fs/ext3/namei.c

   1 /*
   2  *  linux/fs/ext3/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/quotaops.h>
  28 #include "ext3.h"
  29 #include "namei.h"
  30 #include "xattr.h"
  31 #include "acl.h"
  32
  33 /*
  34  * define how far ahead to read directories while searching them.
  35  */
  36 #define NAMEI_RA_CHUNKS  2
  37 #define NAMEI_RA_BLOCKS  4
  38 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  39
  40 static struct buffer_head *ext3_append(handle_t *handle,
  41                                         struct inode *inode,
  42                                         u32 *block, int *err)
  43 {
  44         struct buffer_head *bh;
  45
  46         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  47
  48         if ((bh = ext3_dir_bread(handle, inode, *block, 1, err))) {
  49                 inode->i_size += inode->i_sb->s_blocksize;
  50                 EXT3_I(inode)->i_disksize = inode->i_size;
  51                 *err = ext3_journal_get_write_access(handle, bh);
  52                 if (*err) {
  53                         brelse(bh);
  54                         bh = NULL;
  55                 }
  56         }
  57         return bh;
  58 }
  59
  60 #ifndef assert
  61 #define assert(test) J_ASSERT(test)
  62 #endif
  63
  64 #ifdef DX_DEBUG
  65 #define dxtrace(command) command
  66 #else
  67 #define dxtrace(command)
  68 #endif
  69
  70 struct fake_dirent
  71 {
  72         __le32 inode;
  73         __le16 rec_len;
  74         u8 name_len;
  75         u8 file_type;
  76 };
  77
  78 struct dx_countlimit
  79 {
  80         __le16 limit;
  81         __le16 count;
  82 };
  83
  84 struct dx_entry
  85 {
  86         __le32 hash;
  87         __le32 block;
  88 };
  89
  90 /*
  91  * dx_root_info is laid out so that if it should somehow get overlaid by a
  92  * dirent the two low bits of the hash version will be zero.  Therefore, the
  93  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
  94  */
  95
  96 struct dx_root
  97 {
  98         struct fake_dirent dot;
  99         char dot_name[4];
 100         struct fake_dirent dotdot;
 101         char dotdot_name[4];
 102         struct dx_root_info
 103         {
 104                 __le32 reserved_zero;
 105                 u8 hash_version;
 106                 u8 info_length; /* 8 */
 107                 u8 indirect_levels;
 108                 u8 unused_flags;
 109         }
 110         info;
 111         struct dx_entry entries[0];
 112 };
 113
 114 struct dx_node
 115 {
 116         struct fake_dirent fake;
 117         struct dx_entry entries[0];
 118 };
 119
 120
 121 struct dx_frame
 122 {
 123         struct buffer_head *bh;
 124         struct dx_entry *entries;
 125         struct dx_entry *at;
 126 };
 127
 128 struct dx_map_entry
 129 {
 130         u32 hash;
 131         u16 offs;
 132         u16 size;
 133 };
 134
 135 static inline unsigned dx_get_block (struct dx_entry *entry);
 136 static void dx_set_block (struct dx_entry *entry, unsigned value);
 137 static inline unsigned dx_get_hash (struct dx_entry *entry);
 138 static void dx_set_hash (struct dx_entry *entry, unsigned value);
 139 static unsigned dx_get_count (struct dx_entry *entries);
 140 static unsigned dx_get_limit (struct dx_entry *entries);
 141 static void dx_set_count (struct dx_entry *entries, unsigned value);
 142 static void dx_set_limit (struct dx_entry *entries, unsigned value);
 143 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 144 static unsigned dx_node_limit (struct inode *dir);
 145 static struct dx_frame *dx_probe(struct qstr *entry,
 146                                  struct inode *dir,
 147                                  struct dx_hash_info *hinfo,
 148                                  struct dx_frame *frame,
 149                                  int *err);
 150 static void dx_release (struct dx_frame *frames);
 151 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
 152                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 153 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 154 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
 155                 struct dx_map_entry *offsets, int count);
 156 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
 157 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 158 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 159                                  struct dx_frame *frame,
 160                                  struct dx_frame *frames,
 161                                  __u32 *start_hash);
 162 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
 163                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
 164                         int *err);
 165 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
 166                              struct inode *inode);
 167
 168 /*
 169  * p is at least 6 bytes before the end of page
 170  */
 171 static inline struct ext3_dir_entry_2 *
 172 ext3_next_entry(struct ext3_dir_entry_2 *p)
 173 {
 174         return (struct ext3_dir_entry_2 *)((char *)p +
 175                 ext3_rec_len_from_disk(p->rec_len));
 176 }
 177
 178 /*
 179  * Future: use high four bits of block for coalesce-on-delete flags
 180  * Mask them off for now.
 181  */
 182
 183 static inline unsigned dx_get_block (struct dx_entry *entry)
 184 {
 185         return le32_to_cpu(entry->block) & 0x00ffffff;
 186 }
 187
 188 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 189 {
 190         entry->block = cpu_to_le32(value);
 191 }
 192
 193 static inline unsigned dx_get_hash (struct dx_entry *entry)
 194 {
 195         return le32_to_cpu(entry->hash);
 196 }
 197
 198 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 199 {
 200         entry->hash = cpu_to_le32(value);
 201 }
 202
 203 static inline unsigned dx_get_count (struct dx_entry *entries)
 204 {
 205         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 206 }
 207
 208 static inline unsigned dx_get_limit (struct dx_entry *entries)
 209 {
 210         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 211 }
 212
 213 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 214 {
 215         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 216 }
 217
 218 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 219 {
 220         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 221 }
 222
 223 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 224 {
 225         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 226                 EXT3_DIR_REC_LEN(2) - infosize;
 227         return entry_space / sizeof(struct dx_entry);
 228 }
 229
 230 static inline unsigned dx_node_limit (struct inode *dir)
 231 {
 232         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 233         return entry_space / sizeof(struct dx_entry);
 234 }
 235
 236 /*
 237  * Debug
 238  */
 239 #ifdef DX_DEBUG
 240 static void dx_show_index (char * label, struct dx_entry *entries)
 241 {
 242         int i, n = dx_get_count (entries);
 243         printk("%s index ", label);
 244         for (i = 0; i < n; i++)
 245         {
 246                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 247         }
 248         printk("\n");
 249 }
 250
 251 struct stats
 252 {
 253         unsigned names;
 254         unsigned space;
 255         unsigned bcount;
 256 };
 257
 258 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
 259                                  int size, int show_names)
 260 {
 261         unsigned names = 0, space = 0;
 262         char *base = (char *) de;
 263         struct dx_hash_info h = *hinfo;
 264
 265         printk("names: ");
 266         while ((char *) de < base + size)
 267         {
 268                 if (de->inode)
 269                 {
 270                         if (show_names)
 271                         {
 272                                 int len = de->name_len;
 273                                 char *name = de->name;
 274                                 while (len--) printk("%c", *name++);
 275                                 ext3fs_dirhash(de->name, de->name_len, &h);
 276                                 printk(":%x.%u ", h.hash,
 277                                        (unsigned) ((char *) de - base));
 278                         }
 279                         space += EXT3_DIR_REC_LEN(de->name_len);
 280                         names++;
 281                 }
 282                 de = ext3_next_entry(de);
 283         }
 284         printk("(%i)\n", names);
 285         return (struct stats) { names, space, 1 };
 286 }
 287
 288 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 289                              struct dx_entry *entries, int levels)
 290 {
 291         unsigned blocksize = dir->i_sb->s_blocksize;
 292         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 293         unsigned bcount = 0;
 294         struct buffer_head *bh;
 295         int err;
 296         printk("%i indexed blocks...\n", count);
 297         for (i = 0; i < count; i++, entries++)
 298         {
 299                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 300                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 301                 struct stats stats;
 302                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 303                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 304                 stats = levels?
 305                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 306                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
 307                 names += stats.names;
 308                 space += stats.space;
 309                 bcount += stats.bcount;
 310                 brelse (bh);
 311         }
 312         if (bcount)
 313                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 314                         names, space/bcount,(space/bcount)*100/blocksize);
 315         return (struct stats) { names, space, bcount};
 316 }
 317 #endif /* DX_DEBUG */
 318
 319 /*
 320  * Probe for a directory leaf block to search.
 321  *
 322  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 323  * error in the directory index, and the caller should fall back to
 324  * searching the directory normally.  The callers of dx_probe **MUST**
 325  * check for this error code, and make sure it never gets reflected
 326  * back to userspace.
 327  */
 328 static struct dx_frame *
 329 dx_probe(struct qstr *entry, struct inode *dir,
 330          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 331 {
 332         unsigned count, indirect;
 333         struct dx_entry *at, *entries, *p, *q, *m;
 334         struct dx_root *root;
 335         struct buffer_head *bh;
 336         struct dx_frame *frame = frame_in;
 337         u32 hash;
 338
 339         frame->bh = NULL;
 340         if (!(bh = ext3_dir_bread(NULL, dir, 0, 0, err))) {
 341                 *err = ERR_BAD_DX_DIR;
 342                 goto fail;
 343         }
 344         root = (struct dx_root *) bh->b_data;
 345         if (root->info.hash_version != DX_HASH_TEA &&
 346             root->info.hash_version != DX_HASH_HALF_MD4 &&
 347             root->info.hash_version != DX_HASH_LEGACY) {
 348                 ext3_warning(dir->i_sb, __func__,
 349                              "Unrecognised inode hash code %d",
 350                              root->info.hash_version);
 351                 brelse(bh);
 352                 *err = ERR_BAD_DX_DIR;
 353                 goto fail;
 354         }
 355         hinfo->hash_version = root->info.hash_version;
 356         if (hinfo->hash_version <= DX_HASH_TEA)
 357                 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
 358         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 359         if (entry)
 360                 ext3fs_dirhash(entry->name, entry->len, hinfo);
 361         hash = hinfo->hash;
 362
 363         if (root->info.unused_flags & 1) {
 364                 ext3_warning(dir->i_sb, __func__,
 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                 ext3_warning(dir->i_sb, __func__,
 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                 ext3_warning(dir->i_sb, __func__,
 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                         ext3_warning(dir->i_sb, __func__,
 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 = ext3_dir_bread(NULL, dir, dx_get_block(at), 0, err))) {
 440                         *err = ERR_BAD_DX_DIR;
 441                         goto fail2;
 442                 }
 443                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 444                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 445                         ext3_warning(dir->i_sb, __func__,
 446                                      "dx entry: limit != node limit");
 447                         brelse(bh);
 448                         *err = ERR_BAD_DX_DIR;
 449                         goto fail2;
 450                 }
 451                 frame++;
 452                 frame->bh = NULL;
 453         }
 454 fail2:
 455         while (frame >= frame_in) {
 456                 brelse(frame->bh);
 457                 frame--;
 458         }
 459 fail:
 460         if (*err == ERR_BAD_DX_DIR)
 461                 ext3_warning(dir->i_sb, __func__,
 462                              "Corrupt dir inode %ld, running e2fsck is "
 463                              "recommended.", dir->i_ino);
 464         return NULL;
 465 }
 466
 467 static void dx_release (struct dx_frame *frames)
 468 {
 469         if (frames[0].bh == NULL)
 470                 return;
 471
 472         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 473                 brelse(frames[1].bh);
 474         brelse(frames[0].bh);
 475 }
 476
 477 /*
 478  * This function increments the frame pointer to search the next leaf
 479  * block, and reads in the necessary intervening nodes if the search
 480  * should be necessary.  Whether or not the search is necessary is
 481  * controlled by the hash parameter.  If the hash value is even, then
 482  * the search is only continued if the next block starts with that
 483  * hash value.  This is used if we are searching for a specific file.
 484  *
 485  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 486  *
 487  * This function returns 1 if the caller should continue to search,
 488  * or 0 if it should not.  If there is an error reading one of the
 489  * index blocks, it will a negative error code.
 490  *
 491  * If start_hash is non-null, it will be filled in with the starting
 492  * hash of the next page.
 493  */
 494 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 495                                  struct dx_frame *frame,
 496                                  struct dx_frame *frames,
 497                                  __u32 *start_hash)
 498 {
 499         struct dx_frame *p;
 500         struct buffer_head *bh;
 501         int err, num_frames = 0;
 502         __u32 bhash;
 503
 504         p = frame;
 505         /*
 506          * Find the next leaf page by incrementing the frame pointer.
 507          * If we run out of entries in the interior node, loop around and
 508          * increment pointer in the parent node.  When we break out of
 509          * this loop, num_frames indicates the number of interior
 510          * nodes need to be read.
 511          */
 512         while (1) {
 513                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 514                         break;
 515                 if (p == frames)
 516                         return 0;
 517                 num_frames++;
 518                 p--;
 519         }
 520
 521         /*
 522          * If the hash is 1, then continue only if the next page has a
 523          * continuation hash of any value.  This is used for readdir
 524          * handling.  Otherwise, check to see if the hash matches the
 525          * desired contiuation hash.  If it doesn't, return since
 526          * there's no point to read in the successive index pages.
 527          */
 528         bhash = dx_get_hash(p->at);
 529         if (start_hash)
 530                 *start_hash = bhash;
 531         if ((hash & 1) == 0) {
 532                 if ((bhash & ~1) != hash)
 533                         return 0;
 534         }
 535         /*
 536          * If the hash is HASH_NB_ALWAYS, we always go to the next
 537          * block so no check is necessary
 538          */
 539         while (num_frames--) {
 540                 if (!(bh = ext3_dir_bread(NULL, dir, dx_get_block(p->at),
 541                                           0, &err)))
 542                         return err; /* Failure */
 543                 p++;
 544                 brelse (p->bh);
 545                 p->bh = bh;
 546                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 547         }
 548         return 1;
 549 }
 550
 551
 552 /*
 553  * This function fills a red-black tree with information from a
 554  * directory block.  It returns the number directory entries loaded
 555  * into the tree.  If there is an error it is returned in err.
 556  */
 557 static int htree_dirblock_to_tree(struct file *dir_file,
 558                                   struct inode *dir, int block,
 559                                   struct dx_hash_info *hinfo,
 560                                   __u32 start_hash, __u32 start_minor_hash)
 561 {
 562         struct buffer_head *bh;
 563         struct ext3_dir_entry_2 *de, *top;
 564         int err = 0, count = 0;
 565
 566         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
 567
 568         if (!(bh = ext3_dir_bread(NULL, dir, block, 0, &err)))
 569                 return err;
 570
 571         de = (struct ext3_dir_entry_2 *) bh->b_data;
 572         top = (struct ext3_dir_entry_2 *) ((char *) de +
 573                                            dir->i_sb->s_blocksize -
 574                                            EXT3_DIR_REC_LEN(0));
 575         for (; de < top; de = ext3_next_entry(de)) {
 576                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 577                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
 578                                                 +((char *)de - bh->b_data))) {
 579                         /* On error, skip the f_pos to the next block. */
 580                         dir_file->f_pos = (dir_file->f_pos |
 581                                         (dir->i_sb->s_blocksize - 1)) + 1;
 582                         brelse (bh);
 583                         return count;
 584                 }
 585                 ext3fs_dirhash(de->name, de->name_len, hinfo);
 586                 if ((hinfo->hash < start_hash) ||
 587                     ((hinfo->hash == start_hash) &&
 588                      (hinfo->minor_hash < start_minor_hash)))
 589                         continue;
 590                 if (de->inode == 0)
 591                         continue;
 592                 if ((err = ext3_htree_store_dirent(dir_file,
 593                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 594                         brelse(bh);
 595                         return err;
 596                 }
 597                 count++;
 598         }
 599         brelse(bh);
 600         return count;
 601 }
 602
 603
 604 /*
 605  * This function fills a red-black tree with information from a
 606  * directory.  We start scanning the directory in hash order, starting
 607  * at start_hash and start_minor_hash.
 608  *
 609  * This function returns the number of entries inserted into the tree,
 610  * or a negative error code.
 611  */
 612 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 613                          __u32 start_minor_hash, __u32 *next_hash)
 614 {
 615         struct dx_hash_info hinfo;
 616         struct ext3_dir_entry_2 *de;
 617         struct dx_frame frames[2], *frame;
 618         struct inode *dir;
 619         int block, err;
 620         int count = 0;
 621         int ret;
 622         __u32 hashval;
 623
 624         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
 625                        start_minor_hash));
 626         dir = file_inode(dir_file);
 627         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
 628                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
 629                 if (hinfo.hash_version <= DX_HASH_TEA)
 630                         hinfo.hash_version +=
 631                                 EXT3_SB(dir->i_sb)->s_hash_unsigned;
 632                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 633                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 634                                                start_hash, start_minor_hash);
 635                 *next_hash = ~0;
 636                 return count;
 637         }
 638         hinfo.hash = start_hash;
 639         hinfo.minor_hash = 0;
 640         frame = dx_probe(NULL, file_inode(dir_file), &hinfo, frames, &err);
 641         if (!frame)
 642                 return err;
 643
 644         /* Add '.' and '..' from the htree header */
 645         if (!start_hash && !start_minor_hash) {
 646                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 647                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 648                         goto errout;
 649                 count++;
 650         }
 651         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 652                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 653                 de = ext3_next_entry(de);
 654                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 655                         goto errout;
 656                 count++;
 657         }
 658
 659         while (1) {
 660                 block = dx_get_block(frame->at);
 661                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 662                                              start_hash, start_minor_hash);
 663                 if (ret < 0) {
 664                         err = ret;
 665                         goto errout;
 666                 }
 667                 count += ret;
 668                 hashval = ~0;
 669                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 670                                             frame, frames, &hashval);
 671                 *next_hash = hashval;
 672                 if (ret < 0) {
 673                         err = ret;
 674                         goto errout;
 675                 }
 676                 /*
 677                  * Stop if:  (a) there are no more entries, or
 678                  * (b) we have inserted at least one entry and the
 679                  * next hash value is not a continuation
 680                  */
 681                 if ((ret == 0) ||
 682                     (count && ((hashval & 1) == 0)))
 683                         break;
 684         }
 685         dx_release(frames);
 686         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 687                        count, *next_hash));
 688         return count;
 689 errout:
 690         dx_release(frames);
 691         return (err);
 692 }
 693
 694
 695 /*
 696  * Directory block splitting, compacting
 697  */
 698
 699 /*
 700  * Create map of hash values, offsets, and sizes, stored at end of block.
 701  * Returns number of entries mapped.
 702  */
 703 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
 704                 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 705 {
 706         int count = 0;
 707         char *base = (char *) de;
 708         struct dx_hash_info h = *hinfo;
 709
 710         while ((char *) de < base + blocksize)
 711         {
 712                 if (de->name_len && de->inode) {
 713                         ext3fs_dirhash(de->name, de->name_len, &h);
 714                         map_tail--;
 715                         map_tail->hash = h.hash;
 716                         map_tail->offs = (u16) ((char *) de - base);
 717                         map_tail->size = le16_to_cpu(de->rec_len);
 718                         count++;
 719                         cond_resched();
 720                 }
 721                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 722                 de = ext3_next_entry(de);
 723         }
 724         return count;
 725 }
 726
 727 /* Sort map by hash value */
 728 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 729 {
 730         struct dx_map_entry *p, *q, *top = map + count - 1;
 731         int more;
 732         /* Combsort until bubble sort doesn't suck */
 733         while (count > 2)
 734         {
 735                 count = count*10/13;
 736                 if (count - 9 < 2) /* 9, 10 -> 11 */
 737                         count = 11;
 738                 for (p = top, q = p - count; q >= map; p--, q--)
 739                         if (p->hash < q->hash)
 740                                 swap(*p, *q);
 741         }
 742         /* Garden variety bubble sort */
 743         do {
 744                 more = 0;
 745                 q = top;
 746                 while (q-- > map)
 747                 {
 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
 770 static void ext3_update_dx_flag(struct inode *inode)
 771 {
 772         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 773                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 774                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 775 }
 776
 777 /*
 778  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 779  *
 780  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 781  * `de != NULL' is guaranteed by caller.
 782  */
 783 static inline int ext3_match (int len, const char * const name,
 784                               struct ext3_dir_entry_2 * de)
 785 {
 786         if (len != de->name_len)
 787                 return 0;
 788         if (!de->inode)
 789                 return 0;
 790         return !memcmp(name, de->name, len);
 791 }
 792
 793 /*
 794  * Returns 0 if not found, -1 on failure, and 1 on success
 795  */
 796 static inline int search_dirblock(struct buffer_head * bh,
 797                                   struct inode *dir,
 798                                   struct qstr *child,
 799                                   unsigned long offset,
 800                                   struct ext3_dir_entry_2 ** res_dir)
 801 {
 802         struct ext3_dir_entry_2 * de;
 803         char * dlimit;
 804         int de_len;
 805         const char *name = child->name;
 806         int namelen = child->len;
 807
 808         de = (struct ext3_dir_entry_2 *) bh->b_data;
 809         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 810         while ((char *) de < dlimit) {
 811                 /* this code is executed quadratically often */
 812                 /* do minimal checking `by hand' */
 813
 814                 if ((char *) de + namelen <= dlimit &&
 815                     ext3_match (namelen, name, de)) {
 816                         /* found a match - just to be sure, do a full check */
 817                         if (!ext3_check_dir_entry("ext3_find_entry",
 818                                                   dir, de, bh, offset))
 819                                 return -1;
 820                         *res_dir = de;
 821                         return 1;
 822                 }
 823                 /* prevent looping on a bad block */
 824                 de_len = ext3_rec_len_from_disk(de->rec_len);
 825                 if (de_len <= 0)
 826                         return -1;
 827                 offset += de_len;
 828                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 829         }
 830         return 0;
 831 }
 832
 833
 834 /*
 835  *      ext3_find_entry()
 836  *
 837  * finds an entry in the specified directory with the wanted name. It
 838  * returns the cache buffer in which the entry was found, and the entry
 839  * itself (as a parameter - res_dir). It does NOT read the inode of the
 840  * entry - you'll have to do that yourself if you want to.
 841  *
 842  * The returned buffer_head has ->b_count elevated.  The caller is expected
 843  * to brelse() it when appropriate.
 844  */
 845 static struct buffer_head *ext3_find_entry(struct inode *dir,
 846                                         struct qstr *entry,
 847                                         struct ext3_dir_entry_2 **res_dir)
 848 {
 849         struct super_block * sb;
 850         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 851         struct buffer_head * bh, *ret = NULL;
 852         unsigned long start, block, b;
 853         const u8 *name = entry->name;
 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         int namelen;
 861
 862         *res_dir = NULL;
 863         sb = dir->i_sb;
 864         namelen = entry->len;
 865         if (namelen > EXT3_NAME_LEN)
 866                 return NULL;
 867         if ((namelen <= 2) && (name[0] == '.') &&
 868             (name[1] == '.' || name[1] == 0)) {
 869                 /*
 870                  * "." or ".." will only be in the first block
 871                  * NFS may look up ".."; "." should be handled by the VFS
 872                  */
 873                 block = start = 0;
 874                 nblocks = 1;
 875                 goto restart;
 876         }
 877         if (is_dx(dir)) {
 878                 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
 879                 /*
 880                  * On success, or if the error was file not found,
 881                  * return.  Otherwise, fall back to doing a search the
 882                  * old fashioned way.
 883                  */
 884                 if (bh || (err != ERR_BAD_DX_DIR))
 885                         return bh;
 886                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 887         }
 888         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 889         start = EXT3_I(dir)->i_dir_start_lookup;
 890         if (start >= nblocks)
 891                 start = 0;
 892         block = start;
 893 restart:
 894         do {
 895                 /*
 896                  * We deal with the read-ahead logic here.
 897                  */
 898                 if (ra_ptr >= ra_max) {
 899                         /* Refill the readahead buffer */
 900                         ra_ptr = 0;
 901                         b = block;
 902                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 903                                 /*
 904                                  * Terminate if we reach the end of the
 905                                  * directory and must wrap, or if our
 906                                  * search has finished at this block.
 907                                  */
 908                                 if (b >= nblocks || (num && block == start)) {
 909                                         bh_use[ra_max] = NULL;
 910                                         break;
 911                                 }
 912                                 num++;
 913                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 914                                 bh_use[ra_max] = bh;
 915                                 if (bh && !bh_uptodate_or_lock(bh)) {
 916                                         get_bh(bh);
 917                                         bh->b_end_io = end_buffer_read_sync;
 918                                         submit_bh(READ | REQ_META | REQ_PRIO,
 919                                                   bh);
 920                                 }
 921                         }
 922                 }
 923                 if ((bh = bh_use[ra_ptr++]) == NULL)
 924                         goto next;
 925                 wait_on_buffer(bh);
 926                 if (!buffer_uptodate(bh)) {
 927                         /* read error, skip block & hope for the best */
 928                         ext3_error(sb, __func__, "reading directory #%lu "
 929                                    "offset %lu", dir->i_ino, block);
 930                         brelse(bh);
 931                         goto next;
 932                 }
 933                 i = search_dirblock(bh, dir, entry,
 934                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 935                 if (i == 1) {
 936                         EXT3_I(dir)->i_dir_start_lookup = block;
 937                         ret = bh;
 938                         goto cleanup_and_exit;
 939                 } else {
 940                         brelse(bh);
 941                         if (i < 0)
 942                                 goto cleanup_and_exit;
 943                 }
 944         next:
 945                 if (++block >= nblocks)
 946                         block = 0;
 947         } while (block != start);
 948
 949         /*
 950          * If the directory has grown while we were searching, then
 951          * search the last part of the directory before giving up.
 952          */
 953         block = nblocks;
 954         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 955         if (block < nblocks) {
 956                 start = 0;
 957                 goto restart;
 958         }
 959
 960 cleanup_and_exit:
 961         /* Clean up the read-ahead blocks */
 962         for (; ra_ptr < ra_max; ra_ptr++)
 963                 brelse (bh_use[ra_ptr]);
 964         return ret;
 965 }
 966
 967 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
 968                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
 969                         int *err)
 970 {
 971         struct super_block *sb = dir->i_sb;
 972         struct dx_hash_info     hinfo;
 973         struct dx_frame frames[2], *frame;
 974         struct buffer_head *bh;
 975         unsigned long block;
 976         int retval;
 977
 978         if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
 979                 return NULL;
 980         do {
 981                 block = dx_get_block(frame->at);
 982                 if (!(bh = ext3_dir_bread (NULL, dir, block, 0, err)))
 983                         goto errout;
 984
 985                 retval = search_dirblock(bh, dir, entry,
 986                                          block << EXT3_BLOCK_SIZE_BITS(sb),
 987                                          res_dir);
 988                 if (retval == 1) {
 989                         dx_release(frames);
 990                         return bh;
 991                 }
 992                 brelse(bh);
 993                 if (retval == -1) {
 994                         *err = ERR_BAD_DX_DIR;
 995                         goto errout;
 996                 }
 997
 998                 /* Check to see if we should continue to search */
 999                 retval = ext3_htree_next_block(dir, hinfo.hash, frame,
1000                                                frames, NULL);
1001                 if (retval < 0) {
1002                         ext3_warning(sb, __func__,
1003                              "error reading index page in directory #%lu",
1004                              dir->i_ino);
1005                         *err = retval;
1006                         goto errout;
1007                 }
1008         } while (retval == 1);
1009
1010         *err = -ENOENT;
1011 errout:
1012         dxtrace(printk("%s not found\n", entry->name));
1013         dx_release (frames);
1014         return NULL;
1015 }
1016
1017 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, unsigned int flags)
1018 {
1019         struct inode * inode;
1020         struct ext3_dir_entry_2 * de;
1021         struct buffer_head * bh;
1022
1023         if (dentry->d_name.len > EXT3_NAME_LEN)
1024                 return ERR_PTR(-ENAMETOOLONG);
1025
1026         bh = ext3_find_entry(dir, &dentry->d_name, &de);
1027         inode = NULL;
1028         if (bh) {
1029                 unsigned long ino = le32_to_cpu(de->inode);
1030                 brelse (bh);
1031                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1032                         ext3_error(dir->i_sb, "ext3_lookup",
1033                                    "bad inode number: %lu", ino);
1034                         return ERR_PTR(-EIO);
1035                 }
1036                 inode = ext3_iget(dir->i_sb, ino);
1037                 if (inode == ERR_PTR(-ESTALE)) {
1038                         ext3_error(dir->i_sb, __func__,
1039                                         "deleted inode referenced: %lu",
1040                                         ino);
1041                         return ERR_PTR(-EIO);
1042                 }
1043         }
1044         return d_splice_alias(inode, dentry);
1045 }
1046
1047
1048 struct dentry *ext3_get_parent(struct dentry *child)
1049 {
1050         unsigned long ino;
1051         struct qstr dotdot = QSTR_INIT("..", 2);
1052         struct ext3_dir_entry_2 * de;
1053         struct buffer_head *bh;
1054
1055         bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1056         if (!bh)
1057                 return ERR_PTR(-ENOENT);
1058         ino = le32_to_cpu(de->inode);
1059         brelse(bh);
1060
1061         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1062                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1063                            "bad inode number: %lu", ino);
1064                 return ERR_PTR(-EIO);
1065         }
1066
1067         return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1068 }
1069
1070 #define S_SHIFT 12
1071 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1072         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1073         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1074         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1075         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1076         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1077         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1078         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1079 };
1080
1081 static inline void ext3_set_de_type(struct super_block *sb,
1082                                 struct ext3_dir_entry_2 *de,
1083                                 umode_t mode) {
1084         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1085                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1086 }
1087
1088 /*
1089  * Move count entries from end of map between two memory locations.
1090  * Returns pointer to last entry moved.
1091  */
1092 static struct ext3_dir_entry_2 *
1093 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1094 {
1095         unsigned rec_len = 0;
1096
1097         while (count--) {
1098                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1099                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1100                 memcpy (to, de, rec_len);
1101                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1102                                 ext3_rec_len_to_disk(rec_len);
1103                 de->inode = 0;
1104                 map++;
1105                 to += rec_len;
1106         }
1107         return (struct ext3_dir_entry_2 *) (to - rec_len);
1108 }
1109
1110 /*
1111  * Compact each dir entry in the range to the minimal rec_len.
1112  * Returns pointer to last entry in range.
1113  */
1114 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1115 {
1116         struct ext3_dir_entry_2 *next, *to, *prev;
1117         struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1118         unsigned rec_len = 0;
1119
1120         prev = to = de;
1121         while ((char *)de < base + blocksize) {
1122                 next = ext3_next_entry(de);
1123                 if (de->inode && de->name_len) {
1124                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1125                         if (de > to)
1126                                 memmove(to, de, rec_len);
1127                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1128                         prev = to;
1129                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1130                 }
1131                 de = next;
1132         }
1133         return prev;
1134 }
1135
1136 /*
1137  * Split a full leaf block to make room for a new dir entry.
1138  * Allocate a new block, and move entries so that they are approx. equally full.
1139  * Returns pointer to de in block into which the new entry will be inserted.
1140  */
1141 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1142                         struct buffer_head **bh,struct dx_frame *frame,
1143                         struct dx_hash_info *hinfo, int *error)
1144 {
1145         unsigned blocksize = dir->i_sb->s_blocksize;
1146         unsigned count, continued;
1147         struct buffer_head *bh2;
1148         u32 newblock;
1149         u32 hash2;
1150         struct dx_map_entry *map;
1151         char *data1 = (*bh)->b_data, *data2;
1152         unsigned split, move, size;
1153         struct ext3_dir_entry_2 *de = NULL, *de2;
1154         int     err = 0, i;
1155
1156         bh2 = ext3_append (handle, dir, &newblock, &err);
1157         if (!(bh2)) {
1158                 brelse(*bh);
1159                 *bh = NULL;
1160                 goto errout;
1161         }
1162
1163         BUFFER_TRACE(*bh, "get_write_access");
1164         err = ext3_journal_get_write_access(handle, *bh);
1165         if (err)
1166                 goto journal_error;
1167
1168         BUFFER_TRACE(frame->bh, "get_write_access");
1169         err = ext3_journal_get_write_access(handle, frame->bh);
1170         if (err)
1171                 goto journal_error;
1172
1173         data2 = bh2->b_data;
1174
1175         /* create map in the end of data2 block */
1176         map = (struct dx_map_entry *) (data2 + blocksize);
1177         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1178                              blocksize, hinfo, map);
1179         map -= count;
1180         dx_sort_map (map, count);
1181         /* Split the existing block in the middle, size-wise */
1182         size = 0;
1183         move = 0;
1184         for (i = count-1; i >= 0; i--) {
1185                 /* is more than half of this entry in 2nd half of the block? */
1186                 if (size + map[i].size/2 > blocksize/2)
1187                         break;
1188                 size += map[i].size;
1189                 move++;
1190         }
1191         /* map index at which we will split */
1192         split = count - move;
1193         hash2 = map[split].hash;
1194         continued = hash2 == map[split - 1].hash;
1195         dxtrace(printk("Split block %i at %x, %i/%i\n",
1196                 dx_get_block(frame->at), hash2, split, count-split));
1197
1198         /* Fancy dance to stay within two buffers */
1199         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1200         de = dx_pack_dirents(data1,blocksize);
1201         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1202         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1203         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1204         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1205
1206         /* Which block gets the new entry? */
1207         if (hinfo->hash >= hash2)
1208         {
1209                 swap(*bh, bh2);
1210                 de = de2;
1211         }
1212         dx_insert_block (frame, hash2 + continued, newblock);
1213         err = ext3_journal_dirty_metadata (handle, bh2);
1214         if (err)
1215                 goto journal_error;
1216         err = ext3_journal_dirty_metadata (handle, frame->bh);
1217         if (err)
1218                 goto journal_error;
1219         brelse (bh2);
1220         dxtrace(dx_show_index ("frame", frame->entries));
1221         return de;
1222
1223 journal_error:
1224         brelse(*bh);
1225         brelse(bh2);
1226         *bh = NULL;
1227         ext3_std_error(dir->i_sb, err);
1228 errout:
1229         *error = err;
1230         return NULL;
1231 }
1232
1233
1234 /*
1235  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1236  * it points to a directory entry which is guaranteed to be large
1237  * enough for new directory entry.  If de is NULL, then
1238  * add_dirent_to_buf will attempt search the directory block for
1239  * space.  It will return -ENOSPC if no space is available, and -EIO
1240  * and -EEXIST if directory entry already exists.
1241  *
1242  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1243  * all other cases bh is released.
1244  */
1245 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1246                              struct inode *inode, struct ext3_dir_entry_2 *de,
1247                              struct buffer_head * bh)
1248 {
1249         struct inode    *dir = dentry->d_parent->d_inode;
1250         const char      *name = dentry->d_name.name;
1251         int             namelen = dentry->d_name.len;
1252         unsigned long   offset = 0;
1253         unsigned short  reclen;
1254         int             nlen, rlen, err;
1255         char            *top;
1256
1257         reclen = EXT3_DIR_REC_LEN(namelen);
1258         if (!de) {
1259                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1260                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1261                 while ((char *) de <= top) {
1262                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1263                                                   bh, offset)) {
1264                                 brelse (bh);
1265                                 return -EIO;
1266                         }
1267                         if (ext3_match (namelen, name, de)) {
1268                                 brelse (bh);
1269                                 return -EEXIST;
1270                         }
1271                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1272                         rlen = ext3_rec_len_from_disk(de->rec_len);
1273                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1274                                 break;
1275                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1276                         offset += rlen;
1277                 }
1278                 if ((char *) de > top)
1279                         return -ENOSPC;
1280         }
1281         BUFFER_TRACE(bh, "get_write_access");
1282         err = ext3_journal_get_write_access(handle, bh);
1283         if (err) {
1284                 ext3_std_error(dir->i_sb, err);
1285                 brelse(bh);
1286                 return err;
1287         }
1288
1289         /* By now the buffer is marked for journaling */
1290         nlen = EXT3_DIR_REC_LEN(de->name_len);
1291         rlen = ext3_rec_len_from_disk(de->rec_len);
1292         if (de->inode) {
1293                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1294                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1295                 de->rec_len = ext3_rec_len_to_disk(nlen);
1296                 de = de1;
1297         }
1298         de->file_type = EXT3_FT_UNKNOWN;
1299         if (inode) {
1300                 de->inode = cpu_to_le32(inode->i_ino);
1301                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1302         } else
1303                 de->inode = 0;
1304         de->name_len = namelen;
1305         memcpy (de->name, name, namelen);
1306         /*
1307          * XXX shouldn't update any times until successful
1308          * completion of syscall, but too many callers depend
1309          * on this.
1310          *
1311          * XXX similarly, too many callers depend on
1312          * ext3_new_inode() setting the times, but error
1313          * recovery deletes the inode, so the worst that can
1314          * happen is that the times are slightly out of date
1315          * and/or different from the directory change time.
1316          */
1317         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1318         ext3_update_dx_flag(dir);
1319         dir->i_version++;
1320         ext3_mark_inode_dirty(handle, dir);
1321         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1322         err = ext3_journal_dirty_metadata(handle, bh);
1323         if (err)
1324                 ext3_std_error(dir->i_sb, err);
1325         brelse(bh);
1326         return 0;
1327 }
1328
1329 /*
1330  * This converts a one block unindexed directory to a 3 block indexed
1331  * directory, and adds the dentry to the indexed directory.
1332  */
1333 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1334                             struct inode *inode, struct buffer_head *bh)
1335 {
1336         struct inode    *dir = dentry->d_parent->d_inode;
1337         const char      *name = dentry->d_name.name;
1338         int             namelen = dentry->d_name.len;
1339         struct buffer_head *bh2;
1340         struct dx_root  *root;
1341         struct dx_frame frames[2], *frame;
1342         struct dx_entry *entries;
1343         struct ext3_dir_entry_2 *de, *de2;
1344         char            *data1, *top;
1345         unsigned        len;
1346         int             retval;
1347         unsigned        blocksize;
1348         struct dx_hash_info hinfo;
1349         u32             block;
1350         struct fake_dirent *fde;
1351
1352         blocksize =  dir->i_sb->s_blocksize;
1353         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1354         retval = ext3_journal_get_write_access(handle, bh);
1355         if (retval) {
1356                 ext3_std_error(dir->i_sb, retval);
1357                 brelse(bh);
1358                 return retval;
1359         }
1360         root = (struct dx_root *) bh->b_data;
1361
1362         /* The 0th block becomes the root, move the dirents out */
1363         fde = &root->dotdot;
1364         de = (struct ext3_dir_entry_2 *)((char *)fde +
1365                         ext3_rec_len_from_disk(fde->rec_len));
1366         if ((char *) de >= (((char *) root) + blocksize)) {
1367                 ext3_error(dir->i_sb, __func__,
1368                            "invalid rec_len for '..' in inode %lu",
1369                            dir->i_ino);
1370                 brelse(bh);
1371                 return -EIO;
1372         }
1373         len = ((char *) root) + blocksize - (char *) de;
1374
1375         bh2 = ext3_append (handle, dir, &block, &retval);
1376         if (!(bh2)) {
1377                 brelse(bh);
1378                 return retval;
1379         }
1380         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1381         data1 = bh2->b_data;
1382
1383         memcpy (data1, de, len);
1384         de = (struct ext3_dir_entry_2 *) data1;
1385         top = data1 + len;
1386         while ((char *)(de2 = ext3_next_entry(de)) < top)
1387                 de = de2;
1388         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1389         /* Initialize the root; the dot dirents already exist */
1390         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1391         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1392         memset (&root->info, 0, sizeof(root->info));
1393         root->info.info_length = sizeof(root->info);
1394         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1395         entries = root->entries;
1396         dx_set_block (entries, 1);
1397         dx_set_count (entries, 1);
1398         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1399
1400         /* Initialize as for dx_probe */
1401         hinfo.hash_version = root->info.hash_version;
1402         if (hinfo.hash_version <= DX_HASH_TEA)
1403                 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1404         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1405         ext3fs_dirhash(name, namelen, &hinfo);
1406         frame = frames;
1407         frame->entries = entries;
1408         frame->at = entries;
1409         frame->bh = bh;
1410         bh = bh2;
1411         /*
1412          * Mark buffers dirty here so that if do_split() fails we write a
1413          * consistent set of buffers to disk.
1414          */
1415         ext3_journal_dirty_metadata(handle, frame->bh);
1416         ext3_journal_dirty_metadata(handle, bh);
1417         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1418         if (!de) {
1419                 ext3_mark_inode_dirty(handle, dir);
1420                 dx_release(frames);
1421                 return retval;
1422         }
1423         dx_release(frames);
1424
1425         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1426 }
1427
1428 /*
1429  *      ext3_add_entry()
1430  *
1431  * adds a file entry to the specified directory, using the same
1432  * semantics as ext3_find_entry(). It returns NULL if it failed.
1433  *
1434  * NOTE!! The inode part of 'de' is left at 0 - which means you
1435  * may not sleep between calling this and putting something into
1436  * the entry, as someone else might have used it while you slept.
1437  */
1438 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1439         struct inode *inode)
1440 {
1441         struct inode *dir = dentry->d_parent->d_inode;
1442         struct buffer_head * bh;
1443         struct ext3_dir_entry_2 *de;
1444         struct super_block * sb;
1445         int     retval;
1446         int     dx_fallback=0;
1447         unsigned blocksize;
1448         u32 block, blocks;
1449
1450         sb = dir->i_sb;
1451         blocksize = sb->s_blocksize;
1452         if (!dentry->d_name.len)
1453                 return -EINVAL;
1454         if (is_dx(dir)) {
1455                 retval = ext3_dx_add_entry(handle, dentry, inode);
1456                 if (!retval || (retval != ERR_BAD_DX_DIR))
1457                         return retval;
1458                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1459                 dx_fallback++;
1460                 ext3_mark_inode_dirty(handle, dir);
1461         }
1462         blocks = dir->i_size >> sb->s_blocksize_bits;
1463         for (block = 0; block < blocks; block++) {
1464                 if (!(bh = ext3_dir_bread(handle, dir, block, 0, &retval)))
1465                         return retval;
1466
1467                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1468                 if (retval != -ENOSPC)
1469                         return retval;
1470
1471                 if (blocks == 1 && !dx_fallback &&
1472                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1473                         return make_indexed_dir(handle, dentry, inode, bh);
1474                 brelse(bh);
1475         }
1476         bh = ext3_append(handle, dir, &block, &retval);
1477         if (!bh)
1478                 return retval;
1479         de = (struct ext3_dir_entry_2 *) bh->b_data;
1480         de->inode = 0;
1481         de->rec_len = ext3_rec_len_to_disk(blocksize);
1482         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1483 }
1484
1485 /*
1486  * Returns 0 for success, or a negative error value
1487  */
1488 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1489                              struct inode *inode)
1490 {
1491         struct dx_frame frames[2], *frame;
1492         struct dx_entry *entries, *at;
1493         struct dx_hash_info hinfo;
1494         struct buffer_head * bh;
1495         struct inode *dir = dentry->d_parent->d_inode;
1496         struct super_block * sb = dir->i_sb;
1497         struct ext3_dir_entry_2 *de;
1498         int err;
1499
1500         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1501         if (!frame)
1502                 return err;
1503         entries = frame->entries;
1504         at = frame->at;
1505
1506         if (!(bh = ext3_dir_bread(handle, dir, dx_get_block(frame->at), 0, &err)))
1507                 goto cleanup;
1508
1509         BUFFER_TRACE(bh, "get_write_access");
1510         err = ext3_journal_get_write_access(handle, bh);
1511         if (err)
1512                 goto journal_error;
1513
1514         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1515         if (err != -ENOSPC) {
1516                 bh = NULL;
1517                 goto cleanup;
1518         }
1519
1520         /* Block full, should compress but for now just split */
1521         dxtrace(printk("using %u of %u node entries\n",
1522                        dx_get_count(entries), dx_get_limit(entries)));
1523         /* Need to split index? */
1524         if (dx_get_count(entries) == dx_get_limit(entries)) {
1525                 u32 newblock;
1526                 unsigned icount = dx_get_count(entries);
1527                 int levels = frame - frames;
1528                 struct dx_entry *entries2;
1529                 struct dx_node *node2;
1530                 struct buffer_head *bh2;
1531
1532                 if (levels && (dx_get_count(frames->entries) ==
1533                                dx_get_limit(frames->entries))) {
1534                         ext3_warning(sb, __func__,
1535                                      "Directory index full!");
1536                         err = -ENOSPC;
1537                         goto cleanup;
1538                 }
1539                 bh2 = ext3_append (handle, dir, &newblock, &err);
1540                 if (!(bh2))
1541                         goto cleanup;
1542                 node2 = (struct dx_node *)(bh2->b_data);
1543                 entries2 = node2->entries;
1544                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1545                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1546                 BUFFER_TRACE(frame->bh, "get_write_access");
1547                 err = ext3_journal_get_write_access(handle, frame->bh);
1548                 if (err)
1549                         goto journal_error;
1550                 if (levels) {
1551                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1552                         unsigned hash2 = dx_get_hash(entries + icount1);
1553                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1554
1555                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1556                         err = ext3_journal_get_write_access(handle,
1557                                                              frames[0].bh);
1558                         if (err)
1559                                 goto journal_error;
1560
1561                         memcpy ((char *) entries2, (char *) (entries + icount1),
1562                                 icount2 * sizeof(struct dx_entry));
1563                         dx_set_count (entries, icount1);
1564                         dx_set_count (entries2, icount2);
1565                         dx_set_limit (entries2, dx_node_limit(dir));
1566
1567                         /* Which index block gets the new entry? */
1568                         if (at - entries >= icount1) {
1569                                 frame->at = at = at - entries - icount1 + entries2;
1570                                 frame->entries = entries = entries2;
1571                                 swap(frame->bh, bh2);
1572                         }
1573                         dx_insert_block (frames + 0, hash2, newblock);
1574                         dxtrace(dx_show_index ("node", frames[1].entries));
1575                         dxtrace(dx_show_index ("node",
1576                                ((struct dx_node *) bh2->b_data)->entries));
1577                         err = ext3_journal_dirty_metadata(handle, bh2);
1578                         if (err)
1579                                 goto journal_error;
1580                         brelse (bh2);
1581                 } else {
1582                         dxtrace(printk("Creating second level index...\n"));
1583                         memcpy((char *) entries2, (char *) entries,
1584                                icount * sizeof(struct dx_entry));
1585                         dx_set_limit(entries2, dx_node_limit(dir));
1586
1587                         /* Set up root */
1588                         dx_set_count(entries, 1);
1589                         dx_set_block(entries + 0, newblock);
1590                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1591
1592                         /* Add new access path frame */
1593                         frame = frames + 1;
1594                         frame->at = at = at - entries + entries2;
1595                         frame->entries = entries = entries2;
1596                         frame->bh = bh2;
1597                         err = ext3_journal_get_write_access(handle,
1598                                                              frame->bh);
1599                         if (err)
1600                                 goto journal_error;
1601                 }
1602                 err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1603                 if (err)
1604                         goto journal_error;
1605         }
1606         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1607         if (!de)
1608                 goto cleanup;
1609         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1610         bh = NULL;
1611         goto cleanup;
1612
1613 journal_error:
1614         ext3_std_error(dir->i_sb, err);
1615 cleanup:
1616         if (bh)
1617                 brelse(bh);
1618         dx_release(frames);
1619         return err;
1620 }
1621
1622 /*
1623  * ext3_delete_entry deletes a directory entry by merging it with the
1624  * previous entry
1625  */
1626 static int ext3_delete_entry (handle_t *handle,
1627                               struct inode * dir,
1628                               struct ext3_dir_entry_2 * de_del,
1629                               struct buffer_head * bh)
1630 {
1631         struct ext3_dir_entry_2 * de, * pde;
1632         int i;
1633
1634         i = 0;
1635         pde = NULL;
1636         de = (struct ext3_dir_entry_2 *) bh->b_data;
1637         while (i < bh->b_size) {
1638                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1639                         return -EIO;
1640                 if (de == de_del)  {
1641                         int err;
1642
1643                         BUFFER_TRACE(bh, "get_write_access");
1644                         err = ext3_journal_get_write_access(handle, bh);
1645                         if (err)
1646                                 goto journal_error;
1647
1648                         if (pde)
1649                                 pde->rec_len = ext3_rec_len_to_disk(
1650                                         ext3_rec_len_from_disk(pde->rec_len) +
1651                                         ext3_rec_len_from_disk(de->rec_len));
1652                         else
1653                                 de->inode = 0;
1654                         dir->i_version++;
1655                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1656                         err = ext3_journal_dirty_metadata(handle, bh);
1657                         if (err) {
1658 journal_error:
1659                                 ext3_std_error(dir->i_sb, err);
1660                                 return err;
1661                         }
1662                         return 0;
1663                 }
1664                 i += ext3_rec_len_from_disk(de->rec_len);
1665                 pde = de;
1666                 de = ext3_next_entry(de);
1667         }
1668         return -ENOENT;
1669 }
1670
1671 static int ext3_add_nondir(handle_t *handle,
1672                 struct dentry *dentry, struct inode *inode)
1673 {
1674         int err = ext3_add_entry(handle, dentry, inode);
1675         if (!err) {
1676                 ext3_mark_inode_dirty(handle, inode);
1677                 unlock_new_inode(inode);
1678                 d_instantiate(dentry, inode);
1679                 return 0;
1680         }
1681         drop_nlink(inode);
1682         unlock_new_inode(inode);
1683         iput(inode);
1684         return err;
1685 }
1686
1687 /*
1688  * By the time this is called, we already have created
1689  * the directory cache entry for the new file, but it
1690  * is so far negative - it has no inode.
1691  *
1692  * If the create succeeds, we fill in the inode information
1693  * with d_instantiate().
1694  */
1695 static int ext3_create (struct inode * dir, struct dentry * dentry, umode_t mode,
1696                 bool excl)
1697 {
1698         handle_t *handle;
1699         struct inode * inode;
1700         int err, retries = 0;
1701
1702         dquot_initialize(dir);
1703
1704 retry:
1705         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1706                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1707                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1708         if (IS_ERR(handle))
1709                 return PTR_ERR(handle);
1710
1711         if (IS_DIRSYNC(dir))
1712                 handle->h_sync = 1;
1713
1714         inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1715         err = PTR_ERR(inode);
1716         if (!IS_ERR(inode)) {
1717                 inode->i_op = &ext3_file_inode_operations;
1718                 inode->i_fop = &ext3_file_operations;
1719                 ext3_set_aops(inode);
1720                 err = ext3_add_nondir(handle, dentry, inode);
1721         }
1722         ext3_journal_stop(handle);
1723         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1724                 goto retry;
1725         return err;
1726 }
1727
1728 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1729                         umode_t mode, dev_t rdev)
1730 {
1731         handle_t *handle;
1732         struct inode *inode;
1733         int err, retries = 0;
1734
1735         if (!new_valid_dev(rdev))
1736                 return -EINVAL;
1737
1738         dquot_initialize(dir);
1739
1740 retry:
1741         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1742                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1743                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1744         if (IS_ERR(handle))
1745                 return PTR_ERR(handle);
1746
1747         if (IS_DIRSYNC(dir))
1748                 handle->h_sync = 1;
1749
1750         inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1751         err = PTR_ERR(inode);
1752         if (!IS_ERR(inode)) {
1753                 init_special_inode(inode, inode->i_mode, rdev);
1754 #ifdef CONFIG_EXT3_FS_XATTR
1755                 inode->i_op = &ext3_special_inode_operations;
1756 #endif
1757                 err = ext3_add_nondir(handle, dentry, inode);
1758         }
1759         ext3_journal_stop(handle);
1760         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1761                 goto retry;
1762         return err;
1763 }
1764
1765 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, umode_t mode)
1766 {
1767         handle_t *handle;
1768         struct inode * inode;
1769         struct buffer_head * dir_block = NULL;
1770         struct ext3_dir_entry_2 * de;
1771         int err, retries = 0;
1772
1773         if (dir->i_nlink >= EXT3_LINK_MAX)
1774                 return -EMLINK;
1775
1776         dquot_initialize(dir);
1777
1778 retry:
1779         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1780                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1781                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1782         if (IS_ERR(handle))
1783                 return PTR_ERR(handle);
1784
1785         if (IS_DIRSYNC(dir))
1786                 handle->h_sync = 1;
1787
1788         inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1789         err = PTR_ERR(inode);
1790         if (IS_ERR(inode))
1791                 goto out_stop;
1792
1793         inode->i_op = &ext3_dir_inode_operations;
1794         inode->i_fop = &ext3_dir_operations;
1795         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1796         if (!(dir_block = ext3_dir_bread(handle, inode, 0, 1, &err)))
1797                 goto out_clear_inode;
1798
1799         BUFFER_TRACE(dir_block, "get_write_access");
1800         err = ext3_journal_get_write_access(handle, dir_block);
1801         if (err)
1802                 goto out_clear_inode;
1803
1804         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1805         de->inode = cpu_to_le32(inode->i_ino);
1806         de->name_len = 1;
1807         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1808         strcpy (de->name, ".");
1809         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1810         de = ext3_next_entry(de);
1811         de->inode = cpu_to_le32(dir->i_ino);
1812         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1813                                         EXT3_DIR_REC_LEN(1));
1814         de->name_len = 2;
1815         strcpy (de->name, "..");
1816         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1817         set_nlink(inode, 2);
1818         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1819         err = ext3_journal_dirty_metadata(handle, dir_block);
1820         if (err)
1821                 goto out_clear_inode;
1822
1823         err = ext3_mark_inode_dirty(handle, inode);
1824         if (!err)
1825                 err = ext3_add_entry (handle, dentry, inode);
1826
1827         if (err) {
1828 out_clear_inode:
1829                 clear_nlink(inode);
1830                 unlock_new_inode(inode);
1831                 ext3_mark_inode_dirty(handle, inode);
1832                 iput (inode);
1833                 goto out_stop;
1834         }
1835         inc_nlink(dir);
1836         ext3_update_dx_flag(dir);
1837         err = ext3_mark_inode_dirty(handle, dir);
1838         if (err)
1839                 goto out_clear_inode;
1840
1841         unlock_new_inode(inode);
1842         d_instantiate(dentry, inode);
1843 out_stop:
1844         brelse(dir_block);
1845         ext3_journal_stop(handle);
1846         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1847                 goto retry;
1848         return err;
1849 }
1850
1851 /*
1852  * routine to check that the specified directory is empty (for rmdir)
1853  */
1854 static int empty_dir (struct inode * inode)
1855 {
1856         unsigned long offset;
1857         struct buffer_head * bh;
1858         struct ext3_dir_entry_2 * de, * de1;
1859         struct super_block * sb;
1860         int err = 0;
1861
1862         sb = inode->i_sb;
1863         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1864             !(bh = ext3_dir_bread(NULL, inode, 0, 0, &err))) {
1865                 if (err)
1866                         ext3_error(inode->i_sb, __func__,
1867                                    "error %d reading directory #%lu offset 0",
1868                                    err, inode->i_ino);
1869                 else
1870                         ext3_warning(inode->i_sb, __func__,
1871                                      "bad directory (dir #%lu) - no data block",
1872                                      inode->i_ino);
1873                 return 1;
1874         }
1875         de = (struct ext3_dir_entry_2 *) bh->b_data;
1876         de1 = ext3_next_entry(de);
1877         if (le32_to_cpu(de->inode) != inode->i_ino ||
1878                         !le32_to_cpu(de1->inode) ||
1879                         strcmp (".", de->name) ||
1880                         strcmp ("..", de1->name)) {
1881                 ext3_warning (inode->i_sb, "empty_dir",
1882                               "bad directory (dir #%lu) - no `.' or `..'",
1883                               inode->i_ino);
1884                 brelse (bh);
1885                 return 1;
1886         }
1887         offset = ext3_rec_len_from_disk(de->rec_len) +
1888                         ext3_rec_len_from_disk(de1->rec_len);
1889         de = ext3_next_entry(de1);
1890         while (offset < inode->i_size ) {
1891                 if (!bh ||
1892                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1893                         err = 0;
1894                         brelse (bh);
1895                         if (!(bh = ext3_dir_bread (NULL, inode,
1896                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err))) {
1897                                 if (err)
1898                                         ext3_error(sb, __func__,
1899                                                    "error %d reading directory"
1900                                                    " #%lu offset %lu",
1901                                                    err, inode->i_ino, offset);
1902                                 offset += sb->s_blocksize;
1903                                 continue;
1904                         }
1905                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1906                 }
1907                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1908                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1909                                                          sb->s_blocksize);
1910                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1911                         continue;
1912                 }
1913                 if (le32_to_cpu(de->inode)) {
1914                         brelse (bh);
1915                         return 0;
1916                 }
1917                 offset += ext3_rec_len_from_disk(de->rec_len);
1918                 de = ext3_next_entry(de);
1919         }
1920         brelse (bh);
1921         return 1;
1922 }
1923
1924 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1925  * such inodes, starting at the superblock, in case we crash before the
1926  * file is closed/deleted, or in case the inode truncate spans multiple
1927  * transactions and the last transaction is not recovered after a crash.
1928  *
1929  * At filesystem recovery time, we walk this list deleting unlinked
1930  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1931  */
1932 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1933 {
1934         struct super_block *sb = inode->i_sb;
1935         struct ext3_iloc iloc;
1936         int err = 0, rc;
1937
1938         mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1939         if (!list_empty(&EXT3_I(inode)->i_orphan))
1940                 goto out_unlock;
1941
1942         /* Orphan handling is only valid for files with data blocks
1943          * being truncated, or files being unlinked. */
1944
1945         /* @@@ FIXME: Observation from aviro:
1946          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1947          * here (on s_orphan_lock), so race with ext3_link() which might bump
1948          * ->i_nlink. For, say it, character device. Not a regular file,
1949          * not a directory, not a symlink and ->i_nlink > 0.
1950          *
1951          * tytso, 4/25/2009: I'm not sure how that could happen;
1952          * shouldn't the fs core protect us from these sort of
1953          * unlink()/link() races?
1954          */
1955         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1956                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1957
1958         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1959         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1960         if (err)
1961                 goto out_unlock;
1962
1963         err = ext3_reserve_inode_write(handle, inode, &iloc);
1964         if (err)
1965                 goto out_unlock;
1966
1967         /* Insert this inode at the head of the on-disk orphan list... */
1968         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1969         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1970         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1971         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1972         if (!err)
1973                 err = rc;
1974
1975         /* Only add to the head of the in-memory list if all the
1976          * previous operations succeeded.  If the orphan_add is going to
1977          * fail (possibly taking the journal offline), we can't risk
1978          * leaving the inode on the orphan list: stray orphan-list
1979          * entries can cause panics at unmount time.
1980          *
1981          * This is safe: on error we're going to ignore the orphan list
1982          * anyway on the next recovery. */
1983         if (!err)
1984                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1985
1986         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1987         jbd_debug(4, "orphan inode %lu will point to %d\n",
1988                         inode->i_ino, NEXT_ORPHAN(inode));
1989 out_unlock:
1990         mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
1991         ext3_std_error(inode->i_sb, err);
1992         return err;
1993 }
1994
1995 /*
1996  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1997  * of such inodes stored on disk, because it is finally being cleaned up.
1998  */
1999 int ext3_orphan_del(handle_t *handle, struct inode *inode)
2000 {
2001         struct list_head *prev;
2002         struct ext3_inode_info *ei = EXT3_I(inode);
2003         struct ext3_sb_info *sbi;
2004         unsigned long ino_next;
2005         struct ext3_iloc iloc;
2006         int err = 0;
2007
2008         mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2009         if (list_empty(&ei->i_orphan))
2010                 goto out;
2011
2012         ino_next = NEXT_ORPHAN(inode);
2013         prev = ei->i_orphan.prev;
2014         sbi = EXT3_SB(inode->i_sb);
2015
2016         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2017
2018         list_del_init(&ei->i_orphan);
2019
2020         /* If we're on an error path, we may not have a valid
2021          * transaction handle with which to update the orphan list on
2022          * disk, but we still need to remove the inode from the linked
2023          * list in memory. */
2024         if (!handle)
2025                 goto out;
2026
2027         err = ext3_reserve_inode_write(handle, inode, &iloc);
2028         if (err)
2029                 goto out_err;
2030
2031         if (prev == &sbi->s_orphan) {
2032                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2033                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2034                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2035                 if (err)
2036                         goto out_brelse;
2037                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2038                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2039         } else {
2040                 struct ext3_iloc iloc2;
2041                 struct inode *i_prev =
2042                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2043
2044                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2045                           i_prev->i_ino, ino_next);
2046                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2047                 if (err)
2048                         goto out_brelse;
2049                 NEXT_ORPHAN(i_prev) = ino_next;
2050                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2051         }
2052         if (err)
2053                 goto out_brelse;
2054         NEXT_ORPHAN(inode) = 0;
2055         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2056
2057 out_err:
2058         ext3_std_error(inode->i_sb, err);
2059 out:
2060         mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2061         return err;
2062
2063 out_brelse:
2064         brelse(iloc.bh);
2065         goto out_err;
2066 }
2067
2068 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2069 {
2070         int retval;
2071         struct inode * inode;
2072         struct buffer_head * bh;
2073         struct ext3_dir_entry_2 * de;
2074         handle_t *handle;
2075
2076         /* Initialize quotas before so that eventual writes go in
2077          * separate transaction */
2078         dquot_initialize(dir);
2079         dquot_initialize(dentry->d_inode);
2080
2081         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2082         if (IS_ERR(handle))
2083                 return PTR_ERR(handle);
2084
2085         retval = -ENOENT;
2086         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2087         if (!bh)
2088                 goto end_rmdir;
2089
2090         if (IS_DIRSYNC(dir))
2091                 handle->h_sync = 1;
2092
2093         inode = dentry->d_inode;
2094
2095         retval = -EIO;
2096         if (le32_to_cpu(de->inode) != inode->i_ino)
2097                 goto end_rmdir;
2098
2099         retval = -ENOTEMPTY;
2100         if (!empty_dir (inode))
2101                 goto end_rmdir;
2102
2103         retval = ext3_delete_entry(handle, dir, de, bh);
2104         if (retval)
2105                 goto end_rmdir;
2106         if (inode->i_nlink != 2)
2107                 ext3_warning (inode->i_sb, "ext3_rmdir",
2108                               "empty directory has nlink!=2 (%d)",
2109                               inode->i_nlink);
2110         inode->i_version++;
2111         clear_nlink(inode);
2112         /* There's no need to set i_disksize: the fact that i_nlink is
2113          * zero will ensure that the right thing happens during any
2114          * recovery. */
2115         inode->i_size = 0;
2116         ext3_orphan_add(handle, inode);
2117         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2118         ext3_mark_inode_dirty(handle, inode);
2119         drop_nlink(dir);
2120         ext3_update_dx_flag(dir);
2121         ext3_mark_inode_dirty(handle, dir);
2122
2123 end_rmdir:
2124         ext3_journal_stop(handle);
2125         brelse (bh);
2126         return retval;
2127 }
2128
2129 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2130 {
2131         int retval;
2132         struct inode * inode;
2133         struct buffer_head * bh;
2134         struct ext3_dir_entry_2 * de;
2135         handle_t *handle;
2136
2137         trace_ext3_unlink_enter(dir, dentry);
2138         /* Initialize quotas before so that eventual writes go
2139          * in separate transaction */
2140         dquot_initialize(dir);
2141         dquot_initialize(dentry->d_inode);
2142
2143         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2144         if (IS_ERR(handle))
2145                 return PTR_ERR(handle);
2146
2147         if (IS_DIRSYNC(dir))
2148                 handle->h_sync = 1;
2149
2150         retval = -ENOENT;
2151         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2152         if (!bh)
2153                 goto end_unlink;
2154
2155         inode = dentry->d_inode;
2156
2157         retval = -EIO;
2158         if (le32_to_cpu(de->inode) != inode->i_ino)
2159                 goto end_unlink;
2160
2161         if (!inode->i_nlink) {
2162                 ext3_warning (inode->i_sb, "ext3_unlink",
2163                               "Deleting nonexistent file (%lu), %d",
2164                               inode->i_ino, inode->i_nlink);
2165                 set_nlink(inode, 1);
2166         }
2167         retval = ext3_delete_entry(handle, dir, de, bh);
2168         if (retval)
2169                 goto end_unlink;
2170         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2171         ext3_update_dx_flag(dir);
2172         ext3_mark_inode_dirty(handle, dir);
2173         drop_nlink(inode);
2174         if (!inode->i_nlink)
2175                 ext3_orphan_add(handle, inode);
2176         inode->i_ctime = dir->i_ctime;
2177         ext3_mark_inode_dirty(handle, inode);
2178         retval = 0;
2179
2180 end_unlink:
2181         ext3_journal_stop(handle);
2182         brelse (bh);
2183         trace_ext3_unlink_exit(dentry, retval);
2184         return retval;
2185 }
2186
2187 static int ext3_symlink (struct inode * dir,
2188                 struct dentry *dentry, const char * symname)
2189 {
2190         handle_t *handle;
2191         struct inode * inode;
2192         int l, err, retries = 0;
2193         int credits;
2194
2195         l = strlen(symname)+1;
2196         if (l > dir->i_sb->s_blocksize)
2197                 return -ENAMETOOLONG;
2198
2199         dquot_initialize(dir);
2200
2201         if (l > EXT3_N_BLOCKS * 4) {
2202                 /*
2203                  * For non-fast symlinks, we just allocate inode and put it on
2204                  * orphan list in the first transaction => we need bitmap,
2205                  * group descriptor, sb, inode block, quota blocks, and
2206                  * possibly selinux xattr blocks.
2207                  */
2208                 credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2209                           EXT3_XATTR_TRANS_BLOCKS;
2210         } else {
2211                 /*
2212                  * Fast symlink. We have to add entry to directory
2213                  * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2214                  * allocate new inode (bitmap, group descriptor, inode block,
2215                  * quota blocks, sb is already counted in previous macros).
2216                  */
2217                 credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2218                           EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2219                           EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2220         }
2221 retry:
2222         handle = ext3_journal_start(dir, credits);
2223         if (IS_ERR(handle))
2224                 return PTR_ERR(handle);
2225
2226         if (IS_DIRSYNC(dir))
2227                 handle->h_sync = 1;
2228
2229         inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2230         err = PTR_ERR(inode);
2231         if (IS_ERR(inode))
2232                 goto out_stop;
2233
2234         if (l > EXT3_N_BLOCKS * 4) {
2235                 inode->i_op = &ext3_symlink_inode_operations;
2236                 ext3_set_aops(inode);
2237                 /*
2238                  * We cannot call page_symlink() with transaction started
2239                  * because it calls into ext3_write_begin() which acquires page
2240                  * lock which ranks below transaction start (and it can also
2241                  * wait for journal commit if we are running out of space). So
2242                  * we have to stop transaction now and restart it when symlink
2243                  * contents is written.
2244                  *
2245                  * To keep fs consistent in case of crash, we have to put inode
2246                  * to orphan list in the mean time.
2247                  */
2248                 drop_nlink(inode);
2249                 err = ext3_orphan_add(handle, inode);
2250                 ext3_journal_stop(handle);
2251                 if (err)
2252                         goto err_drop_inode;
2253                 err = __page_symlink(inode, symname, l, 1);
2254                 if (err)
2255                         goto err_drop_inode;
2256                 /*
2257                  * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2258                  * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2259                  */
2260                 handle = ext3_journal_start(dir,
2261                                 EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2262                                 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2263                 if (IS_ERR(handle)) {
2264                         err = PTR_ERR(handle);
2265                         goto err_drop_inode;
2266                 }
2267                 set_nlink(inode, 1);
2268                 err = ext3_orphan_del(handle, inode);
2269                 if (err) {
2270                         ext3_journal_stop(handle);
2271                         drop_nlink(inode);
2272                         goto err_drop_inode;
2273                 }
2274         } else {
2275                 inode->i_op = &ext3_fast_symlink_inode_operations;
2276                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2277                 inode->i_size = l-1;
2278         }
2279         EXT3_I(inode)->i_disksize = inode->i_size;
2280         err = ext3_add_nondir(handle, dentry, inode);
2281 out_stop:
2282         ext3_journal_stop(handle);
2283         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2284                 goto retry;
2285         return err;
2286 err_drop_inode:
2287         unlock_new_inode(inode);
2288         iput(inode);
2289         return err;
2290 }
2291
2292 static int ext3_link (struct dentry * old_dentry,
2293                 struct inode * dir, struct dentry *dentry)
2294 {
2295         handle_t *handle;
2296         struct inode *inode = old_dentry->d_inode;
2297         int err, retries = 0;
2298
2299         if (inode->i_nlink >= EXT3_LINK_MAX)
2300                 return -EMLINK;
2301
2302         dquot_initialize(dir);
2303
2304 retry:
2305         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2306                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2307         if (IS_ERR(handle))
2308                 return PTR_ERR(handle);
2309
2310         if (IS_DIRSYNC(dir))
2311                 handle->h_sync = 1;
2312
2313         inode->i_ctime = CURRENT_TIME_SEC;
2314         inc_nlink(inode);
2315         ihold(inode);
2316
2317         err = ext3_add_entry(handle, dentry, inode);
2318         if (!err) {
2319                 ext3_mark_inode_dirty(handle, inode);
2320                 d_instantiate(dentry, inode);
2321         } else {
2322                 drop_nlink(inode);
2323                 iput(inode);
2324         }
2325         ext3_journal_stop(handle);
2326         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2327                 goto retry;
2328         return err;
2329 }
2330
2331 #define PARENT_INO(buffer) \
2332         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2333
2334 /*
2335  * Anybody can rename anything with this: the permission checks are left to the
2336  * higher-level routines.
2337  */
2338 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2339                            struct inode * new_dir,struct dentry *new_dentry)
2340 {
2341         handle_t *handle;
2342         struct inode * old_inode, * new_inode;
2343         struct buffer_head * old_bh, * new_bh, * dir_bh;
2344         struct ext3_dir_entry_2 * old_de, * new_de;
2345         int retval, flush_file = 0;
2346
2347         dquot_initialize(old_dir);
2348         dquot_initialize(new_dir);
2349
2350         old_bh = new_bh = dir_bh = NULL;
2351
2352         /* Initialize quotas before so that eventual writes go
2353          * in separate transaction */
2354         if (new_dentry->d_inode)
2355                 dquot_initialize(new_dentry->d_inode);
2356         handle = ext3_journal_start(old_dir, 2 *
2357                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2358                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2359         if (IS_ERR(handle))
2360                 return PTR_ERR(handle);
2361
2362         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2363                 handle->h_sync = 1;
2364
2365         old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2366         /*
2367          *  Check for inode number is _not_ due to possible IO errors.
2368          *  We might rmdir the source, keep it as pwd of some process
2369          *  and merrily kill the link to whatever was created under the
2370          *  same name. Goodbye sticky bit ;-<
2371          */
2372         old_inode = old_dentry->d_inode;
2373         retval = -ENOENT;
2374         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2375                 goto end_rename;
2376
2377         new_inode = new_dentry->d_inode;
2378         new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2379         if (new_bh) {
2380                 if (!new_inode) {
2381                         brelse (new_bh);
2382                         new_bh = NULL;
2383                 }
2384         }
2385         if (S_ISDIR(old_inode->i_mode)) {
2386                 if (new_inode) {
2387                         retval = -ENOTEMPTY;
2388                         if (!empty_dir (new_inode))
2389                                 goto end_rename;
2390                 }
2391                 retval = -EIO;
2392                 dir_bh = ext3_dir_bread(handle, old_inode, 0, 0, &retval);
2393                 if (!dir_bh)
2394                         goto end_rename;
2395                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2396                         goto end_rename;
2397                 retval = -EMLINK;
2398                 if (!new_inode && new_dir!=old_dir &&
2399                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2400                         goto end_rename;
2401         }
2402         if (!new_bh) {
2403                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2404                 if (retval)
2405                         goto end_rename;
2406         } else {
2407                 BUFFER_TRACE(new_bh, "get write access");
2408                 retval = ext3_journal_get_write_access(handle, new_bh);
2409                 if (retval)
2410                         goto journal_error;
2411                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2412                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2413                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2414                         new_de->file_type = old_de->file_type;
2415                 new_dir->i_version++;
2416                 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2417                 ext3_mark_inode_dirty(handle, new_dir);
2418                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2419                 retval = ext3_journal_dirty_metadata(handle, new_bh);
2420                 if (retval)
2421                         goto journal_error;
2422                 brelse(new_bh);
2423                 new_bh = NULL;
2424         }
2425
2426         /*
2427          * Like most other Unix systems, set the ctime for inodes on a
2428          * rename.
2429          */
2430         old_inode->i_ctime = CURRENT_TIME_SEC;
2431         ext3_mark_inode_dirty(handle, old_inode);
2432
2433         /*
2434          * ok, that's it
2435          */
2436         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2437             old_de->name_len != old_dentry->d_name.len ||
2438             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2439             (retval = ext3_delete_entry(handle, old_dir,
2440                                         old_de, old_bh)) == -ENOENT) {
2441                 /* old_de could have moved from under us during htree split, so
2442                  * make sure that we are deleting the right entry.  We might
2443                  * also be pointing to a stale entry in the unused part of
2444                  * old_bh so just checking inum and the name isn't enough. */
2445                 struct buffer_head *old_bh2;
2446                 struct ext3_dir_entry_2 *old_de2;
2447
2448                 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2449                                           &old_de2);
2450                 if (old_bh2) {
2451                         retval = ext3_delete_entry(handle, old_dir,
2452                                                    old_de2, old_bh2);
2453                         brelse(old_bh2);
2454                 }
2455         }
2456         if (retval) {
2457                 ext3_warning(old_dir->i_sb, "ext3_rename",
2458                                 "Deleting old file (%lu), %d, error=%d",
2459                                 old_dir->i_ino, old_dir->i_nlink, retval);
2460         }
2461
2462         if (new_inode) {
2463                 drop_nlink(new_inode);
2464                 new_inode->i_ctime = CURRENT_TIME_SEC;
2465         }
2466         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2467         ext3_update_dx_flag(old_dir);
2468         if (dir_bh) {
2469                 BUFFER_TRACE(dir_bh, "get_write_access");
2470                 retval = ext3_journal_get_write_access(handle, dir_bh);
2471                 if (retval)
2472                         goto journal_error;
2473                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2474                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2475                 retval = ext3_journal_dirty_metadata(handle, dir_bh);
2476                 if (retval) {
2477 journal_error:
2478                         ext3_std_error(new_dir->i_sb, retval);
2479                         goto end_rename;
2480                 }
2481                 drop_nlink(old_dir);
2482                 if (new_inode) {
2483                         drop_nlink(new_inode);
2484                 } else {
2485                         inc_nlink(new_dir);
2486                         ext3_update_dx_flag(new_dir);
2487                         ext3_mark_inode_dirty(handle, new_dir);
2488                 }
2489         }
2490         ext3_mark_inode_dirty(handle, old_dir);
2491         if (new_inode) {
2492                 ext3_mark_inode_dirty(handle, new_inode);
2493                 if (!new_inode->i_nlink)
2494                         ext3_orphan_add(handle, new_inode);
2495                 if (ext3_should_writeback_data(new_inode))
2496                         flush_file = 1;
2497         }
2498         retval = 0;
2499
2500 end_rename:
2501         brelse (dir_bh);
2502         brelse (old_bh);
2503         brelse (new_bh);
2504         ext3_journal_stop(handle);
2505         if (retval == 0 && flush_file)
2506                 filemap_flush(old_inode->i_mapping);
2507         return retval;
2508 }
2509
2510 /*
2511  * directories can handle most operations...
2512  */
2513 const struct inode_operations ext3_dir_inode_operations = {
2514         .create         = ext3_create,
2515         .lookup         = ext3_lookup,
2516         .link           = ext3_link,
2517         .unlink         = ext3_unlink,
2518         .symlink        = ext3_symlink,
2519         .mkdir          = ext3_mkdir,
2520         .rmdir          = ext3_rmdir,
2521         .mknod          = ext3_mknod,
2522         .rename         = ext3_rename,
2523         .setattr        = ext3_setattr,
2524 #ifdef CONFIG_EXT3_FS_XATTR
2525         .setxattr       = generic_setxattr,
2526         .getxattr       = generic_getxattr,
2527         .listxattr      = ext3_listxattr,
2528         .removexattr    = generic_removexattr,
2529 #endif
2530         .get_acl        = ext3_get_acl,
2531 };
2532
2533 const struct inode_operations ext3_special_inode_operations = {
2534         .setattr        = ext3_setattr,
2535 #ifdef CONFIG_EXT3_FS_XATTR
2536         .setxattr       = generic_setxattr,
2537         .getxattr       = generic_getxattr,
2538         .listxattr      = ext3_listxattr,
2539         .removexattr    = generic_removexattr,
2540 #endif
2541         .get_acl        = ext3_get_acl,
2542 };