fs/ext4/namei.c

   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/ext4_fs.h>
  32 #include <linux/ext4_jbd2.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/stat.h>
  35 #include <linux/string.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/buffer_head.h>
  38 #include <linux/bio.h>
  39
  40 #include "namei.h"
  41 #include "xattr.h"
  42 #include "acl.h"
  43
  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 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52 static struct buffer_head *ext4_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         u32 *block, int *err)
  55 {
  56         struct buffer_head *bh;
  57
  58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  59
  60         if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
  61                 inode->i_size += inode->i_sb->s_blocksize;
  62                 EXT4_I(inode)->i_disksize = inode->i_size;
  63                 ext4_journal_get_write_access(handle,bh);
  64         }
  65         return bh;
  66 }
  67
  68 #ifndef assert
  69 #define assert(test) J_ASSERT(test)
  70 #endif
  71
  72 #ifndef swap
  73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  74 #endif
  75
  76 #ifdef DX_DEBUG
  77 #define dxtrace(command) command
  78 #else
  79 #define dxtrace(command)
  80 #endif
  81
  82 struct fake_dirent
  83 {
  84         __le32 inode;
  85         __le16 rec_len;
  86         u8 name_len;
  87         u8 file_type;
  88 };
  89
  90 struct dx_countlimit
  91 {
  92         __le16 limit;
  93         __le16 count;
  94 };
  95
  96 struct dx_entry
  97 {
  98         __le32 hash;
  99         __le32 block;
 100 };
 101
 102 /*
 103  * dx_root_info is laid out so that if it should somehow get overlaid by a
 104  * dirent the two low bits of the hash version will be zero.  Therefore, the
 105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 106  */
 107
 108 struct dx_root
 109 {
 110         struct fake_dirent dot;
 111         char dot_name[4];
 112         struct fake_dirent dotdot;
 113         char dotdot_name[4];
 114         struct dx_root_info
 115         {
 116                 __le32 reserved_zero;
 117                 u8 hash_version;
 118                 u8 info_length; /* 8 */
 119                 u8 indirect_levels;
 120                 u8 unused_flags;
 121         }
 122         info;
 123         struct dx_entry entries[0];
 124 };
 125
 126 struct dx_node
 127 {
 128         struct fake_dirent fake;
 129         struct dx_entry entries[0];
 130 };
 131
 132
 133 struct dx_frame
 134 {
 135         struct buffer_head *bh;
 136         struct dx_entry *entries;
 137         struct dx_entry *at;
 138 };
 139
 140 struct dx_map_entry
 141 {
 142         u32 hash;
 143         u16 offs;
 144         u16 size;
 145 };
 146
 147 #ifdef CONFIG_EXT4_INDEX
 148 static inline unsigned dx_get_block (struct dx_entry *entry);
 149 static void dx_set_block (struct dx_entry *entry, unsigned value);
 150 static inline unsigned dx_get_hash (struct dx_entry *entry);
 151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
 152 static unsigned dx_get_count (struct dx_entry *entries);
 153 static unsigned dx_get_limit (struct dx_entry *entries);
 154 static void dx_set_count (struct dx_entry *entries, unsigned value);
 155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
 156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 157 static unsigned dx_node_limit (struct inode *dir);
 158 static struct dx_frame *dx_probe(struct dentry *dentry,
 159                                  struct inode *dir,
 160                                  struct dx_hash_info *hinfo,
 161                                  struct dx_frame *frame,
 162                                  int *err);
 163 static void dx_release (struct dx_frame *frames);
 164 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
 165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 167 static struct ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
 168                 struct dx_map_entry *offsets, int count);
 169 static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
 170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 172                                  struct dx_frame *frame,
 173                                  struct dx_frame *frames,
 174                                  __u32 *start_hash);
 175 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
 176                        struct ext4_dir_entry_2 **res_dir, int *err);
 177 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 178                              struct inode *inode);
 179
 180 /*
 181  * Future: use high four bits of block for coalesce-on-delete flags
 182  * Mask them off for now.
 183  */
 184
 185 static inline unsigned dx_get_block (struct dx_entry *entry)
 186 {
 187         return le32_to_cpu(entry->block) & 0x00ffffff;
 188 }
 189
 190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 191 {
 192         entry->block = cpu_to_le32(value);
 193 }
 194
 195 static inline unsigned dx_get_hash (struct dx_entry *entry)
 196 {
 197         return le32_to_cpu(entry->hash);
 198 }
 199
 200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 201 {
 202         entry->hash = cpu_to_le32(value);
 203 }
 204
 205 static inline unsigned dx_get_count (struct dx_entry *entries)
 206 {
 207         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 208 }
 209
 210 static inline unsigned dx_get_limit (struct dx_entry *entries)
 211 {
 212         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 213 }
 214
 215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 216 {
 217         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 218 }
 219
 220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 221 {
 222         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 223 }
 224
 225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 226 {
 227         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 228                 EXT4_DIR_REC_LEN(2) - infosize;
 229         return 0? 20: entry_space / sizeof(struct dx_entry);
 230 }
 231
 232 static inline unsigned dx_node_limit (struct inode *dir)
 233 {
 234         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 235         return 0? 22: entry_space / sizeof(struct dx_entry);
 236 }
 237
 238 /*
 239  * Debug
 240  */
 241 #ifdef DX_DEBUG
 242 static void dx_show_index (char * label, struct dx_entry *entries)
 243 {
 244         int i, n = dx_get_count (entries);
 245         printk("%s index ", label);
 246         for (i = 0; i < n; i++) {
 247                 printk("%x->%u ", i? dx_get_hash(entries + i) :
 248                                 0, dx_get_block(entries + i));
 249         }
 250         printk("\n");
 251 }
 252
 253 struct stats
 254 {
 255         unsigned names;
 256         unsigned space;
 257         unsigned bcount;
 258 };
 259
 260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 261                                  int size, int show_names)
 262 {
 263         unsigned names = 0, space = 0;
 264         char *base = (char *) de;
 265         struct dx_hash_info h = *hinfo;
 266
 267         printk("names: ");
 268         while ((char *) de < base + size)
 269         {
 270                 if (de->inode)
 271                 {
 272                         if (show_names)
 273                         {
 274                                 int len = de->name_len;
 275                                 char *name = de->name;
 276                                 while (len--) printk("%c", *name++);
 277                                 ext4fs_dirhash(de->name, de->name_len, &h);
 278                                 printk(":%x.%u ", h.hash,
 279                                        ((char *) de - base));
 280                         }
 281                         space += EXT4_DIR_REC_LEN(de->name_len);
 282                         names++;
 283                 }
 284                 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 285         }
 286         printk("(%i)\n", names);
 287         return (struct stats) { names, space, 1 };
 288 }
 289
 290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 291                              struct dx_entry *entries, int levels)
 292 {
 293         unsigned blocksize = dir->i_sb->s_blocksize;
 294         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 295         unsigned bcount = 0;
 296         struct buffer_head *bh;
 297         int err;
 298         printk("%i indexed blocks...\n", count);
 299         for (i = 0; i < count; i++, entries++)
 300         {
 301                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 302                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 303                 struct stats stats;
 304                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 305                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 306                 stats = levels?
 307                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 308                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 309                 names += stats.names;
 310                 space += stats.space;
 311                 bcount += stats.bcount;
 312                 brelse (bh);
 313         }
 314         if (bcount)
 315                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 316                         names, space/bcount,(space/bcount)*100/blocksize);
 317         return (struct stats) { names, space, bcount};
 318 }
 319 #endif /* DX_DEBUG */
 320
 321 /*
 322  * Probe for a directory leaf block to search.
 323  *
 324  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 325  * error in the directory index, and the caller should fall back to
 326  * searching the directory normally.  The callers of dx_probe **MUST**
 327  * check for this error code, and make sure it never gets reflected
 328  * back to userspace.
 329  */
 330 static struct dx_frame *
 331 dx_probe(struct dentry *dentry, struct inode *dir,
 332          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 333 {
 334         unsigned count, indirect;
 335         struct dx_entry *at, *entries, *p, *q, *m;
 336         struct dx_root *root;
 337         struct buffer_head *bh;
 338         struct dx_frame *frame = frame_in;
 339         u32 hash;
 340
 341         frame->bh = NULL;
 342         if (dentry)
 343                 dir = dentry->d_parent->d_inode;
 344         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 345                 goto fail;
 346         root = (struct dx_root *) bh->b_data;
 347         if (root->info.hash_version != DX_HASH_TEA &&
 348             root->info.hash_version != DX_HASH_HALF_MD4 &&
 349             root->info.hash_version != DX_HASH_LEGACY) {
 350                 ext4_warning(dir->i_sb, __FUNCTION__,
 351                              "Unrecognised inode hash code %d",
 352                              root->info.hash_version);
 353                 brelse(bh);
 354                 *err = ERR_BAD_DX_DIR;
 355                 goto fail;
 356         }
 357         hinfo->hash_version = root->info.hash_version;
 358         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 359         if (dentry)
 360                 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
 361         hash = hinfo->hash;
 362
 363         if (root->info.unused_flags & 1) {
 364                 ext4_warning(dir->i_sb, __FUNCTION__,
 365                              "Unimplemented inode hash flags: %#06x",
 366                              root->info.unused_flags);
 367                 brelse(bh);
 368                 *err = ERR_BAD_DX_DIR;
 369                 goto fail;
 370         }
 371
 372         if ((indirect = root->info.indirect_levels) > 1) {
 373                 ext4_warning(dir->i_sb, __FUNCTION__,
 374                              "Unimplemented inode hash depth: %#06x",
 375                              root->info.indirect_levels);
 376                 brelse(bh);
 377                 *err = ERR_BAD_DX_DIR;
 378                 goto fail;
 379         }
 380
 381         entries = (struct dx_entry *) (((char *)&root->info) +
 382                                        root->info.info_length);
 383
 384         if (dx_get_limit(entries) != dx_root_limit(dir,
 385                                                    root->info.info_length)) {
 386                 ext4_warning(dir->i_sb, __FUNCTION__,
 387                              "dx entry: limit != root limit");
 388                 brelse(bh);
 389                 *err = ERR_BAD_DX_DIR;
 390                 goto fail;
 391         }
 392
 393         dxtrace (printk("Look up %x", hash));
 394         while (1)
 395         {
 396                 count = dx_get_count(entries);
 397                 if (!count || count > dx_get_limit(entries)) {
 398                         ext4_warning(dir->i_sb, __FUNCTION__,
 399                                      "dx entry: no count or count > limit");
 400                         brelse(bh);
 401                         *err = ERR_BAD_DX_DIR;
 402                         goto fail2;
 403                 }
 404
 405                 p = entries + 1;
 406                 q = entries + count - 1;
 407                 while (p <= q)
 408                 {
 409                         m = p + (q - p)/2;
 410                         dxtrace(printk("."));
 411                         if (dx_get_hash(m) > hash)
 412                                 q = m - 1;
 413                         else
 414                                 p = m + 1;
 415                 }
 416
 417                 if (0) // linear search cross check
 418                 {
 419                         unsigned n = count - 1;
 420                         at = entries;
 421                         while (n--)
 422                         {
 423                                 dxtrace(printk(","));
 424                                 if (dx_get_hash(++at) > hash)
 425                                 {
 426                                         at--;
 427                                         break;
 428                                 }
 429                         }
 430                         assert (at == p - 1);
 431                 }
 432
 433                 at = p - 1;
 434                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 435                 frame->bh = bh;
 436                 frame->entries = entries;
 437                 frame->at = at;
 438                 if (!indirect--) return frame;
 439                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 440                         goto fail2;
 441                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 442                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 443                         ext4_warning(dir->i_sb, __FUNCTION__,
 444                                      "dx entry: limit != node limit");
 445                         brelse(bh);
 446                         *err = ERR_BAD_DX_DIR;
 447                         goto fail2;
 448                 }
 449                 frame++;
 450                 frame->bh = NULL;
 451         }
 452 fail2:
 453         while (frame >= frame_in) {
 454                 brelse(frame->bh);
 455                 frame--;
 456         }
 457 fail:
 458         if (*err == ERR_BAD_DX_DIR)
 459                 ext4_warning(dir->i_sb, __FUNCTION__,
 460                              "Corrupt dir inode %ld, running e2fsck is "
 461                              "recommended.", dir->i_ino);
 462         return NULL;
 463 }
 464
 465 static void dx_release (struct dx_frame *frames)
 466 {
 467         if (frames[0].bh == NULL)
 468                 return;
 469
 470         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 471                 brelse(frames[1].bh);
 472         brelse(frames[0].bh);
 473 }
 474
 475 /*
 476  * This function increments the frame pointer to search the next leaf
 477  * block, and reads in the necessary intervening nodes if the search
 478  * should be necessary.  Whether or not the search is necessary is
 479  * controlled by the hash parameter.  If the hash value is even, then
 480  * the search is only continued if the next block starts with that
 481  * hash value.  This is used if we are searching for a specific file.
 482  *
 483  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 484  *
 485  * This function returns 1 if the caller should continue to search,
 486  * or 0 if it should not.  If there is an error reading one of the
 487  * index blocks, it will a negative error code.
 488  *
 489  * If start_hash is non-null, it will be filled in with the starting
 490  * hash of the next page.
 491  */
 492 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 493                                  struct dx_frame *frame,
 494                                  struct dx_frame *frames,
 495                                  __u32 *start_hash)
 496 {
 497         struct dx_frame *p;
 498         struct buffer_head *bh;
 499         int err, num_frames = 0;
 500         __u32 bhash;
 501
 502         p = frame;
 503         /*
 504          * Find the next leaf page by incrementing the frame pointer.
 505          * If we run out of entries in the interior node, loop around and
 506          * increment pointer in the parent node.  When we break out of
 507          * this loop, num_frames indicates the number of interior
 508          * nodes need to be read.
 509          */
 510         while (1) {
 511                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 512                         break;
 513                 if (p == frames)
 514                         return 0;
 515                 num_frames++;
 516                 p--;
 517         }
 518
 519         /*
 520          * If the hash is 1, then continue only if the next page has a
 521          * continuation hash of any value.  This is used for readdir
 522          * handling.  Otherwise, check to see if the hash matches the
 523          * desired contiuation hash.  If it doesn't, return since
 524          * there's no point to read in the successive index pages.
 525          */
 526         bhash = dx_get_hash(p->at);
 527         if (start_hash)
 528                 *start_hash = bhash;
 529         if ((hash & 1) == 0) {
 530                 if ((bhash & ~1) != hash)
 531                         return 0;
 532         }
 533         /*
 534          * If the hash is HASH_NB_ALWAYS, we always go to the next
 535          * block so no check is necessary
 536          */
 537         while (num_frames--) {
 538                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 539                                       0, &err)))
 540                         return err; /* Failure */
 541                 p++;
 542                 brelse (p->bh);
 543                 p->bh = bh;
 544                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 545         }
 546         return 1;
 547 }
 548
 549
 550 /*
 551  * p is at least 6 bytes before the end of page
 552  */
 553 static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
 554 {
 555         return (struct ext4_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
 556 }
 557
 558 /*
 559  * This function fills a red-black tree with information from a
 560  * directory block.  It returns the number directory entries loaded
 561  * into the tree.  If there is an error it is returned in err.
 562  */
 563 static int htree_dirblock_to_tree(struct file *dir_file,
 564                                   struct inode *dir, int block,
 565                                   struct dx_hash_info *hinfo,
 566                                   __u32 start_hash, __u32 start_minor_hash)
 567 {
 568         struct buffer_head *bh;
 569         struct ext4_dir_entry_2 *de, *top;
 570         int err, count = 0;
 571
 572         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
 573         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 574                 return err;
 575
 576         de = (struct ext4_dir_entry_2 *) bh->b_data;
 577         top = (struct ext4_dir_entry_2 *) ((char *) de +
 578                                            dir->i_sb->s_blocksize -
 579                                            EXT4_DIR_REC_LEN(0));
 580         for (; de < top; de = ext4_next_entry(de)) {
 581                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 582                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 583                                                 +((char *)de - bh->b_data))) {
 584                         /* On error, skip the f_pos to the next block. */
 585                         dir_file->f_pos = (dir_file->f_pos |
 586                                         (dir->i_sb->s_blocksize - 1)) + 1;
 587                         brelse (bh);
 588                         return count;
 589                 }
 590                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 591                 if ((hinfo->hash < start_hash) ||
 592                     ((hinfo->hash == start_hash) &&
 593                      (hinfo->minor_hash < start_minor_hash)))
 594                         continue;
 595                 if (de->inode == 0)
 596                         continue;
 597                 if ((err = ext4_htree_store_dirent(dir_file,
 598                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 599                         brelse(bh);
 600                         return err;
 601                 }
 602                 count++;
 603         }
 604         brelse(bh);
 605         return count;
 606 }
 607
 608
 609 /*
 610  * This function fills a red-black tree with information from a
 611  * directory.  We start scanning the directory in hash order, starting
 612  * at start_hash and start_minor_hash.
 613  *
 614  * This function returns the number of entries inserted into the tree,
 615  * or a negative error code.
 616  */
 617 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 618                          __u32 start_minor_hash, __u32 *next_hash)
 619 {
 620         struct dx_hash_info hinfo;
 621         struct ext4_dir_entry_2 *de;
 622         struct dx_frame frames[2], *frame;
 623         struct inode *dir;
 624         int block, err;
 625         int count = 0;
 626         int ret;
 627         __u32 hashval;
 628
 629         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
 630                        start_minor_hash));
 631         dir = dir_file->f_path.dentry->d_inode;
 632         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
 633                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 634                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 635                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 636                                                start_hash, start_minor_hash);
 637                 *next_hash = ~0;
 638                 return count;
 639         }
 640         hinfo.hash = start_hash;
 641         hinfo.minor_hash = 0;
 642         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
 643         if (!frame)
 644                 return err;
 645
 646         /* Add '.' and '..' from the htree header */
 647         if (!start_hash && !start_minor_hash) {
 648                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 649                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 650                         goto errout;
 651                 count++;
 652         }
 653         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 654                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 655                 de = ext4_next_entry(de);
 656                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 657                         goto errout;
 658                 count++;
 659         }
 660
 661         while (1) {
 662                 block = dx_get_block(frame->at);
 663                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 664                                              start_hash, start_minor_hash);
 665                 if (ret < 0) {
 666                         err = ret;
 667                         goto errout;
 668                 }
 669                 count += ret;
 670                 hashval = ~0;
 671                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 672                                             frame, frames, &hashval);
 673                 *next_hash = hashval;
 674                 if (ret < 0) {
 675                         err = ret;
 676                         goto errout;
 677                 }
 678                 /*
 679                  * Stop if:  (a) there are no more entries, or
 680                  * (b) we have inserted at least one entry and the
 681                  * next hash value is not a continuation
 682                  */
 683                 if ((ret == 0) ||
 684                     (count && ((hashval & 1) == 0)))
 685                         break;
 686         }
 687         dx_release(frames);
 688         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 689                        count, *next_hash));
 690         return count;
 691 errout:
 692         dx_release(frames);
 693         return (err);
 694 }
 695
 696
 697 /*
 698  * Directory block splitting, compacting
 699  */
 700
 701 /*
 702  * Create map of hash values, offsets, and sizes, stored at end of block.
 703  * Returns number of entries mapped.
 704  */
 705 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
 706                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 707 {
 708         int count = 0;
 709         char *base = (char *) de;
 710         struct dx_hash_info h = *hinfo;
 711
 712         while ((char *) de < base + size)
 713         {
 714                 if (de->name_len && de->inode) {
 715                         ext4fs_dirhash(de->name, de->name_len, &h);
 716                         map_tail--;
 717                         map_tail->hash = h.hash;
 718                         map_tail->offs = (u16) ((char *) de - base);
 719                         map_tail->size = le16_to_cpu(de->rec_len);
 720                         count++;
 721                         cond_resched();
 722                 }
 723                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 724                 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 725         }
 726         return count;
 727 }
 728
 729 /* Sort map by hash value */
 730 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 731 {
 732         struct dx_map_entry *p, *q, *top = map + count - 1;
 733         int more;
 734         /* Combsort until bubble sort doesn't suck */
 735         while (count > 2) {
 736                 count = count*10/13;
 737                 if (count - 9 < 2) /* 9, 10 -> 11 */
 738                         count = 11;
 739                 for (p = top, q = p - count; q >= map; p--, q--)
 740                         if (p->hash < q->hash)
 741                                 swap(*p, *q);
 742         }
 743         /* Garden variety bubble sort */
 744         do {
 745                 more = 0;
 746                 q = top;
 747                 while (q-- > map) {
 748                         if (q[1].hash >= q[0].hash)
 749                                 continue;
 750                         swap(*(q+1), *q);
 751                         more = 1;
 752                 }
 753         } while(more);
 754 }
 755
 756 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 757 {
 758         struct dx_entry *entries = frame->entries;
 759         struct dx_entry *old = frame->at, *new = old + 1;
 760         int count = dx_get_count(entries);
 761
 762         assert(count < dx_get_limit(entries));
 763         assert(old < entries + count);
 764         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 765         dx_set_hash(new, hash);
 766         dx_set_block(new, block);
 767         dx_set_count(entries, count + 1);
 768 }
 769 #endif
 770
 771
 772 static void ext4_update_dx_flag(struct inode *inode)
 773 {
 774         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 775                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 776                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
 777 }
 778
 779 /*
 780  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 781  *
 782  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 783  * `de != NULL' is guaranteed by caller.
 784  */
 785 static inline int ext4_match (int len, const char * const name,
 786                               struct ext4_dir_entry_2 * de)
 787 {
 788         if (len != de->name_len)
 789                 return 0;
 790         if (!de->inode)
 791                 return 0;
 792         return !memcmp(name, de->name, len);
 793 }
 794
 795 /*
 796  * Returns 0 if not found, -1 on failure, and 1 on success
 797  */
 798 static inline int search_dirblock(struct buffer_head * bh,
 799                                   struct inode *dir,
 800                                   struct dentry *dentry,
 801                                   unsigned long offset,
 802                                   struct ext4_dir_entry_2 ** res_dir)
 803 {
 804         struct ext4_dir_entry_2 * de;
 805         char * dlimit;
 806         int de_len;
 807         const char *name = dentry->d_name.name;
 808         int namelen = dentry->d_name.len;
 809
 810         de = (struct ext4_dir_entry_2 *) bh->b_data;
 811         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 812         while ((char *) de < dlimit) {
 813                 /* this code is executed quadratically often */
 814                 /* do minimal checking `by hand' */
 815
 816                 if ((char *) de + namelen <= dlimit &&
 817                     ext4_match (namelen, name, de)) {
 818                         /* found a match - just to be sure, do a full check */
 819                         if (!ext4_check_dir_entry("ext4_find_entry",
 820                                                   dir, de, bh, offset))
 821                                 return -1;
 822                         *res_dir = de;
 823                         return 1;
 824                 }
 825                 /* prevent looping on a bad block */
 826                 de_len = le16_to_cpu(de->rec_len);
 827                 if (de_len <= 0)
 828                         return -1;
 829                 offset += de_len;
 830                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 831         }
 832         return 0;
 833 }
 834
 835
 836 /*
 837  *      ext4_find_entry()
 838  *
 839  * finds an entry in the specified directory with the wanted name. It
 840  * returns the cache buffer in which the entry was found, and the entry
 841  * itself (as a parameter - res_dir). It does NOT read the inode of the
 842  * entry - you'll have to do that yourself if you want to.
 843  *
 844  * The returned buffer_head has ->b_count elevated.  The caller is expected
 845  * to brelse() it when appropriate.
 846  */
 847 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
 848                                         struct ext4_dir_entry_2 ** res_dir)
 849 {
 850         struct super_block * sb;
 851         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 852         struct buffer_head * bh, *ret = NULL;
 853         unsigned long start, block, b;
 854         int ra_max = 0;         /* Number of bh's in the readahead
 855                                    buffer, bh_use[] */
 856         int ra_ptr = 0;         /* Current index into readahead
 857                                    buffer */
 858         int num = 0;
 859         int nblocks, i, err;
 860         struct inode *dir = dentry->d_parent->d_inode;
 861         int namelen;
 862         const u8 *name;
 863         unsigned blocksize;
 864
 865         *res_dir = NULL;
 866         sb = dir->i_sb;
 867         blocksize = sb->s_blocksize;
 868         namelen = dentry->d_name.len;
 869         name = dentry->d_name.name;
 870         if (namelen > EXT4_NAME_LEN)
 871                 return NULL;
 872 #ifdef CONFIG_EXT4_INDEX
 873         if (is_dx(dir)) {
 874                 bh = ext4_dx_find_entry(dentry, res_dir, &err);
 875                 /*
 876                  * On success, or if the error was file not found,
 877                  * return.  Otherwise, fall back to doing a search the
 878                  * old fashioned way.
 879                  */
 880                 if (bh || (err != ERR_BAD_DX_DIR))
 881                         return bh;
 882                 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
 883         }
 884 #endif
 885         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 886         start = EXT4_I(dir)->i_dir_start_lookup;
 887         if (start >= nblocks)
 888                 start = 0;
 889         block = start;
 890 restart:
 891         do {
 892                 /*
 893                  * We deal with the read-ahead logic here.
 894                  */
 895                 if (ra_ptr >= ra_max) {
 896                         /* Refill the readahead buffer */
 897                         ra_ptr = 0;
 898                         b = block;
 899                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 900                                 /*
 901                                  * Terminate if we reach the end of the
 902                                  * directory and must wrap, or if our
 903                                  * search has finished at this block.
 904                                  */
 905                                 if (b >= nblocks || (num && block == start)) {
 906                                         bh_use[ra_max] = NULL;
 907                                         break;
 908                                 }
 909                                 num++;
 910                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
 911                                 bh_use[ra_max] = bh;
 912                                 if (bh)
 913                                         ll_rw_block(READ_META, 1, &bh);
 914                         }
 915                 }
 916                 if ((bh = bh_use[ra_ptr++]) == NULL)
 917                         goto next;
 918                 wait_on_buffer(bh);
 919                 if (!buffer_uptodate(bh)) {
 920                         /* read error, skip block & hope for the best */
 921                         ext4_error(sb, __FUNCTION__, "reading directory #%lu "
 922                                    "offset %lu", dir->i_ino, block);
 923                         brelse(bh);
 924                         goto next;
 925                 }
 926                 i = search_dirblock(bh, dir, dentry,
 927                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
 928                 if (i == 1) {
 929                         EXT4_I(dir)->i_dir_start_lookup = block;
 930                         ret = bh;
 931                         goto cleanup_and_exit;
 932                 } else {
 933                         brelse(bh);
 934                         if (i < 0)
 935                                 goto cleanup_and_exit;
 936                 }
 937         next:
 938                 if (++block >= nblocks)
 939                         block = 0;
 940         } while (block != start);
 941
 942         /*
 943          * If the directory has grown while we were searching, then
 944          * search the last part of the directory before giving up.
 945          */
 946         block = nblocks;
 947         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 948         if (block < nblocks) {
 949                 start = 0;
 950                 goto restart;
 951         }
 952
 953 cleanup_and_exit:
 954         /* Clean up the read-ahead blocks */
 955         for (; ra_ptr < ra_max; ra_ptr++)
 956                 brelse (bh_use[ra_ptr]);
 957         return ret;
 958 }
 959
 960 #ifdef CONFIG_EXT4_INDEX
 961 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
 962                        struct ext4_dir_entry_2 **res_dir, int *err)
 963 {
 964         struct super_block * sb;
 965         struct dx_hash_info     hinfo;
 966         u32 hash;
 967         struct dx_frame frames[2], *frame;
 968         struct ext4_dir_entry_2 *de, *top;
 969         struct buffer_head *bh;
 970         unsigned long block;
 971         int retval;
 972         int namelen = dentry->d_name.len;
 973         const u8 *name = dentry->d_name.name;
 974         struct inode *dir = dentry->d_parent->d_inode;
 975
 976         sb = dir->i_sb;
 977         /* NFS may look up ".." - look at dx_root directory block */
 978         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
 979                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
 980                         return NULL;
 981         } else {
 982                 frame = frames;
 983                 frame->bh = NULL;                       /* for dx_release() */
 984                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
 985                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
 986         }
 987         hash = hinfo.hash;
 988         do {
 989                 block = dx_get_block(frame->at);
 990                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
 991                         goto errout;
 992                 de = (struct ext4_dir_entry_2 *) bh->b_data;
 993                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
 994                                        EXT4_DIR_REC_LEN(0));
 995                 for (; de < top; de = ext4_next_entry(de))
 996                 if (ext4_match (namelen, name, de)) {
 997                         if (!ext4_check_dir_entry("ext4_find_entry",
 998                                                   dir, de, bh,
 999                                   (block<<EXT4_BLOCK_SIZE_BITS(sb))
1000                                           +((char *)de - bh->b_data))) {
1001                                 brelse (bh);
1002                                 *err = ERR_BAD_DX_DIR;
1003                                 goto errout;
1004                         }
1005                         *res_dir = de;
1006                         dx_release (frames);
1007                         return bh;
1008                 }
1009                 brelse (bh);
1010                 /* Check to see if we should continue to search */
1011                 retval = ext4_htree_next_block(dir, hash, frame,
1012                                                frames, NULL);
1013                 if (retval < 0) {
1014                         ext4_warning(sb, __FUNCTION__,
1015                              "error reading index page in directory #%lu",
1016                              dir->i_ino);
1017                         *err = retval;
1018                         goto errout;
1019                 }
1020         } while (retval == 1);
1021
1022         *err = -ENOENT;
1023 errout:
1024         dxtrace(printk("%s not found\n", name));
1025         dx_release (frames);
1026         return NULL;
1027 }
1028 #endif
1029
1030 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1031 {
1032         struct inode * inode;
1033         struct ext4_dir_entry_2 * de;
1034         struct buffer_head * bh;
1035
1036         if (dentry->d_name.len > EXT4_NAME_LEN)
1037                 return ERR_PTR(-ENAMETOOLONG);
1038
1039         bh = ext4_find_entry(dentry, &de);
1040         inode = NULL;
1041         if (bh) {
1042                 unsigned long ino = le32_to_cpu(de->inode);
1043                 brelse (bh);
1044                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1045                         ext4_error(dir->i_sb, "ext4_lookup",
1046                                    "bad inode number: %lu", ino);
1047                         inode = NULL;
1048                 } else
1049                         inode = iget(dir->i_sb, ino);
1050
1051                 if (!inode)
1052                         return ERR_PTR(-EACCES);
1053
1054                 if (is_bad_inode(inode)) {
1055                         iput(inode);
1056                         return ERR_PTR(-ENOENT);
1057                 }
1058         }
1059         return d_splice_alias(inode, dentry);
1060 }
1061
1062
1063 struct dentry *ext4_get_parent(struct dentry *child)
1064 {
1065         unsigned long ino;
1066         struct dentry *parent;
1067         struct inode *inode;
1068         struct dentry dotdot;
1069         struct ext4_dir_entry_2 * de;
1070         struct buffer_head *bh;
1071
1072         dotdot.d_name.name = "..";
1073         dotdot.d_name.len = 2;
1074         dotdot.d_parent = child; /* confusing, isn't it! */
1075
1076         bh = ext4_find_entry(&dotdot, &de);
1077         inode = NULL;
1078         if (!bh)
1079                 return ERR_PTR(-ENOENT);
1080         ino = le32_to_cpu(de->inode);
1081         brelse(bh);
1082
1083         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1084                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1085                            "bad inode number: %lu", ino);
1086                 inode = NULL;
1087         } else
1088                 inode = iget(child->d_inode->i_sb, ino);
1089
1090         if (!inode)
1091                 return ERR_PTR(-EACCES);
1092
1093         if (is_bad_inode(inode)) {
1094                 iput(inode);
1095                 return ERR_PTR(-ENOENT);
1096         }
1097
1098         parent = d_alloc_anon(inode);
1099         if (!parent) {
1100                 iput(inode);
1101                 parent = ERR_PTR(-ENOMEM);
1102         }
1103         return parent;
1104 }
1105
1106 #define S_SHIFT 12
1107 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1108         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1109         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1110         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1111         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1112         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1113         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1114         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1115 };
1116
1117 static inline void ext4_set_de_type(struct super_block *sb,
1118                                 struct ext4_dir_entry_2 *de,
1119                                 umode_t mode) {
1120         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1121                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1122 }
1123
1124 #ifdef CONFIG_EXT4_INDEX
1125 /*
1126  * Move count entries from end of map between two memory locations.
1127  * Returns pointer to last entry moved.
1128  */
1129 static struct ext4_dir_entry_2 *
1130 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1131 {
1132         unsigned rec_len = 0;
1133
1134         while (count--) {
1135                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1136                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1137                 memcpy (to, de, rec_len);
1138                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1139                                 cpu_to_le16(rec_len);
1140                 de->inode = 0;
1141                 map++;
1142                 to += rec_len;
1143         }
1144         return (struct ext4_dir_entry_2 *) (to - rec_len);
1145 }
1146
1147 /*
1148  * Compact each dir entry in the range to the minimal rec_len.
1149  * Returns pointer to last entry in range.
1150  */
1151 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1152 {
1153         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1154         unsigned rec_len = 0;
1155
1156         prev = to = de;
1157         while ((char*)de < base + size) {
1158                 next = (struct ext4_dir_entry_2 *) ((char *) de +
1159                                                     le16_to_cpu(de->rec_len));
1160                 if (de->inode && de->name_len) {
1161                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1162                         if (de > to)
1163                                 memmove(to, de, rec_len);
1164                         to->rec_len = cpu_to_le16(rec_len);
1165                         prev = to;
1166                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1167                 }
1168                 de = next;
1169         }
1170         return prev;
1171 }
1172
1173 /*
1174  * Split a full leaf block to make room for a new dir entry.
1175  * Allocate a new block, and move entries so that they are approx. equally full.
1176  * Returns pointer to de in block into which the new entry will be inserted.
1177  */
1178 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1179                         struct buffer_head **bh,struct dx_frame *frame,
1180                         struct dx_hash_info *hinfo, int *error)
1181 {
1182         unsigned blocksize = dir->i_sb->s_blocksize;
1183         unsigned count, continued;
1184         struct buffer_head *bh2;
1185         u32 newblock;
1186         u32 hash2;
1187         struct dx_map_entry *map;
1188         char *data1 = (*bh)->b_data, *data2;
1189         unsigned split, move, size, i;
1190         struct ext4_dir_entry_2 *de = NULL, *de2;
1191         int     err = 0;
1192
1193         bh2 = ext4_append (handle, dir, &newblock, &err);
1194         if (!(bh2)) {
1195                 brelse(*bh);
1196                 *bh = NULL;
1197                 goto errout;
1198         }
1199
1200         BUFFER_TRACE(*bh, "get_write_access");
1201         err = ext4_journal_get_write_access(handle, *bh);
1202         if (err)
1203                 goto journal_error;
1204
1205         BUFFER_TRACE(frame->bh, "get_write_access");
1206         err = ext4_journal_get_write_access(handle, frame->bh);
1207         if (err)
1208                 goto journal_error;
1209
1210         data2 = bh2->b_data;
1211
1212         /* create map in the end of data2 block */
1213         map = (struct dx_map_entry *) (data2 + blocksize);
1214         count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1215                              blocksize, hinfo, map);
1216         map -= count;
1217         dx_sort_map (map, count);
1218         /* Split the existing block in the middle, size-wise */
1219         size = 0;
1220         move = 0;
1221         for (i = count-1; i >= 0; i--) {
1222                 /* is more than half of this entry in 2nd half of the block? */
1223                 if (size + map[i].size/2 > blocksize/2)
1224                         break;
1225                 size += map[i].size;
1226                 move++;
1227         }
1228         /* map index at which we will split */
1229         split = count - move;
1230         hash2 = map[split].hash;
1231         continued = hash2 == map[split - 1].hash;
1232         dxtrace(printk("Split block %i at %x, %i/%i\n",
1233                 dx_get_block(frame->at), hash2, split, count-split));
1234
1235         /* Fancy dance to stay within two buffers */
1236         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1237         de = dx_pack_dirents(data1,blocksize);
1238         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1239         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1240         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1241         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1242
1243         /* Which block gets the new entry? */
1244         if (hinfo->hash >= hash2)
1245         {
1246                 swap(*bh, bh2);
1247                 de = de2;
1248         }
1249         dx_insert_block (frame, hash2 + continued, newblock);
1250         err = ext4_journal_dirty_metadata (handle, bh2);
1251         if (err)
1252                 goto journal_error;
1253         err = ext4_journal_dirty_metadata (handle, frame->bh);
1254         if (err)
1255                 goto journal_error;
1256         brelse (bh2);
1257         dxtrace(dx_show_index ("frame", frame->entries));
1258         return de;
1259
1260 journal_error:
1261         brelse(*bh);
1262         brelse(bh2);
1263         *bh = NULL;
1264         ext4_std_error(dir->i_sb, err);
1265 errout:
1266         *error = err;
1267         return NULL;
1268 }
1269 #endif
1270
1271
1272 /*
1273  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1274  * it points to a directory entry which is guaranteed to be large
1275  * enough for new directory entry.  If de is NULL, then
1276  * add_dirent_to_buf will attempt search the directory block for
1277  * space.  It will return -ENOSPC if no space is available, and -EIO
1278  * and -EEXIST if directory entry already exists.
1279  *
1280  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1281  * all other cases bh is released.
1282  */
1283 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1284                              struct inode *inode, struct ext4_dir_entry_2 *de,
1285                              struct buffer_head * bh)
1286 {
1287         struct inode    *dir = dentry->d_parent->d_inode;
1288         const char      *name = dentry->d_name.name;
1289         int             namelen = dentry->d_name.len;
1290         unsigned long   offset = 0;
1291         unsigned short  reclen;
1292         int             nlen, rlen, err;
1293         char            *top;
1294
1295         reclen = EXT4_DIR_REC_LEN(namelen);
1296         if (!de) {
1297                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1298                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1299                 while ((char *) de <= top) {
1300                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1301                                                   bh, offset)) {
1302                                 brelse (bh);
1303                                 return -EIO;
1304                         }
1305                         if (ext4_match (namelen, name, de)) {
1306                                 brelse (bh);
1307                                 return -EEXIST;
1308                         }
1309                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1310                         rlen = le16_to_cpu(de->rec_len);
1311                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1312                                 break;
1313                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1314                         offset += rlen;
1315                 }
1316                 if ((char *) de > top)
1317                         return -ENOSPC;
1318         }
1319         BUFFER_TRACE(bh, "get_write_access");
1320         err = ext4_journal_get_write_access(handle, bh);
1321         if (err) {
1322                 ext4_std_error(dir->i_sb, err);
1323                 brelse(bh);
1324                 return err;
1325         }
1326
1327         /* By now the buffer is marked for journaling */
1328         nlen = EXT4_DIR_REC_LEN(de->name_len);
1329         rlen = le16_to_cpu(de->rec_len);
1330         if (de->inode) {
1331                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1332                 de1->rec_len = cpu_to_le16(rlen - nlen);
1333                 de->rec_len = cpu_to_le16(nlen);
1334                 de = de1;
1335         }
1336         de->file_type = EXT4_FT_UNKNOWN;
1337         if (inode) {
1338                 de->inode = cpu_to_le32(inode->i_ino);
1339                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1340         } else
1341                 de->inode = 0;
1342         de->name_len = namelen;
1343         memcpy (de->name, name, namelen);
1344         /*
1345          * XXX shouldn't update any times until successful
1346          * completion of syscall, but too many callers depend
1347          * on this.
1348          *
1349          * XXX similarly, too many callers depend on
1350          * ext4_new_inode() setting the times, but error
1351          * recovery deletes the inode, so the worst that can
1352          * happen is that the times are slightly out of date
1353          * and/or different from the directory change time.
1354          */
1355         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1356         ext4_update_dx_flag(dir);
1357         dir->i_version++;
1358         ext4_mark_inode_dirty(handle, dir);
1359         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1360         err = ext4_journal_dirty_metadata(handle, bh);
1361         if (err)
1362                 ext4_std_error(dir->i_sb, err);
1363         brelse(bh);
1364         return 0;
1365 }
1366
1367 #ifdef CONFIG_EXT4_INDEX
1368 /*
1369  * This converts a one block unindexed directory to a 3 block indexed
1370  * directory, and adds the dentry to the indexed directory.
1371  */
1372 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1373                             struct inode *inode, struct buffer_head *bh)
1374 {
1375         struct inode    *dir = dentry->d_parent->d_inode;
1376         const char      *name = dentry->d_name.name;
1377         int             namelen = dentry->d_name.len;
1378         struct buffer_head *bh2;
1379         struct dx_root  *root;
1380         struct dx_frame frames[2], *frame;
1381         struct dx_entry *entries;
1382         struct ext4_dir_entry_2 *de, *de2;
1383         char            *data1, *top;
1384         unsigned        len;
1385         int             retval;
1386         unsigned        blocksize;
1387         struct dx_hash_info hinfo;
1388         u32             block;
1389         struct fake_dirent *fde;
1390
1391         blocksize =  dir->i_sb->s_blocksize;
1392         dxtrace(printk("Creating index\n"));
1393         retval = ext4_journal_get_write_access(handle, bh);
1394         if (retval) {
1395                 ext4_std_error(dir->i_sb, retval);
1396                 brelse(bh);
1397                 return retval;
1398         }
1399         root = (struct dx_root *) bh->b_data;
1400
1401         bh2 = ext4_append (handle, dir, &block, &retval);
1402         if (!(bh2)) {
1403                 brelse(bh);
1404                 return retval;
1405         }
1406         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1407         data1 = bh2->b_data;
1408
1409         /* The 0th block becomes the root, move the dirents out */
1410         fde = &root->dotdot;
1411         de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1412         len = ((char *) root) + blocksize - (char *) de;
1413         memcpy (data1, de, len);
1414         de = (struct ext4_dir_entry_2 *) data1;
1415         top = data1 + len;
1416         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1417                 de = de2;
1418         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1419         /* Initialize the root; the dot dirents already exist */
1420         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1421         de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1422         memset (&root->info, 0, sizeof(root->info));
1423         root->info.info_length = sizeof(root->info);
1424         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1425         entries = root->entries;
1426         dx_set_block (entries, 1);
1427         dx_set_count (entries, 1);
1428         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1429
1430         /* Initialize as for dx_probe */
1431         hinfo.hash_version = root->info.hash_version;
1432         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1433         ext4fs_dirhash(name, namelen, &hinfo);
1434         frame = frames;
1435         frame->entries = entries;
1436         frame->at = entries;
1437         frame->bh = bh;
1438         bh = bh2;
1439         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1440         dx_release (frames);
1441         if (!(de))
1442                 return retval;
1443
1444         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1445 }
1446 #endif
1447
1448 /*
1449  *      ext4_add_entry()
1450  *
1451  * adds a file entry to the specified directory, using the same
1452  * semantics as ext4_find_entry(). It returns NULL if it failed.
1453  *
1454  * NOTE!! The inode part of 'de' is left at 0 - which means you
1455  * may not sleep between calling this and putting something into
1456  * the entry, as someone else might have used it while you slept.
1457  */
1458 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1459         struct inode *inode)
1460 {
1461         struct inode *dir = dentry->d_parent->d_inode;
1462         unsigned long offset;
1463         struct buffer_head * bh;
1464         struct ext4_dir_entry_2 *de;
1465         struct super_block * sb;
1466         int     retval;
1467 #ifdef CONFIG_EXT4_INDEX
1468         int     dx_fallback=0;
1469 #endif
1470         unsigned blocksize;
1471         u32 block, blocks;
1472
1473         sb = dir->i_sb;
1474         blocksize = sb->s_blocksize;
1475         if (!dentry->d_name.len)
1476                 return -EINVAL;
1477 #ifdef CONFIG_EXT4_INDEX
1478         if (is_dx(dir)) {
1479                 retval = ext4_dx_add_entry(handle, dentry, inode);
1480                 if (!retval || (retval != ERR_BAD_DX_DIR))
1481                         return retval;
1482                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1483                 dx_fallback++;
1484                 ext4_mark_inode_dirty(handle, dir);
1485         }
1486 #endif
1487         blocks = dir->i_size >> sb->s_blocksize_bits;
1488         for (block = 0, offset = 0; block < blocks; block++) {
1489                 bh = ext4_bread(handle, dir, block, 0, &retval);
1490                 if(!bh)
1491                         return retval;
1492                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1493                 if (retval != -ENOSPC)
1494                         return retval;
1495
1496 #ifdef CONFIG_EXT4_INDEX
1497                 if (blocks == 1 && !dx_fallback &&
1498                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1499                         return make_indexed_dir(handle, dentry, inode, bh);
1500 #endif
1501                 brelse(bh);
1502         }
1503         bh = ext4_append(handle, dir, &block, &retval);
1504         if (!bh)
1505                 return retval;
1506         de = (struct ext4_dir_entry_2 *) bh->b_data;
1507         de->inode = 0;
1508         de->rec_len = cpu_to_le16(blocksize);
1509         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1510 }
1511
1512 #ifdef CONFIG_EXT4_INDEX
1513 /*
1514  * Returns 0 for success, or a negative error value
1515  */
1516 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1517                              struct inode *inode)
1518 {
1519         struct dx_frame frames[2], *frame;
1520         struct dx_entry *entries, *at;
1521         struct dx_hash_info hinfo;
1522         struct buffer_head * bh;
1523         struct inode *dir = dentry->d_parent->d_inode;
1524         struct super_block * sb = dir->i_sb;
1525         struct ext4_dir_entry_2 *de;
1526         int err;
1527
1528         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1529         if (!frame)
1530                 return err;
1531         entries = frame->entries;
1532         at = frame->at;
1533
1534         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1535                 goto cleanup;
1536
1537         BUFFER_TRACE(bh, "get_write_access");
1538         err = ext4_journal_get_write_access(handle, bh);
1539         if (err)
1540                 goto journal_error;
1541
1542         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1543         if (err != -ENOSPC) {
1544                 bh = NULL;
1545                 goto cleanup;
1546         }
1547
1548         /* Block full, should compress but for now just split */
1549         dxtrace(printk("using %u of %u node entries\n",
1550                        dx_get_count(entries), dx_get_limit(entries)));
1551         /* Need to split index? */
1552         if (dx_get_count(entries) == dx_get_limit(entries)) {
1553                 u32 newblock;
1554                 unsigned icount = dx_get_count(entries);
1555                 int levels = frame - frames;
1556                 struct dx_entry *entries2;
1557                 struct dx_node *node2;
1558                 struct buffer_head *bh2;
1559
1560                 if (levels && (dx_get_count(frames->entries) ==
1561                                dx_get_limit(frames->entries))) {
1562                         ext4_warning(sb, __FUNCTION__,
1563                                      "Directory index full!");
1564                         err = -ENOSPC;
1565                         goto cleanup;
1566                 }
1567                 bh2 = ext4_append (handle, dir, &newblock, &err);
1568                 if (!(bh2))
1569                         goto cleanup;
1570                 node2 = (struct dx_node *)(bh2->b_data);
1571                 entries2 = node2->entries;
1572                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1573                 node2->fake.inode = 0;
1574                 BUFFER_TRACE(frame->bh, "get_write_access");
1575                 err = ext4_journal_get_write_access(handle, frame->bh);
1576                 if (err)
1577                         goto journal_error;
1578                 if (levels) {
1579                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1580                         unsigned hash2 = dx_get_hash(entries + icount1);
1581                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1582
1583                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1584                         err = ext4_journal_get_write_access(handle,
1585                                                              frames[0].bh);
1586                         if (err)
1587                                 goto journal_error;
1588
1589                         memcpy ((char *) entries2, (char *) (entries + icount1),
1590                                 icount2 * sizeof(struct dx_entry));
1591                         dx_set_count (entries, icount1);
1592                         dx_set_count (entries2, icount2);
1593                         dx_set_limit (entries2, dx_node_limit(dir));
1594
1595                         /* Which index block gets the new entry? */
1596                         if (at - entries >= icount1) {
1597                                 frame->at = at = at - entries - icount1 + entries2;
1598                                 frame->entries = entries = entries2;
1599                                 swap(frame->bh, bh2);
1600                         }
1601                         dx_insert_block (frames + 0, hash2, newblock);
1602                         dxtrace(dx_show_index ("node", frames[1].entries));
1603                         dxtrace(dx_show_index ("node",
1604                                ((struct dx_node *) bh2->b_data)->entries));
1605                         err = ext4_journal_dirty_metadata(handle, bh2);
1606                         if (err)
1607                                 goto journal_error;
1608                         brelse (bh2);
1609                 } else {
1610                         dxtrace(printk("Creating second level index...\n"));
1611                         memcpy((char *) entries2, (char *) entries,
1612                                icount * sizeof(struct dx_entry));
1613                         dx_set_limit(entries2, dx_node_limit(dir));
1614
1615                         /* Set up root */
1616                         dx_set_count(entries, 1);
1617                         dx_set_block(entries + 0, newblock);
1618                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1619
1620                         /* Add new access path frame */
1621                         frame = frames + 1;
1622                         frame->at = at = at - entries + entries2;
1623                         frame->entries = entries = entries2;
1624                         frame->bh = bh2;
1625                         err = ext4_journal_get_write_access(handle,
1626                                                              frame->bh);
1627                         if (err)
1628                                 goto journal_error;
1629                 }
1630                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1631         }
1632         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1633         if (!de)
1634                 goto cleanup;
1635         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1636         bh = NULL;
1637         goto cleanup;
1638
1639 journal_error:
1640         ext4_std_error(dir->i_sb, err);
1641 cleanup:
1642         if (bh)
1643                 brelse(bh);
1644         dx_release(frames);
1645         return err;
1646 }
1647 #endif
1648
1649 /*
1650  * ext4_delete_entry deletes a directory entry by merging it with the
1651  * previous entry
1652  */
1653 static int ext4_delete_entry (handle_t *handle,
1654                               struct inode * dir,
1655                               struct ext4_dir_entry_2 * de_del,
1656                               struct buffer_head * bh)
1657 {
1658         struct ext4_dir_entry_2 * de, * pde;
1659         int i;
1660
1661         i = 0;
1662         pde = NULL;
1663         de = (struct ext4_dir_entry_2 *) bh->b_data;
1664         while (i < bh->b_size) {
1665                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1666                         return -EIO;
1667                 if (de == de_del)  {
1668                         BUFFER_TRACE(bh, "get_write_access");
1669                         ext4_journal_get_write_access(handle, bh);
1670                         if (pde)
1671                                 pde->rec_len =
1672                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1673                                                     le16_to_cpu(de->rec_len));
1674                         else
1675                                 de->inode = 0;
1676                         dir->i_version++;
1677                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1678                         ext4_journal_dirty_metadata(handle, bh);
1679                         return 0;
1680                 }
1681                 i += le16_to_cpu(de->rec_len);
1682                 pde = de;
1683                 de = (struct ext4_dir_entry_2 *)
1684                         ((char *) de + le16_to_cpu(de->rec_len));
1685         }
1686         return -ENOENT;
1687 }
1688
1689 /*
1690  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1691  * since this indicates that nlinks count was previously 1.
1692  */
1693 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1694 {
1695         inc_nlink(inode);
1696         if (is_dx(inode) && inode->i_nlink > 1) {
1697                 /* limit is 16-bit i_links_count */
1698                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1699                         inode->i_nlink = 1;
1700                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1701                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1702                 }
1703         }
1704 }
1705
1706 /*
1707  * If a directory had nlink == 1, then we should let it be 1. This indicates
1708  * directory has >EXT4_LINK_MAX subdirs.
1709  */
1710 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1711 {
1712         drop_nlink(inode);
1713         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1714                 inc_nlink(inode);
1715 }
1716
1717
1718 static int ext4_add_nondir(handle_t *handle,
1719                 struct dentry *dentry, struct inode *inode)
1720 {
1721         int err = ext4_add_entry(handle, dentry, inode);
1722         if (!err) {
1723                 ext4_mark_inode_dirty(handle, inode);
1724                 d_instantiate(dentry, inode);
1725                 return 0;
1726         }
1727         drop_nlink(inode);
1728         iput(inode);
1729         return err;
1730 }
1731
1732 /*
1733  * By the time this is called, we already have created
1734  * the directory cache entry for the new file, but it
1735  * is so far negative - it has no inode.
1736  *
1737  * If the create succeeds, we fill in the inode information
1738  * with d_instantiate().
1739  */
1740 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1741                 struct nameidata *nd)
1742 {
1743         handle_t *handle;
1744         struct inode * inode;
1745         int err, retries = 0;
1746
1747 retry:
1748         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1749                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1750                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1751         if (IS_ERR(handle))
1752                 return PTR_ERR(handle);
1753
1754         if (IS_DIRSYNC(dir))
1755                 handle->h_sync = 1;
1756
1757         inode = ext4_new_inode (handle, dir, mode);
1758         err = PTR_ERR(inode);
1759         if (!IS_ERR(inode)) {
1760                 inode->i_op = &ext4_file_inode_operations;
1761                 inode->i_fop = &ext4_file_operations;
1762                 ext4_set_aops(inode);
1763                 err = ext4_add_nondir(handle, dentry, inode);
1764         }
1765         ext4_journal_stop(handle);
1766         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1767                 goto retry;
1768         return err;
1769 }
1770
1771 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1772                         int mode, dev_t rdev)
1773 {
1774         handle_t *handle;
1775         struct inode *inode;
1776         int err, retries = 0;
1777
1778         if (!new_valid_dev(rdev))
1779                 return -EINVAL;
1780
1781 retry:
1782         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1783                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1784                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1785         if (IS_ERR(handle))
1786                 return PTR_ERR(handle);
1787
1788         if (IS_DIRSYNC(dir))
1789                 handle->h_sync = 1;
1790
1791         inode = ext4_new_inode (handle, dir, mode);
1792         err = PTR_ERR(inode);
1793         if (!IS_ERR(inode)) {
1794                 init_special_inode(inode, inode->i_mode, rdev);
1795 #ifdef CONFIG_EXT4DEV_FS_XATTR
1796                 inode->i_op = &ext4_special_inode_operations;
1797 #endif
1798                 err = ext4_add_nondir(handle, dentry, inode);
1799         }
1800         ext4_journal_stop(handle);
1801         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1802                 goto retry;
1803         return err;
1804 }
1805
1806 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1807 {
1808         handle_t *handle;
1809         struct inode * inode;
1810         struct buffer_head * dir_block;
1811         struct ext4_dir_entry_2 * de;
1812         int err, retries = 0;
1813
1814         if (EXT4_DIR_LINK_MAX(dir))
1815                 return -EMLINK;
1816
1817 retry:
1818         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1819                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1820                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1821         if (IS_ERR(handle))
1822                 return PTR_ERR(handle);
1823
1824         if (IS_DIRSYNC(dir))
1825                 handle->h_sync = 1;
1826
1827         inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1828         err = PTR_ERR(inode);
1829         if (IS_ERR(inode))
1830                 goto out_stop;
1831
1832         inode->i_op = &ext4_dir_inode_operations;
1833         inode->i_fop = &ext4_dir_operations;
1834         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1835         dir_block = ext4_bread (handle, inode, 0, 1, &err);
1836         if (!dir_block) {
1837                 ext4_dec_count(handle, inode); /* is this nlink == 0? */
1838                 ext4_mark_inode_dirty(handle, inode);
1839                 iput (inode);
1840                 goto out_stop;
1841         }
1842         BUFFER_TRACE(dir_block, "get_write_access");
1843         ext4_journal_get_write_access(handle, dir_block);
1844         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1845         de->inode = cpu_to_le32(inode->i_ino);
1846         de->name_len = 1;
1847         de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1848         strcpy (de->name, ".");
1849         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1850         de = (struct ext4_dir_entry_2 *)
1851                         ((char *) de + le16_to_cpu(de->rec_len));
1852         de->inode = cpu_to_le32(dir->i_ino);
1853         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1854         de->name_len = 2;
1855         strcpy (de->name, "..");
1856         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1857         inode->i_nlink = 2;
1858         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1859         ext4_journal_dirty_metadata(handle, dir_block);
1860         brelse (dir_block);
1861         ext4_mark_inode_dirty(handle, inode);
1862         err = ext4_add_entry (handle, dentry, inode);
1863         if (err) {
1864                 inode->i_nlink = 0;
1865                 ext4_mark_inode_dirty(handle, inode);
1866                 iput (inode);
1867                 goto out_stop;
1868         }
1869         ext4_inc_count(handle, dir);
1870         ext4_update_dx_flag(dir);
1871         ext4_mark_inode_dirty(handle, dir);
1872         d_instantiate(dentry, inode);
1873 out_stop:
1874         ext4_journal_stop(handle);
1875         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1876                 goto retry;
1877         return err;
1878 }
1879
1880 /*
1881  * routine to check that the specified directory is empty (for rmdir)
1882  */
1883 static int empty_dir (struct inode * inode)
1884 {
1885         unsigned long offset;
1886         struct buffer_head * bh;
1887         struct ext4_dir_entry_2 * de, * de1;
1888         struct super_block * sb;
1889         int err = 0;
1890
1891         sb = inode->i_sb;
1892         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1893             !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1894                 if (err)
1895                         ext4_error(inode->i_sb, __FUNCTION__,
1896                                    "error %d reading directory #%lu offset 0",
1897                                    err, inode->i_ino);
1898                 else
1899                         ext4_warning(inode->i_sb, __FUNCTION__,
1900                                      "bad directory (dir #%lu) - no data block",
1901                                      inode->i_ino);
1902                 return 1;
1903         }
1904         de = (struct ext4_dir_entry_2 *) bh->b_data;
1905         de1 = (struct ext4_dir_entry_2 *)
1906                         ((char *) de + le16_to_cpu(de->rec_len));
1907         if (le32_to_cpu(de->inode) != inode->i_ino ||
1908                         !le32_to_cpu(de1->inode) ||
1909                         strcmp (".", de->name) ||
1910                         strcmp ("..", de1->name)) {
1911                 ext4_warning (inode->i_sb, "empty_dir",
1912                               "bad directory (dir #%lu) - no `.' or `..'",
1913                               inode->i_ino);
1914                 brelse (bh);
1915                 return 1;
1916         }
1917         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1918         de = (struct ext4_dir_entry_2 *)
1919                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1920         while (offset < inode->i_size ) {
1921                 if (!bh ||
1922                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1923                         err = 0;
1924                         brelse (bh);
1925                         bh = ext4_bread (NULL, inode,
1926                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1927                         if (!bh) {
1928                                 if (err)
1929                                         ext4_error(sb, __FUNCTION__,
1930                                                    "error %d reading directory"
1931                                                    " #%lu offset %lu",
1932                                                    err, inode->i_ino, offset);
1933                                 offset += sb->s_blocksize;
1934                                 continue;
1935                         }
1936                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1937                 }
1938                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1939                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1940                                                          sb->s_blocksize);
1941                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1942                         continue;
1943                 }
1944                 if (le32_to_cpu(de->inode)) {
1945                         brelse (bh);
1946                         return 0;
1947                 }
1948                 offset += le16_to_cpu(de->rec_len);
1949                 de = (struct ext4_dir_entry_2 *)
1950                                 ((char *) de + le16_to_cpu(de->rec_len));
1951         }
1952         brelse (bh);
1953         return 1;
1954 }
1955
1956 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1957  * such inodes, starting at the superblock, in case we crash before the
1958  * file is closed/deleted, or in case the inode truncate spans multiple
1959  * transactions and the last transaction is not recovered after a crash.
1960  *
1961  * At filesystem recovery time, we walk this list deleting unlinked
1962  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1963  */
1964 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1965 {
1966         struct super_block *sb = inode->i_sb;
1967         struct ext4_iloc iloc;
1968         int err = 0, rc;
1969
1970         lock_super(sb);
1971         if (!list_empty(&EXT4_I(inode)->i_orphan))
1972                 goto out_unlock;
1973
1974         /* Orphan handling is only valid for files with data blocks
1975          * being truncated, or files being unlinked. */
1976
1977         /* @@@ FIXME: Observation from aviro:
1978          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1979          * here (on lock_super()), so race with ext4_link() which might bump
1980          * ->i_nlink. For, say it, character device. Not a regular file,
1981          * not a directory, not a symlink and ->i_nlink > 0.
1982          */
1983         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1984                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1985
1986         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1987         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1988         if (err)
1989                 goto out_unlock;
1990
1991         err = ext4_reserve_inode_write(handle, inode, &iloc);
1992         if (err)
1993                 goto out_unlock;
1994
1995         /* Insert this inode at the head of the on-disk orphan list... */
1996         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1997         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1998         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1999         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2000         if (!err)
2001                 err = rc;
2002
2003         /* Only add to the head of the in-memory list if all the
2004          * previous operations succeeded.  If the orphan_add is going to
2005          * fail (possibly taking the journal offline), we can't risk
2006          * leaving the inode on the orphan list: stray orphan-list
2007          * entries can cause panics at unmount time.
2008          *
2009          * This is safe: on error we're going to ignore the orphan list
2010          * anyway on the next recovery. */
2011         if (!err)
2012                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2013
2014         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2015         jbd_debug(4, "orphan inode %lu will point to %d\n",
2016                         inode->i_ino, NEXT_ORPHAN(inode));
2017 out_unlock:
2018         unlock_super(sb);
2019         ext4_std_error(inode->i_sb, err);
2020         return err;
2021 }
2022
2023 /*
2024  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2025  * of such inodes stored on disk, because it is finally being cleaned up.
2026  */
2027 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2028 {
2029         struct list_head *prev;
2030         struct ext4_inode_info *ei = EXT4_I(inode);
2031         struct ext4_sb_info *sbi;
2032         unsigned long ino_next;
2033         struct ext4_iloc iloc;
2034         int err = 0;
2035
2036         lock_super(inode->i_sb);
2037         if (list_empty(&ei->i_orphan)) {
2038                 unlock_super(inode->i_sb);
2039                 return 0;
2040         }
2041
2042         ino_next = NEXT_ORPHAN(inode);
2043         prev = ei->i_orphan.prev;
2044         sbi = EXT4_SB(inode->i_sb);
2045
2046         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2047
2048         list_del_init(&ei->i_orphan);
2049
2050         /* If we're on an error path, we may not have a valid
2051          * transaction handle with which to update the orphan list on
2052          * disk, but we still need to remove the inode from the linked
2053          * list in memory. */
2054         if (!handle)
2055                 goto out;
2056
2057         err = ext4_reserve_inode_write(handle, inode, &iloc);
2058         if (err)
2059                 goto out_err;
2060
2061         if (prev == &sbi->s_orphan) {
2062                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2063                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2064                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2065                 if (err)
2066                         goto out_brelse;
2067                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2068                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2069         } else {
2070                 struct ext4_iloc iloc2;
2071                 struct inode *i_prev =
2072                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2073
2074                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2075                           i_prev->i_ino, ino_next);
2076                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2077                 if (err)
2078                         goto out_brelse;
2079                 NEXT_ORPHAN(i_prev) = ino_next;
2080                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2081         }
2082         if (err)
2083                 goto out_brelse;
2084         NEXT_ORPHAN(inode) = 0;
2085         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2086
2087 out_err:
2088         ext4_std_error(inode->i_sb, err);
2089 out:
2090         unlock_super(inode->i_sb);
2091         return err;
2092
2093 out_brelse:
2094         brelse(iloc.bh);
2095         goto out_err;
2096 }
2097
2098 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2099 {
2100         int retval;
2101         struct inode * inode;
2102         struct buffer_head * bh;
2103         struct ext4_dir_entry_2 * de;
2104         handle_t *handle;
2105
2106         /* Initialize quotas before so that eventual writes go in
2107          * separate transaction */
2108         DQUOT_INIT(dentry->d_inode);
2109         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2110         if (IS_ERR(handle))
2111                 return PTR_ERR(handle);
2112
2113         retval = -ENOENT;
2114         bh = ext4_find_entry (dentry, &de);
2115         if (!bh)
2116                 goto end_rmdir;
2117
2118         if (IS_DIRSYNC(dir))
2119                 handle->h_sync = 1;
2120
2121         inode = dentry->d_inode;
2122
2123         retval = -EIO;
2124         if (le32_to_cpu(de->inode) != inode->i_ino)
2125                 goto end_rmdir;
2126
2127         retval = -ENOTEMPTY;
2128         if (!empty_dir (inode))
2129                 goto end_rmdir;
2130
2131         retval = ext4_delete_entry(handle, dir, de, bh);
2132         if (retval)
2133                 goto end_rmdir;
2134         if (!EXT4_DIR_LINK_EMPTY(inode))
2135                 ext4_warning (inode->i_sb, "ext4_rmdir",
2136                               "empty directory has too many links (%d)",
2137                               inode->i_nlink);
2138         inode->i_version++;
2139         clear_nlink(inode);
2140         /* There's no need to set i_disksize: the fact that i_nlink is
2141          * zero will ensure that the right thing happens during any
2142          * recovery. */
2143         inode->i_size = 0;
2144         ext4_orphan_add(handle, inode);
2145         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2146         ext4_mark_inode_dirty(handle, inode);
2147         ext4_dec_count(handle, dir);
2148         ext4_update_dx_flag(dir);
2149         ext4_mark_inode_dirty(handle, dir);
2150
2151 end_rmdir:
2152         ext4_journal_stop(handle);
2153         brelse (bh);
2154         return retval;
2155 }
2156
2157 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2158 {
2159         int retval;
2160         struct inode * inode;
2161         struct buffer_head * bh;
2162         struct ext4_dir_entry_2 * de;
2163         handle_t *handle;
2164
2165         /* Initialize quotas before so that eventual writes go
2166          * in separate transaction */
2167         DQUOT_INIT(dentry->d_inode);
2168         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2169         if (IS_ERR(handle))
2170                 return PTR_ERR(handle);
2171
2172         if (IS_DIRSYNC(dir))
2173                 handle->h_sync = 1;
2174
2175         retval = -ENOENT;
2176         bh = ext4_find_entry (dentry, &de);
2177         if (!bh)
2178                 goto end_unlink;
2179
2180         inode = dentry->d_inode;
2181
2182         retval = -EIO;
2183         if (le32_to_cpu(de->inode) != inode->i_ino)
2184                 goto end_unlink;
2185
2186         if (!inode->i_nlink) {
2187                 ext4_warning (inode->i_sb, "ext4_unlink",
2188                               "Deleting nonexistent file (%lu), %d",
2189                               inode->i_ino, inode->i_nlink);
2190                 inode->i_nlink = 1;
2191         }
2192         retval = ext4_delete_entry(handle, dir, de, bh);
2193         if (retval)
2194                 goto end_unlink;
2195         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2196         ext4_update_dx_flag(dir);
2197         ext4_mark_inode_dirty(handle, dir);
2198         ext4_dec_count(handle, inode);
2199         if (!inode->i_nlink)
2200                 ext4_orphan_add(handle, inode);
2201         inode->i_ctime = ext4_current_time(inode);
2202         ext4_mark_inode_dirty(handle, inode);
2203         retval = 0;
2204
2205 end_unlink:
2206         ext4_journal_stop(handle);
2207         brelse (bh);
2208         return retval;
2209 }
2210
2211 static int ext4_symlink (struct inode * dir,
2212                 struct dentry *dentry, const char * symname)
2213 {
2214         handle_t *handle;
2215         struct inode * inode;
2216         int l, err, retries = 0;
2217
2218         l = strlen(symname)+1;
2219         if (l > dir->i_sb->s_blocksize)
2220                 return -ENAMETOOLONG;
2221
2222 retry:
2223         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2224                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2225                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2226         if (IS_ERR(handle))
2227                 return PTR_ERR(handle);
2228
2229         if (IS_DIRSYNC(dir))
2230                 handle->h_sync = 1;
2231
2232         inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2233         err = PTR_ERR(inode);
2234         if (IS_ERR(inode))
2235                 goto out_stop;
2236
2237         if (l > sizeof (EXT4_I(inode)->i_data)) {
2238                 inode->i_op = &ext4_symlink_inode_operations;
2239                 ext4_set_aops(inode);
2240                 /*
2241                  * page_symlink() calls into ext4_prepare/commit_write.
2242                  * We have a transaction open.  All is sweetness.  It also sets
2243                  * i_size in generic_commit_write().
2244                  */
2245                 err = __page_symlink(inode, symname, l,
2246                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2247                 if (err) {
2248                         ext4_dec_count(handle, inode);
2249                         ext4_mark_inode_dirty(handle, inode);
2250                         iput (inode);
2251                         goto out_stop;
2252                 }
2253         } else {
2254                 inode->i_op = &ext4_fast_symlink_inode_operations;
2255                 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2256                 inode->i_size = l-1;
2257         }
2258         EXT4_I(inode)->i_disksize = inode->i_size;
2259         err = ext4_add_nondir(handle, dentry, inode);
2260 out_stop:
2261         ext4_journal_stop(handle);
2262         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2263                 goto retry;
2264         return err;
2265 }
2266
2267 static int ext4_link (struct dentry * old_dentry,
2268                 struct inode * dir, struct dentry *dentry)
2269 {
2270         handle_t *handle;
2271         struct inode *inode = old_dentry->d_inode;
2272         int err, retries = 0;
2273
2274         if (EXT4_DIR_LINK_MAX(inode))
2275                 return -EMLINK;
2276
2277         /*
2278          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2279          * otherwise has the potential to corrupt the orphan inode list.
2280          */
2281         if (inode->i_nlink == 0)
2282                 return -ENOENT;
2283
2284 retry:
2285         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2286                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2287         if (IS_ERR(handle))
2288                 return PTR_ERR(handle);
2289
2290         if (IS_DIRSYNC(dir))
2291                 handle->h_sync = 1;
2292
2293         inode->i_ctime = ext4_current_time(inode);
2294         ext4_inc_count(handle, inode);
2295         atomic_inc(&inode->i_count);
2296
2297         err = ext4_add_nondir(handle, dentry, inode);
2298         ext4_journal_stop(handle);
2299         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2300                 goto retry;
2301         return err;
2302 }
2303
2304 #define PARENT_INO(buffer) \
2305         ((struct ext4_dir_entry_2 *) ((char *) buffer + \
2306         le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2307
2308 /*
2309  * Anybody can rename anything with this: the permission checks are left to the
2310  * higher-level routines.
2311  */
2312 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2313                            struct inode * new_dir,struct dentry *new_dentry)
2314 {
2315         handle_t *handle;
2316         struct inode * old_inode, * new_inode;
2317         struct buffer_head * old_bh, * new_bh, * dir_bh;
2318         struct ext4_dir_entry_2 * old_de, * new_de;
2319         int retval;
2320
2321         old_bh = new_bh = dir_bh = NULL;
2322
2323         /* Initialize quotas before so that eventual writes go
2324          * in separate transaction */
2325         if (new_dentry->d_inode)
2326                 DQUOT_INIT(new_dentry->d_inode);
2327         handle = ext4_journal_start(old_dir, 2 *
2328                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2329                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2330         if (IS_ERR(handle))
2331                 return PTR_ERR(handle);
2332
2333         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2334                 handle->h_sync = 1;
2335
2336         old_bh = ext4_find_entry (old_dentry, &old_de);
2337         /*
2338          *  Check for inode number is _not_ due to possible IO errors.
2339          *  We might rmdir the source, keep it as pwd of some process
2340          *  and merrily kill the link to whatever was created under the
2341          *  same name. Goodbye sticky bit ;-<
2342          */
2343         old_inode = old_dentry->d_inode;
2344         retval = -ENOENT;
2345         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2346                 goto end_rename;
2347
2348         new_inode = new_dentry->d_inode;
2349         new_bh = ext4_find_entry (new_dentry, &new_de);
2350         if (new_bh) {
2351                 if (!new_inode) {
2352                         brelse (new_bh);
2353                         new_bh = NULL;
2354                 }
2355         }
2356         if (S_ISDIR(old_inode->i_mode)) {
2357                 if (new_inode) {
2358                         retval = -ENOTEMPTY;
2359                         if (!empty_dir (new_inode))
2360                                 goto end_rename;
2361                 }
2362                 retval = -EIO;
2363                 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2364                 if (!dir_bh)
2365                         goto end_rename;
2366                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2367                         goto end_rename;
2368                 retval = -EMLINK;
2369                 if (!new_inode && new_dir!=old_dir &&
2370                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2371                         goto end_rename;
2372         }
2373         if (!new_bh) {
2374                 retval = ext4_add_entry (handle, new_dentry, old_inode);
2375                 if (retval)
2376                         goto end_rename;
2377         } else {
2378                 BUFFER_TRACE(new_bh, "get write access");
2379                 ext4_journal_get_write_access(handle, new_bh);
2380                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2381                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2382                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2383                         new_de->file_type = old_de->file_type;
2384                 new_dir->i_version++;
2385                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2386                 ext4_journal_dirty_metadata(handle, new_bh);
2387                 brelse(new_bh);
2388                 new_bh = NULL;
2389         }
2390
2391         /*
2392          * Like most other Unix systems, set the ctime for inodes on a
2393          * rename.
2394          */
2395         old_inode->i_ctime = ext4_current_time(old_inode);
2396         ext4_mark_inode_dirty(handle, old_inode);
2397
2398         /*
2399          * ok, that's it
2400          */
2401         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2402             old_de->name_len != old_dentry->d_name.len ||
2403             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2404             (retval = ext4_delete_entry(handle, old_dir,
2405                                         old_de, old_bh)) == -ENOENT) {
2406                 /* old_de could have moved from under us during htree split, so
2407                  * make sure that we are deleting the right entry.  We might
2408                  * also be pointing to a stale entry in the unused part of
2409                  * old_bh so just checking inum and the name isn't enough. */
2410                 struct buffer_head *old_bh2;
2411                 struct ext4_dir_entry_2 *old_de2;
2412
2413                 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2414                 if (old_bh2) {
2415                         retval = ext4_delete_entry(handle, old_dir,
2416                                                    old_de2, old_bh2);
2417                         brelse(old_bh2);
2418                 }
2419         }
2420         if (retval) {
2421                 ext4_warning(old_dir->i_sb, "ext4_rename",
2422                                 "Deleting old file (%lu), %d, error=%d",
2423                                 old_dir->i_ino, old_dir->i_nlink, retval);
2424         }
2425
2426         if (new_inode) {
2427                 ext4_dec_count(handle, new_inode);
2428                 new_inode->i_ctime = ext4_current_time(new_inode);
2429         }
2430         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2431         ext4_update_dx_flag(old_dir);
2432         if (dir_bh) {
2433                 BUFFER_TRACE(dir_bh, "get_write_access");
2434                 ext4_journal_get_write_access(handle, dir_bh);
2435                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2436                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2437                 ext4_journal_dirty_metadata(handle, dir_bh);
2438                 ext4_dec_count(handle, old_dir);
2439                 if (new_inode) {
2440                         /* checked empty_dir above, can't have another parent,
2441                          * ext3_dec_count() won't work for many-linked dirs */
2442                         new_inode->i_nlink = 0;
2443                 } else {
2444                         ext4_inc_count(handle, new_dir);
2445                         ext4_update_dx_flag(new_dir);
2446                         ext4_mark_inode_dirty(handle, new_dir);
2447                 }
2448         }
2449         ext4_mark_inode_dirty(handle, old_dir);
2450         if (new_inode) {
2451                 ext4_mark_inode_dirty(handle, new_inode);
2452                 if (!new_inode->i_nlink)
2453                         ext4_orphan_add(handle, new_inode);
2454         }
2455         retval = 0;
2456
2457 end_rename:
2458         brelse (dir_bh);
2459         brelse (old_bh);
2460         brelse (new_bh);
2461         ext4_journal_stop(handle);
2462         return retval;
2463 }
2464
2465 /*
2466  * directories can handle most operations...
2467  */
2468 const struct inode_operations ext4_dir_inode_operations = {
2469         .create         = ext4_create,
2470         .lookup         = ext4_lookup,
2471         .link           = ext4_link,
2472         .unlink         = ext4_unlink,
2473         .symlink        = ext4_symlink,
2474         .mkdir          = ext4_mkdir,
2475         .rmdir          = ext4_rmdir,
2476         .mknod          = ext4_mknod,
2477         .rename         = ext4_rename,
2478         .setattr        = ext4_setattr,
2479 #ifdef CONFIG_EXT4DEV_FS_XATTR
2480         .setxattr       = generic_setxattr,
2481         .getxattr       = generic_getxattr,
2482         .listxattr      = ext4_listxattr,
2483         .removexattr    = generic_removexattr,
2484 #endif
2485         .permission     = ext4_permission,
2486 };
2487
2488 const struct inode_operations ext4_special_inode_operations = {
2489         .setattr        = ext4_setattr,
2490 #ifdef CONFIG_EXT4DEV_FS_XATTR
2491         .setxattr       = generic_setxattr,
2492         .getxattr       = generic_getxattr,
2493         .listxattr      = ext4_listxattr,
2494         .removexattr    = generic_removexattr,
2495 #endif
2496         .permission     = ext4_permission,
2497 };