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/fs.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/jbd.h>
  30 #include <linux/time.h>
  31 #include <linux/ext3_fs.h>
  32 #include <linux/ext3_jbd.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/stat.h>
  35 #include <linux/string.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/buffer_head.h>
  38 #include <linux/bio.h>
  39
  40 #include "namei.h"
  41 #include "xattr.h"
  42 #include "acl.h"
  43
  44 /*
  45  * define how far ahead to read directories while searching them.
  46  */
  47 #define NAMEI_RA_CHUNKS  2
  48 #define NAMEI_RA_BLOCKS  4
  49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52 static struct buffer_head *ext3_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         u32 *block, int *err)
  55 {
  56         struct buffer_head *bh;
  57
  58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  59
  60         bh = ext3_bread(handle, inode, *block, 1, err);
  61         if (bh) {
  62                 inode->i_size += inode->i_sb->s_blocksize;
  63                 EXT3_I(inode)->i_disksize = inode->i_size;
  64                 *err = ext3_journal_get_write_access(handle, bh);
  65                 if (*err) {
  66                         brelse(bh);
  67                         bh = NULL;
  68                 }
  69         }
  70         return bh;
  71 }
  72
  73 #ifndef assert
  74 #define assert(test) J_ASSERT(test)
  75 #endif
  76
  77 #ifdef DX_DEBUG
  78 #define dxtrace(command) command
  79 #else
  80 #define dxtrace(command)
  81 #endif
  82
  83 struct fake_dirent
  84 {
  85         __le32 inode;
  86         __le16 rec_len;
  87         u8 name_len;
  88         u8 file_type;
  89 };
  90
  91 struct dx_countlimit
  92 {
  93         __le16 limit;
  94         __le16 count;
  95 };
  96
  97 struct dx_entry
  98 {
  99         __le32 hash;
 100         __le32 block;
 101 };
 102
 103 /*
 104  * dx_root_info is laid out so that if it should somehow get overlaid by a
 105  * dirent the two low bits of the hash version will be zero.  Therefore, the
 106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 107  */
 108
 109 struct dx_root
 110 {
 111         struct fake_dirent dot;
 112         char dot_name[4];
 113         struct fake_dirent dotdot;
 114         char dotdot_name[4];
 115         struct dx_root_info
 116         {
 117                 __le32 reserved_zero;
 118                 u8 hash_version;
 119                 u8 info_length; /* 8 */
 120                 u8 indirect_levels;
 121                 u8 unused_flags;
 122         }
 123         info;
 124         struct dx_entry entries[0];
 125 };
 126
 127 struct dx_node
 128 {
 129         struct fake_dirent fake;
 130         struct dx_entry entries[0];
 131 };
 132
 133
 134 struct dx_frame
 135 {
 136         struct buffer_head *bh;
 137         struct dx_entry *entries;
 138         struct dx_entry *at;
 139 };
 140
 141 struct dx_map_entry
 142 {
 143         u32 hash;
 144         u16 offs;
 145         u16 size;
 146 };
 147
 148 static inline unsigned dx_get_block (struct dx_entry *entry);
 149 static void dx_set_block (struct dx_entry *entry, unsigned value);
 150 static inline unsigned dx_get_hash (struct dx_entry *entry);
 151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
 152 static unsigned dx_get_count (struct dx_entry *entries);
 153 static unsigned dx_get_limit (struct dx_entry *entries);
 154 static void dx_set_count (struct dx_entry *entries, unsigned value);
 155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
 156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 157 static unsigned dx_node_limit (struct inode *dir);
 158 static struct dx_frame *dx_probe(struct qstr *entry,
 159                                  struct inode *dir,
 160                                  struct dx_hash_info *hinfo,
 161                                  struct dx_frame *frame,
 162                                  int *err);
 163 static void dx_release (struct dx_frame *frames);
 164 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
 165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
 168                 struct dx_map_entry *offsets, int count);
 169 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
 170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 172                                  struct dx_frame *frame,
 173                                  struct dx_frame *frames,
 174                                  __u32 *start_hash);
 175 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
 176                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
 177                         int *err);
 178 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
 179                              struct inode *inode);
 180
 181 /*
 182  * p is at least 6 bytes before the end of page
 183  */
 184 static inline struct ext3_dir_entry_2 *
 185 ext3_next_entry(struct ext3_dir_entry_2 *p)
 186 {
 187         return (struct ext3_dir_entry_2 *)((char *)p +
 188                 ext3_rec_len_from_disk(p->rec_len));
 189 }
 190
 191 /*
 192  * Future: use high four bits of block for coalesce-on-delete flags
 193  * Mask them off for now.
 194  */
 195
 196 static inline unsigned dx_get_block (struct dx_entry *entry)
 197 {
 198         return le32_to_cpu(entry->block) & 0x00ffffff;
 199 }
 200
 201 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 202 {
 203         entry->block = cpu_to_le32(value);
 204 }
 205
 206 static inline unsigned dx_get_hash (struct dx_entry *entry)
 207 {
 208         return le32_to_cpu(entry->hash);
 209 }
 210
 211 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 212 {
 213         entry->hash = cpu_to_le32(value);
 214 }
 215
 216 static inline unsigned dx_get_count (struct dx_entry *entries)
 217 {
 218         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 219 }
 220
 221 static inline unsigned dx_get_limit (struct dx_entry *entries)
 222 {
 223         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 224 }
 225
 226 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 227 {
 228         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 229 }
 230
 231 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 232 {
 233         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 234 }
 235
 236 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 237 {
 238         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 239                 EXT3_DIR_REC_LEN(2) - infosize;
 240         return entry_space / sizeof(struct dx_entry);
 241 }
 242
 243 static inline unsigned dx_node_limit (struct inode *dir)
 244 {
 245         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 246         return entry_space / sizeof(struct dx_entry);
 247 }
 248
 249 /*
 250  * Debug
 251  */
 252 #ifdef DX_DEBUG
 253 static void dx_show_index (char * label, struct dx_entry *entries)
 254 {
 255         int i, n = dx_get_count (entries);
 256         printk("%s index ", label);
 257         for (i = 0; i < n; i++)
 258         {
 259                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 260         }
 261         printk("\n");
 262 }
 263
 264 struct stats
 265 {
 266         unsigned names;
 267         unsigned space;
 268         unsigned bcount;
 269 };
 270
 271 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
 272                                  int size, int show_names)
 273 {
 274         unsigned names = 0, space = 0;
 275         char *base = (char *) de;
 276         struct dx_hash_info h = *hinfo;
 277
 278         printk("names: ");
 279         while ((char *) de < base + size)
 280         {
 281                 if (de->inode)
 282                 {
 283                         if (show_names)
 284                         {
 285                                 int len = de->name_len;
 286                                 char *name = de->name;
 287                                 while (len--) printk("%c", *name++);
 288                                 ext3fs_dirhash(de->name, de->name_len, &h);
 289                                 printk(":%x.%u ", h.hash,
 290                                        ((char *) de - base));
 291                         }
 292                         space += EXT3_DIR_REC_LEN(de->name_len);
 293                         names++;
 294                 }
 295                 de = ext3_next_entry(de);
 296         }
 297         printk("(%i)\n", names);
 298         return (struct stats) { names, space, 1 };
 299 }
 300
 301 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 302                              struct dx_entry *entries, int levels)
 303 {
 304         unsigned blocksize = dir->i_sb->s_blocksize;
 305         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 306         unsigned bcount = 0;
 307         struct buffer_head *bh;
 308         int err;
 309         printk("%i indexed blocks...\n", count);
 310         for (i = 0; i < count; i++, entries++)
 311         {
 312                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 313                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 314                 struct stats stats;
 315                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 316                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 317                 stats = levels?
 318                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 319                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
 320                 names += stats.names;
 321                 space += stats.space;
 322                 bcount += stats.bcount;
 323                 brelse (bh);
 324         }
 325         if (bcount)
 326                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 327                         names, space/bcount,(space/bcount)*100/blocksize);
 328         return (struct stats) { names, space, bcount};
 329 }
 330 #endif /* DX_DEBUG */
 331
 332 /*
 333  * Probe for a directory leaf block to search.
 334  *
 335  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 336  * error in the directory index, and the caller should fall back to
 337  * searching the directory normally.  The callers of dx_probe **MUST**
 338  * check for this error code, and make sure it never gets reflected
 339  * back to userspace.
 340  */
 341 static struct dx_frame *
 342 dx_probe(struct qstr *entry, struct inode *dir,
 343          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 344 {
 345         unsigned count, indirect;
 346         struct dx_entry *at, *entries, *p, *q, *m;
 347         struct dx_root *root;
 348         struct buffer_head *bh;
 349         struct dx_frame *frame = frame_in;
 350         u32 hash;
 351
 352         frame->bh = NULL;
 353         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
 354                 goto fail;
 355         root = (struct dx_root *) bh->b_data;
 356         if (root->info.hash_version != DX_HASH_TEA &&
 357             root->info.hash_version != DX_HASH_HALF_MD4 &&
 358             root->info.hash_version != DX_HASH_LEGACY) {
 359                 ext3_warning(dir->i_sb, __func__,
 360                              "Unrecognised inode hash code %d",
 361                              root->info.hash_version);
 362                 brelse(bh);
 363                 *err = ERR_BAD_DX_DIR;
 364                 goto fail;
 365         }
 366         hinfo->hash_version = root->info.hash_version;
 367         if (hinfo->hash_version <= DX_HASH_TEA)
 368                 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
 369         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 370         if (entry)
 371                 ext3fs_dirhash(entry->name, entry->len, hinfo);
 372         hash = hinfo->hash;
 373
 374         if (root->info.unused_flags & 1) {
 375                 ext3_warning(dir->i_sb, __func__,
 376                              "Unimplemented inode hash flags: %#06x",
 377                              root->info.unused_flags);
 378                 brelse(bh);
 379                 *err = ERR_BAD_DX_DIR;
 380                 goto fail;
 381         }
 382
 383         if ((indirect = root->info.indirect_levels) > 1) {
 384                 ext3_warning(dir->i_sb, __func__,
 385                              "Unimplemented inode hash depth: %#06x",
 386                              root->info.indirect_levels);
 387                 brelse(bh);
 388                 *err = ERR_BAD_DX_DIR;
 389                 goto fail;
 390         }
 391
 392         entries = (struct dx_entry *) (((char *)&root->info) +
 393                                        root->info.info_length);
 394
 395         if (dx_get_limit(entries) != dx_root_limit(dir,
 396                                                    root->info.info_length)) {
 397                 ext3_warning(dir->i_sb, __func__,
 398                              "dx entry: limit != root limit");
 399                 brelse(bh);
 400                 *err = ERR_BAD_DX_DIR;
 401                 goto fail;
 402         }
 403
 404         dxtrace (printk("Look up %x", hash));
 405         while (1)
 406         {
 407                 count = dx_get_count(entries);
 408                 if (!count || count > dx_get_limit(entries)) {
 409                         ext3_warning(dir->i_sb, __func__,
 410                                      "dx entry: no count or count > limit");
 411                         brelse(bh);
 412                         *err = ERR_BAD_DX_DIR;
 413                         goto fail2;
 414                 }
 415
 416                 p = entries + 1;
 417                 q = entries + count - 1;
 418                 while (p <= q)
 419                 {
 420                         m = p + (q - p)/2;
 421                         dxtrace(printk("."));
 422                         if (dx_get_hash(m) > hash)
 423                                 q = m - 1;
 424                         else
 425                                 p = m + 1;
 426                 }
 427
 428                 if (0) // linear search cross check
 429                 {
 430                         unsigned n = count - 1;
 431                         at = entries;
 432                         while (n--)
 433                         {
 434                                 dxtrace(printk(","));
 435                                 if (dx_get_hash(++at) > hash)
 436                                 {
 437                                         at--;
 438                                         break;
 439                                 }
 440                         }
 441                         assert (at == p - 1);
 442                 }
 443
 444                 at = p - 1;
 445                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 446                 frame->bh = bh;
 447                 frame->entries = entries;
 448                 frame->at = at;
 449                 if (!indirect--) return frame;
 450                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
 451                         goto fail2;
 452                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 453                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 454                         ext3_warning(dir->i_sb, __func__,
 455                                      "dx entry: limit != node limit");
 456                         brelse(bh);
 457                         *err = ERR_BAD_DX_DIR;
 458                         goto fail2;
 459                 }
 460                 frame++;
 461                 frame->bh = NULL;
 462         }
 463 fail2:
 464         while (frame >= frame_in) {
 465                 brelse(frame->bh);
 466                 frame--;
 467         }
 468 fail:
 469         if (*err == ERR_BAD_DX_DIR)
 470                 ext3_warning(dir->i_sb, __func__,
 471                              "Corrupt dir inode %ld, running e2fsck is "
 472                              "recommended.", dir->i_ino);
 473         return NULL;
 474 }
 475
 476 static void dx_release (struct dx_frame *frames)
 477 {
 478         if (frames[0].bh == NULL)
 479                 return;
 480
 481         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 482                 brelse(frames[1].bh);
 483         brelse(frames[0].bh);
 484 }
 485
 486 /*
 487  * This function increments the frame pointer to search the next leaf
 488  * block, and reads in the necessary intervening nodes if the search
 489  * should be necessary.  Whether or not the search is necessary is
 490  * controlled by the hash parameter.  If the hash value is even, then
 491  * the search is only continued if the next block starts with that
 492  * hash value.  This is used if we are searching for a specific file.
 493  *
 494  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 495  *
 496  * This function returns 1 if the caller should continue to search,
 497  * or 0 if it should not.  If there is an error reading one of the
 498  * index blocks, it will a negative error code.
 499  *
 500  * If start_hash is non-null, it will be filled in with the starting
 501  * hash of the next page.
 502  */
 503 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 504                                  struct dx_frame *frame,
 505                                  struct dx_frame *frames,
 506                                  __u32 *start_hash)
 507 {
 508         struct dx_frame *p;
 509         struct buffer_head *bh;
 510         int err, num_frames = 0;
 511         __u32 bhash;
 512
 513         p = frame;
 514         /*
 515          * Find the next leaf page by incrementing the frame pointer.
 516          * If we run out of entries in the interior node, loop around and
 517          * increment pointer in the parent node.  When we break out of
 518          * this loop, num_frames indicates the number of interior
 519          * nodes need to be read.
 520          */
 521         while (1) {
 522                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 523                         break;
 524                 if (p == frames)
 525                         return 0;
 526                 num_frames++;
 527                 p--;
 528         }
 529
 530         /*
 531          * If the hash is 1, then continue only if the next page has a
 532          * continuation hash of any value.  This is used for readdir
 533          * handling.  Otherwise, check to see if the hash matches the
 534          * desired contiuation hash.  If it doesn't, return since
 535          * there's no point to read in the successive index pages.
 536          */
 537         bhash = dx_get_hash(p->at);
 538         if (start_hash)
 539                 *start_hash = bhash;
 540         if ((hash & 1) == 0) {
 541                 if ((bhash & ~1) != hash)
 542                         return 0;
 543         }
 544         /*
 545          * If the hash is HASH_NB_ALWAYS, we always go to the next
 546          * block so no check is necessary
 547          */
 548         while (num_frames--) {
 549                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
 550                                       0, &err)))
 551                         return err; /* Failure */
 552                 p++;
 553                 brelse (p->bh);
 554                 p->bh = bh;
 555                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 556         }
 557         return 1;
 558 }
 559
 560
 561 /*
 562  * This function fills a red-black tree with information from a
 563  * directory block.  It returns the number directory entries loaded
 564  * into the tree.  If there is an error it is returned in err.
 565  */
 566 static int htree_dirblock_to_tree(struct file *dir_file,
 567                                   struct inode *dir, int block,
 568                                   struct dx_hash_info *hinfo,
 569                                   __u32 start_hash, __u32 start_minor_hash)
 570 {
 571         struct buffer_head *bh;
 572         struct ext3_dir_entry_2 *de, *top;
 573         int err, count = 0;
 574
 575         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
 576         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
 577                 return err;
 578
 579         de = (struct ext3_dir_entry_2 *) bh->b_data;
 580         top = (struct ext3_dir_entry_2 *) ((char *) de +
 581                                            dir->i_sb->s_blocksize -
 582                                            EXT3_DIR_REC_LEN(0));
 583         for (; de < top; de = ext3_next_entry(de)) {
 584                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 585                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
 586                                                 +((char *)de - bh->b_data))) {
 587                         /* On error, skip the f_pos to the next block. */
 588                         dir_file->f_pos = (dir_file->f_pos |
 589                                         (dir->i_sb->s_blocksize - 1)) + 1;
 590                         brelse (bh);
 591                         return count;
 592                 }
 593                 ext3fs_dirhash(de->name, de->name_len, hinfo);
 594                 if ((hinfo->hash < start_hash) ||
 595                     ((hinfo->hash == start_hash) &&
 596                      (hinfo->minor_hash < start_minor_hash)))
 597                         continue;
 598                 if (de->inode == 0)
 599                         continue;
 600                 if ((err = ext3_htree_store_dirent(dir_file,
 601                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 602                         brelse(bh);
 603                         return err;
 604                 }
 605                 count++;
 606         }
 607         brelse(bh);
 608         return count;
 609 }
 610
 611
 612 /*
 613  * This function fills a red-black tree with information from a
 614  * directory.  We start scanning the directory in hash order, starting
 615  * at start_hash and start_minor_hash.
 616  *
 617  * This function returns the number of entries inserted into the tree,
 618  * or a negative error code.
 619  */
 620 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 621                          __u32 start_minor_hash, __u32 *next_hash)
 622 {
 623         struct dx_hash_info hinfo;
 624         struct ext3_dir_entry_2 *de;
 625         struct dx_frame frames[2], *frame;
 626         struct inode *dir;
 627         int block, err;
 628         int count = 0;
 629         int ret;
 630         __u32 hashval;
 631
 632         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
 633                        start_minor_hash));
 634         dir = dir_file->f_path.dentry->d_inode;
 635         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
 636                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
 637                 if (hinfo.hash_version <= DX_HASH_TEA)
 638                         hinfo.hash_version +=
 639                                 EXT3_SB(dir->i_sb)->s_hash_unsigned;
 640                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 641                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 642                                                start_hash, start_minor_hash);
 643                 *next_hash = ~0;
 644                 return count;
 645         }
 646         hinfo.hash = start_hash;
 647         hinfo.minor_hash = 0;
 648         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
 649         if (!frame)
 650                 return err;
 651
 652         /* Add '.' and '..' from the htree header */
 653         if (!start_hash && !start_minor_hash) {
 654                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 655                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 656                         goto errout;
 657                 count++;
 658         }
 659         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 660                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 661                 de = ext3_next_entry(de);
 662                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 663                         goto errout;
 664                 count++;
 665         }
 666
 667         while (1) {
 668                 block = dx_get_block(frame->at);
 669                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 670                                              start_hash, start_minor_hash);
 671                 if (ret < 0) {
 672                         err = ret;
 673                         goto errout;
 674                 }
 675                 count += ret;
 676                 hashval = ~0;
 677                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 678                                             frame, frames, &hashval);
 679                 *next_hash = hashval;
 680                 if (ret < 0) {
 681                         err = ret;
 682                         goto errout;
 683                 }
 684                 /*
 685                  * Stop if:  (a) there are no more entries, or
 686                  * (b) we have inserted at least one entry and the
 687                  * next hash value is not a continuation
 688                  */
 689                 if ((ret == 0) ||
 690                     (count && ((hashval & 1) == 0)))
 691                         break;
 692         }
 693         dx_release(frames);
 694         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 695                        count, *next_hash));
 696         return count;
 697 errout:
 698         dx_release(frames);
 699         return (err);
 700 }
 701
 702
 703 /*
 704  * Directory block splitting, compacting
 705  */
 706
 707 /*
 708  * Create map of hash values, offsets, and sizes, stored at end of block.
 709  * Returns number of entries mapped.
 710  */
 711 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
 712                 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 713 {
 714         int count = 0;
 715         char *base = (char *) de;
 716         struct dx_hash_info h = *hinfo;
 717
 718         while ((char *) de < base + blocksize)
 719         {
 720                 if (de->name_len && de->inode) {
 721                         ext3fs_dirhash(de->name, de->name_len, &h);
 722                         map_tail--;
 723                         map_tail->hash = h.hash;
 724                         map_tail->offs = (u16) ((char *) de - base);
 725                         map_tail->size = le16_to_cpu(de->rec_len);
 726                         count++;
 727                         cond_resched();
 728                 }
 729                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 730                 de = ext3_next_entry(de);
 731         }
 732         return count;
 733 }
 734
 735 /* Sort map by hash value */
 736 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 737 {
 738         struct dx_map_entry *p, *q, *top = map + count - 1;
 739         int more;
 740         /* Combsort until bubble sort doesn't suck */
 741         while (count > 2)
 742         {
 743                 count = count*10/13;
 744                 if (count - 9 < 2) /* 9, 10 -> 11 */
 745                         count = 11;
 746                 for (p = top, q = p - count; q >= map; p--, q--)
 747                         if (p->hash < q->hash)
 748                                 swap(*p, *q);
 749         }
 750         /* Garden variety bubble sort */
 751         do {
 752                 more = 0;
 753                 q = top;
 754                 while (q-- > map)
 755                 {
 756                         if (q[1].hash >= q[0].hash)
 757                                 continue;
 758                         swap(*(q+1), *q);
 759                         more = 1;
 760                 }
 761         } while(more);
 762 }
 763
 764 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 765 {
 766         struct dx_entry *entries = frame->entries;
 767         struct dx_entry *old = frame->at, *new = old + 1;
 768         int count = dx_get_count(entries);
 769
 770         assert(count < dx_get_limit(entries));
 771         assert(old < entries + count);
 772         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 773         dx_set_hash(new, hash);
 774         dx_set_block(new, block);
 775         dx_set_count(entries, count + 1);
 776 }
 777
 778 static void ext3_update_dx_flag(struct inode *inode)
 779 {
 780         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 781                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 782                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 783 }
 784
 785 /*
 786  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 787  *
 788  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 789  * `de != NULL' is guaranteed by caller.
 790  */
 791 static inline int ext3_match (int len, const char * const name,
 792                               struct ext3_dir_entry_2 * de)
 793 {
 794         if (len != de->name_len)
 795                 return 0;
 796         if (!de->inode)
 797                 return 0;
 798         return !memcmp(name, de->name, len);
 799 }
 800
 801 /*
 802  * Returns 0 if not found, -1 on failure, and 1 on success
 803  */
 804 static inline int search_dirblock(struct buffer_head * bh,
 805                                   struct inode *dir,
 806                                   struct qstr *child,
 807                                   unsigned long offset,
 808                                   struct ext3_dir_entry_2 ** res_dir)
 809 {
 810         struct ext3_dir_entry_2 * de;
 811         char * dlimit;
 812         int de_len;
 813         const char *name = child->name;
 814         int namelen = child->len;
 815
 816         de = (struct ext3_dir_entry_2 *) bh->b_data;
 817         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 818         while ((char *) de < dlimit) {
 819                 /* this code is executed quadratically often */
 820                 /* do minimal checking `by hand' */
 821
 822                 if ((char *) de + namelen <= dlimit &&
 823                     ext3_match (namelen, name, de)) {
 824                         /* found a match - just to be sure, do a full check */
 825                         if (!ext3_check_dir_entry("ext3_find_entry",
 826                                                   dir, de, bh, offset))
 827                                 return -1;
 828                         *res_dir = de;
 829                         return 1;
 830                 }
 831                 /* prevent looping on a bad block */
 832                 de_len = ext3_rec_len_from_disk(de->rec_len);
 833                 if (de_len <= 0)
 834                         return -1;
 835                 offset += de_len;
 836                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 837         }
 838         return 0;
 839 }
 840
 841
 842 /*
 843  *      ext3_find_entry()
 844  *
 845  * finds an entry in the specified directory with the wanted name. It
 846  * returns the cache buffer in which the entry was found, and the entry
 847  * itself (as a parameter - res_dir). It does NOT read the inode of the
 848  * entry - you'll have to do that yourself if you want to.
 849  *
 850  * The returned buffer_head has ->b_count elevated.  The caller is expected
 851  * to brelse() it when appropriate.
 852  */
 853 static struct buffer_head *ext3_find_entry(struct inode *dir,
 854                                         struct qstr *entry,
 855                                         struct ext3_dir_entry_2 **res_dir)
 856 {
 857         struct super_block * sb;
 858         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 859         struct buffer_head * bh, *ret = NULL;
 860         unsigned long start, block, b;
 861         const u8 *name = entry->name;
 862         int ra_max = 0;         /* Number of bh's in the readahead
 863                                    buffer, bh_use[] */
 864         int ra_ptr = 0;         /* Current index into readahead
 865                                    buffer */
 866         int num = 0;
 867         int nblocks, i, err;
 868         int namelen;
 869
 870         *res_dir = NULL;
 871         sb = dir->i_sb;
 872         namelen = entry->len;
 873         if (namelen > EXT3_NAME_LEN)
 874                 return NULL;
 875         if ((namelen <= 2) && (name[0] == '.') &&
 876             (name[1] == '.' || name[1] == 0)) {
 877                 /*
 878                  * "." or ".." will only be in the first block
 879                  * NFS may look up ".."; "." should be handled by the VFS
 880                  */
 881                 block = start = 0;
 882                 nblocks = 1;
 883                 goto restart;
 884         }
 885         if (is_dx(dir)) {
 886                 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
 887                 /*
 888                  * On success, or if the error was file not found,
 889                  * return.  Otherwise, fall back to doing a search the
 890                  * old fashioned way.
 891                  */
 892                 if (bh || (err != ERR_BAD_DX_DIR))
 893                         return bh;
 894                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 895         }
 896         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 897         start = EXT3_I(dir)->i_dir_start_lookup;
 898         if (start >= nblocks)
 899                 start = 0;
 900         block = start;
 901 restart:
 902         do {
 903                 /*
 904                  * We deal with the read-ahead logic here.
 905                  */
 906                 if (ra_ptr >= ra_max) {
 907                         /* Refill the readahead buffer */
 908                         ra_ptr = 0;
 909                         b = block;
 910                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 911                                 /*
 912                                  * Terminate if we reach the end of the
 913                                  * directory and must wrap, or if our
 914                                  * search has finished at this block.
 915                                  */
 916                                 if (b >= nblocks || (num && block == start)) {
 917                                         bh_use[ra_max] = NULL;
 918                                         break;
 919                                 }
 920                                 num++;
 921                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 922                                 bh_use[ra_max] = bh;
 923                                 if (bh)
 924                                         ll_rw_block(READ_META, 1, &bh);
 925                         }
 926                 }
 927                 if ((bh = bh_use[ra_ptr++]) == NULL)
 928                         goto next;
 929                 wait_on_buffer(bh);
 930                 if (!buffer_uptodate(bh)) {
 931                         /* read error, skip block & hope for the best */
 932                         ext3_error(sb, __func__, "reading directory #%lu "
 933                                    "offset %lu", dir->i_ino, block);
 934                         brelse(bh);
 935                         goto next;
 936                 }
 937                 i = search_dirblock(bh, dir, entry,
 938                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 939                 if (i == 1) {
 940                         EXT3_I(dir)->i_dir_start_lookup = block;
 941                         ret = bh;
 942                         goto cleanup_and_exit;
 943                 } else {
 944                         brelse(bh);
 945                         if (i < 0)
 946                                 goto cleanup_and_exit;
 947                 }
 948         next:
 949                 if (++block >= nblocks)
 950                         block = 0;
 951         } while (block != start);
 952
 953         /*
 954          * If the directory has grown while we were searching, then
 955          * search the last part of the directory before giving up.
 956          */
 957         block = nblocks;
 958         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 959         if (block < nblocks) {
 960                 start = 0;
 961                 goto restart;
 962         }
 963
 964 cleanup_and_exit:
 965         /* Clean up the read-ahead blocks */
 966         for (; ra_ptr < ra_max; ra_ptr++)
 967                 brelse (bh_use[ra_ptr]);
 968         return ret;
 969 }
 970
 971 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
 972                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
 973                         int *err)
 974 {
 975         struct super_block *sb = dir->i_sb;
 976         struct dx_hash_info     hinfo;
 977         struct dx_frame frames[2], *frame;
 978         struct buffer_head *bh;
 979         unsigned long block;
 980         int retval;
 981
 982         if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
 983                 return NULL;
 984         do {
 985                 block = dx_get_block(frame->at);
 986                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
 987                         goto errout;
 988
 989                 retval = search_dirblock(bh, dir, entry,
 990                                          block << EXT3_BLOCK_SIZE_BITS(sb),
 991                                          res_dir);
 992                 if (retval == 1) {
 993                         dx_release(frames);
 994                         return bh;
 995                 }
 996                 brelse(bh);
 997                 if (retval == -1) {
 998                         *err = ERR_BAD_DX_DIR;
 999                         goto errout;
1000                 }
1001
1002                 /* Check to see if we should continue to search */
1003                 retval = ext3_htree_next_block(dir, hinfo.hash, frame,
1004                                                frames, NULL);
1005                 if (retval < 0) {
1006                         ext3_warning(sb, __func__,
1007                              "error reading index page in directory #%lu",
1008                              dir->i_ino);
1009                         *err = retval;
1010                         goto errout;
1011                 }
1012         } while (retval == 1);
1013
1014         *err = -ENOENT;
1015 errout:
1016         dxtrace(printk("%s not found\n", name));
1017         dx_release (frames);
1018         return NULL;
1019 }
1020
1021 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1022 {
1023         struct inode * inode;
1024         struct ext3_dir_entry_2 * de;
1025         struct buffer_head * bh;
1026
1027         if (dentry->d_name.len > EXT3_NAME_LEN)
1028                 return ERR_PTR(-ENAMETOOLONG);
1029
1030         bh = ext3_find_entry(dir, &dentry->d_name, &de);
1031         inode = NULL;
1032         if (bh) {
1033                 unsigned long ino = le32_to_cpu(de->inode);
1034                 brelse (bh);
1035                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1036                         ext3_error(dir->i_sb, "ext3_lookup",
1037                                    "bad inode number: %lu", ino);
1038                         return ERR_PTR(-EIO);
1039                 }
1040                 inode = ext3_iget(dir->i_sb, ino);
1041                 if (inode == ERR_PTR(-ESTALE)) {
1042                         ext3_error(dir->i_sb, __func__,
1043                                         "deleted inode referenced: %lu",
1044                                         ino);
1045                         return ERR_PTR(-EIO);
1046                 }
1047         }
1048         return d_splice_alias(inode, dentry);
1049 }
1050
1051
1052 struct dentry *ext3_get_parent(struct dentry *child)
1053 {
1054         unsigned long ino;
1055         struct qstr dotdot = {.name = "..", .len = 2};
1056         struct ext3_dir_entry_2 * de;
1057         struct buffer_head *bh;
1058
1059         bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1060         if (!bh)
1061                 return ERR_PTR(-ENOENT);
1062         ino = le32_to_cpu(de->inode);
1063         brelse(bh);
1064
1065         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1066                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1067                            "bad inode number: %lu", ino);
1068                 return ERR_PTR(-EIO);
1069         }
1070
1071         return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1072 }
1073
1074 #define S_SHIFT 12
1075 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1076         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1077         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1078         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1079         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1080         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1081         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1082         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1083 };
1084
1085 static inline void ext3_set_de_type(struct super_block *sb,
1086                                 struct ext3_dir_entry_2 *de,
1087                                 umode_t mode) {
1088         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1089                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1090 }
1091
1092 /*
1093  * Move count entries from end of map between two memory locations.
1094  * Returns pointer to last entry moved.
1095  */
1096 static struct ext3_dir_entry_2 *
1097 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1098 {
1099         unsigned rec_len = 0;
1100
1101         while (count--) {
1102                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1103                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1104                 memcpy (to, de, rec_len);
1105                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1106                                 ext3_rec_len_to_disk(rec_len);
1107                 de->inode = 0;
1108                 map++;
1109                 to += rec_len;
1110         }
1111         return (struct ext3_dir_entry_2 *) (to - rec_len);
1112 }
1113
1114 /*
1115  * Compact each dir entry in the range to the minimal rec_len.
1116  * Returns pointer to last entry in range.
1117  */
1118 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1119 {
1120         struct ext3_dir_entry_2 *next, *to, *prev;
1121         struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1122         unsigned rec_len = 0;
1123
1124         prev = to = de;
1125         while ((char *)de < base + blocksize) {
1126                 next = ext3_next_entry(de);
1127                 if (de->inode && de->name_len) {
1128                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1129                         if (de > to)
1130                                 memmove(to, de, rec_len);
1131                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1132                         prev = to;
1133                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1134                 }
1135                 de = next;
1136         }
1137         return prev;
1138 }
1139
1140 /*
1141  * Split a full leaf block to make room for a new dir entry.
1142  * Allocate a new block, and move entries so that they are approx. equally full.
1143  * Returns pointer to de in block into which the new entry will be inserted.
1144  */
1145 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1146                         struct buffer_head **bh,struct dx_frame *frame,
1147                         struct dx_hash_info *hinfo, int *error)
1148 {
1149         unsigned blocksize = dir->i_sb->s_blocksize;
1150         unsigned count, continued;
1151         struct buffer_head *bh2;
1152         u32 newblock;
1153         u32 hash2;
1154         struct dx_map_entry *map;
1155         char *data1 = (*bh)->b_data, *data2;
1156         unsigned split, move, size;
1157         struct ext3_dir_entry_2 *de = NULL, *de2;
1158         int     err = 0, i;
1159
1160         bh2 = ext3_append (handle, dir, &newblock, &err);
1161         if (!(bh2)) {
1162                 brelse(*bh);
1163                 *bh = NULL;
1164                 goto errout;
1165         }
1166
1167         BUFFER_TRACE(*bh, "get_write_access");
1168         err = ext3_journal_get_write_access(handle, *bh);
1169         if (err)
1170                 goto journal_error;
1171
1172         BUFFER_TRACE(frame->bh, "get_write_access");
1173         err = ext3_journal_get_write_access(handle, frame->bh);
1174         if (err)
1175                 goto journal_error;
1176
1177         data2 = bh2->b_data;
1178
1179         /* create map in the end of data2 block */
1180         map = (struct dx_map_entry *) (data2 + blocksize);
1181         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1182                              blocksize, hinfo, map);
1183         map -= count;
1184         dx_sort_map (map, count);
1185         /* Split the existing block in the middle, size-wise */
1186         size = 0;
1187         move = 0;
1188         for (i = count-1; i >= 0; i--) {
1189                 /* is more than half of this entry in 2nd half of the block? */
1190                 if (size + map[i].size/2 > blocksize/2)
1191                         break;
1192                 size += map[i].size;
1193                 move++;
1194         }
1195         /* map index at which we will split */
1196         split = count - move;
1197         hash2 = map[split].hash;
1198         continued = hash2 == map[split - 1].hash;
1199         dxtrace(printk("Split block %i at %x, %i/%i\n",
1200                 dx_get_block(frame->at), hash2, split, count-split));
1201
1202         /* Fancy dance to stay within two buffers */
1203         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1204         de = dx_pack_dirents(data1,blocksize);
1205         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1206         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1207         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1208         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1209
1210         /* Which block gets the new entry? */
1211         if (hinfo->hash >= hash2)
1212         {
1213                 swap(*bh, bh2);
1214                 de = de2;
1215         }
1216         dx_insert_block (frame, hash2 + continued, newblock);
1217         err = ext3_journal_dirty_metadata (handle, bh2);
1218         if (err)
1219                 goto journal_error;
1220         err = ext3_journal_dirty_metadata (handle, frame->bh);
1221         if (err)
1222                 goto journal_error;
1223         brelse (bh2);
1224         dxtrace(dx_show_index ("frame", frame->entries));
1225         return de;
1226
1227 journal_error:
1228         brelse(*bh);
1229         brelse(bh2);
1230         *bh = NULL;
1231         ext3_std_error(dir->i_sb, err);
1232 errout:
1233         *error = err;
1234         return NULL;
1235 }
1236
1237
1238 /*
1239  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1240  * it points to a directory entry which is guaranteed to be large
1241  * enough for new directory entry.  If de is NULL, then
1242  * add_dirent_to_buf will attempt search the directory block for
1243  * space.  It will return -ENOSPC if no space is available, and -EIO
1244  * and -EEXIST if directory entry already exists.
1245  *
1246  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1247  * all other cases bh is released.
1248  */
1249 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1250                              struct inode *inode, struct ext3_dir_entry_2 *de,
1251                              struct buffer_head * bh)
1252 {
1253         struct inode    *dir = dentry->d_parent->d_inode;
1254         const char      *name = dentry->d_name.name;
1255         int             namelen = dentry->d_name.len;
1256         unsigned long   offset = 0;
1257         unsigned short  reclen;
1258         int             nlen, rlen, err;
1259         char            *top;
1260
1261         reclen = EXT3_DIR_REC_LEN(namelen);
1262         if (!de) {
1263                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1264                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1265                 while ((char *) de <= top) {
1266                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1267                                                   bh, offset)) {
1268                                 brelse (bh);
1269                                 return -EIO;
1270                         }
1271                         if (ext3_match (namelen, name, de)) {
1272                                 brelse (bh);
1273                                 return -EEXIST;
1274                         }
1275                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1276                         rlen = ext3_rec_len_from_disk(de->rec_len);
1277                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1278                                 break;
1279                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1280                         offset += rlen;
1281                 }
1282                 if ((char *) de > top)
1283                         return -ENOSPC;
1284         }
1285         BUFFER_TRACE(bh, "get_write_access");
1286         err = ext3_journal_get_write_access(handle, bh);
1287         if (err) {
1288                 ext3_std_error(dir->i_sb, err);
1289                 brelse(bh);
1290                 return err;
1291         }
1292
1293         /* By now the buffer is marked for journaling */
1294         nlen = EXT3_DIR_REC_LEN(de->name_len);
1295         rlen = ext3_rec_len_from_disk(de->rec_len);
1296         if (de->inode) {
1297                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1298                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1299                 de->rec_len = ext3_rec_len_to_disk(nlen);
1300                 de = de1;
1301         }
1302         de->file_type = EXT3_FT_UNKNOWN;
1303         if (inode) {
1304                 de->inode = cpu_to_le32(inode->i_ino);
1305                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1306         } else
1307                 de->inode = 0;
1308         de->name_len = namelen;
1309         memcpy (de->name, name, namelen);
1310         /*
1311          * XXX shouldn't update any times until successful
1312          * completion of syscall, but too many callers depend
1313          * on this.
1314          *
1315          * XXX similarly, too many callers depend on
1316          * ext3_new_inode() setting the times, but error
1317          * recovery deletes the inode, so the worst that can
1318          * happen is that the times are slightly out of date
1319          * and/or different from the directory change time.
1320          */
1321         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1322         ext3_update_dx_flag(dir);
1323         dir->i_version++;
1324         ext3_mark_inode_dirty(handle, dir);
1325         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1326         err = ext3_journal_dirty_metadata(handle, bh);
1327         if (err)
1328                 ext3_std_error(dir->i_sb, err);
1329         brelse(bh);
1330         return 0;
1331 }
1332
1333 /*
1334  * This converts a one block unindexed directory to a 3 block indexed
1335  * directory, and adds the dentry to the indexed directory.
1336  */
1337 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1338                             struct inode *inode, struct buffer_head *bh)
1339 {
1340         struct inode    *dir = dentry->d_parent->d_inode;
1341         const char      *name = dentry->d_name.name;
1342         int             namelen = dentry->d_name.len;
1343         struct buffer_head *bh2;
1344         struct dx_root  *root;
1345         struct dx_frame frames[2], *frame;
1346         struct dx_entry *entries;
1347         struct ext3_dir_entry_2 *de, *de2;
1348         char            *data1, *top;
1349         unsigned        len;
1350         int             retval;
1351         unsigned        blocksize;
1352         struct dx_hash_info hinfo;
1353         u32             block;
1354         struct fake_dirent *fde;
1355
1356         blocksize =  dir->i_sb->s_blocksize;
1357         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1358         retval = ext3_journal_get_write_access(handle, bh);
1359         if (retval) {
1360                 ext3_std_error(dir->i_sb, retval);
1361                 brelse(bh);
1362                 return retval;
1363         }
1364         root = (struct dx_root *) bh->b_data;
1365
1366         /* The 0th block becomes the root, move the dirents out */
1367         fde = &root->dotdot;
1368         de = (struct ext3_dir_entry_2 *)((char *)fde +
1369                         ext3_rec_len_from_disk(fde->rec_len));
1370         if ((char *) de >= (((char *) root) + blocksize)) {
1371                 ext3_error(dir->i_sb, __func__,
1372                            "invalid rec_len for '..' in inode %lu",
1373                            dir->i_ino);
1374                 brelse(bh);
1375                 return -EIO;
1376         }
1377         len = ((char *) root) + blocksize - (char *) de;
1378
1379         bh2 = ext3_append (handle, dir, &block, &retval);
1380         if (!(bh2)) {
1381                 brelse(bh);
1382                 return retval;
1383         }
1384         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1385         data1 = bh2->b_data;
1386
1387         memcpy (data1, de, len);
1388         de = (struct ext3_dir_entry_2 *) data1;
1389         top = data1 + len;
1390         while ((char *)(de2 = ext3_next_entry(de)) < top)
1391                 de = de2;
1392         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1393         /* Initialize the root; the dot dirents already exist */
1394         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1395         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1396         memset (&root->info, 0, sizeof(root->info));
1397         root->info.info_length = sizeof(root->info);
1398         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1399         entries = root->entries;
1400         dx_set_block (entries, 1);
1401         dx_set_count (entries, 1);
1402         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1403
1404         /* Initialize as for dx_probe */
1405         hinfo.hash_version = root->info.hash_version;
1406         if (hinfo.hash_version <= DX_HASH_TEA)
1407                 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1408         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1409         ext3fs_dirhash(name, namelen, &hinfo);
1410         frame = frames;
1411         frame->entries = entries;
1412         frame->at = entries;
1413         frame->bh = bh;
1414         bh = bh2;
1415         /*
1416          * Mark buffers dirty here so that if do_split() fails we write a
1417          * consistent set of buffers to disk.
1418          */
1419         ext3_journal_dirty_metadata(handle, frame->bh);
1420         ext3_journal_dirty_metadata(handle, bh);
1421         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1422         if (!de) {
1423                 ext3_mark_inode_dirty(handle, dir);
1424                 dx_release(frames);
1425                 return retval;
1426         }
1427         dx_release(frames);
1428
1429         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1430 }
1431
1432 /*
1433  *      ext3_add_entry()
1434  *
1435  * adds a file entry to the specified directory, using the same
1436  * semantics as ext3_find_entry(). It returns NULL if it failed.
1437  *
1438  * NOTE!! The inode part of 'de' is left at 0 - which means you
1439  * may not sleep between calling this and putting something into
1440  * the entry, as someone else might have used it while you slept.
1441  */
1442 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1443         struct inode *inode)
1444 {
1445         struct inode *dir = dentry->d_parent->d_inode;
1446         struct buffer_head * bh;
1447         struct ext3_dir_entry_2 *de;
1448         struct super_block * sb;
1449         int     retval;
1450         int     dx_fallback=0;
1451         unsigned blocksize;
1452         u32 block, blocks;
1453
1454         sb = dir->i_sb;
1455         blocksize = sb->s_blocksize;
1456         if (!dentry->d_name.len)
1457                 return -EINVAL;
1458         if (is_dx(dir)) {
1459                 retval = ext3_dx_add_entry(handle, dentry, inode);
1460                 if (!retval || (retval != ERR_BAD_DX_DIR))
1461                         return retval;
1462                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1463                 dx_fallback++;
1464                 ext3_mark_inode_dirty(handle, dir);
1465         }
1466         blocks = dir->i_size >> sb->s_blocksize_bits;
1467         for (block = 0; block < blocks; block++) {
1468                 bh = ext3_bread(handle, dir, block, 0, &retval);
1469                 if(!bh)
1470                         return retval;
1471                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1472                 if (retval != -ENOSPC)
1473                         return retval;
1474
1475                 if (blocks == 1 && !dx_fallback &&
1476                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1477                         return make_indexed_dir(handle, dentry, inode, bh);
1478                 brelse(bh);
1479         }
1480         bh = ext3_append(handle, dir, &block, &retval);
1481         if (!bh)
1482                 return retval;
1483         de = (struct ext3_dir_entry_2 *) bh->b_data;
1484         de->inode = 0;
1485         de->rec_len = ext3_rec_len_to_disk(blocksize);
1486         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1487 }
1488
1489 /*
1490  * Returns 0 for success, or a negative error value
1491  */
1492 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1493                              struct inode *inode)
1494 {
1495         struct dx_frame frames[2], *frame;
1496         struct dx_entry *entries, *at;
1497         struct dx_hash_info hinfo;
1498         struct buffer_head * bh;
1499         struct inode *dir = dentry->d_parent->d_inode;
1500         struct super_block * sb = dir->i_sb;
1501         struct ext3_dir_entry_2 *de;
1502         int err;
1503
1504         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1505         if (!frame)
1506                 return err;
1507         entries = frame->entries;
1508         at = frame->at;
1509
1510         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1511                 goto cleanup;
1512
1513         BUFFER_TRACE(bh, "get_write_access");
1514         err = ext3_journal_get_write_access(handle, bh);
1515         if (err)
1516                 goto journal_error;
1517
1518         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1519         if (err != -ENOSPC) {
1520                 bh = NULL;
1521                 goto cleanup;
1522         }
1523
1524         /* Block full, should compress but for now just split */
1525         dxtrace(printk("using %u of %u node entries\n",
1526                        dx_get_count(entries), dx_get_limit(entries)));
1527         /* Need to split index? */
1528         if (dx_get_count(entries) == dx_get_limit(entries)) {
1529                 u32 newblock;
1530                 unsigned icount = dx_get_count(entries);
1531                 int levels = frame - frames;
1532                 struct dx_entry *entries2;
1533                 struct dx_node *node2;
1534                 struct buffer_head *bh2;
1535
1536                 if (levels && (dx_get_count(frames->entries) ==
1537                                dx_get_limit(frames->entries))) {
1538                         ext3_warning(sb, __func__,
1539                                      "Directory index full!");
1540                         err = -ENOSPC;
1541                         goto cleanup;
1542                 }
1543                 bh2 = ext3_append (handle, dir, &newblock, &err);
1544                 if (!(bh2))
1545                         goto cleanup;
1546                 node2 = (struct dx_node *)(bh2->b_data);
1547                 entries2 = node2->entries;
1548                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1549                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1550                 BUFFER_TRACE(frame->bh, "get_write_access");
1551                 err = ext3_journal_get_write_access(handle, frame->bh);
1552                 if (err)
1553                         goto journal_error;
1554                 if (levels) {
1555                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1556                         unsigned hash2 = dx_get_hash(entries + icount1);
1557                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1558
1559                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1560                         err = ext3_journal_get_write_access(handle,
1561                                                              frames[0].bh);
1562                         if (err)
1563                                 goto journal_error;
1564
1565                         memcpy ((char *) entries2, (char *) (entries + icount1),
1566                                 icount2 * sizeof(struct dx_entry));
1567                         dx_set_count (entries, icount1);
1568                         dx_set_count (entries2, icount2);
1569                         dx_set_limit (entries2, dx_node_limit(dir));
1570
1571                         /* Which index block gets the new entry? */
1572                         if (at - entries >= icount1) {
1573                                 frame->at = at = at - entries - icount1 + entries2;
1574                                 frame->entries = entries = entries2;
1575                                 swap(frame->bh, bh2);
1576                         }
1577                         dx_insert_block (frames + 0, hash2, newblock);
1578                         dxtrace(dx_show_index ("node", frames[1].entries));
1579                         dxtrace(dx_show_index ("node",
1580                                ((struct dx_node *) bh2->b_data)->entries));
1581                         err = ext3_journal_dirty_metadata(handle, bh2);
1582                         if (err)
1583                                 goto journal_error;
1584                         brelse (bh2);
1585                 } else {
1586                         dxtrace(printk("Creating second level index...\n"));
1587                         memcpy((char *) entries2, (char *) entries,
1588                                icount * sizeof(struct dx_entry));
1589                         dx_set_limit(entries2, dx_node_limit(dir));
1590
1591                         /* Set up root */
1592                         dx_set_count(entries, 1);
1593                         dx_set_block(entries + 0, newblock);
1594                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1595
1596                         /* Add new access path frame */
1597                         frame = frames + 1;
1598                         frame->at = at = at - entries + entries2;
1599                         frame->entries = entries = entries2;
1600                         frame->bh = bh2;
1601                         err = ext3_journal_get_write_access(handle,
1602                                                              frame->bh);
1603                         if (err)
1604                                 goto journal_error;
1605                 }
1606                 err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1607                 if (err)
1608                         goto journal_error;
1609         }
1610         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1611         if (!de)
1612                 goto cleanup;
1613         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1614         bh = NULL;
1615         goto cleanup;
1616
1617 journal_error:
1618         ext3_std_error(dir->i_sb, err);
1619 cleanup:
1620         if (bh)
1621                 brelse(bh);
1622         dx_release(frames);
1623         return err;
1624 }
1625
1626 /*
1627  * ext3_delete_entry deletes a directory entry by merging it with the
1628  * previous entry
1629  */
1630 static int ext3_delete_entry (handle_t *handle,
1631                               struct inode * dir,
1632                               struct ext3_dir_entry_2 * de_del,
1633                               struct buffer_head * bh)
1634 {
1635         struct ext3_dir_entry_2 * de, * pde;
1636         int i;
1637
1638         i = 0;
1639         pde = NULL;
1640         de = (struct ext3_dir_entry_2 *) bh->b_data;
1641         while (i < bh->b_size) {
1642                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1643                         return -EIO;
1644                 if (de == de_del)  {
1645                         int err;
1646
1647                         BUFFER_TRACE(bh, "get_write_access");
1648                         err = ext3_journal_get_write_access(handle, bh);
1649                         if (err)
1650                                 goto journal_error;
1651
1652                         if (pde)
1653                                 pde->rec_len = ext3_rec_len_to_disk(
1654                                         ext3_rec_len_from_disk(pde->rec_len) +
1655                                         ext3_rec_len_from_disk(de->rec_len));
1656                         else
1657                                 de->inode = 0;
1658                         dir->i_version++;
1659                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1660                         err = ext3_journal_dirty_metadata(handle, bh);
1661                         if (err) {
1662 journal_error:
1663                                 ext3_std_error(dir->i_sb, err);
1664                                 return err;
1665                         }
1666                         return 0;
1667                 }
1668                 i += ext3_rec_len_from_disk(de->rec_len);
1669                 pde = de;
1670                 de = ext3_next_entry(de);
1671         }
1672         return -ENOENT;
1673 }
1674
1675 static int ext3_add_nondir(handle_t *handle,
1676                 struct dentry *dentry, struct inode *inode)
1677 {
1678         int err = ext3_add_entry(handle, dentry, inode);
1679         if (!err) {
1680                 ext3_mark_inode_dirty(handle, inode);
1681                 d_instantiate(dentry, inode);
1682                 unlock_new_inode(inode);
1683                 return 0;
1684         }
1685         drop_nlink(inode);
1686         unlock_new_inode(inode);
1687         iput(inode);
1688         return err;
1689 }
1690
1691 /*
1692  * By the time this is called, we already have created
1693  * the directory cache entry for the new file, but it
1694  * is so far negative - it has no inode.
1695  *
1696  * If the create succeeds, we fill in the inode information
1697  * with d_instantiate().
1698  */
1699 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1700                 struct nameidata *nd)
1701 {
1702         handle_t *handle;
1703         struct inode * inode;
1704         int err, retries = 0;
1705
1706         dquot_initialize(dir);
1707
1708 retry:
1709         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1710                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1711                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1712         if (IS_ERR(handle))
1713                 return PTR_ERR(handle);
1714
1715         if (IS_DIRSYNC(dir))
1716                 handle->h_sync = 1;
1717
1718         inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1719         err = PTR_ERR(inode);
1720         if (!IS_ERR(inode)) {
1721                 inode->i_op = &ext3_file_inode_operations;
1722                 inode->i_fop = &ext3_file_operations;
1723                 ext3_set_aops(inode);
1724                 err = ext3_add_nondir(handle, dentry, inode);
1725         }
1726         ext3_journal_stop(handle);
1727         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1728                 goto retry;
1729         return err;
1730 }
1731
1732 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1733                         int mode, dev_t rdev)
1734 {
1735         handle_t *handle;
1736         struct inode *inode;
1737         int err, retries = 0;
1738
1739         if (!new_valid_dev(rdev))
1740                 return -EINVAL;
1741
1742         dquot_initialize(dir);
1743
1744 retry:
1745         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1746                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1747                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1748         if (IS_ERR(handle))
1749                 return PTR_ERR(handle);
1750
1751         if (IS_DIRSYNC(dir))
1752                 handle->h_sync = 1;
1753
1754         inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1755         err = PTR_ERR(inode);
1756         if (!IS_ERR(inode)) {
1757                 init_special_inode(inode, inode->i_mode, rdev);
1758 #ifdef CONFIG_EXT3_FS_XATTR
1759                 inode->i_op = &ext3_special_inode_operations;
1760 #endif
1761                 err = ext3_add_nondir(handle, dentry, inode);
1762         }
1763         ext3_journal_stop(handle);
1764         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1765                 goto retry;
1766         return err;
1767 }
1768
1769 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1770 {
1771         handle_t *handle;
1772         struct inode * inode;
1773         struct buffer_head * dir_block = NULL;
1774         struct ext3_dir_entry_2 * de;
1775         int err, retries = 0;
1776
1777         if (dir->i_nlink >= EXT3_LINK_MAX)
1778                 return -EMLINK;
1779
1780         dquot_initialize(dir);
1781
1782 retry:
1783         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1784                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1785                                         EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1786         if (IS_ERR(handle))
1787                 return PTR_ERR(handle);
1788
1789         if (IS_DIRSYNC(dir))
1790                 handle->h_sync = 1;
1791
1792         inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1793         err = PTR_ERR(inode);
1794         if (IS_ERR(inode))
1795                 goto out_stop;
1796
1797         inode->i_op = &ext3_dir_inode_operations;
1798         inode->i_fop = &ext3_dir_operations;
1799         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1800         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1801         if (!dir_block)
1802                 goto out_clear_inode;
1803
1804         BUFFER_TRACE(dir_block, "get_write_access");
1805         err = ext3_journal_get_write_access(handle, dir_block);
1806         if (err)
1807                 goto out_clear_inode;
1808
1809         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1810         de->inode = cpu_to_le32(inode->i_ino);
1811         de->name_len = 1;
1812         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1813         strcpy (de->name, ".");
1814         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1815         de = ext3_next_entry(de);
1816         de->inode = cpu_to_le32(dir->i_ino);
1817         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1818                                         EXT3_DIR_REC_LEN(1));
1819         de->name_len = 2;
1820         strcpy (de->name, "..");
1821         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1822         inode->i_nlink = 2;
1823         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1824         err = ext3_journal_dirty_metadata(handle, dir_block);
1825         if (err)
1826                 goto out_clear_inode;
1827
1828         err = ext3_mark_inode_dirty(handle, inode);
1829         if (!err)
1830                 err = ext3_add_entry (handle, dentry, inode);
1831
1832         if (err) {
1833 out_clear_inode:
1834                 inode->i_nlink = 0;
1835                 unlock_new_inode(inode);
1836                 ext3_mark_inode_dirty(handle, inode);
1837                 iput (inode);
1838                 goto out_stop;
1839         }
1840         inc_nlink(dir);
1841         ext3_update_dx_flag(dir);
1842         err = ext3_mark_inode_dirty(handle, dir);
1843         if (err)
1844                 goto out_clear_inode;
1845
1846         d_instantiate(dentry, inode);
1847         unlock_new_inode(inode);
1848 out_stop:
1849         brelse(dir_block);
1850         ext3_journal_stop(handle);
1851         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1852                 goto retry;
1853         return err;
1854 }
1855
1856 /*
1857  * routine to check that the specified directory is empty (for rmdir)
1858  */
1859 static int empty_dir (struct inode * inode)
1860 {
1861         unsigned long offset;
1862         struct buffer_head * bh;
1863         struct ext3_dir_entry_2 * de, * de1;
1864         struct super_block * sb;
1865         int err = 0;
1866
1867         sb = inode->i_sb;
1868         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1869             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1870                 if (err)
1871                         ext3_error(inode->i_sb, __func__,
1872                                    "error %d reading directory #%lu offset 0",
1873                                    err, inode->i_ino);
1874                 else
1875                         ext3_warning(inode->i_sb, __func__,
1876                                      "bad directory (dir #%lu) - no data block",
1877                                      inode->i_ino);
1878                 return 1;
1879         }
1880         de = (struct ext3_dir_entry_2 *) bh->b_data;
1881         de1 = ext3_next_entry(de);
1882         if (le32_to_cpu(de->inode) != inode->i_ino ||
1883                         !le32_to_cpu(de1->inode) ||
1884                         strcmp (".", de->name) ||
1885                         strcmp ("..", de1->name)) {
1886                 ext3_warning (inode->i_sb, "empty_dir",
1887                               "bad directory (dir #%lu) - no `.' or `..'",
1888                               inode->i_ino);
1889                 brelse (bh);
1890                 return 1;
1891         }
1892         offset = ext3_rec_len_from_disk(de->rec_len) +
1893                         ext3_rec_len_from_disk(de1->rec_len);
1894         de = ext3_next_entry(de1);
1895         while (offset < inode->i_size ) {
1896                 if (!bh ||
1897                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1898                         err = 0;
1899                         brelse (bh);
1900                         bh = ext3_bread (NULL, inode,
1901                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1902                         if (!bh) {
1903                                 if (err)
1904                                         ext3_error(sb, __func__,
1905                                                    "error %d reading directory"
1906                                                    " #%lu offset %lu",
1907                                                    err, inode->i_ino, offset);
1908                                 offset += sb->s_blocksize;
1909                                 continue;
1910                         }
1911                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1912                 }
1913                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1914                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1915                                                          sb->s_blocksize);
1916                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1917                         continue;
1918                 }
1919                 if (le32_to_cpu(de->inode)) {
1920                         brelse (bh);
1921                         return 0;
1922                 }
1923                 offset += ext3_rec_len_from_disk(de->rec_len);
1924                 de = ext3_next_entry(de);
1925         }
1926         brelse (bh);
1927         return 1;
1928 }
1929
1930 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1931  * such inodes, starting at the superblock, in case we crash before the
1932  * file is closed/deleted, or in case the inode truncate spans multiple
1933  * transactions and the last transaction is not recovered after a crash.
1934  *
1935  * At filesystem recovery time, we walk this list deleting unlinked
1936  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1937  */
1938 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1939 {
1940         struct super_block *sb = inode->i_sb;
1941         struct ext3_iloc iloc;
1942         int err = 0, rc;
1943
1944         mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1945         if (!list_empty(&EXT3_I(inode)->i_orphan))
1946                 goto out_unlock;
1947
1948         /* Orphan handling is only valid for files with data blocks
1949          * being truncated, or files being unlinked. */
1950
1951         /* @@@ FIXME: Observation from aviro:
1952          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1953          * here (on s_orphan_lock), so race with ext3_link() which might bump
1954          * ->i_nlink. For, say it, character device. Not a regular file,
1955          * not a directory, not a symlink and ->i_nlink > 0.
1956          *
1957          * tytso, 4/25/2009: I'm not sure how that could happen;
1958          * shouldn't the fs core protect us from these sort of
1959          * unlink()/link() races?
1960          */
1961         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1962                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1963
1964         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1965         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1966         if (err)
1967                 goto out_unlock;
1968
1969         err = ext3_reserve_inode_write(handle, inode, &iloc);
1970         if (err)
1971                 goto out_unlock;
1972
1973         /* Insert this inode at the head of the on-disk orphan list... */
1974         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1975         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1976         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1977         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1978         if (!err)
1979                 err = rc;
1980
1981         /* Only add to the head of the in-memory list if all the
1982          * previous operations succeeded.  If the orphan_add is going to
1983          * fail (possibly taking the journal offline), we can't risk
1984          * leaving the inode on the orphan list: stray orphan-list
1985          * entries can cause panics at unmount time.
1986          *
1987          * This is safe: on error we're going to ignore the orphan list
1988          * anyway on the next recovery. */
1989         if (!err)
1990                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1991
1992         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1993         jbd_debug(4, "orphan inode %lu will point to %d\n",
1994                         inode->i_ino, NEXT_ORPHAN(inode));
1995 out_unlock:
1996         mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
1997         ext3_std_error(inode->i_sb, err);
1998         return err;
1999 }
2000
2001 /*
2002  * ext3_orphan_del() removes an unlinked or truncated inode from the list
2003  * of such inodes stored on disk, because it is finally being cleaned up.
2004  */
2005 int ext3_orphan_del(handle_t *handle, struct inode *inode)
2006 {
2007         struct list_head *prev;
2008         struct ext3_inode_info *ei = EXT3_I(inode);
2009         struct ext3_sb_info *sbi;
2010         unsigned long ino_next;
2011         struct ext3_iloc iloc;
2012         int err = 0;
2013
2014         mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2015         if (list_empty(&ei->i_orphan))
2016                 goto out;
2017
2018         ino_next = NEXT_ORPHAN(inode);
2019         prev = ei->i_orphan.prev;
2020         sbi = EXT3_SB(inode->i_sb);
2021
2022         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2023
2024         list_del_init(&ei->i_orphan);
2025
2026         /* If we're on an error path, we may not have a valid
2027          * transaction handle with which to update the orphan list on
2028          * disk, but we still need to remove the inode from the linked
2029          * list in memory. */
2030         if (!handle)
2031                 goto out;
2032
2033         err = ext3_reserve_inode_write(handle, inode, &iloc);
2034         if (err)
2035                 goto out_err;
2036
2037         if (prev == &sbi->s_orphan) {
2038                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2039                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2040                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2041                 if (err)
2042                         goto out_brelse;
2043                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2044                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2045         } else {
2046                 struct ext3_iloc iloc2;
2047                 struct inode *i_prev =
2048                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2049
2050                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2051                           i_prev->i_ino, ino_next);
2052                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2053                 if (err)
2054                         goto out_brelse;
2055                 NEXT_ORPHAN(i_prev) = ino_next;
2056                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2057         }
2058         if (err)
2059                 goto out_brelse;
2060         NEXT_ORPHAN(inode) = 0;
2061         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2062
2063 out_err:
2064         ext3_std_error(inode->i_sb, err);
2065 out:
2066         mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2067         return err;
2068
2069 out_brelse:
2070         brelse(iloc.bh);
2071         goto out_err;
2072 }
2073
2074 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2075 {
2076         int retval;
2077         struct inode * inode;
2078         struct buffer_head * bh;
2079         struct ext3_dir_entry_2 * de;
2080         handle_t *handle;
2081
2082         /* Initialize quotas before so that eventual writes go in
2083          * separate transaction */
2084         dquot_initialize(dir);
2085         dquot_initialize(dentry->d_inode);
2086
2087         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2088         if (IS_ERR(handle))
2089                 return PTR_ERR(handle);
2090
2091         retval = -ENOENT;
2092         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2093         if (!bh)
2094                 goto end_rmdir;
2095
2096         if (IS_DIRSYNC(dir))
2097                 handle->h_sync = 1;
2098
2099         inode = dentry->d_inode;
2100
2101         retval = -EIO;
2102         if (le32_to_cpu(de->inode) != inode->i_ino)
2103                 goto end_rmdir;
2104
2105         retval = -ENOTEMPTY;
2106         if (!empty_dir (inode))
2107                 goto end_rmdir;
2108
2109         retval = ext3_delete_entry(handle, dir, de, bh);
2110         if (retval)
2111                 goto end_rmdir;
2112         if (inode->i_nlink != 2)
2113                 ext3_warning (inode->i_sb, "ext3_rmdir",
2114                               "empty directory has nlink!=2 (%d)",
2115                               inode->i_nlink);
2116         inode->i_version++;
2117         clear_nlink(inode);
2118         /* There's no need to set i_disksize: the fact that i_nlink is
2119          * zero will ensure that the right thing happens during any
2120          * recovery. */
2121         inode->i_size = 0;
2122         ext3_orphan_add(handle, inode);
2123         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2124         ext3_mark_inode_dirty(handle, inode);
2125         drop_nlink(dir);
2126         ext3_update_dx_flag(dir);
2127         ext3_mark_inode_dirty(handle, dir);
2128
2129 end_rmdir:
2130         ext3_journal_stop(handle);
2131         brelse (bh);
2132         return retval;
2133 }
2134
2135 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2136 {
2137         int retval;
2138         struct inode * inode;
2139         struct buffer_head * bh;
2140         struct ext3_dir_entry_2 * de;
2141         handle_t *handle;
2142
2143         /* Initialize quotas before so that eventual writes go
2144          * in separate transaction */
2145         dquot_initialize(dir);
2146         dquot_initialize(dentry->d_inode);
2147
2148         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2149         if (IS_ERR(handle))
2150                 return PTR_ERR(handle);
2151
2152         if (IS_DIRSYNC(dir))
2153                 handle->h_sync = 1;
2154
2155         retval = -ENOENT;
2156         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2157         if (!bh)
2158                 goto end_unlink;
2159
2160         inode = dentry->d_inode;
2161
2162         retval = -EIO;
2163         if (le32_to_cpu(de->inode) != inode->i_ino)
2164                 goto end_unlink;
2165
2166         if (!inode->i_nlink) {
2167                 ext3_warning (inode->i_sb, "ext3_unlink",
2168                               "Deleting nonexistent file (%lu), %d",
2169                               inode->i_ino, inode->i_nlink);
2170                 inode->i_nlink = 1;
2171         }
2172         retval = ext3_delete_entry(handle, dir, de, bh);
2173         if (retval)
2174                 goto end_unlink;
2175         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2176         ext3_update_dx_flag(dir);
2177         ext3_mark_inode_dirty(handle, dir);
2178         drop_nlink(inode);
2179         if (!inode->i_nlink)
2180                 ext3_orphan_add(handle, inode);
2181         inode->i_ctime = dir->i_ctime;
2182         ext3_mark_inode_dirty(handle, inode);
2183         retval = 0;
2184
2185 end_unlink:
2186         ext3_journal_stop(handle);
2187         brelse (bh);
2188         return retval;
2189 }
2190
2191 static int ext3_symlink (struct inode * dir,
2192                 struct dentry *dentry, const char * symname)
2193 {
2194         handle_t *handle;
2195         struct inode * inode;
2196         int l, err, retries = 0;
2197         int credits;
2198
2199         l = strlen(symname)+1;
2200         if (l > dir->i_sb->s_blocksize)
2201                 return -ENAMETOOLONG;
2202
2203         dquot_initialize(dir);
2204
2205         if (l > EXT3_N_BLOCKS * 4) {
2206                 /*
2207                  * For non-fast symlinks, we just allocate inode and put it on
2208                  * orphan list in the first transaction => we need bitmap,
2209                  * group descriptor, sb, inode block, quota blocks.
2210                  */
2211                 credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2212         } else {
2213                 /*
2214                  * Fast symlink. We have to add entry to directory
2215                  * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2216                  * allocate new inode (bitmap, group descriptor, inode block,
2217                  * quota blocks, sb is already counted in previous macros).
2218                  */
2219                 credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2220                           EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2221                           EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2222         }
2223 retry:
2224         handle = ext3_journal_start(dir, credits);
2225         if (IS_ERR(handle))
2226                 return PTR_ERR(handle);
2227
2228         if (IS_DIRSYNC(dir))
2229                 handle->h_sync = 1;
2230
2231         inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2232         err = PTR_ERR(inode);
2233         if (IS_ERR(inode))
2234                 goto out_stop;
2235
2236         if (l > EXT3_N_BLOCKS * 4) {
2237                 inode->i_op = &ext3_symlink_inode_operations;
2238                 ext3_set_aops(inode);
2239                 /*
2240                  * We cannot call page_symlink() with transaction started
2241                  * because it calls into ext3_write_begin() which acquires page
2242                  * lock which ranks below transaction start (and it can also
2243                  * wait for journal commit if we are running out of space). So
2244                  * we have to stop transaction now and restart it when symlink
2245                  * contents is written.
2246                  *
2247                  * To keep fs consistent in case of crash, we have to put inode
2248                  * to orphan list in the mean time.
2249                  */
2250                 drop_nlink(inode);
2251                 err = ext3_orphan_add(handle, inode);
2252                 ext3_journal_stop(handle);
2253                 if (err)
2254                         goto err_drop_inode;
2255                 err = __page_symlink(inode, symname, l, 1);
2256                 if (err)
2257                         goto err_drop_inode;
2258                 /*
2259                  * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2260                  * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2261                  */
2262                 handle = ext3_journal_start(dir,
2263                                 EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2264                                 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2265                 if (IS_ERR(handle)) {
2266                         err = PTR_ERR(handle);
2267                         goto err_drop_inode;
2268                 }
2269                 inc_nlink(inode);
2270                 err = ext3_orphan_del(handle, inode);
2271                 if (err) {
2272                         ext3_journal_stop(handle);
2273                         drop_nlink(inode);
2274                         goto err_drop_inode;
2275                 }
2276         } else {
2277                 inode->i_op = &ext3_fast_symlink_inode_operations;
2278                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2279                 inode->i_size = l-1;
2280         }
2281         EXT3_I(inode)->i_disksize = inode->i_size;
2282         err = ext3_add_nondir(handle, dentry, inode);
2283 out_stop:
2284         ext3_journal_stop(handle);
2285         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2286                 goto retry;
2287         return err;
2288 err_drop_inode:
2289         unlock_new_inode(inode);
2290         iput(inode);
2291         return err;
2292 }
2293
2294 static int ext3_link (struct dentry * old_dentry,
2295                 struct inode * dir, struct dentry *dentry)
2296 {
2297         handle_t *handle;
2298         struct inode *inode = old_dentry->d_inode;
2299         int err, retries = 0;
2300
2301         if (inode->i_nlink >= EXT3_LINK_MAX)
2302                 return -EMLINK;
2303
2304         dquot_initialize(dir);
2305
2306 retry:
2307         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2308                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2309         if (IS_ERR(handle))
2310                 return PTR_ERR(handle);
2311
2312         if (IS_DIRSYNC(dir))
2313                 handle->h_sync = 1;
2314
2315         inode->i_ctime = CURRENT_TIME_SEC;
2316         inc_nlink(inode);
2317         ihold(inode);
2318
2319         err = ext3_add_entry(handle, dentry, inode);
2320         if (!err) {
2321                 ext3_mark_inode_dirty(handle, inode);
2322                 d_instantiate(dentry, inode);
2323         } else {
2324                 drop_nlink(inode);
2325                 iput(inode);
2326         }
2327         ext3_journal_stop(handle);
2328         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2329                 goto retry;
2330         return err;
2331 }
2332
2333 #define PARENT_INO(buffer) \
2334         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2335
2336 /*
2337  * Anybody can rename anything with this: the permission checks are left to the
2338  * higher-level routines.
2339  */
2340 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2341                            struct inode * new_dir,struct dentry *new_dentry)
2342 {
2343         handle_t *handle;
2344         struct inode * old_inode, * new_inode;
2345         struct buffer_head * old_bh, * new_bh, * dir_bh;
2346         struct ext3_dir_entry_2 * old_de, * new_de;
2347         int retval, flush_file = 0;
2348
2349         dquot_initialize(old_dir);
2350         dquot_initialize(new_dir);
2351
2352         old_bh = new_bh = dir_bh = NULL;
2353
2354         /* Initialize quotas before so that eventual writes go
2355          * in separate transaction */
2356         if (new_dentry->d_inode)
2357                 dquot_initialize(new_dentry->d_inode);
2358         handle = ext3_journal_start(old_dir, 2 *
2359                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2360                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2361         if (IS_ERR(handle))
2362                 return PTR_ERR(handle);
2363
2364         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2365                 handle->h_sync = 1;
2366
2367         old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2368         /*
2369          *  Check for inode number is _not_ due to possible IO errors.
2370          *  We might rmdir the source, keep it as pwd of some process
2371          *  and merrily kill the link to whatever was created under the
2372          *  same name. Goodbye sticky bit ;-<
2373          */
2374         old_inode = old_dentry->d_inode;
2375         retval = -ENOENT;
2376         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2377                 goto end_rename;
2378
2379         new_inode = new_dentry->d_inode;
2380         new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2381         if (new_bh) {
2382                 if (!new_inode) {
2383                         brelse (new_bh);
2384                         new_bh = NULL;
2385                 }
2386         }
2387         if (S_ISDIR(old_inode->i_mode)) {
2388                 if (new_inode) {
2389                         retval = -ENOTEMPTY;
2390                         if (!empty_dir (new_inode))
2391                                 goto end_rename;
2392                 }
2393                 retval = -EIO;
2394                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2395                 if (!dir_bh)
2396                         goto end_rename;
2397                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2398                         goto end_rename;
2399                 retval = -EMLINK;
2400                 if (!new_inode && new_dir!=old_dir &&
2401                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2402                         goto end_rename;
2403         }
2404         if (!new_bh) {
2405                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2406                 if (retval)
2407                         goto end_rename;
2408         } else {
2409                 BUFFER_TRACE(new_bh, "get write access");
2410                 retval = ext3_journal_get_write_access(handle, new_bh);
2411                 if (retval)
2412                         goto journal_error;
2413                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2414                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2415                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2416                         new_de->file_type = old_de->file_type;
2417                 new_dir->i_version++;
2418                 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2419                 ext3_mark_inode_dirty(handle, new_dir);
2420                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2421                 retval = ext3_journal_dirty_metadata(handle, new_bh);
2422                 if (retval)
2423                         goto journal_error;
2424                 brelse(new_bh);
2425                 new_bh = NULL;
2426         }
2427
2428         /*
2429          * Like most other Unix systems, set the ctime for inodes on a
2430          * rename.
2431          */
2432         old_inode->i_ctime = CURRENT_TIME_SEC;
2433         ext3_mark_inode_dirty(handle, old_inode);
2434
2435         /*
2436          * ok, that's it
2437          */
2438         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2439             old_de->name_len != old_dentry->d_name.len ||
2440             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2441             (retval = ext3_delete_entry(handle, old_dir,
2442                                         old_de, old_bh)) == -ENOENT) {
2443                 /* old_de could have moved from under us during htree split, so
2444                  * make sure that we are deleting the right entry.  We might
2445                  * also be pointing to a stale entry in the unused part of
2446                  * old_bh so just checking inum and the name isn't enough. */
2447                 struct buffer_head *old_bh2;
2448                 struct ext3_dir_entry_2 *old_de2;
2449
2450                 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2451                                           &old_de2);
2452                 if (old_bh2) {
2453                         retval = ext3_delete_entry(handle, old_dir,
2454                                                    old_de2, old_bh2);
2455                         brelse(old_bh2);
2456                 }
2457         }
2458         if (retval) {
2459                 ext3_warning(old_dir->i_sb, "ext3_rename",
2460                                 "Deleting old file (%lu), %d, error=%d",
2461                                 old_dir->i_ino, old_dir->i_nlink, retval);
2462         }
2463
2464         if (new_inode) {
2465                 drop_nlink(new_inode);
2466                 new_inode->i_ctime = CURRENT_TIME_SEC;
2467         }
2468         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2469         ext3_update_dx_flag(old_dir);
2470         if (dir_bh) {
2471                 BUFFER_TRACE(dir_bh, "get_write_access");
2472                 retval = ext3_journal_get_write_access(handle, dir_bh);
2473                 if (retval)
2474                         goto journal_error;
2475                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2476                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2477                 retval = ext3_journal_dirty_metadata(handle, dir_bh);
2478                 if (retval) {
2479 journal_error:
2480                         ext3_std_error(new_dir->i_sb, retval);
2481                         goto end_rename;
2482                 }
2483                 drop_nlink(old_dir);
2484                 if (new_inode) {
2485                         drop_nlink(new_inode);
2486                 } else {
2487                         inc_nlink(new_dir);
2488                         ext3_update_dx_flag(new_dir);
2489                         ext3_mark_inode_dirty(handle, new_dir);
2490                 }
2491         }
2492         ext3_mark_inode_dirty(handle, old_dir);
2493         if (new_inode) {
2494                 ext3_mark_inode_dirty(handle, new_inode);
2495                 if (!new_inode->i_nlink)
2496                         ext3_orphan_add(handle, new_inode);
2497                 if (ext3_should_writeback_data(new_inode))
2498                         flush_file = 1;
2499         }
2500         retval = 0;
2501
2502 end_rename:
2503         brelse (dir_bh);
2504         brelse (old_bh);
2505         brelse (new_bh);
2506         ext3_journal_stop(handle);
2507         if (retval == 0 && flush_file)
2508                 filemap_flush(old_inode->i_mapping);
2509         return retval;
2510 }
2511
2512 /*
2513  * directories can handle most operations...
2514  */
2515 const struct inode_operations ext3_dir_inode_operations = {
2516         .create         = ext3_create,
2517         .lookup         = ext3_lookup,
2518         .link           = ext3_link,
2519         .unlink         = ext3_unlink,
2520         .symlink        = ext3_symlink,
2521         .mkdir          = ext3_mkdir,
2522         .rmdir          = ext3_rmdir,
2523         .mknod          = ext3_mknod,
2524         .rename         = ext3_rename,
2525         .setattr        = ext3_setattr,
2526 #ifdef CONFIG_EXT3_FS_XATTR
2527         .setxattr       = generic_setxattr,
2528         .getxattr       = generic_getxattr,
2529         .listxattr      = ext3_listxattr,
2530         .removexattr    = generic_removexattr,
2531 #endif
2532         .get_acl        = ext3_get_acl,
2533 };
2534
2535 const struct inode_operations ext3_special_inode_operations = {
2536         .setattr        = ext3_setattr,
2537 #ifdef CONFIG_EXT3_FS_XATTR
2538         .setxattr       = generic_setxattr,
2539         .getxattr       = generic_getxattr,
2540         .listxattr      = ext3_listxattr,
2541         .removexattr    = generic_removexattr,
2542 #endif
2543         .get_acl        = ext3_get_acl,
2544 };