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