.minor_hash !=
1507                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1508
1509                                         return false;
1510                                 }
1511                         }
1512                         return !ext4_ci_compare(parent, &cf, de->name,
1513                                                         de->name_len, true);
1514                 }
1515                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1516                                                 de->name_len, false);
1517         }
1518 #endif
1519
1520         return fscrypt_match_name(&f, de->name, de->name_len);
1521 }
1522
1523 /*
1524  * Returns 0 if not found, -1 on failure, and 1 on success
1525  */
1526 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1527                     struct inode *dir, struct ext4_filename *fname,
1528                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1529 {
1530         struct ext4_dir_entry_2 * de;
1531         char * dlimit;
1532         int de_len;
1533
1534         de = (struct ext4_dir_entry_2 *)search_buf;
1535         dlimit = search_buf + buf_size;
1536         while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1537                 /* this code is executed quadratically often */
1538                 /* do minimal checking `by hand' */
1539                 if (de->name + de->name_len <= dlimit &&
1540                     ext4_match(dir, fname, de)) {
1541                         /* found a match - just to be sure, do
1542                          * a full check */
1543                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1544                                                  buf_size, offset))
1545                                 return -1;
1546                         *res_dir = de;
1547                         return 1;
1548                 }
1549                 /* prevent looping on a bad block */
1550                 de_len = ext4_rec_len_from_disk(de->rec_len,
1551                                                 dir->i_sb->s_blocksize);
1552                 if (de_len <= 0)
1553                         return -1;
1554                 offset += de_len;
1555                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1556         }
1557         return 0;
1558 }
1559
1560 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1561                                struct ext4_dir_entry *de)
1562 {
1563         struct super_block *sb = dir->i_sb;
1564
1565         if (!is_dx(dir))
1566                 return 0;
1567         if (block == 0)
1568                 return 1;
1569         if (de->inode == 0 &&
1570             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1571                         sb->s_blocksize)
1572                 return 1;
1573         return 0;
1574 }
1575
1576 /*
1577  *      __ext4_find_entry()
1578  *
1579  * finds an entry in the specified directory with the wanted name. It
1580  * returns the cache buffer in which the entry was found, and the entry
1581  * itself (as a parameter - res_dir). It does NOT read the inode of the
1582  * entry - you'll have to do that yourself if you want to.
1583  *
1584  * The returned buffer_head has ->b_count elevated.  The caller is expected
1585  * to brelse() it when appropriate.
1586  */
1587 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1588                                              struct ext4_filename *fname,
1589                                              struct ext4_dir_entry_2 **res_dir,
1590                                              int *inlined)
1591 {
1592         struct super_block *sb;
1593         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1594         struct buffer_head *bh, *ret = NULL;
1595         ext4_lblk_t start, block;
1596         const u8 *name = fname->usr_fname->name;
1597         size_t ra_max = 0;      /* Number of bh's in the readahead
1598                                    buffer, bh_use[] */
1599         size_t ra_ptr = 0;      /* Current index into readahead
1600                                    buffer */
1601         ext4_lblk_t  nblocks;
1602         int i, namelen, retval;
1603
1604         *res_dir = NULL;
1605         sb = dir->i_sb;
1606         namelen = fname->usr_fname->len;
1607         if (namelen > EXT4_NAME_LEN)
1608                 return NULL;
1609
1610         if (ext4_has_inline_data(dir)) {
1611                 int has_inline_data = 1;
1612                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1613                                              &has_inline_data);
1614                 if (inlined)
1615                         *inlined = has_inline_data;
1616                 if (has_inline_data)
1617                         goto cleanup_and_exit;
1618         }
1619
1620         if ((namelen <= 2) && (name[0] == '.') &&
1621             (name[1] == '.' || name[1] == '\0')) {
1622                 /*
1623                  * "." or ".." will only be in the first block
1624                  * NFS may look up ".."; "." should be handled by the VFS
1625                  */
1626                 block = start = 0;
1627                 nblocks = 1;
1628                 goto restart;
1629         }
1630         if (is_dx(dir)) {
1631                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1632                 /*
1633                  * On success, or if the error was file not found,
1634                  * return.  Otherwise, fall back to doing a search the
1635                  * old fashioned way.
1636                  */
1637                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1638                         goto cleanup_and_exit;
1639                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1640                                "falling back\n"));
1641                 ret = NULL;
1642         }
1643         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1644         if (!nblocks) {
1645                 ret = NULL;
1646                 goto cleanup_and_exit;
1647         }
1648         start = EXT4_I(dir)->i_dir_start_lookup;
1649         if (start >= nblocks)
1650                 start = 0;
1651         block = start;
1652 restart:
1653         do {
1654                 /*
1655                  * We deal with the read-ahead logic here.
1656                  */
1657                 cond_resched();
1658                 if (ra_ptr >= ra_max) {
1659                         /* Refill the readahead buffer */
1660                         ra_ptr = 0;
1661                         if (block < start)
1662                                 ra_max = start - block;
1663                         else
1664                                 ra_max = nblocks - block;
1665                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1666                         retval = ext4_bread_batch(dir, block, ra_max,
1667                                                   false /* wait */, bh_use);
1668                         if (retval) {
1669                                 ret = ERR_PTR(retval);
1670                                 ra_max = 0;
1671                                 goto cleanup_and_exit;
1672                         }
1673                 }
1674                 if ((bh = bh_use[ra_ptr++]) == NULL)
1675                         goto next;
1676                 wait_on_buffer(bh);
1677                 if (!buffer_uptodate(bh)) {
1678                         EXT4_ERROR_INODE_ERR(dir, EIO,
1679                                              "reading directory lblock %lu",
1680                                              (unsigned long) block);
1681                         brelse(bh);
1682                         ret = ERR_PTR(-EIO);
1683                         goto cleanup_and_exit;
1684                 }
1685                 if (!buffer_verified(bh) &&
1686                     !is_dx_internal_node(dir, block,
1687                                          (struct ext4_dir_entry *)bh->b_data) &&
1688                     !ext4_dirblock_csum_verify(dir, bh)) {
1689                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1690                                              "checksumming directory "
1691                                              "block %lu", (unsigned long)block);
1692                         brelse(bh);
1693                         ret = ERR_PTR(-EFSBADCRC);
1694                         goto cleanup_and_exit;
1695                 }
1696                 set_buffer_verified(bh);
1697                 i = search_dirblock(bh, dir, fname,
1698                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1699                 if (i == 1) {
1700                         EXT4_I(dir)->i_dir_start_lookup = block;
1701                         ret = bh;
1702                         goto cleanup_and_exit;
1703                 } else {
1704                         brelse(bh);
1705                         if (i < 0)
1706                                 goto cleanup_and_exit;
1707                 }
1708         next:
1709                 if (++block >= nblocks)
1710                         block = 0;
1711         } while (block != start);
1712
1713         /*
1714          * If the directory has grown while we were searching, then
1715          * search the last part of the directory before giving up.
1716          */
1717         block = nblocks;
1718         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1719         if (block < nblocks) {
1720                 start = 0;
1721                 goto restart;
1722         }
1723
1724 cleanup_and_exit:
1725         /* Clean up the read-ahead blocks */
1726         for (; ra_ptr < ra_max; ra_ptr++)
1727                 brelse(bh_use[ra_ptr]);
1728         return ret;
1729 }
1730
1731 static struct buffer_head *ext4_find_entry(struct inode *dir,
1732                                            const struct qstr *d_name,
1733                                            struct ext4_dir_entry_2 **res_dir,
1734                                            int *inlined)
1735 {
1736         int err;
1737         struct ext4_filename fname;
1738         struct buffer_head *bh;
1739
1740         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1741         if (err == -ENOENT)
1742                 return NULL;
1743         if (err)
1744                 return ERR_PTR(err);
1745
1746         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1747
1748         ext4_fname_free_filename(&fname);
1749         return bh;
1750 }
1751
1752 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1753                                              struct dentry *dentry,
1754                                              struct ext4_dir_entry_2 **res_dir)
1755 {
1756         int err;
1757         struct ext4_filename fname;
1758         struct buffer_head *bh;
1759
1760         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1761         generic_set_encrypted_ci_d_ops(dentry);
1762         if (err == -ENOENT)
1763                 return NULL;
1764         if (err)
1765                 return ERR_PTR(err);
1766
1767         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1768
1769         ext4_fname_free_filename(&fname);
1770         return bh;
1771 }
1772
1773 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1774                         struct ext4_filename *fname,
1775                         struct ext4_dir_entry_2 **res_dir)
1776 {
1777         struct super_block * sb = dir->i_sb;
1778         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1779         struct buffer_head *bh;
1780         ext4_lblk_t block;
1781         int retval;
1782
1783 #ifdef CONFIG_FS_ENCRYPTION
1784         *res_dir = NULL;
1785 #endif
1786         frame = dx_probe(fname, dir, NULL, frames);
1787         if (IS_ERR(frame))
1788                 return (struct buffer_head *) frame;
1789         do {
1790                 block = dx_get_block(frame->at);
1791                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1792                 if (IS_ERR(bh))
1793                         goto errout;
1794
1795                 retval = search_dirblock(bh, dir, fname,
1796                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1797                                          res_dir);
1798                 if (retval == 1)
1799                         goto success;
1800                 brelse(bh);
1801                 if (retval == -1) {
1802                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1803                         goto errout;
1804                 }
1805
1806                 /* Check to see if we should continue to search */
1807                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1808                                                frames, NULL);
1809                 if (retval < 0) {
1810                         ext4_warning_inode(dir,
1811                                 "error %d reading directory index block",
1812                                 retval);
1813                         bh = ERR_PTR(retval);
1814                         goto errout;
1815                 }
1816         } while (retval == 1);
1817
1818         bh = NULL;
1819 errout:
1820         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1821 success:
1822         dx_release(frames);
1823         return bh;
1824 }
1825
1826 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1827 {
1828         struct inode *inode;
1829         struct ext4_dir_entry_2 *de;
1830         struct buffer_head *bh;
1831
1832         if (dentry->d_name.len > EXT4_NAME_LEN)
1833                 return ERR_PTR(-ENAMETOOLONG);
1834
1835         bh = ext4_lookup_entry(dir, dentry, &de);
1836         if (IS_ERR(bh))
1837                 return ERR_CAST(bh);
1838         inode = NULL;
1839         if (bh) {
1840                 __u32 ino = le32_to_cpu(de->inode);
1841                 brelse(bh);
1842                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1843                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1844                         return ERR_PTR(-EFSCORRUPTED);
1845                 }
1846                 if (unlikely(ino == dir->i_ino)) {
1847                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1848                                          dentry);
1849                         return ERR_PTR(-EFSCORRUPTED);
1850                 }
1851                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1852                 if (inode == ERR_PTR(-ESTALE)) {
1853                         EXT4_ERROR_INODE(dir,
1854                                          "deleted inode referenced: %u",
1855                                          ino);
1856                         return ERR_PTR(-EFSCORRUPTED);
1857                 }
1858                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1859                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1860                     !fscrypt_has_permitted_context(dir, inode)) {
1861                         ext4_warning(inode->i_sb,
1862                                      "Inconsistent encryption contexts: %lu/%lu",
1863                                      dir->i_ino, inode->i_ino);
1864                         iput(inode);
1865                         return ERR_PTR(-EPERM);
1866                 }
1867         }
1868
1869 #if IS_ENABLED(CONFIG_UNICODE)
1870         if (!inode && IS_CASEFOLDED(dir)) {
1871                 /* Eventually we want to call d_add_ci(dentry, NULL)
1872                  * for negative dentries in the encoding case as
1873                  * well.  For now, prevent the negative dentry
1874                  * from being cached.
1875                  */
1876                 return NULL;
1877         }
1878 #endif
1879         return d_splice_alias(inode, dentry);
1880 }
1881
1882
1883 struct dentry *ext4_get_parent(struct dentry *child)
1884 {
1885         __u32 ino;
1886         struct ext4_dir_entry_2 * de;
1887         struct buffer_head *bh;
1888
1889         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1890         if (IS_ERR(bh))
1891                 return ERR_CAST(bh);
1892         if (!bh)
1893                 return ERR_PTR(-ENOENT);
1894         ino = le32_to_cpu(de->inode);
1895         brelse(bh);
1896
1897         if (!ext4_valid_inum(child->d_sb, ino)) {
1898                 EXT4_ERROR_INODE(d_inode(child),
1899                                  "bad parent inode number: %u", ino);
1900                 return ERR_PTR(-EFSCORRUPTED);
1901         }
1902
1903         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1904 }
1905
1906 /*
1907  * Move count entries from end of map between two memory locations.
1908  * Returns pointer to last entry moved.
1909  */
1910 static struct ext4_dir_entry_2 *
1911 dx_move_dirents(struct inode *dir, char *from, char *to,
1912                 struct dx_map_entry *map, int count,
1913                 unsigned blocksize)
1914 {
1915         unsigned rec_len = 0;
1916
1917         while (count--) {
1918                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1919                                                 (from + (map->offs<<2));
1920                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1921
1922                 memcpy (to, de, rec_len);
1923                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1924                                 ext4_rec_len_to_disk(rec_len, blocksize);
1925
1926                 /* wipe dir_entry excluding the rec_len field */
1927                 de->inode = 0;
1928                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1929                                                                 blocksize) -
1930                                          offsetof(struct ext4_dir_entry_2,
1931                                                                 name_len));
1932
1933                 map++;
1934                 to += rec_len;
1935         }
1936         return (struct ext4_dir_entry_2 *) (to - rec_len);
1937 }
1938
1939 /*
1940  * Compact each dir entry in the range to the minimal rec_len.
1941  * Returns pointer to last entry in range.
1942  */
1943 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1944                                                         unsigned int blocksize)
1945 {
1946         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1947         unsigned rec_len = 0;
1948
1949         prev = to = de;
1950         while ((char*)de < base + blocksize) {
1951                 next = ext4_next_entry(de, blocksize);
1952                 if (de->inode && de->name_len) {
1953                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1954                         if (de > to)
1955                                 memmove(to, de, rec_len);
1956                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1957                         prev = to;
1958                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1959                 }
1960                 de = next;
1961         }
1962         return prev;
1963 }
1964
1965 /*
1966  * Split a full leaf block to make room for a new dir entry.
1967  * Allocate a new block, and move entries so that they are approx. equally full.
1968  * Returns pointer to de in block into which the new entry will be inserted.
1969  */
1970 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1971                         struct buffer_head **bh,struct dx_frame *frame,
1972                         struct dx_hash_info *hinfo)
1973 {
1974         unsigned blocksize = dir->i_sb->s_blocksize;
1975         unsigned continued;
1976         int count;
1977         struct buffer_head *bh2;
1978         ext4_lblk_t newblock;
1979         u32 hash2;
1980         struct dx_map_entry *map;
1981         char *data1 = (*bh)->b_data, *data2;
1982         unsigned split, move, size;
1983         struct ext4_dir_entry_2 *de = NULL, *de2;
1984         int     csum_size = 0;
1985         int     err = 0, i;
1986
1987         if (ext4_has_metadata_csum(dir->i_sb))
1988                 csum_size = sizeof(struct ext4_dir_entry_tail);
1989
1990         bh2 = ext4_append(handle, dir, &newblock);
1991         if (IS_ERR(bh2)) {
1992                 brelse(*bh);
1993                 *bh = NULL;
1994                 return (struct ext4_dir_entry_2 *) bh2;
1995         }
1996
1997         BUFFER_TRACE(*bh, "get_write_access");
1998         err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1999                                             EXT4_JTR_NONE);
2000         if (err)
2001                 goto journal_error;
2002
2003         BUFFER_TRACE(frame->bh, "get_write_access");
2004         err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2005                                             EXT4_JTR_NONE);
2006         if (err)
2007                 goto journal_error;
2008
2009         data2 = bh2->b_data;
2010
2011         /* create map in the end of data2 block */
2012         map = (struct dx_map_entry *) (data2 + blocksize);
2013         count = dx_make_map(dir, *bh, hinfo, map);
2014         if (count < 0) {
2015                 err = count;
2016                 goto journal_error;
2017         }
2018         map -= count;
2019         dx_sort_map(map, count);
2020         /* Ensure that neither split block is over half full */
2021         size = 0;
2022         move = 0;
2023         for (i = count-1; i >= 0; i--) {
2024                 /* is more than half of this entry in 2nd half of the block? */
2025                 if (size + map[i].size/2 > blocksize/2)
2026                         break;
2027                 size += map[i].size;
2028                 move++;
2029         }
2030         /*
2031          * map index at which we will split
2032          *
2033          * If the sum of active entries didn't exceed half the block size, just
2034          * split it in half by count; each resulting block will have at least
2035          * half the space free.
2036          */
2037         if (i > 0)
2038                 split = count - move;
2039         else
2040                 split = count/2;
2041
2042         hash2 = map[split].hash;
2043         continued = hash2 == map[split - 1].hash;
2044         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2045                         (unsigned long)dx_get_block(frame->at),
2046                                         hash2, split, count-split));
2047
2048         /* Fancy dance to stay within two buffers */
2049         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2050                               blocksize);
2051         de = dx_pack_dirents(dir, data1, blocksize);
2052         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2053                                            (char *) de,
2054                                            blocksize);
2055         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2056                                             (char *) de2,
2057                                             blocksize);
2058         if (csum_size) {
2059                 ext4_initialize_dirent_tail(*bh, blocksize);
2060                 ext4_initialize_dirent_tail(bh2, blocksize);
2061         }
2062
2063         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2064                         blocksize, 1));
2065         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2066                         blocksize, 1));
2067
2068         /* Which block gets the new entry? */
2069         if (hinfo->hash >= hash2) {
2070                 swap(*bh, bh2);
2071                 de = de2;
2072         }
2073         dx_insert_block(frame, hash2 + continued, newblock);
2074         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2075         if (err)
2076                 goto journal_error;
2077         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2078         if (err)
2079                 goto journal_error;
2080         brelse(bh2);
2081         dxtrace(dx_show_index("frame", frame->entries));
2082         return de;
2083
2084 journal_error:
2085         brelse(*bh);
2086         brelse(bh2);
2087         *bh = NULL;
2088         ext4_std_error(dir->i_sb, err);
2089         return ERR_PTR(err);
2090 }
2091
2092 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2093                       struct buffer_head *bh,
2094                       void *buf, int buf_size,
2095                       struct ext4_filename *fname,
2096                       struct ext4_dir_entry_2 **dest_de)
2097 {
2098         struct ext4_dir_entry_2 *de;
2099         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2100         int nlen, rlen;
2101         unsigned int offset = 0;
2102         char *top;
2103
2104         de = buf;
2105         top = buf + buf_size - reclen;
2106         while ((char *) de <= top) {
2107                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2108                                          buf, buf_size, offset))
2109                         return -EFSCORRUPTED;
2110                 if (ext4_match(dir, fname, de))
2111                         return -EEXIST;
2112                 nlen = ext4_dir_rec_len(de->name_len, dir);
2113                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2114                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2115                         break;
2116                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2117                 offset += rlen;
2118         }
2119         if ((char *) de > top)
2120                 return -ENOSPC;
2121
2122         *dest_de = de;
2123         return 0;
2124 }
2125
2126 void ext4_insert_dentry(struct inode *dir,
2127                         struct inode *inode,
2128                         struct ext4_dir_entry_2 *de,
2129                         int buf_size,
2130                         struct ext4_filename *fname)
2131 {
2132
2133         int nlen, rlen;
2134
2135         nlen = ext4_dir_rec_len(de->name_len, dir);
2136         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2137         if (de->inode) {
2138                 struct ext4_dir_entry_2 *de1 =
2139                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2140                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2141                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2142                 de = de1;
2143         }
2144         de->file_type = EXT4_FT_UNKNOWN;
2145         de->inode = cpu_to_le32(inode->i_ino);
2146         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2147         de->name_len = fname_len(fname);
2148         memcpy(de->name, fname_name(fname), fname_len(fname));
2149         if (ext4_hash_in_dirent(dir)) {
2150                 struct dx_hash_info *hinfo = &fname->hinfo;
2151
2152                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2153                 EXT4_DIRENT_HASHES(de)->minor_hash =
2154                                                 cpu_to_le32(hinfo->minor_hash);
2155         }
2156 }
2157
2158 /*
2159  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2160  * it points to a directory entry which is guaranteed to be large
2161  * enough for new directory entry.  If de is NULL, then
2162  * add_dirent_to_buf will attempt search the directory block for
2163  * space.  It will return -ENOSPC if no space is available, and -EIO
2164  * and -EEXIST if directory entry already exists.
2165  */
2166 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2167                              struct inode *dir,
2168                              struct inode *inode, struct ext4_dir_entry_2 *de,
2169                              struct buffer_head *bh)
2170 {
2171         unsigned int    blocksize = dir->i_sb->s_blocksize;
2172         int             csum_size = 0;
2173         int             err, err2;
2174
2175         if (ext4_has_metadata_csum(inode->i_sb))
2176                 csum_size = sizeof(struct ext4_dir_entry_tail);
2177
2178         if (!de) {
2179                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2180                                         blocksize - csum_size, fname, &de);
2181                 if (err)
2182                         return err;
2183         }
2184         BUFFER_TRACE(bh, "get_write_access");
2185         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2186                                             EXT4_JTR_NONE);
2187         if (err) {
2188                 ext4_std_error(dir->i_sb, err);
2189                 return err;
2190         }
2191
2192         /* By now the buffer is marked for journaling */
2193         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2194
2195         /*
2196          * XXX shouldn't update any times until successful
2197          * completion of syscall, but too many callers depend
2198          * on this.
2199          *
2200          * XXX similarly, too many callers depend on
2201          * ext4_new_inode() setting the times, but error
2202          * recovery deletes the inode, so the worst that can
2203          * happen is that the times are slightly out of date
2204          * and/or different from the directory change time.
2205          */
2206         dir->i_mtime = dir->i_ctime = current_time(dir);
2207         ext4_update_dx_flag(dir);
2208         inode_inc_iversion(dir);
2209         err2 = ext4_mark_inode_dirty(handle, dir);
2210         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2211         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2212         if (err)
2213                 ext4_std_error(dir->i_sb, err);
2214         return err ? err : err2;
2215 }
2216
2217 /*
2218  * This converts a one block unindexed directory to a 3 block indexed
2219  * directory, and adds the dentry to the indexed directory.
2220  */
2221 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2222                             struct inode *dir,
2223                             struct inode *inode, struct buffer_head *bh)
2224 {
2225         struct buffer_head *bh2;
2226         struct dx_root  *root;
2227         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2228         struct dx_entry *entries;
2229         struct ext4_dir_entry_2 *de, *de2;
2230         char            *data2, *top;
2231         unsigned        len;
2232         int             retval;
2233         unsigned        blocksize;
2234         ext4_lblk_t  block;
2235         struct fake_dirent *fde;
2236         int csum_size = 0;
2237
2238         if (ext4_has_metadata_csum(inode->i_sb))
2239                 csum_size = sizeof(struct ext4_dir_entry_tail);
2240
2241         blocksize =  dir->i_sb->s_blocksize;
2242         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2243         BUFFER_TRACE(bh, "get_write_access");
2244         retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2245                                                EXT4_JTR_NONE);
2246         if (retval) {
2247                 ext4_std_error(dir->i_sb, retval);
2248                 brelse(bh);
2249                 return retval;
2250         }
2251         root = (struct dx_root *) bh->b_data;
2252
2253         /* The 0th block becomes the root, move the dirents out */
2254         fde = &root->dotdot;
2255         de = (struct ext4_dir_entry_2 *)((char *)fde +
2256                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2257         if ((char *) de >= (((char *) root) + blocksize)) {
2258                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2259                 brelse(bh);
2260                 return -EFSCORRUPTED;
2261         }
2262         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2263
2264         /* Allocate new block for the 0th block's dirents */
2265         bh2 = ext4_append(handle, dir, &block);
2266         if (IS_ERR(bh2)) {
2267                 brelse(bh);
2268                 return PTR_ERR(bh2);
2269         }
2270         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2271         data2 = bh2->b_data;
2272
2273         memcpy(data2, de, len);
2274         memset(de, 0, len); /* wipe old data */
2275         de = (struct ext4_dir_entry_2 *) data2;
2276         top = data2 + len;
2277         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2278                 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2279                                          (data2 + (blocksize - csum_size) -
2280                                           (char *) de))) {
2281                         brelse(bh2);
2282                         brelse(bh);
2283                         return -EFSCORRUPTED;
2284                 }
2285                 de = de2;
2286         }
2287         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2288                                            (char *) de, blocksize);
2289
2290         if (csum_size)
2291                 ext4_initialize_dirent_tail(bh2, blocksize);
2292
2293         /* Initialize the root; the dot dirents already exist */
2294         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2295         de->rec_len = ext4_rec_len_to_disk(
2296                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2297         memset (&root->info, 0, sizeof(root->info));
2298         root->info.info_length = sizeof(root->info);
2299         if (ext4_hash_in_dirent(dir))
2300                 root->info.hash_version = DX_HASH_SIPHASH;
2301         else
2302                 root->info.hash_version =
2303                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2304
2305         entries = root->entries;
2306         dx_set_block(entries, 1);
2307         dx_set_count(entries, 1);
2308         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2309
2310         /* Initialize as for dx_probe */
2311         fname->hinfo.hash_version = root->info.hash_version;
2312         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2313                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2314         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2315
2316         /* casefolded encrypted hashes are computed on fname setup */
2317         if (!ext4_hash_in_dirent(dir)) {
2318                 int err = ext4fs_dirhash(dir, fname_name(fname),
2319                                          fname_len(fname), &fname->hinfo);
2320                 if (err < 0) {
2321                         brelse(bh2);
2322                         brelse(bh);
2323                         return err;
2324                 }
2325         }
2326         memset(frames, 0, sizeof(frames));
2327         frame = frames;
2328         frame->entries = entries;
2329         frame->at = entries;
2330         frame->bh = bh;
2331
2332         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2333         if (retval)
2334                 goto out_frames;
2335         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2336         if (retval)
2337                 goto out_frames;
2338
2339         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2340         if (IS_ERR(de)) {
2341                 retval = PTR_ERR(de);
2342                 goto out_frames;
2343         }
2344
2345         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2346 out_frames:
2347         /*
2348          * Even if the block split failed, we have to properly write
2349          * out all the changes we did so far. Otherwise we can end up
2350          * with corrupted filesystem.
2351          */
2352         if (retval)
2353                 ext4_mark_inode_dirty(handle, dir);
2354         dx_release(frames);
2355         brelse(bh2);
2356         return retval;
2357 }
2358
2359 /*
2360  *      ext4_add_entry()
2361  *
2362  * adds a file entry to the specified directory, using the same
2363  * semantics as ext4_find_entry(). It returns NULL if it failed.
2364  *
2365  * NOTE!! The inode part of 'de' is left at 0 - which means you
2366  * may not sleep between calling this and putting something into
2367  * the entry, as someone else might have used it while you slept.
2368  */
2369 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2370                           struct inode *inode)
2371 {
2372         struct inode *dir = d_inode(dentry->d_parent);
2373         struct buffer_head *bh = NULL;
2374         struct ext4_dir_entry_2 *de;
2375         struct super_block *sb;
2376         struct ext4_filename fname;
2377         int     retval;
2378         int     dx_fallback=0;
2379         unsigned blocksize;
2380         ext4_lblk_t block, blocks;
2381         int     csum_size = 0;
2382
2383         if (ext4_has_metadata_csum(inode->i_sb))
2384                 csum_size = sizeof(struct ext4_dir_entry_tail);
2385
2386         sb = dir->i_sb;
2387         blocksize = sb->s_blocksize;
2388         if (!dentry->d_name.len)
2389                 return -EINVAL;
2390
2391         if (fscrypt_is_nokey_name(dentry))
2392                 return -ENOKEY;
2393
2394 #if IS_ENABLED(CONFIG_UNICODE)
2395         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2396             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2397                 return -EINVAL;
2398 #endif
2399
2400         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2401         if (retval)
2402                 return retval;
2403
2404         if (ext4_has_inline_data(dir)) {
2405                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2406                 if (retval < 0)
2407                         goto out;
2408                 if (retval == 1) {
2409                         retval = 0;
2410                         goto out;
2411                 }
2412         }
2413
2414         if (is_dx(dir)) {
2415                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2416                 if (!retval || (retval != ERR_BAD_DX_DIR))
2417                         goto out;
2418                 /* Can we just ignore htree data? */
2419                 if (ext4_has_metadata_csum(sb)) {
2420                         EXT4_ERROR_INODE(dir,
2421                                 "Directory has corrupted htree index.");
2422                         retval = -EFSCORRUPTED;
2423                         goto out;
2424                 }
2425                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2426                 dx_fallback++;
2427                 retval = ext4_mark_inode_dirty(handle, dir);
2428                 if (unlikely(retval))
2429                         goto out;
2430         }
2431         blocks = dir->i_size >> sb->s_blocksize_bits;
2432         for (block = 0; block < blocks; block++) {
2433                 bh = ext4_read_dirblock(dir, block, DIRENT);
2434                 if (bh == NULL) {
2435                         bh = ext4_bread(handle, dir, block,
2436                                         EXT4_GET_BLOCKS_CREATE);
2437                         goto add_to_new_block;
2438                 }
2439                 if (IS_ERR(bh)) {
2440                         retval = PTR_ERR(bh);
2441                         bh = NULL;
2442                         goto out;
2443                 }
2444                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2445                                            NULL, bh);
2446                 if (retval != -ENOSPC)
2447                         goto out;
2448
2449                 if (blocks == 1 && !dx_fallback &&
2450                     ext4_has_feature_dir_index(sb)) {
2451                         retval = make_indexed_dir(handle, &fname, dir,
2452                                                   inode, bh);
2453                         bh = NULL; /* make_indexed_dir releases bh */
2454                         goto out;
2455                 }
2456                 brelse(bh);
2457         }
2458         bh = ext4_append(handle, dir, &block);
2459 add_to_new_block:
2460         if (IS_ERR(bh)) {
2461                 retval = PTR_ERR(bh);
2462                 bh = NULL;
2463                 goto out;
2464         }
2465         de = (struct ext4_dir_entry_2 *) bh->b_data;
2466         de->inode = 0;
2467         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2468
2469         if (csum_size)
2470                 ext4_initialize_dirent_tail(bh, blocksize);
2471
2472         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2473 out:
2474         ext4_fname_free_filename(&fname);
2475         brelse(bh);
2476         if (retval == 0)
2477                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2478         return retval;
2479 }
2480
2481 /*
2482  * Returns 0 for success, or a negative error value
2483  */
2484 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2485                              struct inode *dir, struct inode *inode)
2486 {
2487         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2488         struct dx_entry *entries, *at;
2489         struct buffer_head *bh;
2490         struct super_block *sb = dir->i_sb;
2491         struct ext4_dir_entry_2 *de;
2492         int restart;
2493         int err;
2494
2495 again:
2496         restart = 0;
2497         frame = dx_probe(fname, dir, NULL, frames);
2498         if (IS_ERR(frame))
2499                 return PTR_ERR(frame);
2500         entries = frame->entries;
2501         at = frame->at;
2502         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2503         if (IS_ERR(bh)) {
2504                 err = PTR_ERR(bh);
2505                 bh = NULL;
2506                 goto cleanup;
2507         }
2508
2509         BUFFER_TRACE(bh, "get_write_access");
2510         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2511         if (err)
2512                 goto journal_error;
2513
2514         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2515         if (err != -ENOSPC)
2516                 goto cleanup;
2517
2518         err = 0;
2519         /* Block full, should compress but for now just split */
2520         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2521                        dx_get_count(entries), dx_get_limit(entries)));
2522         /* Need to split index? */
2523         if (dx_get_count(entries) == dx_get_limit(entries)) {
2524                 ext4_lblk_t newblock;
2525                 int levels = frame - frames + 1;
2526                 unsigned int icount;
2527                 int add_level = 1;
2528                 struct dx_entry *entries2;
2529                 struct dx_node *node2;
2530                 struct buffer_head *bh2;
2531
2532                 while (frame > frames) {
2533                         if (dx_get_count((frame - 1)->entries) <
2534                             dx_get_limit((frame - 1)->entries)) {
2535                                 add_level = 0;
2536                                 break;
2537                         }
2538                         frame--; /* split higher index block */
2539                         at = frame->at;
2540                         entries = frame->entries;
2541                         restart = 1;
2542                 }
2543                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2544                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2545                                          "reach max htree level :%d",
2546                                          dir->i_ino, levels);
2547                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2548                                 ext4_warning(sb, "Large directory feature is "
2549                                                  "not enabled on this "
2550                                                  "filesystem");
2551                         }
2552                         err = -ENOSPC;
2553                         goto cleanup;
2554                 }
2555                 icount = dx_get_count(entries);
2556                 bh2 = ext4_append(handle, dir, &newblock);
2557                 if (IS_ERR(bh2)) {
2558                         err = PTR_ERR(bh2);
2559                         goto cleanup;
2560                 }
2561                 node2 = (struct dx_node *)(bh2->b_data);
2562                 entries2 = node2->entries;
2563                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2564                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2565                                                            sb->s_blocksize);
2566                 BUFFER_TRACE(frame->bh, "get_write_access");
2567                 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2568                                                     EXT4_JTR_NONE);
2569                 if (err)
2570                         goto journal_error;
2571                 if (!add_level) {
2572                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2573                         unsigned hash2 = dx_get_hash(entries + icount1);
2574                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2575                                        icount1, icount2));
2576
2577                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2578                         err = ext4_journal_get_write_access(handle, sb,
2579                                                             (frame - 1)->bh,
2580                                                             EXT4_JTR_NONE);
2581                         if (err)
2582                                 goto journal_error;
2583
2584                         memcpy((char *) entries2, (char *) (entries + icount1),
2585                                icount2 * sizeof(struct dx_entry));
2586                         dx_set_count(entries, icount1);
2587                         dx_set_count(entries2, icount2);
2588                         dx_set_limit(entries2, dx_node_limit(dir));
2589
2590                         /* Which index block gets the new entry? */
2591                         if (at - entries >= icount1) {
2592                                 frame->at = at - entries - icount1 + entries2;
2593                                 frame->entries = entries = entries2;
2594                                 swap(frame->bh, bh2);
2595                         }
2596                         dx_insert_block((frame - 1), hash2, newblock);
2597                         dxtrace(dx_show_index("node", frame->entries));
2598                         dxtrace(dx_show_index("node",
2599                                ((struct dx_node *) bh2->b_data)->entries));
2600                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2601                         if (err)
2602                                 goto journal_error;
2603                         brelse (bh2);
2604                         err = ext4_handle_dirty_dx_node(handle, dir,
2605                                                    (frame - 1)->bh);
2606                         if (err)
2607                                 goto journal_error;
2608                         err = ext4_handle_dirty_dx_node(handle, dir,
2609                                                         frame->bh);
2610                         if (restart || err)
2611                                 goto journal_error;
2612                 } else {
2613                         struct dx_root *dxroot;
2614                         memcpy((char *) entries2, (char *) entries,
2615                                icount * sizeof(struct dx_entry));
2616                         dx_set_limit(entries2, dx_node_limit(dir));
2617
2618                         /* Set up root */
2619                         dx_set_count(entries, 1);
2620                         dx_set_block(entries + 0, newblock);
2621                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2622                         dxroot->info.indirect_levels += 1;
2623                         dxtrace(printk(KERN_DEBUG
2624                                        "Creating %d level index...\n",
2625                                        dxroot->info.indirect_levels));
2626                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2627                         if (err)
2628                                 goto journal_error;
2629                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2630                         brelse(bh2);
2631                         restart = 1;
2632                         goto journal_error;
2633                 }
2634         }
2635         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2636         if (IS_ERR(de)) {
2637                 err = PTR_ERR(de);
2638                 goto cleanup;
2639         }
2640         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2641         goto cleanup;
2642
2643 journal_error:
2644         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2645 cleanup:
2646         brelse(bh);
2647         dx_release(frames);
2648         /* @restart is true means htree-path has been changed, we need to
2649          * repeat dx_probe() to find out valid htree-path
2650          */
2651         if (restart && err == 0)
2652                 goto again;
2653         return err;
2654 }
2655
2656 /*
2657  * ext4_generic_delete_entry deletes a directory entry by merging it
2658  * with the previous entry
2659  */
2660 int ext4_generic_delete_entry(struct inode *dir,
2661                               struct ext4_dir_entry_2 *de_del,
2662                               struct buffer_head *bh,
2663                               void *entry_buf,
2664                               int buf_size,
2665                               int csum_size)
2666 {
2667         struct ext4_dir_entry_2 *de, *pde;
2668         unsigned int blocksize = dir->i_sb->s_blocksize;
2669         int i;
2670
2671         i = 0;
2672         pde = NULL;
2673         de = entry_buf;
2674         while (i < buf_size - csum_size) {
2675                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2676                                          entry_buf, buf_size, i))
2677                         return -EFSCORRUPTED;
2678                 if (de == de_del)  {
2679                         if (pde) {
2680                                 pde->rec_len = ext4_rec_len_to_disk(
2681                                         ext4_rec_len_from_disk(pde->rec_len,
2682                                                                blocksize) +
2683                                         ext4_rec_len_from_disk(de->rec_len,
2684                                                                blocksize),
2685                                         blocksize);
2686
2687                                 /* wipe entire dir_entry */
2688                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2689                                                                 blocksize));
2690                         } else {
2691                                 /* wipe dir_entry excluding the rec_len field */
2692                                 de->inode = 0;
2693                                 memset(&de->name_len, 0,
2694                                         ext4_rec_len_from_disk(de->rec_len,
2695                                                                 blocksize) -
2696                                         offsetof(struct ext4_dir_entry_2,
2697                                                                 name_len));
2698                         }
2699
2700                         inode_inc_iversion(dir);
2701                         return 0;
2702                 }
2703                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2704                 pde = de;
2705                 de = ext4_next_entry(de, blocksize);
2706         }
2707         return -ENOENT;
2708 }
2709
2710 static int ext4_delete_entry(handle_t *handle,
2711                              struct inode *dir,
2712                              struct ext4_dir_entry_2 *de_del,
2713                              struct buffer_head *bh)
2714 {
2715         int err, csum_size = 0;
2716
2717         if (ext4_has_inline_data(dir)) {
2718                 int has_inline_data = 1;
2719                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2720                                                &has_inline_data);
2721                 if (has_inline_data)
2722                         return err;
2723         }
2724
2725         if (ext4_has_metadata_csum(dir->i_sb))
2726                 csum_size = sizeof(struct ext4_dir_entry_tail);
2727
2728         BUFFER_TRACE(bh, "get_write_access");
2729         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2730                                             EXT4_JTR_NONE);
2731         if (unlikely(err))
2732                 goto out;
2733
2734         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2735                                         dir->i_sb->s_blocksize, csum_size);
2736         if (err)
2737                 goto out;
2738
2739         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2740         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2741         if (unlikely(err))
2742                 goto out;
2743
2744         return 0;
2745 out:
2746         if (err != -ENOENT)
2747                 ext4_std_error(dir->i_sb, err);
2748         return err;
2749 }
2750
2751 /*
2752  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2753  * since this indicates that nlinks count was previously 1 to avoid overflowing
2754  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2755  * that subdirectory link counts are not being maintained accurately.
2756  *
2757  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2758  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2759  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2760  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2761  */
2762 static void ext4_inc_count(struct inode *inode)
2763 {
2764         inc_nlink(inode);
2765         if (is_dx(inode) &&
2766             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2767                 set_nlink(inode, 1);
2768 }
2769
2770 /*
2771  * If a directory had nlink == 1, then we should let it be 1. This indicates
2772  * directory has >EXT4_LINK_MAX subdirs.
2773  */
2774 static void ext4_dec_count(struct inode *inode)
2775 {
2776         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2777                 drop_nlink(inode);
2778 }
2779
2780
2781 /*
2782  * Add non-directory inode to a directory. On success, the inode reference is
2783  * consumed by dentry is instantiation. This is also indicated by clearing of
2784  * *inodep pointer. On failure, the caller is responsible for dropping the
2785  * inode reference in the safe context.
2786  */
2787 static int ext4_add_nondir(handle_t *handle,
2788                 struct dentry *dentry, struct inode **inodep)
2789 {
2790         struct inode *dir = d_inode(dentry->d_parent);
2791         struct inode *inode = *inodep;
2792         int err = ext4_add_entry(handle, dentry, inode);
2793         if (!err) {
2794                 err = ext4_mark_inode_dirty(handle, inode);
2795                 if (IS_DIRSYNC(dir))
2796                         ext4_handle_sync(handle);
2797                 d_instantiate_new(dentry, inode);
2798                 *inodep = NULL;
2799                 return err;
2800         }
2801         drop_nlink(inode);
2802         ext4_orphan_add(handle, inode);
2803         unlock_new_inode(inode);
2804         return err;
2805 }
2806
2807 /*
2808  * By the time this is called, we already have created
2809  * the directory cache entry for the new file, but it
2810  * is so far negative - it has no inode.
2811  *
2812  * If the create succeeds, we fill in the inode information
2813  * with d_instantiate().
2814  */
2815 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2816                        struct dentry *dentry, umode_t mode, bool excl)
2817 {
2818         handle_t *handle;
2819         struct inode *inode;
2820         int err, credits, retries = 0;
2821
2822         err = dquot_initialize(dir);
2823         if (err)
2824                 return err;
2825
2826         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2827                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2828 retry:
2829         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2830                                             0, NULL, EXT4_HT_DIR, credits);
2831         handle = ext4_journal_current_handle();
2832         err = PTR_ERR(inode);
2833         if (!IS_ERR(inode)) {
2834                 inode->i_op = &ext4_file_inode_operations;
2835                 inode->i_fop = &ext4_file_operations;
2836                 ext4_set_aops(inode);
2837                 err = ext4_add_nondir(handle, dentry, &inode);
2838                 if (!err)
2839                         ext4_fc_track_create(handle, dentry);
2840         }
2841         if (handle)
2842                 ext4_journal_stop(handle);
2843         if (!IS_ERR_OR_NULL(inode))
2844                 iput(inode);
2845         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2846                 goto retry;
2847         return err;
2848 }
2849
2850 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2851                       struct dentry *dentry, umode_t mode, dev_t rdev)
2852 {
2853         handle_t *handle;
2854         struct inode *inode;
2855         int err, credits, retries = 0;
2856
2857         err = dquot_initialize(dir);
2858         if (err)
2859                 return err;
2860
2861         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2862                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2863 retry:
2864         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2865                                             0, NULL, EXT4_HT_DIR, credits);
2866         handle = ext4_journal_current_handle();
2867         err = PTR_ERR(inode);
2868         if (!IS_ERR(inode)) {
2869                 init_special_inode(inode, inode->i_mode, rdev);
2870                 inode->i_op = &ext4_special_inode_operations;
2871                 err = ext4_add_nondir(handle, dentry, &inode);
2872                 if (!err)
2873                         ext4_fc_track_create(handle, dentry);
2874         }
2875         if (handle)
2876                 ext4_journal_stop(handle);
2877         if (!IS_ERR_OR_NULL(inode))
2878                 iput(inode);
2879         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2880                 goto retry;
2881         return err;
2882 }
2883
2884 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2885                         struct file *file, umode_t mode)
2886 {
2887         handle_t *handle;
2888         struct inode *inode;
2889         int err, retries = 0;
2890
2891         err = dquot_initialize(dir);
2892         if (err)
2893                 return err;
2894
2895 retry:
2896         inode = ext4_new_inode_start_handle(idmap, dir, mode,
2897                                             NULL, 0, NULL,
2898                                             EXT4_HT_DIR,
2899                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2900                           4 + EXT4_XATTR_TRANS_BLOCKS);
2901         handle = ext4_journal_current_handle();
2902         err = PTR_ERR(inode);
2903         if (!IS_ERR(inode)) {
2904                 inode->i_op = &ext4_file_inode_operations;
2905                 inode->i_fop = &ext4_file_operations;
2906                 ext4_set_aops(inode);
2907                 d_tmpfile(file, inode);
2908                 err = ext4_orphan_add(handle, inode);
2909                 if (err)
2910                         goto err_unlock_inode;
2911                 mark_inode_dirty(inode);
2912                 unlock_new_inode(inode);
2913         }
2914         if (handle)
2915                 ext4_journal_stop(handle);
2916         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2917                 goto retry;
2918         return finish_open_simple(file, err);
2919 err_unlock_inode:
2920         ext4_journal_stop(handle);
2921         unlock_new_inode(inode);
2922         return err;
2923 }
2924
2925 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2926                           struct ext4_dir_entry_2 *de,
2927                           int blocksize, int csum_size,
2928                           unsigned int parent_ino, int dotdot_real_len)
2929 {
2930         de->inode = cpu_to_le32(inode->i_ino);
2931         de->name_len = 1;
2932         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2933                                            blocksize);
2934         strcpy(de->name, ".");
2935         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2936
2937         de = ext4_next_entry(de, blocksize);
2938         de->inode = cpu_to_le32(parent_ino);
2939         de->name_len = 2;
2940         if (!dotdot_real_len)
2941                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2942                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2943                                         blocksize);
2944         else
2945                 de->rec_len = ext4_rec_len_to_disk(
2946                                         ext4_dir_rec_len(de->name_len, NULL),
2947                                         blocksize);
2948         strcpy(de->name, "..");
2949         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2950
2951         return ext4_next_entry(de, blocksize);
2952 }
2953
2954 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2955                              struct inode *inode)
2956 {
2957         struct buffer_head *dir_block = NULL;
2958         struct ext4_dir_entry_2 *de;
2959         ext4_lblk_t block = 0;
2960         unsigned int blocksize = dir->i_sb->s_blocksize;
2961         int csum_size = 0;
2962         int err;
2963
2964         if (ext4_has_metadata_csum(dir->i_sb))
2965                 csum_size = sizeof(struct ext4_dir_entry_tail);
2966
2967         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2968                 err = ext4_try_create_inline_dir(handle, dir, inode);
2969                 if (err < 0 && err != -ENOSPC)
2970                         goto out;
2971                 if (!err)
2972                         goto out;
2973         }
2974
2975         inode->i_size = 0;
2976         dir_block = ext4_append(handle, inode, &block);
2977         if (IS_ERR(dir_block))
2978                 return PTR_ERR(dir_block);
2979         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2980         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2981         set_nlink(inode, 2);
2982         if (csum_size)
2983                 ext4_initialize_dirent_tail(dir_block, blocksize);
2984
2985         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2986         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2987         if (err)
2988                 goto out;
2989         set_buffer_verified(dir_block);
2990 out:
2991         brelse(dir_block);
2992         return err;
2993 }
2994
2995 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2996                       struct dentry *dentry, umode_t mode)
2997 {
2998         handle_t *handle;
2999         struct inode *inode;
3000         int err, err2 = 0, credits, retries = 0;
3001
3002         if (EXT4_DIR_LINK_MAX(dir))
3003                 return -EMLINK;
3004
3005         err = dquot_initialize(dir);
3006         if (err)
3007                 return err;
3008
3009         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3010                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3011 retry:
3012         inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3013                                             &dentry->d_name,
3014                                             0, NULL, EXT4_HT_DIR, credits);
3015         handle = ext4_journal_current_handle();
3016         err = PTR_ERR(inode);
3017         if (IS_ERR(inode))
3018                 goto out_stop;
3019
3020         inode->i_op = &ext4_dir_inode_operations;
3021         inode->i_fop = &ext4_dir_operations;
3022         err = ext4_init_new_dir(handle, dir, inode);
3023         if (err)
3024                 goto out_clear_inode;
3025         err = ext4_mark_inode_dirty(handle, inode);
3026         if (!err)
3027                 err = ext4_add_entry(handle, dentry, inode);
3028         if (err) {
3029 out_clear_inode:
3030                 clear_nlink(inode);
3031                 ext4_orphan_add(handle, inode);
3032                 unlock_new_inode(inode);
3033                 err2 = ext4_mark_inode_dirty(handle, inode);
3034                 if (unlikely(err2))
3035                         err = err2;
3036                 ext4_journal_stop(handle);
3037                 iput(inode);
3038                 goto out_retry;
3039         }
3040         ext4_inc_count(dir);
3041
3042         ext4_update_dx_flag(dir);
3043         err = ext4_mark_inode_dirty(handle, dir);
3044         if (err)
3045                 goto out_clear_inode;
3046         d_instantiate_new(dentry, inode);
3047         ext4_fc_track_create(handle, dentry);
3048         if (IS_DIRSYNC(dir))
3049                 ext4_handle_sync(handle);
3050
3051 out_stop:
3052         if (handle)
3053                 ext4_journal_stop(handle);
3054 out_retry:
3055         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3056                 goto retry;
3057         return err;
3058 }
3059
3060 /*
3061  * routine to check that the specified directory is empty (for rmdir)
3062  */
3063 bool ext4_empty_dir(struct inode *inode)
3064 {
3065         unsigned int offset;
3066         struct buffer_head *bh;
3067         struct ext4_dir_entry_2 *de;
3068         struct super_block *sb;
3069
3070         if (ext4_has_inline_data(inode)) {
3071                 int has_inline_data = 1;
3072                 int ret;
3073
3074                 ret = empty_inline_dir(inode, &has_inline_data);
3075                 if (has_inline_data)
3076                         return ret;
3077         }
3078
3079         sb = inode->i_sb;
3080         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3081                                         ext4_dir_rec_len(2, NULL)) {
3082                 EXT4_ERROR_INODE(inode, "invalid size");
3083                 return false;
3084         }
3085         /* The first directory block must not be a hole,
3086          * so treat it as DIRENT_HTREE
3087          */
3088         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3089         if (IS_ERR(bh))
3090                 return false;
3091
3092         de = (struct ext4_dir_entry_2 *) bh->b_data;
3093         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3094                                  0) ||
3095             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3096                 ext4_warning_inode(inode, "directory missing '.'");
3097                 brelse(bh);
3098                 return false;
3099         }
3100         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3101         de = ext4_next_entry(de, sb->s_blocksize);
3102         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3103                                  offset) ||
3104             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3105                 ext4_warning_inode(inode, "directory missing '..'");
3106                 brelse(bh);
3107                 return false;
3108         }
3109         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3110         while (offset < inode->i_size) {
3111                 if (!(offset & (sb->s_blocksize - 1))) {
3112                         unsigned int lblock;
3113                         brelse(bh);
3114                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3115                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3116                         if (bh == NULL) {
3117                                 offset += sb->s_blocksize;
3118                                 continue;
3119                         }
3120                         if (IS_ERR(bh))
3121                                 return false;
3122                 }
3123                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3124                                         (offset & (sb->s_blocksize - 1)));
3125                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3126                                          bh->b_data, bh->b_size, offset) ||
3127                     le32_to_cpu(de->inode)) {
3128                         brelse(bh);
3129                         return false;
3130                 }
3131                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3132         }
3133         brelse(bh);
3134         return true;
3135 }
3136
3137 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3138 {
3139         int retval;
3140         struct inode *inode;
3141         struct buffer_head *bh;
3142         struct ext4_dir_entry_2 *de;
3143         handle_t *handle = NULL;
3144
3145         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3146                 return -EIO;
3147
3148         /* Initialize quotas before so that eventual writes go in
3149          * separate transaction */
3150         retval = dquot_initialize(dir);
3151         if (retval)
3152                 return retval;
3153         retval = dquot_initialize(d_inode(dentry));
3154         if (retval)
3155                 return retval;
3156
3157         retval = -ENOENT;
3158         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3159         if (IS_ERR(bh))
3160                 return PTR_ERR(bh);
3161         if (!bh)
3162                 goto end_rmdir;
3163
3164         inode = d_inode(dentry);
3165
3166         retval = -EFSCORRUPTED;
3167         if (le32_to_cpu(de->inode) != inode->i_ino)
3168                 goto end_rmdir;
3169
3170         retval = -ENOTEMPTY;
3171         if (!ext4_empty_dir(inode))
3172                 goto end_rmdir;
3173
3174         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3175                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3176         if (IS_ERR(handle)) {
3177                 retval = PTR_ERR(handle);
3178                 handle = NULL;
3179                 goto end_rmdir;
3180         }
3181
3182         if (IS_DIRSYNC(dir))
3183                 ext4_handle_sync(handle);
3184
3185         retval = ext4_delete_entry(handle, dir, de, bh);
3186         if (retval)
3187                 goto end_rmdir;
3188         if (!EXT4_DIR_LINK_EMPTY(inode))
3189                 ext4_warning_inode(inode,
3190                              "empty directory '%.*s' has too many links (%u)",
3191                              dentry->d_name.len, dentry->d_name.name,
3192                              inode->i_nlink);
3193         inode_inc_iversion(inode);
3194         clear_nlink(inode);
3195         /* There's no need to set i_disksize: the fact that i_nlink is
3196          * zero will ensure that the right thing happens during any
3197          * recovery. */
3198         inode->i_size = 0;
3199         ext4_orphan_add(handle, inode);
3200         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3201         retval = ext4_mark_inode_dirty(handle, inode);
3202         if (retval)
3203                 goto end_rmdir;
3204         ext4_dec_count(dir);
3205         ext4_update_dx_flag(dir);
3206         ext4_fc_track_unlink(handle, dentry);
3207         retval = ext4_mark_inode_dirty(handle, dir);
3208
3209 #if IS_ENABLED(CONFIG_UNICODE)
3210         /* VFS negative dentries are incompatible with Encoding and
3211          * Case-insensitiveness. Eventually we'll want avoid
3212          * invalidating the dentries here, alongside with returning the
3213          * negative dentries at ext4_lookup(), when it is better
3214          * supported by the VFS for the CI case.
3215          */
3216         if (IS_CASEFOLDED(dir))
3217                 d_invalidate(dentry);
3218 #endif
3219
3220 end_rmdir:
3221         brelse(bh);
3222         if (handle)
3223                 ext4_journal_stop(handle);
3224         return retval;
3225 }
3226
3227 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3228                   struct inode *inode,
3229                   struct dentry *dentry /* NULL during fast_commit recovery */)
3230 {
3231         int retval = -ENOENT;
3232         struct buffer_head *bh;
3233         struct ext4_dir_entry_2 *de;
3234         handle_t *handle;
3235         int skip_remove_dentry = 0;
3236
3237         /*
3238          * Keep this outside the transaction; it may have to set up the
3239          * directory's encryption key, which isn't GFP_NOFS-safe.
3240          */
3241         bh = ext4_find_entry(dir, d_name, &de, NULL);
3242         if (IS_ERR(bh))
3243                 return PTR_ERR(bh);
3244
3245         if (!bh)
3246                 return -ENOENT;
3247
3248         if (le32_to_cpu(de->inode) != inode->i_ino) {
3249                 /*
3250                  * It's okay if we find dont find dentry which matches
3251                  * the inode. That's because it might have gotten
3252                  * renamed to a different inode number
3253                  */
3254                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3255                         skip_remove_dentry = 1;
3256                 else
3257                         goto out_bh;
3258         }
3259
3260         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3261                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3262         if (IS_ERR(handle)) {
3263                 retval = PTR_ERR(handle);
3264                 goto out_bh;
3265         }
3266
3267         if (IS_DIRSYNC(dir))
3268                 ext4_handle_sync(handle);
3269
3270         if (!skip_remove_dentry) {
3271                 retval = ext4_delete_entry(handle, dir, de, bh);
3272                 if (retval)
3273                         goto out_handle;
3274                 dir->i_ctime = dir->i_mtime = current_time(dir);
3275                 ext4_update_dx_flag(dir);
3276                 retval = ext4_mark_inode_dirty(handle, dir);
3277                 if (retval)
3278                         goto out_handle;
3279         } else {
3280                 retval = 0;
3281         }
3282         if (inode->i_nlink == 0)
3283                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3284                                    d_name->len, d_name->name);
3285         else
3286                 drop_nlink(inode);
3287         if (!inode->i_nlink)
3288                 ext4_orphan_add(handle, inode);
3289         inode->i_ctime = current_time(inode);
3290         retval = ext4_mark_inode_dirty(handle, inode);
3291         if (dentry && !retval)
3292                 ext4_fc_track_unlink(handle, dentry);
3293 out_handle:
3294         ext4_journal_stop(handle);
3295 out_bh:
3296         brelse(bh);
3297         return retval;
3298 }
3299
3300 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3301 {
3302         int retval;
3303
3304         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3305                 return -EIO;
3306
3307         trace_ext4_unlink_enter(dir, dentry);
3308         /*
3309          * Initialize quotas before so that eventual writes go
3310          * in separate transaction
3311          */
3312         retval = dquot_initialize(dir);
3313         if (retval)
3314                 goto out_trace;
3315         retval = dquot_initialize(d_inode(dentry));
3316         if (retval)
3317                 goto out_trace;
3318
3319         retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3320 #if IS_ENABLED(CONFIG_UNICODE)
3321         /* VFS negative dentries are incompatible with Encoding and
3322          * Case-insensitiveness. Eventually we'll want avoid
3323          * invalidating the dentries here, alongside with returning the
3324          * negative dentries at ext4_lookup(), when it is  better
3325          * supported by the VFS for the CI case.
3326          */
3327         if (IS_CASEFOLDED(dir))
3328                 d_invalidate(dentry);
3329 #endif
3330
3331 out_trace:
3332         trace_ext4_unlink_exit(dentry, retval);
3333         return retval;
3334 }
3335
3336 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3337                                    struct fscrypt_str *disk_link)
3338 {
3339         struct buffer_head *bh;
3340         char *kaddr;
3341         int err = 0;
3342
3343         bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3344         if (IS_ERR(bh))
3345                 return PTR_ERR(bh);
3346
3347         BUFFER_TRACE(bh, "get_write_access");
3348         err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3349         if (err)
3350                 goto out;
3351
3352         kaddr = (char *)bh->b_data;
3353         memcpy(kaddr, disk_link->name, disk_link->len);
3354         inode->i_size = disk_link->len - 1;
3355         EXT4_I(inode)->i_disksize = inode->i_size;
3356         err = ext4_handle_dirty_metadata(handle, inode, bh);
3357 out:
3358         brelse(bh);
3359         return err;
3360 }
3361
3362 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3363                         struct dentry *dentry, const char *symname)
3364 {
3365         handle_t *handle;
3366         struct inode *inode;
3367         int err, len = strlen(symname);
3368         int credits;
3369         struct fscrypt_str disk_link;
3370         int retries = 0;
3371
3372         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3373                 return -EIO;
3374
3375         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3376                                       &disk_link);
3377         if (err)
3378                 return err;
3379
3380         err = dquot_initialize(dir);
3381         if (err)
3382                 return err;
3383
3384         /*
3385          * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3386          * directory. +3 for inode, inode bitmap, group descriptor allocation.
3387          * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3388          * modification.
3389          */
3390         credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3391                   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3392 retry:
3393         inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3394                                             &dentry->d_name, 0, NULL,
3395                                             EXT4_HT_DIR, credits);
3396         handle = ext4_journal_current_handle();
3397         if (IS_ERR(inode)) {
3398                 if (handle)
3399                         ext4_journal_stop(handle);
3400                 err = PTR_ERR(inode);
3401                 goto out_retry;
3402         }
3403
3404         if (IS_ENCRYPTED(inode)) {
3405                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3406                 if (err)
3407                         goto err_drop_inode;
3408                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3409         } else {
3410                 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3411                         inode->i_op = &ext4_symlink_inode_operations;
3412                 } else {
3413                         inode->i_op = &ext4_fast_symlink_inode_operations;
3414                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3415                 }
3416         }
3417
3418         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3419                 /* alloc symlink block and fill it */
3420                 err = ext4_init_symlink_block(handle, inode, &disk_link);
3421                 if (err)
3422                         goto err_drop_inode;
3423         } else {
3424                 /* clear the extent format for fast symlink */
3425                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3426                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3427                        disk_link.len);
3428                 inode->i_size = disk_link.len - 1;
3429                 EXT4_I(inode)->i_disksize = inode->i_size;
3430         }
3431         err = ext4_add_nondir(handle, dentry, &inode);
3432         if (handle)
3433                 ext4_journal_stop(handle);
3434         iput(inode);
3435         goto out_retry;
3436
3437 err_drop_inode:
3438         clear_nlink(inode);
3439         ext4_orphan_add(handle, inode);
3440         unlock_new_inode(inode);
3441         if (handle)
3442                 ext4_journal_stop(handle);
3443         iput(inode);
3444 out_retry:
3445         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3446                 goto retry;
3447         if (disk_link.name != (unsigned char *)symname)
3448                 kfree(disk_link.name);
3449         return err;
3450 }
3451
3452 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3453 {
3454         handle_t *handle;
3455         int err, retries = 0;
3456 retry:
3457         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3458                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3459                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3460         if (IS_ERR(handle))
3461                 return PTR_ERR(handle);
3462
3463         if (IS_DIRSYNC(dir))
3464                 ext4_handle_sync(handle);
3465
3466         inode->i_ctime = current_time(inode);
3467         ext4_inc_count(inode);
3468         ihold(inode);
3469
3470         err = ext4_add_entry(handle, dentry, inode);
3471         if (!err) {
3472                 err = ext4_mark_inode_dirty(handle, inode);
3473                 /* this can happen only for tmpfile being
3474                  * linked the first time
3475                  */
3476                 if (inode->i_nlink == 1)
3477                         ext4_orphan_del(handle, inode);
3478                 d_instantiate(dentry, inode);
3479                 ext4_fc_track_link(handle, dentry);
3480         } else {
3481                 drop_nlink(inode);
3482                 iput(inode);
3483         }
3484         ext4_journal_stop(handle);
3485         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3486                 goto retry;
3487         return err;
3488 }
3489
3490 static int ext4_link(struct dentry *old_dentry,
3491                      struct inode *dir, struct dentry *dentry)
3492 {
3493         struct inode *inode = d_inode(old_dentry);
3494         int err;
3495
3496         if (inode->i_nlink >= EXT4_LINK_MAX)
3497                 return -EMLINK;
3498
3499         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3500         if (err)
3501                 return err;
3502
3503         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3504             (!projid_eq(EXT4_I(dir)->i_projid,
3505                         EXT4_I(old_dentry->d_inode)->i_projid)))
3506                 return -EXDEV;
3507
3508         err = dquot_initialize(dir);
3509         if (err)
3510                 return err;
3511         return __ext4_link(dir, inode, dentry);
3512 }
3513
3514 /*
3515  * Try to find buffer head where contains the parent block.
3516  * It should be the inode block if it is inlined or the 1st block
3517  * if it is a normal dir.
3518  */
3519 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3520                                         struct inode *inode,
3521                                         int *retval,
3522                                         struct ext4_dir_entry_2 **parent_de,
3523                                         int *inlined)
3524 {
3525         struct buffer_head *bh;
3526
3527         if (!ext4_has_inline_data(inode)) {
3528                 struct ext4_dir_entry_2 *de;
3529                 unsigned int offset;
3530
3531                 /* The first directory block must not be a hole, so
3532                  * treat it as DIRENT_HTREE
3533                  */
3534                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3535                 if (IS_ERR(bh)) {
3536                         *retval = PTR_ERR(bh);
3537                         return NULL;
3538                 }
3539
3540                 de = (struct ext4_dir_entry_2 *) bh->b_data;
3541                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3542                                          bh->b_size, 0) ||
3543                     le32_to_cpu(de->inode) != inode->i_ino ||
3544                     strcmp(".", de->name)) {
3545                         EXT4_ERROR_INODE(inode, "directory missing '.'");
3546                         brelse(bh);
3547                         *retval = -EFSCORRUPTED;
3548                         return NULL;
3549                 }
3550                 offset = ext4_rec_len_from_disk(de->rec_len,
3551                                                 inode->i_sb->s_blocksize);
3552                 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3553                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3554                                          bh->b_size, offset) ||
3555                     le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3556                         EXT4_ERROR_INODE(inode, "directory missing '..'");
3557                         brelse(bh);
3558                         *retval = -EFSCORRUPTED;
3559                         return NULL;
3560                 }
3561                 *parent_de = de;
3562
3563                 return bh;
3564         }
3565
3566         *inlined = 1;
3567         return ext4_get_first_inline_block(inode, parent_de, retval);
3568 }
3569
3570 struct ext4_renament {
3571         struct inode *dir;
3572         struct dentry *dentry;
3573         struct inode *inode;
3574         bool is_dir;
3575         int dir_nlink_delta;
3576
3577         /* entry for "dentry" */
3578         struct buffer_head *bh;
3579         struct ext4_dir_entry_2 *de;
3580         int inlined;
3581
3582         /* entry for ".." in inode if it's a directory */
3583         struct buffer_head *dir_bh;
3584         struct ext4_dir_entry_2 *parent_de;
3585         int dir_inlined;
3586 };
3587
3588 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3589 {
3590         int retval;
3591
3592         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3593                                               &retval, &ent->parent_de,
3594                                               &ent->dir_inlined);
3595         if (!ent->dir_bh)
3596                 return retval;
3597         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3598                 return -EFSCORRUPTED;
3599         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3600         return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3601                                              ent->dir_bh, EXT4_JTR_NONE);
3602 }
3603
3604 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3605                                   unsigned dir_ino)
3606 {
3607         int retval;
3608
3609         ent->parent_de->inode = cpu_to_le32(dir_ino);
3610         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3611         if (!ent->dir_inlined) {
3612                 if (is_dx(ent->inode)) {
3613                         retval = ext4_handle_dirty_dx_node(handle,
3614                                                            ent->inode,
3615                                                            ent->dir_bh);
3616                 } else {
3617                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3618                                                             ent->dir_bh);
3619                 }
3620         } else {
3621                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3622         }
3623         if (retval) {
3624                 ext4_std_error(ent->dir->i_sb, retval);
3625                 return retval;
3626         }
3627         return 0;
3628 }
3629
3630 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3631                        unsigned ino, unsigned file_type)
3632 {
3633         int retval, retval2;
3634
3635         BUFFER_TRACE(ent->bh, "get write access");
3636         retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3637                                                EXT4_JTR_NONE);
3638         if (retval)
3639                 return retval;
3640         ent->de->inode = cpu_to_le32(ino);
3641         if (ext4_has_feature_filetype(ent->dir->i_sb))
3642                 ent->de->file_type = file_type;
3643         inode_inc_iversion(ent->dir);
3644         ent->dir->i_ctime = ent->dir->i_mtime =
3645                 current_time(ent->dir);
3646         retval = ext4_mark_inode_dirty(handle, ent->dir);
3647         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3648         if (!ent->inlined) {
3649                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3650                 if (unlikely(retval2)) {
3651                         ext4_std_error(ent->dir->i_sb, retval2);
3652                         return retval2;
3653                 }
3654         }
3655         return retval;
3656 }
3657
3658 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3659                           unsigned ino, unsigned file_type)
3660 {
3661         struct ext4_renament old = *ent;
3662         int retval = 0;
3663
3664         /*
3665          * old->de could have moved from under us during make indexed dir,
3666          * so the old->de may no longer valid and need to find it again
3667          * before reset old inode info.
3668          */
3669         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3670                                  &old.inlined);
3671         if (IS_ERR(old.bh))
3672                 retval = PTR_ERR(old.bh);
3673         if (!old.bh)
3674                 retval = -ENOENT;
3675         if (retval) {
3676                 ext4_std_error(old.dir->i_sb, retval);
3677                 return;
3678         }
3679
3680         ext4_setent(handle, &old, ino, file_type);
3681         brelse(old.bh);
3682 }
3683
3684 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3685                                   const struct qstr *d_name)
3686 {
3687         int retval = -ENOENT;
3688         struct buffer_head *bh;
3689         struct ext4_dir_entry_2 *de;
3690
3691         bh = ext4_find_entry(dir, d_name, &de, NULL);
3692         if (IS_ERR(bh))
3693                 return PTR_ERR(bh);
3694         if (bh) {
3695                 retval = ext4_delete_entry(handle, dir, de, bh);
3696                 brelse(bh);
3697         }
3698         return retval;
3699 }
3700
3701 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3702                                int force_reread)
3703 {
3704         int retval;
3705         /*
3706          * ent->de could have moved from under us during htree split, so make
3707          * sure that we are deleting the right entry.  We might also be pointing
3708          * to a stale entry in the unused part of ent->bh so just checking inum
3709          * and the name isn't enough.
3710          */
3711         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3712             ent->de->name_len != ent->dentry->d_name.len ||
3713             strncmp(ent->de->name, ent->dentry->d_name.name,
3714                     ent->de->name_len) ||
3715             force_reread) {
3716                 retval = ext4_find_delete_entry(handle, ent->dir,
3717                                                 &ent->dentry->d_name);
3718         } else {
3719                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3720                 if (retval == -ENOENT) {
3721                         retval = ext4_find_delete_entry(handle, ent->dir,
3722                                                         &ent->dentry->d_name);
3723                 }
3724         }
3725
3726         if (retval) {
3727                 ext4_warning_inode(ent->dir,
3728                                    "Deleting old file: nlink %d, error=%d",
3729                                    ent->dir->i_nlink, retval);
3730         }
3731 }
3732
3733 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3734 {
3735         if (ent->dir_nlink_delta) {
3736                 if (ent->dir_nlink_delta == -1)
3737                         ext4_dec_count(ent->dir);
3738                 else
3739                         ext4_inc_count(ent->dir);
3740                 ext4_mark_inode_dirty(handle, ent->dir);
3741         }
3742 }
3743
3744 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3745                                               struct ext4_renament *ent,
3746                                               int credits, handle_t **h)
3747 {
3748         struct inode *wh;
3749         handle_t *handle;
3750         int retries = 0;
3751
3752         /*
3753          * for inode block, sb block, group summaries,
3754          * and inode bitmap
3755          */
3756         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3757                     EXT4_XATTR_TRANS_BLOCKS + 4);
3758 retry:
3759         wh = ext4_new_inode_start_handle(idmap, ent->dir,
3760                                          S_IFCHR | WHITEOUT_MODE,
3761                                          &ent->dentry->d_name, 0, NULL,
3762                                          EXT4_HT_DIR, credits);
3763
3764         handle = ext4_journal_current_handle();
3765         if (IS_ERR(wh)) {
3766                 if (handle)
3767                         ext4_journal_stop(handle);
3768                 if (PTR_ERR(wh) == -ENOSPC &&
3769                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3770                         goto retry;
3771         } else {
3772                 *h = handle;
3773                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3774                 wh->i_op = &ext4_special_inode_operations;
3775         }
3776         return wh;
3777 }
3778
3779 /*
3780  * Anybody can rename anything with this: the permission checks are left to the
3781  * higher-level routines.
3782  *
3783  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3784  * while new_{dentry,inode) refers to the destination dentry/inode
3785  * This comes from rename(const char *oldpath, const char *newpath)
3786  */
3787 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3788                        struct dentry *old_dentry, struct inode *new_dir,
3789                        struct dentry *new_dentry, unsigned int flags)
3790 {
3791         handle_t *handle = NULL;
3792         struct ext4_renament old = {
3793                 .dir = old_dir,
3794                 .dentry = old_dentry,
3795                 .inode = d_inode(old_dentry),
3796         };
3797         struct ext4_renament new = {
3798                 .dir = new_dir,
3799                 .dentry = new_dentry,
3800                 .inode = d_inode(new_dentry),
3801         };
3802         int force_reread;
3803         int retval;
3804         struct inode *whiteout = NULL;
3805         int credits;
3806         u8 old_file_type;
3807
3808         if (new.inode && new.inode->i_nlink == 0) {
3809                 EXT4_ERROR_INODE(new.inode,
3810                                  "target of rename is already freed");
3811                 return -EFSCORRUPTED;
3812         }
3813
3814         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3815             (!projid_eq(EXT4_I(new_dir)->i_projid,
3816                         EXT4_I(old_dentry->d_inode)->i_projid)))
3817                 return -EXDEV;
3818
3819         retval = dquot_initialize(old.dir);
3820         if (retval)
3821                 return retval;
3822         retval = dquot_initialize(old.inode);
3823         if (retval)
3824                 return retval;
3825         retval = dquot_initialize(new.dir);
3826         if (retval)
3827                 return retval;
3828
3829         /* Initialize quotas before so that eventual writes go
3830          * in separate transaction */
3831         if (new.inode) {
3832                 retval = dquot_initialize(new.inode);
3833                 if (retval)
3834                         return retval;
3835         }
3836
3837         /*
3838          * We need to protect against old.inode directory getting converted
3839          * from inline directory format into a normal one.
3840          */
3841         if (S_ISDIR(old.inode->i_mode))
3842                 inode_lock_nested(old.inode, I_MUTEX_NONDIR2);
3843
3844         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3845                                  &old.inlined);
3846         if (IS_ERR(old.bh)) {
3847                 retval = PTR_ERR(old.bh);
3848                 goto unlock_moved_dir;
3849         }
3850
3851         /*
3852          *  Check for inode number is _not_ due to possible IO errors.
3853          *  We might rmdir the source, keep it as pwd of some process
3854          *  and merrily kill the link to whatever was created under the
3855          *  same name. Goodbye sticky bit ;-<
3856          */
3857         retval = -ENOENT;
3858         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3859                 goto release_bh;
3860
3861         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3862                                  &new.de, &new.inlined);
3863         if (IS_ERR(new.bh)) {
3864                 retval = PTR_ERR(new.bh);
3865                 new.bh = NULL;
3866                 goto release_bh;
3867         }
3868         if (new.bh) {
3869                 if (!new.inode) {
3870                         brelse(new.bh);
3871                         new.bh = NULL;
3872                 }
3873         }
3874         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3875                 ext4_alloc_da_blocks(old.inode);
3876
3877         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3878                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3879         if (!(flags & RENAME_WHITEOUT)) {
3880                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3881                 if (IS_ERR(handle)) {
3882                         retval = PTR_ERR(handle);
3883                         goto release_bh;
3884                 }
3885         } else {
3886                 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3887                 if (IS_ERR(whiteout)) {
3888                         retval = PTR_ERR(whiteout);
3889                         goto release_bh;
3890                 }
3891         }
3892
3893         old_file_type = old.de->file_type;
3894         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3895                 ext4_handle_sync(handle);
3896
3897         if (S_ISDIR(old.inode->i_mode)) {
3898                 if (new.inode) {
3899                         retval = -ENOTEMPTY;
3900                         if (!ext4_empty_dir(new.inode))
3901                                 goto end_rename;
3902                 } else {
3903                         retval = -EMLINK;
3904                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3905                                 goto end_rename;
3906                 }
3907                 retval = ext4_rename_dir_prepare(handle, &old);
3908                 if (retval)
3909                         goto end_rename;
3910         }
3911         /*
3912          * If we're renaming a file within an inline_data dir and adding or
3913          * setting the new dirent causes a conversion from inline_data to
3914          * extents/blockmap, we need to force the dirent delete code to
3915          * re-read the directory, or else we end up trying to delete a dirent
3916          * from what is now the extent tree root (or a block map).
3917          */
3918         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3919                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3920
3921         if (whiteout) {
3922                 /*
3923                  * Do this before adding a new entry, so the old entry is sure
3924                  * to be still pointing to the valid old entry.
3925                  */
3926                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3927                                      EXT4_FT_CHRDEV);
3928                 if (retval)
3929                         goto end_rename;
3930                 retval = ext4_mark_inode_dirty(handle, whiteout);
3931                 if (unlikely(retval))
3932                         goto end_rename;
3933
3934         }
3935         if (!new.bh) {
3936                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3937                 if (retval)
3938                         goto end_rename;
3939         } else {
3940                 retval = ext4_setent(handle, &new,
3941                                      old.inode->i_ino, old_file_type);
3942                 if (retval)
3943                         goto end_rename;
3944         }
3945         if (force_reread)
3946                 force_reread = !ext4_test_inode_flag(new.dir,
3947                                                      EXT4_INODE_INLINE_DATA);
3948
3949         /*
3950          * Like most other Unix systems, set the ctime for inodes on a
3951          * rename.
3952          */
3953         old.inode->i_ctime = current_time(old.inode);
3954         retval = ext4_mark_inode_dirty(handle, old.inode);
3955         if (unlikely(retval))
3956                 goto end_rename;
3957
3958         if (!whiteout) {
3959                 /*
3960                  * ok, that's it
3961                  */
3962                 ext4_rename_delete(handle, &old, force_reread);
3963         }
3964
3965         if (new.inode) {
3966                 ext4_dec_count(new.inode);
3967                 new.inode->i_ctime = current_time(new.inode);
3968         }
3969         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3970         ext4_update_dx_flag(old.dir);
3971         if (old.dir_bh) {
3972                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3973                 if (retval)
3974                         goto end_rename;
3975
3976                 ext4_dec_count(old.dir);
3977                 if (new.inode) {
3978                         /* checked ext4_empty_dir above, can't have another
3979                          * parent, ext4_dec_count() won't work for many-linked
3980                          * dirs */
3981                         clear_nlink(new.inode);
3982                 } else {
3983                         ext4_inc_count(new.dir);
3984                         ext4_update_dx_flag(new.dir);
3985                         retval = ext4_mark_inode_dirty(handle, new.dir);
3986                         if (unlikely(retval))
3987                                 goto end_rename;
3988                 }
3989         }
3990         retval = ext4_mark_inode_dirty(handle, old.dir);
3991         if (unlikely(retval))
3992                 goto end_rename;
3993
3994         if (S_ISDIR(old.inode->i_mode)) {
3995                 /*
3996                  * We disable fast commits here that's because the
3997                  * replay code is not yet capable of changing dot dot
3998                  * dirents in directories.
3999                  */
4000                 ext4_fc_mark_ineligible(old.inode->i_sb,
4001                         EXT4_FC_REASON_RENAME_DIR, handle);
4002         } else {
4003                 struct super_block *sb = old.inode->i_sb;
4004
4005                 if (new.inode)
4006                         ext4_fc_track_unlink(handle, new.dentry);
4007                 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4008                     !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4009                     !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4010                         __ext4_fc_track_link(handle, old.inode, new.dentry);
4011                         __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4012                         if (whiteout)
4013                                 __ext4_fc_track_create(handle, whiteout,
4014                                                        old.dentry);
4015                 }
4016         }
4017
4018         if (new.inode) {
4019                 retval = ext4_mark_inode_dirty(handle, new.inode);
4020                 if (unlikely(retval))
4021                         goto end_rename;
4022                 if (!new.inode->i_nlink)
4023                         ext4_orphan_add(handle, new.inode);
4024         }
4025         retval = 0;
4026
4027 end_rename:
4028         if (whiteout) {
4029                 if (retval) {
4030                         ext4_resetent(handle, &old,
4031                                       old.inode->i_ino, old_file_type);
4032                         drop_nlink(whiteout);
4033                         ext4_orphan_add(handle, whiteout);
4034                 }
4035                 unlock_new_inode(whiteout);
4036                 ext4_journal_stop(handle);
4037                 iput(whiteout);
4038         } else {
4039                 ext4_journal_stop(handle);
4040         }
4041 release_bh:
4042         brelse(old.dir_bh);
4043         brelse(old.bh);
4044         brelse(new.bh);
4045
4046 unlock_moved_dir:
4047         if (S_ISDIR(old.inode->i_mode))
4048                 inode_unlock(old.inode);
4049
4050         return retval;
4051 }
4052
4053 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4054                              struct inode *new_dir, struct dentry *new_dentry)
4055 {
4056         handle_t *handle = NULL;
4057         struct ext4_renament old = {
4058                 .dir = old_dir,
4059                 .dentry = old_dentry,
4060                 .inode = d_inode(old_dentry),
4061         };
4062         struct ext4_renament new = {
4063                 .dir = new_dir,
4064                 .dentry = new_dentry,
4065                 .inode = d_inode(new_dentry),
4066         };
4067         u8 new_file_type;
4068         int retval;
4069         struct timespec64 ctime;
4070
4071         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4072              !projid_eq(EXT4_I(new_dir)->i_projid,
4073                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
4074             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4075              !projid_eq(EXT4_I(old_dir)->i_projid,
4076                         EXT4_I(new_dentry->d_inode)->i_projid)))
4077                 return -EXDEV;
4078
4079         retval = dquot_initialize(old.dir);
4080         if (retval)
4081                 return retval;
4082         retval = dquot_initialize(new.dir);
4083         if (retval)
4084                 return retval;
4085
4086         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4087                                  &old.de, &old.inlined);
4088         if (IS_ERR(old.bh))
4089                 return PTR_ERR(old.bh);
4090         /*
4091          *  Check for inode number is _not_ due to possible IO errors.
4092          *  We might rmdir the source, keep it as pwd of some process
4093          *  and merrily kill the link to whatever was created under the
4094          *  same name. Goodbye sticky bit ;-<
4095          */
4096         retval = -ENOENT;
4097         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4098                 goto end_rename;
4099
4100         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4101                                  &new.de, &new.inlined);
4102         if (IS_ERR(new.bh)) {
4103                 retval = PTR_ERR(new.bh);
4104                 new.bh = NULL;
4105                 goto end_rename;
4106         }
4107
4108         /* RENAME_EXCHANGE case: old *and* new must both exist */
4109         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4110                 goto end_rename;
4111
4112         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4113                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4114                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4115         if (IS_ERR(handle)) {
4116                 retval = PTR_ERR(handle);
4117                 handle = NULL;
4118                 goto end_rename;
4119         }
4120
4121         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4122                 ext4_handle_sync(handle);
4123
4124         if (S_ISDIR(old.inode->i_mode)) {
4125                 old.is_dir = true;
4126                 retval = ext4_rename_dir_prepare(handle, &old);
4127                 if (retval)
4128                         goto end_rename;
4129         }
4130         if (S_ISDIR(new.inode->i_mode)) {
4131                 new.is_dir = true;
4132                 retval = ext4_rename_dir_prepare(handle, &new);
4133                 if (retval)
4134                         goto end_rename;
4135         }
4136
4137         /*
4138          * Other than the special case of overwriting a directory, parents'
4139          * nlink only needs to be modified if this is a cross directory rename.
4140          */
4141         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4142                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4143                 new.dir_nlink_delta = -old.dir_nlink_delta;
4144                 retval = -EMLINK;
4145                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4146                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4147                         goto end_rename;
4148         }
4149
4150         new_file_type = new.de->file_type;
4151         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4152         if (retval)
4153                 goto end_rename;
4154
4155         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4156         if (retval)
4157                 goto end_rename;
4158
4159         /*
4160          * Like most other Unix systems, set the ctime for inodes on a
4161          * rename.
4162          */
4163         ctime = current_time(old.inode);
4164         old.inode->i_ctime = ctime;
4165         new.inode->i_ctime = ctime;
4166         retval = ext4_mark_inode_dirty(handle, old.inode);
4167         if (unlikely(retval))
4168                 goto end_rename;
4169         retval = ext4_mark_inode_dirty(handle, new.inode);
4170         if (unlikely(retval))
4171                 goto end_rename;
4172         ext4_fc_mark_ineligible(new.inode->i_sb,
4173                                 EXT4_FC_REASON_CROSS_RENAME, handle);
4174         if (old.dir_bh) {
4175                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4176                 if (retval)
4177                         goto end_rename;
4178         }
4179         if (new.dir_bh) {
4180                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4181                 if (retval)
4182                         goto end_rename;
4183         }
4184         ext4_update_dir_count(handle, &old);
4185         ext4_update_dir_count(handle, &new);
4186         retval = 0;
4187
4188 end_rename:
4189         brelse(old.dir_bh);
4190         brelse(new.dir_bh);
4191         brelse(old.bh);
4192         brelse(new.bh);
4193         if (handle)
4194                 ext4_journal_stop(handle);
4195         return retval;
4196 }
4197
4198 static int ext4_rename2(struct mnt_idmap *idmap,
4199                         struct inode *old_dir, struct dentry *old_dentry,
4200                         struct inode *new_dir, struct dentry *new_dentry,
4201                         unsigned int flags)
4202 {
4203         int err;
4204
4205         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4206                 return -EIO;
4207
4208         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4209                 return -EINVAL;
4210
4211         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4212                                      flags);
4213         if (err)
4214                 return err;
4215
4216         if (flags & RENAME_EXCHANGE) {
4217                 return ext4_cross_rename(old_dir, old_dentry,
4218                                          new_dir, new_dentry);
4219         }
4220
4221         return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4222 }
4223
4224 /*
4225  * directories can handle most operations...
4226  */
4227 const struct inode_operations ext4_dir_inode_operations = {
4228         .create         = ext4_create,
4229         .lookup         = ext4_lookup,
4230         .link           = ext4_link,
4231         .unlink         = ext4_unlink,
4232         .symlink        = ext4_symlink,
4233         .mkdir          = ext4_mkdir,
4234         .rmdir          = ext4_rmdir,
4235         .mknod          = ext4_mknod,
4236         .tmpfile        = ext4_tmpfile,
4237         .rename         = ext4_rename2,
4238         .setattr        = ext4_setattr,
4239         .getattr        = ext4_getattr,
4240         .listxattr      = ext4_listxattr,
4241         .get_inode_acl  = ext4_get_acl,
4242         .set_acl        = ext4_set_acl,
4243         .fiemap         = ext4_fiemap,
4244         .fileattr_get   = ext4_fileattr_get,
4245         .fileattr_set   = ext4_fileattr_set,
4246 };
4247
4248 const struct inode_operations ext4_special_inode_operations = {
4249         .setattr        = ext4_setattr,
4250         .getattr        = ext4_getattr,
4251         .listxattr      = ext4_listxattr,
4252         .get_inode_acl  = ext4_get_acl,
4253         .set_acl        = ext4_set_acl,
4254 };