fs/ext4/namei.c

   1 /*
   2  *  linux/fs/ext4/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *  Hash Tree Directory indexing (c)
  20  *      Daniel Phillips, 2001
  21  *  Hash Tree Directory indexing porting
  22  *      Christopher Li, 2002
  23  *  Hash Tree Directory indexing cleanup
  24  *      Theodore Ts'o, 2002
  25  */
  26
  27 #include <linux/fs.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/jbd2.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include "ext4.h"
  38 #include "ext4_jbd2.h"
  39
  40 #include "xattr.h"
  41 #include "acl.h"
  42
  43 #include <trace/events/ext4.h>
  44 /*
  45  * define how far ahead to read directories while searching them.
  46  */
  47 #define NAMEI_RA_CHUNKS  2
  48 #define NAMEI_RA_BLOCKS  4
  49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52 static struct buffer_head *ext4_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         ext4_lblk_t *block, int *err)
  55 {
  56         struct buffer_head *bh;
  57
  58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  59
  60         bh = ext4_bread(handle, inode, *block, 1, err);
  61         if (bh) {
  62                 inode->i_size += inode->i_sb->s_blocksize;
  63                 EXT4_I(inode)->i_disksize = inode->i_size;
  64                 *err = ext4_journal_get_write_access(handle, bh);
  65                 if (*err) {
  66                         brelse(bh);
  67                         bh = NULL;
  68                 }
  69         }
  70         return bh;
  71 }
  72
  73 #ifndef assert
  74 #define assert(test) J_ASSERT(test)
  75 #endif
  76
  77 #ifdef DX_DEBUG
  78 #define dxtrace(command) command
  79 #else
  80 #define dxtrace(command)
  81 #endif
  82
  83 struct fake_dirent
  84 {
  85         __le32 inode;
  86         __le16 rec_len;
  87         u8 name_len;
  88         u8 file_type;
  89 };
  90
  91 struct dx_countlimit
  92 {
  93         __le16 limit;
  94         __le16 count;
  95 };
  96
  97 struct dx_entry
  98 {
  99         __le32 hash;
 100         __le32 block;
 101 };
 102
 103 /*
 104  * dx_root_info is laid out so that if it should somehow get overlaid by a
 105  * dirent the two low bits of the hash version will be zero.  Therefore, the
 106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 107  */
 108
 109 struct dx_root
 110 {
 111         struct fake_dirent dot;
 112         char dot_name[4];
 113         struct fake_dirent dotdot;
 114         char dotdot_name[4];
 115         struct dx_root_info
 116         {
 117                 __le32 reserved_zero;
 118                 u8 hash_version;
 119                 u8 info_length; /* 8 */
 120                 u8 indirect_levels;
 121                 u8 unused_flags;
 122         }
 123         info;
 124         struct dx_entry entries[0];
 125 };
 126
 127 struct dx_node
 128 {
 129         struct fake_dirent fake;
 130         struct dx_entry entries[0];
 131 };
 132
 133
 134 struct dx_frame
 135 {
 136         struct buffer_head *bh;
 137         struct dx_entry *entries;
 138         struct dx_entry *at;
 139 };
 140
 141 struct dx_map_entry
 142 {
 143         u32 hash;
 144         u16 offs;
 145         u16 size;
 146 };
 147
 148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 150 static inline unsigned dx_get_hash(struct dx_entry *entry);
 151 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 152 static unsigned dx_get_count(struct dx_entry *entries);
 153 static unsigned dx_get_limit(struct dx_entry *entries);
 154 static void dx_set_count(struct dx_entry *entries, unsigned value);
 155 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 156 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 157 static unsigned dx_node_limit(struct inode *dir);
 158 static struct dx_frame *dx_probe(const struct qstr *d_name,
 159                                  struct inode *dir,
 160                                  struct dx_hash_info *hinfo,
 161                                  struct dx_frame *frame,
 162                                  int *err);
 163 static void dx_release(struct dx_frame *frames);
 164 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 165                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 167 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 168                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 170 static void dx_insert_block(struct dx_frame *frame,
 171                                         u32 hash, ext4_lblk_t block);
 172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 173                                  struct dx_frame *frame,
 174                                  struct dx_frame *frames,
 175                                  __u32 *start_hash);
 176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 177                 const struct qstr *d_name,
 178                 struct ext4_dir_entry_2 **res_dir,
 179                 int *err);
 180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 181                              struct inode *inode);
 182
 183 /*
 184  * p is at least 6 bytes before the end of page
 185  */
 186 static inline struct ext4_dir_entry_2 *
 187 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 188 {
 189         return (struct ext4_dir_entry_2 *)((char *)p +
 190                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 191 }
 192
 193 /*
 194  * Future: use high four bits of block for coalesce-on-delete flags
 195  * Mask them off for now.
 196  */
 197
 198 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 199 {
 200         return le32_to_cpu(entry->block) & 0x00ffffff;
 201 }
 202
 203 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 204 {
 205         entry->block = cpu_to_le32(value);
 206 }
 207
 208 static inline unsigned dx_get_hash(struct dx_entry *entry)
 209 {
 210         return le32_to_cpu(entry->hash);
 211 }
 212
 213 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 214 {
 215         entry->hash = cpu_to_le32(value);
 216 }
 217
 218 static inline unsigned dx_get_count(struct dx_entry *entries)
 219 {
 220         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 221 }
 222
 223 static inline unsigned dx_get_limit(struct dx_entry *entries)
 224 {
 225         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 226 }
 227
 228 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 229 {
 230         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 231 }
 232
 233 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 234 {
 235         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 236 }
 237
 238 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 239 {
 240         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 241                 EXT4_DIR_REC_LEN(2) - infosize;
 242         return entry_space / sizeof(struct dx_entry);
 243 }
 244
 245 static inline unsigned dx_node_limit(struct inode *dir)
 246 {
 247         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 248         return entry_space / sizeof(struct dx_entry);
 249 }
 250
 251 /*
 252  * Debug
 253  */
 254 #ifdef DX_DEBUG
 255 static void dx_show_index(char * label, struct dx_entry *entries)
 256 {
 257         int i, n = dx_get_count (entries);
 258         printk(KERN_DEBUG "%s index ", label);
 259         for (i = 0; i < n; i++) {
 260                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 261                                 0, (unsigned long)dx_get_block(entries + i));
 262         }
 263         printk("\n");
 264 }
 265
 266 struct stats
 267 {
 268         unsigned names;
 269         unsigned space;
 270         unsigned bcount;
 271 };
 272
 273 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 274                                  int size, int show_names)
 275 {
 276         unsigned names = 0, space = 0;
 277         char *base = (char *) de;
 278         struct dx_hash_info h = *hinfo;
 279
 280         printk("names: ");
 281         while ((char *) de < base + size)
 282         {
 283                 if (de->inode)
 284                 {
 285                         if (show_names)
 286                         {
 287                                 int len = de->name_len;
 288                                 char *name = de->name;
 289                                 while (len--) printk("%c", *name++);
 290                                 ext4fs_dirhash(de->name, de->name_len, &h);
 291                                 printk(":%x.%u ", h.hash,
 292                                        (unsigned) ((char *) de - base));
 293                         }
 294                         space += EXT4_DIR_REC_LEN(de->name_len);
 295                         names++;
 296                 }
 297                 de = ext4_next_entry(de, size);
 298         }
 299         printk("(%i)\n", names);
 300         return (struct stats) { names, space, 1 };
 301 }
 302
 303 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 304                              struct dx_entry *entries, int levels)
 305 {
 306         unsigned blocksize = dir->i_sb->s_blocksize;
 307         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 308         unsigned bcount = 0;
 309         struct buffer_head *bh;
 310         int err;
 311         printk("%i indexed blocks...\n", count);
 312         for (i = 0; i < count; i++, entries++)
 313         {
 314                 ext4_lblk_t block = dx_get_block(entries);
 315                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 316                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 317                 struct stats stats;
 318                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 319                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 320                 stats = levels?
 321                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 322                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 323                 names += stats.names;
 324                 space += stats.space;
 325                 bcount += stats.bcount;
 326                 brelse(bh);
 327         }
 328         if (bcount)
 329                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 330                        levels ? "" : "   ", names, space/bcount,
 331                        (space/bcount)*100/blocksize);
 332         return (struct stats) { names, space, bcount};
 333 }
 334 #endif /* DX_DEBUG */
 335
 336 /*
 337  * Probe for a directory leaf block to search.
 338  *
 339  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 340  * error in the directory index, and the caller should fall back to
 341  * searching the directory normally.  The callers of dx_probe **MUST**
 342  * check for this error code, and make sure it never gets reflected
 343  * back to userspace.
 344  */
 345 static struct dx_frame *
 346 dx_probe(const struct qstr *d_name, struct inode *dir,
 347          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 348 {
 349         unsigned count, indirect;
 350         struct dx_entry *at, *entries, *p, *q, *m;
 351         struct dx_root *root;
 352         struct buffer_head *bh;
 353         struct dx_frame *frame = frame_in;
 354         u32 hash;
 355
 356         frame->bh = NULL;
 357         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 358                 goto fail;
 359         root = (struct dx_root *) bh->b_data;
 360         if (root->info.hash_version != DX_HASH_TEA &&
 361             root->info.hash_version != DX_HASH_HALF_MD4 &&
 362             root->info.hash_version != DX_HASH_LEGACY) {
 363                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 364                              root->info.hash_version);
 365                 brelse(bh);
 366                 *err = ERR_BAD_DX_DIR;
 367                 goto fail;
 368         }
 369         hinfo->hash_version = root->info.hash_version;
 370         if (hinfo->hash_version <= DX_HASH_TEA)
 371                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 372         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 373         if (d_name)
 374                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 375         hash = hinfo->hash;
 376
 377         if (root->info.unused_flags & 1) {
 378                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 379                              root->info.unused_flags);
 380                 brelse(bh);
 381                 *err = ERR_BAD_DX_DIR;
 382                 goto fail;
 383         }
 384
 385         if ((indirect = root->info.indirect_levels) > 1) {
 386                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 387                              root->info.indirect_levels);
 388                 brelse(bh);
 389                 *err = ERR_BAD_DX_DIR;
 390                 goto fail;
 391         }
 392
 393         entries = (struct dx_entry *) (((char *)&root->info) +
 394                                        root->info.info_length);
 395
 396         if (dx_get_limit(entries) != dx_root_limit(dir,
 397                                                    root->info.info_length)) {
 398                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 399                 brelse(bh);
 400                 *err = ERR_BAD_DX_DIR;
 401                 goto fail;
 402         }
 403
 404         dxtrace(printk("Look up %x", hash));
 405         while (1)
 406         {
 407                 count = dx_get_count(entries);
 408                 if (!count || count > dx_get_limit(entries)) {
 409                         ext4_warning(dir->i_sb,
 410                                      "dx entry: no count or count > limit");
 411                         brelse(bh);
 412                         *err = ERR_BAD_DX_DIR;
 413                         goto fail2;
 414                 }
 415
 416                 p = entries + 1;
 417                 q = entries + count - 1;
 418                 while (p <= q)
 419                 {
 420                         m = p + (q - p)/2;
 421                         dxtrace(printk("."));
 422                         if (dx_get_hash(m) > hash)
 423                                 q = m - 1;
 424                         else
 425                                 p = m + 1;
 426                 }
 427
 428                 if (0) // linear search cross check
 429                 {
 430                         unsigned n = count - 1;
 431                         at = entries;
 432                         while (n--)
 433                         {
 434                                 dxtrace(printk(","));
 435                                 if (dx_get_hash(++at) > hash)
 436                                 {
 437                                         at--;
 438                                         break;
 439                                 }
 440                         }
 441                         assert (at == p - 1);
 442                 }
 443
 444                 at = p - 1;
 445                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 446                 frame->bh = bh;
 447                 frame->entries = entries;
 448                 frame->at = at;
 449                 if (!indirect--) return frame;
 450                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 451                         goto fail2;
 452                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 453                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 454                         ext4_warning(dir->i_sb,
 455                                      "dx entry: limit != node limit");
 456                         brelse(bh);
 457                         *err = ERR_BAD_DX_DIR;
 458                         goto fail2;
 459                 }
 460                 frame++;
 461                 frame->bh = NULL;
 462         }
 463 fail2:
 464         while (frame >= frame_in) {
 465                 brelse(frame->bh);
 466                 frame--;
 467         }
 468 fail:
 469         if (*err == ERR_BAD_DX_DIR)
 470                 ext4_warning(dir->i_sb,
 471                              "Corrupt dir inode %ld, running e2fsck is "
 472                              "recommended.", dir->i_ino);
 473         return NULL;
 474 }
 475
 476 static void dx_release (struct dx_frame *frames)
 477 {
 478         if (frames[0].bh == NULL)
 479                 return;
 480
 481         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 482                 brelse(frames[1].bh);
 483         brelse(frames[0].bh);
 484 }
 485
 486 /*
 487  * This function increments the frame pointer to search the next leaf
 488  * block, and reads in the necessary intervening nodes if the search
 489  * should be necessary.  Whether or not the search is necessary is
 490  * controlled by the hash parameter.  If the hash value is even, then
 491  * the search is only continued if the next block starts with that
 492  * hash value.  This is used if we are searching for a specific file.
 493  *
 494  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 495  *
 496  * This function returns 1 if the caller should continue to search,
 497  * or 0 if it should not.  If there is an error reading one of the
 498  * index blocks, it will a negative error code.
 499  *
 500  * If start_hash is non-null, it will be filled in with the starting
 501  * hash of the next page.
 502  */
 503 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 504                                  struct dx_frame *frame,
 505                                  struct dx_frame *frames,
 506                                  __u32 *start_hash)
 507 {
 508         struct dx_frame *p;
 509         struct buffer_head *bh;
 510         int err, num_frames = 0;
 511         __u32 bhash;
 512
 513         p = frame;
 514         /*
 515          * Find the next leaf page by incrementing the frame pointer.
 516          * If we run out of entries in the interior node, loop around and
 517          * increment pointer in the parent node.  When we break out of
 518          * this loop, num_frames indicates the number of interior
 519          * nodes need to be read.
 520          */
 521         while (1) {
 522                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 523                         break;
 524                 if (p == frames)
 525                         return 0;
 526                 num_frames++;
 527                 p--;
 528         }
 529
 530         /*
 531          * If the hash is 1, then continue only if the next page has a
 532          * continuation hash of any value.  This is used for readdir
 533          * handling.  Otherwise, check to see if the hash matches the
 534          * desired contiuation hash.  If it doesn't, return since
 535          * there's no point to read in the successive index pages.
 536          */
 537         bhash = dx_get_hash(p->at);
 538         if (start_hash)
 539                 *start_hash = bhash;
 540         if ((hash & 1) == 0) {
 541                 if ((bhash & ~1) != hash)
 542                         return 0;
 543         }
 544         /*
 545          * If the hash is HASH_NB_ALWAYS, we always go to the next
 546          * block so no check is necessary
 547          */
 548         while (num_frames--) {
 549                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 550                                       0, &err)))
 551                         return err; /* Failure */
 552                 p++;
 553                 brelse(p->bh);
 554                 p->bh = bh;
 555                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 556         }
 557         return 1;
 558 }
 559
 560
 561 /*
 562  * This function fills a red-black tree with information from a
 563  * directory block.  It returns the number directory entries loaded
 564  * into the tree.  If there is an error it is returned in err.
 565  */
 566 static int htree_dirblock_to_tree(struct file *dir_file,
 567                                   struct inode *dir, ext4_lblk_t block,
 568                                   struct dx_hash_info *hinfo,
 569                                   __u32 start_hash, __u32 start_minor_hash)
 570 {
 571         struct buffer_head *bh;
 572         struct ext4_dir_entry_2 *de, *top;
 573         int err, count = 0;
 574
 575         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 576                                                         (unsigned long)block));
 577         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 578                 return err;
 579
 580         de = (struct ext4_dir_entry_2 *) bh->b_data;
 581         top = (struct ext4_dir_entry_2 *) ((char *) de +
 582                                            dir->i_sb->s_blocksize -
 583                                            EXT4_DIR_REC_LEN(0));
 584         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 585                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 586                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 587                                          + ((char *)de - bh->b_data))) {
 588                         /* silently ignore the rest of the block */
 589                         break;
 590                 }
 591                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 592                 if ((hinfo->hash < start_hash) ||
 593                     ((hinfo->hash == start_hash) &&
 594                      (hinfo->minor_hash < start_minor_hash)))
 595                         continue;
 596                 if (de->inode == 0)
 597                         continue;
 598                 if ((err = ext4_htree_store_dirent(dir_file,
 599                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 600                         brelse(bh);
 601                         return err;
 602                 }
 603                 count++;
 604         }
 605         brelse(bh);
 606         return count;
 607 }
 608
 609
 610 /*
 611  * This function fills a red-black tree with information from a
 612  * directory.  We start scanning the directory in hash order, starting
 613  * at start_hash and start_minor_hash.
 614  *
 615  * This function returns the number of entries inserted into the tree,
 616  * or a negative error code.
 617  */
 618 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 619                          __u32 start_minor_hash, __u32 *next_hash)
 620 {
 621         struct dx_hash_info hinfo;
 622         struct ext4_dir_entry_2 *de;
 623         struct dx_frame frames[2], *frame;
 624         struct inode *dir;
 625         ext4_lblk_t block;
 626         int count = 0;
 627         int ret, err;
 628         __u32 hashval;
 629
 630         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 631                        start_hash, start_minor_hash));
 632         dir = dir_file->f_path.dentry->d_inode;
 633         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 634                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 635                 if (hinfo.hash_version <= DX_HASH_TEA)
 636                         hinfo.hash_version +=
 637                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 638                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 639                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 640                                                start_hash, start_minor_hash);
 641                 *next_hash = ~0;
 642                 return count;
 643         }
 644         hinfo.hash = start_hash;
 645         hinfo.minor_hash = 0;
 646         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 647         if (!frame)
 648                 return err;
 649
 650         /* Add '.' and '..' from the htree header */
 651         if (!start_hash && !start_minor_hash) {
 652                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 653                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 654                         goto errout;
 655                 count++;
 656         }
 657         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 658                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 659                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 660                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 661                         goto errout;
 662                 count++;
 663         }
 664
 665         while (1) {
 666                 block = dx_get_block(frame->at);
 667                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 668                                              start_hash, start_minor_hash);
 669                 if (ret < 0) {
 670                         err = ret;
 671                         goto errout;
 672                 }
 673                 count += ret;
 674                 hashval = ~0;
 675                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 676                                             frame, frames, &hashval);
 677                 *next_hash = hashval;
 678                 if (ret < 0) {
 679                         err = ret;
 680                         goto errout;
 681                 }
 682                 /*
 683                  * Stop if:  (a) there are no more entries, or
 684                  * (b) we have inserted at least one entry and the
 685                  * next hash value is not a continuation
 686                  */
 687                 if ((ret == 0) ||
 688                     (count && ((hashval & 1) == 0)))
 689                         break;
 690         }
 691         dx_release(frames);
 692         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 693                        "next hash: %x\n", count, *next_hash));
 694         return count;
 695 errout:
 696         dx_release(frames);
 697         return (err);
 698 }
 699
 700
 701 /*
 702  * Directory block splitting, compacting
 703  */
 704
 705 /*
 706  * Create map of hash values, offsets, and sizes, stored at end of block.
 707  * Returns number of entries mapped.
 708  */
 709 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 710                        struct dx_hash_info *hinfo,
 711                        struct dx_map_entry *map_tail)
 712 {
 713         int count = 0;
 714         char *base = (char *) de;
 715         struct dx_hash_info h = *hinfo;
 716
 717         while ((char *) de < base + blocksize) {
 718                 if (de->name_len && de->inode) {
 719                         ext4fs_dirhash(de->name, de->name_len, &h);
 720                         map_tail--;
 721                         map_tail->hash = h.hash;
 722                         map_tail->offs = ((char *) de - base)>>2;
 723                         map_tail->size = le16_to_cpu(de->rec_len);
 724                         count++;
 725                         cond_resched();
 726                 }
 727                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 728                 de = ext4_next_entry(de, blocksize);
 729         }
 730         return count;
 731 }
 732
 733 /* Sort map by hash value */
 734 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 735 {
 736         struct dx_map_entry *p, *q, *top = map + count - 1;
 737         int more;
 738         /* Combsort until bubble sort doesn't suck */
 739         while (count > 2) {
 740                 count = count*10/13;
 741                 if (count - 9 < 2) /* 9, 10 -> 11 */
 742                         count = 11;
 743                 for (p = top, q = p - count; q >= map; p--, q--)
 744                         if (p->hash < q->hash)
 745                                 swap(*p, *q);
 746         }
 747         /* Garden variety bubble sort */
 748         do {
 749                 more = 0;
 750                 q = top;
 751                 while (q-- > map) {
 752                         if (q[1].hash >= q[0].hash)
 753                                 continue;
 754                         swap(*(q+1), *q);
 755                         more = 1;
 756                 }
 757         } while(more);
 758 }
 759
 760 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
 761 {
 762         struct dx_entry *entries = frame->entries;
 763         struct dx_entry *old = frame->at, *new = old + 1;
 764         int count = dx_get_count(entries);
 765
 766         assert(count < dx_get_limit(entries));
 767         assert(old < entries + count);
 768         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 769         dx_set_hash(new, hash);
 770         dx_set_block(new, block);
 771         dx_set_count(entries, count + 1);
 772 }
 773
 774 static void ext4_update_dx_flag(struct inode *inode)
 775 {
 776         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 777                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 778                 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
 779 }
 780
 781 /*
 782  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 783  *
 784  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 785  * `de != NULL' is guaranteed by caller.
 786  */
 787 static inline int ext4_match (int len, const char * const name,
 788                               struct ext4_dir_entry_2 * de)
 789 {
 790         if (len != de->name_len)
 791                 return 0;
 792         if (!de->inode)
 793                 return 0;
 794         return !memcmp(name, de->name, len);
 795 }
 796
 797 /*
 798  * Returns 0 if not found, -1 on failure, and 1 on success
 799  */
 800 static inline int search_dirblock(struct buffer_head *bh,
 801                                   struct inode *dir,
 802                                   const struct qstr *d_name,
 803                                   unsigned int offset,
 804                                   struct ext4_dir_entry_2 ** res_dir)
 805 {
 806         struct ext4_dir_entry_2 * de;
 807         char * dlimit;
 808         int de_len;
 809         const char *name = d_name->name;
 810         int namelen = d_name->len;
 811
 812         de = (struct ext4_dir_entry_2 *) bh->b_data;
 813         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 814         while ((char *) de < dlimit) {
 815                 /* this code is executed quadratically often */
 816                 /* do minimal checking `by hand' */
 817
 818                 if ((char *) de + namelen <= dlimit &&
 819                     ext4_match (namelen, name, de)) {
 820                         /* found a match - just to be sure, do a full check */
 821                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
 822                                 return -1;
 823                         *res_dir = de;
 824                         return 1;
 825                 }
 826                 /* prevent looping on a bad block */
 827                 de_len = ext4_rec_len_from_disk(de->rec_len,
 828                                                 dir->i_sb->s_blocksize);
 829                 if (de_len <= 0)
 830                         return -1;
 831                 offset += de_len;
 832                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 833         }
 834         return 0;
 835 }
 836
 837
 838 /*
 839  *      ext4_find_entry()
 840  *
 841  * finds an entry in the specified directory with the wanted name. It
 842  * returns the cache buffer in which the entry was found, and the entry
 843  * itself (as a parameter - res_dir). It does NOT read the inode of the
 844  * entry - you'll have to do that yourself if you want to.
 845  *
 846  * The returned buffer_head has ->b_count elevated.  The caller is expected
 847  * to brelse() it when appropriate.
 848  */
 849 static struct buffer_head * ext4_find_entry (struct inode *dir,
 850                                         const struct qstr *d_name,
 851                                         struct ext4_dir_entry_2 ** res_dir)
 852 {
 853         struct super_block *sb;
 854         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 855         struct buffer_head *bh, *ret = NULL;
 856         ext4_lblk_t start, block, b;
 857         const u8 *name = d_name->name;
 858         int ra_max = 0;         /* Number of bh's in the readahead
 859                                    buffer, bh_use[] */
 860         int ra_ptr = 0;         /* Current index into readahead
 861                                    buffer */
 862         int num = 0;
 863         ext4_lblk_t  nblocks;
 864         int i, err;
 865         int namelen;
 866
 867         *res_dir = NULL;
 868         sb = dir->i_sb;
 869         namelen = d_name->len;
 870         if (namelen > EXT4_NAME_LEN)
 871                 return NULL;
 872         if ((namelen <= 2) && (name[0] == '.') &&
 873             (name[1] == '.' || name[1] == '\0')) {
 874                 /*
 875                  * "." or ".." will only be in the first block
 876                  * NFS may look up ".."; "." should be handled by the VFS
 877                  */
 878                 block = start = 0;
 879                 nblocks = 1;
 880                 goto restart;
 881         }
 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 | REQ_META | REQ_PRIO,
 923                                                     1, &bh);
 924                         }
 925                 }
 926                 if ((bh = bh_use[ra_ptr++]) == NULL)
 927                         goto next;
 928                 wait_on_buffer(bh);
 929                 if (!buffer_uptodate(bh)) {
 930                         /* read error, skip block & hope for the best */
 931                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
 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 = dir->i_sb;
 974         struct dx_hash_info     hinfo;
 975         struct dx_frame frames[2], *frame;
 976         struct buffer_head *bh;
 977         ext4_lblk_t block;
 978         int retval;
 979
 980         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
 981                 return NULL;
 982         do {
 983                 block = dx_get_block(frame->at);
 984                 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
 985                         goto errout;
 986
 987                 retval = search_dirblock(bh, dir, d_name,
 988                                          block << EXT4_BLOCK_SIZE_BITS(sb),
 989                                          res_dir);
 990                 if (retval == 1) {      /* Success! */
 991                         dx_release(frames);
 992                         return bh;
 993                 }
 994                 brelse(bh);
 995                 if (retval == -1) {
 996                         *err = ERR_BAD_DX_DIR;
 997                         goto errout;
 998                 }
 999
1000                 /* Check to see if we should continue to search */
1001                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1002                                                frames, NULL);
1003                 if (retval < 0) {
1004                         ext4_warning(sb,
1005                              "error reading index page in directory #%lu",
1006                              dir->i_ino);
1007                         *err = retval;
1008                         goto errout;
1009                 }
1010         } while (retval == 1);
1011
1012         *err = -ENOENT;
1013 errout:
1014         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1015         dx_release (frames);
1016         return NULL;
1017 }
1018
1019 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1020 {
1021         struct inode *inode;
1022         struct ext4_dir_entry_2 *de;
1023         struct buffer_head *bh;
1024
1025         if (dentry->d_name.len > EXT4_NAME_LEN)
1026                 return ERR_PTR(-ENAMETOOLONG);
1027
1028         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1029         inode = NULL;
1030         if (bh) {
1031                 __u32 ino = le32_to_cpu(de->inode);
1032                 brelse(bh);
1033                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1034                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1035                         return ERR_PTR(-EIO);
1036                 }
1037                 if (unlikely(ino == dir->i_ino)) {
1038                         EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1039                                          dentry->d_name.len,
1040                                          dentry->d_name.name);
1041                         return ERR_PTR(-EIO);
1042                 }
1043                 inode = ext4_iget_normal(dir->i_sb, ino);
1044                 if (inode == ERR_PTR(-ESTALE)) {
1045                         EXT4_ERROR_INODE(dir,
1046                                          "deleted inode referenced: %u",
1047                                          ino);
1048                         return ERR_PTR(-EIO);
1049                 }
1050         }
1051         return d_splice_alias(inode, dentry);
1052 }
1053
1054
1055 struct dentry *ext4_get_parent(struct dentry *child)
1056 {
1057         __u32 ino;
1058         static const struct qstr dotdot = {
1059                 .name = "..",
1060                 .len = 2,
1061         };
1062         struct ext4_dir_entry_2 * de;
1063         struct buffer_head *bh;
1064
1065         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1066         if (!bh)
1067                 return ERR_PTR(-ENOENT);
1068         ino = le32_to_cpu(de->inode);
1069         brelse(bh);
1070
1071         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1072                 EXT4_ERROR_INODE(child->d_inode,
1073                                  "bad parent inode number: %u", ino);
1074                 return ERR_PTR(-EIO);
1075         }
1076
1077         return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1078 }
1079
1080 #define S_SHIFT 12
1081 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1082         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1083         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1084         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1085         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1086         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1087         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1088         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1089 };
1090
1091 static inline void ext4_set_de_type(struct super_block *sb,
1092                                 struct ext4_dir_entry_2 *de,
1093                                 umode_t mode) {
1094         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1095                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1096 }
1097
1098 /*
1099  * Move count entries from end of map between two memory locations.
1100  * Returns pointer to last entry moved.
1101  */
1102 static struct ext4_dir_entry_2 *
1103 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1104                 unsigned blocksize)
1105 {
1106         unsigned rec_len = 0;
1107
1108         while (count--) {
1109                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1110                                                 (from + (map->offs<<2));
1111                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1112                 memcpy (to, de, rec_len);
1113                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1114                                 ext4_rec_len_to_disk(rec_len, blocksize);
1115                 de->inode = 0;
1116                 map++;
1117                 to += rec_len;
1118         }
1119         return (struct ext4_dir_entry_2 *) (to - rec_len);
1120 }
1121
1122 /*
1123  * Compact each dir entry in the range to the minimal rec_len.
1124  * Returns pointer to last entry in range.
1125  */
1126 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1127 {
1128         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1129         unsigned rec_len = 0;
1130
1131         prev = to = de;
1132         while ((char*)de < base + blocksize) {
1133                 next = ext4_next_entry(de, blocksize);
1134                 if (de->inode && de->name_len) {
1135                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1136                         if (de > to)
1137                                 memmove(to, de, rec_len);
1138                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1139                         prev = to;
1140                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1141                 }
1142                 de = next;
1143         }
1144         return prev;
1145 }
1146
1147 /*
1148  * Split a full leaf block to make room for a new dir entry.
1149  * Allocate a new block, and move entries so that they are approx. equally full.
1150  * Returns pointer to de in block into which the new entry will be inserted.
1151  */
1152 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1153                         struct buffer_head **bh,struct dx_frame *frame,
1154                         struct dx_hash_info *hinfo, int *error)
1155 {
1156         unsigned blocksize = dir->i_sb->s_blocksize;
1157         unsigned count, continued;
1158         struct buffer_head *bh2;
1159         ext4_lblk_t newblock;
1160         u32 hash2;
1161         struct dx_map_entry *map;
1162         char *data1 = (*bh)->b_data, *data2;
1163         unsigned split, move, size;
1164         struct ext4_dir_entry_2 *de = NULL, *de2;
1165         int     err = 0, i;
1166
1167         bh2 = ext4_append (handle, dir, &newblock, &err);
1168         if (!(bh2)) {
1169                 brelse(*bh);
1170                 *bh = NULL;
1171                 goto errout;
1172         }
1173
1174         BUFFER_TRACE(*bh, "get_write_access");
1175         err = ext4_journal_get_write_access(handle, *bh);
1176         if (err)
1177                 goto journal_error;
1178
1179         BUFFER_TRACE(frame->bh, "get_write_access");
1180         err = ext4_journal_get_write_access(handle, frame->bh);
1181         if (err)
1182                 goto journal_error;
1183
1184         data2 = bh2->b_data;
1185
1186         /* create map in the end of data2 block */
1187         map = (struct dx_map_entry *) (data2 + blocksize);
1188         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1189                              blocksize, hinfo, map);
1190         map -= count;
1191         dx_sort_map(map, count);
1192         /* Split the existing block in the middle, size-wise */
1193         size = 0;
1194         move = 0;
1195         for (i = count-1; i >= 0; i--) {
1196                 /* is more than half of this entry in 2nd half of the block? */
1197                 if (size + map[i].size/2 > blocksize/2)
1198                         break;
1199                 size += map[i].size;
1200                 move++;
1201         }
1202         /* map index at which we will split */
1203         split = count - move;
1204         hash2 = map[split].hash;
1205         continued = hash2 == map[split - 1].hash;
1206         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1207                         (unsigned long)dx_get_block(frame->at),
1208                                         hash2, split, count-split));
1209
1210         /* Fancy dance to stay within two buffers */
1211         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1212         de = dx_pack_dirents(data1, blocksize);
1213         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1214                                            blocksize);
1215         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1216                                             blocksize);
1217         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1218         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1219
1220         /* Which block gets the new entry? */
1221         if (hinfo->hash >= hash2)
1222         {
1223                 swap(*bh, bh2);
1224                 de = de2;
1225         }
1226         dx_insert_block(frame, hash2 + continued, newblock);
1227         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1228         if (err)
1229                 goto journal_error;
1230         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1231         if (err)
1232                 goto journal_error;
1233         brelse(bh2);
1234         dxtrace(dx_show_index("frame", frame->entries));
1235         return de;
1236
1237 journal_error:
1238         brelse(*bh);
1239         brelse(bh2);
1240         *bh = NULL;
1241         ext4_std_error(dir->i_sb, err);
1242 errout:
1243         *error = err;
1244         return NULL;
1245 }
1246
1247 /*
1248  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1249  * it points to a directory entry which is guaranteed to be large
1250  * enough for new directory entry.  If de is NULL, then
1251  * add_dirent_to_buf will attempt search the directory block for
1252  * space.  It will return -ENOSPC if no space is available, and -EIO
1253  * and -EEXIST if directory entry already exists.
1254  */
1255 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1256                              struct inode *inode, struct ext4_dir_entry_2 *de,
1257                              struct buffer_head *bh)
1258 {
1259         struct inode    *dir = dentry->d_parent->d_inode;
1260         const char      *name = dentry->d_name.name;
1261         int             namelen = dentry->d_name.len;
1262         unsigned int    offset = 0;
1263         unsigned int    blocksize = dir->i_sb->s_blocksize;
1264         unsigned short  reclen;
1265         int             nlen, rlen, err;
1266         char            *top;
1267
1268         reclen = EXT4_DIR_REC_LEN(namelen);
1269         if (!de) {
1270                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1271                 top = bh->b_data + blocksize - reclen;
1272                 while ((char *) de <= top) {
1273                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1274                                 return -EIO;
1275                         if (ext4_match(namelen, name, de))
1276                                 return -EEXIST;
1277                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1278                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1279                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1280                                 break;
1281                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1282                         offset += rlen;
1283                 }
1284                 if ((char *) de > top)
1285                         return -ENOSPC;
1286         }
1287         BUFFER_TRACE(bh, "get_write_access");
1288         err = ext4_journal_get_write_access(handle, bh);
1289         if (err) {
1290                 ext4_std_error(dir->i_sb, err);
1291                 return err;
1292         }
1293
1294         /* By now the buffer is marked for journaling */
1295         nlen = EXT4_DIR_REC_LEN(de->name_len);
1296         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1297         if (de->inode) {
1298                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1299                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1300                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1301                 de = de1;
1302         }
1303         de->file_type = EXT4_FT_UNKNOWN;
1304         if (inode) {
1305                 de->inode = cpu_to_le32(inode->i_ino);
1306                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1307         } else
1308                 de->inode = 0;
1309         de->name_len = namelen;
1310         memcpy(de->name, name, namelen);
1311         /*
1312          * XXX shouldn't update any times until successful
1313          * completion of syscall, but too many callers depend
1314          * on this.
1315          *
1316          * XXX similarly, too many callers depend on
1317          * ext4_new_inode() setting the times, but error
1318          * recovery deletes the inode, so the worst that can
1319          * happen is that the times are slightly out of date
1320          * and/or different from the directory change time.
1321          */
1322         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1323         ext4_update_dx_flag(dir);
1324         dir->i_version++;
1325         ext4_mark_inode_dirty(handle, dir);
1326         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1327         err = ext4_handle_dirty_metadata(handle, dir, bh);
1328         if (err)
1329                 ext4_std_error(dir->i_sb, err);
1330         return 0;
1331 }
1332
1333 /*
1334  * This converts a one block unindexed directory to a 3 block indexed
1335  * directory, and adds the dentry to the indexed directory.
1336  */
1337 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1338                             struct inode *inode, struct buffer_head *bh)
1339 {
1340         struct inode    *dir = dentry->d_parent->d_inode;
1341         const char      *name = dentry->d_name.name;
1342         int             namelen = dentry->d_name.len;
1343         struct buffer_head *bh2;
1344         struct dx_root  *root;
1345         struct dx_frame frames[2], *frame;
1346         struct dx_entry *entries;
1347         struct ext4_dir_entry_2 *de, *de2;
1348         char            *data1, *top;
1349         unsigned        len;
1350         int             retval;
1351         unsigned        blocksize;
1352         struct dx_hash_info hinfo;
1353         ext4_lblk_t  block;
1354         struct fake_dirent *fde;
1355
1356         blocksize =  dir->i_sb->s_blocksize;
1357         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1358         retval = ext4_journal_get_write_access(handle, bh);
1359         if (retval) {
1360                 ext4_std_error(dir->i_sb, retval);
1361                 brelse(bh);
1362                 return retval;
1363         }
1364         root = (struct dx_root *) bh->b_data;
1365
1366         /* The 0th block becomes the root, move the dirents out */
1367         fde = &root->dotdot;
1368         de = (struct ext4_dir_entry_2 *)((char *)fde +
1369                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1370         if ((char *) de >= (((char *) root) + blocksize)) {
1371                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1372                 brelse(bh);
1373                 return -EIO;
1374         }
1375         len = ((char *) root) + blocksize - (char *) de;
1376
1377         /* Allocate new block for the 0th block's dirents */
1378         bh2 = ext4_append(handle, dir, &block, &retval);
1379         if (!(bh2)) {
1380                 brelse(bh);
1381                 return retval;
1382         }
1383         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1384         data1 = bh2->b_data;
1385
1386         memcpy (data1, de, len);
1387         de = (struct ext4_dir_entry_2 *) data1;
1388         top = data1 + len;
1389         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1390                 de = de2;
1391         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1392                                            blocksize);
1393         /* Initialize the root; the dot dirents already exist */
1394         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1395         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1396                                            blocksize);
1397         memset (&root->info, 0, sizeof(root->info));
1398         root->info.info_length = sizeof(root->info);
1399         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1400         entries = root->entries;
1401         dx_set_block(entries, 1);
1402         dx_set_count(entries, 1);
1403         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1404
1405         /* Initialize as for dx_probe */
1406         hinfo.hash_version = root->info.hash_version;
1407         if (hinfo.hash_version <= DX_HASH_TEA)
1408                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1409         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1410         ext4fs_dirhash(name, namelen, &hinfo);
1411         memset(frames, 0, sizeof(frames));
1412         frame = frames;
1413         frame->entries = entries;
1414         frame->at = entries;
1415         frame->bh = bh;
1416
1417         retval = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1418         if (retval)
1419                 goto out_frames;
1420         retval = ext4_handle_dirty_metadata(handle, dir, bh2);
1421         if (retval)
1422                 goto out_frames;
1423
1424         de = do_split(handle,dir, &bh2, frame, &hinfo, &retval);
1425         if (!de) {
1426                 goto out_frames;
1427         }
1428
1429         retval = add_dirent_to_buf(handle, dentry, inode, de, bh2);
1430 out_frames:
1431         /*
1432          * Even if the block split failed, we have to properly write
1433          * out all the changes we did so far. Otherwise we can end up
1434          * with corrupted filesystem.
1435          */
1436         if (retval)
1437                 ext4_mark_inode_dirty(handle, dir);
1438         dx_release(frames);
1439         brelse(bh2);
1440         return retval;
1441 }
1442
1443 /*
1444  *      ext4_add_entry()
1445  *
1446  * adds a file entry to the specified directory, using the same
1447  * semantics as ext4_find_entry(). It returns NULL if it failed.
1448  *
1449  * NOTE!! The inode part of 'de' is left at 0 - which means you
1450  * may not sleep between calling this and putting something into
1451  * the entry, as someone else might have used it while you slept.
1452  */
1453 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1454                           struct inode *inode)
1455 {
1456         struct inode *dir = dentry->d_parent->d_inode;
1457         struct buffer_head *bh = NULL;
1458         struct ext4_dir_entry_2 *de;
1459         struct super_block *sb;
1460         int     retval;
1461         int     dx_fallback=0;
1462         unsigned blocksize;
1463         ext4_lblk_t block, blocks;
1464
1465         sb = dir->i_sb;
1466         blocksize = sb->s_blocksize;
1467         if (!dentry->d_name.len)
1468                 return -EINVAL;
1469         if (is_dx(dir)) {
1470                 retval = ext4_dx_add_entry(handle, dentry, inode);
1471                 if (!retval || (retval != ERR_BAD_DX_DIR))
1472                         goto out;
1473                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1474                 dx_fallback++;
1475                 ext4_mark_inode_dirty(handle, dir);
1476         }
1477         blocks = dir->i_size >> sb->s_blocksize_bits;
1478         for (block = 0; block < blocks; block++) {
1479                 bh = ext4_bread(handle, dir, block, 0, &retval);
1480                 if(!bh)
1481                         return retval;
1482                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1483                 if (retval != -ENOSPC)
1484                         goto out;
1485
1486                 if (blocks == 1 && !dx_fallback &&
1487                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
1488                         retval = make_indexed_dir(handle, dentry, inode, bh);
1489                         bh = NULL; /* make_indexed_dir releases bh */
1490                         goto out;
1491                 }
1492                 brelse(bh);
1493         }
1494         bh = ext4_append(handle, dir, &block, &retval);
1495         if (!bh)
1496                 return retval;
1497         de = (struct ext4_dir_entry_2 *) bh->b_data;
1498         de->inode = 0;
1499         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1500         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1501 out:
1502         brelse(bh);
1503         if (retval == 0)
1504                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1505         return retval;
1506 }
1507
1508 /*
1509  * Returns 0 for success, or a negative error value
1510  */
1511 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1512                              struct inode *inode)
1513 {
1514         struct dx_frame frames[2], *frame;
1515         struct dx_entry *entries, *at;
1516         struct dx_hash_info hinfo;
1517         struct buffer_head *bh;
1518         struct inode *dir = dentry->d_parent->d_inode;
1519         struct super_block *sb = dir->i_sb;
1520         struct ext4_dir_entry_2 *de;
1521         int err;
1522
1523         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1524         if (!frame)
1525                 return err;
1526         entries = frame->entries;
1527         at = frame->at;
1528
1529         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1530                 goto cleanup;
1531
1532         BUFFER_TRACE(bh, "get_write_access");
1533         err = ext4_journal_get_write_access(handle, bh);
1534         if (err)
1535                 goto journal_error;
1536
1537         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1538         if (err != -ENOSPC)
1539                 goto cleanup;
1540
1541         /* Block full, should compress but for now just split */
1542         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1543                        dx_get_count(entries), dx_get_limit(entries)));
1544         /* Need to split index? */
1545         if (dx_get_count(entries) == dx_get_limit(entries)) {
1546                 ext4_lblk_t newblock;
1547                 unsigned icount = dx_get_count(entries);
1548                 int levels = frame - frames;
1549                 struct dx_entry *entries2;
1550                 struct dx_node *node2;
1551                 struct buffer_head *bh2;
1552
1553                 if (levels && (dx_get_count(frames->entries) ==
1554                                dx_get_limit(frames->entries))) {
1555                         ext4_warning(sb, "Directory index full!");
1556                         err = -ENOSPC;
1557                         goto cleanup;
1558                 }
1559                 bh2 = ext4_append (handle, dir, &newblock, &err);
1560                 if (!(bh2))
1561                         goto cleanup;
1562                 node2 = (struct dx_node *)(bh2->b_data);
1563                 entries2 = node2->entries;
1564                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1565                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1566                                                            sb->s_blocksize);
1567                 BUFFER_TRACE(frame->bh, "get_write_access");
1568                 err = ext4_journal_get_write_access(handle, frame->bh);
1569                 if (err)
1570                         goto journal_error;
1571                 if (levels) {
1572                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1573                         unsigned hash2 = dx_get_hash(entries + icount1);
1574                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1575                                        icount1, icount2));
1576
1577                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1578                         err = ext4_journal_get_write_access(handle,
1579                                                              frames[0].bh);
1580                         if (err)
1581                                 goto journal_error;
1582
1583                         memcpy((char *) entries2, (char *) (entries + icount1),
1584                                icount2 * sizeof(struct dx_entry));
1585                         dx_set_count(entries, icount1);
1586                         dx_set_count(entries2, icount2);
1587                         dx_set_limit(entries2, dx_node_limit(dir));
1588
1589                         /* Which index block gets the new entry? */
1590                         if (at - entries >= icount1) {
1591                                 frame->at = at = at - entries - icount1 + entries2;
1592                                 frame->entries = entries = entries2;
1593                                 swap(frame->bh, bh2);
1594                         }
1595                         dx_insert_block(frames + 0, hash2, newblock);
1596                         dxtrace(dx_show_index("node", frames[1].entries));
1597                         dxtrace(dx_show_index("node",
1598                                ((struct dx_node *) bh2->b_data)->entries));
1599                         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1600                         if (err)
1601                                 goto journal_error;
1602                         brelse (bh2);
1603                 } else {
1604                         dxtrace(printk(KERN_DEBUG
1605                                        "Creating second level index...\n"));
1606                         memcpy((char *) entries2, (char *) entries,
1607                                icount * sizeof(struct dx_entry));
1608                         dx_set_limit(entries2, dx_node_limit(dir));
1609
1610                         /* Set up root */
1611                         dx_set_count(entries, 1);
1612                         dx_set_block(entries + 0, newblock);
1613                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1614
1615                         /* Add new access path frame */
1616                         frame = frames + 1;
1617                         frame->at = at = at - entries + entries2;
1618                         frame->entries = entries = entries2;
1619                         frame->bh = bh2;
1620                         err = ext4_journal_get_write_access(handle,
1621                                                              frame->bh);
1622                         if (err)
1623                                 goto journal_error;
1624                 }
1625                 err = ext4_handle_dirty_metadata(handle, dir, frames[0].bh);
1626                 if (err) {
1627                         ext4_std_error(inode->i_sb, err);
1628                         goto cleanup;
1629                 }
1630         }
1631         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1632         if (!de)
1633                 goto cleanup;
1634         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1635         goto cleanup;
1636
1637 journal_error:
1638         ext4_std_error(dir->i_sb, err);
1639 cleanup:
1640         if (bh)
1641                 brelse(bh);
1642         dx_release(frames);
1643         return err;
1644 }
1645
1646 /*
1647  * ext4_delete_entry deletes a directory entry by merging it with the
1648  * previous entry
1649  */
1650 static int ext4_delete_entry(handle_t *handle,
1651                              struct inode *dir,
1652                              struct ext4_dir_entry_2 *de_del,
1653                              struct buffer_head *bh)
1654 {
1655         struct ext4_dir_entry_2 *de, *pde;
1656         unsigned int blocksize = dir->i_sb->s_blocksize;
1657         int i, err;
1658
1659         i = 0;
1660         pde = NULL;
1661         de = (struct ext4_dir_entry_2 *) bh->b_data;
1662         while (i < bh->b_size) {
1663                 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1664                         return -EIO;
1665                 if (de == de_del)  {
1666                         BUFFER_TRACE(bh, "get_write_access");
1667                         err = ext4_journal_get_write_access(handle, bh);
1668                         if (unlikely(err)) {
1669                                 ext4_std_error(dir->i_sb, err);
1670                                 return err;
1671                         }
1672                         if (pde)
1673                                 pde->rec_len = ext4_rec_len_to_disk(
1674                                         ext4_rec_len_from_disk(pde->rec_len,
1675                                                                blocksize) +
1676                                         ext4_rec_len_from_disk(de->rec_len,
1677                                                                blocksize),
1678                                         blocksize);
1679                         else
1680                                 de->inode = 0;
1681                         dir->i_version++;
1682                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1683                         err = ext4_handle_dirty_metadata(handle, dir, bh);
1684                         if (unlikely(err)) {
1685                                 ext4_std_error(dir->i_sb, err);
1686                                 return err;
1687                         }
1688                         return 0;
1689                 }
1690                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1691                 pde = de;
1692                 de = ext4_next_entry(de, blocksize);
1693         }
1694         return -ENOENT;
1695 }
1696
1697 /*
1698  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1699  * since this indicates that nlinks count was previously 1.
1700  */
1701 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1702 {
1703         inc_nlink(inode);
1704         if (is_dx(inode) && inode->i_nlink > 1) {
1705                 /* limit is 16-bit i_links_count */
1706                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1707                         set_nlink(inode, 1);
1708                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1709                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1710                 }
1711         }
1712 }
1713
1714 /*
1715  * If a directory had nlink == 1, then we should let it be 1. This indicates
1716  * directory has >EXT4_LINK_MAX subdirs.
1717  */
1718 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1719 {
1720         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
1721                 drop_nlink(inode);
1722 }
1723
1724
1725 static int ext4_add_nondir(handle_t *handle,
1726                 struct dentry *dentry, struct inode *inode)
1727 {
1728         int err = ext4_add_entry(handle, dentry, inode);
1729         if (!err) {
1730                 ext4_mark_inode_dirty(handle, inode);
1731                 d_instantiate(dentry, inode);
1732                 unlock_new_inode(inode);
1733                 return 0;
1734         }
1735         drop_nlink(inode);
1736         unlock_new_inode(inode);
1737         iput(inode);
1738         return err;
1739 }
1740
1741 /*
1742  * By the time this is called, we already have created
1743  * the directory cache entry for the new file, but it
1744  * is so far negative - it has no inode.
1745  *
1746  * If the create succeeds, we fill in the inode information
1747  * with d_instantiate().
1748  */
1749 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1750                        struct nameidata *nd)
1751 {
1752         handle_t *handle;
1753         struct inode *inode;
1754         int err, retries = 0;
1755
1756         dquot_initialize(dir);
1757
1758 retry:
1759         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1760                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1761                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1762         if (IS_ERR(handle))
1763                 return PTR_ERR(handle);
1764
1765         if (IS_DIRSYNC(dir))
1766                 ext4_handle_sync(handle);
1767
1768         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1769         err = PTR_ERR(inode);
1770         if (!IS_ERR(inode)) {
1771                 inode->i_op = &ext4_file_inode_operations;
1772                 inode->i_fop = &ext4_file_operations;
1773                 ext4_set_aops(inode);
1774                 err = ext4_add_nondir(handle, dentry, inode);
1775         }
1776         ext4_journal_stop(handle);
1777         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1778                 goto retry;
1779         return err;
1780 }
1781
1782 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1783                       int mode, dev_t rdev)
1784 {
1785         handle_t *handle;
1786         struct inode *inode;
1787         int err, retries = 0;
1788
1789         if (!new_valid_dev(rdev))
1790                 return -EINVAL;
1791
1792         dquot_initialize(dir);
1793
1794 retry:
1795         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1796                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1797                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1798         if (IS_ERR(handle))
1799                 return PTR_ERR(handle);
1800
1801         if (IS_DIRSYNC(dir))
1802                 ext4_handle_sync(handle);
1803
1804         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1805         err = PTR_ERR(inode);
1806         if (!IS_ERR(inode)) {
1807                 init_special_inode(inode, inode->i_mode, rdev);
1808                 inode->i_op = &ext4_special_inode_operations;
1809                 err = ext4_add_nondir(handle, dentry, inode);
1810         }
1811         ext4_journal_stop(handle);
1812         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1813                 goto retry;
1814         return err;
1815 }
1816
1817 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1818 {
1819         handle_t *handle;
1820         struct inode *inode;
1821         struct buffer_head *dir_block = NULL;
1822         struct ext4_dir_entry_2 *de;
1823         unsigned int blocksize = dir->i_sb->s_blocksize;
1824         int err, retries = 0;
1825
1826         if (EXT4_DIR_LINK_MAX(dir))
1827                 return -EMLINK;
1828
1829         dquot_initialize(dir);
1830
1831 retry:
1832         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1833                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1834                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1835         if (IS_ERR(handle))
1836                 return PTR_ERR(handle);
1837
1838         if (IS_DIRSYNC(dir))
1839                 ext4_handle_sync(handle);
1840
1841         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1842                                &dentry->d_name, 0, NULL);
1843         err = PTR_ERR(inode);
1844         if (IS_ERR(inode))
1845                 goto out_stop;
1846
1847         inode->i_op = &ext4_dir_inode_operations;
1848         inode->i_fop = &ext4_dir_operations;
1849         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1850         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1851         if (!dir_block)
1852                 goto out_clear_inode;
1853         BUFFER_TRACE(dir_block, "get_write_access");
1854         err = ext4_journal_get_write_access(handle, dir_block);
1855         if (err)
1856                 goto out_clear_inode;
1857         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1858         de->inode = cpu_to_le32(inode->i_ino);
1859         de->name_len = 1;
1860         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1861                                            blocksize);
1862         strcpy(de->name, ".");
1863         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1864         de = ext4_next_entry(de, blocksize);
1865         de->inode = cpu_to_le32(dir->i_ino);
1866         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1867                                            blocksize);
1868         de->name_len = 2;
1869         strcpy(de->name, "..");
1870         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1871         set_nlink(inode, 2);
1872         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1873         err = ext4_handle_dirty_metadata(handle, inode, dir_block);
1874         if (err)
1875                 goto out_clear_inode;
1876         err = ext4_mark_inode_dirty(handle, inode);
1877         if (!err)
1878                 err = ext4_add_entry(handle, dentry, inode);
1879         if (err) {
1880 out_clear_inode:
1881                 clear_nlink(inode);
1882                 unlock_new_inode(inode);
1883                 ext4_mark_inode_dirty(handle, inode);
1884                 iput(inode);
1885                 goto out_stop;
1886         }
1887         ext4_inc_count(handle, dir);
1888         ext4_update_dx_flag(dir);
1889         err = ext4_mark_inode_dirty(handle, dir);
1890         if (err)
1891                 goto out_clear_inode;
1892         d_instantiate(dentry, inode);
1893         unlock_new_inode(inode);
1894 out_stop:
1895         brelse(dir_block);
1896         ext4_journal_stop(handle);
1897         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1898                 goto retry;
1899         return err;
1900 }
1901
1902 /*
1903  * routine to check that the specified directory is empty (for rmdir)
1904  */
1905 static int empty_dir(struct inode *inode)
1906 {
1907         unsigned int offset;
1908         struct buffer_head *bh;
1909         struct ext4_dir_entry_2 *de, *de1;
1910         struct super_block *sb;
1911         int err = 0;
1912
1913         sb = inode->i_sb;
1914         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1915             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1916                 if (err)
1917                         EXT4_ERROR_INODE(inode,
1918                                 "error %d reading directory lblock 0", err);
1919                 else
1920                         ext4_warning(inode->i_sb,
1921                                      "bad directory (dir #%lu) - no data block",
1922                                      inode->i_ino);
1923                 return 1;
1924         }
1925         de = (struct ext4_dir_entry_2 *) bh->b_data;
1926         de1 = ext4_next_entry(de, sb->s_blocksize);
1927         if (le32_to_cpu(de->inode) != inode->i_ino ||
1928                         !le32_to_cpu(de1->inode) ||
1929                         strcmp(".", de->name) ||
1930                         strcmp("..", de1->name)) {
1931                 ext4_warning(inode->i_sb,
1932                              "bad directory (dir #%lu) - no `.' or `..'",
1933                              inode->i_ino);
1934                 brelse(bh);
1935                 return 1;
1936         }
1937         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1938                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1939         de = ext4_next_entry(de1, sb->s_blocksize);
1940         while (offset < inode->i_size) {
1941                 if (!bh ||
1942                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1943                         unsigned int lblock;
1944                         err = 0;
1945                         brelse(bh);
1946                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1947                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
1948                         if (!bh) {
1949                                 if (err)
1950                                         EXT4_ERROR_INODE(inode,
1951                                                 "error %d reading directory "
1952                                                 "lblock %u", err, lblock);
1953                                 offset += sb->s_blocksize;
1954                                 continue;
1955                         }
1956                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1957                 }
1958                 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1959                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1960                                                          sb->s_blocksize);
1961                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1962                         continue;
1963                 }
1964                 if (le32_to_cpu(de->inode)) {
1965                         brelse(bh);
1966                         return 0;
1967                 }
1968                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1969                 de = ext4_next_entry(de, sb->s_blocksize);
1970         }
1971         brelse(bh);
1972         return 1;
1973 }
1974
1975 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1976  * such inodes, starting at the superblock, in case we crash before the
1977  * file is closed/deleted, or in case the inode truncate spans multiple
1978  * transactions and the last transaction is not recovered after a crash.
1979  *
1980  * At filesystem recovery time, we walk this list deleting unlinked
1981  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1982  */
1983 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1984 {
1985         struct super_block *sb = inode->i_sb;
1986         struct ext4_iloc iloc;
1987         int err = 0, rc;
1988
1989         if (!EXT4_SB(sb)->s_journal || is_bad_inode(inode))
1990                 return 0;
1991
1992         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1993         if (!list_empty(&EXT4_I(inode)->i_orphan))
1994                 goto out_unlock;
1995
1996         /*
1997          * Orphan handling is only valid for files with data blocks
1998          * being truncated, or files being unlinked. Note that we either
1999          * hold i_mutex, or the inode can not be referenced from outside,
2000          * so i_nlink should not be bumped due to race
2001          */
2002         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2003                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2004
2005         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2006         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2007         if (err)
2008                 goto out_unlock;
2009
2010         err = ext4_reserve_inode_write(handle, inode, &iloc);
2011         if (err)
2012                 goto out_unlock;
2013         /*
2014          * Due to previous errors inode may be already a part of on-disk
2015          * orphan list. If so skip on-disk list modification.
2016          */
2017         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2018                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2019                         goto mem_insert;
2020
2021         /* Insert this inode at the head of the on-disk orphan list... */
2022         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2023         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2024         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2025         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2026         if (!err)
2027                 err = rc;
2028
2029         /* Only add to the head of the in-memory list if all the
2030          * previous operations succeeded.  If the orphan_add is going to
2031          * fail (possibly taking the journal offline), we can't risk
2032          * leaving the inode on the orphan list: stray orphan-list
2033          * entries can cause panics at unmount time.
2034          *
2035          * This is safe: on error we're going to ignore the orphan list
2036          * anyway on the next recovery. */
2037 mem_insert:
2038         if (!err)
2039                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2040
2041         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2042         jbd_debug(4, "orphan inode %lu will point to %d\n",
2043                         inode->i_ino, NEXT_ORPHAN(inode));
2044 out_unlock:
2045         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2046         ext4_std_error(inode->i_sb, err);
2047         return err;
2048 }
2049
2050 /*
2051  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2052  * of such inodes stored on disk, because it is finally being cleaned up.
2053  */
2054 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2055 {
2056         struct list_head *prev;
2057         struct ext4_inode_info *ei = EXT4_I(inode);
2058         struct ext4_sb_info *sbi;
2059         __u32 ino_next;
2060         struct ext4_iloc iloc;
2061         int err = 0;
2062
2063         if (!EXT4_SB(inode->i_sb)->s_journal &&
2064             !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2065                 return 0;
2066
2067         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2068         if (list_empty(&ei->i_orphan))
2069                 goto out;
2070
2071         ino_next = NEXT_ORPHAN(inode);
2072         prev = ei->i_orphan.prev;
2073         sbi = EXT4_SB(inode->i_sb);
2074
2075         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2076
2077         list_del_init(&ei->i_orphan);
2078
2079         /* If we're on an error path, we may not have a valid
2080          * transaction handle with which to update the orphan list on
2081          * disk, but we still need to remove the inode from the linked
2082          * list in memory. */
2083         if (!handle)
2084                 goto out;
2085
2086         err = ext4_reserve_inode_write(handle, inode, &iloc);
2087         if (err)
2088                 goto out_err;
2089
2090         if (prev == &sbi->s_orphan) {
2091                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2092                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2093                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2094                 if (err)
2095                         goto out_brelse;
2096                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2097                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2098         } else {
2099                 struct ext4_iloc iloc2;
2100                 struct inode *i_prev =
2101                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2102
2103                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2104                           i_prev->i_ino, ino_next);
2105                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2106                 if (err)
2107                         goto out_brelse;
2108                 NEXT_ORPHAN(i_prev) = ino_next;
2109                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2110         }
2111         if (err)
2112                 goto out_brelse;
2113         NEXT_ORPHAN(inode) = 0;
2114         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2115
2116 out_err:
2117         ext4_std_error(inode->i_sb, err);
2118 out:
2119         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2120         return err;
2121
2122 out_brelse:
2123         brelse(iloc.bh);
2124         goto out_err;
2125 }
2126
2127 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2128 {
2129         int retval;
2130         struct inode *inode;
2131         struct buffer_head *bh;
2132         struct ext4_dir_entry_2 *de;
2133         handle_t *handle;
2134
2135         /* Initialize quotas before so that eventual writes go in
2136          * separate transaction */
2137         dquot_initialize(dir);
2138         dquot_initialize(dentry->d_inode);
2139
2140         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2141         if (IS_ERR(handle))
2142                 return PTR_ERR(handle);
2143
2144         retval = -ENOENT;
2145         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2146         if (!bh)
2147                 goto end_rmdir;
2148
2149         if (IS_DIRSYNC(dir))
2150                 ext4_handle_sync(handle);
2151
2152         inode = dentry->d_inode;
2153
2154         retval = -EIO;
2155         if (le32_to_cpu(de->inode) != inode->i_ino)
2156                 goto end_rmdir;
2157
2158         retval = -ENOTEMPTY;
2159         if (!empty_dir(inode))
2160                 goto end_rmdir;
2161
2162         retval = ext4_delete_entry(handle, dir, de, bh);
2163         if (retval)
2164                 goto end_rmdir;
2165         if (!EXT4_DIR_LINK_EMPTY(inode))
2166                 ext4_warning(inode->i_sb,
2167                              "empty directory has too many links (%d)",
2168                              inode->i_nlink);
2169         inode->i_version++;
2170         clear_nlink(inode);
2171         /* There's no need to set i_disksize: the fact that i_nlink is
2172          * zero will ensure that the right thing happens during any
2173          * recovery. */
2174         inode->i_size = 0;
2175         ext4_orphan_add(handle, inode);
2176         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2177         ext4_mark_inode_dirty(handle, inode);
2178         ext4_dec_count(handle, dir);
2179         ext4_update_dx_flag(dir);
2180         ext4_mark_inode_dirty(handle, dir);
2181
2182 end_rmdir:
2183         ext4_journal_stop(handle);
2184         brelse(bh);
2185         return retval;
2186 }
2187
2188 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2189 {
2190         int retval;
2191         struct inode *inode;
2192         struct buffer_head *bh;
2193         struct ext4_dir_entry_2 *de;
2194         handle_t *handle;
2195
2196         trace_ext4_unlink_enter(dir, dentry);
2197         /* Initialize quotas before so that eventual writes go
2198          * in separate transaction */
2199         dquot_initialize(dir);
2200         dquot_initialize(dentry->d_inode);
2201
2202         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2203         if (IS_ERR(handle))
2204                 return PTR_ERR(handle);
2205
2206         if (IS_DIRSYNC(dir))
2207                 ext4_handle_sync(handle);
2208
2209         retval = -ENOENT;
2210         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2211         if (!bh)
2212                 goto end_unlink;
2213
2214         inode = dentry->d_inode;
2215
2216         retval = -EIO;
2217         if (le32_to_cpu(de->inode) != inode->i_ino)
2218                 goto end_unlink;
2219
2220         if (!inode->i_nlink) {
2221                 ext4_warning(inode->i_sb,
2222                              "Deleting nonexistent file (%lu), %d",
2223                              inode->i_ino, inode->i_nlink);
2224                 set_nlink(inode, 1);
2225         }
2226         retval = ext4_delete_entry(handle, dir, de, bh);
2227         if (retval)
2228                 goto end_unlink;
2229         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2230         ext4_update_dx_flag(dir);
2231         ext4_mark_inode_dirty(handle, dir);
2232         drop_nlink(inode);
2233         if (!inode->i_nlink)
2234                 ext4_orphan_add(handle, inode);
2235         inode->i_ctime = ext4_current_time(inode);
2236         ext4_mark_inode_dirty(handle, inode);
2237         retval = 0;
2238
2239 end_unlink:
2240         ext4_journal_stop(handle);
2241         brelse(bh);
2242         trace_ext4_unlink_exit(dentry, retval);
2243         return retval;
2244 }
2245
2246 static int ext4_symlink(struct inode *dir,
2247                         struct dentry *dentry, const char *symname)
2248 {
2249         handle_t *handle;
2250         struct inode *inode;
2251         int l, err, retries = 0;
2252         int credits;
2253
2254         l = strlen(symname)+1;
2255         if (l > dir->i_sb->s_blocksize)
2256                 return -ENAMETOOLONG;
2257
2258         dquot_initialize(dir);
2259
2260         if (l > EXT4_N_BLOCKS * 4) {
2261                 /*
2262                  * For non-fast symlinks, we just allocate inode and put it on
2263                  * orphan list in the first transaction => we need bitmap,
2264                  * group descriptor, sb, inode block, quota blocks, and
2265                  * possibly selinux xattr blocks.
2266                  */
2267                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2268                           EXT4_XATTR_TRANS_BLOCKS;
2269         } else {
2270                 /*
2271                  * Fast symlink. We have to add entry to directory
2272                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2273                  * allocate new inode (bitmap, group descriptor, inode block,
2274                  * quota blocks, sb is already counted in previous macros).
2275                  */
2276                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2277                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2278                           EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2279         }
2280 retry:
2281         handle = ext4_journal_start(dir, credits);
2282         if (IS_ERR(handle))
2283                 return PTR_ERR(handle);
2284
2285         if (IS_DIRSYNC(dir))
2286                 ext4_handle_sync(handle);
2287
2288         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2289                                &dentry->d_name, 0, NULL);
2290         err = PTR_ERR(inode);
2291         if (IS_ERR(inode))
2292                 goto out_stop;
2293
2294         if (l > EXT4_N_BLOCKS * 4) {
2295                 inode->i_op = &ext4_symlink_inode_operations;
2296                 ext4_set_aops(inode);
2297                 /*
2298                  * We cannot call page_symlink() with transaction started
2299                  * because it calls into ext4_write_begin() which can wait
2300                  * for transaction commit if we are running out of space
2301                  * and thus we deadlock. So we have to stop transaction now
2302                  * and restart it when symlink contents is written.
2303                  *
2304                  * To keep fs consistent in case of crash, we have to put inode
2305                  * to orphan list in the mean time.
2306                  */
2307                 drop_nlink(inode);
2308                 err = ext4_orphan_add(handle, inode);
2309                 ext4_journal_stop(handle);
2310                 if (err)
2311                         goto err_drop_inode;
2312                 err = __page_symlink(inode, symname, l, 1);
2313                 if (err)
2314                         goto err_drop_inode;
2315                 /*
2316                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2317                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2318                  */
2319                 handle = ext4_journal_start(dir,
2320                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2321                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2322                 if (IS_ERR(handle)) {
2323                         err = PTR_ERR(handle);
2324                         goto err_drop_inode;
2325                 }
2326                 inc_nlink(inode);
2327                 err = ext4_orphan_del(handle, inode);
2328                 if (err) {
2329                         ext4_journal_stop(handle);
2330                         clear_nlink(inode);
2331                         goto err_drop_inode;
2332                 }
2333         } else {
2334                 /* clear the extent format for fast symlink */
2335                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2336                 inode->i_op = &ext4_fast_symlink_inode_operations;
2337                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2338                 inode->i_size = l-1;
2339         }
2340         EXT4_I(inode)->i_disksize = inode->i_size;
2341         err = ext4_add_nondir(handle, dentry, inode);
2342 out_stop:
2343         ext4_journal_stop(handle);
2344         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2345                 goto retry;
2346         return err;
2347 err_drop_inode:
2348         unlock_new_inode(inode);
2349         iput(inode);
2350         return err;
2351 }
2352
2353 static int ext4_link(struct dentry *old_dentry,
2354                      struct inode *dir, struct dentry *dentry)
2355 {
2356         handle_t *handle;
2357         struct inode *inode = old_dentry->d_inode;
2358         int err, retries = 0;
2359
2360         if (inode->i_nlink >= EXT4_LINK_MAX)
2361                 return -EMLINK;
2362
2363         dquot_initialize(dir);
2364
2365 retry:
2366         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2367                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2368         if (IS_ERR(handle))
2369                 return PTR_ERR(handle);
2370
2371         if (IS_DIRSYNC(dir))
2372                 ext4_handle_sync(handle);
2373
2374         inode->i_ctime = ext4_current_time(inode);
2375         ext4_inc_count(handle, inode);
2376         ihold(inode);
2377
2378         err = ext4_add_entry(handle, dentry, inode);
2379         if (!err) {
2380                 ext4_mark_inode_dirty(handle, inode);
2381                 d_instantiate(dentry, inode);
2382         } else {
2383                 drop_nlink(inode);
2384                 iput(inode);
2385         }
2386         ext4_journal_stop(handle);
2387         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2388                 goto retry;
2389         return err;
2390 }
2391
2392 #define PARENT_INO(buffer, size) \
2393         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2394
2395 /*
2396  * Anybody can rename anything with this: the permission checks are left to the
2397  * higher-level routines.
2398  */
2399 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2400                        struct inode *new_dir, struct dentry *new_dentry)
2401 {
2402         handle_t *handle;
2403         struct inode *old_inode, *new_inode;
2404         struct buffer_head *old_bh, *new_bh, *dir_bh;
2405         struct ext4_dir_entry_2 *old_de, *new_de;
2406         int retval, force_da_alloc = 0;
2407
2408         dquot_initialize(old_dir);
2409         dquot_initialize(new_dir);
2410
2411         old_bh = new_bh = dir_bh = NULL;
2412
2413         /* Initialize quotas before so that eventual writes go
2414          * in separate transaction */
2415         if (new_dentry->d_inode)
2416                 dquot_initialize(new_dentry->d_inode);
2417         handle = ext4_journal_start(old_dir, 2 *
2418                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2419                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2420         if (IS_ERR(handle))
2421                 return PTR_ERR(handle);
2422
2423         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2424                 ext4_handle_sync(handle);
2425
2426         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2427         /*
2428          *  Check for inode number is _not_ due to possible IO errors.
2429          *  We might rmdir the source, keep it as pwd of some process
2430          *  and merrily kill the link to whatever was created under the
2431          *  same name. Goodbye sticky bit ;-<
2432          */
2433         old_inode = old_dentry->d_inode;
2434         retval = -ENOENT;
2435         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2436                 goto end_rename;
2437
2438         new_inode = new_dentry->d_inode;
2439         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2440         if (new_bh) {
2441                 if (!new_inode) {
2442                         brelse(new_bh);
2443                         new_bh = NULL;
2444                 }
2445         }
2446         if (S_ISDIR(old_inode->i_mode)) {
2447                 if (new_inode) {
2448                         retval = -ENOTEMPTY;
2449                         if (!empty_dir(new_inode))
2450                                 goto end_rename;
2451                 }
2452                 retval = -EIO;
2453                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2454                 if (!dir_bh)
2455                         goto end_rename;
2456                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2457                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2458                         goto end_rename;
2459                 retval = -EMLINK;
2460                 if (!new_inode && new_dir != old_dir &&
2461                     EXT4_DIR_LINK_MAX(new_dir))
2462                         goto end_rename;
2463                 BUFFER_TRACE(dir_bh, "get_write_access");
2464                 retval = ext4_journal_get_write_access(handle, dir_bh);
2465                 if (retval)
2466                         goto end_rename;
2467         }
2468         if (!new_bh) {
2469                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2470                 if (retval)
2471                         goto end_rename;
2472         } else {
2473                 BUFFER_TRACE(new_bh, "get write access");
2474                 retval = ext4_journal_get_write_access(handle, new_bh);
2475                 if (retval)
2476                         goto end_rename;
2477                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2478                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2479                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2480                         new_de->file_type = old_de->file_type;
2481                 new_dir->i_version++;
2482                 new_dir->i_ctime = new_dir->i_mtime =
2483                                         ext4_current_time(new_dir);
2484                 ext4_mark_inode_dirty(handle, new_dir);
2485                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2486                 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2487                 if (unlikely(retval)) {
2488                         ext4_std_error(new_dir->i_sb, retval);
2489                         goto end_rename;
2490                 }
2491                 brelse(new_bh);
2492                 new_bh = NULL;
2493         }
2494
2495         /*
2496          * Like most other Unix systems, set the ctime for inodes on a
2497          * rename.
2498          */
2499         old_inode->i_ctime = ext4_current_time(old_inode);
2500         ext4_mark_inode_dirty(handle, old_inode);
2501
2502         /*
2503          * ok, that's it
2504          */
2505         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2506             old_de->name_len != old_dentry->d_name.len ||
2507             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2508             (retval = ext4_delete_entry(handle, old_dir,
2509                                         old_de, old_bh)) == -ENOENT) {
2510                 /* old_de could have moved from under us during htree split, so
2511                  * make sure that we are deleting the right entry.  We might
2512                  * also be pointing to a stale entry in the unused part of
2513                  * old_bh so just checking inum and the name isn't enough. */
2514                 struct buffer_head *old_bh2;
2515                 struct ext4_dir_entry_2 *old_de2;
2516
2517                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2518                 if (old_bh2) {
2519                         retval = ext4_delete_entry(handle, old_dir,
2520                                                    old_de2, old_bh2);
2521                         brelse(old_bh2);
2522                 }
2523         }
2524         if (retval) {
2525                 ext4_warning(old_dir->i_sb,
2526                                 "Deleting old file (%lu), %d, error=%d",
2527                                 old_dir->i_ino, old_dir->i_nlink, retval);
2528         }
2529
2530         if (new_inode) {
2531                 ext4_dec_count(handle, new_inode);
2532                 new_inode->i_ctime = ext4_current_time(new_inode);
2533         }
2534         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2535         ext4_update_dx_flag(old_dir);
2536         if (dir_bh) {
2537                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2538                                                 cpu_to_le32(new_dir->i_ino);
2539                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2540                 retval = ext4_handle_dirty_metadata(handle, old_inode, dir_bh);
2541                 if (retval) {
2542                         ext4_std_error(old_dir->i_sb, retval);
2543                         goto end_rename;
2544                 }
2545                 ext4_dec_count(handle, old_dir);
2546                 if (new_inode) {
2547                         /* checked empty_dir above, can't have another parent,
2548                          * ext4_dec_count() won't work for many-linked dirs */
2549                         clear_nlink(new_inode);
2550                 } else {
2551                         ext4_inc_count(handle, new_dir);
2552                         ext4_update_dx_flag(new_dir);
2553                         ext4_mark_inode_dirty(handle, new_dir);
2554                 }
2555         }
2556         ext4_mark_inode_dirty(handle, old_dir);
2557         if (new_inode) {
2558                 ext4_mark_inode_dirty(handle, new_inode);
2559                 if (!new_inode->i_nlink)
2560                         ext4_orphan_add(handle, new_inode);
2561                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2562                         force_da_alloc = 1;
2563         }
2564         retval = 0;
2565
2566 end_rename:
2567         brelse(dir_bh);
2568         brelse(old_bh);
2569         brelse(new_bh);
2570         ext4_journal_stop(handle);
2571         if (retval == 0 && force_da_alloc)
2572                 ext4_alloc_da_blocks(old_inode);
2573         return retval;
2574 }
2575
2576 /*
2577  * directories can handle most operations...
2578  */
2579 const struct inode_operations ext4_dir_inode_operations = {
2580         .create         = ext4_create,
2581         .lookup         = ext4_lookup,
2582         .link           = ext4_link,
2583         .unlink         = ext4_unlink,
2584         .symlink        = ext4_symlink,
2585         .mkdir          = ext4_mkdir,
2586         .rmdir          = ext4_rmdir,
2587         .mknod          = ext4_mknod,
2588         .rename         = ext4_rename,
2589         .setattr        = ext4_setattr,
2590 #ifdef CONFIG_EXT4_FS_XATTR
2591         .setxattr       = generic_setxattr,
2592         .getxattr       = generic_getxattr,
2593         .listxattr      = ext4_listxattr,
2594         .removexattr    = generic_removexattr,
2595 #endif
2596         .get_acl        = ext4_get_acl,
2597         .fiemap         = ext4_fiemap,
2598 };
2599
2600 const struct inode_operations ext4_special_inode_operations = {
2601         .setattr        = ext4_setattr,
2602 #ifdef CONFIG_EXT4_FS_XATTR
2603         .setxattr       = generic_setxattr,
2604         .getxattr       = generic_getxattr,
2605         .listxattr      = ext4_listxattr,
2606         .removexattr    = generic_removexattr,
2607 #endif
2608         .get_acl        = ext4_get_acl,
2609 };