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 (de->inode == 0)
 561                         continue;
 562                 if ((err = ext3_htree_store_dirent(dir_file,
 563                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 564                         brelse(bh);
 565                         return err;
 566                 }
 567                 count++;
 568         }
 569         brelse(bh);
 570         return count;
 571 }
 572
 573
 574 /*
 575  * This function fills a red-black tree with information from a
 576  * directory.  We start scanning the directory in hash order, starting
 577  * at start_hash and start_minor_hash.
 578  *
 579  * This function returns the number of entries inserted into the tree,
 580  * or a negative error code.
 581  */
 582 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 583                          __u32 start_minor_hash, __u32 *next_hash)
 584 {
 585         struct dx_hash_info hinfo;
 586         struct ext3_dir_entry_2 *de;
 587         struct dx_frame frames[2], *frame;
 588         struct inode *dir;
 589         int block, err;
 590         int count = 0;
 591         int ret;
 592         __u32 hashval;
 593
 594         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
 595                        start_minor_hash));
 596         dir = dir_file->f_dentry->d_inode;
 597         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
 598                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
 599                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
 600                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 601                                                start_hash, start_minor_hash);
 602                 *next_hash = ~0;
 603                 return count;
 604         }
 605         hinfo.hash = start_hash;
 606         hinfo.minor_hash = 0;
 607         frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
 608         if (!frame)
 609                 return err;
 610
 611         /* Add '.' and '..' from the htree header */
 612         if (!start_hash && !start_minor_hash) {
 613                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
 614                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 615                         goto errout;
 616                 de = ext3_next_entry(de);
 617                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 618                         goto errout;
 619                 count += 2;
 620         }
 621
 622         while (1) {
 623                 block = dx_get_block(frame->at);
 624                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 625                                              start_hash, start_minor_hash);
 626                 if (ret < 0) {
 627                         err = ret;
 628                         goto errout;
 629                 }
 630                 count += ret;
 631                 hashval = ~0;
 632                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
 633                                             frame, frames, &hashval);
 634                 *next_hash = hashval;
 635                 if (ret < 0) {
 636                         err = ret;
 637                         goto errout;
 638                 }
 639                 /*
 640                  * Stop if:  (a) there are no more entries, or
 641                  * (b) we have inserted at least one entry and the
 642                  * next hash value is not a continuation
 643                  */
 644                 if ((ret == 0) ||
 645                     (count && ((hashval & 1) == 0)))
 646                         break;
 647         }
 648         dx_release(frames);
 649         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
 650                        count, *next_hash));
 651         return count;
 652 errout:
 653         dx_release(frames);
 654         return (err);
 655 }
 656
 657
 658 /*
 659  * Directory block splitting, compacting
 660  */
 661
 662 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 663                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 664 {
 665         int count = 0;
 666         char *base = (char *) de;
 667         struct dx_hash_info h = *hinfo;
 668
 669         while ((char *) de < base + size)
 670         {
 671                 if (de->name_len && de->inode) {
 672                         ext3fs_dirhash(de->name, de->name_len, &h);
 673                         map_tail--;
 674                         map_tail->hash = h.hash;
 675                         map_tail->offs = (u32) ((char *) de - base);
 676                         count++;
 677                 }
 678                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 679                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 680         }
 681         return count;
 682 }
 683
 684 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 685 {
 686         struct dx_map_entry *p, *q, *top = map + count - 1;
 687         int more;
 688         /* Combsort until bubble sort doesn't suck */
 689         while (count > 2)
 690         {
 691                 count = count*10/13;
 692                 if (count - 9 < 2) /* 9, 10 -> 11 */
 693                         count = 11;
 694                 for (p = top, q = p - count; q >= map; p--, q--)
 695                         if (p->hash < q->hash)
 696                                 swap(*p, *q);
 697         }
 698         /* Garden variety bubble sort */
 699         do {
 700                 more = 0;
 701                 q = top;
 702                 while (q-- > map)
 703                 {
 704                         if (q[1].hash >= q[0].hash)
 705                                 continue;
 706                         swap(*(q+1), *q);
 707                         more = 1;
 708                 }
 709         } while(more);
 710 }
 711
 712 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 713 {
 714         struct dx_entry *entries = frame->entries;
 715         struct dx_entry *old = frame->at, *new = old + 1;
 716         int count = dx_get_count(entries);
 717
 718         assert(count < dx_get_limit(entries));
 719         assert(old < entries + count);
 720         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 721         dx_set_hash(new, hash);
 722         dx_set_block(new, block);
 723         dx_set_count(entries, count + 1);
 724 }
 725 #endif
 726
 727
 728 static void ext3_update_dx_flag(struct inode *inode)
 729 {
 730         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 731                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
 732                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 733 }
 734
 735 /*
 736  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 737  *
 738  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
 739  * `de != NULL' is guaranteed by caller.
 740  */
 741 static inline int ext3_match (int len, const char * const name,
 742                               struct ext3_dir_entry_2 * de)
 743 {
 744         if (len != de->name_len)
 745                 return 0;
 746         if (!de->inode)
 747                 return 0;
 748         return !memcmp(name, de->name, len);
 749 }
 750
 751 /*
 752  * Returns 0 if not found, -1 on failure, and 1 on success
 753  */
 754 static inline int search_dirblock(struct buffer_head * bh,
 755                                   struct inode *dir,
 756                                   struct dentry *dentry,
 757                                   unsigned long offset,
 758                                   struct ext3_dir_entry_2 ** res_dir)
 759 {
 760         struct ext3_dir_entry_2 * de;
 761         char * dlimit;
 762         int de_len;
 763         const char *name = dentry->d_name.name;
 764         int namelen = dentry->d_name.len;
 765
 766         de = (struct ext3_dir_entry_2 *) bh->b_data;
 767         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 768         while ((char *) de < dlimit) {
 769                 /* this code is executed quadratically often */
 770                 /* do minimal checking `by hand' */
 771
 772                 if ((char *) de + namelen <= dlimit &&
 773                     ext3_match (namelen, name, de)) {
 774                         /* found a match - just to be sure, do a full check */
 775                         if (!ext3_check_dir_entry("ext3_find_entry",
 776                                                   dir, de, bh, offset))
 777                                 return -1;
 778                         *res_dir = de;
 779                         return 1;
 780                 }
 781                 /* prevent looping on a bad block */
 782                 de_len = le16_to_cpu(de->rec_len);
 783                 if (de_len <= 0)
 784                         return -1;
 785                 offset += de_len;
 786                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
 787         }
 788         return 0;
 789 }
 790
 791
 792 /*
 793  *      ext3_find_entry()
 794  *
 795  * finds an entry in the specified directory with the wanted name. It
 796  * returns the cache buffer in which the entry was found, and the entry
 797  * itself (as a parameter - res_dir). It does NOT read the inode of the
 798  * entry - you'll have to do that yourself if you want to.
 799  *
 800  * The returned buffer_head has ->b_count elevated.  The caller is expected
 801  * to brelse() it when appropriate.
 802  */
 803 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 804                                         struct ext3_dir_entry_2 ** res_dir)
 805 {
 806         struct super_block * sb;
 807         struct buffer_head * bh_use[NAMEI_RA_SIZE];
 808         struct buffer_head * bh, *ret = NULL;
 809         unsigned long start, block, b;
 810         int ra_max = 0;         /* Number of bh's in the readahead
 811                                    buffer, bh_use[] */
 812         int ra_ptr = 0;         /* Current index into readahead
 813                                    buffer */
 814         int num = 0;
 815         int nblocks, i, err;
 816         struct inode *dir = dentry->d_parent->d_inode;
 817         int namelen;
 818         const u8 *name;
 819         unsigned blocksize;
 820
 821         *res_dir = NULL;
 822         sb = dir->i_sb;
 823         blocksize = sb->s_blocksize;
 824         namelen = dentry->d_name.len;
 825         name = dentry->d_name.name;
 826         if (namelen > EXT3_NAME_LEN)
 827                 return NULL;
 828 #ifdef CONFIG_EXT3_INDEX
 829         if (is_dx(dir)) {
 830                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
 831                 /*
 832                  * On success, or if the error was file not found,
 833                  * return.  Otherwise, fall back to doing a search the
 834                  * old fashioned way.
 835                  */
 836                 if (bh || (err != ERR_BAD_DX_DIR))
 837                         return bh;
 838                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
 839         }
 840 #endif
 841         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 842         start = EXT3_I(dir)->i_dir_start_lookup;
 843         if (start >= nblocks)
 844                 start = 0;
 845         block = start;
 846 restart:
 847         do {
 848                 /*
 849                  * We deal with the read-ahead logic here.
 850                  */
 851                 if (ra_ptr >= ra_max) {
 852                         /* Refill the readahead buffer */
 853                         ra_ptr = 0;
 854                         b = block;
 855                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 856                                 /*
 857                                  * Terminate if we reach the end of the
 858                                  * directory and must wrap, or if our
 859                                  * search has finished at this block.
 860                                  */
 861                                 if (b >= nblocks || (num && block == start)) {
 862                                         bh_use[ra_max] = NULL;
 863                                         break;
 864                                 }
 865                                 num++;
 866                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
 867                                 bh_use[ra_max] = bh;
 868                                 if (bh)
 869                                         ll_rw_block(READ, 1, &bh);
 870                         }
 871                 }
 872                 if ((bh = bh_use[ra_ptr++]) == NULL)
 873                         goto next;
 874                 wait_on_buffer(bh);
 875                 if (!buffer_uptodate(bh)) {
 876                         /* read error, skip block & hope for the best */
 877                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
 878                                    "offset %lu\n", dir->i_ino, block);
 879                         brelse(bh);
 880                         goto next;
 881                 }
 882                 i = search_dirblock(bh, dir, dentry,
 883                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
 884                 if (i == 1) {
 885                         EXT3_I(dir)->i_dir_start_lookup = block;
 886                         ret = bh;
 887                         goto cleanup_and_exit;
 888                 } else {
 889                         brelse(bh);
 890                         if (i < 0)
 891                                 goto cleanup_and_exit;
 892                 }
 893         next:
 894                 if (++block >= nblocks)
 895                         block = 0;
 896         } while (block != start);
 897
 898         /*
 899          * If the directory has grown while we were searching, then
 900          * search the last part of the directory before giving up.
 901          */
 902         block = nblocks;
 903         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 904         if (block < nblocks) {
 905                 start = 0;
 906                 goto restart;
 907         }
 908
 909 cleanup_and_exit:
 910         /* Clean up the read-ahead blocks */
 911         for (; ra_ptr < ra_max; ra_ptr++)
 912                 brelse (bh_use[ra_ptr]);
 913         return ret;
 914 }
 915
 916 #ifdef CONFIG_EXT3_INDEX
 917 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 918                        struct ext3_dir_entry_2 **res_dir, int *err)
 919 {
 920         struct super_block * sb;
 921         struct dx_hash_info     hinfo;
 922         u32 hash;
 923         struct dx_frame frames[2], *frame;
 924         struct ext3_dir_entry_2 *de, *top;
 925         struct buffer_head *bh;
 926         unsigned long block;
 927         int retval;
 928         int namelen = dentry->d_name.len;
 929         const u8 *name = dentry->d_name.name;
 930         struct inode *dir = dentry->d_parent->d_inode;
 931
 932         sb = dir->i_sb;
 933         if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
 934                 return NULL;
 935         hash = hinfo.hash;
 936         do {
 937                 block = dx_get_block(frame->at);
 938                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
 939                         goto errout;
 940                 de = (struct ext3_dir_entry_2 *) bh->b_data;
 941                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
 942                                        EXT3_DIR_REC_LEN(0));
 943                 for (; de < top; de = ext3_next_entry(de))
 944                 if (ext3_match (namelen, name, de)) {
 945                         if (!ext3_check_dir_entry("ext3_find_entry",
 946                                                   dir, de, bh,
 947                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
 948                                           +((char *)de - bh->b_data))) {
 949                                 brelse (bh);
 950                                 goto errout;
 951                         }
 952                         *res_dir = de;
 953                         dx_release (frames);
 954                         return bh;
 955                 }
 956                 brelse (bh);
 957                 /* Check to see if we should continue to search */
 958                 retval = ext3_htree_next_block(dir, hash, frame,
 959                                                frames, NULL);
 960                 if (retval < 0) {
 961                         ext3_warning(sb, __FUNCTION__,
 962                              "error reading index page in directory #%lu",
 963                              dir->i_ino);
 964                         *err = retval;
 965                         goto errout;
 966                 }
 967         } while (retval == 1);
 968
 969         *err = -ENOENT;
 970 errout:
 971         dxtrace(printk("%s not found\n", name));
 972         dx_release (frames);
 973         return NULL;
 974 }
 975 #endif
 976
 977 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
 978 {
 979         struct inode * inode;
 980         struct ext3_dir_entry_2 * de;
 981         struct buffer_head * bh;
 982
 983         if (dentry->d_name.len > EXT3_NAME_LEN)
 984                 return ERR_PTR(-ENAMETOOLONG);
 985
 986         bh = ext3_find_entry(dentry, &de);
 987         inode = NULL;
 988         if (bh) {
 989                 unsigned long ino = le32_to_cpu(de->inode);
 990                 brelse (bh);
 991                 inode = iget(dir->i_sb, ino);
 992
 993                 if (!inode)
 994                         return ERR_PTR(-EACCES);
 995         }
 996         if (inode)
 997                 return d_splice_alias(inode, dentry);
 998         d_add(dentry, inode);
 999         return NULL;
1000 }
1001
1002
1003 struct dentry *ext3_get_parent(struct dentry *child)
1004 {
1005         unsigned long ino;
1006         struct dentry *parent;
1007         struct inode *inode;
1008         struct dentry dotdot;
1009         struct ext3_dir_entry_2 * de;
1010         struct buffer_head *bh;
1011
1012         dotdot.d_name.name = "..";
1013         dotdot.d_name.len = 2;
1014         dotdot.d_parent = child; /* confusing, isn't it! */
1015
1016         bh = ext3_find_entry(&dotdot, &de);
1017         inode = NULL;
1018         if (!bh)
1019                 return ERR_PTR(-ENOENT);
1020         ino = le32_to_cpu(de->inode);
1021         brelse(bh);
1022         inode = iget(child->d_inode->i_sb, ino);
1023
1024         if (!inode)
1025                 return ERR_PTR(-EACCES);
1026
1027         parent = d_alloc_anon(inode);
1028         if (!parent) {
1029                 iput(inode);
1030                 parent = ERR_PTR(-ENOMEM);
1031         }
1032         return parent;
1033 }
1034
1035 #define S_SHIFT 12
1036 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1037         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1038         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1039         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1040         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1041         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1042         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1043         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1044 };
1045
1046 static inline void ext3_set_de_type(struct super_block *sb,
1047                                 struct ext3_dir_entry_2 *de,
1048                                 umode_t mode) {
1049         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1050                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1051 }
1052
1053 #ifdef CONFIG_EXT3_INDEX
1054 static struct ext3_dir_entry_2 *
1055 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1056 {
1057         unsigned rec_len = 0;
1058
1059         while (count--) {
1060                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1061                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1062                 memcpy (to, de, rec_len);
1063                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1064                                 cpu_to_le16(rec_len);
1065                 de->inode = 0;
1066                 map++;
1067                 to += rec_len;
1068         }
1069         return (struct ext3_dir_entry_2 *) (to - rec_len);
1070 }
1071
1072 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1073 {
1074         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1075         unsigned rec_len = 0;
1076
1077         prev = to = de;
1078         while ((char*)de < base + size) {
1079                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1080                                                     le16_to_cpu(de->rec_len));
1081                 if (de->inode && de->name_len) {
1082                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1083                         if (de > to)
1084                                 memmove(to, de, rec_len);
1085                         to->rec_len = cpu_to_le16(rec_len);
1086                         prev = to;
1087                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1088                 }
1089                 de = next;
1090         }
1091         return prev;
1092 }
1093
1094 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1095                         struct buffer_head **bh,struct dx_frame *frame,
1096                         struct dx_hash_info *hinfo, int *error)
1097 {
1098         unsigned blocksize = dir->i_sb->s_blocksize;
1099         unsigned count, continued;
1100         struct buffer_head *bh2;
1101         u32 newblock;
1102         u32 hash2;
1103         struct dx_map_entry *map;
1104         char *data1 = (*bh)->b_data, *data2;
1105         unsigned split;
1106         struct ext3_dir_entry_2 *de = NULL, *de2;
1107         int     err;
1108
1109         bh2 = ext3_append (handle, dir, &newblock, error);
1110         if (!(bh2)) {
1111                 brelse(*bh);
1112                 *bh = NULL;
1113                 goto errout;
1114         }
1115
1116         BUFFER_TRACE(*bh, "get_write_access");
1117         err = ext3_journal_get_write_access(handle, *bh);
1118         if (err) {
1119         journal_error:
1120                 brelse(*bh);
1121                 brelse(bh2);
1122                 *bh = NULL;
1123                 ext3_std_error(dir->i_sb, err);
1124                 goto errout;
1125         }
1126         BUFFER_TRACE(frame->bh, "get_write_access");
1127         err = ext3_journal_get_write_access(handle, frame->bh);
1128         if (err)
1129                 goto journal_error;
1130
1131         data2 = bh2->b_data;
1132
1133         /* create map in the end of data2 block */
1134         map = (struct dx_map_entry *) (data2 + blocksize);
1135         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1136                              blocksize, hinfo, map);
1137         map -= count;
1138         split = count/2; // need to adjust to actual middle
1139         dx_sort_map (map, count);
1140         hash2 = map[split].hash;
1141         continued = hash2 == map[split - 1].hash;
1142         dxtrace(printk("Split block %i at %x, %i/%i\n",
1143                 dx_get_block(frame->at), hash2, split, count-split));
1144
1145         /* Fancy dance to stay within two buffers */
1146         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1147         de = dx_pack_dirents(data1,blocksize);
1148         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1149         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1150         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1151         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1152
1153         /* Which block gets the new entry? */
1154         if (hinfo->hash >= hash2)
1155         {
1156                 swap(*bh, bh2);
1157                 de = de2;
1158         }
1159         dx_insert_block (frame, hash2 + continued, newblock);
1160         err = ext3_journal_dirty_metadata (handle, bh2);
1161         if (err)
1162                 goto journal_error;
1163         err = ext3_journal_dirty_metadata (handle, frame->bh);
1164         if (err)
1165                 goto journal_error;
1166         brelse (bh2);
1167         dxtrace(dx_show_index ("frame", frame->entries));
1168 errout:
1169         return de;
1170 }
1171 #endif
1172
1173
1174 /*
1175  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1176  * it points to a directory entry which is guaranteed to be large
1177  * enough for new directory entry.  If de is NULL, then
1178  * add_dirent_to_buf will attempt search the directory block for
1179  * space.  It will return -ENOSPC if no space is available, and -EIO
1180  * and -EEXIST if directory entry already exists.
1181  *
1182  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1183  * all other cases bh is released.
1184  */
1185 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1186                              struct inode *inode, struct ext3_dir_entry_2 *de,
1187                              struct buffer_head * bh)
1188 {
1189         struct inode    *dir = dentry->d_parent->d_inode;
1190         const char      *name = dentry->d_name.name;
1191         int             namelen = dentry->d_name.len;
1192         unsigned long   offset = 0;
1193         unsigned short  reclen;
1194         int             nlen, rlen, err;
1195         char            *top;
1196
1197         reclen = EXT3_DIR_REC_LEN(namelen);
1198         if (!de) {
1199                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1200                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1201                 while ((char *) de <= top) {
1202                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1203                                                   bh, offset)) {
1204                                 brelse (bh);
1205                                 return -EIO;
1206                         }
1207                         if (ext3_match (namelen, name, de)) {
1208                                 brelse (bh);
1209                                 return -EEXIST;
1210                         }
1211                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1212                         rlen = le16_to_cpu(de->rec_len);
1213                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1214                                 break;
1215                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1216                         offset += rlen;
1217                 }
1218                 if ((char *) de > top)
1219                         return -ENOSPC;
1220         }
1221         BUFFER_TRACE(bh, "get_write_access");
1222         err = ext3_journal_get_write_access(handle, bh);
1223         if (err) {
1224                 ext3_std_error(dir->i_sb, err);
1225                 brelse(bh);
1226                 return err;
1227         }
1228
1229         /* By now the buffer is marked for journaling */
1230         nlen = EXT3_DIR_REC_LEN(de->name_len);
1231         rlen = le16_to_cpu(de->rec_len);
1232         if (de->inode) {
1233                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1234                 de1->rec_len = cpu_to_le16(rlen - nlen);
1235                 de->rec_len = cpu_to_le16(nlen);
1236                 de = de1;
1237         }
1238         de->file_type = EXT3_FT_UNKNOWN;
1239         if (inode) {
1240                 de->inode = cpu_to_le32(inode->i_ino);
1241                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1242         } else
1243                 de->inode = 0;
1244         de->name_len = namelen;
1245         memcpy (de->name, name, namelen);
1246         /*
1247          * XXX shouldn't update any times until successful
1248          * completion of syscall, but too many callers depend
1249          * on this.
1250          *
1251          * XXX similarly, too many callers depend on
1252          * ext3_new_inode() setting the times, but error
1253          * recovery deletes the inode, so the worst that can
1254          * happen is that the times are slightly out of date
1255          * and/or different from the directory change time.
1256          */
1257         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1258         ext3_update_dx_flag(dir);
1259         dir->i_version++;
1260         ext3_mark_inode_dirty(handle, dir);
1261         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1262         err = ext3_journal_dirty_metadata(handle, bh);
1263         if (err)
1264                 ext3_std_error(dir->i_sb, err);
1265         brelse(bh);
1266         return 0;
1267 }
1268
1269 #ifdef CONFIG_EXT3_INDEX
1270 /*
1271  * This converts a one block unindexed directory to a 3 block indexed
1272  * directory, and adds the dentry to the indexed directory.
1273  */
1274 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1275                             struct inode *inode, struct buffer_head *bh)
1276 {
1277         struct inode    *dir = dentry->d_parent->d_inode;
1278         const char      *name = dentry->d_name.name;
1279         int             namelen = dentry->d_name.len;
1280         struct buffer_head *bh2;
1281         struct dx_root  *root;
1282         struct dx_frame frames[2], *frame;
1283         struct dx_entry *entries;
1284         struct ext3_dir_entry_2 *de, *de2;
1285         char            *data1, *top;
1286         unsigned        len;
1287         int             retval;
1288         unsigned        blocksize;
1289         struct dx_hash_info hinfo;
1290         u32             block;
1291         struct fake_dirent *fde;
1292
1293         blocksize =  dir->i_sb->s_blocksize;
1294         dxtrace(printk("Creating index\n"));
1295         retval = ext3_journal_get_write_access(handle, bh);
1296         if (retval) {
1297                 ext3_std_error(dir->i_sb, retval);
1298                 brelse(bh);
1299                 return retval;
1300         }
1301         root = (struct dx_root *) bh->b_data;
1302
1303         bh2 = ext3_append (handle, dir, &block, &retval);
1304         if (!(bh2)) {
1305                 brelse(bh);
1306                 return retval;
1307         }
1308         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1309         data1 = bh2->b_data;
1310
1311         /* The 0th block becomes the root, move the dirents out */
1312         fde = &root->dotdot;
1313         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1314         len = ((char *) root) + blocksize - (char *) de;
1315         memcpy (data1, de, len);
1316         de = (struct ext3_dir_entry_2 *) data1;
1317         top = data1 + len;
1318         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1319                 de = de2;
1320         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1321         /* Initialize the root; the dot dirents already exist */
1322         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1323         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1324         memset (&root->info, 0, sizeof(root->info));
1325         root->info.info_length = sizeof(root->info);
1326         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1327         entries = root->entries;
1328         dx_set_block (entries, 1);
1329         dx_set_count (entries, 1);
1330         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1331
1332         /* Initialize as for dx_probe */
1333         hinfo.hash_version = root->info.hash_version;
1334         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1335         ext3fs_dirhash(name, namelen, &hinfo);
1336         frame = frames;
1337         frame->entries = entries;
1338         frame->at = entries;
1339         frame->bh = bh;
1340         bh = bh2;
1341         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1342         dx_release (frames);
1343         if (!(de))
1344                 return retval;
1345
1346         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1347 }
1348 #endif
1349
1350 /*
1351  *      ext3_add_entry()
1352  *
1353  * adds a file entry to the specified directory, using the same
1354  * semantics as ext3_find_entry(). It returns NULL if it failed.
1355  *
1356  * NOTE!! The inode part of 'de' is left at 0 - which means you
1357  * may not sleep between calling this and putting something into
1358  * the entry, as someone else might have used it while you slept.
1359  */
1360 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1361         struct inode *inode)
1362 {
1363         struct inode *dir = dentry->d_parent->d_inode;
1364         unsigned long offset;
1365         struct buffer_head * bh;
1366         struct ext3_dir_entry_2 *de;
1367         struct super_block * sb;
1368         int     retval;
1369 #ifdef CONFIG_EXT3_INDEX
1370         int     dx_fallback=0;
1371 #endif
1372         unsigned blocksize;
1373         unsigned nlen, rlen;
1374         u32 block, blocks;
1375
1376         sb = dir->i_sb;
1377         blocksize = sb->s_blocksize;
1378         if (!dentry->d_name.len)
1379                 return -EINVAL;
1380 #ifdef CONFIG_EXT3_INDEX
1381         if (is_dx(dir)) {
1382                 retval = ext3_dx_add_entry(handle, dentry, inode);
1383                 if (!retval || (retval != ERR_BAD_DX_DIR))
1384                         return retval;
1385                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1386                 dx_fallback++;
1387                 ext3_mark_inode_dirty(handle, dir);
1388         }
1389 #endif
1390         blocks = dir->i_size >> sb->s_blocksize_bits;
1391         for (block = 0, offset = 0; block < blocks; block++) {
1392                 bh = ext3_bread(handle, dir, block, 0, &retval);
1393                 if(!bh)
1394                         return retval;
1395                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1396                 if (retval != -ENOSPC)
1397                         return retval;
1398
1399 #ifdef CONFIG_EXT3_INDEX
1400                 if (blocks == 1 && !dx_fallback &&
1401                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1402                         return make_indexed_dir(handle, dentry, inode, bh);
1403 #endif
1404                 brelse(bh);
1405         }
1406         bh = ext3_append(handle, dir, &block, &retval);
1407         if (!bh)
1408                 return retval;
1409         de = (struct ext3_dir_entry_2 *) bh->b_data;
1410         de->inode = 0;
1411         de->rec_len = cpu_to_le16(rlen = blocksize);
1412         nlen = 0;
1413         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1414 }
1415
1416 #ifdef CONFIG_EXT3_INDEX
1417 /*
1418  * Returns 0 for success, or a negative error value
1419  */
1420 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1421                              struct inode *inode)
1422 {
1423         struct dx_frame frames[2], *frame;
1424         struct dx_entry *entries, *at;
1425         struct dx_hash_info hinfo;
1426         struct buffer_head * bh;
1427         struct inode *dir = dentry->d_parent->d_inode;
1428         struct super_block * sb = dir->i_sb;
1429         struct ext3_dir_entry_2 *de;
1430         int err;
1431
1432         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1433         if (!frame)
1434                 return err;
1435         entries = frame->entries;
1436         at = frame->at;
1437
1438         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1439                 goto cleanup;
1440
1441         BUFFER_TRACE(bh, "get_write_access");
1442         err = ext3_journal_get_write_access(handle, bh);
1443         if (err)
1444                 goto journal_error;
1445
1446         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1447         if (err != -ENOSPC) {
1448                 bh = NULL;
1449                 goto cleanup;
1450         }
1451
1452         /* Block full, should compress but for now just split */
1453         dxtrace(printk("using %u of %u node entries\n",
1454                        dx_get_count(entries), dx_get_limit(entries)));
1455         /* Need to split index? */
1456         if (dx_get_count(entries) == dx_get_limit(entries)) {
1457                 u32 newblock;
1458                 unsigned icount = dx_get_count(entries);
1459                 int levels = frame - frames;
1460                 struct dx_entry *entries2;
1461                 struct dx_node *node2;
1462                 struct buffer_head *bh2;
1463
1464                 if (levels && (dx_get_count(frames->entries) ==
1465                                dx_get_limit(frames->entries))) {
1466                         ext3_warning(sb, __FUNCTION__,
1467                                      "Directory index full!\n");
1468                         err = -ENOSPC;
1469                         goto cleanup;
1470                 }
1471                 bh2 = ext3_append (handle, dir, &newblock, &err);
1472                 if (!(bh2))
1473                         goto cleanup;
1474                 node2 = (struct dx_node *)(bh2->b_data);
1475                 entries2 = node2->entries;
1476                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1477                 node2->fake.inode = 0;
1478                 BUFFER_TRACE(frame->bh, "get_write_access");
1479                 err = ext3_journal_get_write_access(handle, frame->bh);
1480                 if (err)
1481                         goto journal_error;
1482                 if (levels) {
1483                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1484                         unsigned hash2 = dx_get_hash(entries + icount1);
1485                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1486
1487                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1488                         err = ext3_journal_get_write_access(handle,
1489                                                              frames[0].bh);
1490                         if (err)
1491                                 goto journal_error;
1492
1493                         memcpy ((char *) entries2, (char *) (entries + icount1),
1494                                 icount2 * sizeof(struct dx_entry));
1495                         dx_set_count (entries, icount1);
1496                         dx_set_count (entries2, icount2);
1497                         dx_set_limit (entries2, dx_node_limit(dir));
1498
1499                         /* Which index block gets the new entry? */
1500                         if (at - entries >= icount1) {
1501                                 frame->at = at = at - entries - icount1 + entries2;
1502                                 frame->entries = entries = entries2;
1503                                 swap(frame->bh, bh2);
1504                         }
1505                         dx_insert_block (frames + 0, hash2, newblock);
1506                         dxtrace(dx_show_index ("node", frames[1].entries));
1507                         dxtrace(dx_show_index ("node",
1508                                ((struct dx_node *) bh2->b_data)->entries));
1509                         err = ext3_journal_dirty_metadata(handle, bh2);
1510                         if (err)
1511                                 goto journal_error;
1512                         brelse (bh2);
1513                 } else {
1514                         dxtrace(printk("Creating second level index...\n"));
1515                         memcpy((char *) entries2, (char *) entries,
1516                                icount * sizeof(struct dx_entry));
1517                         dx_set_limit(entries2, dx_node_limit(dir));
1518
1519                         /* Set up root */
1520                         dx_set_count(entries, 1);
1521                         dx_set_block(entries + 0, newblock);
1522                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1523
1524                         /* Add new access path frame */
1525                         frame = frames + 1;
1526                         frame->at = at = at - entries + entries2;
1527                         frame->entries = entries = entries2;
1528                         frame->bh = bh2;
1529                         err = ext3_journal_get_write_access(handle,
1530                                                              frame->bh);
1531                         if (err)
1532                                 goto journal_error;
1533                 }
1534                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1535         }
1536         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1537         if (!de)
1538                 goto cleanup;
1539         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1540         bh = NULL;
1541         goto cleanup;
1542
1543 journal_error:
1544         ext3_std_error(dir->i_sb, err);
1545 cleanup:
1546         if (bh)
1547                 brelse(bh);
1548         dx_release(frames);
1549         return err;
1550 }
1551 #endif
1552
1553 /*
1554  * ext3_delete_entry deletes a directory entry by merging it with the
1555  * previous entry
1556  */
1557 static int ext3_delete_entry (handle_t *handle,
1558                               struct inode * dir,
1559                               struct ext3_dir_entry_2 * de_del,
1560                               struct buffer_head * bh)
1561 {
1562         struct ext3_dir_entry_2 * de, * pde;
1563         int i;
1564
1565         i = 0;
1566         pde = NULL;
1567         de = (struct ext3_dir_entry_2 *) bh->b_data;
1568         while (i < bh->b_size) {
1569                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1570                         return -EIO;
1571                 if (de == de_del)  {
1572                         BUFFER_TRACE(bh, "get_write_access");
1573                         ext3_journal_get_write_access(handle, bh);
1574                         if (pde)
1575                                 pde->rec_len =
1576                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1577                                                     le16_to_cpu(de->rec_len));
1578                         else
1579                                 de->inode = 0;
1580                         dir->i_version++;
1581                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1582                         ext3_journal_dirty_metadata(handle, bh);
1583                         return 0;
1584                 }
1585                 i += le16_to_cpu(de->rec_len);
1586                 pde = de;
1587                 de = (struct ext3_dir_entry_2 *)
1588                         ((char *) de + le16_to_cpu(de->rec_len));
1589         }
1590         return -ENOENT;
1591 }
1592
1593 /*
1594  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1595  * do not perform it in these functions.  We perform it at the call site,
1596  * if it is needed.
1597  */
1598 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1599 {
1600         inode->i_nlink++;
1601 }
1602
1603 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1604 {
1605         inode->i_nlink--;
1606 }
1607
1608 static int ext3_add_nondir(handle_t *handle,
1609                 struct dentry *dentry, struct inode *inode)
1610 {
1611         int err = ext3_add_entry(handle, dentry, inode);
1612         if (!err) {
1613                 ext3_mark_inode_dirty(handle, inode);
1614                 d_instantiate(dentry, inode);
1615                 return 0;
1616         }
1617         ext3_dec_count(handle, inode);
1618         iput(inode);
1619         return err;
1620 }
1621
1622 /*
1623  * By the time this is called, we already have created
1624  * the directory cache entry for the new file, but it
1625  * is so far negative - it has no inode.
1626  *
1627  * If the create succeeds, we fill in the inode information
1628  * with d_instantiate().
1629  */
1630 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1631                 struct nameidata *nd)
1632 {
1633         handle_t *handle;
1634         struct inode * inode;
1635         int err, retries = 0;
1636
1637 retry:
1638         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1639                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1640                                         2*EXT3_QUOTA_INIT_BLOCKS);
1641         if (IS_ERR(handle))
1642                 return PTR_ERR(handle);
1643
1644         if (IS_DIRSYNC(dir))
1645                 handle->h_sync = 1;
1646
1647         inode = ext3_new_inode (handle, dir, mode);
1648         err = PTR_ERR(inode);
1649         if (!IS_ERR(inode)) {
1650                 inode->i_op = &ext3_file_inode_operations;
1651                 inode->i_fop = &ext3_file_operations;
1652                 ext3_set_aops(inode);
1653                 err = ext3_add_nondir(handle, dentry, inode);
1654         }
1655         ext3_journal_stop(handle);
1656         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1657                 goto retry;
1658         return err;
1659 }
1660
1661 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1662                         int mode, dev_t rdev)
1663 {
1664         handle_t *handle;
1665         struct inode *inode;
1666         int err, retries = 0;
1667
1668         if (!new_valid_dev(rdev))
1669                 return -EINVAL;
1670
1671 retry:
1672         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1673                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1674                                         2*EXT3_QUOTA_INIT_BLOCKS);
1675         if (IS_ERR(handle))
1676                 return PTR_ERR(handle);
1677
1678         if (IS_DIRSYNC(dir))
1679                 handle->h_sync = 1;
1680
1681         inode = ext3_new_inode (handle, dir, mode);
1682         err = PTR_ERR(inode);
1683         if (!IS_ERR(inode)) {
1684                 init_special_inode(inode, inode->i_mode, rdev);
1685 #ifdef CONFIG_EXT3_FS_XATTR
1686                 inode->i_op = &ext3_special_inode_operations;
1687 #endif
1688                 err = ext3_add_nondir(handle, dentry, inode);
1689         }
1690         ext3_journal_stop(handle);
1691         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1692                 goto retry;
1693         return err;
1694 }
1695
1696 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1697 {
1698         handle_t *handle;
1699         struct inode * inode;
1700         struct buffer_head * dir_block;
1701         struct ext3_dir_entry_2 * de;
1702         int err, retries = 0;
1703
1704         if (dir->i_nlink >= EXT3_LINK_MAX)
1705                 return -EMLINK;
1706
1707 retry:
1708         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1709                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1710                                         2*EXT3_QUOTA_INIT_BLOCKS);
1711         if (IS_ERR(handle))
1712                 return PTR_ERR(handle);
1713
1714         if (IS_DIRSYNC(dir))
1715                 handle->h_sync = 1;
1716
1717         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1718         err = PTR_ERR(inode);
1719         if (IS_ERR(inode))
1720                 goto out_stop;
1721
1722         inode->i_op = &ext3_dir_inode_operations;
1723         inode->i_fop = &ext3_dir_operations;
1724         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1725         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1726         if (!dir_block) {
1727                 inode->i_nlink--; /* is this nlink == 0? */
1728                 ext3_mark_inode_dirty(handle, inode);
1729                 iput (inode);
1730                 goto out_stop;
1731         }
1732         BUFFER_TRACE(dir_block, "get_write_access");
1733         ext3_journal_get_write_access(handle, dir_block);
1734         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1735         de->inode = cpu_to_le32(inode->i_ino);
1736         de->name_len = 1;
1737         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1738         strcpy (de->name, ".");
1739         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1740         de = (struct ext3_dir_entry_2 *)
1741                         ((char *) de + le16_to_cpu(de->rec_len));
1742         de->inode = cpu_to_le32(dir->i_ino);
1743         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1744         de->name_len = 2;
1745         strcpy (de->name, "..");
1746         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1747         inode->i_nlink = 2;
1748         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1749         ext3_journal_dirty_metadata(handle, dir_block);
1750         brelse (dir_block);
1751         ext3_mark_inode_dirty(handle, inode);
1752         err = ext3_add_entry (handle, dentry, inode);
1753         if (err) {
1754                 inode->i_nlink = 0;
1755                 ext3_mark_inode_dirty(handle, inode);
1756                 iput (inode);
1757                 goto out_stop;
1758         }
1759         dir->i_nlink++;
1760         ext3_update_dx_flag(dir);
1761         ext3_mark_inode_dirty(handle, dir);
1762         d_instantiate(dentry, inode);
1763 out_stop:
1764         ext3_journal_stop(handle);
1765         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1766                 goto retry;
1767         return err;
1768 }
1769
1770 /*
1771  * routine to check that the specified directory is empty (for rmdir)
1772  */
1773 static int empty_dir (struct inode * inode)
1774 {
1775         unsigned long offset;
1776         struct buffer_head * bh;
1777         struct ext3_dir_entry_2 * de, * de1;
1778         struct super_block * sb;
1779         int err = 0;
1780
1781         sb = inode->i_sb;
1782         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1783             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1784                 if (err)
1785                         ext3_error(inode->i_sb, __FUNCTION__,
1786                                    "error %d reading directory #%lu offset 0",
1787                                    err, inode->i_ino);
1788                 else
1789                         ext3_warning(inode->i_sb, __FUNCTION__,
1790                                      "bad directory (dir #%lu) - no data block",
1791                                      inode->i_ino);
1792                 return 1;
1793         }
1794         de = (struct ext3_dir_entry_2 *) bh->b_data;
1795         de1 = (struct ext3_dir_entry_2 *)
1796                         ((char *) de + le16_to_cpu(de->rec_len));
1797         if (le32_to_cpu(de->inode) != inode->i_ino ||
1798                         !le32_to_cpu(de1->inode) ||
1799                         strcmp (".", de->name) ||
1800                         strcmp ("..", de1->name)) {
1801                 ext3_warning (inode->i_sb, "empty_dir",
1802                               "bad directory (dir #%lu) - no `.' or `..'",
1803                               inode->i_ino);
1804                 brelse (bh);
1805                 return 1;
1806         }
1807         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1808         de = (struct ext3_dir_entry_2 *)
1809                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1810         while (offset < inode->i_size ) {
1811                 if (!bh ||
1812                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1813                         err = 0;
1814                         brelse (bh);
1815                         bh = ext3_bread (NULL, inode,
1816                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1817                         if (!bh) {
1818                                 if (err)
1819                                         ext3_error(sb, __FUNCTION__,
1820                                                    "error %d reading directory"
1821                                                    " #%lu offset %lu",
1822                                                    err, inode->i_ino, offset);
1823                                 offset += sb->s_blocksize;
1824                                 continue;
1825                         }
1826                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1827                 }
1828                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1829                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1830                                                          sb->s_blocksize);
1831                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1832                         continue;
1833                 }
1834                 if (le32_to_cpu(de->inode)) {
1835                         brelse (bh);
1836                         return 0;
1837                 }
1838                 offset += le16_to_cpu(de->rec_len);
1839                 de = (struct ext3_dir_entry_2 *)
1840                                 ((char *) de + le16_to_cpu(de->rec_len));
1841         }
1842         brelse (bh);
1843         return 1;
1844 }
1845
1846 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1847  * such inodes, starting at the superblock, in case we crash before the
1848  * file is closed/deleted, or in case the inode truncate spans multiple
1849  * transactions and the last transaction is not recovered after a crash.
1850  *
1851  * At filesystem recovery time, we walk this list deleting unlinked
1852  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1853  */
1854 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1855 {
1856         struct super_block *sb = inode->i_sb;
1857         struct ext3_iloc iloc;
1858         int err = 0, rc;
1859
1860         lock_super(sb);
1861         if (!list_empty(&EXT3_I(inode)->i_orphan))
1862                 goto out_unlock;
1863
1864         /* Orphan handling is only valid for files with data blocks
1865          * being truncated, or files being unlinked. */
1866
1867         /* @@@ FIXME: Observation from aviro:
1868          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1869          * here (on lock_super()), so race with ext3_link() which might bump
1870          * ->i_nlink. For, say it, character device. Not a regular file,
1871          * not a directory, not a symlink and ->i_nlink > 0.
1872          */
1873         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1874                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1875
1876         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1877         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1878         if (err)
1879                 goto out_unlock;
1880
1881         err = ext3_reserve_inode_write(handle, inode, &iloc);
1882         if (err)
1883                 goto out_unlock;
1884
1885         /* Insert this inode at the head of the on-disk orphan list... */
1886         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1887         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1888         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1889         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1890         if (!err)
1891                 err = rc;
1892
1893         /* Only add to the head of the in-memory list if all the
1894          * previous operations succeeded.  If the orphan_add is going to
1895          * fail (possibly taking the journal offline), we can't risk
1896          * leaving the inode on the orphan list: stray orphan-list
1897          * entries can cause panics at unmount time.
1898          *
1899          * This is safe: on error we're going to ignore the orphan list
1900          * anyway on the next recovery. */
1901         if (!err)
1902                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1903
1904         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1905         jbd_debug(4, "orphan inode %ld will point to %d\n",
1906                         inode->i_ino, NEXT_ORPHAN(inode));
1907 out_unlock:
1908         unlock_super(sb);
1909         ext3_std_error(inode->i_sb, err);
1910         return err;
1911 }
1912
1913 /*
1914  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1915  * of such inodes stored on disk, because it is finally being cleaned up.
1916  */
1917 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1918 {
1919         struct list_head *prev;
1920         struct ext3_inode_info *ei = EXT3_I(inode);
1921         struct ext3_sb_info *sbi;
1922         unsigned long ino_next;
1923         struct ext3_iloc iloc;
1924         int err = 0;
1925
1926         lock_super(inode->i_sb);
1927         if (list_empty(&ei->i_orphan)) {
1928                 unlock_super(inode->i_sb);
1929                 return 0;
1930         }
1931
1932         ino_next = NEXT_ORPHAN(inode);
1933         prev = ei->i_orphan.prev;
1934         sbi = EXT3_SB(inode->i_sb);
1935
1936         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1937
1938         list_del_init(&ei->i_orphan);
1939
1940         /* If we're on an error path, we may not have a valid
1941          * transaction handle with which to update the orphan list on
1942          * disk, but we still need to remove the inode from the linked
1943          * list in memory. */
1944         if (!handle)
1945                 goto out;
1946
1947         err = ext3_reserve_inode_write(handle, inode, &iloc);
1948         if (err)
1949                 goto out_err;
1950
1951         if (prev == &sbi->s_orphan) {
1952                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1953                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1954                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1955                 if (err)
1956                         goto out_brelse;
1957                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1958                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1959         } else {
1960                 struct ext3_iloc iloc2;
1961                 struct inode *i_prev =
1962                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1963
1964                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1965                           i_prev->i_ino, ino_next);
1966                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1967                 if (err)
1968                         goto out_brelse;
1969                 NEXT_ORPHAN(i_prev) = ino_next;
1970                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1971         }
1972         if (err)
1973                 goto out_brelse;
1974         NEXT_ORPHAN(inode) = 0;
1975         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1976
1977 out_err:
1978         ext3_std_error(inode->i_sb, err);
1979 out:
1980         unlock_super(inode->i_sb);
1981         return err;
1982
1983 out_brelse:
1984         brelse(iloc.bh);
1985         goto out_err;
1986 }
1987
1988 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1989 {
1990         int retval;
1991         struct inode * inode;
1992         struct buffer_head * bh;
1993         struct ext3_dir_entry_2 * de;
1994         handle_t *handle;
1995
1996         /* Initialize quotas before so that eventual writes go in
1997          * separate transaction */
1998         DQUOT_INIT(dentry->d_inode);
1999         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2000         if (IS_ERR(handle))
2001                 return PTR_ERR(handle);
2002
2003         retval = -ENOENT;
2004         bh = ext3_find_entry (dentry, &de);
2005         if (!bh)
2006                 goto end_rmdir;
2007
2008         if (IS_DIRSYNC(dir))
2009                 handle->h_sync = 1;
2010
2011         inode = dentry->d_inode;
2012
2013         retval = -EIO;
2014         if (le32_to_cpu(de->inode) != inode->i_ino)
2015                 goto end_rmdir;
2016
2017         retval = -ENOTEMPTY;
2018         if (!empty_dir (inode))
2019                 goto end_rmdir;
2020
2021         retval = ext3_delete_entry(handle, dir, de, bh);
2022         if (retval)
2023                 goto end_rmdir;
2024         if (inode->i_nlink != 2)
2025                 ext3_warning (inode->i_sb, "ext3_rmdir",
2026                               "empty directory has nlink!=2 (%d)",
2027                               inode->i_nlink);
2028         inode->i_version++;
2029         inode->i_nlink = 0;
2030         /* There's no need to set i_disksize: the fact that i_nlink is
2031          * zero will ensure that the right thing happens during any
2032          * recovery. */
2033         inode->i_size = 0;
2034         ext3_orphan_add(handle, inode);
2035         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2036         ext3_mark_inode_dirty(handle, inode);
2037         dir->i_nlink--;
2038         ext3_update_dx_flag(dir);
2039         ext3_mark_inode_dirty(handle, dir);
2040
2041 end_rmdir:
2042         ext3_journal_stop(handle);
2043         brelse (bh);
2044         return retval;
2045 }
2046
2047 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2048 {
2049         int retval;
2050         struct inode * inode;
2051         struct buffer_head * bh;
2052         struct ext3_dir_entry_2 * de;
2053         handle_t *handle;
2054
2055         /* Initialize quotas before so that eventual writes go
2056          * in separate transaction */
2057         DQUOT_INIT(dentry->d_inode);
2058         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2059         if (IS_ERR(handle))
2060                 return PTR_ERR(handle);
2061
2062         if (IS_DIRSYNC(dir))
2063                 handle->h_sync = 1;
2064
2065         retval = -ENOENT;
2066         bh = ext3_find_entry (dentry, &de);
2067         if (!bh)
2068                 goto end_unlink;
2069
2070         inode = dentry->d_inode;
2071
2072         retval = -EIO;
2073         if (le32_to_cpu(de->inode) != inode->i_ino)
2074                 goto end_unlink;
2075
2076         if (!inode->i_nlink) {
2077                 ext3_warning (inode->i_sb, "ext3_unlink",
2078                               "Deleting nonexistent file (%lu), %d",
2079                               inode->i_ino, inode->i_nlink);
2080                 inode->i_nlink = 1;
2081         }
2082         retval = ext3_delete_entry(handle, dir, de, bh);
2083         if (retval)
2084                 goto end_unlink;
2085         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2086         ext3_update_dx_flag(dir);
2087         ext3_mark_inode_dirty(handle, dir);
2088         inode->i_nlink--;
2089         if (!inode->i_nlink)
2090                 ext3_orphan_add(handle, inode);
2091         inode->i_ctime = dir->i_ctime;
2092         ext3_mark_inode_dirty(handle, inode);
2093         retval = 0;
2094
2095 end_unlink:
2096         ext3_journal_stop(handle);
2097         brelse (bh);
2098         return retval;
2099 }
2100
2101 static int ext3_symlink (struct inode * dir,
2102                 struct dentry *dentry, const char * symname)
2103 {
2104         handle_t *handle;
2105         struct inode * inode;
2106         int l, err, retries = 0;
2107
2108         l = strlen(symname)+1;
2109         if (l > dir->i_sb->s_blocksize)
2110                 return -ENAMETOOLONG;
2111
2112 retry:
2113         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2114                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2115                                         2*EXT3_QUOTA_INIT_BLOCKS);
2116         if (IS_ERR(handle))
2117                 return PTR_ERR(handle);
2118
2119         if (IS_DIRSYNC(dir))
2120                 handle->h_sync = 1;
2121
2122         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2123         err = PTR_ERR(inode);
2124         if (IS_ERR(inode))
2125                 goto out_stop;
2126
2127         if (l > sizeof (EXT3_I(inode)->i_data)) {
2128                 inode->i_op = &ext3_symlink_inode_operations;
2129                 ext3_set_aops(inode);
2130                 /*
2131                  * page_symlink() calls into ext3_prepare/commit_write.
2132                  * We have a transaction open.  All is sweetness.  It also sets
2133                  * i_size in generic_commit_write().
2134                  */
2135                 err = page_symlink(inode, symname, l);
2136                 if (err) {
2137                         ext3_dec_count(handle, inode);
2138                         ext3_mark_inode_dirty(handle, inode);
2139                         iput (inode);
2140                         goto out_stop;
2141                 }
2142         } else {
2143                 inode->i_op = &ext3_fast_symlink_inode_operations;
2144                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2145                 inode->i_size = l-1;
2146         }
2147         EXT3_I(inode)->i_disksize = inode->i_size;
2148         err = ext3_add_nondir(handle, dentry, inode);
2149 out_stop:
2150         ext3_journal_stop(handle);
2151         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2152                 goto retry;
2153         return err;
2154 }
2155
2156 static int ext3_link (struct dentry * old_dentry,
2157                 struct inode * dir, struct dentry *dentry)
2158 {
2159         handle_t *handle;
2160         struct inode *inode = old_dentry->d_inode;
2161         int err, retries = 0;
2162
2163         if (inode->i_nlink >= EXT3_LINK_MAX)
2164                 return -EMLINK;
2165
2166 retry:
2167         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2168                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2169         if (IS_ERR(handle))
2170                 return PTR_ERR(handle);
2171
2172         if (IS_DIRSYNC(dir))
2173                 handle->h_sync = 1;
2174
2175         inode->i_ctime = CURRENT_TIME;
2176         ext3_inc_count(handle, inode);
2177         atomic_inc(&inode->i_count);
2178
2179         err = ext3_add_nondir(handle, dentry, inode);
2180         ext3_journal_stop(handle);
2181         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2182                 goto retry;
2183         return err;
2184 }
2185
2186 #define PARENT_INO(buffer) \
2187         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2188         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2189
2190 /*
2191  * Anybody can rename anything with this: the permission checks are left to the
2192  * higher-level routines.
2193  */
2194 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2195                            struct inode * new_dir,struct dentry *new_dentry)
2196 {
2197         handle_t *handle;
2198         struct inode * old_inode, * new_inode;
2199         struct buffer_head * old_bh, * new_bh, * dir_bh;
2200         struct ext3_dir_entry_2 * old_de, * new_de;
2201         int retval;
2202
2203         old_bh = new_bh = dir_bh = NULL;
2204
2205         /* Initialize quotas before so that eventual writes go
2206          * in separate transaction */
2207         if (new_dentry->d_inode)
2208                 DQUOT_INIT(new_dentry->d_inode);
2209         handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2210                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2211         if (IS_ERR(handle))
2212                 return PTR_ERR(handle);
2213
2214         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2215                 handle->h_sync = 1;
2216
2217         old_bh = ext3_find_entry (old_dentry, &old_de);
2218         /*
2219          *  Check for inode number is _not_ due to possible IO errors.
2220          *  We might rmdir the source, keep it as pwd of some process
2221          *  and merrily kill the link to whatever was created under the
2222          *  same name. Goodbye sticky bit ;-<
2223          */
2224         old_inode = old_dentry->d_inode;
2225         retval = -ENOENT;
2226         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2227                 goto end_rename;
2228
2229         new_inode = new_dentry->d_inode;
2230         new_bh = ext3_find_entry (new_dentry, &new_de);
2231         if (new_bh) {
2232                 if (!new_inode) {
2233                         brelse (new_bh);
2234                         new_bh = NULL;
2235                 }
2236         }
2237         if (S_ISDIR(old_inode->i_mode)) {
2238                 if (new_inode) {
2239                         retval = -ENOTEMPTY;
2240                         if (!empty_dir (new_inode))
2241                                 goto end_rename;
2242                 }
2243                 retval = -EIO;
2244                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2245                 if (!dir_bh)
2246                         goto end_rename;
2247                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2248                         goto end_rename;
2249                 retval = -EMLINK;
2250                 if (!new_inode && new_dir!=old_dir &&
2251                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2252                         goto end_rename;
2253         }
2254         if (!new_bh) {
2255                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2256                 if (retval)
2257                         goto end_rename;
2258         } else {
2259                 BUFFER_TRACE(new_bh, "get write access");
2260                 ext3_journal_get_write_access(handle, new_bh);
2261                 new_de->inode = le32_to_cpu(old_inode->i_ino);
2262                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2263                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2264                         new_de->file_type = old_de->file_type;
2265                 new_dir->i_version++;
2266                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2267                 ext3_journal_dirty_metadata(handle, new_bh);
2268                 brelse(new_bh);
2269                 new_bh = NULL;
2270         }
2271
2272         /*
2273          * Like most other Unix systems, set the ctime for inodes on a
2274          * rename.
2275          */
2276         old_inode->i_ctime = CURRENT_TIME;
2277         ext3_mark_inode_dirty(handle, old_inode);
2278
2279         /*
2280          * ok, that's it
2281          */
2282         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2283             old_de->name_len != old_dentry->d_name.len ||
2284             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2285             (retval = ext3_delete_entry(handle, old_dir,
2286                                         old_de, old_bh)) == -ENOENT) {
2287                 /* old_de could have moved from under us during htree split, so
2288                  * make sure that we are deleting the right entry.  We might
2289                  * also be pointing to a stale entry in the unused part of
2290                  * old_bh so just checking inum and the name isn't enough. */
2291                 struct buffer_head *old_bh2;
2292                 struct ext3_dir_entry_2 *old_de2;
2293
2294                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2295                 if (old_bh2) {
2296                         retval = ext3_delete_entry(handle, old_dir,
2297                                                    old_de2, old_bh2);
2298                         brelse(old_bh2);
2299                 }
2300         }
2301         if (retval) {
2302                 ext3_warning(old_dir->i_sb, "ext3_rename",
2303                                 "Deleting old file (%lu), %d, error=%d",
2304                                 old_dir->i_ino, old_dir->i_nlink, retval);
2305         }
2306
2307         if (new_inode) {
2308                 new_inode->i_nlink--;
2309                 new_inode->i_ctime = CURRENT_TIME;
2310         }
2311         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
2312         ext3_update_dx_flag(old_dir);
2313         if (dir_bh) {
2314                 BUFFER_TRACE(dir_bh, "get_write_access");
2315                 ext3_journal_get_write_access(handle, dir_bh);
2316                 PARENT_INO(dir_bh->b_data) = le32_to_cpu(new_dir->i_ino);
2317                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2318                 ext3_journal_dirty_metadata(handle, dir_bh);
2319                 old_dir->i_nlink--;
2320                 if (new_inode) {
2321                         new_inode->i_nlink--;
2322                 } else {
2323                         new_dir->i_nlink++;
2324                         ext3_update_dx_flag(new_dir);
2325                         ext3_mark_inode_dirty(handle, new_dir);
2326                 }
2327         }
2328         ext3_mark_inode_dirty(handle, old_dir);
2329         if (new_inode) {
2330                 ext3_mark_inode_dirty(handle, new_inode);
2331                 if (!new_inode->i_nlink)
2332                         ext3_orphan_add(handle, new_inode);
2333         }
2334         retval = 0;
2335
2336 end_rename:
2337         brelse (dir_bh);
2338         brelse (old_bh);
2339         brelse (new_bh);
2340         ext3_journal_stop(handle);
2341         return retval;
2342 }
2343
2344 /*
2345  * directories can handle most operations...
2346  */
2347 struct inode_operations ext3_dir_inode_operations = {
2348         .create         = ext3_create,
2349         .lookup         = ext3_lookup,
2350         .link           = ext3_link,
2351         .unlink         = ext3_unlink,
2352         .symlink        = ext3_symlink,
2353         .mkdir          = ext3_mkdir,
2354         .rmdir          = ext3_rmdir,
2355         .mknod          = ext3_mknod,
2356         .rename         = ext3_rename,
2357         .setattr        = ext3_setattr,
2358         .setxattr       = ext3_setxattr,
2359         .getxattr       = ext3_getxattr,
2360         .listxattr      = ext3_listxattr,
2361         .removexattr    = ext3_removexattr,
2362         .permission     = ext3_permission,
2363 };
2364
2365 struct inode_operations ext3_special_inode_operations = {
2366         .setattr        = ext3_setattr,
2367         .setxattr       = ext3_setxattr,
2368         .getxattr       = ext3_getxattr,
2369         .listxattr      = ext3_listxattr,
2370         .removexattr    = ext3_removexattr,
2371         .permission     = ext3_permission,
2372 };