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                 count++;
 615         }
 616         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 617                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 618                 de = ext3_next_entry(de);
 619                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 620                         goto errout;
 621                 count++;
 622         }
 623
 624         while (1) {
 625                 block = dx_get_block(frame->at);
 626                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 627                                              start_hash, start_minor_hash);
 628                 if (ret < 0) {
 629                         err = ret;
 630                         goto errout;
 631                 }
 632                 count += ret;
 633                 hashval = ~0;
 634                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 635                                             frame, frames, &hashval);
 636                 *next_hash = hashval;
 637                 if (ret < 0) {
 638                         err = ret;
 639                         goto errout;
 640                 }
 641                 /*
 642                  * Stop if:  (a) there are no more entries, or
 643                  * (b) we have inserted at least one entry and the
 644                  * next hash value is not a continuation
 645                  */
 646                 if ((ret == 0) ||
 647                     (count && ((hashval & 1) == 0)))
 648                         break;
 649         }
 650         dx_release(frames);
 651         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 652                        count, *next_hash));
 653         return count;
 654 errout:
 655         dx_release(frames);
 656         return (err);
 657 }
 658
 659
 660 /*
 661  * Directory block splitting, compacting
 662  */
 663
 664 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 665                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 666 {
 667         int count = 0;
 668         char *base = (char *) de;
 669         struct dx_hash_info h = *hinfo;
 670
 671         while ((char *) de < base + size)
 672         {
 673                 if (de->name_len && de->inode) {
 674                         ext3fs_dirhash(de->name, de->name_len, &h);
 675                         map_tail--;
 676                         map_tail->hash = h.hash;
 677                         map_tail->offs = (u32) ((char *) de - base);
 678                         count++;
 679                         cond_resched();
 680                 }
 681                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 682                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 683         }
 684         return count;
 685 }
 686
 687 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 688 {
 689         struct dx_map_entry *p, *q, *top = map + count - 1;
 690         int more;
 691         /* Combsort until bubble sort doesn't suck */
 692         while (count > 2)
 693         {
 694                 count = count*10/13;
 695                 if (count - 9 < 2) /* 9, 10 -> 11 */
 696                         count = 11;
 697                 for (p = top, q = p - count; q >= map; p--, q--)
 698                         if (p->hash < q->hash)
 699                                 swap(*p, *q);
 700         }
 701         /* Garden variety bubble sort */
 702         do {
 703                 more = 0;
 704                 q = top;
 705                 while (q-- > map)
 706                 {
 707                         if (q[1].hash >= q[0].hash)
 708                                 continue;
 709                         swap(*(q+1), *q);
 710                         more = 1;
 711                 }
 712         } while(more);
 713 }
 714
 715 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 716 {
 717         struct dx_entry *entries = frame->entries;
 718         struct dx_entry *old = frame->at, *new = old + 1;
 719         int count = dx_get_count(entries);
 720
 721         assert(count < dx_get_limit(entries));
 722         assert(old < entries + count);
 723         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 724         dx_set_hash(new, hash);
 725         dx_set_block(new, block);
 726         dx_set_count(entries, count + 1);
 727 }
 728 #endif
 729
 730
 731 static void ext3_update_dx_flag(struct inode *inode)
 732 {
 733         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 734                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 735                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 736 }
 737
 738 /*
 739  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 740  *
 741  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 742  * `de != NULL' is guaranteed by caller.
 743  */
 744 static inline int ext3_match (int len, const char * const name,
 745                               struct ext3_dir_entry_2 * de)
 746 {
 747         if (len != de->name_len)
 748                 return 0;
 749         if (!de->inode)
 750                 return 0;
 751         return !memcmp(name, de->name, len);
 752 }
 753
 754 /*
 755  * Returns 0 if not found, -1 on failure, and 1 on success
 756  */
 757 static inline int search_dirblock(struct buffer_head * bh,
 758                                   struct inode *dir,
 759                                   struct dentry *dentry,
 760                                   unsigned long offset,
 761                                   struct ext3_dir_entry_2 ** res_dir)
 762 {
 763         struct ext3_dir_entry_2 * de;
 764         char * dlimit;
 765         int de_len;
 766         const char *name = dentry->d_name.name;
 767         int namelen = dentry->d_name.len;
 768
 769         de = (struct ext3_dir_entry_2 *) bh->b_data;
 770         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 771         while ((char *) de < dlimit) {
 772                 /* this code is executed quadratically often */
 773                 /* do minimal checking `by hand' */
 774
 775                 if ((char *) de + namelen <= dlimit &&
 776                     ext3_match (namelen, name, de)) {
 777                         /* found a match - just to be sure, do a full check */
 778                         if (!ext3_check_dir_entry("ext3_find_entry",
 779                                                   dir, de, bh, offset))
 780                                 return -1;
 781                         *res_dir = de;
 782                         return 1;
 783                 }
 784                 /* prevent looping on a bad block */
 785                 de_len = le16_to_cpu(de->rec_len);
 786                 if (de_len <= 0)
 787                         return -1;
 788                 offset += de_len;
 789                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 790         }
 791         return 0;
 792 }
 793
 794
 795 /*
 796  *      ext3_find_entry()
 797  *
 798  * finds an entry in the specified directory with the wanted name. It
 799  * returns the cache buffer in which the entry was found, and the entry
 800  * itself (as a parameter - res_dir). It does NOT read the inode of the
 801  * entry - you'll have to do that yourself if you want to.
 802  *
 803  * The returned buffer_head has ->b_count elevated.  The caller is expected
 804  * to brelse() it when appropriate.
 805  */
 806 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 807                                         struct ext3_dir_entry_2 ** res_dir)
 808 {
 809         struct super_block * sb;
 810         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 811         struct buffer_head * bh, *ret = NULL;
 812         unsigned long start, block, b;
 813         int ra_max = 0;         /* Number of bh's in the readahead
 814                                    buffer, bh_use[] */
 815         int ra_ptr = 0;         /* Current index into readahead
 816                                    buffer */
 817         int num = 0;
 818         int nblocks, i, err;
 819         struct inode *dir = dentry->d_parent->d_inode;
 820         int namelen;
 821         const u8 *name;
 822         unsigned blocksize;
 823
 824         *res_dir = NULL;
 825         sb = dir->i_sb;
 826         blocksize = sb->s_blocksize;
 827         namelen = dentry->d_name.len;
 828         name = dentry->d_name.name;
 829         if (namelen > EXT3_NAME_LEN)
 830                 return NULL;
 831 #ifdef CONFIG_EXT3_INDEX
 832         if (is_dx(dir)) {
 833                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
 834                 /*
 835                  * On success, or if the error was file not found,
 836                  * return.  Otherwise, fall back to doing a search the
 837                  * old fashioned way.
 838                  */
 839                 if (bh || (err != ERR_BAD_DX_DIR))
 840                         return bh;
 841                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 842         }
 843 #endif
 844         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 845         start = EXT3_I(dir)->i_dir_start_lookup;
 846         if (start >= nblocks)
 847                 start = 0;
 848         block = start;
 849 restart:
 850         do {
 851                 /*
 852                  * We deal with the read-ahead logic here.
 853                  */
 854                 if (ra_ptr >= ra_max) {
 855                         /* Refill the readahead buffer */
 856                         ra_ptr = 0;
 857                         b = block;
 858                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 859                                 /*
 860                                  * Terminate if we reach the end of the
 861                                  * directory and must wrap, or if our
 862                                  * search has finished at this block.
 863                                  */
 864                                 if (b >= nblocks || (num && block == start)) {
 865                                         bh_use[ra_max] = NULL;
 866                                         break;
 867                                 }
 868                                 num++;
 869                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 870                                 bh_use[ra_max] = bh;
 871                                 if (bh)
 872                                         ll_rw_block(READ, 1, &bh);
 873                         }
 874                 }
 875                 if ((bh = bh_use[ra_ptr++]) == NULL)
 876                         goto next;
 877                 wait_on_buffer(bh);
 878                 if (!buffer_uptodate(bh)) {
 879                         /* read error, skip block & hope for the best */
 880                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
 881                                    "offset %lu", dir->i_ino, block);
 882                         brelse(bh);
 883                         goto next;
 884                 }
 885                 i = search_dirblock(bh, dir, dentry,
 886                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 887                 if (i == 1) {
 888                         EXT3_I(dir)->i_dir_start_lookup = block;
 889                         ret = bh;
 890                         goto cleanup_and_exit;
 891                 } else {
 892                         brelse(bh);
 893                         if (i < 0)
 894                                 goto cleanup_and_exit;
 895                 }
 896         next:
 897                 if (++block >= nblocks)
 898                         block = 0;
 899         } while (block != start);
 900
 901         /*
 902          * If the directory has grown while we were searching, then
 903          * search the last part of the directory before giving up.
 904          */
 905         block = nblocks;
 906         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 907         if (block < nblocks) {
 908                 start = 0;
 909                 goto restart;
 910         }
 911
 912 cleanup_and_exit:
 913         /* Clean up the read-ahead blocks */
 914         for (; ra_ptr < ra_max; ra_ptr++)
 915                 brelse (bh_use[ra_ptr]);
 916         return ret;
 917 }
 918
 919 #ifdef CONFIG_EXT3_INDEX
 920 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 921                        struct ext3_dir_entry_2 **res_dir, int *err)
 922 {
 923         struct super_block * sb;
 924         struct dx_hash_info     hinfo;
 925         u32 hash;
 926         struct dx_frame frames[2], *frame;
 927         struct ext3_dir_entry_2 *de, *top;
 928         struct buffer_head *bh;
 929         unsigned long block;
 930         int retval;
 931         int namelen = dentry->d_name.len;
 932         const u8 *name = dentry->d_name.name;
 933         struct inode *dir = dentry->d_parent->d_inode;
 934
 935         sb = dir->i_sb;
 936         if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
 937                 return NULL;
 938         hash = hinfo.hash;
 939         do {
 940                 block = dx_get_block(frame->at);
 941                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
 942                         goto errout;
 943                 de = (struct ext3_dir_entry_2 *) bh->b_data;
 944                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
 945                                        EXT3_DIR_REC_LEN(0));
 946                 for (; de < top; de = ext3_next_entry(de))
 947                 if (ext3_match (namelen, name, de)) {
 948                         if (!ext3_check_dir_entry("ext3_find_entry",
 949                                                   dir, de, bh,
 950                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
 951                                           +((char *)de - bh->b_data))) {
 952                                 brelse (bh);
 953                                 goto errout;
 954                         }
 955                         *res_dir = de;
 956                         dx_release (frames);
 957                         return bh;
 958                 }
 959                 brelse (bh);
 960                 /* Check to see if we should continue to search */
 961                 retval = ext3_htree_next_block(dir, hash, frame,
 962                                                frames, NULL);
 963                 if (retval < 0) {
 964                         ext3_warning(sb, __FUNCTION__,
 965                              "error reading index page in directory #%lu",
 966                              dir->i_ino);
 967                         *err = retval;
 968                         goto errout;
 969                 }
 970         } while (retval == 1);
 971
 972         *err = -ENOENT;
 973 errout:
 974         dxtrace(printk("%s not found\n", name));
 975         dx_release (frames);
 976         return NULL;
 977 }
 978 #endif
 979
 980 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
 981 {
 982         struct inode * inode;
 983         struct ext3_dir_entry_2 * de;
 984         struct buffer_head * bh;
 985
 986         if (dentry->d_name.len > EXT3_NAME_LEN)
 987                 return ERR_PTR(-ENAMETOOLONG);
 988
 989         bh = ext3_find_entry(dentry, &de);
 990         inode = NULL;
 991         if (bh) {
 992                 unsigned long ino = le32_to_cpu(de->inode);
 993                 brelse (bh);
 994                 inode = iget(dir->i_sb, ino);
 995
 996                 if (!inode)
 997                         return ERR_PTR(-EACCES);
 998                 vx_propagate_xid(nd, inode);
 999         }
1000         if (inode)
1001                 return d_splice_alias(inode, dentry);
1002         d_add(dentry, inode);
1003         return NULL;
1004 }
1005
1006
1007 struct dentry *ext3_get_parent(struct dentry *child)
1008 {
1009         unsigned long ino;
1010         struct dentry *parent;
1011         struct inode *inode;
1012         struct dentry dotdot;
1013         struct ext3_dir_entry_2 * de;
1014         struct buffer_head *bh;
1015
1016         dotdot.d_name.name = "..";
1017         dotdot.d_name.len = 2;
1018         dotdot.d_parent = child; /* confusing, isn't it! */
1019
1020         bh = ext3_find_entry(&dotdot, &de);
1021         inode = NULL;
1022         if (!bh)
1023                 return ERR_PTR(-ENOENT);
1024         ino = le32_to_cpu(de->inode);
1025         brelse(bh);
1026         inode = iget(child->d_inode->i_sb, ino);
1027
1028         if (!inode)
1029                 return ERR_PTR(-EACCES);
1030
1031         parent = d_alloc_anon(inode);
1032         if (!parent) {
1033                 iput(inode);
1034                 parent = ERR_PTR(-ENOMEM);
1035         }
1036         return parent;
1037 }
1038
1039 #define S_SHIFT 12
1040 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1041         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1042         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1043         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1044         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1045         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1046         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1047         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1048 };
1049
1050 static inline void ext3_set_de_type(struct super_block *sb,
1051                                 struct ext3_dir_entry_2 *de,
1052                                 umode_t mode) {
1053         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1054                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1055 }
1056
1057 #ifdef CONFIG_EXT3_INDEX
1058 static struct ext3_dir_entry_2 *
1059 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1060 {
1061         unsigned rec_len = 0;
1062
1063         while (count--) {
1064                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1065                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1066                 memcpy (to, de, rec_len);
1067                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1068                                 cpu_to_le16(rec_len);
1069                 de->inode = 0;
1070                 map++;
1071                 to += rec_len;
1072         }
1073         return (struct ext3_dir_entry_2 *) (to - rec_len);
1074 }
1075
1076 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1077 {
1078         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1079         unsigned rec_len = 0;
1080
1081         prev = to = de;
1082         while ((char*)de < base + size) {
1083                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1084                                                     le16_to_cpu(de->rec_len));
1085                 if (de->inode && de->name_len) {
1086                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1087                         if (de > to)
1088                                 memmove(to, de, rec_len);
1089                         to->rec_len = cpu_to_le16(rec_len);
1090                         prev = to;
1091                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1092                 }
1093                 de = next;
1094         }
1095         return prev;
1096 }
1097
1098 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1099                         struct buffer_head **bh,struct dx_frame *frame,
1100                         struct dx_hash_info *hinfo, int *error)
1101 {
1102         unsigned blocksize = dir->i_sb->s_blocksize;
1103         unsigned count, continued;
1104         struct buffer_head *bh2;
1105         u32 newblock;
1106         u32 hash2;
1107         struct dx_map_entry *map;
1108         char *data1 = (*bh)->b_data, *data2;
1109         unsigned split;
1110         struct ext3_dir_entry_2 *de = NULL, *de2;
1111         int     err;
1112
1113         bh2 = ext3_append (handle, dir, &newblock, error);
1114         if (!(bh2)) {
1115                 brelse(*bh);
1116                 *bh = NULL;
1117                 goto errout;
1118         }
1119
1120         BUFFER_TRACE(*bh, "get_write_access");
1121         err = ext3_journal_get_write_access(handle, *bh);
1122         if (err) {
1123         journal_error:
1124                 brelse(*bh);
1125                 brelse(bh2);
1126                 *bh = NULL;
1127                 ext3_std_error(dir->i_sb, err);
1128                 goto errout;
1129         }
1130         BUFFER_TRACE(frame->bh, "get_write_access");
1131         err = ext3_journal_get_write_access(handle, frame->bh);
1132         if (err)
1133                 goto journal_error;
1134
1135         data2 = bh2->b_data;
1136
1137         /* create map in the end of data2 block */
1138         map = (struct dx_map_entry *) (data2 + blocksize);
1139         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1140                              blocksize, hinfo, map);
1141         map -= count;
1142         split = count/2; // need to adjust to actual middle
1143         dx_sort_map (map, count);
1144         hash2 = map[split].hash;
1145         continued = hash2 == map[split - 1].hash;
1146         dxtrace(printk("Split block %i at %x, %i/%i\n",
1147                 dx_get_block(frame->at), hash2, split, count-split));
1148
1149         /* Fancy dance to stay within two buffers */
1150         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1151         de = dx_pack_dirents(data1,blocksize);
1152         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1153         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1154         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1155         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1156
1157         /* Which block gets the new entry? */
1158         if (hinfo->hash >= hash2)
1159         {
1160                 swap(*bh, bh2);
1161                 de = de2;
1162         }
1163         dx_insert_block (frame, hash2 + continued, newblock);
1164         err = ext3_journal_dirty_metadata (handle, bh2);
1165         if (err)
1166                 goto journal_error;
1167         err = ext3_journal_dirty_metadata (handle, frame->bh);
1168         if (err)
1169                 goto journal_error;
1170         brelse (bh2);
1171         dxtrace(dx_show_index ("frame", frame->entries));
1172 errout:
1173         return de;
1174 }
1175 #endif
1176
1177
1178 /*
1179  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1180  * it points to a directory entry which is guaranteed to be large
1181  * enough for new directory entry.  If de is NULL, then
1182  * add_dirent_to_buf will attempt search the directory block for
1183  * space.  It will return -ENOSPC if no space is available, and -EIO
1184  * and -EEXIST if directory entry already exists.
1185  *
1186  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1187  * all other cases bh is released.
1188  */
1189 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1190                              struct inode *inode, struct ext3_dir_entry_2 *de,
1191                              struct buffer_head * bh)
1192 {
1193         struct inode    *dir = dentry->d_parent->d_inode;
1194         const char      *name = dentry->d_name.name;
1195         int             namelen = dentry->d_name.len;
1196         unsigned long   offset = 0;
1197         unsigned short  reclen;
1198         int             nlen, rlen, err;
1199         char            *top;
1200
1201         reclen = EXT3_DIR_REC_LEN(namelen);
1202         if (!de) {
1203                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1204                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1205                 while ((char *) de <= top) {
1206                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1207                                                   bh, offset)) {
1208                                 brelse (bh);
1209                                 return -EIO;
1210                         }
1211                         if (ext3_match (namelen, name, de)) {
1212                                 brelse (bh);
1213                                 return -EEXIST;
1214                         }
1215                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1216                         rlen = le16_to_cpu(de->rec_len);
1217                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1218                                 break;
1219                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1220                         offset += rlen;
1221                 }
1222                 if ((char *) de > top)
1223                         return -ENOSPC;
1224         }
1225         BUFFER_TRACE(bh, "get_write_access");
1226         err = ext3_journal_get_write_access(handle, bh);
1227         if (err) {
1228                 ext3_std_error(dir->i_sb, err);
1229                 brelse(bh);
1230                 return err;
1231         }
1232
1233         /* By now the buffer is marked for journaling */
1234         nlen = EXT3_DIR_REC_LEN(de->name_len);
1235         rlen = le16_to_cpu(de->rec_len);
1236         if (de->inode) {
1237                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1238                 de1->rec_len = cpu_to_le16(rlen - nlen);
1239                 de->rec_len = cpu_to_le16(nlen);
1240                 de = de1;
1241         }
1242         de->file_type = EXT3_FT_UNKNOWN;
1243         if (inode) {
1244                 de->inode = cpu_to_le32(inode->i_ino);
1245                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1246         } else
1247                 de->inode = 0;
1248         de->name_len = namelen;
1249         memcpy (de->name, name, namelen);
1250         /*
1251          * XXX shouldn't update any times until successful
1252          * completion of syscall, but too many callers depend
1253          * on this.
1254          *
1255          * XXX similarly, too many callers depend on
1256          * ext3_new_inode() setting the times, but error
1257          * recovery deletes the inode, so the worst that can
1258          * happen is that the times are slightly out of date
1259          * and/or different from the directory change time.
1260          */
1261         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1262         ext3_update_dx_flag(dir);
1263         dir->i_version++;
1264         ext3_mark_inode_dirty(handle, dir);
1265         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1266         err = ext3_journal_dirty_metadata(handle, bh);
1267         if (err)
1268                 ext3_std_error(dir->i_sb, err);
1269         brelse(bh);
1270         return 0;
1271 }
1272
1273 #ifdef CONFIG_EXT3_INDEX
1274 /*
1275  * This converts a one block unindexed directory to a 3 block indexed
1276  * directory, and adds the dentry to the indexed directory.
1277  */
1278 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1279                             struct inode *inode, struct buffer_head *bh)
1280 {
1281         struct inode    *dir = dentry->d_parent->d_inode;
1282         const char      *name = dentry->d_name.name;
1283         int             namelen = dentry->d_name.len;
1284         struct buffer_head *bh2;
1285         struct dx_root  *root;
1286         struct dx_frame frames[2], *frame;
1287         struct dx_entry *entries;
1288         struct ext3_dir_entry_2 *de, *de2;
1289         char            *data1, *top;
1290         unsigned        len;
1291         int             retval;
1292         unsigned        blocksize;
1293         struct dx_hash_info hinfo;
1294         u32             block;
1295         struct fake_dirent *fde;
1296
1297         blocksize =  dir->i_sb->s_blocksize;
1298         dxtrace(printk("Creating index\n"));
1299         retval = ext3_journal_get_write_access(handle, bh);
1300         if (retval) {
1301                 ext3_std_error(dir->i_sb, retval);
1302                 brelse(bh);
1303                 return retval;
1304         }
1305         root = (struct dx_root *) bh->b_data;
1306
1307         bh2 = ext3_append (handle, dir, &block, &retval);
1308         if (!(bh2)) {
1309                 brelse(bh);
1310                 return retval;
1311         }
1312         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1313         data1 = bh2->b_data;
1314
1315         /* The 0th block becomes the root, move the dirents out */
1316         fde = &root->dotdot;
1317         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1318         len = ((char *) root) + blocksize - (char *) de;
1319         memcpy (data1, de, len);
1320         de = (struct ext3_dir_entry_2 *) data1;
1321         top = data1 + len;
1322         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1323                 de = de2;
1324         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1325         /* Initialize the root; the dot dirents already exist */
1326         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1327         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1328         memset (&root->info, 0, sizeof(root->info));
1329         root->info.info_length = sizeof(root->info);
1330         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1331         entries = root->entries;
1332         dx_set_block (entries, 1);
1333         dx_set_count (entries, 1);
1334         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1335
1336         /* Initialize as for dx_probe */
1337         hinfo.hash_version = root->info.hash_version;
1338         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1339         ext3fs_dirhash(name, namelen, &hinfo);
1340         frame = frames;
1341         frame->entries = entries;
1342         frame->at = entries;
1343         frame->bh = bh;
1344         bh = bh2;
1345         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1346         dx_release (frames);
1347         if (!(de))
1348                 return retval;
1349
1350         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1351 }
1352 #endif
1353
1354 /*
1355  *      ext3_add_entry()
1356  *
1357  * adds a file entry to the specified directory, using the same
1358  * semantics as ext3_find_entry(). It returns NULL if it failed.
1359  *
1360  * NOTE!! The inode part of 'de' is left at 0 - which means you
1361  * may not sleep between calling this and putting something into
1362  * the entry, as someone else might have used it while you slept.
1363  */
1364 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1365         struct inode *inode)
1366 {
1367         struct inode *dir = dentry->d_parent->d_inode;
1368         unsigned long offset;
1369         struct buffer_head * bh;
1370         struct ext3_dir_entry_2 *de;
1371         struct super_block * sb;
1372         int     retval;
1373 #ifdef CONFIG_EXT3_INDEX
1374         int     dx_fallback=0;
1375 #endif
1376         unsigned blocksize;
1377         unsigned nlen, rlen;
1378         u32 block, blocks;
1379
1380         sb = dir->i_sb;
1381         blocksize = sb->s_blocksize;
1382         if (!dentry->d_name.len)
1383                 return -EINVAL;
1384 #ifdef CONFIG_EXT3_INDEX
1385         if (is_dx(dir)) {
1386                 retval = ext3_dx_add_entry(handle, dentry, inode);
1387                 if (!retval || (retval != ERR_BAD_DX_DIR))
1388                         return retval;
1389                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1390                 dx_fallback++;
1391                 ext3_mark_inode_dirty(handle, dir);
1392         }
1393 #endif
1394         blocks = dir->i_size >> sb->s_blocksize_bits;
1395         for (block = 0, offset = 0; block < blocks; block++) {
1396                 bh = ext3_bread(handle, dir, block, 0, &retval);
1397                 if(!bh)
1398                         return retval;
1399                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1400                 if (retval != -ENOSPC)
1401                         return retval;
1402
1403 #ifdef CONFIG_EXT3_INDEX
1404                 if (blocks == 1 && !dx_fallback &&
1405                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1406                         return make_indexed_dir(handle, dentry, inode, bh);
1407 #endif
1408                 brelse(bh);
1409         }
1410         bh = ext3_append(handle, dir, &block, &retval);
1411         if (!bh)
1412                 return retval;
1413         de = (struct ext3_dir_entry_2 *) bh->b_data;
1414         de->inode = 0;
1415         de->rec_len = cpu_to_le16(rlen = blocksize);
1416         nlen = 0;
1417         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1418 }
1419
1420 #ifdef CONFIG_EXT3_INDEX
1421 /*
1422  * Returns 0 for success, or a negative error value
1423  */
1424 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1425                              struct inode *inode)
1426 {
1427         struct dx_frame frames[2], *frame;
1428         struct dx_entry *entries, *at;
1429         struct dx_hash_info hinfo;
1430         struct buffer_head * bh;
1431         struct inode *dir = dentry->d_parent->d_inode;
1432         struct super_block * sb = dir->i_sb;
1433         struct ext3_dir_entry_2 *de;
1434         int err;
1435
1436         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1437         if (!frame)
1438                 return err;
1439         entries = frame->entries;
1440         at = frame->at;
1441
1442         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1443                 goto cleanup;
1444
1445         BUFFER_TRACE(bh, "get_write_access");
1446         err = ext3_journal_get_write_access(handle, bh);
1447         if (err)
1448                 goto journal_error;
1449
1450         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1451         if (err != -ENOSPC) {
1452                 bh = NULL;
1453                 goto cleanup;
1454         }
1455
1456         /* Block full, should compress but for now just split */
1457         dxtrace(printk("using %u of %u node entries\n",
1458                        dx_get_count(entries), dx_get_limit(entries)));
1459         /* Need to split index? */
1460         if (dx_get_count(entries) == dx_get_limit(entries)) {
1461                 u32 newblock;
1462                 unsigned icount = dx_get_count(entries);
1463                 int levels = frame - frames;
1464                 struct dx_entry *entries2;
1465                 struct dx_node *node2;
1466                 struct buffer_head *bh2;
1467
1468                 if (levels && (dx_get_count(frames->entries) ==
1469                                dx_get_limit(frames->entries))) {
1470                         ext3_warning(sb, __FUNCTION__,
1471                                      "Directory index full!\n");
1472                         err = -ENOSPC;
1473                         goto cleanup;
1474                 }
1475                 bh2 = ext3_append (handle, dir, &newblock, &err);
1476                 if (!(bh2))
1477                         goto cleanup;
1478                 node2 = (struct dx_node *)(bh2->b_data);
1479                 entries2 = node2->entries;
1480                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1481                 node2->fake.inode = 0;
1482                 BUFFER_TRACE(frame->bh, "get_write_access");
1483                 err = ext3_journal_get_write_access(handle, frame->bh);
1484                 if (err)
1485                         goto journal_error;
1486                 if (levels) {
1487                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1488                         unsigned hash2 = dx_get_hash(entries + icount1);
1489                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1490
1491                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1492                         err = ext3_journal_get_write_access(handle,
1493                                                              frames[0].bh);
1494                         if (err)
1495                                 goto journal_error;
1496
1497                         memcpy ((char *) entries2, (char *) (entries + icount1),
1498                                 icount2 * sizeof(struct dx_entry));
1499                         dx_set_count (entries, icount1);
1500                         dx_set_count (entries2, icount2);
1501                         dx_set_limit (entries2, dx_node_limit(dir));
1502
1503                         /* Which index block gets the new entry? */
1504                         if (at - entries >= icount1) {
1505                                 frame->at = at = at - entries - icount1 + entries2;
1506                                 frame->entries = entries = entries2;
1507                                 swap(frame->bh, bh2);
1508                         }
1509                         dx_insert_block (frames + 0, hash2, newblock);
1510                         dxtrace(dx_show_index ("node", frames[1].entries));
1511                         dxtrace(dx_show_index ("node",
1512                                ((struct dx_node *) bh2->b_data)->entries));
1513                         err = ext3_journal_dirty_metadata(handle, bh2);
1514                         if (err)
1515                                 goto journal_error;
1516                         brelse (bh2);
1517                 } else {
1518                         dxtrace(printk("Creating second level index...\n"));
1519                         memcpy((char *) entries2, (char *) entries,
1520                                icount * sizeof(struct dx_entry));
1521                         dx_set_limit(entries2, dx_node_limit(dir));
1522
1523                         /* Set up root */
1524                         dx_set_count(entries, 1);
1525                         dx_set_block(entries + 0, newblock);
1526                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1527
1528                         /* Add new access path frame */
1529                         frame = frames + 1;
1530                         frame->at = at = at - entries + entries2;
1531                         frame->entries = entries = entries2;
1532                         frame->bh = bh2;
1533                         err = ext3_journal_get_write_access(handle,
1534                                                              frame->bh);
1535                         if (err)
1536                                 goto journal_error;
1537                 }
1538                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1539         }
1540         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1541         if (!de)
1542                 goto cleanup;
1543         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1544         bh = NULL;
1545         goto cleanup;
1546
1547 journal_error:
1548         ext3_std_error(dir->i_sb, err);
1549 cleanup:
1550         if (bh)
1551                 brelse(bh);
1552         dx_release(frames);
1553         return err;
1554 }
1555 #endif
1556
1557 /*
1558  * ext3_delete_entry deletes a directory entry by merging it with the
1559  * previous entry
1560  */
1561 static int ext3_delete_entry (handle_t *handle,
1562                               struct inode * dir,
1563                               struct ext3_dir_entry_2 * de_del,
1564                               struct buffer_head * bh)
1565 {
1566         struct ext3_dir_entry_2 * de, * pde;
1567         int i;
1568
1569         i = 0;
1570         pde = NULL;
1571         de = (struct ext3_dir_entry_2 *) bh->b_data;
1572         while (i < bh->b_size) {
1573                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1574                         return -EIO;
1575                 if (de == de_del)  {
1576                         BUFFER_TRACE(bh, "get_write_access");
1577                         ext3_journal_get_write_access(handle, bh);
1578                         if (pde)
1579                                 pde->rec_len =
1580                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1581                                                     le16_to_cpu(de->rec_len));
1582                         else
1583                                 de->inode = 0;
1584                         dir->i_version++;
1585                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1586                         ext3_journal_dirty_metadata(handle, bh);
1587                         return 0;
1588                 }
1589                 i += le16_to_cpu(de->rec_len);
1590                 pde = de;
1591                 de = (struct ext3_dir_entry_2 *)
1592                         ((char *) de + le16_to_cpu(de->rec_len));
1593         }
1594         return -ENOENT;
1595 }
1596
1597 /*
1598  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1599  * do not perform it in these functions.  We perform it at the call site,
1600  * if it is needed.
1601  */
1602 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1603 {
1604         inode->i_nlink++;
1605 }
1606
1607 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1608 {
1609         inode->i_nlink--;
1610 }
1611
1612 static int ext3_add_nondir(handle_t *handle,
1613                 struct dentry *dentry, struct inode *inode)
1614 {
1615         int err = ext3_add_entry(handle, dentry, inode);
1616         if (!err) {
1617                 ext3_mark_inode_dirty(handle, inode);
1618                 d_instantiate(dentry, inode);
1619                 return 0;
1620         }
1621         ext3_dec_count(handle, inode);
1622         iput(inode);
1623         return err;
1624 }
1625
1626 /*
1627  * By the time this is called, we already have created
1628  * the directory cache entry for the new file, but it
1629  * is so far negative - it has no inode.
1630  *
1631  * If the create succeeds, we fill in the inode information
1632  * with d_instantiate().
1633  */
1634 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1635                 struct nameidata *nd)
1636 {
1637         handle_t *handle;
1638         struct inode * inode;
1639         int err, retries = 0;
1640
1641 retry:
1642         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1643                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1644                                         2*EXT3_QUOTA_INIT_BLOCKS);
1645         if (IS_ERR(handle))
1646                 return PTR_ERR(handle);
1647
1648         if (IS_DIRSYNC(dir))
1649                 handle->h_sync = 1;
1650
1651         inode = ext3_new_inode (handle, dir, mode);
1652         err = PTR_ERR(inode);
1653         if (!IS_ERR(inode)) {
1654                 inode->i_op = &ext3_file_inode_operations;
1655                 inode->i_fop = &ext3_file_operations;
1656                 ext3_set_aops(inode);
1657                 err = ext3_add_nondir(handle, dentry, inode);
1658         }
1659         ext3_journal_stop(handle);
1660         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1661                 goto retry;
1662         return err;
1663 }
1664
1665 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1666                         int mode, dev_t rdev)
1667 {
1668         handle_t *handle;
1669         struct inode *inode;
1670         int err, retries = 0;
1671
1672         if (!new_valid_dev(rdev))
1673                 return -EINVAL;
1674
1675 retry:
1676         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1677                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1678                                         2*EXT3_QUOTA_INIT_BLOCKS);
1679         if (IS_ERR(handle))
1680                 return PTR_ERR(handle);
1681
1682         if (IS_DIRSYNC(dir))
1683                 handle->h_sync = 1;
1684
1685         inode = ext3_new_inode (handle, dir, mode);
1686         err = PTR_ERR(inode);
1687         if (!IS_ERR(inode)) {
1688                 init_special_inode(inode, inode->i_mode, rdev);
1689 #ifdef CONFIG_EXT3_FS_XATTR
1690                 inode->i_op = &ext3_special_inode_operations;
1691 #endif
1692                 err = ext3_add_nondir(handle, dentry, inode);
1693         }
1694         ext3_journal_stop(handle);
1695         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1696                 goto retry;
1697         return err;
1698 }
1699
1700 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1701 {
1702         handle_t *handle;
1703         struct inode * inode;
1704         struct buffer_head * dir_block;
1705         struct ext3_dir_entry_2 * de;
1706         int err, retries = 0;
1707
1708         if (dir->i_nlink >= EXT3_LINK_MAX)
1709                 return -EMLINK;
1710
1711 retry:
1712         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1713                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1714                                         2*EXT3_QUOTA_INIT_BLOCKS);
1715         if (IS_ERR(handle))
1716                 return PTR_ERR(handle);
1717
1718         if (IS_DIRSYNC(dir))
1719                 handle->h_sync = 1;
1720
1721         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1722         err = PTR_ERR(inode);
1723         if (IS_ERR(inode))
1724                 goto out_stop;
1725
1726         inode->i_op = &ext3_dir_inode_operations;
1727         inode->i_fop = &ext3_dir_operations;
1728         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1729         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1730         if (!dir_block) {
1731                 inode->i_nlink--; /* is this nlink == 0? */
1732                 ext3_mark_inode_dirty(handle, inode);
1733                 iput (inode);
1734                 goto out_stop;
1735         }
1736         BUFFER_TRACE(dir_block, "get_write_access");
1737         ext3_journal_get_write_access(handle, dir_block);
1738         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1739         de->inode = cpu_to_le32(inode->i_ino);
1740         de->name_len = 1;
1741         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1742         strcpy (de->name, ".");
1743         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1744         de = (struct ext3_dir_entry_2 *)
1745                         ((char *) de + le16_to_cpu(de->rec_len));
1746         de->inode = cpu_to_le32(dir->i_ino);
1747         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1748         de->name_len = 2;
1749         strcpy (de->name, "..");
1750         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1751         inode->i_nlink = 2;
1752         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1753         ext3_journal_dirty_metadata(handle, dir_block);
1754         brelse (dir_block);
1755         ext3_mark_inode_dirty(handle, inode);
1756         err = ext3_add_entry (handle, dentry, inode);
1757         if (err) {
1758                 inode->i_nlink = 0;
1759                 ext3_mark_inode_dirty(handle, inode);
1760                 iput (inode);
1761                 goto out_stop;
1762         }
1763         dir->i_nlink++;
1764         ext3_update_dx_flag(dir);
1765         ext3_mark_inode_dirty(handle, dir);
1766         d_instantiate(dentry, inode);
1767 out_stop:
1768         ext3_journal_stop(handle);
1769         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1770                 goto retry;
1771         return err;
1772 }
1773
1774 /*
1775  * routine to check that the specified directory is empty (for rmdir)
1776  */
1777 static int empty_dir (struct inode * inode)
1778 {
1779         unsigned long offset;
1780         struct buffer_head * bh;
1781         struct ext3_dir_entry_2 * de, * de1;
1782         struct super_block * sb;
1783         int err = 0;
1784
1785         sb = inode->i_sb;
1786         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1787             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1788                 if (err)
1789                         ext3_error(inode->i_sb, __FUNCTION__,
1790                                    "error %d reading directory #%lu offset 0",
1791                                    err, inode->i_ino);
1792                 else
1793                         ext3_warning(inode->i_sb, __FUNCTION__,
1794                                      "bad directory (dir #%lu) - no data block",
1795                                      inode->i_ino);
1796                 return 1;
1797         }
1798         de = (struct ext3_dir_entry_2 *) bh->b_data;
1799         de1 = (struct ext3_dir_entry_2 *)
1800                         ((char *) de + le16_to_cpu(de->rec_len));
1801         if (le32_to_cpu(de->inode) != inode->i_ino ||
1802                         !le32_to_cpu(de1->inode) ||
1803                         strcmp (".", de->name) ||
1804                         strcmp ("..", de1->name)) {
1805                 ext3_warning (inode->i_sb, "empty_dir",
1806                               "bad directory (dir #%lu) - no `.' or `..'",
1807                               inode->i_ino);
1808                 brelse (bh);
1809                 return 1;
1810         }
1811         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1812         de = (struct ext3_dir_entry_2 *)
1813                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1814         while (offset < inode->i_size ) {
1815                 if (!bh ||
1816                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1817                         err = 0;
1818                         brelse (bh);
1819                         bh = ext3_bread (NULL, inode,
1820                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1821                         if (!bh) {
1822                                 if (err)
1823                                         ext3_error(sb, __FUNCTION__,
1824                                                    "error %d reading directory"
1825                                                    " #%lu offset %lu",
1826                                                    err, inode->i_ino, offset);
1827                                 offset += sb->s_blocksize;
1828                                 continue;
1829                         }
1830                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1831                 }
1832                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1833                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1834                                                          sb->s_blocksize);
1835                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1836                         continue;
1837                 }
1838                 if (le32_to_cpu(de->inode)) {
1839                         brelse (bh);
1840                         return 0;
1841                 }
1842                 offset += le16_to_cpu(de->rec_len);
1843                 de = (struct ext3_dir_entry_2 *)
1844                                 ((char *) de + le16_to_cpu(de->rec_len));
1845         }
1846         brelse (bh);
1847         return 1;
1848 }
1849
1850 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1851  * such inodes, starting at the superblock, in case we crash before the
1852  * file is closed/deleted, or in case the inode truncate spans multiple
1853  * transactions and the last transaction is not recovered after a crash.
1854  *
1855  * At filesystem recovery time, we walk this list deleting unlinked
1856  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1857  */
1858 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1859 {
1860         struct super_block *sb = inode->i_sb;
1861         struct ext3_iloc iloc;
1862         int err = 0, rc;
1863
1864         lock_super(sb);
1865         if (!list_empty(&EXT3_I(inode)->i_orphan))
1866                 goto out_unlock;
1867
1868         /* Orphan handling is only valid for files with data blocks
1869          * being truncated, or files being unlinked. */
1870
1871         /* @@@ FIXME: Observation from aviro:
1872          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1873          * here (on lock_super()), so race with ext3_link() which might bump
1874          * ->i_nlink. For, say it, character device. Not a regular file,
1875          * not a directory, not a symlink and ->i_nlink > 0.
1876          */
1877         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1878                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1879
1880         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1881         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1882         if (err)
1883                 goto out_unlock;
1884
1885         err = ext3_reserve_inode_write(handle, inode, &iloc);
1886         if (err)
1887                 goto out_unlock;
1888
1889         /* Insert this inode at the head of the on-disk orphan list... */
1890         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1891         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1892         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1893         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1894         if (!err)
1895                 err = rc;
1896
1897         /* Only add to the head of the in-memory list if all the
1898          * previous operations succeeded.  If the orphan_add is going to
1899          * fail (possibly taking the journal offline), we can't risk
1900          * leaving the inode on the orphan list: stray orphan-list
1901          * entries can cause panics at unmount time.
1902          *
1903          * This is safe: on error we're going to ignore the orphan list
1904          * anyway on the next recovery. */
1905         if (!err)
1906                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1907
1908         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1909         jbd_debug(4, "orphan inode %ld will point to %d\n",
1910                         inode->i_ino, NEXT_ORPHAN(inode));
1911 out_unlock:
1912         unlock_super(sb);
1913         ext3_std_error(inode->i_sb, err);
1914         return err;
1915 }
1916
1917 /*
1918  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1919  * of such inodes stored on disk, because it is finally being cleaned up.
1920  */
1921 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1922 {
1923         struct list_head *prev;
1924         struct ext3_inode_info *ei = EXT3_I(inode);
1925         struct ext3_sb_info *sbi;
1926         unsigned long ino_next;
1927         struct ext3_iloc iloc;
1928         int err = 0;
1929
1930         lock_super(inode->i_sb);
1931         if (list_empty(&ei->i_orphan)) {
1932                 unlock_super(inode->i_sb);
1933                 return 0;
1934         }
1935
1936         ino_next = NEXT_ORPHAN(inode);
1937         prev = ei->i_orphan.prev;
1938         sbi = EXT3_SB(inode->i_sb);
1939
1940         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1941
1942         list_del_init(&ei->i_orphan);
1943
1944         /* If we're on an error path, we may not have a valid
1945          * transaction handle with which to update the orphan list on
1946          * disk, but we still need to remove the inode from the linked
1947          * list in memory. */
1948         if (!handle)
1949                 goto out;
1950
1951         err = ext3_reserve_inode_write(handle, inode, &iloc);
1952         if (err)
1953                 goto out_err;
1954
1955         if (prev == &sbi->s_orphan) {
1956                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1957                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1958                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1959                 if (err)
1960                         goto out_brelse;
1961                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1962                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1963         } else {
1964                 struct ext3_iloc iloc2;
1965                 struct inode *i_prev =
1966                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1967
1968                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1969                           i_prev->i_ino, ino_next);
1970                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1971                 if (err)
1972                         goto out_brelse;
1973                 NEXT_ORPHAN(i_prev) = ino_next;
1974                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1975         }
1976         if (err)
1977                 goto out_brelse;
1978         NEXT_ORPHAN(inode) = 0;
1979         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1980
1981 out_err:
1982         ext3_std_error(inode->i_sb, err);
1983 out:
1984         unlock_super(inode->i_sb);
1985         return err;
1986
1987 out_brelse:
1988         brelse(iloc.bh);
1989         goto out_err;
1990 }
1991
1992 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1993 {
1994         int retval;
1995         struct inode * inode;
1996         struct buffer_head * bh;
1997         struct ext3_dir_entry_2 * de;
1998         handle_t *handle;
1999
2000         /* Initialize quotas before so that eventual writes go in
2001          * separate transaction */
2002         DQUOT_INIT(dentry->d_inode);
2003         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2004         if (IS_ERR(handle))
2005                 return PTR_ERR(handle);
2006
2007         retval = -ENOENT;
2008         bh = ext3_find_entry (dentry, &de);
2009         if (!bh)
2010                 goto end_rmdir;
2011
2012         if (IS_DIRSYNC(dir))
2013                 handle->h_sync = 1;
2014
2015         inode = dentry->d_inode;
2016
2017         retval = -EIO;
2018         if (le32_to_cpu(de->inode) != inode->i_ino)
2019                 goto end_rmdir;
2020
2021         retval = -ENOTEMPTY;
2022         if (!empty_dir (inode))
2023                 goto end_rmdir;
2024
2025         retval = ext3_delete_entry(handle, dir, de, bh);
2026         if (retval)
2027                 goto end_rmdir;
2028         if (inode->i_nlink != 2)
2029                 ext3_warning (inode->i_sb, "ext3_rmdir",
2030                               "empty directory has nlink!=2 (%d)",
2031                               inode->i_nlink);
2032         inode->i_version++;
2033         inode->i_nlink = 0;
2034         /* There's no need to set i_disksize: the fact that i_nlink is
2035          * zero will ensure that the right thing happens during any
2036          * recovery. */
2037         inode->i_size = 0;
2038         ext3_orphan_add(handle, inode);
2039         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2040         ext3_mark_inode_dirty(handle, inode);
2041         dir->i_nlink--;
2042         ext3_update_dx_flag(dir);
2043         ext3_mark_inode_dirty(handle, dir);
2044
2045 end_rmdir:
2046         ext3_journal_stop(handle);
2047         brelse (bh);
2048         return retval;
2049 }
2050
2051 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2052 {
2053         int retval;
2054         struct inode * inode;
2055         struct buffer_head * bh;
2056         struct ext3_dir_entry_2 * de;
2057         handle_t *handle;
2058
2059         /* Initialize quotas before so that eventual writes go
2060          * in separate transaction */
2061         DQUOT_INIT(dentry->d_inode);
2062         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2063         if (IS_ERR(handle))
2064                 return PTR_ERR(handle);
2065
2066         if (IS_DIRSYNC(dir))
2067                 handle->h_sync = 1;
2068
2069         retval = -ENOENT;
2070         bh = ext3_find_entry (dentry, &de);
2071         if (!bh)
2072                 goto end_unlink;
2073
2074         inode = dentry->d_inode;
2075
2076         retval = -EIO;
2077         if (le32_to_cpu(de->inode) != inode->i_ino)
2078                 goto end_unlink;
2079
2080         if (!inode->i_nlink) {
2081                 ext3_warning (inode->i_sb, "ext3_unlink",
2082                               "Deleting nonexistent file (%lu), %d",
2083                               inode->i_ino, inode->i_nlink);
2084                 inode->i_nlink = 1;
2085         }
2086         retval = ext3_delete_entry(handle, dir, de, bh);
2087         if (retval)
2088                 goto end_unlink;
2089         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2090         ext3_update_dx_flag(dir);
2091         ext3_mark_inode_dirty(handle, dir);
2092         inode->i_nlink--;
2093         if (!inode->i_nlink)
2094                 ext3_orphan_add(handle, inode);
2095         inode->i_ctime = dir->i_ctime;
2096         ext3_mark_inode_dirty(handle, inode);
2097         retval = 0;
2098
2099 end_unlink:
2100         ext3_journal_stop(handle);
2101         brelse (bh);
2102         return retval;
2103 }
2104
2105 static int ext3_symlink (struct inode * dir,
2106                 struct dentry *dentry, const char * symname)
2107 {
2108         handle_t *handle;
2109         struct inode * inode;
2110         int l, err, retries = 0;
2111
2112         l = strlen(symname)+1;
2113         if (l > dir->i_sb->s_blocksize)
2114                 return -ENAMETOOLONG;
2115
2116 retry:
2117         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2118                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2119                                         2*EXT3_QUOTA_INIT_BLOCKS);
2120         if (IS_ERR(handle))
2121                 return PTR_ERR(handle);
2122
2123         if (IS_DIRSYNC(dir))
2124                 handle->h_sync = 1;
2125
2126         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2127         err = PTR_ERR(inode);
2128         if (IS_ERR(inode))
2129                 goto out_stop;
2130
2131         if (l > sizeof (EXT3_I(inode)->i_data)) {
2132                 inode->i_op = &ext3_symlink_inode_operations;
2133                 ext3_set_aops(inode);
2134                 /*
2135                  * page_symlink() calls into ext3_prepare/commit_write.
2136                  * We have a transaction open.  All is sweetness.  It also sets
2137                  * i_size in generic_commit_write().
2138                  */
2139                 err = page_symlink(inode, symname, l);
2140                 if (err) {
2141                         ext3_dec_count(handle, inode);
2142                         ext3_mark_inode_dirty(handle, inode);
2143                         iput (inode);
2144                         goto out_stop;
2145                 }
2146         } else {
2147                 inode->i_op = &ext3_fast_symlink_inode_operations;
2148                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2149                 inode->i_size = l-1;
2150         }
2151         EXT3_I(inode)->i_disksize = inode->i_size;
2152         err = ext3_add_nondir(handle, dentry, inode);
2153 out_stop:
2154         ext3_journal_stop(handle);
2155         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2156                 goto retry;
2157         return err;
2158 }
2159
2160 static int ext3_link (struct dentry * old_dentry,
2161                 struct inode * dir, struct dentry *dentry)
2162 {
2163         handle_t *handle;
2164         struct inode *inode = old_dentry->d_inode;
2165         int err, retries = 0;
2166
2167         if (inode->i_nlink >= EXT3_LINK_MAX)
2168                 return -EMLINK;
2169
2170 retry:
2171         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2172                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2173         if (IS_ERR(handle))
2174                 return PTR_ERR(handle);
2175
2176         if (IS_DIRSYNC(dir))
2177                 handle->h_sync = 1;
2178
2179         inode->i_ctime = CURRENT_TIME_SEC;
2180         ext3_inc_count(handle, inode);
2181         atomic_inc(&inode->i_count);
2182
2183         err = ext3_add_nondir(handle, dentry, inode);
2184         ext3_journal_stop(handle);
2185         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2186                 goto retry;
2187         return err;
2188 }
2189
2190 #define PARENT_INO(buffer) \
2191         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2192         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2193
2194 /*
2195  * Anybody can rename anything with this: the permission checks are left to the
2196  * higher-level routines.
2197  */
2198 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2199                            struct inode * new_dir,struct dentry *new_dentry)
2200 {
2201         handle_t *handle;
2202         struct inode * old_inode, * new_inode;
2203         struct buffer_head * old_bh, * new_bh, * dir_bh;
2204         struct ext3_dir_entry_2 * old_de, * new_de;
2205         int retval;
2206
2207         old_bh = new_bh = dir_bh = NULL;
2208
2209         /* Initialize quotas before so that eventual writes go
2210          * in separate transaction */
2211         if (new_dentry->d_inode)
2212                 DQUOT_INIT(new_dentry->d_inode);
2213         handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2214                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2215         if (IS_ERR(handle))
2216                 return PTR_ERR(handle);
2217
2218         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2219                 handle->h_sync = 1;
2220
2221         old_bh = ext3_find_entry (old_dentry, &old_de);
2222         /*
2223          *  Check for inode number is _not_ due to possible IO errors.
2224          *  We might rmdir the source, keep it as pwd of some process
2225          *  and merrily kill the link to whatever was created under the
2226          *  same name. Goodbye sticky bit ;-<
2227          */
2228         old_inode = old_dentry->d_inode;
2229         retval = -ENOENT;
2230         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2231                 goto end_rename;
2232
2233         new_inode = new_dentry->d_inode;
2234         new_bh = ext3_find_entry (new_dentry, &new_de);
2235         if (new_bh) {
2236                 if (!new_inode) {
2237                         brelse (new_bh);
2238                         new_bh = NULL;
2239                 }
2240         }
2241         if (S_ISDIR(old_inode->i_mode)) {
2242                 if (new_inode) {
2243                         retval = -ENOTEMPTY;
2244                         if (!empty_dir (new_inode))
2245                                 goto end_rename;
2246                 }
2247                 retval = -EIO;
2248                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2249                 if (!dir_bh)
2250                         goto end_rename;
2251                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2252                         goto end_rename;
2253                 retval = -EMLINK;
2254                 if (!new_inode && new_dir!=old_dir &&
2255                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2256                         goto end_rename;
2257         }
2258         if (!new_bh) {
2259                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2260                 if (retval)
2261                         goto end_rename;
2262         } else {
2263                 BUFFER_TRACE(new_bh, "get write access");
2264                 ext3_journal_get_write_access(handle, new_bh);
2265                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2266                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2267                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2268                         new_de->file_type = old_de->file_type;
2269                 new_dir->i_version++;
2270                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2271                 ext3_journal_dirty_metadata(handle, new_bh);
2272                 brelse(new_bh);
2273                 new_bh = NULL;
2274         }
2275
2276         /*
2277          * Like most other Unix systems, set the ctime for inodes on a
2278          * rename.
2279          */
2280         old_inode->i_ctime = CURRENT_TIME_SEC;
2281         ext3_mark_inode_dirty(handle, old_inode);
2282
2283         /*
2284          * ok, that's it
2285          */
2286         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2287             old_de->name_len != old_dentry->d_name.len ||
2288             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2289             (retval = ext3_delete_entry(handle, old_dir,
2290                                         old_de, old_bh)) == -ENOENT) {
2291                 /* old_de could have moved from under us during htree split, so
2292                  * make sure that we are deleting the right entry.  We might
2293                  * also be pointing to a stale entry in the unused part of
2294                  * old_bh so just checking inum and the name isn't enough. */
2295                 struct buffer_head *old_bh2;
2296                 struct ext3_dir_entry_2 *old_de2;
2297
2298                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2299                 if (old_bh2) {
2300                         retval = ext3_delete_entry(handle, old_dir,
2301                                                    old_de2, old_bh2);
2302                         brelse(old_bh2);
2303                 }
2304         }
2305         if (retval) {
2306                 ext3_warning(old_dir->i_sb, "ext3_rename",
2307                                 "Deleting old file (%lu), %d, error=%d",
2308                                 old_dir->i_ino, old_dir->i_nlink, retval);
2309         }
2310
2311         if (new_inode) {
2312                 new_inode->i_nlink--;
2313                 new_inode->i_ctime = CURRENT_TIME_SEC;
2314         }
2315         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2316         ext3_update_dx_flag(old_dir);
2317         if (dir_bh) {
2318                 BUFFER_TRACE(dir_bh, "get_write_access");
2319                 ext3_journal_get_write_access(handle, dir_bh);
2320                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2321                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2322                 ext3_journal_dirty_metadata(handle, dir_bh);
2323                 old_dir->i_nlink--;
2324                 if (new_inode) {
2325                         new_inode->i_nlink--;
2326                 } else {
2327                         new_dir->i_nlink++;
2328                         ext3_update_dx_flag(new_dir);
2329                         ext3_mark_inode_dirty(handle, new_dir);
2330                 }
2331         }
2332         ext3_mark_inode_dirty(handle, old_dir);
2333         if (new_inode) {
2334                 ext3_mark_inode_dirty(handle, new_inode);
2335                 if (!new_inode->i_nlink)
2336                         ext3_orphan_add(handle, new_inode);
2337         }
2338         retval = 0;
2339
2340 end_rename:
2341         brelse (dir_bh);
2342         brelse (old_bh);
2343         brelse (new_bh);
2344         ext3_journal_stop(handle);
2345         return retval;
2346 }
2347
2348 /*
2349  * directories can handle most operations...
2350  */
2351 struct inode_operations ext3_dir_inode_operations = {
2352         .create         = ext3_create,
2353         .lookup         = ext3_lookup,
2354         .link           = ext3_link,
2355         .unlink         = ext3_unlink,
2356         .symlink        = ext3_symlink,
2357         .mkdir          = ext3_mkdir,
2358         .rmdir          = ext3_rmdir,
2359         .mknod          = ext3_mknod,
2360         .rename         = ext3_rename,
2361         .setattr        = ext3_setattr,
2362 #ifdef CONFIG_EXT3_FS_XATTR
2363         .setxattr       = generic_setxattr,
2364         .getxattr       = generic_getxattr,
2365         .listxattr      = ext3_listxattr,
2366         .removexattr    = generic_removexattr,
2367 #endif
2368         .permission     = ext3_permission,
2369 };
2370
2371 struct inode_operations ext3_special_inode_operations = {
2372         .setattr        = ext3_setattr,
2373 #ifdef CONFIG_EXT3_FS_XATTR
2374         .setxattr       = generic_setxattr,
2375         .getxattr       = generic_getxattr,
2376         .listxattr      = ext3_listxattr,
2377         .removexattr    = generic_removexattr,
2378 #endif
2379         .permission     = ext3_permission,
2380 };