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 "namei.h"
  41 #include "xattr.h"
  42 #include "acl.h"
  43
  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 #ifndef swap
  78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  79 #endif
  80
  81 #ifdef DX_DEBUG
  82 #define dxtrace(command) command
  83 #else
  84 #define dxtrace(command)
  85 #endif
  86
  87 struct fake_dirent
  88 {
  89         __le32 inode;
  90         __le16 rec_len;
  91         u8 name_len;
  92         u8 file_type;
  93 };
  94
  95 struct dx_countlimit
  96 {
  97         __le16 limit;
  98         __le16 count;
  99 };
 100
 101 struct dx_entry
 102 {
 103         __le32 hash;
 104         __le32 block;
 105 };
 106
 107 /*
 108  * dx_root_info is laid out so that if it should somehow get overlaid by a
 109  * dirent the two low bits of the hash version will be zero.  Therefore, the
 110  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 111  */
 112
 113 struct dx_root
 114 {
 115         struct fake_dirent dot;
 116         char dot_name[4];
 117         struct fake_dirent dotdot;
 118         char dotdot_name[4];
 119         struct dx_root_info
 120         {
 121                 __le32 reserved_zero;
 122                 u8 hash_version;
 123                 u8 info_length; /* 8 */
 124                 u8 indirect_levels;
 125                 u8 unused_flags;
 126         }
 127         info;
 128         struct dx_entry entries[0];
 129 };
 130
 131 struct dx_node
 132 {
 133         struct fake_dirent fake;
 134         struct dx_entry entries[0];
 135 };
 136
 137
 138 struct dx_frame
 139 {
 140         struct buffer_head *bh;
 141         struct dx_entry *entries;
 142         struct dx_entry *at;
 143 };
 144
 145 struct dx_map_entry
 146 {
 147         u32 hash;
 148         u16 offs;
 149         u16 size;
 150 };
 151
 152 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 153 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 154 static inline unsigned dx_get_hash(struct dx_entry *entry);
 155 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 156 static unsigned dx_get_count(struct dx_entry *entries);
 157 static unsigned dx_get_limit(struct dx_entry *entries);
 158 static void dx_set_count(struct dx_entry *entries, unsigned value);
 159 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 160 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 161 static unsigned dx_node_limit(struct inode *dir);
 162 static struct dx_frame *dx_probe(const struct qstr *d_name,
 163                                  struct inode *dir,
 164                                  struct dx_hash_info *hinfo,
 165                                  struct dx_frame *frame,
 166                                  int *err);
 167 static void dx_release(struct dx_frame *frames);
 168 static int dx_make_map(struct ext4_dir_entry_2 *de, int size,
 169                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 170 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 171 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 172                 struct dx_map_entry *offsets, int count);
 173 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size);
 174 static void dx_insert_block(struct dx_frame *frame,
 175                                         u32 hash, ext4_lblk_t block);
 176 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 177                                  struct dx_frame *frame,
 178                                  struct dx_frame *frames,
 179                                  __u32 *start_hash);
 180 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 181                 const struct qstr *d_name,
 182                 struct ext4_dir_entry_2 **res_dir,
 183                 int *err);
 184 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 185                              struct inode *inode);
 186
 187 /*
 188  * p is at least 6 bytes before the end of page
 189  */
 190 static inline struct ext4_dir_entry_2 *
 191 ext4_next_entry(struct ext4_dir_entry_2 *p)
 192 {
 193         return (struct ext4_dir_entry_2 *)((char *)p +
 194                 ext4_rec_len_from_disk(p->rec_len));
 195 }
 196
 197 /*
 198  * Future: use high four bits of block for coalesce-on-delete flags
 199  * Mask them off for now.
 200  */
 201
 202 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 203 {
 204         return le32_to_cpu(entry->block) & 0x00ffffff;
 205 }
 206
 207 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 208 {
 209         entry->block = cpu_to_le32(value);
 210 }
 211
 212 static inline unsigned dx_get_hash(struct dx_entry *entry)
 213 {
 214         return le32_to_cpu(entry->hash);
 215 }
 216
 217 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 218 {
 219         entry->hash = cpu_to_le32(value);
 220 }
 221
 222 static inline unsigned dx_get_count(struct dx_entry *entries)
 223 {
 224         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 225 }
 226
 227 static inline unsigned dx_get_limit(struct dx_entry *entries)
 228 {
 229         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 230 }
 231
 232 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 233 {
 234         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 235 }
 236
 237 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 238 {
 239         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 240 }
 241
 242 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 243 {
 244         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 245                 EXT4_DIR_REC_LEN(2) - infosize;
 246         return entry_space / sizeof(struct dx_entry);
 247 }
 248
 249 static inline unsigned dx_node_limit(struct inode *dir)
 250 {
 251         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 252         return entry_space / sizeof(struct dx_entry);
 253 }
 254
 255 /*
 256  * Debug
 257  */
 258 #ifdef DX_DEBUG
 259 static void dx_show_index(char * label, struct dx_entry *entries)
 260 {
 261         int i, n = dx_get_count (entries);
 262         printk(KERN_DEBUG "%s index ", label);
 263         for (i = 0; i < n; i++) {
 264                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 265                                 0, (unsigned long)dx_get_block(entries + i));
 266         }
 267         printk("\n");
 268 }
 269
 270 struct stats
 271 {
 272         unsigned names;
 273         unsigned space;
 274         unsigned bcount;
 275 };
 276
 277 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 278                                  int size, int show_names)
 279 {
 280         unsigned names = 0, space = 0;
 281         char *base = (char *) de;
 282         struct dx_hash_info h = *hinfo;
 283
 284         printk("names: ");
 285         while ((char *) de < base + size)
 286         {
 287                 if (de->inode)
 288                 {
 289                         if (show_names)
 290                         {
 291                                 int len = de->name_len;
 292                                 char *name = de->name;
 293                                 while (len--) printk("%c", *name++);
 294                                 ext4fs_dirhash(de->name, de->name_len, &h);
 295                                 printk(":%x.%u ", h.hash,
 296                                        ((char *) de - base));
 297                         }
 298                         space += EXT4_DIR_REC_LEN(de->name_len);
 299                         names++;
 300                 }
 301                 de = ext4_next_entry(de);
 302         }
 303         printk("(%i)\n", names);
 304         return (struct stats) { names, space, 1 };
 305 }
 306
 307 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 308                              struct dx_entry *entries, int levels)
 309 {
 310         unsigned blocksize = dir->i_sb->s_blocksize;
 311         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 312         unsigned bcount = 0;
 313         struct buffer_head *bh;
 314         int err;
 315         printk("%i indexed blocks...\n", count);
 316         for (i = 0; i < count; i++, entries++)
 317         {
 318                 ext4_lblk_t block = dx_get_block(entries);
 319                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 320                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 321                 struct stats stats;
 322                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 323                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 324                 stats = levels?
 325                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 326                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 327                 names += stats.names;
 328                 space += stats.space;
 329                 bcount += stats.bcount;
 330                 brelse(bh);
 331         }
 332         if (bcount)
 333                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 334                        levels ? "" : "   ", names, space/bcount,
 335                        (space/bcount)*100/blocksize);
 336         return (struct stats) { names, space, bcount};
 337 }
 338 #endif /* DX_DEBUG */
 339
 340 /*
 341  * Probe for a directory leaf block to search.
 342  *
 343  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 344  * error in the directory index, and the caller should fall back to
 345  * searching the directory normally.  The callers of dx_probe **MUST**
 346  * check for this error code, and make sure it never gets reflected
 347  * back to userspace.
 348  */
 349 static struct dx_frame *
 350 dx_probe(const struct qstr *d_name, struct inode *dir,
 351          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 352 {
 353         unsigned count, indirect;
 354         struct dx_entry *at, *entries, *p, *q, *m;
 355         struct dx_root *root;
 356         struct buffer_head *bh;
 357         struct dx_frame *frame = frame_in;
 358         u32 hash;
 359
 360         frame->bh = NULL;
 361         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 362                 goto fail;
 363         root = (struct dx_root *) bh->b_data;
 364         if (root->info.hash_version != DX_HASH_TEA &&
 365             root->info.hash_version != DX_HASH_HALF_MD4 &&
 366             root->info.hash_version != DX_HASH_LEGACY) {
 367                 ext4_warning(dir->i_sb, __func__,
 368                              "Unrecognised inode hash code %d",
 369                              root->info.hash_version);
 370                 brelse(bh);
 371                 *err = ERR_BAD_DX_DIR;
 372                 goto fail;
 373         }
 374         hinfo->hash_version = root->info.hash_version;
 375         if (hinfo->hash_version <= DX_HASH_TEA)
 376                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 377         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 378         if (d_name)
 379                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 380         hash = hinfo->hash;
 381
 382         if (root->info.unused_flags & 1) {
 383                 ext4_warning(dir->i_sb, __func__,
 384                              "Unimplemented inode hash flags: %#06x",
 385                              root->info.unused_flags);
 386                 brelse(bh);
 387                 *err = ERR_BAD_DX_DIR;
 388                 goto fail;
 389         }
 390
 391         if ((indirect = root->info.indirect_levels) > 1) {
 392                 ext4_warning(dir->i_sb, __func__,
 393                              "Unimplemented inode hash depth: %#06x",
 394                              root->info.indirect_levels);
 395                 brelse(bh);
 396                 *err = ERR_BAD_DX_DIR;
 397                 goto fail;
 398         }
 399
 400         entries = (struct dx_entry *) (((char *)&root->info) +
 401                                        root->info.info_length);
 402
 403         if (dx_get_limit(entries) != dx_root_limit(dir,
 404                                                    root->info.info_length)) {
 405                 ext4_warning(dir->i_sb, __func__,
 406                              "dx entry: limit != root limit");
 407                 brelse(bh);
 408                 *err = ERR_BAD_DX_DIR;
 409                 goto fail;
 410         }
 411
 412         dxtrace(printk("Look up %x", hash));
 413         while (1)
 414         {
 415                 count = dx_get_count(entries);
 416                 if (!count || count > dx_get_limit(entries)) {
 417                         ext4_warning(dir->i_sb, __func__,
 418                                      "dx entry: no count or count > limit");
 419                         brelse(bh);
 420                         *err = ERR_BAD_DX_DIR;
 421                         goto fail2;
 422                 }
 423
 424                 p = entries + 1;
 425                 q = entries + count - 1;
 426                 while (p <= q)
 427                 {
 428                         m = p + (q - p)/2;
 429                         dxtrace(printk("."));
 430                         if (dx_get_hash(m) > hash)
 431                                 q = m - 1;
 432                         else
 433                                 p = m + 1;
 434                 }
 435
 436                 if (0) // linear search cross check
 437                 {
 438                         unsigned n = count - 1;
 439                         at = entries;
 440                         while (n--)
 441                         {
 442                                 dxtrace(printk(","));
 443                                 if (dx_get_hash(++at) > hash)
 444                                 {
 445                                         at--;
 446                                         break;
 447                                 }
 448                         }
 449                         assert (at == p - 1);
 450                 }
 451
 452                 at = p - 1;
 453                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 454                 frame->bh = bh;
 455                 frame->entries = entries;
 456                 frame->at = at;
 457                 if (!indirect--) return frame;
 458                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 459                         goto fail2;
 460                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 461                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 462                         ext4_warning(dir->i_sb, __func__,
 463                                      "dx entry: limit != node limit");
 464                         brelse(bh);
 465                         *err = ERR_BAD_DX_DIR;
 466                         goto fail2;
 467                 }
 468                 frame++;
 469                 frame->bh = NULL;
 470         }
 471 fail2:
 472         while (frame >= frame_in) {
 473                 brelse(frame->bh);
 474                 frame--;
 475         }
 476 fail:
 477         if (*err == ERR_BAD_DX_DIR)
 478                 ext4_warning(dir->i_sb, __func__,
 479                              "Corrupt dir inode %ld, running e2fsck is "
 480                              "recommended.", dir->i_ino);
 481         return NULL;
 482 }
 483
 484 static void dx_release (struct dx_frame *frames)
 485 {
 486         if (frames[0].bh == NULL)
 487                 return;
 488
 489         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 490                 brelse(frames[1].bh);
 491         brelse(frames[0].bh);
 492 }
 493
 494 /*
 495  * This function increments the frame pointer to search the next leaf
 496  * block, and reads in the necessary intervening nodes if the search
 497  * should be necessary.  Whether or not the search is necessary is
 498  * controlled by the hash parameter.  If the hash value is even, then
 499  * the search is only continued if the next block starts with that
 500  * hash value.  This is used if we are searching for a specific file.
 501  *
 502  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 503  *
 504  * This function returns 1 if the caller should continue to search,
 505  * or 0 if it should not.  If there is an error reading one of the
 506  * index blocks, it will a negative error code.
 507  *
 508  * If start_hash is non-null, it will be filled in with the starting
 509  * hash of the next page.
 510  */
 511 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 512                                  struct dx_frame *frame,
 513                                  struct dx_frame *frames,
 514                                  __u32 *start_hash)
 515 {
 516         struct dx_frame *p;
 517         struct buffer_head *bh;
 518         int err, num_frames = 0;
 519         __u32 bhash;
 520
 521         p = frame;
 522         /*
 523          * Find the next leaf page by incrementing the frame pointer.
 524          * If we run out of entries in the interior node, loop around and
 525          * increment pointer in the parent node.  When we break out of
 526          * this loop, num_frames indicates the number of interior
 527          * nodes need to be read.
 528          */
 529         while (1) {
 530                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 531                         break;
 532                 if (p == frames)
 533                         return 0;
 534                 num_frames++;
 535                 p--;
 536         }
 537
 538         /*
 539          * If the hash is 1, then continue only if the next page has a
 540          * continuation hash of any value.  This is used for readdir
 541          * handling.  Otherwise, check to see if the hash matches the
 542          * desired contiuation hash.  If it doesn't, return since
 543          * there's no point to read in the successive index pages.
 544          */
 545         bhash = dx_get_hash(p->at);
 546         if (start_hash)
 547                 *start_hash = bhash;
 548         if ((hash & 1) == 0) {
 549                 if ((bhash & ~1) != hash)
 550                         return 0;
 551         }
 552         /*
 553          * If the hash is HASH_NB_ALWAYS, we always go to the next
 554          * block so no check is necessary
 555          */
 556         while (num_frames--) {
 557                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 558                                       0, &err)))
 559                         return err; /* Failure */
 560                 p++;
 561                 brelse(p->bh);
 562                 p->bh = bh;
 563                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 564         }
 565         return 1;
 566 }
 567
 568
 569 /*
 570  * This function fills a red-black tree with information from a
 571  * directory block.  It returns the number directory entries loaded
 572  * into the tree.  If there is an error it is returned in err.
 573  */
 574 static int htree_dirblock_to_tree(struct file *dir_file,
 575                                   struct inode *dir, ext4_lblk_t block,
 576                                   struct dx_hash_info *hinfo,
 577                                   __u32 start_hash, __u32 start_minor_hash)
 578 {
 579         struct buffer_head *bh;
 580         struct ext4_dir_entry_2 *de, *top;
 581         int err, count = 0;
 582
 583         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 584                                                         (unsigned long)block));
 585         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 586                 return err;
 587
 588         de = (struct ext4_dir_entry_2 *) bh->b_data;
 589         top = (struct ext4_dir_entry_2 *) ((char *) de +
 590                                            dir->i_sb->s_blocksize -
 591                                            EXT4_DIR_REC_LEN(0));
 592         for (; de < top; de = ext4_next_entry(de)) {
 593                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 594                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 595                                                 +((char *)de - bh->b_data))) {
 596                         /* On error, skip the f_pos to the next block. */
 597                         dir_file->f_pos = (dir_file->f_pos |
 598                                         (dir->i_sb->s_blocksize - 1)) + 1;
 599                         brelse(bh);
 600                         return count;
 601                 }
 602                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 603                 if ((hinfo->hash < start_hash) ||
 604                     ((hinfo->hash == start_hash) &&
 605                      (hinfo->minor_hash < start_minor_hash)))
 606                         continue;
 607                 if (de->inode == 0)
 608                         continue;
 609                 if ((err = ext4_htree_store_dirent(dir_file,
 610                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 611                         brelse(bh);
 612                         return err;
 613                 }
 614                 count++;
 615         }
 616         brelse(bh);
 617         return count;
 618 }
 619
 620
 621 /*
 622  * This function fills a red-black tree with information from a
 623  * directory.  We start scanning the directory in hash order, starting
 624  * at start_hash and start_minor_hash.
 625  *
 626  * This function returns the number of entries inserted into the tree,
 627  * or a negative error code.
 628  */
 629 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 630                          __u32 start_minor_hash, __u32 *next_hash)
 631 {
 632         struct dx_hash_info hinfo;
 633         struct ext4_dir_entry_2 *de;
 634         struct dx_frame frames[2], *frame;
 635         struct inode *dir;
 636         ext4_lblk_t block;
 637         int count = 0;
 638         int ret, err;
 639         __u32 hashval;
 640
 641         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 642                        start_hash, start_minor_hash));
 643         dir = dir_file->f_path.dentry->d_inode;
 644         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
 645                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 646                 if (hinfo.hash_version <= DX_HASH_TEA)
 647                         hinfo.hash_version +=
 648                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 649                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 650                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 651                                                start_hash, start_minor_hash);
 652                 *next_hash = ~0;
 653                 return count;
 654         }
 655         hinfo.hash = start_hash;
 656         hinfo.minor_hash = 0;
 657         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 658         if (!frame)
 659                 return err;
 660
 661         /* Add '.' and '..' from the htree header */
 662         if (!start_hash && !start_minor_hash) {
 663                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 664                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 665                         goto errout;
 666                 count++;
 667         }
 668         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 669                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 670                 de = ext4_next_entry(de);
 671                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 672                         goto errout;
 673                 count++;
 674         }
 675
 676         while (1) {
 677                 block = dx_get_block(frame->at);
 678                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 679                                              start_hash, start_minor_hash);
 680                 if (ret < 0) {
 681                         err = ret;
 682                         goto errout;
 683                 }
 684                 count += ret;
 685                 hashval = ~0;
 686                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 687                                             frame, frames, &hashval);
 688                 *next_hash = hashval;
 689                 if (ret < 0) {
 690                         err = ret;
 691                         goto errout;
 692                 }
 693                 /*
 694                  * Stop if:  (a) there are no more entries, or
 695                  * (b) we have inserted at least one entry and the
 696                  * next hash value is not a continuation
 697                  */
 698                 if ((ret == 0) ||
 699                     (count && ((hashval & 1) == 0)))
 700                         break;
 701         }
 702         dx_release(frames);
 703         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 704                        "next hash: %x\n", count, *next_hash));
 705         return count;
 706 errout:
 707         dx_release(frames);
 708         return (err);
 709 }
 710
 711
 712 /*
 713  * Directory block splitting, compacting
 714  */
 715
 716 /*
 717  * Create map of hash values, offsets, and sizes, stored at end of block.
 718  * Returns number of entries mapped.
 719  */
 720 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
 721                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
 722 {
 723         int count = 0;
 724         char *base = (char *) de;
 725         struct dx_hash_info h = *hinfo;
 726
 727         while ((char *) de < base + size)
 728         {
 729                 if (de->name_len && de->inode) {
 730                         ext4fs_dirhash(de->name, de->name_len, &h);
 731                         map_tail--;
 732                         map_tail->hash = h.hash;
 733                         map_tail->offs = (u16) ((char *) de - base);
 734                         map_tail->size = le16_to_cpu(de->rec_len);
 735                         count++;
 736                         cond_resched();
 737                 }
 738                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 739                 de = ext4_next_entry(de);
 740         }
 741         return count;
 742 }
 743
 744 /* Sort map by hash value */
 745 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 746 {
 747         struct dx_map_entry *p, *q, *top = map + count - 1;
 748         int more;
 749         /* Combsort until bubble sort doesn't suck */
 750         while (count > 2) {
 751                 count = count*10/13;
 752                 if (count - 9 < 2) /* 9, 10 -> 11 */
 753                         count = 11;
 754                 for (p = top, q = p - count; q >= map; p--, q--)
 755                         if (p->hash < q->hash)
 756                                 swap(*p, *q);
 757         }
 758         /* Garden variety bubble sort */
 759         do {
 760                 more = 0;
 761                 q = top;
 762                 while (q-- > map) {
 763                         if (q[1].hash >= q[0].hash)
 764                                 continue;
 765                         swap(*(q+1), *q);
 766                         more = 1;
 767                 }
 768         } while(more);
 769 }
 770
 771 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
 772 {
 773         struct dx_entry *entries = frame->entries;
 774         struct dx_entry *old = frame->at, *new = old + 1;
 775         int count = dx_get_count(entries);
 776
 777         assert(count < dx_get_limit(entries));
 778         assert(old < entries + count);
 779         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 780         dx_set_hash(new, hash);
 781         dx_set_block(new, block);
 782         dx_set_count(entries, count + 1);
 783 }
 784
 785 static void ext4_update_dx_flag(struct inode *inode)
 786 {
 787         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 788                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 789                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
 790 }
 791
 792 /*
 793  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 794  *
 795  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 796  * `de != NULL' is guaranteed by caller.
 797  */
 798 static inline int ext4_match (int len, const char * const name,
 799                               struct ext4_dir_entry_2 * de)
 800 {
 801         if (len != de->name_len)
 802                 return 0;
 803         if (!de->inode)
 804                 return 0;
 805         return !memcmp(name, de->name, len);
 806 }
 807
 808 /*
 809  * Returns 0 if not found, -1 on failure, and 1 on success
 810  */
 811 static inline int search_dirblock(struct buffer_head *bh,
 812                                   struct inode *dir,
 813                                   const struct qstr *d_name,
 814                                   unsigned long offset,
 815                                   struct ext4_dir_entry_2 ** res_dir)
 816 {
 817         struct ext4_dir_entry_2 * de;
 818         char * dlimit;
 819         int de_len;
 820         const char *name = d_name->name;
 821         int namelen = d_name->len;
 822
 823         de = (struct ext4_dir_entry_2 *) bh->b_data;
 824         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 825         while ((char *) de < dlimit) {
 826                 /* this code is executed quadratically often */
 827                 /* do minimal checking `by hand' */
 828
 829                 if ((char *) de + namelen <= dlimit &&
 830                     ext4_match (namelen, name, de)) {
 831                         /* found a match - just to be sure, do a full check */
 832                         if (!ext4_check_dir_entry("ext4_find_entry",
 833                                                   dir, de, bh, offset))
 834                                 return -1;
 835                         *res_dir = de;
 836                         return 1;
 837                 }
 838                 /* prevent looping on a bad block */
 839                 de_len = ext4_rec_len_from_disk(de->rec_len);
 840                 if (de_len <= 0)
 841                         return -1;
 842                 offset += de_len;
 843                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 844         }
 845         return 0;
 846 }
 847
 848
 849 /*
 850  *      ext4_find_entry()
 851  *
 852  * finds an entry in the specified directory with the wanted name. It
 853  * returns the cache buffer in which the entry was found, and the entry
 854  * itself (as a parameter - res_dir). It does NOT read the inode of the
 855  * entry - you'll have to do that yourself if you want to.
 856  *
 857  * The returned buffer_head has ->b_count elevated.  The caller is expected
 858  * to brelse() it when appropriate.
 859  */
 860 static struct buffer_head * ext4_find_entry (struct inode *dir,
 861                                         const struct qstr *d_name,
 862                                         struct ext4_dir_entry_2 ** res_dir)
 863 {
 864         struct super_block *sb;
 865         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 866         struct buffer_head *bh, *ret = NULL;
 867         ext4_lblk_t start, block, b;
 868         int ra_max = 0;         /* Number of bh's in the readahead
 869                                    buffer, bh_use[] */
 870         int ra_ptr = 0;         /* Current index into readahead
 871                                    buffer */
 872         int num = 0;
 873         ext4_lblk_t  nblocks;
 874         int i, err;
 875         int namelen;
 876
 877         *res_dir = NULL;
 878         sb = dir->i_sb;
 879         namelen = d_name->len;
 880         if (namelen > EXT4_NAME_LEN)
 881                 return NULL;
 882         if (is_dx(dir)) {
 883                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
 884                 /*
 885                  * On success, or if the error was file not found,
 886                  * return.  Otherwise, fall back to doing a search the
 887                  * old fashioned way.
 888                  */
 889                 if (bh || (err != ERR_BAD_DX_DIR))
 890                         return bh;
 891                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
 892                                "falling back\n"));
 893         }
 894         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 895         start = EXT4_I(dir)->i_dir_start_lookup;
 896         if (start >= nblocks)
 897                 start = 0;
 898         block = start;
 899 restart:
 900         do {
 901                 /*
 902                  * We deal with the read-ahead logic here.
 903                  */
 904                 if (ra_ptr >= ra_max) {
 905                         /* Refill the readahead buffer */
 906                         ra_ptr = 0;
 907                         b = block;
 908                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 909                                 /*
 910                                  * Terminate if we reach the end of the
 911                                  * directory and must wrap, or if our
 912                                  * search has finished at this block.
 913                                  */
 914                                 if (b >= nblocks || (num && block == start)) {
 915                                         bh_use[ra_max] = NULL;
 916                                         break;
 917                                 }
 918                                 num++;
 919                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
 920                                 bh_use[ra_max] = bh;
 921                                 if (bh)
 922                                         ll_rw_block(READ_META, 1, &bh);
 923                         }
 924                 }
 925                 if ((bh = bh_use[ra_ptr++]) == NULL)
 926                         goto next;
 927                 wait_on_buffer(bh);
 928                 if (!buffer_uptodate(bh)) {
 929                         /* read error, skip block & hope for the best */
 930                         ext4_error(sb, __func__, "reading directory #%lu "
 931                                    "offset %lu", dir->i_ino,
 932                                    (unsigned long)block);
 933                         brelse(bh);
 934                         goto next;
 935                 }
 936                 i = search_dirblock(bh, dir, d_name,
 937                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
 938                 if (i == 1) {
 939                         EXT4_I(dir)->i_dir_start_lookup = block;
 940                         ret = bh;
 941                         goto cleanup_and_exit;
 942                 } else {
 943                         brelse(bh);
 944                         if (i < 0)
 945                                 goto cleanup_and_exit;
 946                 }
 947         next:
 948                 if (++block >= nblocks)
 949                         block = 0;
 950         } while (block != start);
 951
 952         /*
 953          * If the directory has grown while we were searching, then
 954          * search the last part of the directory before giving up.
 955          */
 956         block = nblocks;
 957         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 958         if (block < nblocks) {
 959                 start = 0;
 960                 goto restart;
 961         }
 962
 963 cleanup_and_exit:
 964         /* Clean up the read-ahead blocks */
 965         for (; ra_ptr < ra_max; ra_ptr++)
 966                 brelse(bh_use[ra_ptr]);
 967         return ret;
 968 }
 969
 970 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
 971                        struct ext4_dir_entry_2 **res_dir, int *err)
 972 {
 973         struct super_block * sb;
 974         struct dx_hash_info     hinfo;
 975         u32 hash;
 976         struct dx_frame frames[2], *frame;
 977         struct ext4_dir_entry_2 *de, *top;
 978         struct buffer_head *bh;
 979         ext4_lblk_t block;
 980         int retval;
 981         int namelen = d_name->len;
 982         const u8 *name = d_name->name;
 983
 984         sb = dir->i_sb;
 985         /* NFS may look up ".." - look at dx_root directory block */
 986         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
 987                 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
 988                         return NULL;
 989         } else {
 990                 frame = frames;
 991                 frame->bh = NULL;                       /* for dx_release() */
 992                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
 993                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
 994         }
 995         hash = hinfo.hash;
 996         do {
 997                 block = dx_get_block(frame->at);
 998                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
 999                         goto errout;
1000                 de = (struct ext4_dir_entry_2 *) bh->b_data;
1001                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1002                                        EXT4_DIR_REC_LEN(0));
1003                 for (; de < top; de = ext4_next_entry(de)) {
1004                         int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1005                                   + ((char *) de - bh->b_data);
1006
1007                         if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1008                                 brelse(bh);
1009                                 *err = ERR_BAD_DX_DIR;
1010                                 goto errout;
1011                         }
1012
1013                         if (ext4_match(namelen, name, de)) {
1014                                 *res_dir = de;
1015                                 dx_release(frames);
1016                                 return bh;
1017                         }
1018                 }
1019                 brelse(bh);
1020                 /* Check to see if we should continue to search */
1021                 retval = ext4_htree_next_block(dir, hash, frame,
1022                                                frames, NULL);
1023                 if (retval < 0) {
1024                         ext4_warning(sb, __func__,
1025                              "error reading index page in directory #%lu",
1026                              dir->i_ino);
1027                         *err = retval;
1028                         goto errout;
1029                 }
1030         } while (retval == 1);
1031
1032         *err = -ENOENT;
1033 errout:
1034         dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1035         dx_release (frames);
1036         return NULL;
1037 }
1038
1039 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1040 {
1041         struct inode *inode;
1042         struct ext4_dir_entry_2 *de;
1043         struct buffer_head *bh;
1044
1045         if (dentry->d_name.len > EXT4_NAME_LEN)
1046                 return ERR_PTR(-ENAMETOOLONG);
1047
1048         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1049         inode = NULL;
1050         if (bh) {
1051                 unsigned long ino = le32_to_cpu(de->inode);
1052                 brelse(bh);
1053                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1054                         ext4_error(dir->i_sb, "ext4_lookup",
1055                                    "bad inode number: %lu", ino);
1056                         return ERR_PTR(-EIO);
1057                 }
1058                 inode = ext4_iget(dir->i_sb, ino);
1059                 if (IS_ERR(inode))
1060                         return ERR_CAST(inode);
1061         }
1062         return d_splice_alias(inode, dentry);
1063 }
1064
1065
1066 struct dentry *ext4_get_parent(struct dentry *child)
1067 {
1068         unsigned long ino;
1069         struct inode *inode;
1070         static const struct qstr dotdot = {
1071                 .name = "..",
1072                 .len = 2,
1073         };
1074         struct ext4_dir_entry_2 * de;
1075         struct buffer_head *bh;
1076
1077         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1078         inode = NULL;
1079         if (!bh)
1080                 return ERR_PTR(-ENOENT);
1081         ino = le32_to_cpu(de->inode);
1082         brelse(bh);
1083
1084         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1085                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1086                            "bad inode number: %lu", ino);
1087                 return ERR_PTR(-EIO);
1088         }
1089
1090         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1091 }
1092
1093 #define S_SHIFT 12
1094 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1095         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1096         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1097         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1098         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1099         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1100         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1101         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1102 };
1103
1104 static inline void ext4_set_de_type(struct super_block *sb,
1105                                 struct ext4_dir_entry_2 *de,
1106                                 umode_t mode) {
1107         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1108                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1109 }
1110
1111 /*
1112  * Move count entries from end of map between two memory locations.
1113  * Returns pointer to last entry moved.
1114  */
1115 static struct ext4_dir_entry_2 *
1116 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1117 {
1118         unsigned rec_len = 0;
1119
1120         while (count--) {
1121                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1122                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1123                 memcpy (to, de, rec_len);
1124                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1125                                 ext4_rec_len_to_disk(rec_len);
1126                 de->inode = 0;
1127                 map++;
1128                 to += rec_len;
1129         }
1130         return (struct ext4_dir_entry_2 *) (to - rec_len);
1131 }
1132
1133 /*
1134  * Compact each dir entry in the range to the minimal rec_len.
1135  * Returns pointer to last entry in range.
1136  */
1137 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1138 {
1139         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1140         unsigned rec_len = 0;
1141
1142         prev = to = de;
1143         while ((char*)de < base + size) {
1144                 next = ext4_next_entry(de);
1145                 if (de->inode && de->name_len) {
1146                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1147                         if (de > to)
1148                                 memmove(to, de, rec_len);
1149                         to->rec_len = ext4_rec_len_to_disk(rec_len);
1150                         prev = to;
1151                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1152                 }
1153                 de = next;
1154         }
1155         return prev;
1156 }
1157
1158 /*
1159  * Split a full leaf block to make room for a new dir entry.
1160  * Allocate a new block, and move entries so that they are approx. equally full.
1161  * Returns pointer to de in block into which the new entry will be inserted.
1162  */
1163 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1164                         struct buffer_head **bh,struct dx_frame *frame,
1165                         struct dx_hash_info *hinfo, int *error)
1166 {
1167         unsigned blocksize = dir->i_sb->s_blocksize;
1168         unsigned count, continued;
1169         struct buffer_head *bh2;
1170         ext4_lblk_t newblock;
1171         u32 hash2;
1172         struct dx_map_entry *map;
1173         char *data1 = (*bh)->b_data, *data2;
1174         unsigned split, move, size, i;
1175         struct ext4_dir_entry_2 *de = NULL, *de2;
1176         int     err = 0;
1177
1178         bh2 = ext4_append (handle, dir, &newblock, &err);
1179         if (!(bh2)) {
1180                 brelse(*bh);
1181                 *bh = NULL;
1182                 goto errout;
1183         }
1184
1185         BUFFER_TRACE(*bh, "get_write_access");
1186         err = ext4_journal_get_write_access(handle, *bh);
1187         if (err)
1188                 goto journal_error;
1189
1190         BUFFER_TRACE(frame->bh, "get_write_access");
1191         err = ext4_journal_get_write_access(handle, frame->bh);
1192         if (err)
1193                 goto journal_error;
1194
1195         data2 = bh2->b_data;
1196
1197         /* create map in the end of data2 block */
1198         map = (struct dx_map_entry *) (data2 + blocksize);
1199         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1200                              blocksize, hinfo, map);
1201         map -= count;
1202         dx_sort_map(map, count);
1203         /* Split the existing block in the middle, size-wise */
1204         size = 0;
1205         move = 0;
1206         for (i = count-1; i >= 0; i--) {
1207                 /* is more than half of this entry in 2nd half of the block? */
1208                 if (size + map[i].size/2 > blocksize/2)
1209                         break;
1210                 size += map[i].size;
1211                 move++;
1212         }
1213         /* map index at which we will split */
1214         split = count - move;
1215         hash2 = map[split].hash;
1216         continued = hash2 == map[split - 1].hash;
1217         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1218                         (unsigned long)dx_get_block(frame->at),
1219                                         hash2, split, count-split));
1220
1221         /* Fancy dance to stay within two buffers */
1222         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1223         de = dx_pack_dirents(data1, blocksize);
1224         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1225         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
1226         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1227         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1228
1229         /* Which block gets the new entry? */
1230         if (hinfo->hash >= hash2)
1231         {
1232                 swap(*bh, bh2);
1233                 de = de2;
1234         }
1235         dx_insert_block(frame, hash2 + continued, newblock);
1236         err = ext4_journal_dirty_metadata(handle, bh2);
1237         if (err)
1238                 goto journal_error;
1239         err = ext4_journal_dirty_metadata(handle, frame->bh);
1240         if (err)
1241                 goto journal_error;
1242         brelse(bh2);
1243         dxtrace(dx_show_index("frame", frame->entries));
1244         return de;
1245
1246 journal_error:
1247         brelse(*bh);
1248         brelse(bh2);
1249         *bh = NULL;
1250         ext4_std_error(dir->i_sb, err);
1251 errout:
1252         *error = err;
1253         return NULL;
1254 }
1255
1256 /*
1257  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1258  * it points to a directory entry which is guaranteed to be large
1259  * enough for new directory entry.  If de is NULL, then
1260  * add_dirent_to_buf will attempt search the directory block for
1261  * space.  It will return -ENOSPC if no space is available, and -EIO
1262  * and -EEXIST if directory entry already exists.
1263  *
1264  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1265  * all other cases bh is released.
1266  */
1267 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1268                              struct inode *inode, struct ext4_dir_entry_2 *de,
1269                              struct buffer_head *bh)
1270 {
1271         struct inode    *dir = dentry->d_parent->d_inode;
1272         const char      *name = dentry->d_name.name;
1273         int             namelen = dentry->d_name.len;
1274         unsigned long   offset = 0;
1275         unsigned short  reclen;
1276         int             nlen, rlen, err;
1277         char            *top;
1278
1279         reclen = EXT4_DIR_REC_LEN(namelen);
1280         if (!de) {
1281                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1282                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1283                 while ((char *) de <= top) {
1284                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1285                                                   bh, offset)) {
1286                                 brelse(bh);
1287                                 return -EIO;
1288                         }
1289                         if (ext4_match(namelen, name, de)) {
1290                                 brelse(bh);
1291                                 return -EEXIST;
1292                         }
1293                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1294                         rlen = ext4_rec_len_from_disk(de->rec_len);
1295                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1296                                 break;
1297                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1298                         offset += rlen;
1299                 }
1300                 if ((char *) de > top)
1301                         return -ENOSPC;
1302         }
1303         BUFFER_TRACE(bh, "get_write_access");
1304         err = ext4_journal_get_write_access(handle, bh);
1305         if (err) {
1306                 ext4_std_error(dir->i_sb, err);
1307                 brelse(bh);
1308                 return err;
1309         }
1310
1311         /* By now the buffer is marked for journaling */
1312         nlen = EXT4_DIR_REC_LEN(de->name_len);
1313         rlen = ext4_rec_len_from_disk(de->rec_len);
1314         if (de->inode) {
1315                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1316                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1317                 de->rec_len = ext4_rec_len_to_disk(nlen);
1318                 de = de1;
1319         }
1320         de->file_type = EXT4_FT_UNKNOWN;
1321         if (inode) {
1322                 de->inode = cpu_to_le32(inode->i_ino);
1323                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1324         } else
1325                 de->inode = 0;
1326         de->name_len = namelen;
1327         memcpy(de->name, name, namelen);
1328         /*
1329          * XXX shouldn't update any times until successful
1330          * completion of syscall, but too many callers depend
1331          * on this.
1332          *
1333          * XXX similarly, too many callers depend on
1334          * ext4_new_inode() setting the times, but error
1335          * recovery deletes the inode, so the worst that can
1336          * happen is that the times are slightly out of date
1337          * and/or different from the directory change time.
1338          */
1339         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1340         ext4_update_dx_flag(dir);
1341         dir->i_version++;
1342         ext4_mark_inode_dirty(handle, dir);
1343         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1344         err = ext4_journal_dirty_metadata(handle, bh);
1345         if (err)
1346                 ext4_std_error(dir->i_sb, err);
1347         brelse(bh);
1348         return 0;
1349 }
1350
1351 /*
1352  * This converts a one block unindexed directory to a 3 block indexed
1353  * directory, and adds the dentry to the indexed directory.
1354  */
1355 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1356                             struct inode *inode, struct buffer_head *bh)
1357 {
1358         struct inode    *dir = dentry->d_parent->d_inode;
1359         const char      *name = dentry->d_name.name;
1360         int             namelen = dentry->d_name.len;
1361         struct buffer_head *bh2;
1362         struct dx_root  *root;
1363         struct dx_frame frames[2], *frame;
1364         struct dx_entry *entries;
1365         struct ext4_dir_entry_2 *de, *de2;
1366         char            *data1, *top;
1367         unsigned        len;
1368         int             retval;
1369         unsigned        blocksize;
1370         struct dx_hash_info hinfo;
1371         ext4_lblk_t  block;
1372         struct fake_dirent *fde;
1373
1374         blocksize =  dir->i_sb->s_blocksize;
1375         dxtrace(printk(KERN_DEBUG "Creating index\n"));
1376         retval = ext4_journal_get_write_access(handle, bh);
1377         if (retval) {
1378                 ext4_std_error(dir->i_sb, retval);
1379                 brelse(bh);
1380                 return retval;
1381         }
1382         root = (struct dx_root *) bh->b_data;
1383
1384         bh2 = ext4_append(handle, dir, &block, &retval);
1385         if (!(bh2)) {
1386                 brelse(bh);
1387                 return retval;
1388         }
1389         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1390         data1 = bh2->b_data;
1391
1392         /* The 0th block becomes the root, move the dirents out */
1393         fde = &root->dotdot;
1394         de = (struct ext4_dir_entry_2 *)((char *)fde +
1395                 ext4_rec_len_from_disk(fde->rec_len));
1396         len = ((char *) root) + blocksize - (char *) de;
1397         memcpy (data1, de, len);
1398         de = (struct ext4_dir_entry_2 *) data1;
1399         top = data1 + len;
1400         while ((char *)(de2 = ext4_next_entry(de)) < top)
1401                 de = de2;
1402         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1403         /* Initialize the root; the dot dirents already exist */
1404         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1405         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
1406         memset (&root->info, 0, sizeof(root->info));
1407         root->info.info_length = sizeof(root->info);
1408         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1409         entries = root->entries;
1410         dx_set_block(entries, 1);
1411         dx_set_count(entries, 1);
1412         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1413
1414         /* Initialize as for dx_probe */
1415         hinfo.hash_version = root->info.hash_version;
1416         if (hinfo.hash_version <= DX_HASH_TEA)
1417                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1418         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1419         ext4fs_dirhash(name, namelen, &hinfo);
1420         frame = frames;
1421         frame->entries = entries;
1422         frame->at = entries;
1423         frame->bh = bh;
1424         bh = bh2;
1425         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1426         dx_release (frames);
1427         if (!(de))
1428                 return retval;
1429
1430         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1431 }
1432
1433 /*
1434  *      ext4_add_entry()
1435  *
1436  * adds a file entry to the specified directory, using the same
1437  * semantics as ext4_find_entry(). It returns NULL if it failed.
1438  *
1439  * NOTE!! The inode part of 'de' is left at 0 - which means you
1440  * may not sleep between calling this and putting something into
1441  * the entry, as someone else might have used it while you slept.
1442  */
1443 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1444                           struct inode *inode)
1445 {
1446         struct inode *dir = dentry->d_parent->d_inode;
1447         unsigned long offset;
1448         struct buffer_head *bh;
1449         struct ext4_dir_entry_2 *de;
1450         struct super_block *sb;
1451         int     retval;
1452         int     dx_fallback=0;
1453         unsigned blocksize;
1454         ext4_lblk_t block, blocks;
1455
1456         sb = dir->i_sb;
1457         blocksize = sb->s_blocksize;
1458         if (!dentry->d_name.len)
1459                 return -EINVAL;
1460         if (is_dx(dir)) {
1461                 retval = ext4_dx_add_entry(handle, dentry, inode);
1462                 if (!retval || (retval != ERR_BAD_DX_DIR))
1463                         return retval;
1464                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1465                 dx_fallback++;
1466                 ext4_mark_inode_dirty(handle, dir);
1467         }
1468         blocks = dir->i_size >> sb->s_blocksize_bits;
1469         for (block = 0, offset = 0; block < blocks; block++) {
1470                 bh = ext4_bread(handle, dir, block, 0, &retval);
1471                 if(!bh)
1472                         return retval;
1473                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1474                 if (retval != -ENOSPC)
1475                         return retval;
1476
1477                 if (blocks == 1 && !dx_fallback &&
1478                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1479                         return make_indexed_dir(handle, dentry, inode, bh);
1480                 brelse(bh);
1481         }
1482         bh = ext4_append(handle, dir, &block, &retval);
1483         if (!bh)
1484                 return retval;
1485         de = (struct ext4_dir_entry_2 *) bh->b_data;
1486         de->inode = 0;
1487         de->rec_len = ext4_rec_len_to_disk(blocksize);
1488         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1489 }
1490
1491 /*
1492  * Returns 0 for success, or a negative error value
1493  */
1494 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1495                              struct inode *inode)
1496 {
1497         struct dx_frame frames[2], *frame;
1498         struct dx_entry *entries, *at;
1499         struct dx_hash_info hinfo;
1500         struct buffer_head *bh;
1501         struct inode *dir = dentry->d_parent->d_inode;
1502         struct super_block *sb = dir->i_sb;
1503         struct ext4_dir_entry_2 *de;
1504         int err;
1505
1506         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1507         if (!frame)
1508                 return err;
1509         entries = frame->entries;
1510         at = frame->at;
1511
1512         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1513                 goto cleanup;
1514
1515         BUFFER_TRACE(bh, "get_write_access");
1516         err = ext4_journal_get_write_access(handle, bh);
1517         if (err)
1518                 goto journal_error;
1519
1520         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1521         if (err != -ENOSPC) {
1522                 bh = NULL;
1523                 goto cleanup;
1524         }
1525
1526         /* Block full, should compress but for now just split */
1527         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1528                        dx_get_count(entries), dx_get_limit(entries)));
1529         /* Need to split index? */
1530         if (dx_get_count(entries) == dx_get_limit(entries)) {
1531                 ext4_lblk_t newblock;
1532                 unsigned icount = dx_get_count(entries);
1533                 int levels = frame - frames;
1534                 struct dx_entry *entries2;
1535                 struct dx_node *node2;
1536                 struct buffer_head *bh2;
1537
1538                 if (levels && (dx_get_count(frames->entries) ==
1539                                dx_get_limit(frames->entries))) {
1540                         ext4_warning(sb, __func__,
1541                                      "Directory index full!");
1542                         err = -ENOSPC;
1543                         goto cleanup;
1544                 }
1545                 bh2 = ext4_append (handle, dir, &newblock, &err);
1546                 if (!(bh2))
1547                         goto cleanup;
1548                 node2 = (struct dx_node *)(bh2->b_data);
1549                 entries2 = node2->entries;
1550                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
1551                 node2->fake.inode = 0;
1552                 BUFFER_TRACE(frame->bh, "get_write_access");
1553                 err = ext4_journal_get_write_access(handle, frame->bh);
1554                 if (err)
1555                         goto journal_error;
1556                 if (levels) {
1557                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1558                         unsigned hash2 = dx_get_hash(entries + icount1);
1559                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1560                                        icount1, icount2));
1561
1562                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1563                         err = ext4_journal_get_write_access(handle,
1564                                                              frames[0].bh);
1565                         if (err)
1566                                 goto journal_error;
1567
1568                         memcpy((char *) entries2, (char *) (entries + icount1),
1569                                icount2 * sizeof(struct dx_entry));
1570                         dx_set_count(entries, icount1);
1571                         dx_set_count(entries2, icount2);
1572                         dx_set_limit(entries2, dx_node_limit(dir));
1573
1574                         /* Which index block gets the new entry? */
1575                         if (at - entries >= icount1) {
1576                                 frame->at = at = at - entries - icount1 + entries2;
1577                                 frame->entries = entries = entries2;
1578                                 swap(frame->bh, bh2);
1579                         }
1580                         dx_insert_block(frames + 0, hash2, newblock);
1581                         dxtrace(dx_show_index("node", frames[1].entries));
1582                         dxtrace(dx_show_index("node",
1583                                ((struct dx_node *) bh2->b_data)->entries));
1584                         err = ext4_journal_dirty_metadata(handle, bh2);
1585                         if (err)
1586                                 goto journal_error;
1587                         brelse (bh2);
1588                 } else {
1589                         dxtrace(printk(KERN_DEBUG
1590                                        "Creating second level index...\n"));
1591                         memcpy((char *) entries2, (char *) entries,
1592                                icount * sizeof(struct dx_entry));
1593                         dx_set_limit(entries2, dx_node_limit(dir));
1594
1595                         /* Set up root */
1596                         dx_set_count(entries, 1);
1597                         dx_set_block(entries + 0, newblock);
1598                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1599
1600                         /* Add new access path frame */
1601                         frame = frames + 1;
1602                         frame->at = at = at - entries + entries2;
1603                         frame->entries = entries = entries2;
1604                         frame->bh = bh2;
1605                         err = ext4_journal_get_write_access(handle,
1606                                                              frame->bh);
1607                         if (err)
1608                                 goto journal_error;
1609                 }
1610                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1611         }
1612         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1613         if (!de)
1614                 goto cleanup;
1615         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1616         bh = NULL;
1617         goto cleanup;
1618
1619 journal_error:
1620         ext4_std_error(dir->i_sb, err);
1621 cleanup:
1622         if (bh)
1623                 brelse(bh);
1624         dx_release(frames);
1625         return err;
1626 }
1627
1628 /*
1629  * ext4_delete_entry deletes a directory entry by merging it with the
1630  * previous entry
1631  */
1632 static int ext4_delete_entry(handle_t *handle,
1633                              struct inode *dir,
1634                              struct ext4_dir_entry_2 *de_del,
1635                              struct buffer_head *bh)
1636 {
1637         struct ext4_dir_entry_2 *de, *pde;
1638         int i;
1639
1640         i = 0;
1641         pde = NULL;
1642         de = (struct ext4_dir_entry_2 *) bh->b_data;
1643         while (i < bh->b_size) {
1644                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1645                         return -EIO;
1646                 if (de == de_del)  {
1647                         BUFFER_TRACE(bh, "get_write_access");
1648                         ext4_journal_get_write_access(handle, bh);
1649                         if (pde)
1650                                 pde->rec_len = ext4_rec_len_to_disk(
1651                                         ext4_rec_len_from_disk(pde->rec_len) +
1652                                         ext4_rec_len_from_disk(de->rec_len));
1653                         else
1654                                 de->inode = 0;
1655                         dir->i_version++;
1656                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1657                         ext4_journal_dirty_metadata(handle, bh);
1658                         return 0;
1659                 }
1660                 i += ext4_rec_len_from_disk(de->rec_len);
1661                 pde = de;
1662                 de = ext4_next_entry(de);
1663         }
1664         return -ENOENT;
1665 }
1666
1667 /*
1668  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1669  * since this indicates that nlinks count was previously 1.
1670  */
1671 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1672 {
1673         inc_nlink(inode);
1674         if (is_dx(inode) && inode->i_nlink > 1) {
1675                 /* limit is 16-bit i_links_count */
1676                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1677                         inode->i_nlink = 1;
1678                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1679                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1680                 }
1681         }
1682 }
1683
1684 /*
1685  * If a directory had nlink == 1, then we should let it be 1. This indicates
1686  * directory has >EXT4_LINK_MAX subdirs.
1687  */
1688 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1689 {
1690         drop_nlink(inode);
1691         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1692                 inc_nlink(inode);
1693 }
1694
1695
1696 static int ext4_add_nondir(handle_t *handle,
1697                 struct dentry *dentry, struct inode *inode)
1698 {
1699         int err = ext4_add_entry(handle, dentry, inode);
1700         if (!err) {
1701                 ext4_mark_inode_dirty(handle, inode);
1702                 d_instantiate(dentry, inode);
1703                 unlock_new_inode(inode);
1704                 return 0;
1705         }
1706         drop_nlink(inode);
1707         unlock_new_inode(inode);
1708         iput(inode);
1709         return err;
1710 }
1711
1712 /*
1713  * By the time this is called, we already have created
1714  * the directory cache entry for the new file, but it
1715  * is so far negative - it has no inode.
1716  *
1717  * If the create succeeds, we fill in the inode information
1718  * with d_instantiate().
1719  */
1720 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1721                        struct nameidata *nd)
1722 {
1723         handle_t *handle;
1724         struct inode *inode;
1725         int err, retries = 0;
1726
1727 retry:
1728         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1729                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1730                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1731         if (IS_ERR(handle))
1732                 return PTR_ERR(handle);
1733
1734         if (IS_DIRSYNC(dir))
1735                 handle->h_sync = 1;
1736
1737         inode = ext4_new_inode (handle, dir, mode);
1738         err = PTR_ERR(inode);
1739         if (!IS_ERR(inode)) {
1740                 inode->i_op = &ext4_file_inode_operations;
1741                 inode->i_fop = &ext4_file_operations;
1742                 ext4_set_aops(inode);
1743                 err = ext4_add_nondir(handle, dentry, inode);
1744         }
1745         ext4_journal_stop(handle);
1746         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1747                 goto retry;
1748         return err;
1749 }
1750
1751 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1752                       int mode, dev_t rdev)
1753 {
1754         handle_t *handle;
1755         struct inode *inode;
1756         int err, retries = 0;
1757
1758         if (!new_valid_dev(rdev))
1759                 return -EINVAL;
1760
1761 retry:
1762         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1763                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1764                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1765         if (IS_ERR(handle))
1766                 return PTR_ERR(handle);
1767
1768         if (IS_DIRSYNC(dir))
1769                 handle->h_sync = 1;
1770
1771         inode = ext4_new_inode(handle, dir, mode);
1772         err = PTR_ERR(inode);
1773         if (!IS_ERR(inode)) {
1774                 init_special_inode(inode, inode->i_mode, rdev);
1775 #ifdef CONFIG_EXT4_FS_XATTR
1776                 inode->i_op = &ext4_special_inode_operations;
1777 #endif
1778                 err = ext4_add_nondir(handle, dentry, inode);
1779         }
1780         ext4_journal_stop(handle);
1781         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1782                 goto retry;
1783         return err;
1784 }
1785
1786 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1787 {
1788         handle_t *handle;
1789         struct inode *inode;
1790         struct buffer_head *dir_block;
1791         struct ext4_dir_entry_2 *de;
1792         int err, retries = 0;
1793
1794         if (EXT4_DIR_LINK_MAX(dir))
1795                 return -EMLINK;
1796
1797 retry:
1798         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1799                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1800                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1801         if (IS_ERR(handle))
1802                 return PTR_ERR(handle);
1803
1804         if (IS_DIRSYNC(dir))
1805                 handle->h_sync = 1;
1806
1807         inode = ext4_new_inode(handle, dir, S_IFDIR | mode);
1808         err = PTR_ERR(inode);
1809         if (IS_ERR(inode))
1810                 goto out_stop;
1811
1812         inode->i_op = &ext4_dir_inode_operations;
1813         inode->i_fop = &ext4_dir_operations;
1814         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1815         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1816         if (!dir_block)
1817                 goto out_clear_inode;
1818         BUFFER_TRACE(dir_block, "get_write_access");
1819         ext4_journal_get_write_access(handle, dir_block);
1820         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1821         de->inode = cpu_to_le32(inode->i_ino);
1822         de->name_len = 1;
1823         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
1824         strcpy(de->name, ".");
1825         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1826         de = ext4_next_entry(de);
1827         de->inode = cpu_to_le32(dir->i_ino);
1828         de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1829                                                 EXT4_DIR_REC_LEN(1));
1830         de->name_len = 2;
1831         strcpy(de->name, "..");
1832         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1833         inode->i_nlink = 2;
1834         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1835         ext4_journal_dirty_metadata(handle, dir_block);
1836         brelse(dir_block);
1837         ext4_mark_inode_dirty(handle, inode);
1838         err = ext4_add_entry(handle, dentry, inode);
1839         if (err) {
1840 out_clear_inode:
1841                 clear_nlink(inode);
1842                 unlock_new_inode(inode);
1843                 ext4_mark_inode_dirty(handle, inode);
1844                 iput(inode);
1845                 goto out_stop;
1846         }
1847         ext4_inc_count(handle, dir);
1848         ext4_update_dx_flag(dir);
1849         ext4_mark_inode_dirty(handle, dir);
1850         d_instantiate(dentry, inode);
1851         unlock_new_inode(inode);
1852 out_stop:
1853         ext4_journal_stop(handle);
1854         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1855                 goto retry;
1856         return err;
1857 }
1858
1859 /*
1860  * routine to check that the specified directory is empty (for rmdir)
1861  */
1862 static int empty_dir(struct inode *inode)
1863 {
1864         unsigned long offset;
1865         struct buffer_head *bh;
1866         struct ext4_dir_entry_2 *de, *de1;
1867         struct super_block *sb;
1868         int err = 0;
1869
1870         sb = inode->i_sb;
1871         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1872             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1873                 if (err)
1874                         ext4_error(inode->i_sb, __func__,
1875                                    "error %d reading directory #%lu offset 0",
1876                                    err, inode->i_ino);
1877                 else
1878                         ext4_warning(inode->i_sb, __func__,
1879                                      "bad directory (dir #%lu) - no data block",
1880                                      inode->i_ino);
1881                 return 1;
1882         }
1883         de = (struct ext4_dir_entry_2 *) bh->b_data;
1884         de1 = ext4_next_entry(de);
1885         if (le32_to_cpu(de->inode) != inode->i_ino ||
1886                         !le32_to_cpu(de1->inode) ||
1887                         strcmp(".", de->name) ||
1888                         strcmp("..", de1->name)) {
1889                 ext4_warning(inode->i_sb, "empty_dir",
1890                              "bad directory (dir #%lu) - no `.' or `..'",
1891                              inode->i_ino);
1892                 brelse(bh);
1893                 return 1;
1894         }
1895         offset = ext4_rec_len_from_disk(de->rec_len) +
1896                  ext4_rec_len_from_disk(de1->rec_len);
1897         de = ext4_next_entry(de1);
1898         while (offset < inode->i_size) {
1899                 if (!bh ||
1900                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1901                         err = 0;
1902                         brelse(bh);
1903                         bh = ext4_bread(NULL, inode,
1904                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1905                         if (!bh) {
1906                                 if (err)
1907                                         ext4_error(sb, __func__,
1908                                                    "error %d reading directory"
1909                                                    " #%lu offset %lu",
1910                                                    err, inode->i_ino, offset);
1911                                 offset += sb->s_blocksize;
1912                                 continue;
1913                         }
1914                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1915                 }
1916                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1917                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1918                                                          sb->s_blocksize);
1919                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1920                         continue;
1921                 }
1922                 if (le32_to_cpu(de->inode)) {
1923                         brelse(bh);
1924                         return 0;
1925                 }
1926                 offset += ext4_rec_len_from_disk(de->rec_len);
1927                 de = ext4_next_entry(de);
1928         }
1929         brelse(bh);
1930         return 1;
1931 }
1932
1933 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1934  * such inodes, starting at the superblock, in case we crash before the
1935  * file is closed/deleted, or in case the inode truncate spans multiple
1936  * transactions and the last transaction is not recovered after a crash.
1937  *
1938  * At filesystem recovery time, we walk this list deleting unlinked
1939  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1940  */
1941 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1942 {
1943         struct super_block *sb = inode->i_sb;
1944         struct ext4_iloc iloc;
1945         int err = 0, rc;
1946
1947         lock_super(sb);
1948         if (!list_empty(&EXT4_I(inode)->i_orphan))
1949                 goto out_unlock;
1950
1951         /* Orphan handling is only valid for files with data blocks
1952          * being truncated, or files being unlinked. */
1953
1954         /* @@@ FIXME: Observation from aviro:
1955          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1956          * here (on lock_super()), so race with ext4_link() which might bump
1957          * ->i_nlink. For, say it, character device. Not a regular file,
1958          * not a directory, not a symlink and ->i_nlink > 0.
1959          */
1960         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1961                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1962
1963         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1964         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1965         if (err)
1966                 goto out_unlock;
1967
1968         err = ext4_reserve_inode_write(handle, inode, &iloc);
1969         if (err)
1970                 goto out_unlock;
1971
1972         /* Insert this inode at the head of the on-disk orphan list... */
1973         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1974         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1975         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1976         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1977         if (!err)
1978                 err = rc;
1979
1980         /* Only add to the head of the in-memory list if all the
1981          * previous operations succeeded.  If the orphan_add is going to
1982          * fail (possibly taking the journal offline), we can't risk
1983          * leaving the inode on the orphan list: stray orphan-list
1984          * entries can cause panics at unmount time.
1985          *
1986          * This is safe: on error we're going to ignore the orphan list
1987          * anyway on the next recovery. */
1988         if (!err)
1989                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1990
1991         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1992         jbd_debug(4, "orphan inode %lu will point to %d\n",
1993                         inode->i_ino, NEXT_ORPHAN(inode));
1994 out_unlock:
1995         unlock_super(sb);
1996         ext4_std_error(inode->i_sb, err);
1997         return err;
1998 }
1999
2000 /*
2001  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2002  * of such inodes stored on disk, because it is finally being cleaned up.
2003  */
2004 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2005 {
2006         struct list_head *prev;
2007         struct ext4_inode_info *ei = EXT4_I(inode);
2008         struct ext4_sb_info *sbi;
2009         unsigned long ino_next;
2010         struct ext4_iloc iloc;
2011         int err = 0;
2012
2013         lock_super(inode->i_sb);
2014         if (list_empty(&ei->i_orphan)) {
2015                 unlock_super(inode->i_sb);
2016                 return 0;
2017         }
2018
2019         ino_next = NEXT_ORPHAN(inode);
2020         prev = ei->i_orphan.prev;
2021         sbi = EXT4_SB(inode->i_sb);
2022
2023         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2024
2025         list_del_init(&ei->i_orphan);
2026
2027         /* If we're on an error path, we may not have a valid
2028          * transaction handle with which to update the orphan list on
2029          * disk, but we still need to remove the inode from the linked
2030          * list in memory. */
2031         if (!handle)
2032                 goto out;
2033
2034         err = ext4_reserve_inode_write(handle, inode, &iloc);
2035         if (err)
2036                 goto out_err;
2037
2038         if (prev == &sbi->s_orphan) {
2039                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2040                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2041                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2042                 if (err)
2043                         goto out_brelse;
2044                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2045                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2046         } else {
2047                 struct ext4_iloc iloc2;
2048                 struct inode *i_prev =
2049                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2050
2051                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2052                           i_prev->i_ino, ino_next);
2053                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2054                 if (err)
2055                         goto out_brelse;
2056                 NEXT_ORPHAN(i_prev) = ino_next;
2057                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2058         }
2059         if (err)
2060                 goto out_brelse;
2061         NEXT_ORPHAN(inode) = 0;
2062         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2063
2064 out_err:
2065         ext4_std_error(inode->i_sb, err);
2066 out:
2067         unlock_super(inode->i_sb);
2068         return err;
2069
2070 out_brelse:
2071         brelse(iloc.bh);
2072         goto out_err;
2073 }
2074
2075 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2076 {
2077         int retval;
2078         struct inode *inode;
2079         struct buffer_head *bh;
2080         struct ext4_dir_entry_2 *de;
2081         handle_t *handle;
2082
2083         /* Initialize quotas before so that eventual writes go in
2084          * separate transaction */
2085         DQUOT_INIT(dentry->d_inode);
2086         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2087         if (IS_ERR(handle))
2088                 return PTR_ERR(handle);
2089
2090         retval = -ENOENT;
2091         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2092         if (!bh)
2093                 goto end_rmdir;
2094
2095         if (IS_DIRSYNC(dir))
2096                 handle->h_sync = 1;
2097
2098         inode = dentry->d_inode;
2099
2100         retval = -EIO;
2101         if (le32_to_cpu(de->inode) != inode->i_ino)
2102                 goto end_rmdir;
2103
2104         retval = -ENOTEMPTY;
2105         if (!empty_dir(inode))
2106                 goto end_rmdir;
2107
2108         retval = ext4_delete_entry(handle, dir, de, bh);
2109         if (retval)
2110                 goto end_rmdir;
2111         if (!EXT4_DIR_LINK_EMPTY(inode))
2112                 ext4_warning(inode->i_sb, "ext4_rmdir",
2113                              "empty directory has too many links (%d)",
2114                              inode->i_nlink);
2115         inode->i_version++;
2116         clear_nlink(inode);
2117         /* There's no need to set i_disksize: the fact that i_nlink is
2118          * zero will ensure that the right thing happens during any
2119          * recovery. */
2120         inode->i_size = 0;
2121         ext4_orphan_add(handle, inode);
2122         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2123         ext4_mark_inode_dirty(handle, inode);
2124         ext4_dec_count(handle, dir);
2125         ext4_update_dx_flag(dir);
2126         ext4_mark_inode_dirty(handle, dir);
2127
2128 end_rmdir:
2129         ext4_journal_stop(handle);
2130         brelse(bh);
2131         return retval;
2132 }
2133
2134 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2135 {
2136         int retval;
2137         struct inode *inode;
2138         struct buffer_head *bh;
2139         struct ext4_dir_entry_2 *de;
2140         handle_t *handle;
2141
2142         /* Initialize quotas before so that eventual writes go
2143          * in separate transaction */
2144         DQUOT_INIT(dentry->d_inode);
2145         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2146         if (IS_ERR(handle))
2147                 return PTR_ERR(handle);
2148
2149         if (IS_DIRSYNC(dir))
2150                 handle->h_sync = 1;
2151
2152         retval = -ENOENT;
2153         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2154         if (!bh)
2155                 goto end_unlink;
2156
2157         inode = dentry->d_inode;
2158
2159         retval = -EIO;
2160         if (le32_to_cpu(de->inode) != inode->i_ino)
2161                 goto end_unlink;
2162
2163         if (!inode->i_nlink) {
2164                 ext4_warning(inode->i_sb, "ext4_unlink",
2165                              "Deleting nonexistent file (%lu), %d",
2166                              inode->i_ino, inode->i_nlink);
2167                 inode->i_nlink = 1;
2168         }
2169         retval = ext4_delete_entry(handle, dir, de, bh);
2170         if (retval)
2171                 goto end_unlink;
2172         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2173         ext4_update_dx_flag(dir);
2174         ext4_mark_inode_dirty(handle, dir);
2175         drop_nlink(inode);
2176         if (!inode->i_nlink)
2177                 ext4_orphan_add(handle, inode);
2178         inode->i_ctime = ext4_current_time(inode);
2179         ext4_mark_inode_dirty(handle, inode);
2180         retval = 0;
2181
2182 end_unlink:
2183         ext4_journal_stop(handle);
2184         brelse(bh);
2185         return retval;
2186 }
2187
2188 static int ext4_symlink(struct inode *dir,
2189                         struct dentry *dentry, const char *symname)
2190 {
2191         handle_t *handle;
2192         struct inode *inode;
2193         int l, err, retries = 0;
2194
2195         l = strlen(symname)+1;
2196         if (l > dir->i_sb->s_blocksize)
2197                 return -ENAMETOOLONG;
2198
2199 retry:
2200         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2201                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2202                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2203         if (IS_ERR(handle))
2204                 return PTR_ERR(handle);
2205
2206         if (IS_DIRSYNC(dir))
2207                 handle->h_sync = 1;
2208
2209         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO);
2210         err = PTR_ERR(inode);
2211         if (IS_ERR(inode))
2212                 goto out_stop;
2213
2214         if (l > sizeof(EXT4_I(inode)->i_data)) {
2215                 inode->i_op = &ext4_symlink_inode_operations;
2216                 ext4_set_aops(inode);
2217                 /*
2218                  * page_symlink() calls into ext4_prepare/commit_write.
2219                  * We have a transaction open.  All is sweetness.  It also sets
2220                  * i_size in generic_commit_write().
2221                  */
2222                 err = __page_symlink(inode, symname, l, 1);
2223                 if (err) {
2224                         clear_nlink(inode);
2225                         unlock_new_inode(inode);
2226                         ext4_mark_inode_dirty(handle, inode);
2227                         iput(inode);
2228                         goto out_stop;
2229                 }
2230         } else {
2231                 /* clear the extent format for fast symlink */
2232                 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2233                 inode->i_op = &ext4_fast_symlink_inode_operations;
2234                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2235                 inode->i_size = l-1;
2236         }
2237         EXT4_I(inode)->i_disksize = inode->i_size;
2238         err = ext4_add_nondir(handle, dentry, inode);
2239 out_stop:
2240         ext4_journal_stop(handle);
2241         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2242                 goto retry;
2243         return err;
2244 }
2245
2246 static int ext4_link(struct dentry *old_dentry,
2247                      struct inode *dir, struct dentry *dentry)
2248 {
2249         handle_t *handle;
2250         struct inode *inode = old_dentry->d_inode;
2251         int err, retries = 0;
2252
2253         if (EXT4_DIR_LINK_MAX(inode))
2254                 return -EMLINK;
2255
2256         /*
2257          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2258          * otherwise has the potential to corrupt the orphan inode list.
2259          */
2260         if (inode->i_nlink == 0)
2261                 return -ENOENT;
2262
2263 retry:
2264         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2265                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2266         if (IS_ERR(handle))
2267                 return PTR_ERR(handle);
2268
2269         if (IS_DIRSYNC(dir))
2270                 handle->h_sync = 1;
2271
2272         inode->i_ctime = ext4_current_time(inode);
2273         ext4_inc_count(handle, inode);
2274         atomic_inc(&inode->i_count);
2275
2276         err = ext4_add_entry(handle, dentry, inode);
2277         if (!err) {
2278                 ext4_mark_inode_dirty(handle, inode);
2279                 d_instantiate(dentry, inode);
2280         } else {
2281                 drop_nlink(inode);
2282                 iput(inode);
2283         }
2284         ext4_journal_stop(handle);
2285         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2286                 goto retry;
2287         return err;
2288 }
2289
2290 #define PARENT_INO(buffer) \
2291         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2292
2293 /*
2294  * Anybody can rename anything with this: the permission checks are left to the
2295  * higher-level routines.
2296  */
2297 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2298                        struct inode *new_dir, struct dentry *new_dentry)
2299 {
2300         handle_t *handle;
2301         struct inode *old_inode, *new_inode;
2302         struct buffer_head *old_bh, *new_bh, *dir_bh;
2303         struct ext4_dir_entry_2 *old_de, *new_de;
2304         int retval;
2305
2306         old_bh = new_bh = dir_bh = NULL;
2307
2308         /* Initialize quotas before so that eventual writes go
2309          * in separate transaction */
2310         if (new_dentry->d_inode)
2311                 DQUOT_INIT(new_dentry->d_inode);
2312         handle = ext4_journal_start(old_dir, 2 *
2313                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2314                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2315         if (IS_ERR(handle))
2316                 return PTR_ERR(handle);
2317
2318         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2319                 handle->h_sync = 1;
2320
2321         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2322         /*
2323          *  Check for inode number is _not_ due to possible IO errors.
2324          *  We might rmdir the source, keep it as pwd of some process
2325          *  and merrily kill the link to whatever was created under the
2326          *  same name. Goodbye sticky bit ;-<
2327          */
2328         old_inode = old_dentry->d_inode;
2329         retval = -ENOENT;
2330         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2331                 goto end_rename;
2332
2333         new_inode = new_dentry->d_inode;
2334         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2335         if (new_bh) {
2336                 if (!new_inode) {
2337                         brelse(new_bh);
2338                         new_bh = NULL;
2339                 }
2340         }
2341         if (S_ISDIR(old_inode->i_mode)) {
2342                 if (new_inode) {
2343                         retval = -ENOTEMPTY;
2344                         if (!empty_dir(new_inode))
2345                                 goto end_rename;
2346                 }
2347                 retval = -EIO;
2348                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2349                 if (!dir_bh)
2350                         goto end_rename;
2351                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2352                         goto end_rename;
2353                 retval = -EMLINK;
2354                 if (!new_inode && new_dir != old_dir &&
2355                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2356                         goto end_rename;
2357         }
2358         if (!new_bh) {
2359                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2360                 if (retval)
2361                         goto end_rename;
2362         } else {
2363                 BUFFER_TRACE(new_bh, "get write access");
2364                 ext4_journal_get_write_access(handle, new_bh);
2365                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2366                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2367                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2368                         new_de->file_type = old_de->file_type;
2369                 new_dir->i_version++;
2370                 new_dir->i_ctime = new_dir->i_mtime =
2371                                         ext4_current_time(new_dir);
2372                 ext4_mark_inode_dirty(handle, new_dir);
2373                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2374                 ext4_journal_dirty_metadata(handle, new_bh);
2375                 brelse(new_bh);
2376                 new_bh = NULL;
2377         }
2378
2379         /*
2380          * Like most other Unix systems, set the ctime for inodes on a
2381          * rename.
2382          */
2383         old_inode->i_ctime = ext4_current_time(old_inode);
2384         ext4_mark_inode_dirty(handle, old_inode);
2385
2386         /*
2387          * ok, that's it
2388          */
2389         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2390             old_de->name_len != old_dentry->d_name.len ||
2391             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2392             (retval = ext4_delete_entry(handle, old_dir,
2393                                         old_de, old_bh)) == -ENOENT) {
2394                 /* old_de could have moved from under us during htree split, so
2395                  * make sure that we are deleting the right entry.  We might
2396                  * also be pointing to a stale entry in the unused part of
2397                  * old_bh so just checking inum and the name isn't enough. */
2398                 struct buffer_head *old_bh2;
2399                 struct ext4_dir_entry_2 *old_de2;
2400
2401                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2402                 if (old_bh2) {
2403                         retval = ext4_delete_entry(handle, old_dir,
2404                                                    old_de2, old_bh2);
2405                         brelse(old_bh2);
2406                 }
2407         }
2408         if (retval) {
2409                 ext4_warning(old_dir->i_sb, "ext4_rename",
2410                                 "Deleting old file (%lu), %d, error=%d",
2411                                 old_dir->i_ino, old_dir->i_nlink, retval);
2412         }
2413
2414         if (new_inode) {
2415                 ext4_dec_count(handle, new_inode);
2416                 new_inode->i_ctime = ext4_current_time(new_inode);
2417         }
2418         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2419         ext4_update_dx_flag(old_dir);
2420         if (dir_bh) {
2421                 BUFFER_TRACE(dir_bh, "get_write_access");
2422                 ext4_journal_get_write_access(handle, dir_bh);
2423                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2424                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2425                 ext4_journal_dirty_metadata(handle, dir_bh);
2426                 ext4_dec_count(handle, old_dir);
2427                 if (new_inode) {
2428                         /* checked empty_dir above, can't have another parent,
2429                          * ext4_dec_count() won't work for many-linked dirs */
2430                         new_inode->i_nlink = 0;
2431                 } else {
2432                         ext4_inc_count(handle, new_dir);
2433                         ext4_update_dx_flag(new_dir);
2434                         ext4_mark_inode_dirty(handle, new_dir);
2435                 }
2436         }
2437         ext4_mark_inode_dirty(handle, old_dir);
2438         if (new_inode) {
2439                 ext4_mark_inode_dirty(handle, new_inode);
2440                 if (!new_inode->i_nlink)
2441                         ext4_orphan_add(handle, new_inode);
2442         }
2443         retval = 0;
2444
2445 end_rename:
2446         brelse(dir_bh);
2447         brelse(old_bh);
2448         brelse(new_bh);
2449         ext4_journal_stop(handle);
2450         return retval;
2451 }
2452
2453 /*
2454  * directories can handle most operations...
2455  */
2456 const struct inode_operations ext4_dir_inode_operations = {
2457         .create         = ext4_create,
2458         .lookup         = ext4_lookup,
2459         .link           = ext4_link,
2460         .unlink         = ext4_unlink,
2461         .symlink        = ext4_symlink,
2462         .mkdir          = ext4_mkdir,
2463         .rmdir          = ext4_rmdir,
2464         .mknod          = ext4_mknod,
2465         .rename         = ext4_rename,
2466         .setattr        = ext4_setattr,
2467 #ifdef CONFIG_EXT4_FS_XATTR
2468         .setxattr       = generic_setxattr,
2469         .getxattr       = generic_getxattr,
2470         .listxattr      = ext4_listxattr,
2471         .removexattr    = generic_removexattr,
2472 #endif
2473         .permission     = ext4_permission,
2474 };
2475
2476 const struct inode_operations ext4_special_inode_operations = {
2477         .setattr        = ext4_setattr,
2478 #ifdef CONFIG_EXT4_FS_XATTR
2479         .setxattr       = generic_setxattr,
2480         .getxattr       = generic_getxattr,
2481         .listxattr      = ext4_listxattr,
2482         .removexattr    = generic_removexattr,
2483 #endif
2484         .permission     = ext4_permission,
2485 };