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 #include <linux/smp_lock.h>
  40 #include <linux/vs_tag.h>
  41
  42 #include "namei.h"
  43 #include "xattr.h"
  44 #include "acl.h"
  45
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  53
  54 static struct buffer_head *ext3_append(handle_t *handle,
  55                                         struct inode *inode,
  56                                         u32 *block, int *err)
  57 {
  58         struct buffer_head *bh;
  59
  60         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  61
  62         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
  63                 inode->i_size += inode->i_sb->s_blocksize;
  64                 EXT3_I(inode)->i_disksize = inode->i_size;
  65                 ext3_journal_get_write_access(handle,bh);
  66         }
  67         return bh;
  68 }
  69
  70 #ifndef assert
  71 #define assert(test) J_ASSERT(test)
  72 #endif
  73
  74 #ifndef swap
  75 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  76 #endif
  77
  78 #ifdef DX_DEBUG
  79 #define dxtrace(command) command
  80 #else
  81 #define dxtrace(command)
  82 #endif
  83
  84 struct fake_dirent
  85 {
  86         __le32 inode;
  87         __le16 rec_len;
  88         u8 name_len;
  89         u8 file_type;
  90 };
  91
  92 struct dx_countlimit
  93 {
  94         __le16 limit;
  95         __le16 count;
  96 };
  97
  98 struct dx_entry
  99 {
 100         __le32 hash;
 101         __le32 block;
 102 };
 103
 104 /*
 105  * dx_root_info is laid out so that if it should somehow get overlaid by a
 106  * dirent the two low bits of the hash version will be zero.  Therefore, the
 107  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 108  */
 109
 110 struct dx_root
 111 {
 112         struct fake_dirent dot;
 113         char dot_name[4];
 114         struct fake_dirent dotdot;
 115         char dotdot_name[4];
 116         struct dx_root_info
 117         {
 118                 __le32 reserved_zero;
 119                 u8 hash_version;
 120                 u8 info_length; /* 8 */
 121                 u8 indirect_levels;
 122                 u8 unused_flags;
 123         }
 124         info;
 125         struct dx_entry entries[0];
 126 };
 127
 128 struct dx_node
 129 {
 130         struct fake_dirent fake;
 131         struct dx_entry entries[0];
 132 };
 133
 134
 135 struct dx_frame
 136 {
 137         struct buffer_head *bh;
 138         struct dx_entry *entries;
 139         struct dx_entry *at;
 140 };
 141
 142 struct dx_map_entry
 143 {
 144         u32 hash;
 145         u32 offs;
 146 };
 147
 148 #ifdef CONFIG_EXT3_INDEX
 149 static inline unsigned dx_get_block (struct dx_entry *entry);
 150 static void dx_set_block (struct dx_entry *entry, unsigned value);
 151 static inline unsigned dx_get_hash (struct dx_entry *entry);
 152 static void dx_set_hash (struct dx_entry *entry, unsigned value);
 153 static unsigned dx_get_count (struct dx_entry *entries);
 154 static unsigned dx_get_limit (struct dx_entry *entries);
 155 static void dx_set_count (struct dx_entry *entries, unsigned value);
 156 static void dx_set_limit (struct dx_entry *entries, unsigned value);
 157 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 158 static unsigned dx_node_limit (struct inode *dir);
 159 static struct dx_frame *dx_probe(struct dentry *dentry,
 160                                  struct inode *dir,
 161                                  struct dx_hash_info *hinfo,
 162                                  struct dx_frame *frame,
 163                                  int *err);
 164 static void dx_release (struct dx_frame *frames);
 165 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 166                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 167 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 168 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
 169                 struct dx_map_entry *offsets, int count);
 170 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
 171 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 172 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 173                                  struct dx_frame *frame,
 174                                  struct dx_frame *frames,
 175                                  __u32 *start_hash);
 176 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 177                        struct ext3_dir_entry_2 **res_dir, int *err);
 178 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
 179                              struct inode *inode);
 180
 181 /*
 182  * Future: use high four bits of block for coalesce-on-delete flags
 183  * Mask them off for now.
 184  */
 185
 186 static inline unsigned dx_get_block (struct dx_entry *entry)
 187 {
 188         return le32_to_cpu(entry->block) & 0x00ffffff;
 189 }
 190
 191 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 192 {
 193         entry->block = cpu_to_le32(value);
 194 }
 195
 196 static inline unsigned dx_get_hash (struct dx_entry *entry)
 197 {
 198         return le32_to_cpu(entry->hash);
 199 }
 200
 201 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 202 {
 203         entry->hash = cpu_to_le32(value);
 204 }
 205
 206 static inline unsigned dx_get_count (struct dx_entry *entries)
 207 {
 208         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 209 }
 210
 211 static inline unsigned dx_get_limit (struct dx_entry *entries)
 212 {
 213         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 214 }
 215
 216 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 217 {
 218         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 219 }
 220
 221 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 222 {
 223         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 224 }
 225
 226 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 227 {
 228         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 229                 EXT3_DIR_REC_LEN(2) - infosize;
 230         return 0? 20: entry_space / sizeof(struct dx_entry);
 231 }
 232
 233 static inline unsigned dx_node_limit (struct inode *dir)
 234 {
 235         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 236         return 0? 22: entry_space / sizeof(struct dx_entry);
 237 }
 238
 239 /*
 240  * Debug
 241  */
 242 #ifdef DX_DEBUG
 243 static void dx_show_index (char * label, struct dx_entry *entries)
 244 {
 245         int i, n = dx_get_count (entries);
 246         printk("%s index ", label);
 247         for (i = 0; i < n; i++)
 248         {
 249                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 250         }
 251         printk("\n");
 252 }
 253
 254 struct stats
 255 {
 256         unsigned names;
 257         unsigned space;
 258         unsigned bcount;
 259 };
 260
 261 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
 262                                  int size, int show_names)
 263 {
 264         unsigned names = 0, space = 0;
 265         char *base = (char *) de;
 266         struct dx_hash_info h = *hinfo;
 267
 268         printk("names: ");
 269         while ((char *) de < base + size)
 270         {
 271                 if (de->inode)
 272                 {
 273                         if (show_names)
 274                         {
 275                                 int len = de->name_len;
 276                                 char *name = de->name;
 277                                 while (len--) printk("%c", *name++);
 278                                 ext3fs_dirhash(de->name, de->name_len, &h);
 279                                 printk(":%x.%u ", h.hash,
 280                                        ((char *) de - base));
 281                         }
 282                         space += EXT3_DIR_REC_LEN(de->name_len);
 283                         names++;
 284                 }
 285                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 286         }
 287         printk("(%i)\n", names);
 288         return (struct stats) { names, space, 1 };
 289 }
 290
 291 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 292                              struct dx_entry *entries, int levels)
 293 {
 294         unsigned blocksize = dir->i_sb->s_blocksize;
 295         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 296         unsigned bcount = 0;
 297         struct buffer_head *bh;
 298         int err;
 299         printk("%i indexed blocks...\n", count);
 300         for (i = 0; i < count; i++, entries++)
 301         {
 302                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 303                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 304                 struct stats stats;
 305                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 306                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 307                 stats = levels?
 308                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 309                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
 310                 names += stats.names;
 311                 space += stats.space;
 312                 bcount += stats.bcount;
 313                 brelse (bh);
 314         }
 315         if (bcount)
 316                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 317                         names, space/bcount,(space/bcount)*100/blocksize);
 318         return (struct stats) { names, space, bcount};
 319 }
 320 #endif /* DX_DEBUG */
 321
 322 /*
 323  * Probe for a directory leaf block to search.
 324  *
 325  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 326  * error in the directory index, and the caller should fall back to
 327  * searching the directory normally.  The callers of dx_probe **MUST**
 328  * check for this error code, and make sure it never gets reflected
 329  * back to userspace.
 330  */
 331 static struct dx_frame *
 332 dx_probe(struct dentry *dentry, struct inode *dir,
 333          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 334 {
 335         unsigned count, indirect;
 336         struct dx_entry *at, *entries, *p, *q, *m;
 337         struct dx_root *root;
 338         struct buffer_head *bh;
 339         struct dx_frame *frame = frame_in;
 340         u32 hash;
 341
 342         frame->bh = NULL;
 343         if (dentry)
 344                 dir = dentry->d_parent->d_inode;
 345         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
 346                 goto fail;
 347         root = (struct dx_root *) bh->b_data;
 348         if (root->info.hash_version != DX_HASH_TEA &&
 349             root->info.hash_version != DX_HASH_HALF_MD4 &&
 350             root->info.hash_version != DX_HASH_LEGACY) {
 351                 ext3_warning(dir->i_sb, __FUNCTION__,
 352                              "Unrecognised inode hash code %d",
 353                              root->info.hash_version);
 354                 brelse(bh);
 355                 *err = ERR_BAD_DX_DIR;
 356                 goto fail;
 357         }
 358         hinfo->hash_version = root->info.hash_version;
 359         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 360         if (dentry)
 361                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
 362         hash = hinfo->hash;
 363
 364         if (root->info.unused_flags & 1) {
 365                 ext3_warning(dir->i_sb, __FUNCTION__,
 366                              "Unimplemented inode hash flags: %#06x",
 367                              root->info.unused_flags);
 368                 brelse(bh);
 369                 *err = ERR_BAD_DX_DIR;
 370                 goto fail;
 371         }
 372
 373         if ((indirect = root->info.indirect_levels) > 1) {
 374                 ext3_warning(dir->i_sb, __FUNCTION__,
 375                              "Unimplemented inode hash depth: %#06x",
 376                              root->info.indirect_levels);
 377                 brelse(bh);
 378                 *err = ERR_BAD_DX_DIR;
 379                 goto fail;
 380         }
 381
 382         entries = (struct dx_entry *) (((char *)&root->info) +
 383                                        root->info.info_length);
 384         assert(dx_get_limit(entries) == dx_root_limit(dir,
 385                                                       root->info.info_length));
 386         dxtrace (printk("Look up %x", hash));
 387         while (1)
 388         {
 389                 count = dx_get_count(entries);
 390                 assert (count && count <= dx_get_limit(entries));
 391                 p = entries + 1;
 392                 q = entries + count - 1;
 393                 while (p <= q)
 394                 {
 395                         m = p + (q - p)/2;
 396                         dxtrace(printk("."));
 397                         if (dx_get_hash(m) > hash)
 398                                 q = m - 1;
 399                         else
 400                                 p = m + 1;
 401                 }
 402
 403                 if (0) // linear search cross check
 404                 {
 405                         unsigned n = count - 1;
 406                         at = entries;
 407                         while (n--)
 408                         {
 409                                 dxtrace(printk(","));
 410                                 if (dx_get_hash(++at) > hash)
 411                                 {
 412                                         at--;
 413                                         break;
 414                                 }
 415                         }
 416                         assert (at == p - 1);
 417                 }
 418
 419                 at = p - 1;
 420                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 421                 frame->bh = bh;
 422                 frame->entries = entries;
 423                 frame->at = at;
 424                 if (!indirect--) return frame;
 425                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
 426                         goto fail2;
 427                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 428                 assert (dx_get_limit(entries) == dx_node_limit (dir));
 429                 frame++;
 430         }
 431 fail2:
 432         while (frame >= frame_in) {
 433                 brelse(frame->bh);
 434                 frame--;
 435         }
 436 fail:
 437         return NULL;
 438 }
 439
 440 static void dx_release (struct dx_frame *frames)
 441 {
 442         if (frames[0].bh == NULL)
 443                 return;
 444
 445         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 446                 brelse(frames[1].bh);
 447         brelse(frames[0].bh);
 448 }
 449
 450 /*
 451  * This function increments the frame pointer to search the next leaf
 452  * block, and reads in the necessary intervening nodes if the search
 453  * should be necessary.  Whether or not the search is necessary is
 454  * controlled by the hash parameter.  If the hash value is even, then
 455  * the search is only continued if the next block starts with that
 456  * hash value.  This is used if we are searching for a specific file.
 457  *
 458  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 459  *
 460  * This function returns 1 if the caller should continue to search,
 461  * or 0 if it should not.  If there is an error reading one of the
 462  * index blocks, it will a negative error code.
 463  *
 464  * If start_hash is non-null, it will be filled in with the starting
 465  * hash of the next page.
 466  */
 467 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 468                                  struct dx_frame *frame,
 469                                  struct dx_frame *frames,
 470                                  __u32 *start_hash)
 471 {
 472         struct dx_frame *p;
 473         struct buffer_head *bh;
 474         int err, num_frames = 0;
 475         __u32 bhash;
 476
 477         p = frame;
 478         /*
 479          * Find the next leaf page by incrementing the frame pointer.
 480          * If we run out of entries in the interior node, loop around and
 481          * increment pointer in the parent node.  When we break out of
 482          * this loop, num_frames indicates the number of interior
 483          * nodes need to be read.
 484          */
 485         while (1) {
 486                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 487                         break;
 488                 if (p == frames)
 489                         return 0;
 490                 num_frames++;
 491                 p--;
 492         }
 493
 494         /*
 495          * If the hash is 1, then continue only if the next page has a
 496          * continuation hash of any value.  This is used for readdir
 497          * handling.  Otherwise, check to see if the hash matches the
 498          * desired contiuation hash.  If it doesn't, return since
 499          * there's no point to read in the successive index pages.
 500          */
 501         bhash = dx_get_hash(p->at);
 502         if (start_hash)
 503                 *start_hash = bhash;
 504         if ((hash & 1) == 0) {
 505                 if ((bhash & ~1) != hash)
 506                         return 0;
 507         }
 508         /*
 509          * If the hash is HASH_NB_ALWAYS, we always go to the next
 510          * block so no check is necessary
 511          */
 512         while (num_frames--) {
 513                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
 514                                       0, &err)))
 515                         return err; /* Failure */
 516                 p++;
 517                 brelse (p->bh);
 518                 p->bh = bh;
 519                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 520         }
 521         return 1;
 522 }
 523
 524
 525 /*
 526  * p is at least 6 bytes before the end of page
 527  */
 528 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
 529 {
 530         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
 531 }
 532
 533 /*
 534  * This function fills a red-black tree with information from a
 535  * directory block.  It returns the number directory entries loaded
 536  * into the tree.  If there is an error it is returned in err.
 537  */
 538 static int htree_dirblock_to_tree(struct file *dir_file,
 539                                   struct inode *dir, int block,
 540                                   struct dx_hash_info *hinfo,
 541                                   __u32 start_hash, __u32 start_minor_hash)
 542 {
 543         struct buffer_head *bh;
 544         struct ext3_dir_entry_2 *de, *top;
 545         int err, count = 0;
 546
 547         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
 548         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
 549                 return err;
 550
 551         de = (struct ext3_dir_entry_2 *) bh->b_data;
 552         top = (struct ext3_dir_entry_2 *) ((char *) de +
 553                                            dir->i_sb->s_blocksize -
 554                                            EXT3_DIR_REC_LEN(0));
 555         for (; de < top; de = ext3_next_entry(de)) {
 556                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 557                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
 558                                                 +((char *)de - bh->b_data))) {
 559                         /* On error, skip the f_pos to the next block. */
 560                         dir_file->f_pos = (dir_file->f_pos |
 561                                         (dir->i_sb->s_blocksize - 1)) + 1;
 562                         brelse (bh);
 563                         return count;
 564                 }
 565                 ext3fs_dirhash(de->name, de->name_len, hinfo);
 566                 if ((hinfo->hash < start_hash) ||
 567                     ((hinfo->hash == start_hash) &&
 568                      (hinfo->minor_hash < start_minor_hash)))
 569                         continue;
 570                 if (de->inode == 0)
 571                         continue;
 572                 if ((err = ext3_htree_store_dirent(dir_file,
 573                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 574                         brelse(bh);
 575                         return err;
 576                 }
 577                 count++;
 578         }
 579         brelse(bh);
 580         return count;
 581 }
 582
 583
 584 /*
 585  * This function fills a red-black tree with information from a
 586  * directory.  We start scanning the directory in hash order, starting
 587  * at start_hash and start_minor_hash.
 588  *
 589  * This function returns the number of entries inserted into the tree,
 590  * or a negative error code.
 591  */
 592 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 593                          __u32 start_minor_hash, __u32 *next_hash)
 594 {
 595         struct dx_hash_info hinfo;
 596         struct ext3_dir_entry_2 *de;
 597         struct dx_frame frames[2], *frame;
 598         struct inode *dir;
 599         int block, err;
 600         int count = 0;
 601         int ret;
 602         __u32 hashval;
 603
 604         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
 605                        start_minor_hash));
 606         dir = dir_file->f_path.dentry->d_inode;
 607         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
 608                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
 609                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 610                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 611                                                start_hash, start_minor_hash);
 612                 *next_hash = ~0;
 613                 return count;
 614         }
 615         hinfo.hash = start_hash;
 616         hinfo.minor_hash = 0;
 617         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
 618         if (!frame)
 619                 return err;
 620
 621         /* Add '.' and '..' from the htree header */
 622         if (!start_hash && !start_minor_hash) {
 623                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 624                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 625                         goto errout;
 626                 count++;
 627         }
 628         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 629                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 630                 de = ext3_next_entry(de);
 631                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 632                         goto errout;
 633                 count++;
 634         }
 635
 636         while (1) {
 637                 block = dx_get_block(frame->at);
 638                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 639                                              start_hash, start_minor_hash);
 640                 if (ret < 0) {
 641                         err = ret;
 642                         goto errout;
 643                 }
 644                 count += ret;
 645                 hashval = ~0;
 646                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 647                                             frame, frames, &hashval);
 648                 *next_hash = hashval;
 649                 if (ret < 0) {
 650                         err = ret;
 651                         goto errout;
 652                 }
 653                 /*
 654                  * Stop if:  (a) there are no more entries, or
 655                  * (b) we have inserted at least one entry and the
 656                  * next hash value is not a continuation
 657                  */
 658                 if ((ret == 0) ||
 659                     (count && ((hashval & 1) == 0)))
 660                         break;
 661         }
 662         dx_release(frames);
 663         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 664                        count, *next_hash));
 665         return count;
 666 errout:
 667         dx_release(frames);
 668         return (err);
 669 }
 670
 671
 672 /*
 673  * Directory block splitting, compacting
 674  */
 675
 676 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 677                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 678 {
 679         int count = 0;
 680         char *base = (char *) de;
 681         struct dx_hash_info h = *hinfo;
 682
 683         while ((char *) de < base + size)
 684         {
 685                 if (de->name_len && de->inode) {
 686                         ext3fs_dirhash(de->name, de->name_len, &h);
 687                         map_tail--;
 688                         map_tail->hash = h.hash;
 689                         map_tail->offs = (u32) ((char *) de - base);
 690                         count++;
 691                         cond_resched();
 692                 }
 693                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 694                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 695         }
 696         return count;
 697 }
 698
 699 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 700 {
 701         struct dx_map_entry *p, *q, *top = map + count - 1;
 702         int more;
 703         /* Combsort until bubble sort doesn't suck */
 704         while (count > 2)
 705         {
 706                 count = count*10/13;
 707                 if (count - 9 < 2) /* 9, 10 -> 11 */
 708                         count = 11;
 709                 for (p = top, q = p - count; q >= map; p--, q--)
 710                         if (p->hash < q->hash)
 711                                 swap(*p, *q);
 712         }
 713         /* Garden variety bubble sort */
 714         do {
 715                 more = 0;
 716                 q = top;
 717                 while (q-- > map)
 718                 {
 719                         if (q[1].hash >= q[0].hash)
 720                                 continue;
 721                         swap(*(q+1), *q);
 722                         more = 1;
 723                 }
 724         } while(more);
 725 }
 726
 727 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 728 {
 729         struct dx_entry *entries = frame->entries;
 730         struct dx_entry *old = frame->at, *new = old + 1;
 731         int count = dx_get_count(entries);
 732
 733         assert(count < dx_get_limit(entries));
 734         assert(old < entries + count);
 735         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 736         dx_set_hash(new, hash);
 737         dx_set_block(new, block);
 738         dx_set_count(entries, count + 1);
 739 }
 740 #endif
 741
 742
 743 static void ext3_update_dx_flag(struct inode *inode)
 744 {
 745         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 746                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 747                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 748 }
 749
 750 /*
 751  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 752  *
 753  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 754  * `de != NULL' is guaranteed by caller.
 755  */
 756 static inline int ext3_match (int len, const char * const name,
 757                               struct ext3_dir_entry_2 * de)
 758 {
 759         if (len != de->name_len)
 760                 return 0;
 761         if (!de->inode)
 762                 return 0;
 763         return !memcmp(name, de->name, len);
 764 }
 765
 766 /*
 767  * Returns 0 if not found, -1 on failure, and 1 on success
 768  */
 769 static inline int search_dirblock(struct buffer_head * bh,
 770                                   struct inode *dir,
 771                                   struct dentry *dentry,
 772                                   unsigned long offset,
 773                                   struct ext3_dir_entry_2 ** res_dir)
 774 {
 775         struct ext3_dir_entry_2 * de;
 776         char * dlimit;
 777         int de_len;
 778         const char *name = dentry->d_name.name;
 779         int namelen = dentry->d_name.len;
 780
 781         de = (struct ext3_dir_entry_2 *) bh->b_data;
 782         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 783         while ((char *) de < dlimit) {
 784                 /* this code is executed quadratically often */
 785                 /* do minimal checking `by hand' */
 786
 787                 if ((char *) de + namelen <= dlimit &&
 788                     ext3_match (namelen, name, de)) {
 789                         /* found a match - just to be sure, do a full check */
 790                         if (!ext3_check_dir_entry("ext3_find_entry",
 791                                                   dir, de, bh, offset))
 792                                 return -1;
 793                         *res_dir = de;
 794                         return 1;
 795                 }
 796                 /* prevent looping on a bad block */
 797                 de_len = le16_to_cpu(de->rec_len);
 798                 if (de_len <= 0)
 799                         return -1;
 800                 offset += de_len;
 801                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 802         }
 803         return 0;
 804 }
 805
 806
 807 /*
 808  *      ext3_find_entry()
 809  *
 810  * finds an entry in the specified directory with the wanted name. It
 811  * returns the cache buffer in which the entry was found, and the entry
 812  * itself (as a parameter - res_dir). It does NOT read the inode of the
 813  * entry - you'll have to do that yourself if you want to.
 814  *
 815  * The returned buffer_head has ->b_count elevated.  The caller is expected
 816  * to brelse() it when appropriate.
 817  */
 818 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 819                                         struct ext3_dir_entry_2 ** res_dir)
 820 {
 821         struct super_block * sb;
 822         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 823         struct buffer_head * bh, *ret = NULL;
 824         unsigned long start, block, b;
 825         int ra_max = 0;         /* Number of bh's in the readahead
 826                                    buffer, bh_use[] */
 827         int ra_ptr = 0;         /* Current index into readahead
 828                                    buffer */
 829         int num = 0;
 830         int nblocks, i, err;
 831         struct inode *dir = dentry->d_parent->d_inode;
 832         int namelen;
 833         const u8 *name;
 834         unsigned blocksize;
 835
 836         *res_dir = NULL;
 837         sb = dir->i_sb;
 838         blocksize = sb->s_blocksize;
 839         namelen = dentry->d_name.len;
 840         name = dentry->d_name.name;
 841         if (namelen > EXT3_NAME_LEN)
 842                 return NULL;
 843 #ifdef CONFIG_EXT3_INDEX
 844         if (is_dx(dir)) {
 845                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
 846                 /*
 847                  * On success, or if the error was file not found,
 848                  * return.  Otherwise, fall back to doing a search the
 849                  * old fashioned way.
 850                  */
 851                 if (bh || (err != ERR_BAD_DX_DIR))
 852                         return bh;
 853                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 854         }
 855 #endif
 856         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 857         start = EXT3_I(dir)->i_dir_start_lookup;
 858         if (start >= nblocks)
 859                 start = 0;
 860         block = start;
 861 restart:
 862         do {
 863                 /*
 864                  * We deal with the read-ahead logic here.
 865                  */
 866                 if (ra_ptr >= ra_max) {
 867                         /* Refill the readahead buffer */
 868                         ra_ptr = 0;
 869                         b = block;
 870                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 871                                 /*
 872                                  * Terminate if we reach the end of the
 873                                  * directory and must wrap, or if our
 874                                  * search has finished at this block.
 875                                  */
 876                                 if (b >= nblocks || (num && block == start)) {
 877                                         bh_use[ra_max] = NULL;
 878                                         break;
 879                                 }
 880                                 num++;
 881                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 882                                 bh_use[ra_max] = bh;
 883                                 if (bh)
 884                                         ll_rw_block(READ_META, 1, &bh);
 885                         }
 886                 }
 887                 if ((bh = bh_use[ra_ptr++]) == NULL)
 888                         goto next;
 889                 wait_on_buffer(bh);
 890                 if (!buffer_uptodate(bh)) {
 891                         /* read error, skip block & hope for the best */
 892                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
 893                                    "offset %lu", dir->i_ino, block);
 894                         brelse(bh);
 895                         goto next;
 896                 }
 897                 i = search_dirblock(bh, dir, dentry,
 898                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 899                 if (i == 1) {
 900                         EXT3_I(dir)->i_dir_start_lookup = block;
 901                         ret = bh;
 902                         goto cleanup_and_exit;
 903                 } else {
 904                         brelse(bh);
 905                         if (i < 0)
 906                                 goto cleanup_and_exit;
 907                 }
 908         next:
 909                 if (++block >= nblocks)
 910                         block = 0;
 911         } while (block != start);
 912
 913         /*
 914          * If the directory has grown while we were searching, then
 915          * search the last part of the directory before giving up.
 916          */
 917         block = nblocks;
 918         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 919         if (block < nblocks) {
 920                 start = 0;
 921                 goto restart;
 922         }
 923
 924 cleanup_and_exit:
 925         /* Clean up the read-ahead blocks */
 926         for (; ra_ptr < ra_max; ra_ptr++)
 927                 brelse (bh_use[ra_ptr]);
 928         return ret;
 929 }
 930
 931 #ifdef CONFIG_EXT3_INDEX
 932 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 933                        struct ext3_dir_entry_2 **res_dir, int *err)
 934 {
 935         struct super_block * sb;
 936         struct dx_hash_info     hinfo;
 937         u32 hash;
 938         struct dx_frame frames[2], *frame;
 939         struct ext3_dir_entry_2 *de, *top;
 940         struct buffer_head *bh;
 941         unsigned long block;
 942         int retval;
 943         int namelen = dentry->d_name.len;
 944         const u8 *name = dentry->d_name.name;
 945         struct inode *dir = dentry->d_parent->d_inode;
 946
 947         sb = dir->i_sb;
 948         /* NFS may look up ".." - look at dx_root directory block */
 949         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
 950                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
 951                         return NULL;
 952         } else {
 953                 frame = frames;
 954                 frame->bh = NULL;                       /* for dx_release() */
 955                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
 956                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
 957         }
 958         hash = hinfo.hash;
 959         do {
 960                 block = dx_get_block(frame->at);
 961                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
 962                         goto errout;
 963                 de = (struct ext3_dir_entry_2 *) bh->b_data;
 964                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
 965                                        EXT3_DIR_REC_LEN(0));
 966                 for (; de < top; de = ext3_next_entry(de))
 967                 if (ext3_match (namelen, name, de)) {
 968                         if (!ext3_check_dir_entry("ext3_find_entry",
 969                                                   dir, de, bh,
 970                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
 971                                           +((char *)de - bh->b_data))) {
 972                                 brelse (bh);
 973                                 goto errout;
 974                         }
 975                         *res_dir = de;
 976                         dx_release (frames);
 977                         return bh;
 978                 }
 979                 brelse (bh);
 980                 /* Check to see if we should continue to search */
 981                 retval = ext3_htree_next_block(dir, hash, frame,
 982                                                frames, NULL);
 983                 if (retval < 0) {
 984                         ext3_warning(sb, __FUNCTION__,
 985                              "error reading index page in directory #%lu",
 986                              dir->i_ino);
 987                         *err = retval;
 988                         goto errout;
 989                 }
 990         } while (retval == 1);
 991
 992         *err = -ENOENT;
 993 errout:
 994         dxtrace(printk("%s not found\n", name));
 995         dx_release (frames);
 996         return NULL;
 997 }
 998 #endif
 999
1000 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1001 {
1002         struct inode * inode;
1003         struct ext3_dir_entry_2 * de;
1004         struct buffer_head * bh;
1005
1006         if (dentry->d_name.len > EXT3_NAME_LEN)
1007                 return ERR_PTR(-ENAMETOOLONG);
1008
1009         bh = ext3_find_entry(dentry, &de);
1010         inode = NULL;
1011         if (bh) {
1012                 unsigned long ino = le32_to_cpu(de->inode);
1013                 brelse (bh);
1014                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1015                         ext3_error(dir->i_sb, "ext3_lookup",
1016                                    "bad inode number: %lu", ino);
1017                         inode = NULL;
1018                 } else
1019                         inode = iget(dir->i_sb, ino);
1020
1021                 if (!inode)
1022                         return ERR_PTR(-EACCES);
1023                 dx_propagate_tag(nd, inode);
1024         }
1025         return d_splice_alias(inode, dentry);
1026 }
1027
1028
1029 struct dentry *ext3_get_parent(struct dentry *child)
1030 {
1031         unsigned long ino;
1032         struct dentry *parent;
1033         struct inode *inode;
1034         struct dentry dotdot;
1035         struct ext3_dir_entry_2 * de;
1036         struct buffer_head *bh;
1037
1038         dotdot.d_name.name = "..";
1039         dotdot.d_name.len = 2;
1040         dotdot.d_parent = child; /* confusing, isn't it! */
1041
1042         bh = ext3_find_entry(&dotdot, &de);
1043         inode = NULL;
1044         if (!bh)
1045                 return ERR_PTR(-ENOENT);
1046         ino = le32_to_cpu(de->inode);
1047         brelse(bh);
1048
1049         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1050                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1051                            "bad inode number: %lu", ino);
1052                 inode = NULL;
1053         } else
1054                 inode = iget(child->d_inode->i_sb, ino);
1055
1056         if (!inode)
1057                 return ERR_PTR(-EACCES);
1058
1059         parent = d_alloc_anon(inode);
1060         if (!parent) {
1061                 iput(inode);
1062                 parent = ERR_PTR(-ENOMEM);
1063         }
1064         return parent;
1065 }
1066
1067 #define S_SHIFT 12
1068 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1069         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1070         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1071         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1072         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1073         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1074         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1075         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1076 };
1077
1078 static inline void ext3_set_de_type(struct super_block *sb,
1079                                 struct ext3_dir_entry_2 *de,
1080                                 umode_t mode) {
1081         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1082                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1083 }
1084
1085 #ifdef CONFIG_EXT3_INDEX
1086 static struct ext3_dir_entry_2 *
1087 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1088 {
1089         unsigned rec_len = 0;
1090
1091         while (count--) {
1092                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1093                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1094                 memcpy (to, de, rec_len);
1095                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1096                                 cpu_to_le16(rec_len);
1097                 de->inode = 0;
1098                 map++;
1099                 to += rec_len;
1100         }
1101         return (struct ext3_dir_entry_2 *) (to - rec_len);
1102 }
1103
1104 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1105 {
1106         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1107         unsigned rec_len = 0;
1108
1109         prev = to = de;
1110         while ((char*)de < base + size) {
1111                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1112                                                     le16_to_cpu(de->rec_len));
1113                 if (de->inode && de->name_len) {
1114                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1115                         if (de > to)
1116                                 memmove(to, de, rec_len);
1117                         to->rec_len = cpu_to_le16(rec_len);
1118                         prev = to;
1119                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1120                 }
1121                 de = next;
1122         }
1123         return prev;
1124 }
1125
1126 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1127                         struct buffer_head **bh,struct dx_frame *frame,
1128                         struct dx_hash_info *hinfo, int *error)
1129 {
1130         unsigned blocksize = dir->i_sb->s_blocksize;
1131         unsigned count, continued;
1132         struct buffer_head *bh2;
1133         u32 newblock;
1134         u32 hash2;
1135         struct dx_map_entry *map;
1136         char *data1 = (*bh)->b_data, *data2;
1137         unsigned split;
1138         struct ext3_dir_entry_2 *de = NULL, *de2;
1139         int     err;
1140
1141         bh2 = ext3_append (handle, dir, &newblock, error);
1142         if (!(bh2)) {
1143                 brelse(*bh);
1144                 *bh = NULL;
1145                 goto errout;
1146         }
1147
1148         BUFFER_TRACE(*bh, "get_write_access");
1149         err = ext3_journal_get_write_access(handle, *bh);
1150         if (err) {
1151         journal_error:
1152                 brelse(*bh);
1153                 brelse(bh2);
1154                 *bh = NULL;
1155                 ext3_std_error(dir->i_sb, err);
1156                 goto errout;
1157         }
1158         BUFFER_TRACE(frame->bh, "get_write_access");
1159         err = ext3_journal_get_write_access(handle, frame->bh);
1160         if (err)
1161                 goto journal_error;
1162
1163         data2 = bh2->b_data;
1164
1165         /* create map in the end of data2 block */
1166         map = (struct dx_map_entry *) (data2 + blocksize);
1167         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1168                              blocksize, hinfo, map);
1169         map -= count;
1170         split = count/2; // need to adjust to actual middle
1171         dx_sort_map (map, count);
1172         hash2 = map[split].hash;
1173         continued = hash2 == map[split - 1].hash;
1174         dxtrace(printk("Split block %i at %x, %i/%i\n",
1175                 dx_get_block(frame->at), hash2, split, count-split));
1176
1177         /* Fancy dance to stay within two buffers */
1178         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1179         de = dx_pack_dirents(data1,blocksize);
1180         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1181         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1182         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1183         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1184
1185         /* Which block gets the new entry? */
1186         if (hinfo->hash >= hash2)
1187         {
1188                 swap(*bh, bh2);
1189                 de = de2;
1190         }
1191         dx_insert_block (frame, hash2 + continued, newblock);
1192         err = ext3_journal_dirty_metadata (handle, bh2);
1193         if (err)
1194                 goto journal_error;
1195         err = ext3_journal_dirty_metadata (handle, frame->bh);
1196         if (err)
1197                 goto journal_error;
1198         brelse (bh2);
1199         dxtrace(dx_show_index ("frame", frame->entries));
1200 errout:
1201         return de;
1202 }
1203 #endif
1204
1205
1206 /*
1207  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1208  * it points to a directory entry which is guaranteed to be large
1209  * enough for new directory entry.  If de is NULL, then
1210  * add_dirent_to_buf will attempt search the directory block for
1211  * space.  It will return -ENOSPC if no space is available, and -EIO
1212  * and -EEXIST if directory entry already exists.
1213  *
1214  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1215  * all other cases bh is released.
1216  */
1217 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1218                              struct inode *inode, struct ext3_dir_entry_2 *de,
1219                              struct buffer_head * bh)
1220 {
1221         struct inode    *dir = dentry->d_parent->d_inode;
1222         const char      *name = dentry->d_name.name;
1223         int             namelen = dentry->d_name.len;
1224         unsigned long   offset = 0;
1225         unsigned short  reclen;
1226         int             nlen, rlen, err;
1227         char            *top;
1228
1229         reclen = EXT3_DIR_REC_LEN(namelen);
1230         if (!de) {
1231                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1232                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1233                 while ((char *) de <= top) {
1234                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1235                                                   bh, offset)) {
1236                                 brelse (bh);
1237                                 return -EIO;
1238                         }
1239                         if (ext3_match (namelen, name, de)) {
1240                                 brelse (bh);
1241                                 return -EEXIST;
1242                         }
1243                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1244                         rlen = le16_to_cpu(de->rec_len);
1245                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1246                                 break;
1247                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1248                         offset += rlen;
1249                 }
1250                 if ((char *) de > top)
1251                         return -ENOSPC;
1252         }
1253         BUFFER_TRACE(bh, "get_write_access");
1254         err = ext3_journal_get_write_access(handle, bh);
1255         if (err) {
1256                 ext3_std_error(dir->i_sb, err);
1257                 brelse(bh);
1258                 return err;
1259         }
1260
1261         /* By now the buffer is marked for journaling */
1262         nlen = EXT3_DIR_REC_LEN(de->name_len);
1263         rlen = le16_to_cpu(de->rec_len);
1264         if (de->inode) {
1265                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1266                 de1->rec_len = cpu_to_le16(rlen - nlen);
1267                 de->rec_len = cpu_to_le16(nlen);
1268                 de = de1;
1269         }
1270         de->file_type = EXT3_FT_UNKNOWN;
1271         if (inode) {
1272                 de->inode = cpu_to_le32(inode->i_ino);
1273                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1274         } else
1275                 de->inode = 0;
1276         de->name_len = namelen;
1277         memcpy (de->name, name, namelen);
1278         /*
1279          * XXX shouldn't update any times until successful
1280          * completion of syscall, but too many callers depend
1281          * on this.
1282          *
1283          * XXX similarly, too many callers depend on
1284          * ext3_new_inode() setting the times, but error
1285          * recovery deletes the inode, so the worst that can
1286          * happen is that the times are slightly out of date
1287          * and/or different from the directory change time.
1288          */
1289         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1290         ext3_update_dx_flag(dir);
1291         dir->i_version++;
1292         ext3_mark_inode_dirty(handle, dir);
1293         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1294         err = ext3_journal_dirty_metadata(handle, bh);
1295         if (err)
1296                 ext3_std_error(dir->i_sb, err);
1297         brelse(bh);
1298         return 0;
1299 }
1300
1301 #ifdef CONFIG_EXT3_INDEX
1302 /*
1303  * This converts a one block unindexed directory to a 3 block indexed
1304  * directory, and adds the dentry to the indexed directory.
1305  */
1306 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1307                             struct inode *inode, struct buffer_head *bh)
1308 {
1309         struct inode    *dir = dentry->d_parent->d_inode;
1310         const char      *name = dentry->d_name.name;
1311         int             namelen = dentry->d_name.len;
1312         struct buffer_head *bh2;
1313         struct dx_root  *root;
1314         struct dx_frame frames[2], *frame;
1315         struct dx_entry *entries;
1316         struct ext3_dir_entry_2 *de, *de2;
1317         char            *data1, *top;
1318         unsigned        len;
1319         int             retval;
1320         unsigned        blocksize;
1321         struct dx_hash_info hinfo;
1322         u32             block;
1323         struct fake_dirent *fde;
1324
1325         blocksize =  dir->i_sb->s_blocksize;
1326         dxtrace(printk("Creating index\n"));
1327         retval = ext3_journal_get_write_access(handle, bh);
1328         if (retval) {
1329                 ext3_std_error(dir->i_sb, retval);
1330                 brelse(bh);
1331                 return retval;
1332         }
1333         root = (struct dx_root *) bh->b_data;
1334
1335         bh2 = ext3_append (handle, dir, &block, &retval);
1336         if (!(bh2)) {
1337                 brelse(bh);
1338                 return retval;
1339         }
1340         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1341         data1 = bh2->b_data;
1342
1343         /* The 0th block becomes the root, move the dirents out */
1344         fde = &root->dotdot;
1345         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1346         len = ((char *) root) + blocksize - (char *) de;
1347         memcpy (data1, de, len);
1348         de = (struct ext3_dir_entry_2 *) data1;
1349         top = data1 + len;
1350         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1351                 de = de2;
1352         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1353         /* Initialize the root; the dot dirents already exist */
1354         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1355         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1356         memset (&root->info, 0, sizeof(root->info));
1357         root->info.info_length = sizeof(root->info);
1358         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1359         entries = root->entries;
1360         dx_set_block (entries, 1);
1361         dx_set_count (entries, 1);
1362         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1363
1364         /* Initialize as for dx_probe */
1365         hinfo.hash_version = root->info.hash_version;
1366         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1367         ext3fs_dirhash(name, namelen, &hinfo);
1368         frame = frames;
1369         frame->entries = entries;
1370         frame->at = entries;
1371         frame->bh = bh;
1372         bh = bh2;
1373         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1374         dx_release (frames);
1375         if (!(de))
1376                 return retval;
1377
1378         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1379 }
1380 #endif
1381
1382 /*
1383  *      ext3_add_entry()
1384  *
1385  * adds a file entry to the specified directory, using the same
1386  * semantics as ext3_find_entry(). It returns NULL if it failed.
1387  *
1388  * NOTE!! The inode part of 'de' is left at 0 - which means you
1389  * may not sleep between calling this and putting something into
1390  * the entry, as someone else might have used it while you slept.
1391  */
1392 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1393         struct inode *inode)
1394 {
1395         struct inode *dir = dentry->d_parent->d_inode;
1396         unsigned long offset;
1397         struct buffer_head * bh;
1398         struct ext3_dir_entry_2 *de;
1399         struct super_block * sb;
1400         int     retval;
1401 #ifdef CONFIG_EXT3_INDEX
1402         int     dx_fallback=0;
1403 #endif
1404         unsigned blocksize;
1405         u32 block, blocks;
1406
1407         sb = dir->i_sb;
1408         blocksize = sb->s_blocksize;
1409         if (!dentry->d_name.len)
1410                 return -EINVAL;
1411 #ifdef CONFIG_EXT3_INDEX
1412         if (is_dx(dir)) {
1413                 retval = ext3_dx_add_entry(handle, dentry, inode);
1414                 if (!retval || (retval != ERR_BAD_DX_DIR))
1415                         return retval;
1416                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1417                 dx_fallback++;
1418                 ext3_mark_inode_dirty(handle, dir);
1419         }
1420 #endif
1421         blocks = dir->i_size >> sb->s_blocksize_bits;
1422         for (block = 0, offset = 0; block < blocks; block++) {
1423                 bh = ext3_bread(handle, dir, block, 0, &retval);
1424                 if(!bh)
1425                         return retval;
1426                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1427                 if (retval != -ENOSPC)
1428                         return retval;
1429
1430 #ifdef CONFIG_EXT3_INDEX
1431                 if (blocks == 1 && !dx_fallback &&
1432                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1433                         return make_indexed_dir(handle, dentry, inode, bh);
1434 #endif
1435                 brelse(bh);
1436         }
1437         bh = ext3_append(handle, dir, &block, &retval);
1438         if (!bh)
1439                 return retval;
1440         de = (struct ext3_dir_entry_2 *) bh->b_data;
1441         de->inode = 0;
1442         de->rec_len = cpu_to_le16(blocksize);
1443         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1444 }
1445
1446 #ifdef CONFIG_EXT3_INDEX
1447 /*
1448  * Returns 0 for success, or a negative error value
1449  */
1450 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1451                              struct inode *inode)
1452 {
1453         struct dx_frame frames[2], *frame;
1454         struct dx_entry *entries, *at;
1455         struct dx_hash_info hinfo;
1456         struct buffer_head * bh;
1457         struct inode *dir = dentry->d_parent->d_inode;
1458         struct super_block * sb = dir->i_sb;
1459         struct ext3_dir_entry_2 *de;
1460         int err;
1461
1462         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1463         if (!frame)
1464                 return err;
1465         entries = frame->entries;
1466         at = frame->at;
1467
1468         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1469                 goto cleanup;
1470
1471         BUFFER_TRACE(bh, "get_write_access");
1472         err = ext3_journal_get_write_access(handle, bh);
1473         if (err)
1474                 goto journal_error;
1475
1476         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477         if (err != -ENOSPC) {
1478                 bh = NULL;
1479                 goto cleanup;
1480         }
1481
1482         /* Block full, should compress but for now just split */
1483         dxtrace(printk("using %u of %u node entries\n",
1484                        dx_get_count(entries), dx_get_limit(entries)));
1485         /* Need to split index? */
1486         if (dx_get_count(entries) == dx_get_limit(entries)) {
1487                 u32 newblock;
1488                 unsigned icount = dx_get_count(entries);
1489                 int levels = frame - frames;
1490                 struct dx_entry *entries2;
1491                 struct dx_node *node2;
1492                 struct buffer_head *bh2;
1493
1494                 if (levels && (dx_get_count(frames->entries) ==
1495                                dx_get_limit(frames->entries))) {
1496                         ext3_warning(sb, __FUNCTION__,
1497                                      "Directory index full!");
1498                         err = -ENOSPC;
1499                         goto cleanup;
1500                 }
1501                 bh2 = ext3_append (handle, dir, &newblock, &err);
1502                 if (!(bh2))
1503                         goto cleanup;
1504                 node2 = (struct dx_node *)(bh2->b_data);
1505                 entries2 = node2->entries;
1506                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1507                 node2->fake.inode = 0;
1508                 BUFFER_TRACE(frame->bh, "get_write_access");
1509                 err = ext3_journal_get_write_access(handle, frame->bh);
1510                 if (err)
1511                         goto journal_error;
1512                 if (levels) {
1513                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1514                         unsigned hash2 = dx_get_hash(entries + icount1);
1515                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1516
1517                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1518                         err = ext3_journal_get_write_access(handle,
1519                                                              frames[0].bh);
1520                         if (err)
1521                                 goto journal_error;
1522
1523                         memcpy ((char *) entries2, (char *) (entries + icount1),
1524                                 icount2 * sizeof(struct dx_entry));
1525                         dx_set_count (entries, icount1);
1526                         dx_set_count (entries2, icount2);
1527                         dx_set_limit (entries2, dx_node_limit(dir));
1528
1529                         /* Which index block gets the new entry? */
1530                         if (at - entries >= icount1) {
1531                                 frame->at = at = at - entries - icount1 + entries2;
1532                                 frame->entries = entries = entries2;
1533                                 swap(frame->bh, bh2);
1534                         }
1535                         dx_insert_block (frames + 0, hash2, newblock);
1536                         dxtrace(dx_show_index ("node", frames[1].entries));
1537                         dxtrace(dx_show_index ("node",
1538                                ((struct dx_node *) bh2->b_data)->entries));
1539                         err = ext3_journal_dirty_metadata(handle, bh2);
1540                         if (err)
1541                                 goto journal_error;
1542                         brelse (bh2);
1543                 } else {
1544                         dxtrace(printk("Creating second level index...\n"));
1545                         memcpy((char *) entries2, (char *) entries,
1546                                icount * sizeof(struct dx_entry));
1547                         dx_set_limit(entries2, dx_node_limit(dir));
1548
1549                         /* Set up root */
1550                         dx_set_count(entries, 1);
1551                         dx_set_block(entries + 0, newblock);
1552                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1553
1554                         /* Add new access path frame */
1555                         frame = frames + 1;
1556                         frame->at = at = at - entries + entries2;
1557                         frame->entries = entries = entries2;
1558                         frame->bh = bh2;
1559                         err = ext3_journal_get_write_access(handle,
1560                                                              frame->bh);
1561                         if (err)
1562                                 goto journal_error;
1563                 }
1564                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1565         }
1566         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1567         if (!de)
1568                 goto cleanup;
1569         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1570         bh = NULL;
1571         goto cleanup;
1572
1573 journal_error:
1574         ext3_std_error(dir->i_sb, err);
1575 cleanup:
1576         if (bh)
1577                 brelse(bh);
1578         dx_release(frames);
1579         return err;
1580 }
1581 #endif
1582
1583 /*
1584  * ext3_delete_entry deletes a directory entry by merging it with the
1585  * previous entry
1586  */
1587 static int ext3_delete_entry (handle_t *handle,
1588                               struct inode * dir,
1589                               struct ext3_dir_entry_2 * de_del,
1590                               struct buffer_head * bh)
1591 {
1592         struct ext3_dir_entry_2 * de, * pde;
1593         int i;
1594
1595         i = 0;
1596         pde = NULL;
1597         de = (struct ext3_dir_entry_2 *) bh->b_data;
1598         while (i < bh->b_size) {
1599                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1600                         return -EIO;
1601                 if (de == de_del)  {
1602                         BUFFER_TRACE(bh, "get_write_access");
1603                         ext3_journal_get_write_access(handle, bh);
1604                         if (pde)
1605                                 pde->rec_len =
1606                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1607                                                     le16_to_cpu(de->rec_len));
1608                         else
1609                                 de->inode = 0;
1610                         dir->i_version++;
1611                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1612                         ext3_journal_dirty_metadata(handle, bh);
1613                         return 0;
1614                 }
1615                 i += le16_to_cpu(de->rec_len);
1616                 pde = de;
1617                 de = (struct ext3_dir_entry_2 *)
1618                         ((char *) de + le16_to_cpu(de->rec_len));
1619         }
1620         return -ENOENT;
1621 }
1622
1623 /*
1624  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1625  * do not perform it in these functions.  We perform it at the call site,
1626  * if it is needed.
1627  */
1628 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1629 {
1630         inc_nlink(inode);
1631 }
1632
1633 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1634 {
1635         drop_nlink(inode);
1636 }
1637
1638 static int ext3_add_nondir(handle_t *handle,
1639                 struct dentry *dentry, struct inode *inode)
1640 {
1641         int err = ext3_add_entry(handle, dentry, inode);
1642         if (!err) {
1643                 ext3_mark_inode_dirty(handle, inode);
1644                 d_instantiate(dentry, inode);
1645                 return 0;
1646         }
1647         ext3_dec_count(handle, inode);
1648         iput(inode);
1649         return err;
1650 }
1651
1652 /*
1653  * By the time this is called, we already have created
1654  * the directory cache entry for the new file, but it
1655  * is so far negative - it has no inode.
1656  *
1657  * If the create succeeds, we fill in the inode information
1658  * with d_instantiate().
1659  */
1660 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1661                 struct nameidata *nd)
1662 {
1663         handle_t *handle;
1664         struct inode * inode;
1665         int err, retries = 0;
1666
1667 retry:
1668         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1669                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1670                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1671         if (IS_ERR(handle))
1672                 return PTR_ERR(handle);
1673
1674         if (IS_DIRSYNC(dir))
1675                 handle->h_sync = 1;
1676
1677         inode = ext3_new_inode (handle, dir, mode);
1678         err = PTR_ERR(inode);
1679         if (!IS_ERR(inode)) {
1680                 inode->i_op = &ext3_file_inode_operations;
1681                 inode->i_fop = &ext3_file_operations;
1682                 ext3_set_aops(inode);
1683                 err = ext3_add_nondir(handle, dentry, inode);
1684         }
1685         ext3_journal_stop(handle);
1686         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1687                 goto retry;
1688         return err;
1689 }
1690
1691 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1692                         int mode, dev_t rdev)
1693 {
1694         handle_t *handle;
1695         struct inode *inode;
1696         int err, retries = 0;
1697
1698         if (!new_valid_dev(rdev))
1699                 return -EINVAL;
1700
1701 retry:
1702         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1703                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1704                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1705         if (IS_ERR(handle))
1706                 return PTR_ERR(handle);
1707
1708         if (IS_DIRSYNC(dir))
1709                 handle->h_sync = 1;
1710
1711         inode = ext3_new_inode (handle, dir, mode);
1712         err = PTR_ERR(inode);
1713         if (!IS_ERR(inode)) {
1714                 init_special_inode(inode, inode->i_mode, rdev);
1715 #ifdef CONFIG_EXT3_FS_XATTR
1716                 inode->i_op = &ext3_special_inode_operations;
1717 #endif
1718                 err = ext3_add_nondir(handle, dentry, inode);
1719         }
1720         ext3_journal_stop(handle);
1721         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1722                 goto retry;
1723         return err;
1724 }
1725
1726 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1727 {
1728         handle_t *handle;
1729         struct inode * inode;
1730         struct buffer_head * dir_block;
1731         struct ext3_dir_entry_2 * de;
1732         int err, retries = 0;
1733
1734         if (dir->i_nlink >= EXT3_LINK_MAX)
1735                 return -EMLINK;
1736
1737 retry:
1738         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1739                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1740                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1741         if (IS_ERR(handle))
1742                 return PTR_ERR(handle);
1743
1744         if (IS_DIRSYNC(dir))
1745                 handle->h_sync = 1;
1746
1747         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1748         err = PTR_ERR(inode);
1749         if (IS_ERR(inode))
1750                 goto out_stop;
1751
1752         inode->i_op = &ext3_dir_inode_operations;
1753         inode->i_fop = &ext3_dir_operations;
1754         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1755         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1756         if (!dir_block) {
1757                 drop_nlink(inode); /* is this nlink == 0? */
1758                 ext3_mark_inode_dirty(handle, inode);
1759                 iput (inode);
1760                 goto out_stop;
1761         }
1762         BUFFER_TRACE(dir_block, "get_write_access");
1763         ext3_journal_get_write_access(handle, dir_block);
1764         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1765         de->inode = cpu_to_le32(inode->i_ino);
1766         de->name_len = 1;
1767         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1768         strcpy (de->name, ".");
1769         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1770         de = (struct ext3_dir_entry_2 *)
1771                         ((char *) de + le16_to_cpu(de->rec_len));
1772         de->inode = cpu_to_le32(dir->i_ino);
1773         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1774         de->name_len = 2;
1775         strcpy (de->name, "..");
1776         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1777         inode->i_nlink = 2;
1778         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1779         ext3_journal_dirty_metadata(handle, dir_block);
1780         brelse (dir_block);
1781         ext3_mark_inode_dirty(handle, inode);
1782         err = ext3_add_entry (handle, dentry, inode);
1783         if (err) {
1784                 inode->i_nlink = 0;
1785                 ext3_mark_inode_dirty(handle, inode);
1786                 iput (inode);
1787                 goto out_stop;
1788         }
1789         inc_nlink(dir);
1790         ext3_update_dx_flag(dir);
1791         ext3_mark_inode_dirty(handle, dir);
1792         d_instantiate(dentry, inode);
1793 out_stop:
1794         ext3_journal_stop(handle);
1795         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1796                 goto retry;
1797         return err;
1798 }
1799
1800 /*
1801  * routine to check that the specified directory is empty (for rmdir)
1802  */
1803 static int empty_dir (struct inode * inode)
1804 {
1805         unsigned long offset;
1806         struct buffer_head * bh;
1807         struct ext3_dir_entry_2 * de, * de1;
1808         struct super_block * sb;
1809         int err = 0;
1810
1811         sb = inode->i_sb;
1812         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1813             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1814                 if (err)
1815                         ext3_error(inode->i_sb, __FUNCTION__,
1816                                    "error %d reading directory #%lu offset 0",
1817                                    err, inode->i_ino);
1818                 else
1819                         ext3_warning(inode->i_sb, __FUNCTION__,
1820                                      "bad directory (dir #%lu) - no data block",
1821                                      inode->i_ino);
1822                 return 1;
1823         }
1824         de = (struct ext3_dir_entry_2 *) bh->b_data;
1825         de1 = (struct ext3_dir_entry_2 *)
1826                         ((char *) de + le16_to_cpu(de->rec_len));
1827         if (le32_to_cpu(de->inode) != inode->i_ino ||
1828                         !le32_to_cpu(de1->inode) ||
1829                         strcmp (".", de->name) ||
1830                         strcmp ("..", de1->name)) {
1831                 ext3_warning (inode->i_sb, "empty_dir",
1832                               "bad directory (dir #%lu) - no `.' or `..'",
1833                               inode->i_ino);
1834                 brelse (bh);
1835                 return 1;
1836         }
1837         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1838         de = (struct ext3_dir_entry_2 *)
1839                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1840         while (offset < inode->i_size ) {
1841                 if (!bh ||
1842                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1843                         err = 0;
1844                         brelse (bh);
1845                         bh = ext3_bread (NULL, inode,
1846                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1847                         if (!bh) {
1848                                 if (err)
1849                                         ext3_error(sb, __FUNCTION__,
1850                                                    "error %d reading directory"
1851                                                    " #%lu offset %lu",
1852                                                    err, inode->i_ino, offset);
1853                                 offset += sb->s_blocksize;
1854                                 continue;
1855                         }
1856                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1857                 }
1858                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1859                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1860                                                          sb->s_blocksize);
1861                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1862                         continue;
1863                 }
1864                 if (le32_to_cpu(de->inode)) {
1865                         brelse (bh);
1866                         return 0;
1867                 }
1868                 offset += le16_to_cpu(de->rec_len);
1869                 de = (struct ext3_dir_entry_2 *)
1870                                 ((char *) de + le16_to_cpu(de->rec_len));
1871         }
1872         brelse (bh);
1873         return 1;
1874 }
1875
1876 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1877  * such inodes, starting at the superblock, in case we crash before the
1878  * file is closed/deleted, or in case the inode truncate spans multiple
1879  * transactions and the last transaction is not recovered after a crash.
1880  *
1881  * At filesystem recovery time, we walk this list deleting unlinked
1882  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1883  */
1884 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1885 {
1886         struct super_block *sb = inode->i_sb;
1887         struct ext3_iloc iloc;
1888         int err = 0, rc;
1889
1890         lock_super(sb);
1891         if (!list_empty(&EXT3_I(inode)->i_orphan))
1892                 goto out_unlock;
1893
1894         /* Orphan handling is only valid for files with data blocks
1895          * being truncated, or files being unlinked. */
1896
1897         /* @@@ FIXME: Observation from aviro:
1898          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1899          * here (on lock_super()), so race with ext3_link() which might bump
1900          * ->i_nlink. For, say it, character device. Not a regular file,
1901          * not a directory, not a symlink and ->i_nlink > 0.
1902          */
1903         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1904                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1905
1906         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1907         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1908         if (err)
1909                 goto out_unlock;
1910
1911         err = ext3_reserve_inode_write(handle, inode, &iloc);
1912         if (err)
1913                 goto out_unlock;
1914
1915         /* Insert this inode at the head of the on-disk orphan list... */
1916         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1917         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1918         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1919         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1920         if (!err)
1921                 err = rc;
1922
1923         /* Only add to the head of the in-memory list if all the
1924          * previous operations succeeded.  If the orphan_add is going to
1925          * fail (possibly taking the journal offline), we can't risk
1926          * leaving the inode on the orphan list: stray orphan-list
1927          * entries can cause panics at unmount time.
1928          *
1929          * This is safe: on error we're going to ignore the orphan list
1930          * anyway on the next recovery. */
1931         if (!err)
1932                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1933
1934         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1935         jbd_debug(4, "orphan inode %lu will point to %d\n",
1936                         inode->i_ino, NEXT_ORPHAN(inode));
1937 out_unlock:
1938         unlock_super(sb);
1939         ext3_std_error(inode->i_sb, err);
1940         return err;
1941 }
1942
1943 /*
1944  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1945  * of such inodes stored on disk, because it is finally being cleaned up.
1946  */
1947 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1948 {
1949         struct list_head *prev;
1950         struct ext3_inode_info *ei = EXT3_I(inode);
1951         struct ext3_sb_info *sbi;
1952         unsigned long ino_next;
1953         struct ext3_iloc iloc;
1954         int err = 0;
1955
1956         lock_super(inode->i_sb);
1957         if (list_empty(&ei->i_orphan)) {
1958                 unlock_super(inode->i_sb);
1959                 return 0;
1960         }
1961
1962         ino_next = NEXT_ORPHAN(inode);
1963         prev = ei->i_orphan.prev;
1964         sbi = EXT3_SB(inode->i_sb);
1965
1966         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1967
1968         list_del_init(&ei->i_orphan);
1969
1970         /* If we're on an error path, we may not have a valid
1971          * transaction handle with which to update the orphan list on
1972          * disk, but we still need to remove the inode from the linked
1973          * list in memory. */
1974         if (!handle)
1975                 goto out;
1976
1977         err = ext3_reserve_inode_write(handle, inode, &iloc);
1978         if (err)
1979                 goto out_err;
1980
1981         if (prev == &sbi->s_orphan) {
1982                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1983                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1984                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1985                 if (err)
1986                         goto out_brelse;
1987                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1988                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1989         } else {
1990                 struct ext3_iloc iloc2;
1991                 struct inode *i_prev =
1992                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1993
1994                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1995                           i_prev->i_ino, ino_next);
1996                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1997                 if (err)
1998                         goto out_brelse;
1999                 NEXT_ORPHAN(i_prev) = ino_next;
2000                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2001         }
2002         if (err)
2003                 goto out_brelse;
2004         NEXT_ORPHAN(inode) = 0;
2005         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2006
2007 out_err:
2008         ext3_std_error(inode->i_sb, err);
2009 out:
2010         unlock_super(inode->i_sb);
2011         return err;
2012
2013 out_brelse:
2014         brelse(iloc.bh);
2015         goto out_err;
2016 }
2017
2018 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2019 {
2020         int retval;
2021         struct inode * inode;
2022         struct buffer_head * bh;
2023         struct ext3_dir_entry_2 * de;
2024         handle_t *handle;
2025
2026         /* Initialize quotas before so that eventual writes go in
2027          * separate transaction */
2028         DQUOT_INIT(dentry->d_inode);
2029         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2030         if (IS_ERR(handle))
2031                 return PTR_ERR(handle);
2032
2033         retval = -ENOENT;
2034         bh = ext3_find_entry (dentry, &de);
2035         if (!bh)
2036                 goto end_rmdir;
2037
2038         if (IS_DIRSYNC(dir))
2039                 handle->h_sync = 1;
2040
2041         inode = dentry->d_inode;
2042
2043         retval = -EIO;
2044         if (le32_to_cpu(de->inode) != inode->i_ino)
2045                 goto end_rmdir;
2046
2047         retval = -ENOTEMPTY;
2048         if (!empty_dir (inode))
2049                 goto end_rmdir;
2050
2051         retval = ext3_delete_entry(handle, dir, de, bh);
2052         if (retval)
2053                 goto end_rmdir;
2054         if (inode->i_nlink != 2)
2055                 ext3_warning (inode->i_sb, "ext3_rmdir",
2056                               "empty directory has nlink!=2 (%d)",
2057                               inode->i_nlink);
2058         inode->i_version++;
2059         clear_nlink(inode);
2060         /* There's no need to set i_disksize: the fact that i_nlink is
2061          * zero will ensure that the right thing happens during any
2062          * recovery. */
2063         inode->i_size = 0;
2064         ext3_orphan_add(handle, inode);
2065         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2066         ext3_mark_inode_dirty(handle, inode);
2067         drop_nlink(dir);
2068         ext3_update_dx_flag(dir);
2069         ext3_mark_inode_dirty(handle, dir);
2070
2071 end_rmdir:
2072         ext3_journal_stop(handle);
2073         brelse (bh);
2074         return retval;
2075 }
2076
2077 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2078 {
2079         int retval;
2080         struct inode * inode;
2081         struct buffer_head * bh;
2082         struct ext3_dir_entry_2 * de;
2083         handle_t *handle;
2084
2085         /* Initialize quotas before so that eventual writes go
2086          * in separate transaction */
2087         DQUOT_INIT(dentry->d_inode);
2088         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2089         if (IS_ERR(handle))
2090                 return PTR_ERR(handle);
2091
2092         if (IS_DIRSYNC(dir))
2093                 handle->h_sync = 1;
2094
2095         retval = -ENOENT;
2096         bh = ext3_find_entry (dentry, &de);
2097         if (!bh)
2098                 goto end_unlink;
2099
2100         inode = dentry->d_inode;
2101
2102         retval = -EIO;
2103         if (le32_to_cpu(de->inode) != inode->i_ino)
2104                 goto end_unlink;
2105
2106         if (!inode->i_nlink) {
2107                 ext3_warning (inode->i_sb, "ext3_unlink",
2108                               "Deleting nonexistent file (%lu), %d",
2109                               inode->i_ino, inode->i_nlink);
2110                 inode->i_nlink = 1;
2111         }
2112         retval = ext3_delete_entry(handle, dir, de, bh);
2113         if (retval)
2114                 goto end_unlink;
2115         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2116         ext3_update_dx_flag(dir);
2117         ext3_mark_inode_dirty(handle, dir);
2118         drop_nlink(inode);
2119         if (!inode->i_nlink)
2120                 ext3_orphan_add(handle, inode);
2121         inode->i_ctime = dir->i_ctime;
2122         ext3_mark_inode_dirty(handle, inode);
2123         retval = 0;
2124
2125 end_unlink:
2126         ext3_journal_stop(handle);
2127         brelse (bh);
2128         return retval;
2129 }
2130
2131 static int ext3_symlink (struct inode * dir,
2132                 struct dentry *dentry, const char * symname)
2133 {
2134         handle_t *handle;
2135         struct inode * inode;
2136         int l, err, retries = 0;
2137
2138         l = strlen(symname)+1;
2139         if (l > dir->i_sb->s_blocksize)
2140                 return -ENAMETOOLONG;
2141
2142 retry:
2143         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2144                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2145                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2146         if (IS_ERR(handle))
2147                 return PTR_ERR(handle);
2148
2149         if (IS_DIRSYNC(dir))
2150                 handle->h_sync = 1;
2151
2152         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2153         err = PTR_ERR(inode);
2154         if (IS_ERR(inode))
2155                 goto out_stop;
2156
2157         if (l > sizeof (EXT3_I(inode)->i_data)) {
2158                 inode->i_op = &ext3_symlink_inode_operations;
2159                 ext3_set_aops(inode);
2160                 /*
2161                  * page_symlink() calls into ext3_prepare/commit_write.
2162                  * We have a transaction open.  All is sweetness.  It also sets
2163                  * i_size in generic_commit_write().
2164                  */
2165                 err = __page_symlink(inode, symname, l,
2166                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2167                 if (err) {
2168                         ext3_dec_count(handle, inode);
2169                         ext3_mark_inode_dirty(handle, inode);
2170                         iput (inode);
2171                         goto out_stop;
2172                 }
2173         } else {
2174                 inode->i_op = &ext3_fast_symlink_inode_operations;
2175                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2176                 inode->i_size = l-1;
2177         }
2178         EXT3_I(inode)->i_disksize = inode->i_size;
2179         err = ext3_add_nondir(handle, dentry, inode);
2180 out_stop:
2181         ext3_journal_stop(handle);
2182         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2183                 goto retry;
2184         return err;
2185 }
2186
2187 static int ext3_link (struct dentry * old_dentry,
2188                 struct inode * dir, struct dentry *dentry)
2189 {
2190         handle_t *handle;
2191         struct inode *inode = old_dentry->d_inode;
2192         int err, retries = 0;
2193
2194         if (inode->i_nlink >= EXT3_LINK_MAX)
2195                 return -EMLINK;
2196
2197 retry:
2198         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2199                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2200         if (IS_ERR(handle))
2201                 return PTR_ERR(handle);
2202
2203         if (IS_DIRSYNC(dir))
2204                 handle->h_sync = 1;
2205
2206         inode->i_ctime = CURRENT_TIME_SEC;
2207         ext3_inc_count(handle, inode);
2208         atomic_inc(&inode->i_count);
2209
2210         err = ext3_add_nondir(handle, dentry, inode);
2211         ext3_journal_stop(handle);
2212         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2213                 goto retry;
2214         return err;
2215 }
2216
2217 #define PARENT_INO(buffer) \
2218         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2219         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2220
2221 /*
2222  * Anybody can rename anything with this: the permission checks are left to the
2223  * higher-level routines.
2224  */
2225 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2226                            struct inode * new_dir,struct dentry *new_dentry)
2227 {
2228         handle_t *handle;
2229         struct inode * old_inode, * new_inode;
2230         struct buffer_head * old_bh, * new_bh, * dir_bh;
2231         struct ext3_dir_entry_2 * old_de, * new_de;
2232         int retval;
2233
2234         old_bh = new_bh = dir_bh = NULL;
2235
2236         /* Initialize quotas before so that eventual writes go
2237          * in separate transaction */
2238         if (new_dentry->d_inode)
2239                 DQUOT_INIT(new_dentry->d_inode);
2240         handle = ext3_journal_start(old_dir, 2 *
2241                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2242                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2243         if (IS_ERR(handle))
2244                 return PTR_ERR(handle);
2245
2246         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2247                 handle->h_sync = 1;
2248
2249         old_bh = ext3_find_entry (old_dentry, &old_de);
2250         /*
2251          *  Check for inode number is _not_ due to possible IO errors.
2252          *  We might rmdir the source, keep it as pwd of some process
2253          *  and merrily kill the link to whatever was created under the
2254          *  same name. Goodbye sticky bit ;-<
2255          */
2256         old_inode = old_dentry->d_inode;
2257         retval = -ENOENT;
2258         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2259                 goto end_rename;
2260
2261         new_inode = new_dentry->d_inode;
2262         new_bh = ext3_find_entry (new_dentry, &new_de);
2263         if (new_bh) {
2264                 if (!new_inode) {
2265                         brelse (new_bh);
2266                         new_bh = NULL;
2267                 }
2268         }
2269         if (S_ISDIR(old_inode->i_mode)) {
2270                 if (new_inode) {
2271                         retval = -ENOTEMPTY;
2272                         if (!empty_dir (new_inode))
2273                                 goto end_rename;
2274                 }
2275                 retval = -EIO;
2276                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2277                 if (!dir_bh)
2278                         goto end_rename;
2279                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2280                         goto end_rename;
2281                 retval = -EMLINK;
2282                 if (!new_inode && new_dir!=old_dir &&
2283                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2284                         goto end_rename;
2285         }
2286         if (!new_bh) {
2287                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2288                 if (retval)
2289                         goto end_rename;
2290         } else {
2291                 BUFFER_TRACE(new_bh, "get write access");
2292                 ext3_journal_get_write_access(handle, new_bh);
2293                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2294                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2295                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2296                         new_de->file_type = old_de->file_type;
2297                 new_dir->i_version++;
2298                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2299                 ext3_journal_dirty_metadata(handle, new_bh);
2300                 brelse(new_bh);
2301                 new_bh = NULL;
2302         }
2303
2304         /*
2305          * Like most other Unix systems, set the ctime for inodes on a
2306          * rename.
2307          */
2308         old_inode->i_ctime = CURRENT_TIME_SEC;
2309         ext3_mark_inode_dirty(handle, old_inode);
2310
2311         /*
2312          * ok, that's it
2313          */
2314         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2315             old_de->name_len != old_dentry->d_name.len ||
2316             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2317             (retval = ext3_delete_entry(handle, old_dir,
2318                                         old_de, old_bh)) == -ENOENT) {
2319                 /* old_de could have moved from under us during htree split, so
2320                  * make sure that we are deleting the right entry.  We might
2321                  * also be pointing to a stale entry in the unused part of
2322                  * old_bh so just checking inum and the name isn't enough. */
2323                 struct buffer_head *old_bh2;
2324                 struct ext3_dir_entry_2 *old_de2;
2325
2326                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2327                 if (old_bh2) {
2328                         retval = ext3_delete_entry(handle, old_dir,
2329                                                    old_de2, old_bh2);
2330                         brelse(old_bh2);
2331                 }
2332         }
2333         if (retval) {
2334                 ext3_warning(old_dir->i_sb, "ext3_rename",
2335                                 "Deleting old file (%lu), %d, error=%d",
2336                                 old_dir->i_ino, old_dir->i_nlink, retval);
2337         }
2338
2339         if (new_inode) {
2340                 drop_nlink(new_inode);
2341                 new_inode->i_ctime = CURRENT_TIME_SEC;
2342         }
2343         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2344         ext3_update_dx_flag(old_dir);
2345         if (dir_bh) {
2346                 BUFFER_TRACE(dir_bh, "get_write_access");
2347                 ext3_journal_get_write_access(handle, dir_bh);
2348                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2349                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2350                 ext3_journal_dirty_metadata(handle, dir_bh);
2351                 drop_nlink(old_dir);
2352                 if (new_inode) {
2353                         drop_nlink(new_inode);
2354                 } else {
2355                         inc_nlink(new_dir);
2356                         ext3_update_dx_flag(new_dir);
2357                         ext3_mark_inode_dirty(handle, new_dir);
2358                 }
2359         }
2360         ext3_mark_inode_dirty(handle, old_dir);
2361         if (new_inode) {
2362                 ext3_mark_inode_dirty(handle, new_inode);
2363                 if (!new_inode->i_nlink)
2364                         ext3_orphan_add(handle, new_inode);
2365         }
2366         retval = 0;
2367
2368 end_rename:
2369         brelse (dir_bh);
2370         brelse (old_bh);
2371         brelse (new_bh);
2372         ext3_journal_stop(handle);
2373         return retval;
2374 }
2375
2376 /*
2377  * directories can handle most operations...
2378  */
2379 struct inode_operations ext3_dir_inode_operations = {
2380         .create         = ext3_create,
2381         .lookup         = ext3_lookup,
2382         .link           = ext3_link,
2383         .unlink         = ext3_unlink,
2384         .symlink        = ext3_symlink,
2385         .mkdir          = ext3_mkdir,
2386         .rmdir          = ext3_rmdir,
2387         .mknod          = ext3_mknod,
2388         .rename         = ext3_rename,
2389         .setattr        = ext3_setattr,
2390 #ifdef CONFIG_EXT3_FS_XATTR
2391         .setxattr       = generic_setxattr,
2392         .getxattr       = generic_getxattr,
2393         .listxattr      = ext3_listxattr,
2394         .removexattr    = generic_removexattr,
2395 #endif
2396         .permission     = ext3_permission,
2397         .sync_flags     = ext3_sync_flags,
2398 };
2399
2400 struct inode_operations ext3_special_inode_operations = {
2401         .setattr        = ext3_setattr,
2402 #ifdef CONFIG_EXT3_FS_XATTR
2403         .setxattr       = generic_setxattr,
2404         .getxattr       = generic_getxattr,
2405         .listxattr      = ext3_listxattr,
2406         .removexattr    = generic_removexattr,
2407 #endif
2408         .permission     = ext3_permission,
2409         .sync_flags     = ext3_sync_flags,
2410 };