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 "xattr.h"
  40 #include "acl.h"
  41
  42 /*
  43  * define how far ahead to read directories while searching them.
  44  */
  45 #define NAMEI_RA_CHUNKS  2
  46 #define NAMEI_RA_BLOCKS  4
  47 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  48 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  49
  50 static struct buffer_head *ext3_append(handle_t *handle,
  51                                         struct inode *inode,
  52                                         u32 *block, int *err)
  53 {
  54         struct buffer_head *bh;
  55
  56         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  57
  58         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
  59                 inode->i_size += inode->i_sb->s_blocksize;
  60                 EXT3_I(inode)->i_disksize = inode->i_size;
  61                 ext3_journal_get_write_access(handle,bh);
  62         }
  63         return bh;
  64 }
  65
  66 #ifndef assert
  67 #define assert(test) J_ASSERT(test)
  68 #endif
  69
  70 #ifndef swap
  71 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  72 #endif
  73
  74 typedef struct { u32 v; } le_u32;
  75 typedef struct { u16 v; } le_u16;
  76
  77 #ifdef DX_DEBUG
  78 #define dxtrace(command) command
  79 #else
  80 #define dxtrace(command)
  81 #endif
  82
  83 struct fake_dirent
  84 {
  85         /*le*/u32 inode;
  86         /*le*/u16 rec_len;
  87         u8 name_len;
  88         u8 file_type;
  89 };
  90
  91 struct dx_countlimit
  92 {
  93         le_u16 limit;
  94         le_u16 count;
  95 };
  96
  97 struct dx_entry
  98 {
  99         le_u32 hash;
 100         le_u32 block;
 101 };
 102
 103 /*
 104  * dx_root_info is laid out so that if it should somehow get overlaid by a
 105  * dirent the two low bits of the hash version will be zero.  Therefore, the
 106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 107  */
 108
 109 struct dx_root
 110 {
 111         struct fake_dirent dot;
 112         char dot_name[4];
 113         struct fake_dirent dotdot;
 114         char dotdot_name[4];
 115         struct dx_root_info
 116         {
 117                 le_u32 reserved_zero;
 118                 u8 hash_version;
 119                 u8 info_length; /* 8 */
 120                 u8 indirect_levels;
 121                 u8 unused_flags;
 122         }
 123         info;
 124         struct dx_entry entries[0];
 125 };
 126
 127 struct dx_node
 128 {
 129         struct fake_dirent fake;
 130         struct dx_entry entries[0];
 131 };
 132
 133
 134 struct dx_frame
 135 {
 136         struct buffer_head *bh;
 137         struct dx_entry *entries;
 138         struct dx_entry *at;
 139 };
 140
 141 struct dx_map_entry
 142 {
 143         u32 hash;
 144         u32 offs;
 145 };
 146
 147 #ifdef CONFIG_EXT3_INDEX
 148 static inline unsigned dx_get_block (struct dx_entry *entry);
 149 static void dx_set_block (struct dx_entry *entry, unsigned value);
 150 static inline unsigned dx_get_hash (struct dx_entry *entry);
 151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
 152 static unsigned dx_get_count (struct dx_entry *entries);
 153 static unsigned dx_get_limit (struct dx_entry *entries);
 154 static void dx_set_count (struct dx_entry *entries, unsigned value);
 155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
 156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 157 static unsigned dx_node_limit (struct inode *dir);
 158 static struct dx_frame *dx_probe(struct dentry *dentry,
 159                                  struct inode *dir,
 160                                  struct dx_hash_info *hinfo,
 161                                  struct dx_frame *frame,
 162                                  int *err);
 163 static void dx_release (struct dx_frame *frames);
 164 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
 168                 struct dx_map_entry *offsets, int count);
 169 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
 170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 172                                  struct dx_frame *frame,
 173                                  struct dx_frame *frames,
 174                                  __u32 *start_hash);
 175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 176                        struct ext3_dir_entry_2 **res_dir, int *err);
 177 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
 178                              struct inode *inode);
 179
 180 /*
 181  * Future: use high four bits of block for coalesce-on-delete flags
 182  * Mask them off for now.
 183  */
 184
 185 static inline unsigned dx_get_block (struct dx_entry *entry)
 186 {
 187         return le32_to_cpu(entry->block.v) & 0x00ffffff;
 188 }
 189
 190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 191 {
 192         entry->block.v = cpu_to_le32(value);
 193 }
 194
 195 static inline unsigned dx_get_hash (struct dx_entry *entry)
 196 {
 197         return le32_to_cpu(entry->hash.v);
 198 }
 199
 200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 201 {
 202         entry->hash.v = cpu_to_le32(value);
 203 }
 204
 205 static inline unsigned dx_get_count (struct dx_entry *entries)
 206 {
 207         return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
 208 }
 209
 210 static inline unsigned dx_get_limit (struct dx_entry *entries)
 211 {
 212         return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
 213 }
 214
 215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 216 {
 217         ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
 218 }
 219
 220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 221 {
 222         ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
 223 }
 224
 225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 226 {
 227         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 228                 EXT3_DIR_REC_LEN(2) - infosize;
 229         return 0? 20: entry_space / sizeof(struct dx_entry);
 230 }
 231
 232 static inline unsigned dx_node_limit (struct inode *dir)
 233 {
 234         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 235         return 0? 22: entry_space / sizeof(struct dx_entry);
 236 }
 237
 238 /*
 239  * Debug
 240  */
 241 #ifdef DX_DEBUG
 242 static void dx_show_index (char * label, struct dx_entry *entries)
 243 {
 244         int i, n = dx_get_count (entries);
 245         printk("%s index ", label);
 246         for (i = 0; i < n; i++)
 247         {
 248                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 249         }
 250         printk("\n");
 251 }
 252
 253 struct stats
 254 {
 255         unsigned names;
 256         unsigned space;
 257         unsigned bcount;
 258 };
 259
 260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
 261                                  int size, int show_names)
 262 {
 263         unsigned names = 0, space = 0;
 264         char *base = (char *) de;
 265         struct dx_hash_info h = *hinfo;
 266
 267         printk("names: ");
 268         while ((char *) de < base + size)
 269         {
 270                 if (de->inode)
 271                 {
 272                         if (show_names)
 273                         {
 274                                 int len = de->name_len;
 275                                 char *name = de->name;
 276                                 while (len--) printk("%c", *name++);
 277                                 ext3fs_dirhash(de->name, de->name_len, &h);
 278                                 printk(":%x.%u ", h.hash,
 279                                        ((char *) de - base));
 280                         }
 281                         space += EXT3_DIR_REC_LEN(de->name_len);
 282                         names++;
 283                 }
 284                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 285         }
 286         printk("(%i)\n", names);
 287         return (struct stats) { names, space, 1 };
 288 }
 289
 290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 291                              struct dx_entry *entries, int levels)
 292 {
 293         unsigned blocksize = dir->i_sb->s_blocksize;
 294         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 295         unsigned bcount = 0;
 296         struct buffer_head *bh;
 297         int err;
 298         printk("%i indexed blocks...\n", count);
 299         for (i = 0; i < count; i++, entries++)
 300         {
 301                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 302                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 303                 struct stats stats;
 304                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 305                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 306                 stats = levels?
 307                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 308                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
 309                 names += stats.names;
 310                 space += stats.space;
 311                 bcount += stats.bcount;
 312                 brelse (bh);
 313         }
 314         if (bcount)
 315                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 316                         names, space/bcount,(space/bcount)*100/blocksize);
 317         return (struct stats) { names, space, bcount};
 318 }
 319 #endif /* DX_DEBUG */
 320
 321 /*
 322  * Probe for a directory leaf block to search.
 323  *
 324  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 325  * error in the directory index, and the caller should fall back to
 326  * searching the directory normally.  The callers of dx_probe **MUST**
 327  * check for this error code, and make sure it never gets reflected
 328  * back to userspace.
 329  */
 330 static struct dx_frame *
 331 dx_probe(struct dentry *dentry, struct inode *dir,
 332          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 333 {
 334         unsigned count, indirect;
 335         struct dx_entry *at, *entries, *p, *q, *m;
 336         struct dx_root *root;
 337         struct buffer_head *bh;
 338         struct dx_frame *frame = frame_in;
 339         u32 hash;
 340
 341         frame->bh = NULL;
 342         if (dentry)
 343                 dir = dentry->d_parent->d_inode;
 344         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
 345                 goto fail;
 346         root = (struct dx_root *) bh->b_data;
 347         if (root->info.hash_version != DX_HASH_TEA &&
 348             root->info.hash_version != DX_HASH_HALF_MD4 &&
 349             root->info.hash_version != DX_HASH_LEGACY) {
 350                 ext3_warning(dir->i_sb, __FUNCTION__,
 351                              "Unrecognised inode hash code %d",
 352                              root->info.hash_version);
 353                 brelse(bh);
 354                 *err = ERR_BAD_DX_DIR;
 355                 goto fail;
 356         }
 357         hinfo->hash_version = root->info.hash_version;
 358         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 359         if (dentry)
 360                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
 361         hash = hinfo->hash;
 362
 363         if (root->info.unused_flags & 1) {
 364                 ext3_warning(dir->i_sb, __FUNCTION__,
 365                              "Unimplemented inode hash flags: %#06x",
 366                              root->info.unused_flags);
 367                 brelse(bh);
 368                 *err = ERR_BAD_DX_DIR;
 369                 goto fail;
 370         }
 371
 372         if ((indirect = root->info.indirect_levels) > 1) {
 373                 ext3_warning(dir->i_sb, __FUNCTION__,
 374                              "Unimplemented inode hash depth: %#06x",
 375                              root->info.indirect_levels);
 376                 brelse(bh);
 377                 *err = ERR_BAD_DX_DIR;
 378                 goto fail;
 379         }
 380
 381         entries = (struct dx_entry *) (((char *)&root->info) +
 382                                        root->info.info_length);
 383         assert(dx_get_limit(entries) == dx_root_limit(dir,
 384                                                       root->info.info_length));
 385         dxtrace (printk("Look up %x", hash));
 386         while (1)
 387         {
 388                 count = dx_get_count(entries);
 389                 assert (count && count <= dx_get_limit(entries));
 390                 p = entries + 1;
 391                 q = entries + count - 1;
 392                 while (p <= q)
 393                 {
 394                         m = p + (q - p)/2;
 395                         dxtrace(printk("."));
 396                         if (dx_get_hash(m) > hash)
 397                                 q = m - 1;
 398                         else
 399                                 p = m + 1;
 400                 }
 401
 402                 if (0) // linear search cross check
 403                 {
 404                         unsigned n = count - 1;
 405                         at = entries;
 406                         while (n--)
 407                         {
 408                                 dxtrace(printk(","));
 409                                 if (dx_get_hash(++at) > hash)
 410                                 {
 411                                         at--;
 412                                         break;
 413                                 }
 414                         }
 415                         assert (at == p - 1);
 416                 }
 417
 418                 at = p - 1;
 419                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 420                 frame->bh = bh;
 421                 frame->entries = entries;
 422                 frame->at = at;
 423                 if (!indirect--) return frame;
 424                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
 425                         goto fail2;
 426                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 427                 assert (dx_get_limit(entries) == dx_node_limit (dir));
 428                 frame++;
 429         }
 430 fail2:
 431         while (frame >= frame_in) {
 432                 brelse(frame->bh);
 433                 frame--;
 434         }
 435 fail:
 436         return NULL;
 437 }
 438
 439 static void dx_release (struct dx_frame *frames)
 440 {
 441         if (frames[0].bh == NULL)
 442                 return;
 443
 444         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 445                 brelse(frames[1].bh);
 446         brelse(frames[0].bh);
 447 }
 448
 449 /*
 450  * This function increments the frame pointer to search the next leaf
 451  * block, and reads in the necessary intervening nodes if the search
 452  * should be necessary.  Whether or not the search is necessary is
 453  * controlled by the hash parameter.  If the hash value is even, then
 454  * the search is only continued if the next block starts with that
 455  * hash value.  This is used if we are searching for a specific file.
 456  *
 457  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 458  *
 459  * This function returns 1 if the caller should continue to search,
 460  * or 0 if it should not.  If there is an error reading one of the
 461  * index blocks, it will a negative error code.
 462  *
 463  * If start_hash is non-null, it will be filled in with the starting
 464  * hash of the next page.
 465  */
 466 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 467                                  struct dx_frame *frame,
 468                                  struct dx_frame *frames,
 469                                  __u32 *start_hash)
 470 {
 471         struct dx_frame *p;
 472         struct buffer_head *bh;
 473         int err, num_frames = 0;
 474         __u32 bhash;
 475
 476         p = frame;
 477         /*
 478          * Find the next leaf page by incrementing the frame pointer.
 479          * If we run out of entries in the interior node, loop around and
 480          * increment pointer in the parent node.  When we break out of
 481          * this loop, num_frames indicates the number of interior
 482          * nodes need to be read.
 483          */
 484         while (1) {
 485                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 486                         break;
 487                 if (p == frames)
 488                         return 0;
 489                 num_frames++;
 490                 p--;
 491         }
 492
 493         /*
 494          * If the hash is 1, then continue only if the next page has a
 495          * continuation hash of any value.  This is used for readdir
 496          * handling.  Otherwise, check to see if the hash matches the
 497          * desired contiuation hash.  If it doesn't, return since
 498          * there's no point to read in the successive index pages.
 499          */
 500         bhash = dx_get_hash(p->at);
 501         if (start_hash)
 502                 *start_hash = bhash;
 503         if ((hash & 1) == 0) {
 504                 if ((bhash & ~1) != hash)
 505                         return 0;
 506         }
 507         /*
 508          * If the hash is HASH_NB_ALWAYS, we always go to the next
 509          * block so no check is necessary
 510          */
 511         while (num_frames--) {
 512                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
 513                                       0, &err)))
 514                         return err; /* Failure */
 515                 p++;
 516                 brelse (p->bh);
 517                 p->bh = bh;
 518                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 519         }
 520         return 1;
 521 }
 522
 523
 524 /*
 525  * p is at least 6 bytes before the end of page
 526  */
 527 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
 528 {
 529         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
 530 }
 531
 532 /*
 533  * This function fills a red-black tree with information from a
 534  * directory block.  It returns the number directory entries loaded
 535  * into the tree.  If there is an error it is returned in err.
 536  */
 537 static int htree_dirblock_to_tree(struct file *dir_file,
 538                                   struct inode *dir, int block,
 539                                   struct dx_hash_info *hinfo,
 540                                   __u32 start_hash, __u32 start_minor_hash)
 541 {
 542         struct buffer_head *bh;
 543         struct ext3_dir_entry_2 *de, *top;
 544         int err, count = 0;
 545
 546         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
 547         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
 548                 return err;
 549
 550         de = (struct ext3_dir_entry_2 *) bh->b_data;
 551         top = (struct ext3_dir_entry_2 *) ((char *) de +
 552                                            dir->i_sb->s_blocksize -
 553                                            EXT3_DIR_REC_LEN(0));
 554         for (; de < top; de = ext3_next_entry(de)) {
 555                 ext3fs_dirhash(de->name, de->name_len, hinfo);
 556                 if ((hinfo->hash < start_hash) ||
 557                     ((hinfo->hash == start_hash) &&
 558                      (hinfo->minor_hash < start_minor_hash)))
 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(0, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
 606         if (!frame)
 607                 return err;
 608
 609         /* Add '.' and '..' from the htree header */
 610         if (!start_hash && !start_minor_hash) {
 611                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 612                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 613                         goto errout;
 614                 de = ext3_next_entry(de);
 615                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 616                         goto errout;
 617                 count += 2;
 618         }
 619
 620         while (1) {
 621                 block = dx_get_block(frame->at);
 622                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 623                                              start_hash, start_minor_hash);
 624                 if (ret < 0) {
 625                         err = ret;
 626                         goto errout;
 627                 }
 628                 count += ret;
 629                 hashval = ~0;
 630                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 631                                             frame, frames, &hashval);
 632                 *next_hash = hashval;
 633                 if (ret < 0) {
 634                         err = ret;
 635                         goto errout;
 636                 }
 637                 /*
 638                  * Stop if:  (a) there are no more entries, or
 639                  * (b) we have inserted at least one entry and the
 640                  * next hash value is not a continuation
 641                  */
 642                 if ((ret == 0) ||
 643                     (count && ((hashval & 1) == 0)))
 644                         break;
 645         }
 646         dx_release(frames);
 647         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 648                        count, *next_hash));
 649         return count;
 650 errout:
 651         dx_release(frames);
 652         return (err);
 653 }
 654
 655
 656 /*
 657  * Directory block splitting, compacting
 658  */
 659
 660 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 661                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 662 {
 663         int count = 0;
 664         char *base = (char *) de;
 665         struct dx_hash_info h = *hinfo;
 666
 667         while ((char *) de < base + size)
 668         {
 669                 if (de->name_len && de->inode) {
 670                         ext3fs_dirhash(de->name, de->name_len, &h);
 671                         map_tail--;
 672                         map_tail->hash = h.hash;
 673                         map_tail->offs = (u32) ((char *) de - base);
 674                         count++;
 675                 }
 676                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 677                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 678         }
 679         return count;
 680 }
 681
 682 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 683 {
 684         struct dx_map_entry *p, *q, *top = map + count - 1;
 685         int more;
 686         /* Combsort until bubble sort doesn't suck */
 687         while (count > 2)
 688         {
 689                 count = count*10/13;
 690                 if (count - 9 < 2) /* 9, 10 -> 11 */
 691                         count = 11;
 692                 for (p = top, q = p - count; q >= map; p--, q--)
 693                         if (p->hash < q->hash)
 694                                 swap(*p, *q);
 695         }
 696         /* Garden variety bubble sort */
 697         do {
 698                 more = 0;
 699                 q = top;
 700                 while (q-- > map)
 701                 {
 702                         if (q[1].hash >= q[0].hash)
 703                                 continue;
 704                         swap(*(q+1), *q);
 705                         more = 1;
 706                 }
 707         } while(more);
 708 }
 709
 710 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 711 {
 712         struct dx_entry *entries = frame->entries;
 713         struct dx_entry *old = frame->at, *new = old + 1;
 714         int count = dx_get_count(entries);
 715
 716         assert(count < dx_get_limit(entries));
 717         assert(old < entries + count);
 718         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 719         dx_set_hash(new, hash);
 720         dx_set_block(new, block);
 721         dx_set_count(entries, count + 1);
 722 }
 723 #endif
 724
 725
 726 static void ext3_update_dx_flag(struct inode *inode)
 727 {
 728         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 729                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 730                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 731 }
 732
 733 /*
 734  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 735  *
 736  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 737  * `de != NULL' is guaranteed by caller.
 738  */
 739 static inline int ext3_match (int len, const char * const name,
 740                               struct ext3_dir_entry_2 * de)
 741 {
 742         if (len != de->name_len)
 743                 return 0;
 744         if (!de->inode)
 745                 return 0;
 746         return !memcmp(name, de->name, len);
 747 }
 748
 749 /*
 750  * Returns 0 if not found, -1 on failure, and 1 on success
 751  */
 752 static inline int search_dirblock(struct buffer_head * bh,
 753                                   struct inode *dir,
 754                                   struct dentry *dentry,
 755                                   unsigned long offset,
 756                                   struct ext3_dir_entry_2 ** res_dir)
 757 {
 758         struct ext3_dir_entry_2 * de;
 759         char * dlimit;
 760         int de_len;
 761         const char *name = dentry->d_name.name;
 762         int namelen = dentry->d_name.len;
 763
 764         de = (struct ext3_dir_entry_2 *) bh->b_data;
 765         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 766         while ((char *) de < dlimit) {
 767                 /* this code is executed quadratically often */
 768                 /* do minimal checking `by hand' */
 769
 770                 if ((char *) de + namelen <= dlimit &&
 771                     ext3_match (namelen, name, de)) {
 772                         /* found a match - just to be sure, do a full check */
 773                         if (!ext3_check_dir_entry("ext3_find_entry",
 774                                                   dir, de, bh, offset))
 775                                 return -1;
 776                         *res_dir = de;
 777                         return 1;
 778                 }
 779                 /* prevent looping on a bad block */
 780                 de_len = le16_to_cpu(de->rec_len);
 781                 if (de_len <= 0)
 782                         return -1;
 783                 offset += de_len;
 784                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 785         }
 786         return 0;
 787 }
 788
 789
 790 /*
 791  *      ext3_find_entry()
 792  *
 793  * finds an entry in the specified directory with the wanted name. It
 794  * returns the cache buffer in which the entry was found, and the entry
 795  * itself (as a parameter - res_dir). It does NOT read the inode of the
 796  * entry - you'll have to do that yourself if you want to.
 797  *
 798  * The returned buffer_head has ->b_count elevated.  The caller is expected
 799  * to brelse() it when appropriate.
 800  */
 801 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 802                                         struct ext3_dir_entry_2 ** res_dir)
 803 {
 804         struct super_block * sb;
 805         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 806         struct buffer_head * bh, *ret = NULL;
 807         unsigned long start, block, b;
 808         int ra_max = 0;         /* Number of bh's in the readahead
 809                                    buffer, bh_use[] */
 810         int ra_ptr = 0;         /* Current index into readahead
 811                                    buffer */
 812         int num = 0;
 813         int nblocks, i, err;
 814         struct inode *dir = dentry->d_parent->d_inode;
 815         int namelen;
 816         const u8 *name;
 817         unsigned blocksize;
 818
 819         *res_dir = NULL;
 820         sb = dir->i_sb;
 821         blocksize = sb->s_blocksize;
 822         namelen = dentry->d_name.len;
 823         name = dentry->d_name.name;
 824         if (namelen > EXT3_NAME_LEN)
 825                 return NULL;
 826 #ifdef CONFIG_EXT3_INDEX
 827         if (is_dx(dir)) {
 828                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
 829                 /*
 830                  * On success, or if the error was file not found,
 831                  * return.  Otherwise, fall back to doing a search the
 832                  * old fashioned way.
 833                  */
 834                 if (bh || (err != ERR_BAD_DX_DIR))
 835                         return bh;
 836                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 837         }
 838 #endif
 839         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 840         start = EXT3_I(dir)->i_dir_start_lookup;
 841         if (start >= nblocks)
 842                 start = 0;
 843         block = start;
 844 restart:
 845         do {
 846                 /*
 847                  * We deal with the read-ahead logic here.
 848                  */
 849                 if (ra_ptr >= ra_max) {
 850                         /* Refill the readahead buffer */
 851                         ra_ptr = 0;
 852                         b = block;
 853                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 854                                 /*
 855                                  * Terminate if we reach the end of the
 856                                  * directory and must wrap, or if our
 857                                  * search has finished at this block.
 858                                  */
 859                                 if (b >= nblocks || (num && block == start)) {
 860                                         bh_use[ra_max] = NULL;
 861                                         break;
 862                                 }
 863                                 num++;
 864                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 865                                 bh_use[ra_max] = bh;
 866                                 if (bh)
 867                                         ll_rw_block(READ, 1, &bh);
 868                         }
 869                 }
 870                 if ((bh = bh_use[ra_ptr++]) == NULL)
 871                         goto next;
 872                 wait_on_buffer(bh);
 873                 if (!buffer_uptodate(bh)) {
 874                         /* read error, skip block & hope for the best */
 875                         brelse(bh);
 876                         goto next;
 877                 }
 878                 i = search_dirblock(bh, dir, dentry,
 879                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 880                 if (i == 1) {
 881                         EXT3_I(dir)->i_dir_start_lookup = block;
 882                         ret = bh;
 883                         goto cleanup_and_exit;
 884                 } else {
 885                         brelse(bh);
 886                         if (i < 0)
 887                                 goto cleanup_and_exit;
 888                 }
 889         next:
 890                 if (++block >= nblocks)
 891                         block = 0;
 892         } while (block != start);
 893
 894         /*
 895          * If the directory has grown while we were searching, then
 896          * search the last part of the directory before giving up.
 897          */
 898         block = nblocks;
 899         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 900         if (block < nblocks) {
 901                 start = 0;
 902                 goto restart;
 903         }
 904
 905 cleanup_and_exit:
 906         /* Clean up the read-ahead blocks */
 907         for (; ra_ptr < ra_max; ra_ptr++)
 908                 brelse (bh_use[ra_ptr]);
 909         return ret;
 910 }
 911
 912 #ifdef CONFIG_EXT3_INDEX
 913 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 914                        struct ext3_dir_entry_2 **res_dir, int *err)
 915 {
 916         struct super_block * sb;
 917         struct dx_hash_info     hinfo;
 918         u32 hash;
 919         struct dx_frame frames[2], *frame;
 920         struct ext3_dir_entry_2 *de, *top;
 921         struct buffer_head *bh;
 922         unsigned long block;
 923         int retval;
 924         int namelen = dentry->d_name.len;
 925         const u8 *name = dentry->d_name.name;
 926         struct inode *dir = dentry->d_parent->d_inode;
 927
 928         sb = dir->i_sb;
 929         if (!(frame = dx_probe (dentry, 0, &hinfo, frames, err)))
 930                 return NULL;
 931         hash = hinfo.hash;
 932         do {
 933                 block = dx_get_block(frame->at);
 934                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
 935                         goto errout;
 936                 de = (struct ext3_dir_entry_2 *) bh->b_data;
 937                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
 938                                        EXT3_DIR_REC_LEN(0));
 939                 for (; de < top; de = ext3_next_entry(de))
 940                 if (ext3_match (namelen, name, de)) {
 941                         if (!ext3_check_dir_entry("ext3_find_entry",
 942                                                   dir, de, bh,
 943                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
 944                                           +((char *)de - bh->b_data))) {
 945                                 brelse (bh);
 946                                 goto errout;
 947                         }
 948                         *res_dir = de;
 949                         dx_release (frames);
 950                         return bh;
 951                 }
 952                 brelse (bh);
 953                 /* Check to see if we should continue to search */
 954                 retval = ext3_htree_next_block(dir, hash, frame,
 955                                                frames, 0);
 956                 if (retval < 0) {
 957                         ext3_warning(sb, __FUNCTION__,
 958                              "error reading index page in directory #%lu",
 959                              dir->i_ino);
 960                         *err = retval;
 961                         goto errout;
 962                 }
 963         } while (retval == 1);
 964
 965         *err = -ENOENT;
 966 errout:
 967         dxtrace(printk("%s not found\n", name));
 968         dx_release (frames);
 969         return NULL;
 970 }
 971 #endif
 972
 973 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
 974 {
 975         struct inode * inode;
 976         struct ext3_dir_entry_2 * de;
 977         struct buffer_head * bh;
 978
 979         if (dentry->d_name.len > EXT3_NAME_LEN)
 980                 return ERR_PTR(-ENAMETOOLONG);
 981
 982         bh = ext3_find_entry(dentry, &de);
 983         inode = NULL;
 984         if (bh) {
 985                 unsigned long ino = le32_to_cpu(de->inode);
 986                 brelse (bh);
 987                 inode = iget(dir->i_sb, ino);
 988
 989                 if (!inode)
 990                         return ERR_PTR(-EACCES);
 991         }
 992         if (inode)
 993                 return d_splice_alias(inode, dentry);
 994         d_add(dentry, inode);
 995         return NULL;
 996 }
 997
 998
 999 struct dentry *ext3_get_parent(struct dentry *child)
1000 {
1001         unsigned long ino;
1002         struct dentry *parent;
1003         struct inode *inode;
1004         struct dentry dotdot;
1005         struct ext3_dir_entry_2 * de;
1006         struct buffer_head *bh;
1007
1008         dotdot.d_name.name = "..";
1009         dotdot.d_name.len = 2;
1010         dotdot.d_parent = child; /* confusing, isn't it! */
1011
1012         bh = ext3_find_entry(&dotdot, &de);
1013         inode = NULL;
1014         if (!bh)
1015                 return ERR_PTR(-ENOENT);
1016         ino = le32_to_cpu(de->inode);
1017         brelse(bh);
1018         inode = iget(child->d_inode->i_sb, ino);
1019
1020         if (!inode)
1021                 return ERR_PTR(-EACCES);
1022
1023         parent = d_alloc_anon(inode);
1024         if (!parent) {
1025                 iput(inode);
1026                 parent = ERR_PTR(-ENOMEM);
1027         }
1028         return parent;
1029 }
1030
1031 #define S_SHIFT 12
1032 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1033         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1034         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1035         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1036         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1037         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1038         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1039         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1040 };
1041
1042 static inline void ext3_set_de_type(struct super_block *sb,
1043                                 struct ext3_dir_entry_2 *de,
1044                                 umode_t mode) {
1045         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1046                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1047 }
1048
1049 #ifdef CONFIG_EXT3_INDEX
1050 static struct ext3_dir_entry_2 *
1051 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1052 {
1053         unsigned rec_len = 0;
1054
1055         while (count--) {
1056                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1057                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1058                 memcpy (to, de, rec_len);
1059                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1060                                 cpu_to_le16(rec_len);
1061                 de->inode = 0;
1062                 map++;
1063                 to += rec_len;
1064         }
1065         return (struct ext3_dir_entry_2 *) (to - rec_len);
1066 }
1067
1068 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1069 {
1070         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1071         unsigned rec_len = 0;
1072
1073         prev = to = de;
1074         while ((char*)de < base + size) {
1075                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1076                                                     le16_to_cpu(de->rec_len));
1077                 if (de->inode && de->name_len) {
1078                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1079                         if (de > to)
1080                                 memmove(to, de, rec_len);
1081                         to->rec_len = cpu_to_le16(rec_len);
1082                         prev = to;
1083                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1084                 }
1085                 de = next;
1086         }
1087         return prev;
1088 }
1089
1090 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1091                         struct buffer_head **bh,struct dx_frame *frame,
1092                         struct dx_hash_info *hinfo, int *error)
1093 {
1094         unsigned blocksize = dir->i_sb->s_blocksize;
1095         unsigned count, continued;
1096         struct buffer_head *bh2;
1097         u32 newblock;
1098         u32 hash2;
1099         struct dx_map_entry *map;
1100         char *data1 = (*bh)->b_data, *data2;
1101         unsigned split;
1102         struct ext3_dir_entry_2 *de = NULL, *de2;
1103         int     err;
1104
1105         bh2 = ext3_append (handle, dir, &newblock, error);
1106         if (!(bh2)) {
1107                 brelse(*bh);
1108                 *bh = NULL;
1109                 goto errout;
1110         }
1111
1112         BUFFER_TRACE(*bh, "get_write_access");
1113         err = ext3_journal_get_write_access(handle, *bh);
1114         if (err) {
1115         journal_error:
1116                 brelse(*bh);
1117                 brelse(bh2);
1118                 *bh = NULL;
1119                 ext3_std_error(dir->i_sb, err);
1120                 goto errout;
1121         }
1122         BUFFER_TRACE(frame->bh, "get_write_access");
1123         err = ext3_journal_get_write_access(handle, frame->bh);
1124         if (err)
1125                 goto journal_error;
1126
1127         data2 = bh2->b_data;
1128
1129         /* create map in the end of data2 block */
1130         map = (struct dx_map_entry *) (data2 + blocksize);
1131         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1132                              blocksize, hinfo, map);
1133         map -= count;
1134         split = count/2; // need to adjust to actual middle
1135         dx_sort_map (map, count);
1136         hash2 = map[split].hash;
1137         continued = hash2 == map[split - 1].hash;
1138         dxtrace(printk("Split block %i at %x, %i/%i\n",
1139                 dx_get_block(frame->at), hash2, split, count-split));
1140
1141         /* Fancy dance to stay within two buffers */
1142         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1143         de = dx_pack_dirents(data1,blocksize);
1144         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1145         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1146         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1147         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1148
1149         /* Which block gets the new entry? */
1150         if (hinfo->hash >= hash2)
1151         {
1152                 swap(*bh, bh2);
1153                 de = de2;
1154         }
1155         dx_insert_block (frame, hash2 + continued, newblock);
1156         err = ext3_journal_dirty_metadata (handle, bh2);
1157         if (err)
1158                 goto journal_error;
1159         err = ext3_journal_dirty_metadata (handle, frame->bh);
1160         if (err)
1161                 goto journal_error;
1162         brelse (bh2);
1163         dxtrace(dx_show_index ("frame", frame->entries));
1164 errout:
1165         return de;
1166 }
1167 #endif
1168
1169
1170 /*
1171  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1172  * it points to a directory entry which is guaranteed to be large
1173  * enough for new directory entry.  If de is NULL, then
1174  * add_dirent_to_buf will attempt search the directory block for
1175  * space.  It will return -ENOSPC if no space is available, and -EIO
1176  * and -EEXIST if directory entry already exists.
1177  *
1178  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1179  * all other cases bh is released.
1180  */
1181 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1182                              struct inode *inode, struct ext3_dir_entry_2 *de,
1183                              struct buffer_head * bh)
1184 {
1185         struct inode    *dir = dentry->d_parent->d_inode;
1186         const char      *name = dentry->d_name.name;
1187         int             namelen = dentry->d_name.len;
1188         unsigned long   offset = 0;
1189         unsigned short  reclen;
1190         int             nlen, rlen, err;
1191         char            *top;
1192
1193         reclen = EXT3_DIR_REC_LEN(namelen);
1194         if (!de) {
1195                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1196                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1197                 while ((char *) de <= top) {
1198                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1199                                                   bh, offset)) {
1200                                 brelse (bh);
1201                                 return -EIO;
1202                         }
1203                         if (ext3_match (namelen, name, de)) {
1204                                 brelse (bh);
1205                                 return -EEXIST;
1206                         }
1207                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1208                         rlen = le16_to_cpu(de->rec_len);
1209                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1210                                 break;
1211                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1212                         offset += rlen;
1213                 }
1214                 if ((char *) de > top)
1215                         return -ENOSPC;
1216         }
1217         BUFFER_TRACE(bh, "get_write_access");
1218         err = ext3_journal_get_write_access(handle, bh);
1219         if (err) {
1220                 ext3_std_error(dir->i_sb, err);
1221                 brelse(bh);
1222                 return err;
1223         }
1224
1225         /* By now the buffer is marked for journaling */
1226         nlen = EXT3_DIR_REC_LEN(de->name_len);
1227         rlen = le16_to_cpu(de->rec_len);
1228         if (de->inode) {
1229                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1230                 de1->rec_len = cpu_to_le16(rlen - nlen);
1231                 de->rec_len = cpu_to_le16(nlen);
1232                 de = de1;
1233         }
1234         de->file_type = EXT3_FT_UNKNOWN;
1235         if (inode) {
1236                 de->inode = cpu_to_le32(inode->i_ino);
1237                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1238         } else
1239                 de->inode = 0;
1240         de->name_len = namelen;
1241         memcpy (de->name, name, namelen);
1242         /*
1243          * XXX shouldn't update any times until successful
1244          * completion of syscall, but too many callers depend
1245          * on this.
1246          *
1247          * XXX similarly, too many callers depend on
1248          * ext3_new_inode() setting the times, but error
1249          * recovery deletes the inode, so the worst that can
1250          * happen is that the times are slightly out of date
1251          * and/or different from the directory change time.
1252          */
1253         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1254         ext3_update_dx_flag(dir);
1255         dir->i_version++;
1256         ext3_mark_inode_dirty(handle, dir);
1257         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1258         err = ext3_journal_dirty_metadata(handle, bh);
1259         if (err)
1260                 ext3_std_error(dir->i_sb, err);
1261         brelse(bh);
1262         return 0;
1263 }
1264
1265 #ifdef CONFIG_EXT3_INDEX
1266 /*
1267  * This converts a one block unindexed directory to a 3 block indexed
1268  * directory, and adds the dentry to the indexed directory.
1269  */
1270 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1271                             struct inode *inode, struct buffer_head *bh)
1272 {
1273         struct inode    *dir = dentry->d_parent->d_inode;
1274         const char      *name = dentry->d_name.name;
1275         int             namelen = dentry->d_name.len;
1276         struct buffer_head *bh2;
1277         struct dx_root  *root;
1278         struct dx_frame frames[2], *frame;
1279         struct dx_entry *entries;
1280         struct ext3_dir_entry_2 *de, *de2;
1281         char            *data1, *top;
1282         unsigned        len;
1283         int             retval;
1284         unsigned        blocksize;
1285         struct dx_hash_info hinfo;
1286         u32             block;
1287         struct fake_dirent *fde;
1288
1289         blocksize =  dir->i_sb->s_blocksize;
1290         dxtrace(printk("Creating index\n"));
1291         retval = ext3_journal_get_write_access(handle, bh);
1292         if (retval) {
1293                 ext3_std_error(dir->i_sb, retval);
1294                 brelse(bh);
1295                 return retval;
1296         }
1297         root = (struct dx_root *) bh->b_data;
1298
1299         bh2 = ext3_append (handle, dir, &block, &retval);
1300         if (!(bh2)) {
1301                 brelse(bh);
1302                 return retval;
1303         }
1304         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1305         data1 = bh2->b_data;
1306
1307         /* The 0th block becomes the root, move the dirents out */
1308         fde = &root->dotdot;
1309         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1310         len = ((char *) root) + blocksize - (char *) de;
1311         memcpy (data1, de, len);
1312         de = (struct ext3_dir_entry_2 *) data1;
1313         top = data1 + len;
1314         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1315                 de = de2;
1316         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1317         /* Initialize the root; the dot dirents already exist */
1318         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1319         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1320         memset (&root->info, 0, sizeof(root->info));
1321         root->info.info_length = sizeof(root->info);
1322         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1323         entries = root->entries;
1324         dx_set_block (entries, 1);
1325         dx_set_count (entries, 1);
1326         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1327
1328         /* Initialize as for dx_probe */
1329         hinfo.hash_version = root->info.hash_version;
1330         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1331         ext3fs_dirhash(name, namelen, &hinfo);
1332         frame = frames;
1333         frame->entries = entries;
1334         frame->at = entries;
1335         frame->bh = bh;
1336         bh = bh2;
1337         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1338         dx_release (frames);
1339         if (!(de))
1340                 return retval;
1341
1342         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1343 }
1344 #endif
1345
1346 /*
1347  *      ext3_add_entry()
1348  *
1349  * adds a file entry to the specified directory, using the same
1350  * semantics as ext3_find_entry(). It returns NULL if it failed.
1351  *
1352  * NOTE!! The inode part of 'de' is left at 0 - which means you
1353  * may not sleep between calling this and putting something into
1354  * the entry, as someone else might have used it while you slept.
1355  */
1356 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1357         struct inode *inode)
1358 {
1359         struct inode *dir = dentry->d_parent->d_inode;
1360         unsigned long offset;
1361         struct buffer_head * bh;
1362         struct ext3_dir_entry_2 *de;
1363         struct super_block * sb;
1364         int     retval;
1365 #ifdef CONFIG_EXT3_INDEX
1366         int     dx_fallback=0;
1367 #endif
1368         unsigned blocksize;
1369         unsigned nlen, rlen;
1370         u32 block, blocks;
1371
1372         sb = dir->i_sb;
1373         blocksize = sb->s_blocksize;
1374         if (!dentry->d_name.len)
1375                 return -EINVAL;
1376 #ifdef CONFIG_EXT3_INDEX
1377         if (is_dx(dir)) {
1378                 retval = ext3_dx_add_entry(handle, dentry, inode);
1379                 if (!retval || (retval != ERR_BAD_DX_DIR))
1380                         return retval;
1381                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1382                 dx_fallback++;
1383                 ext3_mark_inode_dirty(handle, dir);
1384         }
1385 #endif
1386         blocks = dir->i_size >> sb->s_blocksize_bits;
1387         for (block = 0, offset = 0; block < blocks; block++) {
1388                 bh = ext3_bread(handle, dir, block, 0, &retval);
1389                 if(!bh)
1390                         return retval;
1391                 retval = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1392                 if (retval != -ENOSPC)
1393                         return retval;
1394
1395 #ifdef CONFIG_EXT3_INDEX
1396                 if (blocks == 1 && !dx_fallback &&
1397                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1398                         return make_indexed_dir(handle, dentry, inode, bh);
1399 #endif
1400                 brelse(bh);
1401         }
1402         bh = ext3_append(handle, dir, &block, &retval);
1403         if (!bh)
1404                 return retval;
1405         de = (struct ext3_dir_entry_2 *) bh->b_data;
1406         de->inode = 0;
1407         de->rec_len = cpu_to_le16(rlen = blocksize);
1408         nlen = 0;
1409         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1410 }
1411
1412 #ifdef CONFIG_EXT3_INDEX
1413 /*
1414  * Returns 0 for success, or a negative error value
1415  */
1416 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1417                              struct inode *inode)
1418 {
1419         struct dx_frame frames[2], *frame;
1420         struct dx_entry *entries, *at;
1421         struct dx_hash_info hinfo;
1422         struct buffer_head * bh;
1423         struct inode *dir = dentry->d_parent->d_inode;
1424         struct super_block * sb = dir->i_sb;
1425         struct ext3_dir_entry_2 *de;
1426         int err;
1427
1428         frame = dx_probe(dentry, 0, &hinfo, frames, &err);
1429         if (!frame)
1430                 return err;
1431         entries = frame->entries;
1432         at = frame->at;
1433
1434         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1435                 goto cleanup;
1436
1437         BUFFER_TRACE(bh, "get_write_access");
1438         err = ext3_journal_get_write_access(handle, bh);
1439         if (err)
1440                 goto journal_error;
1441
1442         err = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1443         if (err != -ENOSPC) {
1444                 bh = 0;
1445                 goto cleanup;
1446         }
1447
1448         /* Block full, should compress but for now just split */
1449         dxtrace(printk("using %u of %u node entries\n",
1450                        dx_get_count(entries), dx_get_limit(entries)));
1451         /* Need to split index? */
1452         if (dx_get_count(entries) == dx_get_limit(entries)) {
1453                 u32 newblock;
1454                 unsigned icount = dx_get_count(entries);
1455                 int levels = frame - frames;
1456                 struct dx_entry *entries2;
1457                 struct dx_node *node2;
1458                 struct buffer_head *bh2;
1459
1460                 if (levels && (dx_get_count(frames->entries) ==
1461                                dx_get_limit(frames->entries))) {
1462                         ext3_warning(sb, __FUNCTION__,
1463                                      "Directory index full!\n");
1464                         err = -ENOSPC;
1465                         goto cleanup;
1466                 }
1467                 bh2 = ext3_append (handle, dir, &newblock, &err);
1468                 if (!(bh2))
1469                         goto cleanup;
1470                 node2 = (struct dx_node *)(bh2->b_data);
1471                 entries2 = node2->entries;
1472                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1473                 node2->fake.inode = 0;
1474                 BUFFER_TRACE(frame->bh, "get_write_access");
1475                 err = ext3_journal_get_write_access(handle, frame->bh);
1476                 if (err)
1477                         goto journal_error;
1478                 if (levels) {
1479                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1480                         unsigned hash2 = dx_get_hash(entries + icount1);
1481                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1482
1483                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1484                         err = ext3_journal_get_write_access(handle,
1485                                                              frames[0].bh);
1486                         if (err)
1487                                 goto journal_error;
1488
1489                         memcpy ((char *) entries2, (char *) (entries + icount1),
1490                                 icount2 * sizeof(struct dx_entry));
1491                         dx_set_count (entries, icount1);
1492                         dx_set_count (entries2, icount2);
1493                         dx_set_limit (entries2, dx_node_limit(dir));
1494
1495                         /* Which index block gets the new entry? */
1496                         if (at - entries >= icount1) {
1497                                 frame->at = at = at - entries - icount1 + entries2;
1498                                 frame->entries = entries = entries2;
1499                                 swap(frame->bh, bh2);
1500                         }
1501                         dx_insert_block (frames + 0, hash2, newblock);
1502                         dxtrace(dx_show_index ("node", frames[1].entries));
1503                         dxtrace(dx_show_index ("node",
1504                                ((struct dx_node *) bh2->b_data)->entries));
1505                         err = ext3_journal_dirty_metadata(handle, bh2);
1506                         if (err)
1507                                 goto journal_error;
1508                         brelse (bh2);
1509                 } else {
1510                         dxtrace(printk("Creating second level index...\n"));
1511                         memcpy((char *) entries2, (char *) entries,
1512                                icount * sizeof(struct dx_entry));
1513                         dx_set_limit(entries2, dx_node_limit(dir));
1514
1515                         /* Set up root */
1516                         dx_set_count(entries, 1);
1517                         dx_set_block(entries + 0, newblock);
1518                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1519
1520                         /* Add new access path frame */
1521                         frame = frames + 1;
1522                         frame->at = at = at - entries + entries2;
1523                         frame->entries = entries = entries2;
1524                         frame->bh = bh2;
1525                         err = ext3_journal_get_write_access(handle,
1526                                                              frame->bh);
1527                         if (err)
1528                                 goto journal_error;
1529                 }
1530                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1531         }
1532         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1533         if (!de)
1534                 goto cleanup;
1535         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1536         bh = 0;
1537         goto cleanup;
1538
1539 journal_error:
1540         ext3_std_error(dir->i_sb, err);
1541 cleanup:
1542         if (bh)
1543                 brelse(bh);
1544         dx_release(frames);
1545         return err;
1546 }
1547 #endif
1548
1549 /*
1550  * ext3_delete_entry deletes a directory entry by merging it with the
1551  * previous entry
1552  */
1553 static int ext3_delete_entry (handle_t *handle,
1554                               struct inode * dir,
1555                               struct ext3_dir_entry_2 * de_del,
1556                               struct buffer_head * bh)
1557 {
1558         struct ext3_dir_entry_2 * de, * pde;
1559         int i;
1560
1561         i = 0;
1562         pde = NULL;
1563         de = (struct ext3_dir_entry_2 *) bh->b_data;
1564         while (i < bh->b_size) {
1565                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1566                         return -EIO;
1567                 if (de == de_del)  {
1568                         BUFFER_TRACE(bh, "get_write_access");
1569                         ext3_journal_get_write_access(handle, bh);
1570                         if (pde)
1571                                 pde->rec_len =
1572                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1573                                                     le16_to_cpu(de->rec_len));
1574                         else
1575                                 de->inode = 0;
1576                         dir->i_version++;
1577                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1578                         ext3_journal_dirty_metadata(handle, bh);
1579                         return 0;
1580                 }
1581                 i += le16_to_cpu(de->rec_len);
1582                 pde = de;
1583                 de = (struct ext3_dir_entry_2 *)
1584                         ((char *) de + le16_to_cpu(de->rec_len));
1585         }
1586         return -ENOENT;
1587 }
1588
1589 /*
1590  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1591  * do not perform it in these functions.  We perform it at the call site,
1592  * if it is needed.
1593  */
1594 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1595 {
1596         inode->i_nlink++;
1597 }
1598
1599 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1600 {
1601         inode->i_nlink--;
1602 }
1603
1604 static int ext3_add_nondir(handle_t *handle,
1605                 struct dentry *dentry, struct inode *inode)
1606 {
1607         int err = ext3_add_entry(handle, dentry, inode);
1608         if (!err) {
1609                 ext3_mark_inode_dirty(handle, inode);
1610                 d_instantiate(dentry, inode);
1611                 return 0;
1612         }
1613         ext3_dec_count(handle, inode);
1614         iput(inode);
1615         return err;
1616 }
1617
1618 /*
1619  * By the time this is called, we already have created
1620  * the directory cache entry for the new file, but it
1621  * is so far negative - it has no inode.
1622  *
1623  * If the create succeeds, we fill in the inode information
1624  * with d_instantiate().
1625  */
1626 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1627                 struct nameidata *nd)
1628 {
1629         handle_t *handle;
1630         struct inode * inode;
1631         int err;
1632
1633         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1634                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1635                                         2*EXT3_QUOTA_INIT_BLOCKS);
1636         if (IS_ERR(handle))
1637                 return PTR_ERR(handle);
1638
1639         if (IS_DIRSYNC(dir))
1640                 handle->h_sync = 1;
1641
1642         inode = ext3_new_inode (handle, dir, mode);
1643         err = PTR_ERR(inode);
1644         if (!IS_ERR(inode)) {
1645                 inode->i_op = &ext3_file_inode_operations;
1646                 inode->i_fop = &ext3_file_operations;
1647                 ext3_set_aops(inode);
1648                 err = ext3_add_nondir(handle, dentry, inode);
1649         }
1650         ext3_journal_stop(handle);
1651         return err;
1652 }
1653
1654 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1655                         int mode, dev_t rdev)
1656 {
1657         handle_t *handle;
1658         struct inode *inode;
1659         int err;
1660
1661         if (!new_valid_dev(rdev))
1662                 return -EINVAL;
1663
1664         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1665                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1666                                         2*EXT3_QUOTA_INIT_BLOCKS);
1667         if (IS_ERR(handle))
1668                 return PTR_ERR(handle);
1669
1670         if (IS_DIRSYNC(dir))
1671                 handle->h_sync = 1;
1672
1673         inode = ext3_new_inode (handle, dir, mode);
1674         err = PTR_ERR(inode);
1675         if (!IS_ERR(inode)) {
1676                 init_special_inode(inode, inode->i_mode, rdev);
1677 #ifdef CONFIG_EXT3_FS_XATTR
1678                 inode->i_op = &ext3_special_inode_operations;
1679 #endif
1680                 err = ext3_add_nondir(handle, dentry, inode);
1681         }
1682         ext3_journal_stop(handle);
1683         return err;
1684 }
1685
1686 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1687 {
1688         handle_t *handle;
1689         struct inode * inode;
1690         struct buffer_head * dir_block;
1691         struct ext3_dir_entry_2 * de;
1692         int err;
1693
1694         if (dir->i_nlink >= EXT3_LINK_MAX)
1695                 return -EMLINK;
1696
1697         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1698                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1699                                         2*EXT3_QUOTA_INIT_BLOCKS);
1700         if (IS_ERR(handle))
1701                 return PTR_ERR(handle);
1702
1703         if (IS_DIRSYNC(dir))
1704                 handle->h_sync = 1;
1705
1706         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1707         err = PTR_ERR(inode);
1708         if (IS_ERR(inode))
1709                 goto out_stop;
1710
1711         inode->i_op = &ext3_dir_inode_operations;
1712         inode->i_fop = &ext3_dir_operations;
1713         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1714         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1715         if (!dir_block) {
1716                 inode->i_nlink--; /* is this nlink == 0? */
1717                 ext3_mark_inode_dirty(handle, inode);
1718                 iput (inode);
1719                 goto out_stop;
1720         }
1721         BUFFER_TRACE(dir_block, "get_write_access");
1722         ext3_journal_get_write_access(handle, dir_block);
1723         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1724         de->inode = cpu_to_le32(inode->i_ino);
1725         de->name_len = 1;
1726         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1727         strcpy (de->name, ".");
1728         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1729         de = (struct ext3_dir_entry_2 *)
1730                         ((char *) de + le16_to_cpu(de->rec_len));
1731         de->inode = cpu_to_le32(dir->i_ino);
1732         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1733         de->name_len = 2;
1734         strcpy (de->name, "..");
1735         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1736         inode->i_nlink = 2;
1737         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1738         ext3_journal_dirty_metadata(handle, dir_block);
1739         brelse (dir_block);
1740         ext3_mark_inode_dirty(handle, inode);
1741         err = ext3_add_entry (handle, dentry, inode);
1742         if (err) {
1743                 inode->i_nlink = 0;
1744                 ext3_mark_inode_dirty(handle, inode);
1745                 iput (inode);
1746                 goto out_stop;
1747         }
1748         dir->i_nlink++;
1749         ext3_update_dx_flag(dir);
1750         ext3_mark_inode_dirty(handle, dir);
1751         d_instantiate(dentry, inode);
1752 out_stop:
1753         ext3_journal_stop(handle);
1754         return err;
1755 }
1756
1757 /*
1758  * routine to check that the specified directory is empty (for rmdir)
1759  */
1760 static int empty_dir (struct inode * inode)
1761 {
1762         unsigned long offset;
1763         struct buffer_head * bh;
1764         struct ext3_dir_entry_2 * de, * de1;
1765         struct super_block * sb;
1766         int err;
1767
1768         sb = inode->i_sb;
1769         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1770             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1771                 ext3_warning (inode->i_sb, "empty_dir",
1772                               "bad directory (dir #%lu) - no data block",
1773                               inode->i_ino);
1774                 return 1;
1775         }
1776         de = (struct ext3_dir_entry_2 *) bh->b_data;
1777         de1 = (struct ext3_dir_entry_2 *)
1778                         ((char *) de + le16_to_cpu(de->rec_len));
1779         if (le32_to_cpu(de->inode) != inode->i_ino ||
1780                         !le32_to_cpu(de1->inode) ||
1781                         strcmp (".", de->name) ||
1782                         strcmp ("..", de1->name)) {
1783                 ext3_warning (inode->i_sb, "empty_dir",
1784                               "bad directory (dir #%lu) - no `.' or `..'",
1785                               inode->i_ino);
1786                 brelse (bh);
1787                 return 1;
1788         }
1789         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1790         de = (struct ext3_dir_entry_2 *)
1791                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1792         while (offset < inode->i_size ) {
1793                 if (!bh ||
1794                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1795                         brelse (bh);
1796                         bh = ext3_bread (NULL, inode,
1797                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1798                         if (!bh) {
1799 #if 0
1800                                 ext3_error (sb, "empty_dir",
1801                                 "directory #%lu contains a hole at offset %lu",
1802                                         inode->i_ino, offset);
1803 #endif
1804                                 offset += sb->s_blocksize;
1805                                 continue;
1806                         }
1807                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1808                 }
1809                 if (!ext3_check_dir_entry ("empty_dir", inode, de, bh,
1810                                            offset)) {
1811                         brelse (bh);
1812                         return 1;
1813                 }
1814                 if (le32_to_cpu(de->inode)) {
1815                         brelse (bh);
1816                         return 0;
1817                 }
1818                 offset += le16_to_cpu(de->rec_len);
1819                 de = (struct ext3_dir_entry_2 *)
1820                                 ((char *) de + le16_to_cpu(de->rec_len));
1821         }
1822         brelse (bh);
1823         return 1;
1824 }
1825
1826 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1827  * such inodes, starting at the superblock, in case we crash before the
1828  * file is closed/deleted, or in case the inode truncate spans multiple
1829  * transactions and the last transaction is not recovered after a crash.
1830  *
1831  * At filesystem recovery time, we walk this list deleting unlinked
1832  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1833  */
1834 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1835 {
1836         struct super_block *sb = inode->i_sb;
1837         struct ext3_iloc iloc;
1838         int err = 0, rc;
1839
1840         lock_super(sb);
1841         if (!list_empty(&EXT3_I(inode)->i_orphan))
1842                 goto out_unlock;
1843
1844         /* Orphan handling is only valid for files with data blocks
1845          * being truncated, or files being unlinked. */
1846
1847         /* @@@ FIXME: Observation from aviro:
1848          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1849          * here (on lock_super()), so race with ext3_link() which might bump
1850          * ->i_nlink. For, say it, character device. Not a regular file,
1851          * not a directory, not a symlink and ->i_nlink > 0.
1852          */
1853         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1854                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1855
1856         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1857         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1858         if (err)
1859                 goto out_unlock;
1860
1861         err = ext3_reserve_inode_write(handle, inode, &iloc);
1862         if (err)
1863                 goto out_unlock;
1864
1865         /* Insert this inode at the head of the on-disk orphan list... */
1866         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1867         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1868         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1869         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1870         if (!err)
1871                 err = rc;
1872
1873         /* Only add to the head of the in-memory list if all the
1874          * previous operations succeeded.  If the orphan_add is going to
1875          * fail (possibly taking the journal offline), we can't risk
1876          * leaving the inode on the orphan list: stray orphan-list
1877          * entries can cause panics at unmount time.
1878          *
1879          * This is safe: on error we're going to ignore the orphan list
1880          * anyway on the next recovery. */
1881         if (!err)
1882                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1883
1884         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1885         jbd_debug(4, "orphan inode %ld will point to %d\n",
1886                         inode->i_ino, NEXT_ORPHAN(inode));
1887 out_unlock:
1888         unlock_super(sb);
1889         ext3_std_error(inode->i_sb, err);
1890         return err;
1891 }
1892
1893 /*
1894  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1895  * of such inodes stored on disk, because it is finally being cleaned up.
1896  */
1897 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1898 {
1899         struct list_head *prev;
1900         struct ext3_inode_info *ei = EXT3_I(inode);
1901         struct ext3_sb_info *sbi;
1902         unsigned long ino_next;
1903         struct ext3_iloc iloc;
1904         int err = 0;
1905
1906         lock_super(inode->i_sb);
1907         if (list_empty(&ei->i_orphan)) {
1908                 unlock_super(inode->i_sb);
1909                 return 0;
1910         }
1911
1912         ino_next = NEXT_ORPHAN(inode);
1913         prev = ei->i_orphan.prev;
1914         sbi = EXT3_SB(inode->i_sb);
1915
1916         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1917
1918         list_del_init(&ei->i_orphan);
1919
1920         /* If we're on an error path, we may not have a valid
1921          * transaction handle with which to update the orphan list on
1922          * disk, but we still need to remove the inode from the linked
1923          * list in memory. */
1924         if (!handle)
1925                 goto out;
1926
1927         err = ext3_reserve_inode_write(handle, inode, &iloc);
1928         if (err)
1929                 goto out_err;
1930
1931         if (prev == &sbi->s_orphan) {
1932                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1933                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1934                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1935                 if (err)
1936                         goto out_brelse;
1937                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1938                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1939         } else {
1940                 struct ext3_iloc iloc2;
1941                 struct inode *i_prev =
1942                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1943
1944                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1945                           i_prev->i_ino, ino_next);
1946                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1947                 if (err)
1948                         goto out_brelse;
1949                 NEXT_ORPHAN(i_prev) = ino_next;
1950                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1951         }
1952         if (err)
1953                 goto out_brelse;
1954         NEXT_ORPHAN(inode) = 0;
1955         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1956         if (err)
1957                 goto out_brelse;
1958
1959 out_err:
1960         ext3_std_error(inode->i_sb, err);
1961 out:
1962         unlock_super(inode->i_sb);
1963         return err;
1964
1965 out_brelse:
1966         brelse(iloc.bh);
1967         goto out_err;
1968 }
1969
1970 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1971 {
1972         int retval;
1973         struct inode * inode;
1974         struct buffer_head * bh;
1975         struct ext3_dir_entry_2 * de;
1976         handle_t *handle;
1977
1978         /* Initialize quotas before so that eventual writes go in
1979          * separate transaction */
1980         DQUOT_INIT(dentry->d_inode);
1981         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
1982         if (IS_ERR(handle))
1983                 return PTR_ERR(handle);
1984
1985         retval = -ENOENT;
1986         bh = ext3_find_entry (dentry, &de);
1987         if (!bh)
1988                 goto end_rmdir;
1989
1990         if (IS_DIRSYNC(dir))
1991                 handle->h_sync = 1;
1992
1993         inode = dentry->d_inode;
1994
1995         retval = -EIO;
1996         if (le32_to_cpu(de->inode) != inode->i_ino)
1997                 goto end_rmdir;
1998
1999         retval = -ENOTEMPTY;
2000         if (!empty_dir (inode))
2001                 goto end_rmdir;
2002
2003         retval = ext3_delete_entry(handle, dir, de, bh);
2004         if (retval)
2005                 goto end_rmdir;
2006         if (inode->i_nlink != 2)
2007                 ext3_warning (inode->i_sb, "ext3_rmdir",
2008                               "empty directory has nlink!=2 (%d)",
2009                               inode->i_nlink);
2010         inode->i_version++;
2011         inode->i_nlink = 0;
2012         /* There's no need to set i_disksize: the fact that i_nlink is
2013          * zero will ensure that the right thing happens during any
2014          * recovery. */
2015         inode->i_size = 0;
2016         ext3_orphan_add(handle, inode);
2017         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2018         ext3_mark_inode_dirty(handle, inode);
2019         dir->i_nlink--;
2020         ext3_update_dx_flag(dir);
2021         ext3_mark_inode_dirty(handle, dir);
2022
2023 end_rmdir:
2024         ext3_journal_stop(handle);
2025         brelse (bh);
2026         return retval;
2027 }
2028
2029 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2030 {
2031         int retval;
2032         struct inode * inode;
2033         struct buffer_head * bh;
2034         struct ext3_dir_entry_2 * de;
2035         handle_t *handle;
2036
2037         /* Initialize quotas before so that eventual writes go
2038          * in separate transaction */
2039         DQUOT_INIT(dentry->d_inode);
2040         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2041         if (IS_ERR(handle))
2042                 return PTR_ERR(handle);
2043
2044         if (IS_DIRSYNC(dir))
2045                 handle->h_sync = 1;
2046
2047         retval = -ENOENT;
2048         bh = ext3_find_entry (dentry, &de);
2049         if (!bh)
2050                 goto end_unlink;
2051
2052         inode = dentry->d_inode;
2053
2054         retval = -EIO;
2055         if (le32_to_cpu(de->inode) != inode->i_ino)
2056                 goto end_unlink;
2057
2058         if (!inode->i_nlink) {
2059                 ext3_warning (inode->i_sb, "ext3_unlink",
2060                               "Deleting nonexistent file (%lu), %d",
2061                               inode->i_ino, inode->i_nlink);
2062                 inode->i_nlink = 1;
2063         }
2064         retval = ext3_delete_entry(handle, dir, de, bh);
2065         if (retval)
2066                 goto end_unlink;
2067         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2068         ext3_update_dx_flag(dir);
2069         ext3_mark_inode_dirty(handle, dir);
2070         inode->i_nlink--;
2071         if (!inode->i_nlink)
2072                 ext3_orphan_add(handle, inode);
2073         inode->i_ctime = dir->i_ctime;
2074         ext3_mark_inode_dirty(handle, inode);
2075         retval = 0;
2076
2077 end_unlink:
2078         ext3_journal_stop(handle);
2079         brelse (bh);
2080         return retval;
2081 }
2082
2083 static int ext3_symlink (struct inode * dir,
2084                 struct dentry *dentry, const char * symname)
2085 {
2086         handle_t *handle;
2087         struct inode * inode;
2088         int l, err;
2089
2090         l = strlen(symname)+1;
2091         if (l > dir->i_sb->s_blocksize)
2092                 return -ENAMETOOLONG;
2093
2094         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2095                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2096                                         2*EXT3_QUOTA_INIT_BLOCKS);
2097         if (IS_ERR(handle))
2098                 return PTR_ERR(handle);
2099
2100         if (IS_DIRSYNC(dir))
2101                 handle->h_sync = 1;
2102
2103         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2104         err = PTR_ERR(inode);
2105         if (IS_ERR(inode))
2106                 goto out_stop;
2107
2108         if (l > sizeof (EXT3_I(inode)->i_data)) {
2109                 inode->i_op = &ext3_symlink_inode_operations;
2110                 ext3_set_aops(inode);
2111                 /*
2112                  * page_symlink() calls into ext3_prepare/commit_write.
2113                  * We have a transaction open.  All is sweetness.  It also sets
2114                  * i_size in generic_commit_write().
2115                  */
2116                 err = page_symlink(inode, symname, l);
2117                 if (err) {
2118                         ext3_dec_count(handle, inode);
2119                         ext3_mark_inode_dirty(handle, inode);
2120                         iput (inode);
2121                         goto out_stop;
2122                 }
2123         } else {
2124                 inode->i_op = &ext3_fast_symlink_inode_operations;
2125                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2126                 inode->i_size = l-1;
2127         }
2128         EXT3_I(inode)->i_disksize = inode->i_size;
2129         err = ext3_add_nondir(handle, dentry, inode);
2130 out_stop:
2131         ext3_journal_stop(handle);
2132         return err;
2133 }
2134
2135 static int ext3_link (struct dentry * old_dentry,
2136                 struct inode * dir, struct dentry *dentry)
2137 {
2138         handle_t *handle;
2139         struct inode *inode = old_dentry->d_inode;
2140         int err;
2141
2142         if (inode->i_nlink >= EXT3_LINK_MAX)
2143                 return -EMLINK;
2144
2145         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2146                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2147         if (IS_ERR(handle))
2148                 return PTR_ERR(handle);
2149
2150         if (IS_DIRSYNC(dir))
2151                 handle->h_sync = 1;
2152
2153         inode->i_ctime = CURRENT_TIME;
2154         ext3_inc_count(handle, inode);
2155         atomic_inc(&inode->i_count);
2156
2157         err = ext3_add_nondir(handle, dentry, inode);
2158         ext3_journal_stop(handle);
2159         return err;
2160 }
2161
2162 #define PARENT_INO(buffer) \
2163         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2164         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2165
2166 /*
2167  * Anybody can rename anything with this: the permission checks are left to the
2168  * higher-level routines.
2169  */
2170 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2171                            struct inode * new_dir,struct dentry *new_dentry)
2172 {
2173         handle_t *handle;
2174         struct inode * old_inode, * new_inode;
2175         struct buffer_head * old_bh, * new_bh, * dir_bh;
2176         struct ext3_dir_entry_2 * old_de, * new_de;
2177         int retval;
2178
2179         old_bh = new_bh = dir_bh = NULL;
2180
2181         /* Initialize quotas before so that eventual writes go
2182          * in separate transaction */
2183         if (new_dentry->d_inode)
2184                 DQUOT_INIT(new_dentry->d_inode);
2185         handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2186                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2187         if (IS_ERR(handle))
2188                 return PTR_ERR(handle);
2189
2190         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2191                 handle->h_sync = 1;
2192
2193         old_bh = ext3_find_entry (old_dentry, &old_de);
2194         /*
2195          *  Check for inode number is _not_ due to possible IO errors.
2196          *  We might rmdir the source, keep it as pwd of some process
2197          *  and merrily kill the link to whatever was created under the
2198          *  same name. Goodbye sticky bit ;-<
2199          */
2200         old_inode = old_dentry->d_inode;
2201         retval = -ENOENT;
2202         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2203                 goto end_rename;
2204
2205         new_inode = new_dentry->d_inode;
2206         new_bh = ext3_find_entry (new_dentry, &new_de);
2207         if (new_bh) {
2208                 if (!new_inode) {
2209                         brelse (new_bh);
2210                         new_bh = NULL;
2211                 }
2212         }
2213         if (S_ISDIR(old_inode->i_mode)) {
2214                 if (new_inode) {
2215                         retval = -ENOTEMPTY;
2216                         if (!empty_dir (new_inode))
2217                                 goto end_rename;
2218                 }
2219                 retval = -EIO;
2220                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2221                 if (!dir_bh)
2222                         goto end_rename;
2223                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2224                         goto end_rename;
2225                 retval = -EMLINK;
2226                 if (!new_inode && new_dir!=old_dir &&
2227                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2228                         goto end_rename;
2229         }
2230         if (!new_bh) {
2231                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2232                 if (retval)
2233                         goto end_rename;
2234         } else {
2235                 BUFFER_TRACE(new_bh, "get write access");
2236                 ext3_journal_get_write_access(handle, new_bh);
2237                 new_de->inode = le32_to_cpu(old_inode->i_ino);
2238                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2239                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2240                         new_de->file_type = old_de->file_type;
2241                 new_dir->i_version++;
2242                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2243                 ext3_journal_dirty_metadata(handle, new_bh);
2244                 brelse(new_bh);
2245                 new_bh = NULL;
2246         }
2247
2248         /*
2249          * Like most other Unix systems, set the ctime for inodes on a
2250          * rename.
2251          */
2252         old_inode->i_ctime = CURRENT_TIME;
2253         ext3_mark_inode_dirty(handle, old_inode);
2254
2255         /*
2256          * ok, that's it
2257          */
2258         retval = ext3_delete_entry(handle, old_dir, old_de, old_bh);
2259         if (retval == -ENOENT) {
2260                 /*
2261                  * old_de could have moved out from under us.
2262                  */
2263                 struct buffer_head *old_bh2;
2264                 struct ext3_dir_entry_2 *old_de2;
2265
2266                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2267                 if (old_bh2) {
2268                         retval = ext3_delete_entry(handle, old_dir,
2269                                                    old_de2, old_bh2);
2270                         brelse(old_bh2);
2271                 }
2272         }
2273         if (retval) {
2274                 ext3_warning(old_dir->i_sb, "ext3_rename",
2275                                 "Deleting old file (%lu), %d, error=%d",
2276                                 old_dir->i_ino, old_dir->i_nlink, retval);
2277         }
2278
2279         if (new_inode) {
2280                 new_inode->i_nlink--;
2281                 new_inode->i_ctime = CURRENT_TIME;
2282         }
2283         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
2284         ext3_update_dx_flag(old_dir);
2285         if (dir_bh) {
2286                 BUFFER_TRACE(dir_bh, "get_write_access");
2287                 ext3_journal_get_write_access(handle, dir_bh);
2288                 PARENT_INO(dir_bh->b_data) = le32_to_cpu(new_dir->i_ino);
2289                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2290                 ext3_journal_dirty_metadata(handle, dir_bh);
2291                 old_dir->i_nlink--;
2292                 if (new_inode) {
2293                         new_inode->i_nlink--;
2294                 } else {
2295                         new_dir->i_nlink++;
2296                         ext3_update_dx_flag(new_dir);
2297                         ext3_mark_inode_dirty(handle, new_dir);
2298                 }
2299         }
2300         ext3_mark_inode_dirty(handle, old_dir);
2301         if (new_inode) {
2302                 ext3_mark_inode_dirty(handle, new_inode);
2303                 if (!new_inode->i_nlink)
2304                         ext3_orphan_add(handle, new_inode);
2305         }
2306         retval = 0;
2307
2308 end_rename:
2309         brelse (dir_bh);
2310         brelse (old_bh);
2311         brelse (new_bh);
2312         ext3_journal_stop(handle);
2313         return retval;
2314 }
2315
2316 /*
2317  * directories can handle most operations...
2318  */
2319 struct inode_operations ext3_dir_inode_operations = {
2320         .create         = ext3_create,
2321         .lookup         = ext3_lookup,
2322         .link           = ext3_link,
2323         .unlink         = ext3_unlink,
2324         .symlink        = ext3_symlink,
2325         .mkdir          = ext3_mkdir,
2326         .rmdir          = ext3_rmdir,
2327         .mknod          = ext3_mknod,
2328         .rename         = ext3_rename,
2329         .setattr        = ext3_setattr,
2330         .setxattr       = ext3_setxattr,
2331         .getxattr       = ext3_getxattr,
2332         .listxattr      = ext3_listxattr,
2333         .removexattr    = ext3_removexattr,
2334         .permission     = ext3_permission,
2335 };
2336
2337 struct inode_operations ext3_special_inode_operations = {
2338         .setattr        = ext3_setattr,
2339         .setxattr       = ext3_setxattr,
2340         .getxattr       = ext3_getxattr,
2341         .listxattr      = ext3_listxattr,
2342         .removexattr    = ext3_removexattr,
2343         .permission     = ext3_permission,
2344 };