~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ext4/namei.c

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/fs/ext4/namei.c
  3  *
  4  * Copyright (C) 1992, 1993, 1994, 1995
  5  * Remy Card (card@masi.ibp.fr)
  6  * Laboratoire MASI - Institut Blaise Pascal
  7  * Universite Pierre et Marie Curie (Paris VI)
  8  *
  9  *  from
 10  *
 11  *  linux/fs/minix/namei.c
 12  *
 13  *  Copyright (C) 1991, 1992  Linus Torvalds
 14  *
 15  *  Big-endian to little-endian byte-swapping/bitmaps by
 16  *        David S. Miller (davem@caip.rutgers.edu), 1995
 17  *  Directory entry file type support and forward compatibility hooks
 18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
 19  *  Hash Tree Directory indexing (c)
 20  *      Daniel Phillips, 2001
 21  *  Hash Tree Directory indexing porting
 22  *      Christopher Li, 2002
 23  *  Hash Tree Directory indexing cleanup
 24  *      Theodore Ts'o, 2002
 25  */
 26 
 27 #include <linux/fs.h>
 28 #include <linux/pagemap.h>
 29 #include <linux/jbd2.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 "ext4.h"
 38 #include "ext4_jbd2.h"
 39 
 40 #include "xattr.h"
 41 #include "acl.h"
 42 
 43 #include <trace/events/ext4.h>
 44 /*
 45  * define how far ahead to read directories while searching them.
 46  */
 47 #define NAMEI_RA_CHUNKS  2
 48 #define NAMEI_RA_BLOCKS  4
 49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
 50 
 51 static struct buffer_head *ext4_append(handle_t *handle,
 52                                         struct inode *inode,
 53                                         ext4_lblk_t *block)
 54 {
 55         struct buffer_head *bh;
 56         int err;
 57 
 58         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
 59                      ((inode->i_size >> 10) >=
 60                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
 61                 return ERR_PTR(-ENOSPC);
 62 
 63         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
 64 
 65         bh = ext4_bread(handle, inode, *block, 1);
 66         if (IS_ERR(bh))
 67                 return bh;
 68         inode->i_size += inode->i_sb->s_blocksize;
 69         EXT4_I(inode)->i_disksize = inode->i_size;
 70         BUFFER_TRACE(bh, "get_write_access");
 71         err = ext4_journal_get_write_access(handle, bh);
 72         if (err) {
 73                 brelse(bh);
 74                 ext4_std_error(inode->i_sb, err);
 75                 return ERR_PTR(err);
 76         }
 77         return bh;
 78 }
 79 
 80 static int ext4_dx_csum_verify(struct inode *inode,
 81                                struct ext4_dir_entry *dirent);
 82 
 83 typedef enum {
 84         EITHER, INDEX, DIRENT
 85 } dirblock_type_t;
 86 
 87 #define ext4_read_dirblock(inode, block, type) \
 88         __ext4_read_dirblock((inode), (block), (type), __LINE__)
 89 
 90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 91                                               ext4_lblk_t block,
 92                                               dirblock_type_t type,
 93                                               unsigned int line)
 94 {
 95         struct buffer_head *bh;
 96         struct ext4_dir_entry *dirent;
 97         int is_dx_block = 0;
 98 
 99         bh = ext4_bread(NULL, inode, block, 0);
100         if (IS_ERR(bh)) {
101                 __ext4_warning(inode->i_sb, __func__, line,
102                                "error %ld reading directory block "
103                                "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
104                                (unsigned long) block);
105 
106                 return bh;
107         }
108         if (!bh) {
109                 ext4_error_inode(inode, __func__, line, block, "Directory hole found");
110                 return ERR_PTR(-EIO);
111         }
112         dirent = (struct ext4_dir_entry *) bh->b_data;
113         /* Determine whether or not we have an index block */
114         if (is_dx(inode)) {
115                 if (block == 0)
116                         is_dx_block = 1;
117                 else if (ext4_rec_len_from_disk(dirent->rec_len,
118                                                 inode->i_sb->s_blocksize) ==
119                          inode->i_sb->s_blocksize)
120                         is_dx_block = 1;
121         }
122         if (!is_dx_block && type == INDEX) {
123                 ext4_error_inode(inode, __func__, line, block,
124                        "directory leaf block found instead of index block");
125                 return ERR_PTR(-EIO);
126         }
127         if (!ext4_has_metadata_csum(inode->i_sb) ||
128             buffer_verified(bh))
129                 return bh;
130 
131         /*
132          * An empty leaf block can get mistaken for a index block; for
133          * this reason, we can only check the index checksum when the
134          * caller is sure it should be an index block.
135          */
136         if (is_dx_block && type == INDEX) {
137                 if (ext4_dx_csum_verify(inode, dirent))
138                         set_buffer_verified(bh);
139                 else {
140                         ext4_error_inode(inode, __func__, line, block,
141                                 "Directory index failed checksum");
142                         brelse(bh);
143                         return ERR_PTR(-EIO);
144                 }
145         }
146         if (!is_dx_block) {
147                 if (ext4_dirent_csum_verify(inode, dirent))
148                         set_buffer_verified(bh);
149                 else {
150                         ext4_error_inode(inode, __func__, line, block,
151                                 "Directory block failed checksum");
152                         brelse(bh);
153                         return ERR_PTR(-EIO);
154                 }
155         }
156         return bh;
157 }
158 
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162 
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168 
169 struct fake_dirent
170 {
171         __le32 inode;
172         __le16 rec_len;
173         u8 name_len;
174         u8 file_type;
175 };
176 
177 struct dx_countlimit
178 {
179         __le16 limit;
180         __le16 count;
181 };
182 
183 struct dx_entry
184 {
185         __le32 hash;
186         __le32 block;
187 };
188 
189 /*
190  * dx_root_info is laid out so that if it should somehow get overlaid by a
191  * dirent the two low bits of the hash version will be zero.  Therefore, the
192  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
193  */
194 
195 struct dx_root
196 {
197         struct fake_dirent dot;
198         char dot_name[4];
199         struct fake_dirent dotdot;
200         char dotdot_name[4];
201         struct dx_root_info
202         {
203                 __le32 reserved_zero;
204                 u8 hash_version;
205                 u8 info_length; /* 8 */
206                 u8 indirect_levels;
207                 u8 unused_flags;
208         }
209         info;
210         struct dx_entry entries[0];
211 };
212 
213 struct dx_node
214 {
215         struct fake_dirent fake;
216         struct dx_entry entries[0];
217 };
218 
219 
220 struct dx_frame
221 {
222         struct buffer_head *bh;
223         struct dx_entry *entries;
224         struct dx_entry *at;
225 };
226 
227 struct dx_map_entry
228 {
229         u32 hash;
230         u16 offs;
231         u16 size;
232 };
233 
234 /*
235  * This goes at the end of each htree block.
236  */
237 struct dx_tail {
238         u32 dt_reserved;
239         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
240 };
241 
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253                                  struct inode *dir,
254                                  struct dx_hash_info *hinfo,
255                                  struct dx_frame *frame);
256 static void dx_release(struct dx_frame *frames);
257 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
258                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
259 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
260 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
261                 struct dx_map_entry *offsets, int count, unsigned blocksize);
262 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
263 static void dx_insert_block(struct dx_frame *frame,
264                                         u32 hash, ext4_lblk_t block);
265 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
266                                  struct dx_frame *frame,
267                                  struct dx_frame *frames,
268                                  __u32 *start_hash);
269 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
270                 const struct qstr *d_name,
271                 struct ext4_dir_entry_2 **res_dir);
272 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
273                              struct inode *inode);
274 
275 /* checksumming functions */
276 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
277                             unsigned int blocksize)
278 {
279         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
280         t->det_rec_len = ext4_rec_len_to_disk(
281                         sizeof(struct ext4_dir_entry_tail), blocksize);
282         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
283 }
284 
285 /* Walk through a dirent block to find a checksum "dirent" at the tail */
286 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
287                                                    struct ext4_dir_entry *de)
288 {
289         struct ext4_dir_entry_tail *t;
290 
291 #ifdef PARANOID
292         struct ext4_dir_entry *d, *top;
293 
294         d = de;
295         top = (struct ext4_dir_entry *)(((void *)de) +
296                 (EXT4_BLOCK_SIZE(inode->i_sb) -
297                 sizeof(struct ext4_dir_entry_tail)));
298         while (d < top && d->rec_len)
299                 d = (struct ext4_dir_entry *)(((void *)d) +
300                     le16_to_cpu(d->rec_len));
301 
302         if (d != top)
303                 return NULL;
304 
305         t = (struct ext4_dir_entry_tail *)d;
306 #else
307         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
308 #endif
309 
310         if (t->det_reserved_zero1 ||
311             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
312             t->det_reserved_zero2 ||
313             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
314                 return NULL;
315 
316         return t;
317 }
318 
319 static __le32 ext4_dirent_csum(struct inode *inode,
320                                struct ext4_dir_entry *dirent, int size)
321 {
322         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
323         struct ext4_inode_info *ei = EXT4_I(inode);
324         __u32 csum;
325 
326         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
327         return cpu_to_le32(csum);
328 }
329 
330 static void warn_no_space_for_csum(struct inode *inode)
331 {
332         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
333                      "checksum.  Please run e2fsck -D.", inode->i_ino);
334 }
335 
336 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
337 {
338         struct ext4_dir_entry_tail *t;
339 
340         if (!ext4_has_metadata_csum(inode->i_sb))
341                 return 1;
342 
343         t = get_dirent_tail(inode, dirent);
344         if (!t) {
345                 warn_no_space_for_csum(inode);
346                 return 0;
347         }
348 
349         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
350                                                 (void *)t - (void *)dirent))
351                 return 0;
352 
353         return 1;
354 }
355 
356 static void ext4_dirent_csum_set(struct inode *inode,
357                                  struct ext4_dir_entry *dirent)
358 {
359         struct ext4_dir_entry_tail *t;
360 
361         if (!ext4_has_metadata_csum(inode->i_sb))
362                 return;
363 
364         t = get_dirent_tail(inode, dirent);
365         if (!t) {
366                 warn_no_space_for_csum(inode);
367                 return;
368         }
369 
370         t->det_checksum = ext4_dirent_csum(inode, dirent,
371                                            (void *)t - (void *)dirent);
372 }
373 
374 int ext4_handle_dirty_dirent_node(handle_t *handle,
375                                   struct inode *inode,
376                                   struct buffer_head *bh)
377 {
378         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
379         return ext4_handle_dirty_metadata(handle, inode, bh);
380 }
381 
382 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
383                                                struct ext4_dir_entry *dirent,
384                                                int *offset)
385 {
386         struct ext4_dir_entry *dp;
387         struct dx_root_info *root;
388         int count_offset;
389 
390         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
391                 count_offset = 8;
392         else if (le16_to_cpu(dirent->rec_len) == 12) {
393                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
394                 if (le16_to_cpu(dp->rec_len) !=
395                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
396                         return NULL;
397                 root = (struct dx_root_info *)(((void *)dp + 12));
398                 if (root->reserved_zero ||
399                     root->info_length != sizeof(struct dx_root_info))
400                         return NULL;
401                 count_offset = 32;
402         } else
403                 return NULL;
404 
405         if (offset)
406                 *offset = count_offset;
407         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
408 }
409 
410 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
411                            int count_offset, int count, struct dx_tail *t)
412 {
413         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
414         struct ext4_inode_info *ei = EXT4_I(inode);
415         __u32 csum;
416         __le32 save_csum;
417         int size;
418 
419         size = count_offset + (count * sizeof(struct dx_entry));
420         save_csum = t->dt_checksum;
421         t->dt_checksum = 0;
422         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
423         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
424         t->dt_checksum = save_csum;
425 
426         return cpu_to_le32(csum);
427 }
428 
429 static int ext4_dx_csum_verify(struct inode *inode,
430                                struct ext4_dir_entry *dirent)
431 {
432         struct dx_countlimit *c;
433         struct dx_tail *t;
434         int count_offset, limit, count;
435 
436         if (!ext4_has_metadata_csum(inode->i_sb))
437                 return 1;
438 
439         c = get_dx_countlimit(inode, dirent, &count_offset);
440         if (!c) {
441                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
442                 return 1;
443         }
444         limit = le16_to_cpu(c->limit);
445         count = le16_to_cpu(c->count);
446         if (count_offset + (limit * sizeof(struct dx_entry)) >
447             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
448                 warn_no_space_for_csum(inode);
449                 return 1;
450         }
451         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
452 
453         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
454                                             count, t))
455                 return 0;
456         return 1;
457 }
458 
459 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
460 {
461         struct dx_countlimit *c;
462         struct dx_tail *t;
463         int count_offset, limit, count;
464 
465         if (!ext4_has_metadata_csum(inode->i_sb))
466                 return;
467 
468         c = get_dx_countlimit(inode, dirent, &count_offset);
469         if (!c) {
470                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
471                 return;
472         }
473         limit = le16_to_cpu(c->limit);
474         count = le16_to_cpu(c->count);
475         if (count_offset + (limit * sizeof(struct dx_entry)) >
476             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
477                 warn_no_space_for_csum(inode);
478                 return;
479         }
480         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
481 
482         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
483 }
484 
485 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
486                                             struct inode *inode,
487                                             struct buffer_head *bh)
488 {
489         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
490         return ext4_handle_dirty_metadata(handle, inode, bh);
491 }
492 
493 /*
494  * p is at least 6 bytes before the end of page
495  */
496 static inline struct ext4_dir_entry_2 *
497 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
498 {
499         return (struct ext4_dir_entry_2 *)((char *)p +
500                 ext4_rec_len_from_disk(p->rec_len, blocksize));
501 }
502 
503 /*
504  * Future: use high four bits of block for coalesce-on-delete flags
505  * Mask them off for now.
506  */
507 
508 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
509 {
510         return le32_to_cpu(entry->block) & 0x00ffffff;
511 }
512 
513 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
514 {
515         entry->block = cpu_to_le32(value);
516 }
517 
518 static inline unsigned dx_get_hash(struct dx_entry *entry)
519 {
520         return le32_to_cpu(entry->hash);
521 }
522 
523 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
524 {
525         entry->hash = cpu_to_le32(value);
526 }
527 
528 static inline unsigned dx_get_count(struct dx_entry *entries)
529 {
530         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
531 }
532 
533 static inline unsigned dx_get_limit(struct dx_entry *entries)
534 {
535         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
536 }
537 
538 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
539 {
540         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
541 }
542 
543 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
544 {
545         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
546 }
547 
548 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
549 {
550         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
551                 EXT4_DIR_REC_LEN(2) - infosize;
552 
553         if (ext4_has_metadata_csum(dir->i_sb))
554                 entry_space -= sizeof(struct dx_tail);
555         return entry_space / sizeof(struct dx_entry);
556 }
557 
558 static inline unsigned dx_node_limit(struct inode *dir)
559 {
560         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
561 
562         if (ext4_has_metadata_csum(dir->i_sb))
563                 entry_space -= sizeof(struct dx_tail);
564         return entry_space / sizeof(struct dx_entry);
565 }
566 
567 /*
568  * Debug
569  */
570 #ifdef DX_DEBUG
571 static void dx_show_index(char * label, struct dx_entry *entries)
572 {
573         int i, n = dx_get_count (entries);
574         printk(KERN_DEBUG "%s index ", label);
575         for (i = 0; i < n; i++) {
576                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
577                                 0, (unsigned long)dx_get_block(entries + i));
578         }
579         printk("\n");
580 }
581 
582 struct stats
583 {
584         unsigned names;
585         unsigned space;
586         unsigned bcount;
587 };
588 
589 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
590                                  int size, int show_names)
591 {
592         unsigned names = 0, space = 0;
593         char *base = (char *) de;
594         struct dx_hash_info h = *hinfo;
595 
596         printk("names: ");
597         while ((char *) de < base + size)
598         {
599                 if (de->inode)
600                 {
601                         if (show_names)
602                         {
603                                 int len = de->name_len;
604                                 char *name = de->name;
605                                 while (len--) printk("%c", *name++);
606                                 ext4fs_dirhash(de->name, de->name_len, &h);
607                                 printk(":%x.%u ", h.hash,
608                                        (unsigned) ((char *) de - base));
609                         }
610                         space += EXT4_DIR_REC_LEN(de->name_len);
611                         names++;
612                 }
613                 de = ext4_next_entry(de, size);
614         }
615         printk("(%i)\n", names);
616         return (struct stats) { names, space, 1 };
617 }
618 
619 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
620                              struct dx_entry *entries, int levels)
621 {
622         unsigned blocksize = dir->i_sb->s_blocksize;
623         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
624         unsigned bcount = 0;
625         struct buffer_head *bh;
626         int err;
627         printk("%i indexed blocks...\n", count);
628         for (i = 0; i < count; i++, entries++)
629         {
630                 ext4_lblk_t block = dx_get_block(entries);
631                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
632                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
633                 struct stats stats;
634                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
635                 bh = ext4_bread(NULL,dir, block, 0);
636                 if (!bh || IS_ERR(bh))
637                         continue;
638                 stats = levels?
639                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
640                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
641                 names += stats.names;
642                 space += stats.space;
643                 bcount += stats.bcount;
644                 brelse(bh);
645         }
646         if (bcount)
647                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
648                        levels ? "" : "   ", names, space/bcount,
649                        (space/bcount)*100/blocksize);
650         return (struct stats) { names, space, bcount};
651 }
652 #endif /* DX_DEBUG */
653 
654 /*
655  * Probe for a directory leaf block to search.
656  *
657  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
658  * error in the directory index, and the caller should fall back to
659  * searching the directory normally.  The callers of dx_probe **MUST**
660  * check for this error code, and make sure it never gets reflected
661  * back to userspace.
662  */
663 static struct dx_frame *
664 dx_probe(const struct qstr *d_name, struct inode *dir,
665          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
666 {
667         unsigned count, indirect;
668         struct dx_entry *at, *entries, *p, *q, *m;
669         struct dx_root *root;
670         struct dx_frame *frame = frame_in;
671         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
672         u32 hash;
673 
674         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
675         if (IS_ERR(frame->bh))
676                 return (struct dx_frame *) frame->bh;
677 
678         root = (struct dx_root *) frame->bh->b_data;
679         if (root->info.hash_version != DX_HASH_TEA &&
680             root->info.hash_version != DX_HASH_HALF_MD4 &&
681             root->info.hash_version != DX_HASH_LEGACY) {
682                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
683                              root->info.hash_version);
684                 goto fail;
685         }
686         hinfo->hash_version = root->info.hash_version;
687         if (hinfo->hash_version <= DX_HASH_TEA)
688                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
689         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
690         if (d_name)
691                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
692         hash = hinfo->hash;
693 
694         if (root->info.unused_flags & 1) {
695                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
696                              root->info.unused_flags);
697                 goto fail;
698         }
699 
700         if ((indirect = root->info.indirect_levels) > 1) {
701                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
702                              root->info.indirect_levels);
703                 goto fail;
704         }
705 
706         entries = (struct dx_entry *) (((char *)&root->info) +
707                                        root->info.info_length);
708 
709         if (dx_get_limit(entries) != dx_root_limit(dir,
710                                                    root->info.info_length)) {
711                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
712                 goto fail;
713         }
714 
715         dxtrace(printk("Look up %x", hash));
716         while (1) {
717                 count = dx_get_count(entries);
718                 if (!count || count > dx_get_limit(entries)) {
719                         ext4_warning(dir->i_sb,
720                                      "dx entry: no count or count > limit");
721                         goto fail;
722                 }
723 
724                 p = entries + 1;
725                 q = entries + count - 1;
726                 while (p <= q) {
727                         m = p + (q - p)/2;
728                         dxtrace(printk("."));
729                         if (dx_get_hash(m) > hash)
730                                 q = m - 1;
731                         else
732                                 p = m + 1;
733                 }
734 
735                 if (0) { // linear search cross check
736                         unsigned n = count - 1;
737                         at = entries;
738                         while (n--)
739                         {
740                                 dxtrace(printk(","));
741                                 if (dx_get_hash(++at) > hash)
742                                 {
743                                         at--;
744                                         break;
745                                 }
746                         }
747                         assert (at == p - 1);
748                 }
749 
750                 at = p - 1;
751                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
752                 frame->entries = entries;
753                 frame->at = at;
754                 if (!indirect--)
755                         return frame;
756                 frame++;
757                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
758                 if (IS_ERR(frame->bh)) {
759                         ret_err = (struct dx_frame *) frame->bh;
760                         frame->bh = NULL;
761                         goto fail;
762                 }
763                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
764 
765                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
766                         ext4_warning(dir->i_sb,
767                                      "dx entry: limit != node limit");
768                         goto fail;
769                 }
770         }
771 fail:
772         while (frame >= frame_in) {
773                 brelse(frame->bh);
774                 frame--;
775         }
776         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
777                 ext4_warning(dir->i_sb,
778                              "Corrupt dir inode %lu, running e2fsck is "
779                              "recommended.", dir->i_ino);
780         return ret_err;
781 }
782 
783 static void dx_release (struct dx_frame *frames)
784 {
785         if (frames[0].bh == NULL)
786                 return;
787 
788         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
789                 brelse(frames[1].bh);
790         brelse(frames[0].bh);
791 }
792 
793 /*
794  * This function increments the frame pointer to search the next leaf
795  * block, and reads in the necessary intervening nodes if the search
796  * should be necessary.  Whether or not the search is necessary is
797  * controlled by the hash parameter.  If the hash value is even, then
798  * the search is only continued if the next block starts with that
799  * hash value.  This is used if we are searching for a specific file.
800  *
801  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
802  *
803  * This function returns 1 if the caller should continue to search,
804  * or 0 if it should not.  If there is an error reading one of the
805  * index blocks, it will a negative error code.
806  *
807  * If start_hash is non-null, it will be filled in with the starting
808  * hash of the next page.
809  */
810 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
811                                  struct dx_frame *frame,
812                                  struct dx_frame *frames,
813                                  __u32 *start_hash)
814 {
815         struct dx_frame *p;
816         struct buffer_head *bh;
817         int num_frames = 0;
818         __u32 bhash;
819 
820         p = frame;
821         /*
822          * Find the next leaf page by incrementing the frame pointer.
823          * If we run out of entries in the interior node, loop around and
824          * increment pointer in the parent node.  When we break out of
825          * this loop, num_frames indicates the number of interior
826          * nodes need to be read.
827          */
828         while (1) {
829                 if (++(p->at) < p->entries + dx_get_count(p->entries))
830                         break;
831                 if (p == frames)
832                         return 0;
833                 num_frames++;
834                 p--;
835         }
836 
837         /*
838          * If the hash is 1, then continue only if the next page has a
839          * continuation hash of any value.  This is used for readdir
840          * handling.  Otherwise, check to see if the hash matches the
841          * desired contiuation hash.  If it doesn't, return since
842          * there's no point to read in the successive index pages.
843          */
844         bhash = dx_get_hash(p->at);
845         if (start_hash)
846                 *start_hash = bhash;
847         if ((hash & 1) == 0) {
848                 if ((bhash & ~1) != hash)
849                         return 0;
850         }
851         /*
852          * If the hash is HASH_NB_ALWAYS, we always go to the next
853          * block so no check is necessary
854          */
855         while (num_frames--) {
856                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
857                 if (IS_ERR(bh))
858                         return PTR_ERR(bh);
859                 p++;
860                 brelse(p->bh);
861                 p->bh = bh;
862                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
863         }
864         return 1;
865 }
866 
867 
868 /*
869  * This function fills a red-black tree with information from a
870  * directory block.  It returns the number directory entries loaded
871  * into the tree.  If there is an error it is returned in err.
872  */
873 static int htree_dirblock_to_tree(struct file *dir_file,
874                                   struct inode *dir, ext4_lblk_t block,
875                                   struct dx_hash_info *hinfo,
876                                   __u32 start_hash, __u32 start_minor_hash)
877 {
878         struct buffer_head *bh;
879         struct ext4_dir_entry_2 *de, *top;
880         int err = 0, count = 0;
881 
882         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
883                                                         (unsigned long)block));
884         bh = ext4_read_dirblock(dir, block, DIRENT);
885         if (IS_ERR(bh))
886                 return PTR_ERR(bh);
887 
888         de = (struct ext4_dir_entry_2 *) bh->b_data;
889         top = (struct ext4_dir_entry_2 *) ((char *) de +
890                                            dir->i_sb->s_blocksize -
891                                            EXT4_DIR_REC_LEN(0));
892         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
893                 if (ext4_check_dir_entry(dir, NULL, de, bh,
894                                 bh->b_data, bh->b_size,
895                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
896                                          + ((char *)de - bh->b_data))) {
897                         /* silently ignore the rest of the block */
898                         break;
899                 }
900                 ext4fs_dirhash(de->name, de->name_len, hinfo);
901                 if ((hinfo->hash < start_hash) ||
902                     ((hinfo->hash == start_hash) &&
903                      (hinfo->minor_hash < start_minor_hash)))
904                         continue;
905                 if (de->inode == 0)
906                         continue;
907                 if ((err = ext4_htree_store_dirent(dir_file,
908                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
909                         brelse(bh);
910                         return err;
911                 }
912                 count++;
913         }
914         brelse(bh);
915         return count;
916 }
917 
918 
919 /*
920  * This function fills a red-black tree with information from a
921  * directory.  We start scanning the directory in hash order, starting
922  * at start_hash and start_minor_hash.
923  *
924  * This function returns the number of entries inserted into the tree,
925  * or a negative error code.
926  */
927 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
928                          __u32 start_minor_hash, __u32 *next_hash)
929 {
930         struct dx_hash_info hinfo;
931         struct ext4_dir_entry_2 *de;
932         struct dx_frame frames[2], *frame;
933         struct inode *dir;
934         ext4_lblk_t block;
935         int count = 0;
936         int ret, err;
937         __u32 hashval;
938 
939         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
940                        start_hash, start_minor_hash));
941         dir = file_inode(dir_file);
942         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
943                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
944                 if (hinfo.hash_version <= DX_HASH_TEA)
945                         hinfo.hash_version +=
946                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
947                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
948                 if (ext4_has_inline_data(dir)) {
949                         int has_inline_data = 1;
950                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
951                                                         &hinfo, start_hash,
952                                                         start_minor_hash,
953                                                         &has_inline_data);
954                         if (has_inline_data) {
955                                 *next_hash = ~0;
956                                 return count;
957                         }
958                 }
959                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
960                                                start_hash, start_minor_hash);
961                 *next_hash = ~0;
962                 return count;
963         }
964         hinfo.hash = start_hash;
965         hinfo.minor_hash = 0;
966         frame = dx_probe(NULL, dir, &hinfo, frames);
967         if (IS_ERR(frame))
968                 return PTR_ERR(frame);
969 
970         /* Add '.' and '..' from the htree header */
971         if (!start_hash && !start_minor_hash) {
972                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
973                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
974                         goto errout;
975                 count++;
976         }
977         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
978                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
979                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
980                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
981                         goto errout;
982                 count++;
983         }
984 
985         while (1) {
986                 block = dx_get_block(frame->at);
987                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
988                                              start_hash, start_minor_hash);
989                 if (ret < 0) {
990                         err = ret;
991                         goto errout;
992                 }
993                 count += ret;
994                 hashval = ~0;
995                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
996                                             frame, frames, &hashval);
997                 *next_hash = hashval;
998                 if (ret < 0) {
999                         err = ret;
1000                         goto errout;
1001                 }
1002                 /*
1003                  * Stop if:  (a) there are no more entries, or
1004                  * (b) we have inserted at least one entry and the
1005                  * next hash value is not a continuation
1006                  */
1007                 if ((ret == 0) ||
1008                     (count && ((hashval & 1) == 0)))
1009                         break;
1010         }
1011         dx_release(frames);
1012         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1013                        "next hash: %x\n", count, *next_hash));
1014         return count;
1015 errout:
1016         dx_release(frames);
1017         return (err);
1018 }
1019 
1020 static inline int search_dirblock(struct buffer_head *bh,
1021                                   struct inode *dir,
1022                                   const struct qstr *d_name,
1023                                   unsigned int offset,
1024                                   struct ext4_dir_entry_2 **res_dir)
1025 {
1026         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1027                           d_name, offset, res_dir);
1028 }
1029 
1030 /*
1031  * Directory block splitting, compacting
1032  */
1033 
1034 /*
1035  * Create map of hash values, offsets, and sizes, stored at end of block.
1036  * Returns number of entries mapped.
1037  */
1038 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1039                        struct dx_hash_info *hinfo,
1040                        struct dx_map_entry *map_tail)
1041 {
1042         int count = 0;
1043         char *base = (char *) de;
1044         struct dx_hash_info h = *hinfo;
1045 
1046         while ((char *) de < base + blocksize) {
1047                 if (de->name_len && de->inode) {
1048                         ext4fs_dirhash(de->name, de->name_len, &h);
1049                         map_tail--;
1050                         map_tail->hash = h.hash;
1051                         map_tail->offs = ((char *) de - base)>>2;
1052                         map_tail->size = le16_to_cpu(de->rec_len);
1053                         count++;
1054                         cond_resched();
1055                 }
1056                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1057                 de = ext4_next_entry(de, blocksize);
1058         }
1059         return count;
1060 }
1061 
1062 /* Sort map by hash value */
1063 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1064 {
1065         struct dx_map_entry *p, *q, *top = map + count - 1;
1066         int more;
1067         /* Combsort until bubble sort doesn't suck */
1068         while (count > 2) {
1069                 count = count*10/13;
1070                 if (count - 9 < 2) /* 9, 10 -> 11 */
1071                         count = 11;
1072                 for (p = top, q = p - count; q >= map; p--, q--)
1073                         if (p->hash < q->hash)
1074                                 swap(*p, *q);
1075         }
1076         /* Garden variety bubble sort */
1077         do {
1078                 more = 0;
1079                 q = top;
1080                 while (q-- > map) {
1081                         if (q[1].hash >= q[0].hash)
1082                                 continue;
1083                         swap(*(q+1), *q);
1084                         more = 1;
1085                 }
1086         } while(more);
1087 }
1088 
1089 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1090 {
1091         struct dx_entry *entries = frame->entries;
1092         struct dx_entry *old = frame->at, *new = old + 1;
1093         int count = dx_get_count(entries);
1094 
1095         assert(count < dx_get_limit(entries));
1096         assert(old < entries + count);
1097         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1098         dx_set_hash(new, hash);
1099         dx_set_block(new, block);
1100         dx_set_count(entries, count + 1);
1101 }
1102 
1103 /*
1104  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1105  *
1106  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1107  * `de != NULL' is guaranteed by caller.
1108  */
1109 static inline int ext4_match (int len, const char * const name,
1110                               struct ext4_dir_entry_2 * de)
1111 {
1112         if (len != de->name_len)
1113                 return 0;
1114         if (!de->inode)
1115                 return 0;
1116         return !memcmp(name, de->name, len);
1117 }
1118 
1119 /*
1120  * Returns 0 if not found, -1 on failure, and 1 on success
1121  */
1122 int search_dir(struct buffer_head *bh,
1123                char *search_buf,
1124                int buf_size,
1125                struct inode *dir,
1126                const struct qstr *d_name,
1127                unsigned int offset,
1128                struct ext4_dir_entry_2 **res_dir)
1129 {
1130         struct ext4_dir_entry_2 * de;
1131         char * dlimit;
1132         int de_len;
1133         const char *name = d_name->name;
1134         int namelen = d_name->len;
1135 
1136         de = (struct ext4_dir_entry_2 *)search_buf;
1137         dlimit = search_buf + buf_size;
1138         while ((char *) de < dlimit) {
1139                 /* this code is executed quadratically often */
1140                 /* do minimal checking `by hand' */
1141 
1142                 if ((char *) de + namelen <= dlimit &&
1143                     ext4_match (namelen, name, de)) {
1144                         /* found a match - just to be sure, do a full check */
1145                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1146                                                  bh->b_size, offset))
1147                                 return -1;
1148                         *res_dir = de;
1149                         return 1;
1150                 }
1151                 /* prevent looping on a bad block */
1152                 de_len = ext4_rec_len_from_disk(de->rec_len,
1153                                                 dir->i_sb->s_blocksize);
1154                 if (de_len <= 0)
1155                         return -1;
1156                 offset += de_len;
1157                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1158         }
1159         return 0;
1160 }
1161 
1162 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1163                                struct ext4_dir_entry *de)
1164 {
1165         struct super_block *sb = dir->i_sb;
1166 
1167         if (!is_dx(dir))
1168                 return 0;
1169         if (block == 0)
1170                 return 1;
1171         if (de->inode == 0 &&
1172             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1173                         sb->s_blocksize)
1174                 return 1;
1175         return 0;
1176 }
1177 
1178 /*
1179  *      ext4_find_entry()
1180  *
1181  * finds an entry in the specified directory with the wanted name. It
1182  * returns the cache buffer in which the entry was found, and the entry
1183  * itself (as a parameter - res_dir). It does NOT read the inode of the
1184  * entry - you'll have to do that yourself if you want to.
1185  *
1186  * The returned buffer_head has ->b_count elevated.  The caller is expected
1187  * to brelse() it when appropriate.
1188  */
1189 static struct buffer_head * ext4_find_entry (struct inode *dir,
1190                                         const struct qstr *d_name,
1191                                         struct ext4_dir_entry_2 **res_dir,
1192                                         int *inlined)
1193 {
1194         struct super_block *sb;
1195         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1196         struct buffer_head *bh, *ret = NULL;
1197         ext4_lblk_t start, block, b;
1198         const u8 *name = d_name->name;
1199         int ra_max = 0;         /* Number of bh's in the readahead
1200                                    buffer, bh_use[] */
1201         int ra_ptr = 0;         /* Current index into readahead
1202                                    buffer */
1203         int num = 0;
1204         ext4_lblk_t  nblocks;
1205         int i, namelen;
1206 
1207         *res_dir = NULL;
1208         sb = dir->i_sb;
1209         namelen = d_name->len;
1210         if (namelen > EXT4_NAME_LEN)
1211                 return NULL;
1212 
1213         if (ext4_has_inline_data(dir)) {
1214                 int has_inline_data = 1;
1215                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1216                                              &has_inline_data);
1217                 if (has_inline_data) {
1218                         if (inlined)
1219                                 *inlined = 1;
1220                         return ret;
1221                 }
1222         }
1223 
1224         if ((namelen <= 2) && (name[0] == '.') &&
1225             (name[1] == '.' || name[1] == '\0')) {
1226                 /*
1227                  * "." or ".." will only be in the first block
1228                  * NFS may look up ".."; "." should be handled by the VFS
1229                  */
1230                 block = start = 0;
1231                 nblocks = 1;
1232                 goto restart;
1233         }
1234         if (is_dx(dir)) {
1235                 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1236                 /*
1237                  * On success, or if the error was file not found,
1238                  * return.  Otherwise, fall back to doing a search the
1239                  * old fashioned way.
1240                  */
1241                 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1242                         return bh;
1243                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1244                                "falling back\n"));
1245         }
1246         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1247         start = EXT4_I(dir)->i_dir_start_lookup;
1248         if (start >= nblocks)
1249                 start = 0;
1250         block = start;
1251 restart:
1252         do {
1253                 /*
1254                  * We deal with the read-ahead logic here.
1255                  */
1256                 if (ra_ptr >= ra_max) {
1257                         /* Refill the readahead buffer */
1258                         ra_ptr = 0;
1259                         b = block;
1260                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1261                                 /*
1262                                  * Terminate if we reach the end of the
1263                                  * directory and must wrap, or if our
1264                                  * search has finished at this block.
1265                                  */
1266                                 if (b >= nblocks || (num && block == start)) {
1267                                         bh_use[ra_max] = NULL;
1268                                         break;
1269                                 }
1270                                 num++;
1271                                 bh = ext4_getblk(NULL, dir, b++, 0);
1272                                 if (unlikely(IS_ERR(bh))) {
1273                                         if (ra_max == 0)
1274                                                 return bh;
1275                                         break;
1276                                 }
1277                                 bh_use[ra_max] = bh;
1278                                 if (bh)
1279                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1280                                                     1, &bh);
1281                         }
1282                 }
1283                 if ((bh = bh_use[ra_ptr++]) == NULL)
1284                         goto next;
1285                 wait_on_buffer(bh);
1286                 if (!buffer_uptodate(bh)) {
1287                         /* read error, skip block & hope for the best */
1288                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1289                                          (unsigned long) block);
1290                         brelse(bh);
1291                         goto next;
1292                 }
1293                 if (!buffer_verified(bh) &&
1294                     !is_dx_internal_node(dir, block,
1295                                          (struct ext4_dir_entry *)bh->b_data) &&
1296                     !ext4_dirent_csum_verify(dir,
1297                                 (struct ext4_dir_entry *)bh->b_data)) {
1298                         EXT4_ERROR_INODE(dir, "checksumming directory "
1299                                          "block %lu", (unsigned long)block);
1300                         brelse(bh);
1301                         goto next;
1302                 }
1303                 set_buffer_verified(bh);
1304                 i = search_dirblock(bh, dir, d_name,
1305                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1306                 if (i == 1) {
1307                         EXT4_I(dir)->i_dir_start_lookup = block;
1308                         ret = bh;
1309                         goto cleanup_and_exit;
1310                 } else {
1311                         brelse(bh);
1312                         if (i < 0)
1313                                 goto cleanup_and_exit;
1314                 }
1315         next:
1316                 if (++block >= nblocks)
1317                         block = 0;
1318         } while (block != start);
1319 
1320         /*
1321          * If the directory has grown while we were searching, then
1322          * search the last part of the directory before giving up.
1323          */
1324         block = nblocks;
1325         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1326         if (block < nblocks) {
1327                 start = 0;
1328                 goto restart;
1329         }
1330 
1331 cleanup_and_exit:
1332         /* Clean up the read-ahead blocks */
1333         for (; ra_ptr < ra_max; ra_ptr++)
1334                 brelse(bh_use[ra_ptr]);
1335         return ret;
1336 }
1337 
1338 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1339                        struct ext4_dir_entry_2 **res_dir)
1340 {
1341         struct super_block * sb = dir->i_sb;
1342         struct dx_hash_info     hinfo;
1343         struct dx_frame frames[2], *frame;
1344         struct buffer_head *bh;
1345         ext4_lblk_t block;
1346         int retval;
1347 
1348         frame = dx_probe(d_name, dir, &hinfo, frames);
1349         if (IS_ERR(frame))
1350                 return (struct buffer_head *) frame;
1351         do {
1352                 block = dx_get_block(frame->at);
1353                 bh = ext4_read_dirblock(dir, block, DIRENT);
1354                 if (IS_ERR(bh))
1355                         goto errout;
1356 
1357                 retval = search_dirblock(bh, dir, d_name,
1358                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1359                                          res_dir);
1360                 if (retval == 1)
1361                         goto success;
1362                 brelse(bh);
1363                 if (retval == -1) {
1364                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1365                         goto errout;
1366                 }
1367 
1368                 /* Check to see if we should continue to search */
1369                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1370                                                frames, NULL);
1371                 if (retval < 0) {
1372                         ext4_warning(sb,
1373                              "error %d reading index page in directory #%lu",
1374                              retval, dir->i_ino);
1375                         bh = ERR_PTR(retval);
1376                         goto errout;
1377                 }
1378         } while (retval == 1);
1379 
1380         bh = NULL;
1381 errout:
1382         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1383 success:
1384         dx_release(frames);
1385         return bh;
1386 }
1387 
1388 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1389 {
1390         struct inode *inode;
1391         struct ext4_dir_entry_2 *de;
1392         struct buffer_head *bh;
1393 
1394         if (dentry->d_name.len > EXT4_NAME_LEN)
1395                 return ERR_PTR(-ENAMETOOLONG);
1396 
1397         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1398         if (IS_ERR(bh))
1399                 return (struct dentry *) bh;
1400         inode = NULL;
1401         if (bh) {
1402                 __u32 ino = le32_to_cpu(de->inode);
1403                 brelse(bh);
1404                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1405                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1406                         return ERR_PTR(-EIO);
1407                 }
1408                 if (unlikely(ino == dir->i_ino)) {
1409                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1410                                          dentry);
1411                         return ERR_PTR(-EIO);
1412                 }
1413                 inode = ext4_iget_normal(dir->i_sb, ino);
1414                 if (inode == ERR_PTR(-ESTALE)) {
1415                         EXT4_ERROR_INODE(dir,
1416                                          "deleted inode referenced: %u",
1417                                          ino);
1418                         return ERR_PTR(-EIO);
1419                 }
1420         }
1421         return d_splice_alias(inode, dentry);
1422 }
1423 
1424 
1425 struct dentry *ext4_get_parent(struct dentry *child)
1426 {
1427         __u32 ino;
1428         static const struct qstr dotdot = QSTR_INIT("..", 2);
1429         struct ext4_dir_entry_2 * de;
1430         struct buffer_head *bh;
1431 
1432         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1433         if (IS_ERR(bh))
1434                 return (struct dentry *) bh;
1435         if (!bh)
1436                 return ERR_PTR(-ENOENT);
1437         ino = le32_to_cpu(de->inode);
1438         brelse(bh);
1439 
1440         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1441                 EXT4_ERROR_INODE(child->d_inode,
1442                                  "bad parent inode number: %u", ino);
1443                 return ERR_PTR(-EIO);
1444         }
1445 
1446         return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1447 }
1448 
1449 /*
1450  * Move count entries from end of map between two memory locations.
1451  * Returns pointer to last entry moved.
1452  */
1453 static struct ext4_dir_entry_2 *
1454 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1455                 unsigned blocksize)
1456 {
1457         unsigned rec_len = 0;
1458 
1459         while (count--) {
1460                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1461                                                 (from + (map->offs<<2));
1462                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1463                 memcpy (to, de, rec_len);
1464                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1465                                 ext4_rec_len_to_disk(rec_len, blocksize);
1466                 de->inode = 0;
1467                 map++;
1468                 to += rec_len;
1469         }
1470         return (struct ext4_dir_entry_2 *) (to - rec_len);
1471 }
1472 
1473 /*
1474  * Compact each dir entry in the range to the minimal rec_len.
1475  * Returns pointer to last entry in range.
1476  */
1477 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1478 {
1479         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1480         unsigned rec_len = 0;
1481 
1482         prev = to = de;
1483         while ((char*)de < base + blocksize) {
1484                 next = ext4_next_entry(de, blocksize);
1485                 if (de->inode && de->name_len) {
1486                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1487                         if (de > to)
1488                                 memmove(to, de, rec_len);
1489                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1490                         prev = to;
1491                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1492                 }
1493                 de = next;
1494         }
1495         return prev;
1496 }
1497 
1498 /*
1499  * Split a full leaf block to make room for a new dir entry.
1500  * Allocate a new block, and move entries so that they are approx. equally full.
1501  * Returns pointer to de in block into which the new entry will be inserted.
1502  */
1503 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1504                         struct buffer_head **bh,struct dx_frame *frame,
1505                         struct dx_hash_info *hinfo)
1506 {
1507         unsigned blocksize = dir->i_sb->s_blocksize;
1508         unsigned count, continued;
1509         struct buffer_head *bh2;
1510         ext4_lblk_t newblock;
1511         u32 hash2;
1512         struct dx_map_entry *map;
1513         char *data1 = (*bh)->b_data, *data2;
1514         unsigned split, move, size;
1515         struct ext4_dir_entry_2 *de = NULL, *de2;
1516         struct ext4_dir_entry_tail *t;
1517         int     csum_size = 0;
1518         int     err = 0, i;
1519 
1520         if (ext4_has_metadata_csum(dir->i_sb))
1521                 csum_size = sizeof(struct ext4_dir_entry_tail);
1522 
1523         bh2 = ext4_append(handle, dir, &newblock);
1524         if (IS_ERR(bh2)) {
1525                 brelse(*bh);
1526                 *bh = NULL;
1527                 return (struct ext4_dir_entry_2 *) bh2;
1528         }
1529 
1530         BUFFER_TRACE(*bh, "get_write_access");
1531         err = ext4_journal_get_write_access(handle, *bh);
1532         if (err)
1533                 goto journal_error;
1534 
1535         BUFFER_TRACE(frame->bh, "get_write_access");
1536         err = ext4_journal_get_write_access(handle, frame->bh);
1537         if (err)
1538                 goto journal_error;
1539 
1540         data2 = bh2->b_data;
1541 
1542         /* create map in the end of data2 block */
1543         map = (struct dx_map_entry *) (data2 + blocksize);
1544         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1545                              blocksize, hinfo, map);
1546         map -= count;
1547         dx_sort_map(map, count);
1548         /* Split the existing block in the middle, size-wise */
1549         size = 0;
1550         move = 0;
1551         for (i = count-1; i >= 0; i--) {
1552                 /* is more than half of this entry in 2nd half of the block? */
1553                 if (size + map[i].size/2 > blocksize/2)
1554                         break;
1555                 size += map[i].size;
1556                 move++;
1557         }
1558         /* map index at which we will split */
1559         split = count - move;
1560         hash2 = map[split].hash;
1561         continued = hash2 == map[split - 1].hash;
1562         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1563                         (unsigned long)dx_get_block(frame->at),
1564                                         hash2, split, count-split));
1565 
1566         /* Fancy dance to stay within two buffers */
1567         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1568         de = dx_pack_dirents(data1, blocksize);
1569         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1570                                            (char *) de,
1571                                            blocksize);
1572         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1573                                             (char *) de2,
1574                                             blocksize);
1575         if (csum_size) {
1576                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1577                 initialize_dirent_tail(t, blocksize);
1578 
1579                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1580                 initialize_dirent_tail(t, blocksize);
1581         }
1582 
1583         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1584         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1585 
1586         /* Which block gets the new entry? */
1587         if (hinfo->hash >= hash2) {
1588                 swap(*bh, bh2);
1589                 de = de2;
1590         }
1591         dx_insert_block(frame, hash2 + continued, newblock);
1592         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1593         if (err)
1594                 goto journal_error;
1595         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1596         if (err)
1597                 goto journal_error;
1598         brelse(bh2);
1599         dxtrace(dx_show_index("frame", frame->entries));
1600         return de;
1601 
1602 journal_error:
1603         brelse(*bh);
1604         brelse(bh2);
1605         *bh = NULL;
1606         ext4_std_error(dir->i_sb, err);
1607         return ERR_PTR(err);
1608 }
1609 
1610 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1611                       struct buffer_head *bh,
1612                       void *buf, int buf_size,
1613                       const char *name, int namelen,
1614                       struct ext4_dir_entry_2 **dest_de)
1615 {
1616         struct ext4_dir_entry_2 *de;
1617         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1618         int nlen, rlen;
1619         unsigned int offset = 0;
1620         char *top;
1621 
1622         de = (struct ext4_dir_entry_2 *)buf;
1623         top = buf + buf_size - reclen;
1624         while ((char *) de <= top) {
1625                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1626                                          buf, buf_size, offset))
1627                         return -EIO;
1628                 if (ext4_match(namelen, name, de))
1629                         return -EEXIST;
1630                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1631                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1632                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1633                         break;
1634                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1635                 offset += rlen;
1636         }
1637         if ((char *) de > top)
1638                 return -ENOSPC;
1639 
1640         *dest_de = de;
1641         return 0;
1642 }
1643 
1644 void ext4_insert_dentry(struct inode *inode,
1645                         struct ext4_dir_entry_2 *de,
1646                         int buf_size,
1647                         const char *name, int namelen)
1648 {
1649 
1650         int nlen, rlen;
1651 
1652         nlen = EXT4_DIR_REC_LEN(de->name_len);
1653         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1654         if (de->inode) {
1655                 struct ext4_dir_entry_2 *de1 =
1656                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1657                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1658                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1659                 de = de1;
1660         }
1661         de->file_type = EXT4_FT_UNKNOWN;
1662         de->inode = cpu_to_le32(inode->i_ino);
1663         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1664         de->name_len = namelen;
1665         memcpy(de->name, name, namelen);
1666 }
1667 /*
1668  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1669  * it points to a directory entry which is guaranteed to be large
1670  * enough for new directory entry.  If de is NULL, then
1671  * add_dirent_to_buf will attempt search the directory block for
1672  * space.  It will return -ENOSPC if no space is available, and -EIO
1673  * and -EEXIST if directory entry already exists.
1674  */
1675 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1676                              struct inode *inode, struct ext4_dir_entry_2 *de,
1677                              struct buffer_head *bh)
1678 {
1679         struct inode    *dir = dentry->d_parent->d_inode;
1680         const char      *name = dentry->d_name.name;
1681         int             namelen = dentry->d_name.len;
1682         unsigned int    blocksize = dir->i_sb->s_blocksize;
1683         int             csum_size = 0;
1684         int             err;
1685 
1686         if (ext4_has_metadata_csum(inode->i_sb))
1687                 csum_size = sizeof(struct ext4_dir_entry_tail);
1688 
1689         if (!de) {
1690                 err = ext4_find_dest_de(dir, inode,
1691                                         bh, bh->b_data, blocksize - csum_size,
1692                                         name, namelen, &de);
1693                 if (err)
1694                         return err;
1695         }
1696         BUFFER_TRACE(bh, "get_write_access");
1697         err = ext4_journal_get_write_access(handle, bh);
1698         if (err) {
1699                 ext4_std_error(dir->i_sb, err);
1700                 return err;
1701         }
1702 
1703         /* By now the buffer is marked for journaling */
1704         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1705 
1706         /*
1707          * XXX shouldn't update any times until successful
1708          * completion of syscall, but too many callers depend
1709          * on this.
1710          *
1711          * XXX similarly, too many callers depend on
1712          * ext4_new_inode() setting the times, but error
1713          * recovery deletes the inode, so the worst that can
1714          * happen is that the times are slightly out of date
1715          * and/or different from the directory change time.
1716          */
1717         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1718         ext4_update_dx_flag(dir);
1719         dir->i_version++;
1720         ext4_mark_inode_dirty(handle, dir);
1721         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1722         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1723         if (err)
1724                 ext4_std_error(dir->i_sb, err);
1725         return 0;
1726 }
1727 
1728 /*
1729  * This converts a one block unindexed directory to a 3 block indexed
1730  * directory, and adds the dentry to the indexed directory.
1731  */
1732 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1733                             struct inode *inode, struct buffer_head *bh)
1734 {
1735         struct inode    *dir = dentry->d_parent->d_inode;
1736         const char      *name = dentry->d_name.name;
1737         int             namelen = dentry->d_name.len;
1738         struct buffer_head *bh2;
1739         struct dx_root  *root;
1740         struct dx_frame frames[2], *frame;
1741         struct dx_entry *entries;
1742         struct ext4_dir_entry_2 *de, *de2;
1743         struct ext4_dir_entry_tail *t;
1744         char            *data1, *top;
1745         unsigned        len;
1746         int             retval;
1747         unsigned        blocksize;
1748         struct dx_hash_info hinfo;
1749         ext4_lblk_t  block;
1750         struct fake_dirent *fde;
1751         int             csum_size = 0;
1752 
1753         if (ext4_has_metadata_csum(inode->i_sb))
1754                 csum_size = sizeof(struct ext4_dir_entry_tail);
1755 
1756         blocksize =  dir->i_sb->s_blocksize;
1757         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1758         BUFFER_TRACE(bh, "get_write_access");
1759         retval = ext4_journal_get_write_access(handle, bh);
1760         if (retval) {
1761                 ext4_std_error(dir->i_sb, retval);
1762                 brelse(bh);
1763                 return retval;
1764         }
1765         root = (struct dx_root *) bh->b_data;
1766 
1767         /* The 0th block becomes the root, move the dirents out */
1768         fde = &root->dotdot;
1769         de = (struct ext4_dir_entry_2 *)((char *)fde +
1770                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1771         if ((char *) de >= (((char *) root) + blocksize)) {
1772                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1773                 brelse(bh);
1774                 return -EIO;
1775         }
1776         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1777 
1778         /* Allocate new block for the 0th block's dirents */
1779         bh2 = ext4_append(handle, dir, &block);
1780         if (IS_ERR(bh2)) {
1781                 brelse(bh);
1782                 return PTR_ERR(bh2);
1783         }
1784         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1785         data1 = bh2->b_data;
1786 
1787         memcpy (data1, de, len);
1788         de = (struct ext4_dir_entry_2 *) data1;
1789         top = data1 + len;
1790         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1791                 de = de2;
1792         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1793                                            (char *) de,
1794                                            blocksize);
1795 
1796         if (csum_size) {
1797                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1798                 initialize_dirent_tail(t, blocksize);
1799         }
1800 
1801         /* Initialize the root; the dot dirents already exist */
1802         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1803         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1804                                            blocksize);
1805         memset (&root->info, 0, sizeof(root->info));
1806         root->info.info_length = sizeof(root->info);
1807         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1808         entries = root->entries;
1809         dx_set_block(entries, 1);
1810         dx_set_count(entries, 1);
1811         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1812 
1813         /* Initialize as for dx_probe */
1814         hinfo.hash_version = root->info.hash_version;
1815         if (hinfo.hash_version <= DX_HASH_TEA)
1816                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1817         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1818         ext4fs_dirhash(name, namelen, &hinfo);
1819         memset(frames, 0, sizeof(frames));
1820         frame = frames;
1821         frame->entries = entries;
1822         frame->at = entries;
1823         frame->bh = bh;
1824         bh = bh2;
1825 
1826         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1827         if (retval)
1828                 goto out_frames;        
1829         retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
1830         if (retval)
1831                 goto out_frames;        
1832 
1833         de = do_split(handle,dir, &bh, frame, &hinfo);
1834         if (IS_ERR(de)) {
1835                 retval = PTR_ERR(de);
1836                 goto out_frames;
1837         }
1838         dx_release(frames);
1839 
1840         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1841         brelse(bh);
1842         return retval;
1843 out_frames:
1844         /*
1845          * Even if the block split failed, we have to properly write
1846          * out all the changes we did so far. Otherwise we can end up
1847          * with corrupted filesystem.
1848          */
1849         ext4_mark_inode_dirty(handle, dir);
1850         dx_release(frames);
1851         return retval;
1852 }
1853 
1854 /*
1855  *      ext4_add_entry()
1856  *
1857  * adds a file entry to the specified directory, using the same
1858  * semantics as ext4_find_entry(). It returns NULL if it failed.
1859  *
1860  * NOTE!! The inode part of 'de' is left at 0 - which means you
1861  * may not sleep between calling this and putting something into
1862  * the entry, as someone else might have used it while you slept.
1863  */
1864 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1865                           struct inode *inode)
1866 {
1867         struct inode *dir = dentry->d_parent->d_inode;
1868         struct buffer_head *bh = NULL;
1869         struct ext4_dir_entry_2 *de;
1870         struct ext4_dir_entry_tail *t;
1871         struct super_block *sb;
1872         int     retval;
1873         int     dx_fallback=0;
1874         unsigned blocksize;
1875         ext4_lblk_t block, blocks;
1876         int     csum_size = 0;
1877 
1878         if (ext4_has_metadata_csum(inode->i_sb))
1879                 csum_size = sizeof(struct ext4_dir_entry_tail);
1880 
1881         sb = dir->i_sb;
1882         blocksize = sb->s_blocksize;
1883         if (!dentry->d_name.len)
1884                 return -EINVAL;
1885 
1886         if (ext4_has_inline_data(dir)) {
1887                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1888                 if (retval < 0)
1889                         return retval;
1890                 if (retval == 1) {
1891                         retval = 0;
1892                         goto out;
1893                 }
1894         }
1895 
1896         if (is_dx(dir)) {
1897                 retval = ext4_dx_add_entry(handle, dentry, inode);
1898                 if (!retval || (retval != ERR_BAD_DX_DIR))
1899                         goto out;
1900                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1901                 dx_fallback++;
1902                 ext4_mark_inode_dirty(handle, dir);
1903         }
1904         blocks = dir->i_size >> sb->s_blocksize_bits;
1905         for (block = 0; block < blocks; block++) {
1906                 bh = ext4_read_dirblock(dir, block, DIRENT);
1907                 if (IS_ERR(bh))
1908                         return PTR_ERR(bh);
1909 
1910                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1911                 if (retval != -ENOSPC)
1912                         goto out;
1913 
1914                 if (blocks == 1 && !dx_fallback &&
1915                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
1916                         retval = make_indexed_dir(handle, dentry, inode, bh);
1917                         bh = NULL; /* make_indexed_dir releases bh */
1918                         goto out;
1919                 }
1920                 brelse(bh);
1921         }
1922         bh = ext4_append(handle, dir, &block);
1923         if (IS_ERR(bh))
1924                 return PTR_ERR(bh);
1925         de = (struct ext4_dir_entry_2 *) bh->b_data;
1926         de->inode = 0;
1927         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1928 
1929         if (csum_size) {
1930                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1931                 initialize_dirent_tail(t, blocksize);
1932         }
1933 
1934         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1935 out:
1936         brelse(bh);
1937         if (retval == 0)
1938                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1939         return retval;
1940 }
1941 
1942 /*
1943  * Returns 0 for success, or a negative error value
1944  */
1945 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1946                              struct inode *inode)
1947 {
1948         struct dx_frame frames[2], *frame;
1949         struct dx_entry *entries, *at;
1950         struct dx_hash_info hinfo;
1951         struct buffer_head *bh;
1952         struct inode *dir = dentry->d_parent->d_inode;
1953         struct super_block *sb = dir->i_sb;
1954         struct ext4_dir_entry_2 *de;
1955         int err;
1956 
1957         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
1958         if (IS_ERR(frame))
1959                 return PTR_ERR(frame);
1960         entries = frame->entries;
1961         at = frame->at;
1962         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1963         if (IS_ERR(bh)) {
1964                 err = PTR_ERR(bh);
1965                 bh = NULL;
1966                 goto cleanup;
1967         }
1968 
1969         BUFFER_TRACE(bh, "get_write_access");
1970         err = ext4_journal_get_write_access(handle, bh);
1971         if (err)
1972                 goto journal_error;
1973 
1974         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1975         if (err != -ENOSPC)
1976                 goto cleanup;
1977 
1978         /* Block full, should compress but for now just split */
1979         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1980                        dx_get_count(entries), dx_get_limit(entries)));
1981         /* Need to split index? */
1982         if (dx_get_count(entries) == dx_get_limit(entries)) {
1983                 ext4_lblk_t newblock;
1984                 unsigned icount = dx_get_count(entries);
1985                 int levels = frame - frames;
1986                 struct dx_entry *entries2;
1987                 struct dx_node *node2;
1988                 struct buffer_head *bh2;
1989 
1990                 if (levels && (dx_get_count(frames->entries) ==
1991                                dx_get_limit(frames->entries))) {
1992                         ext4_warning(sb, "Directory index full!");
1993                         err = -ENOSPC;
1994                         goto cleanup;
1995                 }
1996                 bh2 = ext4_append(handle, dir, &newblock);
1997                 if (IS_ERR(bh2)) {
1998                         err = PTR_ERR(bh2);
1999                         goto cleanup;
2000                 }
2001                 node2 = (struct dx_node *)(bh2->b_data);
2002                 entries2 = node2->entries;
2003                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2004                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2005                                                            sb->s_blocksize);
2006                 BUFFER_TRACE(frame->bh, "get_write_access");
2007                 err = ext4_journal_get_write_access(handle, frame->bh);
2008                 if (err)
2009                         goto journal_error;
2010                 if (levels) {
2011                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2012                         unsigned hash2 = dx_get_hash(entries + icount1);
2013                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2014                                        icount1, icount2));
2015 
2016                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2017                         err = ext4_journal_get_write_access(handle,
2018                                                              frames[0].bh);
2019                         if (err)
2020                                 goto journal_error;
2021 
2022                         memcpy((char *) entries2, (char *) (entries + icount1),
2023                                icount2 * sizeof(struct dx_entry));
2024                         dx_set_count(entries, icount1);
2025                         dx_set_count(entries2, icount2);
2026                         dx_set_limit(entries2, dx_node_limit(dir));
2027 
2028                         /* Which index block gets the new entry? */
2029                         if (at - entries >= icount1) {
2030                                 frame->at = at = at - entries - icount1 + entries2;
2031                                 frame->entries = entries = entries2;
2032                                 swap(frame->bh, bh2);
2033                         }
2034                         dx_insert_block(frames + 0, hash2, newblock);
2035                         dxtrace(dx_show_index("node", frames[1].entries));
2036                         dxtrace(dx_show_index("node",
2037                                ((struct dx_node *) bh2->b_data)->entries));
2038                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2039                         if (err)
2040                                 goto journal_error;
2041                         brelse (bh2);
2042                 } else {
2043                         dxtrace(printk(KERN_DEBUG
2044                                        "Creating second level index...\n"));
2045                         memcpy((char *) entries2, (char *) entries,
2046                                icount * sizeof(struct dx_entry));
2047                         dx_set_limit(entries2, dx_node_limit(dir));
2048 
2049                         /* Set up root */
2050                         dx_set_count(entries, 1);
2051                         dx_set_block(entries + 0, newblock);
2052                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2053 
2054                         /* Add new access path frame */
2055                         frame = frames + 1;
2056                         frame->at = at = at - entries + entries2;
2057                         frame->entries = entries = entries2;
2058                         frame->bh = bh2;
2059                         err = ext4_journal_get_write_access(handle,
2060                                                              frame->bh);
2061                         if (err)
2062                                 goto journal_error;
2063                 }
2064                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2065                 if (err) {
2066                         ext4_std_error(inode->i_sb, err);
2067                         goto cleanup;
2068                 }
2069         }
2070         de = do_split(handle, dir, &bh, frame, &hinfo);
2071         if (IS_ERR(de)) {
2072                 err = PTR_ERR(de);
2073                 goto cleanup;
2074         }
2075         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2076         goto cleanup;
2077 
2078 journal_error:
2079         ext4_std_error(dir->i_sb, err);
2080 cleanup:
2081         brelse(bh);
2082         dx_release(frames);
2083         return err;
2084 }
2085 
2086 /*
2087  * ext4_generic_delete_entry deletes a directory entry by merging it
2088  * with the previous entry
2089  */
2090 int ext4_generic_delete_entry(handle_t *handle,
2091                               struct inode *dir,
2092                               struct ext4_dir_entry_2 *de_del,
2093                               struct buffer_head *bh,
2094                               void *entry_buf,
2095                               int buf_size,
2096                               int csum_size)
2097 {
2098         struct ext4_dir_entry_2 *de, *pde;
2099         unsigned int blocksize = dir->i_sb->s_blocksize;
2100         int i;
2101 
2102         i = 0;
2103         pde = NULL;
2104         de = (struct ext4_dir_entry_2 *)entry_buf;
2105         while (i < buf_size - csum_size) {
2106                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2107                                          bh->b_data, bh->b_size, i))
2108                         return -EIO;
2109                 if (de == de_del)  {
2110                         if (pde)
2111                                 pde->rec_len = ext4_rec_len_to_disk(
2112                                         ext4_rec_len_from_disk(pde->rec_len,
2113                                                                blocksize) +
2114                                         ext4_rec_len_from_disk(de->rec_len,
2115                                                                blocksize),
2116                                         blocksize);
2117                         else
2118                                 de->inode = 0;
2119                         dir->i_version++;
2120                         return 0;
2121                 }
2122                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2123                 pde = de;
2124                 de = ext4_next_entry(de, blocksize);
2125         }
2126         return -ENOENT;
2127 }
2128 
2129 static int ext4_delete_entry(handle_t *handle,
2130                              struct inode *dir,
2131                              struct ext4_dir_entry_2 *de_del,
2132                              struct buffer_head *bh)
2133 {
2134         int err, csum_size = 0;
2135 
2136         if (ext4_has_inline_data(dir)) {
2137                 int has_inline_data = 1;
2138                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2139                                                &has_inline_data);
2140                 if (has_inline_data)
2141                         return err;
2142         }
2143 
2144         if (ext4_has_metadata_csum(dir->i_sb))
2145                 csum_size = sizeof(struct ext4_dir_entry_tail);
2146 
2147         BUFFER_TRACE(bh, "get_write_access");
2148         err = ext4_journal_get_write_access(handle, bh);
2149         if (unlikely(err))
2150                 goto out;
2151 
2152         err = ext4_generic_delete_entry(handle, dir, de_del,
2153                                         bh, bh->b_data,
2154                                         dir->i_sb->s_blocksize, csum_size);
2155         if (err)
2156                 goto out;
2157 
2158         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2159         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2160         if (unlikely(err))
2161                 goto out;
2162 
2163         return 0;
2164 out:
2165         if (err != -ENOENT)
2166                 ext4_std_error(dir->i_sb, err);
2167         return err;
2168 }
2169 
2170 /*
2171  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2172  * since this indicates that nlinks count was previously 1.
2173  */
2174 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2175 {
2176         inc_nlink(inode);
2177         if (is_dx(inode) && inode->i_nlink > 1) {
2178                 /* limit is 16-bit i_links_count */
2179                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2180                         set_nlink(inode, 1);
2181                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2182                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2183                 }
2184         }
2185 }
2186 
2187 /*
2188  * If a directory had nlink == 1, then we should let it be 1. This indicates
2189  * directory has >EXT4_LINK_MAX subdirs.
2190  */
2191 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2192 {
2193         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2194                 drop_nlink(inode);
2195 }
2196 
2197 
2198 static int ext4_add_nondir(handle_t *handle,
2199                 struct dentry *dentry, struct inode *inode)
2200 {
2201         int err = ext4_add_entry(handle, dentry, inode);
2202         if (!err) {
2203                 ext4_mark_inode_dirty(handle, inode);
2204                 unlock_new_inode(inode);
2205                 d_instantiate(dentry, inode);
2206                 return 0;
2207         }
2208         drop_nlink(inode);
2209         unlock_new_inode(inode);
2210         iput(inode);
2211         return err;
2212 }
2213 
2214 /*
2215  * By the time this is called, we already have created
2216  * the directory cache entry for the new file, but it
2217  * is so far negative - it has no inode.
2218  *
2219  * If the create succeeds, we fill in the inode information
2220  * with d_instantiate().
2221  */
2222 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2223                        bool excl)
2224 {
2225         handle_t *handle;
2226         struct inode *inode;
2227         int err, credits, retries = 0;
2228 
2229         dquot_initialize(dir);
2230 
2231         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2232                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2233 retry:
2234         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2235                                             NULL, EXT4_HT_DIR, credits);
2236         handle = ext4_journal_current_handle();
2237         err = PTR_ERR(inode);
2238         if (!IS_ERR(inode)) {
2239                 inode->i_op = &ext4_file_inode_operations;
2240                 if (test_opt(inode->i_sb, DAX))
2241                         inode->i_fop = &ext4_dax_file_operations;
2242                 else
2243                         inode->i_fop = &ext4_file_operations;
2244                 ext4_set_aops(inode);
2245                 err = ext4_add_nondir(handle, dentry, inode);
2246                 if (!err && IS_DIRSYNC(dir))
2247                         ext4_handle_sync(handle);
2248         }
2249         if (handle)
2250                 ext4_journal_stop(handle);
2251         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2252                 goto retry;
2253         return err;
2254 }
2255 
2256 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2257                       umode_t mode, dev_t rdev)
2258 {
2259         handle_t *handle;
2260         struct inode *inode;
2261         int err, credits, retries = 0;
2262 
2263         if (!new_valid_dev(rdev))
2264                 return -EINVAL;
2265 
2266         dquot_initialize(dir);
2267 
2268         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2269                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2270 retry:
2271         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2272                                             NULL, EXT4_HT_DIR, credits);
2273         handle = ext4_journal_current_handle();
2274         err = PTR_ERR(inode);
2275         if (!IS_ERR(inode)) {
2276                 init_special_inode(inode, inode->i_mode, rdev);
2277                 inode->i_op = &ext4_special_inode_operations;
2278                 err = ext4_add_nondir(handle, dentry, inode);
2279                 if (!err && IS_DIRSYNC(dir))
2280                         ext4_handle_sync(handle);
2281         }
2282         if (handle)
2283                 ext4_journal_stop(handle);
2284         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2285                 goto retry;
2286         return err;
2287 }
2288 
2289 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2290 {
2291         handle_t *handle;
2292         struct inode *inode;
2293         int err, retries = 0;
2294 
2295         dquot_initialize(dir);
2296 
2297 retry:
2298         inode = ext4_new_inode_start_handle(dir, mode,
2299                                             NULL, 0, NULL,
2300                                             EXT4_HT_DIR,
2301                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2302                           4 + EXT4_XATTR_TRANS_BLOCKS);
2303         handle = ext4_journal_current_handle();
2304         err = PTR_ERR(inode);
2305         if (!IS_ERR(inode)) {
2306                 inode->i_op = &ext4_file_inode_operations;
2307                 if (test_opt(inode->i_sb, DAX))
2308                         inode->i_fop = &ext4_dax_file_operations;
2309                 else
2310                         inode->i_fop = &ext4_file_operations;
2311                 ext4_set_aops(inode);
2312                 d_tmpfile(dentry, inode);
2313                 err = ext4_orphan_add(handle, inode);
2314                 if (err)
2315                         goto err_unlock_inode;
2316                 mark_inode_dirty(inode);
2317                 unlock_new_inode(inode);
2318         }
2319         if (handle)
2320                 ext4_journal_stop(handle);
2321         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2322                 goto retry;
2323         return err;
2324 err_unlock_inode:
2325         ext4_journal_stop(handle);
2326         unlock_new_inode(inode);
2327         return err;
2328 }
2329 
2330 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2331                           struct ext4_dir_entry_2 *de,
2332                           int blocksize, int csum_size,
2333                           unsigned int parent_ino, int dotdot_real_len)
2334 {
2335         de->inode = cpu_to_le32(inode->i_ino);
2336         de->name_len = 1;
2337         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2338                                            blocksize);
2339         strcpy(de->name, ".");
2340         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2341 
2342         de = ext4_next_entry(de, blocksize);
2343         de->inode = cpu_to_le32(parent_ino);
2344         de->name_len = 2;
2345         if (!dotdot_real_len)
2346                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2347                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2348                                         blocksize);
2349         else
2350                 de->rec_len = ext4_rec_len_to_disk(
2351                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2352         strcpy(de->name, "..");
2353         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2354 
2355         return ext4_next_entry(de, blocksize);
2356 }
2357 
2358 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2359                              struct inode *inode)
2360 {
2361         struct buffer_head *dir_block = NULL;
2362         struct ext4_dir_entry_2 *de;
2363         struct ext4_dir_entry_tail *t;
2364         ext4_lblk_t block = 0;
2365         unsigned int blocksize = dir->i_sb->s_blocksize;
2366         int csum_size = 0;
2367         int err;
2368 
2369         if (ext4_has_metadata_csum(dir->i_sb))
2370                 csum_size = sizeof(struct ext4_dir_entry_tail);
2371 
2372         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2373                 err = ext4_try_create_inline_dir(handle, dir, inode);
2374                 if (err < 0 && err != -ENOSPC)
2375                         goto out;
2376                 if (!err)
2377                         goto out;
2378         }
2379 
2380         inode->i_size = 0;
2381         dir_block = ext4_append(handle, inode, &block);
2382         if (IS_ERR(dir_block))
2383                 return PTR_ERR(dir_block);
2384         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2385         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2386         set_nlink(inode, 2);
2387         if (csum_size) {
2388                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2389                 initialize_dirent_tail(t, blocksize);
2390         }
2391 
2392         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2393         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2394         if (err)
2395                 goto out;
2396         set_buffer_verified(dir_block);
2397 out:
2398         brelse(dir_block);
2399         return err;
2400 }
2401 
2402 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2403 {
2404         handle_t *handle;
2405         struct inode *inode;
2406         int err, credits, retries = 0;
2407 
2408         if (EXT4_DIR_LINK_MAX(dir))
2409                 return -EMLINK;
2410 
2411         dquot_initialize(dir);
2412 
2413         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2414                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2415 retry:
2416         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2417                                             &dentry->d_name,
2418                                             0, NULL, EXT4_HT_DIR, credits);
2419         handle = ext4_journal_current_handle();
2420         err = PTR_ERR(inode);
2421         if (IS_ERR(inode))
2422                 goto out_stop;
2423 
2424         inode->i_op = &ext4_dir_inode_operations;
2425         inode->i_fop = &ext4_dir_operations;
2426         err = ext4_init_new_dir(handle, dir, inode);
2427         if (err)
2428                 goto out_clear_inode;
2429         err = ext4_mark_inode_dirty(handle, inode);
2430         if (!err)
2431                 err = ext4_add_entry(handle, dentry, inode);
2432         if (err) {
2433 out_clear_inode:
2434                 clear_nlink(inode);
2435                 unlock_new_inode(inode);
2436                 ext4_mark_inode_dirty(handle, inode);
2437                 iput(inode);
2438                 goto out_stop;
2439         }
2440         ext4_inc_count(handle, dir);
2441         ext4_update_dx_flag(dir);
2442         err = ext4_mark_inode_dirty(handle, dir);
2443         if (err)
2444                 goto out_clear_inode;
2445         unlock_new_inode(inode);
2446         d_instantiate(dentry, inode);
2447         if (IS_DIRSYNC(dir))
2448                 ext4_handle_sync(handle);
2449 
2450 out_stop:
2451         if (handle)
2452                 ext4_journal_stop(handle);
2453         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2454                 goto retry;
2455         return err;
2456 }
2457 
2458 /*
2459  * routine to check that the specified directory is empty (for rmdir)
2460  */
2461 static int empty_dir(struct inode *inode)
2462 {
2463         unsigned int offset;
2464         struct buffer_head *bh;
2465         struct ext4_dir_entry_2 *de, *de1;
2466         struct super_block *sb;
2467         int err = 0;
2468 
2469         if (ext4_has_inline_data(inode)) {
2470                 int has_inline_data = 1;
2471 
2472                 err = empty_inline_dir(inode, &has_inline_data);
2473                 if (has_inline_data)
2474                         return err;
2475         }
2476 
2477         sb = inode->i_sb;
2478         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2479                 EXT4_ERROR_INODE(inode, "invalid size");
2480                 return 1;
2481         }
2482         bh = ext4_read_dirblock(inode, 0, EITHER);
2483         if (IS_ERR(bh))
2484                 return 1;
2485 
2486         de = (struct ext4_dir_entry_2 *) bh->b_data;
2487         de1 = ext4_next_entry(de, sb->s_blocksize);
2488         if (le32_to_cpu(de->inode) != inode->i_ino ||
2489                         !le32_to_cpu(de1->inode) ||
2490                         strcmp(".", de->name) ||
2491                         strcmp("..", de1->name)) {
2492                 ext4_warning(inode->i_sb,
2493                              "bad directory (dir #%lu) - no `.' or `..'",
2494                              inode->i_ino);
2495                 brelse(bh);
2496                 return 1;
2497         }
2498         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2499                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2500         de = ext4_next_entry(de1, sb->s_blocksize);
2501         while (offset < inode->i_size) {
2502                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2503                         unsigned int lblock;
2504                         err = 0;
2505                         brelse(bh);
2506                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2507                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2508                         if (IS_ERR(bh))
2509                                 return 1;
2510                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2511                 }
2512                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2513                                          bh->b_data, bh->b_size, offset)) {
2514                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2515                                                          sb->s_blocksize);
2516                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2517                         continue;
2518                 }
2519                 if (le32_to_cpu(de->inode)) {
2520                         brelse(bh);
2521                         return 0;
2522                 }
2523                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2524                 de = ext4_next_entry(de, sb->s_blocksize);
2525         }
2526         brelse(bh);
2527         return 1;
2528 }
2529 
2530 /*
2531  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2532  * such inodes, starting at the superblock, in case we crash before the
2533  * file is closed/deleted, or in case the inode truncate spans multiple
2534  * transactions and the last transaction is not recovered after a crash.
2535  *
2536  * At filesystem recovery time, we walk this list deleting unlinked
2537  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2538  *
2539  * Orphan list manipulation functions must be called under i_mutex unless
2540  * we are just creating the inode or deleting it.
2541  */
2542 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2543 {
2544         struct super_block *sb = inode->i_sb;
2545         struct ext4_sb_info *sbi = EXT4_SB(sb);
2546         struct ext4_iloc iloc;
2547         int err = 0, rc;
2548         bool dirty = false;
2549 
2550         if (!sbi->s_journal || is_bad_inode(inode))
2551                 return 0;
2552 
2553         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2554                      !mutex_is_locked(&inode->i_mutex));
2555         /*
2556          * Exit early if inode already is on orphan list. This is a big speedup
2557          * since we don't have to contend on the global s_orphan_lock.
2558          */
2559         if (!list_empty(&EXT4_I(inode)->i_orphan))
2560                 return 0;
2561 
2562         /*
2563          * Orphan handling is only valid for files with data blocks
2564          * being truncated, or files being unlinked. Note that we either
2565          * hold i_mutex, or the inode can not be referenced from outside,
2566          * so i_nlink should not be bumped due to race
2567          */
2568         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2569                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2570 
2571         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2572         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2573         if (err)
2574                 goto out;
2575 
2576         err = ext4_reserve_inode_write(handle, inode, &iloc);
2577         if (err)
2578                 goto out;
2579 
2580         mutex_lock(&sbi->s_orphan_lock);
2581         /*
2582          * Due to previous errors inode may be already a part of on-disk
2583          * orphan list. If so skip on-disk list modification.
2584          */
2585         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2586             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2587                 /* Insert this inode at the head of the on-disk orphan list */
2588                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2589                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2590                 dirty = true;
2591         }
2592         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2593         mutex_unlock(&sbi->s_orphan_lock);
2594 
2595         if (dirty) {
2596                 err = ext4_handle_dirty_super(handle, sb);
2597                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2598                 if (!err)
2599                         err = rc;
2600                 if (err) {
2601                         /*
2602                          * We have to remove inode from in-memory list if
2603                          * addition to on disk orphan list failed. Stray orphan
2604                          * list entries can cause panics at unmount time.
2605                          */
2606                         mutex_lock(&sbi->s_orphan_lock);
2607                         list_del(&EXT4_I(inode)->i_orphan);
2608                         mutex_unlock(&sbi->s_orphan_lock);
2609                 }
2610         }
2611         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2612         jbd_debug(4, "orphan inode %lu will point to %d\n",
2613                         inode->i_ino, NEXT_ORPHAN(inode));
2614 out:
2615         ext4_std_error(sb, err);
2616         return err;
2617 }
2618 
2619 /*
2620  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2621  * of such inodes stored on disk, because it is finally being cleaned up.
2622  */
2623 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2624 {
2625         struct list_head *prev;
2626         struct ext4_inode_info *ei = EXT4_I(inode);
2627         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2628         __u32 ino_next;
2629         struct ext4_iloc iloc;
2630         int err = 0;
2631 
2632         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2633                 return 0;
2634 
2635         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2636                      !mutex_is_locked(&inode->i_mutex));
2637         /* Do this quick check before taking global s_orphan_lock. */
2638         if (list_empty(&ei->i_orphan))
2639                 return 0;
2640 
2641         if (handle) {
2642                 /* Grab inode buffer early before taking global s_orphan_lock */
2643                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2644         }
2645 
2646         mutex_lock(&sbi->s_orphan_lock);
2647         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2648 
2649         prev = ei->i_orphan.prev;
2650         list_del_init(&ei->i_orphan);
2651 
2652         /* If we're on an error path, we may not have a valid
2653          * transaction handle with which to update the orphan list on
2654          * disk, but we still need to remove the inode from the linked
2655          * list in memory. */
2656         if (!handle || err) {
2657                 mutex_unlock(&sbi->s_orphan_lock);
2658                 goto out_err;
2659         }
2660 
2661         ino_next = NEXT_ORPHAN(inode);
2662         if (prev == &sbi->s_orphan) {
2663                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2664                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2665                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2666                 if (err) {
2667                         mutex_unlock(&sbi->s_orphan_lock);
2668                         goto out_brelse;
2669                 }
2670                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2671                 mutex_unlock(&sbi->s_orphan_lock);
2672                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2673         } else {
2674                 struct ext4_iloc iloc2;
2675                 struct inode *i_prev =
2676                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2677 
2678                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2679                           i_prev->i_ino, ino_next);
2680                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2681                 if (err) {
2682                         mutex_unlock(&sbi->s_orphan_lock);
2683                         goto out_brelse;
2684                 }
2685                 NEXT_ORPHAN(i_prev) = ino_next;
2686                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2687                 mutex_unlock(&sbi->s_orphan_lock);
2688         }
2689         if (err)
2690                 goto out_brelse;
2691         NEXT_ORPHAN(inode) = 0;
2692         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2693 out_err:
2694         ext4_std_error(inode->i_sb, err);
2695         return err;
2696 
2697 out_brelse:
2698         brelse(iloc.bh);
2699         goto out_err;
2700 }
2701 
2702 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2703 {
2704         int retval;
2705         struct inode *inode;
2706         struct buffer_head *bh;
2707         struct ext4_dir_entry_2 *de;
2708         handle_t *handle = NULL;
2709 
2710         /* Initialize quotas before so that eventual writes go in
2711          * separate transaction */
2712         dquot_initialize(dir);
2713         dquot_initialize(dentry->d_inode);
2714 
2715         retval = -ENOENT;
2716         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2717         if (IS_ERR(bh))
2718                 return PTR_ERR(bh);
2719         if (!bh)
2720                 goto end_rmdir;
2721 
2722         inode = dentry->d_inode;
2723 
2724         retval = -EIO;
2725         if (le32_to_cpu(de->inode) != inode->i_ino)
2726                 goto end_rmdir;
2727 
2728         retval = -ENOTEMPTY;
2729         if (!empty_dir(inode))
2730                 goto end_rmdir;
2731 
2732         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2733                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2734         if (IS_ERR(handle)) {
2735                 retval = PTR_ERR(handle);
2736                 handle = NULL;
2737                 goto end_rmdir;
2738         }
2739 
2740         if (IS_DIRSYNC(dir))
2741                 ext4_handle_sync(handle);
2742 
2743         retval = ext4_delete_entry(handle, dir, de, bh);
2744         if (retval)
2745                 goto end_rmdir;
2746         if (!EXT4_DIR_LINK_EMPTY(inode))
2747                 ext4_warning(inode->i_sb,
2748                              "empty directory has too many links (%d)",
2749                              inode->i_nlink);
2750         inode->i_version++;
2751         clear_nlink(inode);
2752         /* There's no need to set i_disksize: the fact that i_nlink is
2753          * zero will ensure that the right thing happens during any
2754          * recovery. */
2755         inode->i_size = 0;
2756         ext4_orphan_add(handle, inode);
2757         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2758         ext4_mark_inode_dirty(handle, inode);
2759         ext4_dec_count(handle, dir);
2760         ext4_update_dx_flag(dir);
2761         ext4_mark_inode_dirty(handle, dir);
2762 
2763 end_rmdir:
2764         brelse(bh);
2765         if (handle)
2766                 ext4_journal_stop(handle);
2767         return retval;
2768 }
2769 
2770 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2771 {
2772         int retval;
2773         struct inode *inode;
2774         struct buffer_head *bh;
2775         struct ext4_dir_entry_2 *de;
2776         handle_t *handle = NULL;
2777 
2778         trace_ext4_unlink_enter(dir, dentry);
2779         /* Initialize quotas before so that eventual writes go
2780          * in separate transaction */
2781         dquot_initialize(dir);
2782         dquot_initialize(dentry->d_inode);
2783 
2784         retval = -ENOENT;
2785         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2786         if (IS_ERR(bh))
2787                 return PTR_ERR(bh);
2788         if (!bh)
2789                 goto end_unlink;
2790 
2791         inode = dentry->d_inode;
2792 
2793         retval = -EIO;
2794         if (le32_to_cpu(de->inode) != inode->i_ino)
2795                 goto end_unlink;
2796 
2797         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2798                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2799         if (IS_ERR(handle)) {
2800                 retval = PTR_ERR(handle);
2801                 handle = NULL;
2802                 goto end_unlink;
2803         }
2804 
2805         if (IS_DIRSYNC(dir))
2806                 ext4_handle_sync(handle);
2807 
2808         if (!inode->i_nlink) {
2809                 ext4_warning(inode->i_sb,
2810                              "Deleting nonexistent file (%lu), %d",
2811                              inode->i_ino, inode->i_nlink);
2812                 set_nlink(inode, 1);
2813         }
2814         retval = ext4_delete_entry(handle, dir, de, bh);
2815         if (retval)
2816                 goto end_unlink;
2817         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2818         ext4_update_dx_flag(dir);
2819         ext4_mark_inode_dirty(handle, dir);
2820         drop_nlink(inode);
2821         if (!inode->i_nlink)
2822                 ext4_orphan_add(handle, inode);
2823         inode->i_ctime = ext4_current_time(inode);
2824         ext4_mark_inode_dirty(handle, inode);
2825 
2826 end_unlink:
2827         brelse(bh);
2828         if (handle)
2829                 ext4_journal_stop(handle);
2830         trace_ext4_unlink_exit(dentry, retval);
2831         return retval;
2832 }
2833 
2834 static int ext4_symlink(struct inode *dir,
2835                         struct dentry *dentry, const char *symname)
2836 {
2837         handle_t *handle;
2838         struct inode *inode;
2839         int l, err, retries = 0;
2840         int credits;
2841 
2842         l = strlen(symname)+1;
2843         if (l > dir->i_sb->s_blocksize)
2844                 return -ENAMETOOLONG;
2845 
2846         dquot_initialize(dir);
2847 
2848         if (l > EXT4_N_BLOCKS * 4) {
2849                 /*
2850                  * For non-fast symlinks, we just allocate inode and put it on
2851                  * orphan list in the first transaction => we need bitmap,
2852                  * group descriptor, sb, inode block, quota blocks, and
2853                  * possibly selinux xattr blocks.
2854                  */
2855                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2856                           EXT4_XATTR_TRANS_BLOCKS;
2857         } else {
2858                 /*
2859                  * Fast symlink. We have to add entry to directory
2860                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2861                  * allocate new inode (bitmap, group descriptor, inode block,
2862                  * quota blocks, sb is already counted in previous macros).
2863                  */
2864                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2865                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2866         }
2867 retry:
2868         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2869                                             &dentry->d_name, 0, NULL,
2870                                             EXT4_HT_DIR, credits);
2871         handle = ext4_journal_current_handle();
2872         err = PTR_ERR(inode);
2873         if (IS_ERR(inode))
2874                 goto out_stop;
2875 
2876         if (l > EXT4_N_BLOCKS * 4) {
2877                 inode->i_op = &ext4_symlink_inode_operations;
2878                 ext4_set_aops(inode);
2879                 /*
2880                  * We cannot call page_symlink() with transaction started
2881                  * because it calls into ext4_write_begin() which can wait
2882                  * for transaction commit if we are running out of space
2883                  * and thus we deadlock. So we have to stop transaction now
2884                  * and restart it when symlink contents is written.
2885                  * 
2886                  * To keep fs consistent in case of crash, we have to put inode
2887                  * to orphan list in the mean time.
2888                  */
2889                 drop_nlink(inode);
2890                 err = ext4_orphan_add(handle, inode);
2891                 ext4_journal_stop(handle);
2892                 if (err)
2893                         goto err_drop_inode;
2894                 err = __page_symlink(inode, symname, l, 1);
2895                 if (err)
2896                         goto err_drop_inode;
2897                 /*
2898                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2899                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2900                  */
2901                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2902                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2903                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2904                 if (IS_ERR(handle)) {
2905                         err = PTR_ERR(handle);
2906                         goto err_drop_inode;
2907                 }
2908                 set_nlink(inode, 1);
2909                 err = ext4_orphan_del(handle, inode);
2910                 if (err) {
2911                         ext4_journal_stop(handle);
2912                         clear_nlink(inode);
2913                         goto err_drop_inode;
2914                 }
2915         } else {
2916                 /* clear the extent format for fast symlink */
2917                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2918                 inode->i_op = &ext4_fast_symlink_inode_operations;
2919                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2920                 inode->i_size = l-1;
2921         }
2922         EXT4_I(inode)->i_disksize = inode->i_size;
2923         err = ext4_add_nondir(handle, dentry, inode);
2924         if (!err && IS_DIRSYNC(dir))
2925                 ext4_handle_sync(handle);
2926 
2927 out_stop:
2928         if (handle)
2929                 ext4_journal_stop(handle);
2930         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2931                 goto retry;
2932         return err;
2933 err_drop_inode:
2934         unlock_new_inode(inode);
2935         iput(inode);
2936         return err;
2937 }
2938 
2939 static int ext4_link(struct dentry *old_dentry,
2940                      struct inode *dir, struct dentry *dentry)
2941 {
2942         handle_t *handle;
2943         struct inode *inode = old_dentry->d_inode;
2944         int err, retries = 0;
2945 
2946         if (inode->i_nlink >= EXT4_LINK_MAX)
2947                 return -EMLINK;
2948 
2949         dquot_initialize(dir);
2950 
2951 retry:
2952         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2953                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2954                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2955         if (IS_ERR(handle))
2956                 return PTR_ERR(handle);
2957 
2958         if (IS_DIRSYNC(dir))
2959                 ext4_handle_sync(handle);
2960 
2961         inode->i_ctime = ext4_current_time(inode);
2962         ext4_inc_count(handle, inode);
2963         ihold(inode);
2964 
2965         err = ext4_add_entry(handle, dentry, inode);
2966         if (!err) {
2967                 ext4_mark_inode_dirty(handle, inode);
2968                 /* this can happen only for tmpfile being
2969                  * linked the first time
2970                  */
2971                 if (inode->i_nlink == 1)
2972                         ext4_orphan_del(handle, inode);
2973                 d_instantiate(dentry, inode);
2974         } else {
2975                 drop_nlink(inode);
2976                 iput(inode);
2977         }
2978         ext4_journal_stop(handle);
2979         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2980                 goto retry;
2981         return err;
2982 }
2983 
2984 
2985 /*
2986  * Try to find buffer head where contains the parent block.
2987  * It should be the inode block if it is inlined or the 1st block
2988  * if it is a normal dir.
2989  */
2990 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2991                                         struct inode *inode,
2992                                         int *retval,
2993                                         struct ext4_dir_entry_2 **parent_de,
2994                                         int *inlined)
2995 {
2996         struct buffer_head *bh;
2997 
2998         if (!ext4_has_inline_data(inode)) {
2999                 bh = ext4_read_dirblock(inode, 0, EITHER);
3000                 if (IS_ERR(bh)) {
3001                         *retval = PTR_ERR(bh);
3002                         return NULL;
3003                 }
3004                 *parent_de = ext4_next_entry(
3005                                         (struct ext4_dir_entry_2 *)bh->b_data,
3006                                         inode->i_sb->s_blocksize);
3007                 return bh;
3008         }
3009 
3010         *inlined = 1;
3011         return ext4_get_first_inline_block(inode, parent_de, retval);
3012 }
3013 
3014 struct ext4_renament {
3015         struct inode *dir;
3016         struct dentry *dentry;
3017         struct inode *inode;
3018         bool is_dir;
3019         int dir_nlink_delta;
3020 
3021         /* entry for "dentry" */
3022         struct buffer_head *bh;
3023         struct ext4_dir_entry_2 *de;
3024         int inlined;
3025 
3026         /* entry for ".." in inode if it's a directory */
3027         struct buffer_head *dir_bh;
3028         struct ext4_dir_entry_2 *parent_de;
3029         int dir_inlined;
3030 };
3031 
3032 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3033 {
3034         int retval;
3035 
3036         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3037                                               &retval, &ent->parent_de,
3038                                               &ent->dir_inlined);
3039         if (!ent->dir_bh)
3040                 return retval;
3041         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3042                 return -EIO;
3043         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3044         return ext4_journal_get_write_access(handle, ent->dir_bh);
3045 }
3046 
3047 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3048                                   unsigned dir_ino)
3049 {
3050         int retval;
3051 
3052         ent->parent_de->inode = cpu_to_le32(dir_ino);
3053         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3054         if (!ent->dir_inlined) {
3055                 if (is_dx(ent->inode)) {
3056                         retval = ext4_handle_dirty_dx_node(handle,
3057                                                            ent->inode,
3058                                                            ent->dir_bh);
3059                 } else {
3060                         retval = ext4_handle_dirty_dirent_node(handle,
3061                                                                ent->inode,
3062                                                                ent->dir_bh);
3063                 }
3064         } else {
3065                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3066         }
3067         if (retval) {
3068                 ext4_std_error(ent->dir->i_sb, retval);
3069                 return retval;
3070         }
3071         return 0;
3072 }
3073 
3074 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3075                        unsigned ino, unsigned file_type)
3076 {
3077         int retval;
3078 
3079         BUFFER_TRACE(ent->bh, "get write access");
3080         retval = ext4_journal_get_write_access(handle, ent->bh);
3081         if (retval)
3082                 return retval;
3083         ent->de->inode = cpu_to_le32(ino);
3084         if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3085                                       EXT4_FEATURE_INCOMPAT_FILETYPE))
3086                 ent->de->file_type = file_type;
3087         ent->dir->i_version++;
3088         ent->dir->i_ctime = ent->dir->i_mtime =
3089                 ext4_current_time(ent->dir);
3090         ext4_mark_inode_dirty(handle, ent->dir);
3091         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3092         if (!ent->inlined) {
3093                 retval = ext4_handle_dirty_dirent_node(handle,
3094                                                        ent->dir, ent->bh);
3095                 if (unlikely(retval)) {
3096                         ext4_std_error(ent->dir->i_sb, retval);
3097                         return retval;
3098                 }
3099         }
3100         brelse(ent->bh);
3101         ent->bh = NULL;
3102 
3103         return 0;
3104 }
3105 
3106 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3107                                   const struct qstr *d_name)
3108 {
3109         int retval = -ENOENT;
3110         struct buffer_head *bh;
3111         struct ext4_dir_entry_2 *de;
3112 
3113         bh = ext4_find_entry(dir, d_name, &de, NULL);
3114         if (IS_ERR(bh))
3115                 return PTR_ERR(bh);
3116         if (bh) {
3117                 retval = ext4_delete_entry(handle, dir, de, bh);
3118                 brelse(bh);
3119         }
3120         return retval;
3121 }
3122 
3123 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3124                                int force_reread)
3125 {
3126         int retval;
3127         /*
3128          * ent->de could have moved from under us during htree split, so make
3129          * sure that we are deleting the right entry.  We might also be pointing
3130          * to a stale entry in the unused part of ent->bh so just checking inum
3131          * and the name isn't enough.
3132          */
3133         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3134             ent->de->name_len != ent->dentry->d_name.len ||
3135             strncmp(ent->de->name, ent->dentry->d_name.name,
3136                     ent->de->name_len) ||
3137             force_reread) {
3138                 retval = ext4_find_delete_entry(handle, ent->dir,
3139                                                 &ent->dentry->d_name);
3140         } else {
3141                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3142                 if (retval == -ENOENT) {
3143                         retval = ext4_find_delete_entry(handle, ent->dir,
3144                                                         &ent->dentry->d_name);
3145                 }
3146         }
3147 
3148         if (retval) {
3149                 ext4_warning(ent->dir->i_sb,
3150                                 "Deleting old file (%lu), %d, error=%d",
3151                                 ent->dir->i_ino, ent->dir->i_nlink, retval);
3152         }
3153 }
3154 
3155 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3156 {
3157         if (ent->dir_nlink_delta) {
3158                 if (ent->dir_nlink_delta == -1)
3159                         ext4_dec_count(handle, ent->dir);
3160                 else
3161                         ext4_inc_count(handle, ent->dir);
3162                 ext4_mark_inode_dirty(handle, ent->dir);
3163         }
3164 }
3165 
3166 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3167                                               int credits, handle_t **h)
3168 {
3169         struct inode *wh;
3170         handle_t *handle;
3171         int retries = 0;
3172 
3173         /*
3174          * for inode block, sb block, group summaries,
3175          * and inode bitmap
3176          */
3177         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3178                     EXT4_XATTR_TRANS_BLOCKS + 4);
3179 retry:
3180         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3181                                          &ent->dentry->d_name, 0, NULL,
3182                                          EXT4_HT_DIR, credits);
3183 
3184         handle = ext4_journal_current_handle();
3185         if (IS_ERR(wh)) {
3186                 if (handle)
3187                         ext4_journal_stop(handle);
3188                 if (PTR_ERR(wh) == -ENOSPC &&
3189                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3190                         goto retry;
3191         } else {
3192                 *h = handle;
3193                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3194                 wh->i_op = &ext4_special_inode_operations;
3195         }
3196         return wh;
3197 }
3198 
3199 /*
3200  * Anybody can rename anything with this: the permission checks are left to the
3201  * higher-level routines.
3202  *
3203  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3204  * while new_{dentry,inode) refers to the destination dentry/inode
3205  * This comes from rename(const char *oldpath, const char *newpath)
3206  */
3207 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3208                        struct inode *new_dir, struct dentry *new_dentry,
3209                        unsigned int flags)
3210 {
3211         handle_t *handle = NULL;
3212         struct ext4_renament old = {
3213                 .dir = old_dir,
3214                 .dentry = old_dentry,
3215                 .inode = old_dentry->d_inode,
3216         };
3217         struct ext4_renament new = {
3218                 .dir = new_dir,
3219                 .dentry = new_dentry,
3220                 .inode = new_dentry->d_inode,
3221         };
3222         int force_reread;
3223         int retval;
3224         struct inode *whiteout = NULL;
3225         int credits;
3226         u8 old_file_type;
3227 
3228         dquot_initialize(old.dir);
3229         dquot_initialize(new.dir);
3230 
3231         /* Initialize quotas before so that eventual writes go
3232          * in separate transaction */
3233         if (new.inode)
3234                 dquot_initialize(new.inode);
3235 
3236         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3237         if (IS_ERR(old.bh))
3238                 return PTR_ERR(old.bh);
3239         /*
3240          *  Check for inode number is _not_ due to possible IO errors.
3241          *  We might rmdir the source, keep it as pwd of some process
3242          *  and merrily kill the link to whatever was created under the
3243          *  same name. Goodbye sticky bit ;-<
3244          */
3245         retval = -ENOENT;
3246         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3247                 goto end_rename;
3248 
3249         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3250                                  &new.de, &new.inlined);
3251         if (IS_ERR(new.bh)) {
3252                 retval = PTR_ERR(new.bh);
3253                 new.bh = NULL;
3254                 goto end_rename;
3255         }
3256         if (new.bh) {
3257                 if (!new.inode) {
3258                         brelse(new.bh);
3259                         new.bh = NULL;
3260                 }
3261         }
3262         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3263                 ext4_alloc_da_blocks(old.inode);
3264 
3265         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3266                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3267         if (!(flags & RENAME_WHITEOUT)) {
3268                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3269                 if (IS_ERR(handle))
3270                         return PTR_ERR(handle);
3271         } else {
3272                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3273                 if (IS_ERR(whiteout))
3274                         return PTR_ERR(whiteout);
3275         }
3276 
3277         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3278                 ext4_handle_sync(handle);
3279 
3280         if (S_ISDIR(old.inode->i_mode)) {
3281                 if (new.inode) {
3282                         retval = -ENOTEMPTY;
3283                         if (!empty_dir(new.inode))
3284                                 goto end_rename;
3285                 } else {
3286                         retval = -EMLINK;
3287                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3288                                 goto end_rename;
3289                 }
3290                 retval = ext4_rename_dir_prepare(handle, &old);
3291                 if (retval)
3292                         goto end_rename;
3293         }
3294         /*
3295          * If we're renaming a file within an inline_data dir and adding or
3296          * setting the new dirent causes a conversion from inline_data to
3297          * extents/blockmap, we need to force the dirent delete code to
3298          * re-read the directory, or else we end up trying to delete a dirent
3299          * from what is now the extent tree root (or a block map).
3300          */
3301         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3302                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3303 
3304         old_file_type = old.de->file_type;
3305         if (whiteout) {
3306                 /*
3307                  * Do this before adding a new entry, so the old entry is sure
3308                  * to be still pointing to the valid old entry.
3309                  */
3310                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3311                                      EXT4_FT_CHRDEV);
3312                 if (retval)
3313                         goto end_rename;
3314                 ext4_mark_inode_dirty(handle, whiteout);
3315         }
3316         if (!new.bh) {
3317                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3318                 if (retval)
3319                         goto end_rename;
3320         } else {
3321                 retval = ext4_setent(handle, &new,
3322                                      old.inode->i_ino, old_file_type);
3323                 if (retval)
3324                         goto end_rename;
3325         }
3326         if (force_reread)
3327                 force_reread = !ext4_test_inode_flag(new.dir,
3328                                                      EXT4_INODE_INLINE_DATA);
3329 
3330         /*
3331          * Like most other Unix systems, set the ctime for inodes on a
3332          * rename.
3333          */
3334         old.inode->i_ctime = ext4_current_time(old.inode);
3335         ext4_mark_inode_dirty(handle, old.inode);
3336 
3337         if (!whiteout) {
3338                 /*
3339                  * ok, that's it
3340                  */
3341                 ext4_rename_delete(handle, &old, force_reread);
3342         }
3343 
3344         if (new.inode) {
3345                 ext4_dec_count(handle, new.inode);
3346                 new.inode->i_ctime = ext4_current_time(new.inode);
3347         }
3348         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3349         ext4_update_dx_flag(old.dir);
3350         if (old.dir_bh) {
3351                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3352                 if (retval)
3353                         goto end_rename;
3354 
3355                 ext4_dec_count(handle, old.dir);
3356                 if (new.inode) {
3357                         /* checked empty_dir above, can't have another parent,
3358                          * ext4_dec_count() won't work for many-linked dirs */
3359                         clear_nlink(new.inode);
3360                 } else {
3361                         ext4_inc_count(handle, new.dir);
3362                         ext4_update_dx_flag(new.dir);
3363                         ext4_mark_inode_dirty(handle, new.dir);
3364                 }
3365         }
3366         ext4_mark_inode_dirty(handle, old.dir);
3367         if (new.inode) {
3368                 ext4_mark_inode_dirty(handle, new.inode);
3369                 if (!new.inode->i_nlink)
3370                         ext4_orphan_add(handle, new.inode);
3371         }
3372         retval = 0;
3373 
3374 end_rename:
3375         brelse(old.dir_bh);
3376         brelse(old.bh);
3377         brelse(new.bh);
3378         if (whiteout) {
3379                 if (retval)
3380                         drop_nlink(whiteout);
3381                 unlock_new_inode(whiteout);
3382                 iput(whiteout);
3383         }
3384         if (handle)
3385                 ext4_journal_stop(handle);
3386         return retval;
3387 }
3388 
3389 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3390                              struct inode *new_dir, struct dentry *new_dentry)
3391 {
3392         handle_t *handle = NULL;
3393         struct ext4_renament old = {
3394                 .dir = old_dir,
3395                 .dentry = old_dentry,
3396                 .inode = old_dentry->d_inode,
3397         };
3398         struct ext4_renament new = {
3399                 .dir = new_dir,
3400                 .dentry = new_dentry,
3401                 .inode = new_dentry->d_inode,
3402         };
3403         u8 new_file_type;
3404         int retval;
3405 
3406         dquot_initialize(old.dir);
3407         dquot_initialize(new.dir);
3408 
3409         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3410                                  &old.de, &old.inlined);
3411         if (IS_ERR(old.bh))
3412                 return PTR_ERR(old.bh);
3413         /*
3414          *  Check for inode number is _not_ due to possible IO errors.
3415          *  We might rmdir the source, keep it as pwd of some process
3416          *  and merrily kill the link to whatever was created under the
3417          *  same name. Goodbye sticky bit ;-<
3418          */
3419         retval = -ENOENT;
3420         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3421                 goto end_rename;
3422 
3423         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3424                                  &new.de, &new.inlined);
3425         if (IS_ERR(new.bh)) {
3426                 retval = PTR_ERR(new.bh);
3427                 new.bh = NULL;
3428                 goto end_rename;
3429         }
3430 
3431         /* RENAME_EXCHANGE case: old *and* new must both exist */
3432         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3433                 goto end_rename;
3434 
3435         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3436                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3437                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3438         if (IS_ERR(handle))
3439                 return PTR_ERR(handle);
3440 
3441         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3442                 ext4_handle_sync(handle);
3443 
3444         if (S_ISDIR(old.inode->i_mode)) {
3445                 old.is_dir = true;
3446                 retval = ext4_rename_dir_prepare(handle, &old);
3447                 if (retval)
3448                         goto end_rename;
3449         }
3450         if (S_ISDIR(new.inode->i_mode)) {
3451                 new.is_dir = true;
3452                 retval = ext4_rename_dir_prepare(handle, &new);
3453                 if (retval)
3454                         goto end_rename;
3455         }
3456 
3457         /*
3458          * Other than the special case of overwriting a directory, parents'
3459          * nlink only needs to be modified if this is a cross directory rename.
3460          */
3461         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3462                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3463                 new.dir_nlink_delta = -old.dir_nlink_delta;
3464                 retval = -EMLINK;
3465                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3466                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3467                         goto end_rename;
3468         }
3469 
3470         new_file_type = new.de->file_type;
3471         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3472         if (retval)
3473                 goto end_rename;
3474 
3475         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3476         if (retval)
3477                 goto end_rename;
3478 
3479         /*
3480          * Like most other Unix systems, set the ctime for inodes on a
3481          * rename.
3482          */
3483         old.inode->i_ctime = ext4_current_time(old.inode);
3484         new.inode->i_ctime = ext4_current_time(new.inode);
3485         ext4_mark_inode_dirty(handle, old.inode);
3486         ext4_mark_inode_dirty(handle, new.inode);
3487 
3488         if (old.dir_bh) {
3489                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3490                 if (retval)
3491                         goto end_rename;
3492         }
3493         if (new.dir_bh) {
3494                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3495                 if (retval)
3496                         goto end_rename;
3497         }
3498         ext4_update_dir_count(handle, &old);
3499         ext4_update_dir_count(handle, &new);
3500         retval = 0;
3501 
3502 end_rename:
3503         brelse(old.dir_bh);
3504         brelse(new.dir_bh);
3505         brelse(old.bh);
3506         brelse(new.bh);
3507         if (handle)
3508                 ext4_journal_stop(handle);
3509         return retval;
3510 }
3511 
3512 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3513                         struct inode *new_dir, struct dentry *new_dentry,
3514                         unsigned int flags)
3515 {
3516         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3517                 return -EINVAL;
3518 
3519         if (flags & RENAME_EXCHANGE) {
3520                 return ext4_cross_rename(old_dir, old_dentry,
3521                                          new_dir, new_dentry);
3522         }
3523 
3524         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3525 }
3526 
3527 /*
3528  * directories can handle most operations...
3529  */
3530 const struct inode_operations ext4_dir_inode_operations = {
3531         .create         = ext4_create,
3532         .lookup         = ext4_lookup,
3533         .link           = ext4_link,
3534         .unlink         = ext4_unlink,
3535         .symlink        = ext4_symlink,
3536         .mkdir          = ext4_mkdir,
3537         .rmdir          = ext4_rmdir,
3538         .mknod          = ext4_mknod,
3539         .tmpfile        = ext4_tmpfile,
3540         .rename2        = ext4_rename2,
3541         .setattr        = ext4_setattr,
3542         .setxattr       = generic_setxattr,
3543         .getxattr       = generic_getxattr,
3544         .listxattr      = ext4_listxattr,
3545         .removexattr    = generic_removexattr,
3546         .get_acl        = ext4_get_acl,
3547         .set_acl        = ext4_set_acl,
3548         .fiemap         = ext4_fiemap,
3549 };
3550 
3551 const struct inode_operations ext4_special_inode_operations = {
3552         .setattr        = ext4_setattr,
3553         .setxattr       = generic_setxattr,
3554         .getxattr       = generic_getxattr,
3555         .listxattr      = ext4_listxattr,
3556         .removexattr    = generic_removexattr,
3557         .get_acl        = ext4_get_acl,
3558         .set_acl        = ext4_set_acl,
3559 };
3560 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp