fs/ext4/namei.c

   1 /*
   2  *  linux/fs/ext4/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/jbd2.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include "ext4.h"
  38 #include "ext4_jbd2.h"
  39
  40 #include "xattr.h"
  41 #include "acl.h"
  42
  43 #include <trace/events/ext4.h>
  44 /*
  45  * define how far ahead to read directories while searching them.
  46  */
  47 #define NAMEI_RA_CHUNKS  2
  48 #define NAMEI_RA_BLOCKS  4
  49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52 static struct buffer_head *ext4_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         ext4_lblk_t *block, int *err)
  55 {
  56         struct buffer_head *bh;
  57
  58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  59
  60         bh = ext4_bread(handle, inode, *block, 1, err);
  61         if (bh) {
  62                 inode->i_size += inode->i_sb->s_blocksize;
  63                 EXT4_I(inode)->i_disksize = inode->i_size;
  64                 *err = ext4_journal_get_write_access(handle, bh);
  65                 if (*err) {
  66                         brelse(bh);
  67                         bh = NULL;
  68                 }
  69         }
  70         return bh;
  71 }
  72
  73 #ifndef assert
  74 #define assert(test) J_ASSERT(test)
  75 #endif
  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         __le32 inode;
  86         __le16 rec_len;
  87         u8 name_len;
  88         u8 file_type;
  89 };
  90
  91 struct dx_countlimit
  92 {
  93         __le16 limit;
  94         __le16 count;
  95 };
  96
  97 struct dx_entry
  98 {
  99         __le32 hash;
 100         __le32 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                 __le32 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         u16 offs;
 145         u16 size;
 146 };
 147
 148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t 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(const struct qstr *d_name,
 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 ext4_dir_entry_2 *de, unsigned blocksize,
 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 ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 168                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 170 static void dx_insert_block(struct dx_frame *frame,
 171                                         u32 hash, ext4_lblk_t block);
 172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 173                                  struct dx_frame *frame,
 174                                  struct dx_frame *frames,
 175                                  __u32 *start_hash);
 176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 177                 const struct qstr *d_name,
 178                 struct ext4_dir_entry_2 **res_dir,
 179                 int *err);
 180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 181                              struct inode *inode);
 182
 183 /*
 184  * p is at least 6 bytes before the end of page
 185  */
 186 static inline struct ext4_dir_entry_2 *
 187 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 188 {
 189         return (struct ext4_dir_entry_2 *)((char *)p +
 190                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 191 }
 192
 193 /*
 194  * Future: use high four bits of block for coalesce-on-delete flags
 195  * Mask them off for now.
 196  */
 197
 198 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 199 {
 200         return le32_to_cpu(entry->block) & 0x00ffffff;
 201 }
 202
 203 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 204 {
 205         entry->block = cpu_to_le32(value);
 206 }
 207
 208 static inline unsigned dx_get_hash(struct dx_entry *entry)
 209 {
 210         return le32_to_cpu(entry->hash);
 211 }
 212
 213 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 214 {
 215         entry->hash = cpu_to_le32(value);
 216 }
 217
 218 static inline unsigned dx_get_count(struct dx_entry *entries)
 219 {
 220         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 221 }
 222
 223 static inline unsigned dx_get_limit(struct dx_entry *entries)
 224 {
 225         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 226 }
 227
 228 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 229 {
 230         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 231 }
 232
 233 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 234 {
 235         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 236 }
 237
 238 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 239 {
 240         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 241                 EXT4_DIR_REC_LEN(2) - infosize;
 242         return entry_space / sizeof(struct dx_entry);
 243 }
 244
 245 static inline unsigned dx_node_limit(struct inode *dir)
 246 {
 247         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 248         return entry_space / sizeof(struct dx_entry);
 249 }
 250
 251 /*
 252  * Debug
 253  */
 254 #ifdef DX_DEBUG
 255 static void dx_show_index(char * label, struct dx_entry *entries)
 256 {
 257         int i, n = dx_get_count (entries);
 258         printk(KERN_DEBUG "%s index ", label);
 259         for (i = 0; i < n; i++) {
 260                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 261                                 0, (unsigned long)dx_get_block(entries + i));
 262         }
 263         printk("\n");
 264 }
 265
 266 struct stats
 267 {
 268         unsigned names;
 269         unsigned space;
 270         unsigned bcount;
 271 };
 272
 273 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 274                                  int size, int show_names)
 275 {
 276         unsigned names = 0, space = 0;
 277         char *base = (char *) de;
 278         struct dx_hash_info h = *hinfo;
 279
 280         printk("names: ");
 281         while ((char *) de < base + size)
 282         {
 283                 if (de->inode)
 284                 {
 285                         if (show_names)
 286                         {
 287                                 int len = de->name_len;
 288                                 char *name = de->name;
 289                                 while (len--) printk("%c", *name++);
 290                                 ext4fs_dirhash(de->name, de->name_len, &h);
 291                                 printk(":%x.%u ", h.hash,
 292                                        ((char *) de - base));
 293                         }
 294                         space += EXT4_DIR_REC_LEN(de->name_len);
 295                         names++;
 296                 }
 297                 de = ext4_next_entry(de, size);
 298         }
 299         printk("(%i)\n", names);
 300         return (struct stats) { names, space, 1 };
 301 }
 302
 303 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 304                              struct dx_entry *entries, int levels)
 305 {
 306         unsigned blocksize = dir->i_sb->s_blocksize;
 307         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 308         unsigned bcount = 0;
 309         struct buffer_head *bh;
 310         int err;
 311         printk("%i indexed blocks...\n", count);
 312         for (i = 0; i < count; i++, entries++)
 313         {
 314                 ext4_lblk_t block = dx_get_block(entries);
 315                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 316                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 317                 struct stats stats;
 318                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 319                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 320                 stats = levels?
 321                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 322                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 323                 names += stats.names;
 324                 space += stats.space;
 325                 bcount += stats.bcount;
 326                 brelse(bh);
 327         }
 328         if (bcount)
 329                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 330                        levels ? "" : "   ", names, space/bcount,
 331                        (space/bcount)*100/blocksize);
 332         return (struct stats) { names, space, bcount};
 333 }
 334 #endif /* DX_DEBUG */
 335
 336 /*
 337  * Probe for a directory leaf block to search.
 338  *
 339  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 340  * error in the directory index, and the caller should fall back to
 341  * searching the directory normally.  The callers of dx_probe **MUST**
 342  * check for this error code, and make sure it never gets reflected
 343  * back to userspace.
 344  */
 345 static struct dx_frame *
 346 dx_probe(const struct qstr *d_name, struct inode *dir,
 347          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 348 {
 349         unsigned count, indirect;
 350         struct dx_entry *at, *entries, *p, *q, *m;
 351         struct dx_root *root;
 352         struct buffer_head *bh;
 353         struct dx_frame *frame = frame_in;
 354         u32 hash;
 355
 356         frame->bh = NULL;
 357         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 358                 goto fail;
 359         root = (struct dx_root *) bh->b_data;
 360         if (root->info.hash_version != DX_HASH_TEA &&
 361             root->info.hash_version != DX_HASH_HALF_MD4 &&
 362             root->info.hash_version != DX_HASH_LEGACY) {
 363                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 364                              root->info.hash_version);
 365                 brelse(bh);
 366                 *err = ERR_BAD_DX_DIR;
 367                 goto fail;
 368         }
 369         hinfo->hash_version = root->info.hash_version;
 370         if (hinfo->hash_version <= DX_HASH_TEA)
 371                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 372         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 373         if (d_name)
 374                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 375         hash = hinfo->hash;
 376
 377         if (root->info.unused_flags & 1) {
 378                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 379                              root->info.unused_flags);
 380                 brelse(bh);
 381                 *err = ERR_BAD_DX_DIR;
 382                 goto fail;
 383         }
 384
 385         if ((indirect = root->info.indirect_levels) > 1) {
 386                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 387                              root->info.indirect_levels);
 388                 brelse(bh);
 389                 *err = ERR_BAD_DX_DIR;
 390                 goto fail;
 391         }
 392
 393         entries = (struct dx_entry *) (((char *)&root->info) +
 394                                        root->info.info_length);
 395
 396         if (dx_get_limit(entries) != dx_root_limit(dir,
 397                                                    root->info.info_length)) {
 398                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 399                 brelse(bh);
 400                 *err = ERR_BAD_DX_DIR;
 401                 goto fail;
 402         }
 403
 404         dxtrace(printk("Look up %x", hash));
 405         while (1)
 406         {
 407                 count = dx_get_count(entries);
 408                 if (!count || count > dx_get_limit(entries)) {
 409                         ext4_warning(dir->i_sb,
 410                                      "dx entry: no count or count > limit");
 411                         brelse(bh);
 412                         *err = ERR_BAD_DX_DIR;
 413                         goto fail2;
 414                 }
 415
 416                 p = entries + 1;
 417                 q = entries + count - 1;
 418                 while (p <= q)
 419                 {
 420                         m = p + (q - p)/2;
 421                         dxtrace(printk("."));
 422                         if (dx_get_hash(m) > hash)
 423                                 q = m - 1;
 424                         else
 425                                 p = m + 1;
 426                 }
 427
 428                 if (0) // linear search cross check
 429                 {
 430                         unsigned n = count - 1;
 431                         at = entries;
 432                         while (n--)
 433                         {
 434                                 dxtrace(printk(","));
 435                                 if (dx_get_hash(++at) > hash)
 436                                 {
 437                                         at--;
 438                                         break;
 439                                 }
 440                         }
 441                         assert (at == p - 1);
 442                 }
 443
 444                 at = p - 1;
 445                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 446                 frame->bh = bh;
 447                 frame->entries = entries;
 448                 frame->at = at;
 449                 if (!indirect--) return frame;
 450                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 451                         goto fail2;
 452                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 453                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 454                         ext4_warning(dir->i_sb,
 455                                      "dx entry: limit != node limit");
 456                         brelse(bh);
 457                         *err = ERR_BAD_DX_DIR;
 458                         goto fail2;
 459                 }
 460                 frame++;
 461                 frame->bh = NULL;
 462         }
 463 fail2:
 464         while (frame >= frame_in) {
 465                 brelse(frame->bh);
 466                 frame--;
 467         }
 468 fail:
 469         if (*err == ERR_BAD_DX_DIR)
 470                 ext4_warning(dir->i_sb,
 471                              "Corrupt dir inode %ld, running e2fsck is "
 472                              "recommended.", dir->i_ino);
 473         return NULL;
 474 }
 475
 476 static void dx_release (struct dx_frame *frames)
 477 {
 478         if (frames[0].bh == NULL)
 479                 return;
 480
 481         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 482                 brelse(frames[1].bh);
 483         brelse(frames[0].bh);
 484 }
 485
 486 /*
 487  * This function increments the frame pointer to search the next leaf
 488  * block, and reads in the necessary intervening nodes if the search
 489  * should be necessary.  Whether or not the search is necessary is
 490  * controlled by the hash parameter.  If the hash value is even, then
 491  * the search is only continued if the next block starts with that
 492  * hash value.  This is used if we are searching for a specific file.
 493  *
 494  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 495  *
 496  * This function returns 1 if the caller should continue to search,
 497  * or 0 if it should not.  If there is an error reading one of the
 498  * index blocks, it will a negative error code.
 499  *
 500  * If start_hash is non-null, it will be filled in with the starting
 501  * hash of the next page.
 502  */
 503 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 504                                  struct dx_frame *frame,
 505                                  struct dx_frame *frames,
 506                                  __u32 *start_hash)
 507 {
 508         struct dx_frame *p;
 509         struct buffer_head *bh;
 510         int err, num_frames = 0;
 511         __u32 bhash;
 512
 513         p = frame;
 514         /*
 515          * Find the next leaf page by incrementing the frame pointer.
 516          * If we run out of entries in the interior node, loop around and
 517          * increment pointer in the parent node.  When we break out of
 518          * this loop, num_frames indicates the number of interior
 519          * nodes need to be read.
 520          */
 521         while (1) {
 522                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 523                         break;
 524                 if (p == frames)
 525                         return 0;
 526                 num_frames++;
 527                 p--;
 528         }
 529
 530         /*
 531          * If the hash is 1, then continue only if the next page has a
 532          * continuation hash of any value.  This is used for readdir
 533          * handling.  Otherwise, check to see if the hash matches the
 534          * desired contiuation hash.  If it doesn't, return since
 535          * there's no point to read in the successive index pages.
 536          */
 537         bhash = dx_get_hash(p->at);
 538         if (start_hash)
 539                 *start_hash = bhash;
 540         if ((hash & 1) == 0) {
 541                 if ((bhash & ~1) != hash)
 542                         return 0;
 543         }
 544         /*
 545          * If the hash is HASH_NB_ALWAYS, we always go to the next
 546          * block so no check is necessary
 547          */
 548         while (num_frames--) {
 549                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 550                                       0, &err)))
 551                         return err; /* Failure */
 552                 p++;
 553                 brelse(p->bh);
 554                 p->bh = bh;
 555                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 556         }
 557         return 1;
 558 }
 559
 560
 561 /*
 562  * This function fills a red-black tree with information from a
 563  * directory block.  It returns the number directory entries loaded
 564  * into the tree.  If there is an error it is returned in err.
 565  */
 566 static int htree_dirblock_to_tree(struct file *dir_file,
 567                                   struct inode *dir, ext4_lblk_t block,
 568                                   struct dx_hash_info *hinfo,
 569                                   __u32 start_hash, __u32 start_minor_hash)
 570 {
 571         struct buffer_head *bh;
 572         struct ext4_dir_entry_2 *de, *top;
 573         int err, count = 0;
 574
 575         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 576                                                         (unsigned long)block));
 577         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 578                 return err;
 579
 580         de = (struct ext4_dir_entry_2 *) bh->b_data;
 581         top = (struct ext4_dir_entry_2 *) ((char *) de +
 582                                            dir->i_sb->s_blocksize -
 583                                            EXT4_DIR_REC_LEN(0));
 584         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 585                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 586                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 587                                          + ((char *)de - bh->b_data))) {
 588                         /* On error, skip the f_pos to the next block. */
 589                         dir_file->f_pos = (dir_file->f_pos |
 590                                         (dir->i_sb->s_blocksize - 1)) + 1;
 591                         brelse(bh);
 592                         return count;
 593                 }
 594                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 595                 if ((hinfo->hash < start_hash) ||
 596                     ((hinfo->hash == start_hash) &&
 597                      (hinfo->minor_hash < start_minor_hash)))
 598                         continue;
 599                 if (de->inode == 0)
 600                         continue;
 601                 if ((err = ext4_htree_store_dirent(dir_file,
 602                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 603                         brelse(bh);
 604                         return err;
 605                 }
 606                 count++;
 607         }
 608         brelse(bh);
 609         return count;
 610 }
 611
 612
 613 /*
 614  * This function fills a red-black tree with information from a
 615  * directory.  We start scanning the directory in hash order, starting
 616  * at start_hash and start_minor_hash.
 617  *
 618  * This function returns the number of entries inserted into the tree,
 619  * or a negative error code.
 620  */
 621 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 622                          __u32 start_minor_hash, __u32 *next_hash)
 623 {
 624         struct dx_hash_info hinfo;
 625         struct ext4_dir_entry_2 *de;
 626         struct dx_frame frames[2], *frame;
 627         struct inode *dir;
 628         ext4_lblk_t block;
 629         int count = 0;
 630         int ret, err;
 631         __u32 hashval;
 632
 633         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 634                        start_hash, start_minor_hash));
 635         dir = dir_file->f_path.dentry->d_inode;
 636         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 637                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 638                 if (hinfo.hash_version <= DX_HASH_TEA)
 639                         hinfo.hash_version +=
 640                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 641                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 642                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 643                                                start_hash, start_minor_hash);
 644                 *next_hash = ~0;
 645                 return count;
 646         }
 647         hinfo.hash = start_hash;
 648         hinfo.minor_hash = 0;
 649         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 650         if (!frame)
 651                 return err;
 652
 653         /* Add '.' and '..' from the htree header */
 654         if (!start_hash && !start_minor_hash) {
 655                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 656                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 657                         goto errout;
 658                 count++;
 659         }
 660         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 661                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 662                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 663                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 664                         goto errout;
 665                 count++;
 666         }
 667
 668         while (1) {
 669                 block = dx_get_block(frame->at);
 670                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 671                                              start_hash, start_minor_hash);
 672                 if (ret < 0) {
 673                         err = ret;
 674                         goto errout;
 675                 }
 676                 count += ret;
 677                 hashval = ~0;
 678                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 679                                             frame, frames, &hashval);
 680                 *next_hash = hashval;
 681                 if (ret < 0) {
 682                         err = ret;
 683                         goto errout;
 684                 }
 685                 /*
 686                  * Stop if:  (a) there are no more entries, or
 687                  * (b) we have inserted at least one entry and the
 688                  * next hash value is not a continuation
 689                  */
 690                 if ((ret == 0) ||
 691                     (count && ((hashval & 1) == 0)))
 692                         break;
 693         }
 694         dx_release(frames);
 695         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 696                        "next hash: %x\n", count, *next_hash));
 697         return count;
 698 errout:
 699         dx_release(frames);
 700         return (err);
 701 }
 702
 703
 704 /*
 705  * Directory block splitting, compacting
 706  */
 707
 708 /*
 709  * Create map of hash values, offsets, and sizes, stored at end of block.
 710  * Returns number of entries mapped.
 711  */
 712 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 713                        struct dx_hash_info *hinfo,
 714                        struct dx_map_entry *map_tail)
 715 {
 716         int count = 0;
 717         char *base = (char *) de;
 718         struct dx_hash_info h = *hinfo;
 719
 720         while ((char *) de < base + blocksize) {
 721                 if (de->name_len && de->inode) {
 722                         ext4fs_dirhash(de->name, de->name_len, &h);
 723                         map_tail--;
 724                         map_tail->hash = h.hash;
 725                         map_tail->offs = ((char *) de - base)>>2;
 726                         map_tail->size = le16_to_cpu(de->rec_len);
 727                         count++;
 728                         cond_resched();
 729                 }
 730                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 731                 de = ext4_next_entry(de, blocksize);
 732         }
 733         return count;
 734 }
 735
 736 /* Sort map by hash value */
 737 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 738 {
 739         struct dx_map_entry *p, *q, *top = map + count - 1;
 740         int more;
 741         /* Combsort until bubble sort doesn't suck */
 742         while (count > 2) {
 743                 count = count*10/13;
 744                 if (count - 9 < 2) /* 9, 10 -> 11 */
 745                         count = 11;
 746                 for (p = top, q = p - count; q >= map; p--, q--)
 747                         if (p->hash < q->hash)
 748                                 swap(*p, *q);
 749         }
 750         /* Garden variety bubble sort */
 751         do {
 752                 more = 0;
 753                 q = top;
 754                 while (q-- > map) {
 755                         if (q[1].hash >= q[0].hash)
 756                                 continue;
 757                         swap(*(q+1), *q);
 758                         more = 1;
 759                 }
 760         } while(more);
 761 }
 762
 763 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
 764 {
 765         struct dx_entry *entries = frame->entries;
 766         struct dx_entry *old = frame->at, *new = old + 1;
 767         int count = dx_get_count(entries);
 768
 769         assert(count < dx_get_limit(entries));
 770         assert(old < entries + count);
 771         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 772         dx_set_hash(new, hash);
 773         dx_set_block(new, block);
 774         dx_set_count(entries, count + 1);
 775 }
 776
 777 static void ext4_update_dx_flag(struct inode *inode)
 778 {
 779         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 780                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 781                 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
 782 }
 783
 784 /*
 785  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 786  *
 787  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 788  * `de != NULL' is guaranteed by caller.
 789  */
 790 static inline int ext4_match (int len, const char * const name,
 791                               struct ext4_dir_entry_2 * de)
 792 {
 793         if (len != de->name_len)
 794                 return 0;
 795         if (!de->inode)
 796                 return 0;
 797         return !memcmp(name, de->name, len);
 798 }
 799
 800 /*
 801  * Returns 0 if not found, -1 on failure, and 1 on success
 802  */
 803 static inline int search_dirblock(struct buffer_head *bh,
 804                                   struct inode *dir,
 805                                   const struct qstr *d_name,
 806                                   unsigned int offset,
 807                                   struct ext4_dir_entry_2 ** res_dir)
 808 {
 809         struct ext4_dir_entry_2 * de;
 810         char * dlimit;
 811         int de_len;
 812         const char *name = d_name->name;
 813         int namelen = d_name->len;
 814
 815         de = (struct ext4_dir_entry_2 *) bh->b_data;
 816         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 817         while ((char *) de < dlimit) {
 818                 /* this code is executed quadratically often */
 819                 /* do minimal checking `by hand' */
 820
 821                 if ((char *) de + namelen <= dlimit &&
 822                     ext4_match (namelen, name, de)) {
 823                         /* found a match - just to be sure, do a full check */
 824                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
 825                                 return -1;
 826                         *res_dir = de;
 827                         return 1;
 828                 }
 829                 /* prevent looping on a bad block */
 830                 de_len = ext4_rec_len_from_disk(de->rec_len,
 831                                                 dir->i_sb->s_blocksize);
 832                 if (de_len <= 0)
 833                         return -1;
 834                 offset += de_len;
 835                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 836         }
 837         return 0;
 838 }
 839
 840
 841 /*
 842  *      ext4_find_entry()
 843  *
 844  * finds an entry in the specified directory with the wanted name. It
 845  * returns the cache buffer in which the entry was found, and the entry
 846  * itself (as a parameter - res_dir). It does NOT read the inode of the
 847  * entry - you'll have to do that yourself if you want to.
 848  *
 849  * The returned buffer_head has ->b_count elevated.  The caller is expected
 850  * to brelse() it when appropriate.
 851  */
 852 static struct buffer_head * ext4_find_entry (struct inode *dir,
 853                                         const struct qstr *d_name,
 854                                         struct ext4_dir_entry_2 ** res_dir)
 855 {
 856         struct super_block *sb;
 857         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 858         struct buffer_head *bh, *ret = NULL;
 859         ext4_lblk_t start, block, b;
 860         const u8 *name = d_name->name;
 861         int ra_max = 0;         /* Number of bh's in the readahead
 862                                    buffer, bh_use[] */
 863         int ra_ptr = 0;         /* Current index into readahead
 864                                    buffer */
 865         int num = 0;
 866         ext4_lblk_t  nblocks;
 867         int i, err;
 868         int namelen;
 869
 870         *res_dir = NULL;
 871         sb = dir->i_sb;
 872         namelen = d_name->len;
 873         if (namelen > EXT4_NAME_LEN)
 874                 return NULL;
 875         if ((namelen <= 2) && (name[0] == '.') &&
 876             (name[1] == '.' || name[1] == '\0')) {
 877                 /*
 878                  * "." or ".." will only be in the first block
 879                  * NFS may look up ".."; "." should be handled by the VFS
 880                  */
 881                 block = start = 0;
 882                 nblocks = 1;
 883                 goto restart;
 884         }
 885         if (is_dx(dir)) {
 886                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
 887                 /*
 888                  * On success, or if the error was file not found,
 889                  * return.  Otherwise, fall back to doing a search the
 890                  * old fashioned way.
 891                  */
 892                 if (bh || (err != ERR_BAD_DX_DIR))
 893                         return bh;
 894                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
 895                                "falling back\n"));
 896         }
 897         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 898         start = EXT4_I(dir)->i_dir_start_lookup;
 899         if (start >= nblocks)
 900                 start = 0;
 901         block = start;
 902 restart:
 903         do {
 904                 /*
 905                  * We deal with the read-ahead logic here.
 906                  */
 907                 if (ra_ptr >= ra_max) {
 908                         /* Refill the readahead buffer */
 909                         ra_ptr = 0;
 910                         b = block;
 911                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 912                                 /*
 913                                  * Terminate if we reach the end of the
 914                                  * directory and must wrap, or if our
 915                                  * search has finished at this block.
 916                                  */
 917                                 if (b >= nblocks || (num && block == start)) {
 918                                         bh_use[ra_max] = NULL;
 919                                         break;
 920                                 }
 921                                 num++;
 922                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
 923                                 bh_use[ra_max] = bh;
 924                                 if (bh)
 925                                         ll_rw_block(READ_META, 1, &bh);
 926                         }
 927                 }
 928                 if ((bh = bh_use[ra_ptr++]) == NULL)
 929                         goto next;
 930                 wait_on_buffer(bh);
 931                 if (!buffer_uptodate(bh)) {
 932                         /* read error, skip block & hope for the best */
 933                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
 934                                          (unsigned long) block);
 935                         brelse(bh);
 936                         goto next;
 937                 }
 938                 i = search_dirblock(bh, dir, d_name,
 939                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
 940                 if (i == 1) {
 941                         EXT4_I(dir)->i_dir_start_lookup = block;
 942                         ret = bh;
 943                         goto cleanup_and_exit;
 944                 } else {
 945                         brelse(bh);
 946                         if (i < 0)
 947                                 goto cleanup_and_exit;
 948                 }
 949         next:
 950                 if (++block >= nblocks)
 951                         block = 0;
 952         } while (block != start);
 953
 954         /*
 955          * If the directory has grown while we were searching, then
 956          * search the last part of the directory before giving up.
 957          */
 958         block = nblocks;
 959         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 960         if (block < nblocks) {
 961                 start = 0;
 962                 goto restart;
 963         }
 964
 965 cleanup_and_exit:
 966         /* Clean up the read-ahead blocks */
 967         for (; ra_ptr < ra_max; ra_ptr++)
 968                 brelse(bh_use[ra_ptr]);
 969         return ret;
 970 }
 971
 972 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
 973                        struct ext4_dir_entry_2 **res_dir, int *err)
 974 {
 975         struct super_block * sb = dir->i_sb;
 976         struct dx_hash_info     hinfo;
 977         struct dx_frame frames[2], *frame;
 978         struct buffer_head *bh;
 979         ext4_lblk_t block;
 980         int retval;
 981
 982         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
 983                 return NULL;
 984         do {
 985                 block = dx_get_block(frame->at);
 986                 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
 987                         goto errout;
 988
 989                 retval = search_dirblock(bh, dir, d_name,
 990                                          block << EXT4_BLOCK_SIZE_BITS(sb),
 991                                          res_dir);
 992                 if (retval == 1) {      /* Success! */
 993                         dx_release(frames);
 994                         return bh;
 995                 }
 996                 brelse(bh);
 997                 if (retval == -1) {
 998                         *err = ERR_BAD_DX_DIR;
 999                         goto errout;
1000                 }
1001
1002                 /* Check to see if we should continue to search */
1003                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1004                                                frames, NULL);
1005                 if (retval < 0) {
1006                         ext4_warning(sb,
1007                              "error reading index page in directory #%lu",
1008                              dir->i_ino);
1009                         *err = retval;
1010                         goto errout;
1011                 }
1012         } while (retval == 1);
1013
1014         *err = -ENOENT;
1015 errout:
1016         dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1017         dx_release (frames);
1018         return NULL;
1019 }
1020
1021 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1022 {
1023         struct inode *inode;
1024         struct ext4_dir_entry_2 *de;
1025         struct buffer_head *bh;
1026
1027         if (dentry->d_name.len > EXT4_NAME_LEN)
1028                 return ERR_PTR(-ENAMETOOLONG);
1029
1030         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1031         inode = NULL;
1032         if (bh) {
1033                 __u32 ino = le32_to_cpu(de->inode);
1034                 brelse(bh);
1035                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1036                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1037                         return ERR_PTR(-EIO);
1038                 }
1039                 inode = ext4_iget(dir->i_sb, ino);
1040                 if (IS_ERR(inode)) {
1041                         if (PTR_ERR(inode) == -ESTALE) {
1042                                 EXT4_ERROR_INODE(dir,
1043                                                  "deleted inode referenced: %u",
1044                                                  ino);
1045                                 return ERR_PTR(-EIO);
1046                         } else {
1047                                 return ERR_CAST(inode);
1048                         }
1049                 }
1050         }
1051         return d_splice_alias(inode, dentry);
1052 }
1053
1054
1055 struct dentry *ext4_get_parent(struct dentry *child)
1056 {
1057         __u32 ino;
1058         static const struct qstr dotdot = {
1059                 .name = "..",
1060                 .len = 2,
1061         };
1062         struct ext4_dir_entry_2 * de;
1063         struct buffer_head *bh;
1064
1065         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1066         if (!bh)
1067                 return ERR_PTR(-ENOENT);
1068         ino = le32_to_cpu(de->inode);
1069         brelse(bh);
1070
1071         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1072                 EXT4_ERROR_INODE(child->d_inode,
1073                                  "bad parent inode number: %u", ino);
1074                 return ERR_PTR(-EIO);
1075         }
1076
1077         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1078 }
1079
1080 #define S_SHIFT 12
1081 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1082         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1083         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1084         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1085         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1086         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1087         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1088         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1089 };
1090
1091 static inline void ext4_set_de_type(struct super_block *sb,
1092                                 struct ext4_dir_entry_2 *de,
1093                                 umode_t mode) {
1094         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1095                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1096 }
1097
1098 /*
1099  * Move count entries from end of map between two memory locations.
1100  * Returns pointer to last entry moved.
1101  */
1102 static struct ext4_dir_entry_2 *
1103 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1104                 unsigned blocksize)
1105 {
1106         unsigned rec_len = 0;
1107
1108         while (count--) {
1109                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1110                                                 (from + (map->offs<<2));
1111                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1112                 memcpy (to, de, rec_len);
1113                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1114                                 ext4_rec_len_to_disk(rec_len, blocksize);
1115                 de->inode = 0;
1116                 map++;
1117                 to += rec_len;
1118         }
1119         return (struct ext4_dir_entry_2 *) (to - rec_len);
1120 }
1121
1122 /*
1123  * Compact each dir entry in the range to the minimal rec_len.
1124  * Returns pointer to last entry in range.
1125  */
1126 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1127 {
1128         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1129         unsigned rec_len = 0;
1130
1131         prev = to = de;
1132         while ((char*)de < base + blocksize) {
1133                 next = ext4_next_entry(de, blocksize);
1134                 if (de->inode && de->name_len) {
1135                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1136                         if (de > to)
1137                                 memmove(to, de, rec_len);
1138                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1139                         prev = to;
1140                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1141                 }
1142                 de = next;
1143         }
1144         return prev;
1145 }
1146
1147 /*
1148  * Split a full leaf block to make room for a new dir entry.
1149  * Allocate a new block, and move entries so that they are approx. equally full.
1150  * Returns pointer to de in block into which the new entry will be inserted.
1151  */
1152 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1153                         struct buffer_head **bh,struct dx_frame *frame,
1154                         struct dx_hash_info *hinfo, int *error)
1155 {
1156         unsigned blocksize = dir->i_sb->s_blocksize;
1157         unsigned count, continued;
1158         struct buffer_head *bh2;
1159         ext4_lblk_t newblock;
1160         u32 hash2;
1161         struct dx_map_entry *map;
1162         char *data1 = (*bh)->b_data, *data2;
1163         unsigned split, move, size;
1164         struct ext4_dir_entry_2 *de = NULL, *de2;
1165         int     err = 0, i;
1166
1167         bh2 = ext4_append (handle, dir, &newblock, &err);
1168         if (!(bh2)) {
1169                 brelse(*bh);
1170                 *bh = NULL;
1171                 goto errout;
1172         }
1173
1174         BUFFER_TRACE(*bh, "get_write_access");
1175         err = ext4_journal_get_write_access(handle, *bh);
1176         if (err)
1177                 goto journal_error;
1178
1179         BUFFER_TRACE(frame->bh, "get_write_access");
1180         err = ext4_journal_get_write_access(handle, frame->bh);
1181         if (err)
1182                 goto journal_error;
1183
1184         data2 = bh2->b_data;
1185
1186         /* create map in the end of data2 block */
1187         map = (struct dx_map_entry *) (data2 + blocksize);
1188         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1189                              blocksize, hinfo, map);
1190         map -= count;
1191         dx_sort_map(map, count);
1192         /* Split the existing block in the middle, size-wise */
1193         size = 0;
1194         move = 0;
1195         for (i = count-1; i >= 0; i--) {
1196                 /* is more than half of this entry in 2nd half of the block? */
1197                 if (size + map[i].size/2 > blocksize/2)
1198                         break;
1199                 size += map[i].size;
1200                 move++;
1201         }
1202         /* map index at which we will split */
1203         split = count - move;
1204         hash2 = map[split].hash;
1205         continued = hash2 == map[split - 1].hash;
1206         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1207                         (unsigned long)dx_get_block(frame->at),
1208                                         hash2, split, count-split));
1209
1210         /* Fancy dance to stay within two buffers */
1211         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1212         de = dx_pack_dirents(data1, blocksize);
1213         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1214                                            blocksize);
1215         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1216                                             blocksize);
1217         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1218         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1219
1220         /* Which block gets the new entry? */
1221         if (hinfo->hash >= hash2)
1222         {
1223                 swap(*bh, bh2);
1224                 de = de2;
1225         }
1226         dx_insert_block(frame, hash2 + continued, newblock);
1227         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1228         if (err)
1229                 goto journal_error;
1230         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1231         if (err)
1232                 goto journal_error;
1233         brelse(bh2);
1234         dxtrace(dx_show_index("frame", frame->entries));
1235         return de;
1236
1237 journal_error:
1238         brelse(*bh);
1239         brelse(bh2);
1240         *bh = NULL;
1241         ext4_std_error(dir->i_sb, err);
1242 errout:
1243         *error = err;
1244         return NULL;
1245 }
1246
1247 /*
1248  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1249  * it points to a directory entry which is guaranteed to be large
1250  * enough for new directory entry.  If de is NULL, then
1251  * add_dirent_to_buf will attempt search the directory block for
1252  * space.  It will return -ENOSPC if no space is available, and -EIO
1253  * and -EEXIST if directory entry already exists.
1254  */
1255 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1256                              struct inode *inode, struct ext4_dir_entry_2 *de,
1257                              struct buffer_head *bh)
1258 {
1259         struct inode    *dir = dentry->d_parent->d_inode;
1260         const char      *name = dentry->d_name.name;
1261         int             namelen = dentry->d_name.len;
1262         unsigned int    offset = 0;
1263         unsigned int    blocksize = dir->i_sb->s_blocksize;
1264         unsigned short  reclen;
1265         int             nlen, rlen, err;
1266         char            *top;
1267
1268         reclen = EXT4_DIR_REC_LEN(namelen);
1269         if (!de) {
1270                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1271                 top = bh->b_data + blocksize - reclen;
1272                 while ((char *) de <= top) {
1273                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1274                                 return -EIO;
1275                         if (ext4_match(namelen, name, de))
1276                                 return -EEXIST;
1277                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1278                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1279                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1280                                 break;
1281                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1282                         offset += rlen;
1283                 }
1284                 if ((char *) de > top)
1285                         return -ENOSPC;
1286         }
1287         BUFFER_TRACE(bh, "get_write_access");
1288         err = ext4_journal_get_write_access(handle, bh);
1289         if (err) {
1290                 ext4_std_error(dir->i_sb, err);
1291                 return err;
1292         }
1293
1294         /* By now the buffer is marked for journaling */
1295         nlen = EXT4_DIR_REC_LEN(de->name_len);
1296         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1297         if (de->inode) {
1298                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1299                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1300                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1301                 de = de1;
1302         }
1303         de->file_type = EXT4_FT_UNKNOWN;
1304         if (inode) {
1305                 de->inode = cpu_to_le32(inode->i_ino);
1306                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1307         } else
1308                 de->inode = 0;
1309         de->name_len = namelen;
1310         memcpy(de->name, name, namelen);
1311         /*
1312          * XXX shouldn't update any times until successful
1313          * completion of syscall, but too many callers depend
1314          * on this.
1315          *
1316          * XXX similarly, too many callers depend on
1317          * ext4_new_inode() setting the times, but error
1318          * recovery deletes the inode, so the worst that can
1319          * happen is that the times are slightly out of date
1320          * and/or different from the directory change time.
1321          */
1322         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1323         ext4_update_dx_flag(dir);
1324         dir->i_version++;
1325         ext4_mark_inode_dirty(handle, dir);
1326         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1327         err = ext4_handle_dirty_metadata(handle, dir, bh);
1328         if (err)
1329                 ext4_std_error(dir->i_sb, err);
1330         return 0;
1331 }
1332
1333 /*
1334  * This converts a one block unindexed directory to a 3 block indexed
1335  * directory, and adds the dentry to the indexed directory.
1336  */
1337 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1338                             struct inode *inode, struct buffer_head *bh)
1339 {
1340         struct inode    *dir = dentry->d_parent->d_inode;
1341         const char      *name = dentry->d_name.name;
1342         int             namelen = dentry->d_name.len;
1343         struct buffer_head *bh2;
1344         struct dx_root  *root;
1345         struct dx_frame frames[2], *frame;
1346         struct dx_entry *entries;
1347         struct ext4_dir_entry_2 *de, *de2;
1348         char            *data1, *top;
1349         unsigned        len;
1350         int             retval;
1351         unsigned        blocksize;
1352         struct dx_hash_info hinfo;
1353         ext4_lblk_t  block;
1354         struct fake_dirent *fde;
1355
1356         blocksize =  dir->i_sb->s_blocksize;
1357         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1358         retval = ext4_journal_get_write_access(handle, bh);
1359         if (retval) {
1360                 ext4_std_error(dir->i_sb, retval);
1361                 brelse(bh);
1362                 return retval;
1363         }
1364         root = (struct dx_root *) bh->b_data;
1365
1366         /* The 0th block becomes the root, move the dirents out */
1367         fde = &root->dotdot;
1368         de = (struct ext4_dir_entry_2 *)((char *)fde +
1369                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1370         if ((char *) de >= (((char *) root) + blocksize)) {
1371                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1372                 brelse(bh);
1373                 return -EIO;
1374         }
1375         len = ((char *) root) + blocksize - (char *) de;
1376
1377         /* Allocate new block for the 0th block's dirents */
1378         bh2 = ext4_append(handle, dir, &block, &retval);
1379         if (!(bh2)) {
1380                 brelse(bh);
1381                 return retval;
1382         }
1383         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1384         data1 = bh2->b_data;
1385
1386         memcpy (data1, de, len);
1387         de = (struct ext4_dir_entry_2 *) data1;
1388         top = data1 + len;
1389         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1390                 de = de2;
1391         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1392                                            blocksize);
1393         /* Initialize the root; the dot dirents already exist */
1394         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1395         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1396                                            blocksize);
1397         memset (&root->info, 0, sizeof(root->info));
1398         root->info.info_length = sizeof(root->info);
1399         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1400         entries = root->entries;
1401         dx_set_block(entries, 1);
1402         dx_set_count(entries, 1);
1403         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1404
1405         /* Initialize as for dx_probe */
1406         hinfo.hash_version = root->info.hash_version;
1407         if (hinfo.hash_version <= DX_HASH_TEA)
1408                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1409         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1410         ext4fs_dirhash(name, namelen, &hinfo);
1411         frame = frames;
1412         frame->entries = entries;
1413         frame->at = entries;
1414         frame->bh = bh;
1415         bh = bh2;
1416         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1417         if (!de) {
1418                 /*
1419                  * Even if the block split failed, we have to properly write
1420                  * out all the changes we did so far. Otherwise we can end up
1421                  * with corrupted filesystem.
1422                  */
1423                 ext4_mark_inode_dirty(handle, dir);
1424                 ext4_handle_dirty_metadata(handle, dir, frame->bh);
1425                 ext4_handle_dirty_metadata(handle, dir, bh);
1426                 dx_release(frames);
1427                 return retval;
1428         }
1429         dx_release(frames);
1430
1431         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1432         brelse(bh);
1433         return retval;
1434 }
1435
1436 /*
1437  *      ext4_add_entry()
1438  *
1439  * adds a file entry to the specified directory, using the same
1440  * semantics as ext4_find_entry(). It returns NULL if it failed.
1441  *
1442  * NOTE!! The inode part of 'de' is left at 0 - which means you
1443  * may not sleep between calling this and putting something into
1444  * the entry, as someone else might have used it while you slept.
1445  */
1446 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1447                           struct inode *inode)
1448 {
1449         struct inode *dir = dentry->d_parent->d_inode;
1450         struct buffer_head *bh;
1451         struct ext4_dir_entry_2 *de;
1452         struct super_block *sb;
1453         int     retval;
1454         int     dx_fallback=0;
1455         unsigned blocksize;
1456         ext4_lblk_t block, blocks;
1457
1458         sb = dir->i_sb;
1459         blocksize = sb->s_blocksize;
1460         if (!dentry->d_name.len)
1461                 return -EINVAL;
1462         if (is_dx(dir)) {
1463                 retval = ext4_dx_add_entry(handle, dentry, inode);
1464                 if (!retval || (retval != ERR_BAD_DX_DIR))
1465                         return retval;
1466                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1467                 dx_fallback++;
1468                 ext4_mark_inode_dirty(handle, dir);
1469         }
1470         blocks = dir->i_size >> sb->s_blocksize_bits;
1471         for (block = 0; block < blocks; block++) {
1472                 bh = ext4_bread(handle, dir, block, 0, &retval);
1473                 if(!bh)
1474                         return retval;
1475                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1476                 if (retval != -ENOSPC) {
1477                         brelse(bh);
1478                         return retval;
1479                 }
1480
1481                 if (blocks == 1 && !dx_fallback &&
1482                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1483                         return make_indexed_dir(handle, dentry, inode, bh);
1484                 brelse(bh);
1485         }
1486         bh = ext4_append(handle, dir, &block, &retval);
1487         if (!bh)
1488                 return retval;
1489         de = (struct ext4_dir_entry_2 *) bh->b_data;
1490         de->inode = 0;
1491         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1492         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1493         brelse(bh);
1494         if (retval == 0)
1495                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1496         return retval;
1497 }
1498
1499 /*
1500  * Returns 0 for success, or a negative error value
1501  */
1502 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1503                              struct inode *inode)
1504 {
1505         struct dx_frame frames[2], *frame;
1506         struct dx_entry *entries, *at;
1507         struct dx_hash_info hinfo;
1508         struct buffer_head *bh;
1509         struct inode *dir = dentry->d_parent->d_inode;
1510         struct super_block *sb = dir->i_sb;
1511         struct ext4_dir_entry_2 *de;
1512         int err;
1513
1514         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1515         if (!frame)
1516                 return err;
1517         entries = frame->entries;
1518         at = frame->at;
1519
1520         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1521                 goto cleanup;
1522
1523         BUFFER_TRACE(bh, "get_write_access");
1524         err = ext4_journal_get_write_access(handle, bh);
1525         if (err)
1526                 goto journal_error;
1527
1528         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1529         if (err != -ENOSPC)
1530                 goto cleanup;
1531
1532         /* Block full, should compress but for now just split */
1533         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1534                        dx_get_count(entries), dx_get_limit(entries)));
1535         /* Need to split index? */
1536         if (dx_get_count(entries) == dx_get_limit(entries)) {
1537                 ext4_lblk_t newblock;
1538                 unsigned icount = dx_get_count(entries);
1539                 int levels = frame - frames;
1540                 struct dx_entry *entries2;
1541                 struct dx_node *node2;
1542                 struct buffer_head *bh2;
1543
1544                 if (levels && (dx_get_count(frames->entries) ==
1545                                dx_get_limit(frames->entries))) {
1546                         ext4_warning(sb, "Directory index full!");
1547                         err = -ENOSPC;
1548                         goto cleanup;
1549                 }
1550                 bh2 = ext4_append (handle, dir, &newblock, &err);
1551                 if (!(bh2))
1552                         goto cleanup;
1553                 node2 = (struct dx_node *)(bh2->b_data);
1554                 entries2 = node2->entries;
1555                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1556                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1557                                                            sb->s_blocksize);
1558                 BUFFER_TRACE(frame->bh, "get_write_access");
1559                 err = ext4_journal_get_write_access(handle, frame->bh);
1560                 if (err)
1561                         goto journal_error;
1562                 if (levels) {
1563                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1564                         unsigned hash2 = dx_get_hash(entries + icount1);
1565                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1566                                        icount1, icount2));
1567
1568                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1569                         err = ext4_journal_get_write_access(handle,
1570                                                              frames[0].bh);
1571                         if (err)
1572                                 goto journal_error;
1573
1574                         memcpy((char *) entries2, (char *) (entries + icount1),
1575                                icount2 * sizeof(struct dx_entry));
1576                         dx_set_count(entries, icount1);
1577                         dx_set_count(entries2, icount2);
1578                         dx_set_limit(entries2, dx_node_limit(dir));
1579
1580                         /* Which index block gets the new entry? */
1581                         if (at - entries >= icount1) {
1582                                 frame->at = at = at - entries - icount1 + entries2;
1583                                 frame->entries = entries = entries2;
1584                                 swap(frame->bh, bh2);
1585                         }
1586                         dx_insert_block(frames + 0, hash2, newblock);
1587                         dxtrace(dx_show_index("node", frames[1].entries));
1588                         dxtrace(dx_show_index("node",
1589                                ((struct dx_node *) bh2->b_data)->entries));
1590                         err = ext4_handle_dirty_metadata(handle, inode, bh2);
1591                         if (err)
1592                                 goto journal_error;
1593                         brelse (bh2);
1594                 } else {
1595                         dxtrace(printk(KERN_DEBUG
1596                                        "Creating second level index...\n"));
1597                         memcpy((char *) entries2, (char *) entries,
1598                                icount * sizeof(struct dx_entry));
1599                         dx_set_limit(entries2, dx_node_limit(dir));
1600
1601                         /* Set up root */
1602                         dx_set_count(entries, 1);
1603                         dx_set_block(entries + 0, newblock);
1604                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1605
1606                         /* Add new access path frame */
1607                         frame = frames + 1;
1608                         frame->at = at = at - entries + entries2;
1609                         frame->entries = entries = entries2;
1610                         frame->bh = bh2;
1611                         err = ext4_journal_get_write_access(handle,
1612                                                              frame->bh);
1613                         if (err)
1614                                 goto journal_error;
1615                 }
1616                 err = ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1617                 if (err) {
1618                         ext4_std_error(inode->i_sb, err);
1619                         goto cleanup;
1620                 }
1621         }
1622         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1623         if (!de)
1624                 goto cleanup;
1625         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1626         goto cleanup;
1627
1628 journal_error:
1629         ext4_std_error(dir->i_sb, err);
1630 cleanup:
1631         if (bh)
1632                 brelse(bh);
1633         dx_release(frames);
1634         return err;
1635 }
1636
1637 /*
1638  * ext4_delete_entry deletes a directory entry by merging it with the
1639  * previous entry
1640  */
1641 static int ext4_delete_entry(handle_t *handle,
1642                              struct inode *dir,
1643                              struct ext4_dir_entry_2 *de_del,
1644                              struct buffer_head *bh)
1645 {
1646         struct ext4_dir_entry_2 *de, *pde;
1647         unsigned int blocksize = dir->i_sb->s_blocksize;
1648         int i, err;
1649
1650         i = 0;
1651         pde = NULL;
1652         de = (struct ext4_dir_entry_2 *) bh->b_data;
1653         while (i < bh->b_size) {
1654                 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1655                         return -EIO;
1656                 if (de == de_del)  {
1657                         BUFFER_TRACE(bh, "get_write_access");
1658                         err = ext4_journal_get_write_access(handle, bh);
1659                         if (unlikely(err)) {
1660                                 ext4_std_error(dir->i_sb, err);
1661                                 return err;
1662                         }
1663                         if (pde)
1664                                 pde->rec_len = ext4_rec_len_to_disk(
1665                                         ext4_rec_len_from_disk(pde->rec_len,
1666                                                                blocksize) +
1667                                         ext4_rec_len_from_disk(de->rec_len,
1668                                                                blocksize),
1669                                         blocksize);
1670                         else
1671                                 de->inode = 0;
1672                         dir->i_version++;
1673                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1674                         err = ext4_handle_dirty_metadata(handle, dir, bh);
1675                         if (unlikely(err)) {
1676                                 ext4_std_error(dir->i_sb, err);
1677                                 return err;
1678                         }
1679                         return 0;
1680                 }
1681                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1682                 pde = de;
1683                 de = ext4_next_entry(de, blocksize);
1684         }
1685         return -ENOENT;
1686 }
1687
1688 /*
1689  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1690  * since this indicates that nlinks count was previously 1.
1691  */
1692 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1693 {
1694         inc_nlink(inode);
1695         if (is_dx(inode) && inode->i_nlink > 1) {
1696                 /* limit is 16-bit i_links_count */
1697                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1698                         inode->i_nlink = 1;
1699                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1700                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1701                 }
1702         }
1703 }
1704
1705 /*
1706  * If a directory had nlink == 1, then we should let it be 1. This indicates
1707  * directory has >EXT4_LINK_MAX subdirs.
1708  */
1709 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1710 {
1711         drop_nlink(inode);
1712         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1713                 inc_nlink(inode);
1714 }
1715
1716
1717 static int ext4_add_nondir(handle_t *handle,
1718                 struct dentry *dentry, struct inode *inode)
1719 {
1720         int err = ext4_add_entry(handle, dentry, inode);
1721         if (!err) {
1722                 ext4_mark_inode_dirty(handle, inode);
1723                 d_instantiate(dentry, inode);
1724                 unlock_new_inode(inode);
1725                 return 0;
1726         }
1727         drop_nlink(inode);
1728         unlock_new_inode(inode);
1729         iput(inode);
1730         return err;
1731 }
1732
1733 /*
1734  * By the time this is called, we already have created
1735  * the directory cache entry for the new file, but it
1736  * is so far negative - it has no inode.
1737  *
1738  * If the create succeeds, we fill in the inode information
1739  * with d_instantiate().
1740  */
1741 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1742                        struct nameidata *nd)
1743 {
1744         handle_t *handle;
1745         struct inode *inode;
1746         int err, retries = 0;
1747
1748         dquot_initialize(dir);
1749
1750 retry:
1751         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1752                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1753                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1754         if (IS_ERR(handle))
1755                 return PTR_ERR(handle);
1756
1757         if (IS_DIRSYNC(dir))
1758                 ext4_handle_sync(handle);
1759
1760         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1761         err = PTR_ERR(inode);
1762         if (!IS_ERR(inode)) {
1763                 inode->i_op = &ext4_file_inode_operations;
1764                 inode->i_fop = &ext4_file_operations;
1765                 ext4_set_aops(inode);
1766                 err = ext4_add_nondir(handle, dentry, inode);
1767         }
1768         ext4_journal_stop(handle);
1769         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1770                 goto retry;
1771         return err;
1772 }
1773
1774 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1775                       int mode, dev_t rdev)
1776 {
1777         handle_t *handle;
1778         struct inode *inode;
1779         int err, retries = 0;
1780
1781         if (!new_valid_dev(rdev))
1782                 return -EINVAL;
1783
1784         dquot_initialize(dir);
1785
1786 retry:
1787         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1788                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1789                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1790         if (IS_ERR(handle))
1791                 return PTR_ERR(handle);
1792
1793         if (IS_DIRSYNC(dir))
1794                 ext4_handle_sync(handle);
1795
1796         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1797         err = PTR_ERR(inode);
1798         if (!IS_ERR(inode)) {
1799                 init_special_inode(inode, inode->i_mode, rdev);
1800 #ifdef CONFIG_EXT4_FS_XATTR
1801                 inode->i_op = &ext4_special_inode_operations;
1802 #endif
1803                 err = ext4_add_nondir(handle, dentry, inode);
1804         }
1805         ext4_journal_stop(handle);
1806         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1807                 goto retry;
1808         return err;
1809 }
1810
1811 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1812 {
1813         handle_t *handle;
1814         struct inode *inode;
1815         struct buffer_head *dir_block = NULL;
1816         struct ext4_dir_entry_2 *de;
1817         unsigned int blocksize = dir->i_sb->s_blocksize;
1818         int err, retries = 0;
1819
1820         if (EXT4_DIR_LINK_MAX(dir))
1821                 return -EMLINK;
1822
1823         dquot_initialize(dir);
1824
1825 retry:
1826         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1827                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1828                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1829         if (IS_ERR(handle))
1830                 return PTR_ERR(handle);
1831
1832         if (IS_DIRSYNC(dir))
1833                 ext4_handle_sync(handle);
1834
1835         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1836                                &dentry->d_name, 0);
1837         err = PTR_ERR(inode);
1838         if (IS_ERR(inode))
1839                 goto out_stop;
1840
1841         inode->i_op = &ext4_dir_inode_operations;
1842         inode->i_fop = &ext4_dir_operations;
1843         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1844         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1845         if (!dir_block)
1846                 goto out_clear_inode;
1847         BUFFER_TRACE(dir_block, "get_write_access");
1848         err = ext4_journal_get_write_access(handle, dir_block);
1849         if (err)
1850                 goto out_clear_inode;
1851         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1852         de->inode = cpu_to_le32(inode->i_ino);
1853         de->name_len = 1;
1854         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1855                                            blocksize);
1856         strcpy(de->name, ".");
1857         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1858         de = ext4_next_entry(de, blocksize);
1859         de->inode = cpu_to_le32(dir->i_ino);
1860         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1861                                            blocksize);
1862         de->name_len = 2;
1863         strcpy(de->name, "..");
1864         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1865         inode->i_nlink = 2;
1866         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1867         err = ext4_handle_dirty_metadata(handle, dir, dir_block);
1868         if (err)
1869                 goto out_clear_inode;
1870         err = ext4_mark_inode_dirty(handle, inode);
1871         if (!err)
1872                 err = ext4_add_entry(handle, dentry, inode);
1873         if (err) {
1874 out_clear_inode:
1875                 clear_nlink(inode);
1876                 unlock_new_inode(inode);
1877                 ext4_mark_inode_dirty(handle, inode);
1878                 iput(inode);
1879                 goto out_stop;
1880         }
1881         ext4_inc_count(handle, dir);
1882         ext4_update_dx_flag(dir);
1883         err = ext4_mark_inode_dirty(handle, dir);
1884         if (err)
1885                 goto out_clear_inode;
1886         d_instantiate(dentry, inode);
1887         unlock_new_inode(inode);
1888 out_stop:
1889         brelse(dir_block);
1890         ext4_journal_stop(handle);
1891         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1892                 goto retry;
1893         return err;
1894 }
1895
1896 /*
1897  * routine to check that the specified directory is empty (for rmdir)
1898  */
1899 static int empty_dir(struct inode *inode)
1900 {
1901         unsigned int offset;
1902         struct buffer_head *bh;
1903         struct ext4_dir_entry_2 *de, *de1;
1904         struct super_block *sb;
1905         int err = 0;
1906
1907         sb = inode->i_sb;
1908         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1909             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1910                 if (err)
1911                         EXT4_ERROR_INODE(inode,
1912                                 "error %d reading directory lblock 0", err);
1913                 else
1914                         ext4_warning(inode->i_sb,
1915                                      "bad directory (dir #%lu) - no data block",
1916                                      inode->i_ino);
1917                 return 1;
1918         }
1919         de = (struct ext4_dir_entry_2 *) bh->b_data;
1920         de1 = ext4_next_entry(de, sb->s_blocksize);
1921         if (le32_to_cpu(de->inode) != inode->i_ino ||
1922                         !le32_to_cpu(de1->inode) ||
1923                         strcmp(".", de->name) ||
1924                         strcmp("..", de1->name)) {
1925                 ext4_warning(inode->i_sb,
1926                              "bad directory (dir #%lu) - no `.' or `..'",
1927                              inode->i_ino);
1928                 brelse(bh);
1929                 return 1;
1930         }
1931         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1932                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1933         de = ext4_next_entry(de1, sb->s_blocksize);
1934         while (offset < inode->i_size) {
1935                 if (!bh ||
1936                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1937                         unsigned int lblock;
1938                         err = 0;
1939                         brelse(bh);
1940                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1941                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
1942                         if (!bh) {
1943                                 if (err)
1944                                         EXT4_ERROR_INODE(inode,
1945                                                 "error %d reading directory "
1946                                                 "lblock %u", err, lblock);
1947                                 offset += sb->s_blocksize;
1948                                 continue;
1949                         }
1950                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1951                 }
1952                 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1953                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1954                                                          sb->s_blocksize);
1955                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1956                         continue;
1957                 }
1958                 if (le32_to_cpu(de->inode)) {
1959                         brelse(bh);
1960                         return 0;
1961                 }
1962                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1963                 de = ext4_next_entry(de, sb->s_blocksize);
1964         }
1965         brelse(bh);
1966         return 1;
1967 }
1968
1969 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1970  * such inodes, starting at the superblock, in case we crash before the
1971  * file is closed/deleted, or in case the inode truncate spans multiple
1972  * transactions and the last transaction is not recovered after a crash.
1973  *
1974  * At filesystem recovery time, we walk this list deleting unlinked
1975  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1976  */
1977 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1978 {
1979         struct super_block *sb = inode->i_sb;
1980         struct ext4_iloc iloc;
1981         int err = 0, rc;
1982
1983         if (!ext4_handle_valid(handle))
1984                 return 0;
1985
1986         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1987         if (!list_empty(&EXT4_I(inode)->i_orphan))
1988                 goto out_unlock;
1989
1990         /* Orphan handling is only valid for files with data blocks
1991          * being truncated, or files being unlinked. */
1992
1993         /* @@@ FIXME: Observation from aviro:
1994          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1995          * here (on s_orphan_lock), so race with ext4_link() which might bump
1996          * ->i_nlink. For, say it, character device. Not a regular file,
1997          * not a directory, not a symlink and ->i_nlink > 0.
1998          *
1999          * tytso, 4/25/2009: I'm not sure how that could happen;
2000          * shouldn't the fs core protect us from these sort of
2001          * unlink()/link() races?
2002          */
2003         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2004                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2005
2006         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2007         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2008         if (err)
2009                 goto out_unlock;
2010
2011         err = ext4_reserve_inode_write(handle, inode, &iloc);
2012         if (err)
2013                 goto out_unlock;
2014         /*
2015          * Due to previous errors inode may be already a part of on-disk
2016          * orphan list. If so skip on-disk list modification.
2017          */
2018         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2019                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2020                         goto mem_insert;
2021
2022         /* Insert this inode at the head of the on-disk orphan list... */
2023         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2024         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2025         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2026         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2027         if (!err)
2028                 err = rc;
2029
2030         /* Only add to the head of the in-memory list if all the
2031          * previous operations succeeded.  If the orphan_add is going to
2032          * fail (possibly taking the journal offline), we can't risk
2033          * leaving the inode on the orphan list: stray orphan-list
2034          * entries can cause panics at unmount time.
2035          *
2036          * This is safe: on error we're going to ignore the orphan list
2037          * anyway on the next recovery. */
2038 mem_insert:
2039         if (!err)
2040                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2041
2042         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2043         jbd_debug(4, "orphan inode %lu will point to %d\n",
2044                         inode->i_ino, NEXT_ORPHAN(inode));
2045 out_unlock:
2046         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2047         ext4_std_error(inode->i_sb, err);
2048         return err;
2049 }
2050
2051 /*
2052  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2053  * of such inodes stored on disk, because it is finally being cleaned up.
2054  */
2055 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2056 {
2057         struct list_head *prev;
2058         struct ext4_inode_info *ei = EXT4_I(inode);
2059         struct ext4_sb_info *sbi;
2060         __u32 ino_next;
2061         struct ext4_iloc iloc;
2062         int err = 0;
2063
2064         /* ext4_handle_valid() assumes a valid handle_t pointer */
2065         if (handle && !ext4_handle_valid(handle))
2066                 return 0;
2067
2068         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2069         if (list_empty(&ei->i_orphan))
2070                 goto out;
2071
2072         ino_next = NEXT_ORPHAN(inode);
2073         prev = ei->i_orphan.prev;
2074         sbi = EXT4_SB(inode->i_sb);
2075
2076         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2077
2078         list_del_init(&ei->i_orphan);
2079
2080         /* If we're on an error path, we may not have a valid
2081          * transaction handle with which to update the orphan list on
2082          * disk, but we still need to remove the inode from the linked
2083          * list in memory. */
2084         if (sbi->s_journal && !handle)
2085                 goto out;
2086
2087         err = ext4_reserve_inode_write(handle, inode, &iloc);
2088         if (err)
2089                 goto out_err;
2090
2091         if (prev == &sbi->s_orphan) {
2092                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2093                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2094                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2095                 if (err)
2096                         goto out_brelse;
2097                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2098                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2099         } else {
2100                 struct ext4_iloc iloc2;
2101                 struct inode *i_prev =
2102                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2103
2104                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2105                           i_prev->i_ino, ino_next);
2106                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2107                 if (err)
2108                         goto out_brelse;
2109                 NEXT_ORPHAN(i_prev) = ino_next;
2110                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2111         }
2112         if (err)
2113                 goto out_brelse;
2114         NEXT_ORPHAN(inode) = 0;
2115         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2116
2117 out_err:
2118         ext4_std_error(inode->i_sb, err);
2119 out:
2120         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2121         return err;
2122
2123 out_brelse:
2124         brelse(iloc.bh);
2125         goto out_err;
2126 }
2127
2128 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2129 {
2130         int retval;
2131         struct inode *inode;
2132         struct buffer_head *bh;
2133         struct ext4_dir_entry_2 *de;
2134         handle_t *handle;
2135
2136         /* Initialize quotas before so that eventual writes go in
2137          * separate transaction */
2138         dquot_initialize(dir);
2139         dquot_initialize(dentry->d_inode);
2140
2141         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2142         if (IS_ERR(handle))
2143                 return PTR_ERR(handle);
2144
2145         retval = -ENOENT;
2146         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2147         if (!bh)
2148                 goto end_rmdir;
2149
2150         if (IS_DIRSYNC(dir))
2151                 ext4_handle_sync(handle);
2152
2153         inode = dentry->d_inode;
2154
2155         retval = -EIO;
2156         if (le32_to_cpu(de->inode) != inode->i_ino)
2157                 goto end_rmdir;
2158
2159         retval = -ENOTEMPTY;
2160         if (!empty_dir(inode))
2161                 goto end_rmdir;
2162
2163         retval = ext4_delete_entry(handle, dir, de, bh);
2164         if (retval)
2165                 goto end_rmdir;
2166         if (!EXT4_DIR_LINK_EMPTY(inode))
2167                 ext4_warning(inode->i_sb,
2168                              "empty directory has too many links (%d)",
2169                              inode->i_nlink);
2170         inode->i_version++;
2171         clear_nlink(inode);
2172         /* There's no need to set i_disksize: the fact that i_nlink is
2173          * zero will ensure that the right thing happens during any
2174          * recovery. */
2175         inode->i_size = 0;
2176         ext4_orphan_add(handle, inode);
2177         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2178         ext4_mark_inode_dirty(handle, inode);
2179         ext4_dec_count(handle, dir);
2180         ext4_update_dx_flag(dir);
2181         ext4_mark_inode_dirty(handle, dir);
2182
2183 end_rmdir:
2184         ext4_journal_stop(handle);
2185         brelse(bh);
2186         return retval;
2187 }
2188
2189 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2190 {
2191         int retval;
2192         struct inode *inode;
2193         struct buffer_head *bh;
2194         struct ext4_dir_entry_2 *de;
2195         handle_t *handle;
2196
2197         trace_ext4_unlink_enter(dir, dentry);
2198         /* Initialize quotas before so that eventual writes go
2199          * in separate transaction */
2200         dquot_initialize(dir);
2201         dquot_initialize(dentry->d_inode);
2202
2203         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2204         if (IS_ERR(handle))
2205                 return PTR_ERR(handle);
2206
2207         if (IS_DIRSYNC(dir))
2208                 ext4_handle_sync(handle);
2209
2210         retval = -ENOENT;
2211         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2212         if (!bh)
2213                 goto end_unlink;
2214
2215         inode = dentry->d_inode;
2216
2217         retval = -EIO;
2218         if (le32_to_cpu(de->inode) != inode->i_ino)
2219                 goto end_unlink;
2220
2221         if (!inode->i_nlink) {
2222                 ext4_warning(inode->i_sb,
2223                              "Deleting nonexistent file (%lu), %d",
2224                              inode->i_ino, inode->i_nlink);
2225                 inode->i_nlink = 1;
2226         }
2227         retval = ext4_delete_entry(handle, dir, de, bh);
2228         if (retval)
2229                 goto end_unlink;
2230         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2231         ext4_update_dx_flag(dir);
2232         ext4_mark_inode_dirty(handle, dir);
2233         drop_nlink(inode);
2234         if (!inode->i_nlink)
2235                 ext4_orphan_add(handle, inode);
2236         inode->i_ctime = ext4_current_time(inode);
2237         ext4_mark_inode_dirty(handle, inode);
2238         retval = 0;
2239
2240 end_unlink:
2241         ext4_journal_stop(handle);
2242         brelse(bh);
2243         trace_ext4_unlink_exit(dentry, retval);
2244         return retval;
2245 }
2246
2247 static int ext4_symlink(struct inode *dir,
2248                         struct dentry *dentry, const char *symname)
2249 {
2250         handle_t *handle;
2251         struct inode *inode;
2252         int l, err, retries = 0;
2253         int credits;
2254
2255         l = strlen(symname)+1;
2256         if (l > dir->i_sb->s_blocksize)
2257                 return -ENAMETOOLONG;
2258
2259         dquot_initialize(dir);
2260
2261         if (l > EXT4_N_BLOCKS * 4) {
2262                 /*
2263                  * For non-fast symlinks, we just allocate inode and put it on
2264                  * orphan list in the first transaction => we need bitmap,
2265                  * group descriptor, sb, inode block, quota blocks.
2266                  */
2267                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2268         } else {
2269                 /*
2270                  * Fast symlink. We have to add entry to directory
2271                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2272                  * allocate new inode (bitmap, group descriptor, inode block,
2273                  * quota blocks, sb is already counted in previous macros).
2274                  */
2275                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2276                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2277                           EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2278         }
2279 retry:
2280         handle = ext4_journal_start(dir, credits);
2281         if (IS_ERR(handle))
2282                 return PTR_ERR(handle);
2283
2284         if (IS_DIRSYNC(dir))
2285                 ext4_handle_sync(handle);
2286
2287         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2288                                &dentry->d_name, 0);
2289         err = PTR_ERR(inode);
2290         if (IS_ERR(inode))
2291                 goto out_stop;
2292
2293         if (l > EXT4_N_BLOCKS * 4) {
2294                 inode->i_op = &ext4_symlink_inode_operations;
2295                 ext4_set_aops(inode);
2296                 /*
2297                  * We cannot call page_symlink() with transaction started
2298                  * because it calls into ext4_write_begin() which can wait
2299                  * for transaction commit if we are running out of space
2300                  * and thus we deadlock. So we have to stop transaction now
2301                  * and restart it when symlink contents is written.
2302                  *
2303                  * To keep fs consistent in case of crash, we have to put inode
2304                  * to orphan list in the mean time.
2305                  */
2306                 drop_nlink(inode);
2307                 err = ext4_orphan_add(handle, inode);
2308                 ext4_journal_stop(handle);
2309                 if (err)
2310                         goto err_drop_inode;
2311                 err = __page_symlink(inode, symname, l, 1);
2312                 if (err)
2313                         goto err_drop_inode;
2314                 /*
2315                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2316                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2317                  */
2318                 handle = ext4_journal_start(dir,
2319                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2320                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2321                 if (IS_ERR(handle)) {
2322                         err = PTR_ERR(handle);
2323                         goto err_drop_inode;
2324                 }
2325                 inc_nlink(inode);
2326                 err = ext4_orphan_del(handle, inode);
2327                 if (err) {
2328                         ext4_journal_stop(handle);
2329                         clear_nlink(inode);
2330                         goto err_drop_inode;
2331                 }
2332         } else {
2333                 /* clear the extent format for fast symlink */
2334                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2335                 inode->i_op = &ext4_fast_symlink_inode_operations;
2336                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2337                 inode->i_size = l-1;
2338         }
2339         EXT4_I(inode)->i_disksize = inode->i_size;
2340         err = ext4_add_nondir(handle, dentry, inode);
2341 out_stop:
2342         ext4_journal_stop(handle);
2343         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2344                 goto retry;
2345         return err;
2346 err_drop_inode:
2347         unlock_new_inode(inode);
2348         iput(inode);
2349         return err;
2350 }
2351
2352 static int ext4_link(struct dentry *old_dentry,
2353                      struct inode *dir, struct dentry *dentry)
2354 {
2355         handle_t *handle;
2356         struct inode *inode = old_dentry->d_inode;
2357         int err, retries = 0;
2358
2359         if (inode->i_nlink >= EXT4_LINK_MAX)
2360                 return -EMLINK;
2361
2362         dquot_initialize(dir);
2363
2364 retry:
2365         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2366                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2367         if (IS_ERR(handle))
2368                 return PTR_ERR(handle);
2369
2370         if (IS_DIRSYNC(dir))
2371                 ext4_handle_sync(handle);
2372
2373         inode->i_ctime = ext4_current_time(inode);
2374         ext4_inc_count(handle, inode);
2375         ihold(inode);
2376
2377         err = ext4_add_entry(handle, dentry, inode);
2378         if (!err) {
2379                 ext4_mark_inode_dirty(handle, inode);
2380                 d_instantiate(dentry, inode);
2381         } else {
2382                 drop_nlink(inode);
2383                 iput(inode);
2384         }
2385         ext4_journal_stop(handle);
2386         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2387                 goto retry;
2388         return err;
2389 }
2390
2391 #define PARENT_INO(buffer, size) \
2392         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2393
2394 /*
2395  * Anybody can rename anything with this: the permission checks are left to the
2396  * higher-level routines.
2397  */
2398 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2399                        struct inode *new_dir, struct dentry *new_dentry)
2400 {
2401         handle_t *handle;
2402         struct inode *old_inode, *new_inode;
2403         struct buffer_head *old_bh, *new_bh, *dir_bh;
2404         struct ext4_dir_entry_2 *old_de, *new_de;
2405         int retval, force_da_alloc = 0;
2406
2407         dquot_initialize(old_dir);
2408         dquot_initialize(new_dir);
2409
2410         old_bh = new_bh = dir_bh = NULL;
2411
2412         /* Initialize quotas before so that eventual writes go
2413          * in separate transaction */
2414         if (new_dentry->d_inode)
2415                 dquot_initialize(new_dentry->d_inode);
2416         handle = ext4_journal_start(old_dir, 2 *
2417                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2418                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2419         if (IS_ERR(handle))
2420                 return PTR_ERR(handle);
2421
2422         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2423                 ext4_handle_sync(handle);
2424
2425         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2426         /*
2427          *  Check for inode number is _not_ due to possible IO errors.
2428          *  We might rmdir the source, keep it as pwd of some process
2429          *  and merrily kill the link to whatever was created under the
2430          *  same name. Goodbye sticky bit ;-<
2431          */
2432         old_inode = old_dentry->d_inode;
2433         retval = -ENOENT;
2434         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2435                 goto end_rename;
2436
2437         new_inode = new_dentry->d_inode;
2438         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2439         if (new_bh) {
2440                 if (!new_inode) {
2441                         brelse(new_bh);
2442                         new_bh = NULL;
2443                 }
2444         }
2445         if (S_ISDIR(old_inode->i_mode)) {
2446                 if (new_inode) {
2447                         retval = -ENOTEMPTY;
2448                         if (!empty_dir(new_inode))
2449                                 goto end_rename;
2450                 }
2451                 retval = -EIO;
2452                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2453                 if (!dir_bh)
2454                         goto end_rename;
2455                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2456                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2457                         goto end_rename;
2458                 retval = -EMLINK;
2459                 if (!new_inode && new_dir != old_dir &&
2460                     EXT4_DIR_LINK_MAX(new_dir))
2461                         goto end_rename;
2462                 BUFFER_TRACE(dir_bh, "get_write_access");
2463                 retval = ext4_journal_get_write_access(handle, dir_bh);
2464                 if (retval)
2465                         goto end_rename;
2466         }
2467         if (!new_bh) {
2468                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2469                 if (retval)
2470                         goto end_rename;
2471         } else {
2472                 BUFFER_TRACE(new_bh, "get write access");
2473                 retval = ext4_journal_get_write_access(handle, new_bh);
2474                 if (retval)
2475                         goto end_rename;
2476                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2477                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2478                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2479                         new_de->file_type = old_de->file_type;
2480                 new_dir->i_version++;
2481                 new_dir->i_ctime = new_dir->i_mtime =
2482                                         ext4_current_time(new_dir);
2483                 ext4_mark_inode_dirty(handle, new_dir);
2484                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2485                 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2486                 if (unlikely(retval)) {
2487                         ext4_std_error(new_dir->i_sb, retval);
2488                         goto end_rename;
2489                 }
2490                 brelse(new_bh);
2491                 new_bh = NULL;
2492         }
2493
2494         /*
2495          * Like most other Unix systems, set the ctime for inodes on a
2496          * rename.
2497          */
2498         old_inode->i_ctime = ext4_current_time(old_inode);
2499         ext4_mark_inode_dirty(handle, old_inode);
2500
2501         /*
2502          * ok, that's it
2503          */
2504         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2505             old_de->name_len != old_dentry->d_name.len ||
2506             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2507             (retval = ext4_delete_entry(handle, old_dir,
2508                                         old_de, old_bh)) == -ENOENT) {
2509                 /* old_de could have moved from under us during htree split, so
2510                  * make sure that we are deleting the right entry.  We might
2511                  * also be pointing to a stale entry in the unused part of
2512                  * old_bh so just checking inum and the name isn't enough. */
2513                 struct buffer_head *old_bh2;
2514                 struct ext4_dir_entry_2 *old_de2;
2515
2516                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2517                 if (old_bh2) {
2518                         retval = ext4_delete_entry(handle, old_dir,
2519                                                    old_de2, old_bh2);
2520                         brelse(old_bh2);
2521                 }
2522         }
2523         if (retval) {
2524                 ext4_warning(old_dir->i_sb,
2525                                 "Deleting old file (%lu), %d, error=%d",
2526                                 old_dir->i_ino, old_dir->i_nlink, retval);
2527         }
2528
2529         if (new_inode) {
2530                 ext4_dec_count(handle, new_inode);
2531                 new_inode->i_ctime = ext4_current_time(new_inode);
2532         }
2533         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2534         ext4_update_dx_flag(old_dir);
2535         if (dir_bh) {
2536                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2537                                                 cpu_to_le32(new_dir->i_ino);
2538                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2539                 retval = ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2540                 if (retval) {
2541                         ext4_std_error(old_dir->i_sb, retval);
2542                         goto end_rename;
2543                 }
2544                 ext4_dec_count(handle, old_dir);
2545                 if (new_inode) {
2546                         /* checked empty_dir above, can't have another parent,
2547                          * ext4_dec_count() won't work for many-linked dirs */
2548                         new_inode->i_nlink = 0;
2549                 } else {
2550                         ext4_inc_count(handle, new_dir);
2551                         ext4_update_dx_flag(new_dir);
2552                         ext4_mark_inode_dirty(handle, new_dir);
2553                 }
2554         }
2555         ext4_mark_inode_dirty(handle, old_dir);
2556         if (new_inode) {
2557                 ext4_mark_inode_dirty(handle, new_inode);
2558                 if (!new_inode->i_nlink)
2559                         ext4_orphan_add(handle, new_inode);
2560                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2561                         force_da_alloc = 1;
2562         }
2563         retval = 0;
2564
2565 end_rename:
2566         brelse(dir_bh);
2567         brelse(old_bh);
2568         brelse(new_bh);
2569         ext4_journal_stop(handle);
2570         if (retval == 0 && force_da_alloc)
2571                 ext4_alloc_da_blocks(old_inode);
2572         return retval;
2573 }
2574
2575 /*
2576  * directories can handle most operations...
2577  */
2578 const struct inode_operations ext4_dir_inode_operations = {
2579         .create         = ext4_create,
2580         .lookup         = ext4_lookup,
2581         .link           = ext4_link,
2582         .unlink         = ext4_unlink,
2583         .symlink        = ext4_symlink,
2584         .mkdir          = ext4_mkdir,
2585         .rmdir          = ext4_rmdir,
2586         .mknod          = ext4_mknod,
2587         .rename         = ext4_rename,
2588         .setattr        = ext4_setattr,
2589 #ifdef CONFIG_EXT4_FS_XATTR
2590         .setxattr       = generic_setxattr,
2591         .getxattr       = generic_getxattr,
2592         .listxattr      = ext4_listxattr,
2593         .removexattr    = generic_removexattr,
2594 #endif
2595         .check_acl      = ext4_check_acl,
2596         .fiemap         = ext4_fiemap,
2597 };
2598
2599 const struct inode_operations ext4_special_inode_operations = {
2600         .setattr        = ext4_setattr,
2601 #ifdef CONFIG_EXT4_FS_XATTR
2602         .setxattr       = generic_setxattr,
2603         .getxattr       = generic_getxattr,
2604         .listxattr      = ext4_listxattr,
2605         .removexattr    = generic_removexattr,
2606 #endif
2607         .check_acl      = ext4_check_acl,
2608 };