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 /*
  44  * define how far ahead to read directories while searching them.
  45  */
  46 #define NAMEI_RA_CHUNKS  2
  47 #define NAMEI_RA_BLOCKS  4
  48 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  49 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  50
  51 static struct buffer_head *ext4_append(handle_t *handle,
  52                                         struct inode *inode,
  53                                         ext4_lblk_t *block, int *err)
  54 {
  55         struct buffer_head *bh;
  56
  57         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  58
  59         bh = ext4_bread(handle, inode, *block, 1, err);
  60         if (bh) {
  61                 inode->i_size += inode->i_sb->s_blocksize;
  62                 EXT4_I(inode)->i_disksize = inode->i_size;
  63                 *err = ext4_journal_get_write_access(handle, bh);
  64                 if (*err) {
  65                         brelse(bh);
  66                         bh = NULL;
  67                 }
  68         }
  69         return bh;
  70 }
  71
  72 #ifndef assert
  73 #define assert(test) J_ASSERT(test)
  74 #endif
  75
  76 #ifdef DX_DEBUG
  77 #define dxtrace(command) command
  78 #else
  79 #define dxtrace(command)
  80 #endif
  81
  82 struct fake_dirent
  83 {
  84         __le32 inode;
  85         __le16 rec_len;
  86         u8 name_len;
  87         u8 file_type;
  88 };
  89
  90 struct dx_countlimit
  91 {
  92         __le16 limit;
  93         __le16 count;
  94 };
  95
  96 struct dx_entry
  97 {
  98         __le32 hash;
  99         __le32 block;
 100 };
 101
 102 /*
 103  * dx_root_info is laid out so that if it should somehow get overlaid by a
 104  * dirent the two low bits of the hash version will be zero.  Therefore, the
 105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 106  */
 107
 108 struct dx_root
 109 {
 110         struct fake_dirent dot;
 111         char dot_name[4];
 112         struct fake_dirent dotdot;
 113         char dotdot_name[4];
 114         struct dx_root_info
 115         {
 116                 __le32 reserved_zero;
 117                 u8 hash_version;
 118                 u8 info_length; /* 8 */
 119                 u8 indirect_levels;
 120                 u8 unused_flags;
 121         }
 122         info;
 123         struct dx_entry entries[0];
 124 };
 125
 126 struct dx_node
 127 {
 128         struct fake_dirent fake;
 129         struct dx_entry entries[0];
 130 };
 131
 132
 133 struct dx_frame
 134 {
 135         struct buffer_head *bh;
 136         struct dx_entry *entries;
 137         struct dx_entry *at;
 138 };
 139
 140 struct dx_map_entry
 141 {
 142         u32 hash;
 143         u16 offs;
 144         u16 size;
 145 };
 146
 147 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 148 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 149 static inline unsigned dx_get_hash(struct dx_entry *entry);
 150 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 151 static unsigned dx_get_count(struct dx_entry *entries);
 152 static unsigned dx_get_limit(struct dx_entry *entries);
 153 static void dx_set_count(struct dx_entry *entries, unsigned value);
 154 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 155 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 156 static unsigned dx_node_limit(struct inode *dir);
 157 static struct dx_frame *dx_probe(const struct qstr *d_name,
 158                                  struct inode *dir,
 159                                  struct dx_hash_info *hinfo,
 160                                  struct dx_frame *frame,
 161                                  int *err);
 162 static void dx_release(struct dx_frame *frames);
 163 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 164                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 166 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 167                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 168 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 169 static void dx_insert_block(struct dx_frame *frame,
 170                                         u32 hash, ext4_lblk_t block);
 171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 172                                  struct dx_frame *frame,
 173                                  struct dx_frame *frames,
 174                                  __u32 *start_hash);
 175 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 176                 const struct qstr *d_name,
 177                 struct ext4_dir_entry_2 **res_dir,
 178                 int *err);
 179 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 180                              struct inode *inode);
 181
 182 unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
 183 {
 184         unsigned len = le16_to_cpu(dlen);
 185
 186         if (len == EXT4_MAX_REC_LEN || len == 0)
 187                 return blocksize;
 188         return (len & 65532) | ((len & 3) << 16);
 189 }
 190
 191 __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
 192 {
 193         if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
 194                 BUG();
 195         if (len < 65536)
 196                 return cpu_to_le16(len);
 197         if (len == blocksize) {
 198                 if (blocksize == 65536)
 199                         return cpu_to_le16(EXT4_MAX_REC_LEN);
 200                 else
 201                         return cpu_to_le16(0);
 202         }
 203         return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
 204 }
 205
 206 /*
 207  * p is at least 6 bytes before the end of page
 208  */
 209 static inline struct ext4_dir_entry_2 *
 210 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 211 {
 212         return (struct ext4_dir_entry_2 *)((char *)p +
 213                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 214 }
 215
 216 /*
 217  * Future: use high four bits of block for coalesce-on-delete flags
 218  * Mask them off for now.
 219  */
 220
 221 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 222 {
 223         return le32_to_cpu(entry->block) & 0x00ffffff;
 224 }
 225
 226 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 227 {
 228         entry->block = cpu_to_le32(value);
 229 }
 230
 231 static inline unsigned dx_get_hash(struct dx_entry *entry)
 232 {
 233         return le32_to_cpu(entry->hash);
 234 }
 235
 236 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 237 {
 238         entry->hash = cpu_to_le32(value);
 239 }
 240
 241 static inline unsigned dx_get_count(struct dx_entry *entries)
 242 {
 243         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 244 }
 245
 246 static inline unsigned dx_get_limit(struct dx_entry *entries)
 247 {
 248         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 249 }
 250
 251 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 252 {
 253         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 254 }
 255
 256 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 257 {
 258         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 259 }
 260
 261 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 262 {
 263         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 264                 EXT4_DIR_REC_LEN(2) - infosize;
 265         return entry_space / sizeof(struct dx_entry);
 266 }
 267
 268 static inline unsigned dx_node_limit(struct inode *dir)
 269 {
 270         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 271         return entry_space / sizeof(struct dx_entry);
 272 }
 273
 274 /*
 275  * Debug
 276  */
 277 #ifdef DX_DEBUG
 278 static void dx_show_index(char * label, struct dx_entry *entries)
 279 {
 280         int i, n = dx_get_count (entries);
 281         printk(KERN_DEBUG "%s index ", label);
 282         for (i = 0; i < n; i++) {
 283                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 284                                 0, (unsigned long)dx_get_block(entries + i));
 285         }
 286         printk("\n");
 287 }
 288
 289 struct stats
 290 {
 291         unsigned names;
 292         unsigned space;
 293         unsigned bcount;
 294 };
 295
 296 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 297                                  int size, int show_names)
 298 {
 299         unsigned names = 0, space = 0;
 300         char *base = (char *) de;
 301         struct dx_hash_info h = *hinfo;
 302
 303         printk("names: ");
 304         while ((char *) de < base + size)
 305         {
 306                 if (de->inode)
 307                 {
 308                         if (show_names)
 309                         {
 310                                 int len = de->name_len;
 311                                 char *name = de->name;
 312                                 while (len--) printk("%c", *name++);
 313                                 ext4fs_dirhash(de->name, de->name_len, &h);
 314                                 printk(":%x.%u ", h.hash,
 315                                        ((char *) de - base));
 316                         }
 317                         space += EXT4_DIR_REC_LEN(de->name_len);
 318                         names++;
 319                 }
 320                 de = ext4_next_entry(de, size);
 321         }
 322         printk("(%i)\n", names);
 323         return (struct stats) { names, space, 1 };
 324 }
 325
 326 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 327                              struct dx_entry *entries, int levels)
 328 {
 329         unsigned blocksize = dir->i_sb->s_blocksize;
 330         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 331         unsigned bcount = 0;
 332         struct buffer_head *bh;
 333         int err;
 334         printk("%i indexed blocks...\n", count);
 335         for (i = 0; i < count; i++, entries++)
 336         {
 337                 ext4_lblk_t block = dx_get_block(entries);
 338                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 339                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 340                 struct stats stats;
 341                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 342                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 343                 stats = levels?
 344                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 345                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 346                 names += stats.names;
 347                 space += stats.space;
 348                 bcount += stats.bcount;
 349                 brelse(bh);
 350         }
 351         if (bcount)
 352                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 353                        levels ? "" : "   ", names, space/bcount,
 354                        (space/bcount)*100/blocksize);
 355         return (struct stats) { names, space, bcount};
 356 }
 357 #endif /* DX_DEBUG */
 358
 359 /*
 360  * Probe for a directory leaf block to search.
 361  *
 362  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 363  * error in the directory index, and the caller should fall back to
 364  * searching the directory normally.  The callers of dx_probe **MUST**
 365  * check for this error code, and make sure it never gets reflected
 366  * back to userspace.
 367  */
 368 static struct dx_frame *
 369 dx_probe(const struct qstr *d_name, struct inode *dir,
 370          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 371 {
 372         unsigned count, indirect;
 373         struct dx_entry *at, *entries, *p, *q, *m;
 374         struct dx_root *root;
 375         struct buffer_head *bh;
 376         struct dx_frame *frame = frame_in;
 377         u32 hash;
 378
 379         frame->bh = NULL;
 380         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 381                 goto fail;
 382         root = (struct dx_root *) bh->b_data;
 383         if (root->info.hash_version != DX_HASH_TEA &&
 384             root->info.hash_version != DX_HASH_HALF_MD4 &&
 385             root->info.hash_version != DX_HASH_LEGACY) {
 386                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 387                              root->info.hash_version);
 388                 brelse(bh);
 389                 *err = ERR_BAD_DX_DIR;
 390                 goto fail;
 391         }
 392         hinfo->hash_version = root->info.hash_version;
 393         if (hinfo->hash_version <= DX_HASH_TEA)
 394                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 395         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 396         if (d_name)
 397                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 398         hash = hinfo->hash;
 399
 400         if (root->info.unused_flags & 1) {
 401                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 402                              root->info.unused_flags);
 403                 brelse(bh);
 404                 *err = ERR_BAD_DX_DIR;
 405                 goto fail;
 406         }
 407
 408         if ((indirect = root->info.indirect_levels) > 1) {
 409                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 410                              root->info.indirect_levels);
 411                 brelse(bh);
 412                 *err = ERR_BAD_DX_DIR;
 413                 goto fail;
 414         }
 415
 416         entries = (struct dx_entry *) (((char *)&root->info) +
 417                                        root->info.info_length);
 418
 419         if (dx_get_limit(entries) != dx_root_limit(dir,
 420                                                    root->info.info_length)) {
 421                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 422                 brelse(bh);
 423                 *err = ERR_BAD_DX_DIR;
 424                 goto fail;
 425         }
 426
 427         dxtrace(printk("Look up %x", hash));
 428         while (1)
 429         {
 430                 count = dx_get_count(entries);
 431                 if (!count || count > dx_get_limit(entries)) {
 432                         ext4_warning(dir->i_sb,
 433                                      "dx entry: no count or count > limit");
 434                         brelse(bh);
 435                         *err = ERR_BAD_DX_DIR;
 436                         goto fail2;
 437                 }
 438
 439                 p = entries + 1;
 440                 q = entries + count - 1;
 441                 while (p <= q)
 442                 {
 443                         m = p + (q - p)/2;
 444                         dxtrace(printk("."));
 445                         if (dx_get_hash(m) > hash)
 446                                 q = m - 1;
 447                         else
 448                                 p = m + 1;
 449                 }
 450
 451                 if (0) // linear search cross check
 452                 {
 453                         unsigned n = count - 1;
 454                         at = entries;
 455                         while (n--)
 456                         {
 457                                 dxtrace(printk(","));
 458                                 if (dx_get_hash(++at) > hash)
 459                                 {
 460                                         at--;
 461                                         break;
 462                                 }
 463                         }
 464                         assert (at == p - 1);
 465                 }
 466
 467                 at = p - 1;
 468                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 469                 frame->bh = bh;
 470                 frame->entries = entries;
 471                 frame->at = at;
 472                 if (!indirect--) return frame;
 473                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 474                         goto fail2;
 475                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 476                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 477                         ext4_warning(dir->i_sb,
 478                                      "dx entry: limit != node limit");
 479                         brelse(bh);
 480                         *err = ERR_BAD_DX_DIR;
 481                         goto fail2;
 482                 }
 483                 frame++;
 484                 frame->bh = NULL;
 485         }
 486 fail2:
 487         while (frame >= frame_in) {
 488                 brelse(frame->bh);
 489                 frame--;
 490         }
 491 fail:
 492         if (*err == ERR_BAD_DX_DIR)
 493                 ext4_warning(dir->i_sb,
 494                              "Corrupt dir inode %ld, running e2fsck is "
 495                              "recommended.", dir->i_ino);
 496         return NULL;
 497 }
 498
 499 static void dx_release (struct dx_frame *frames)
 500 {
 501         if (frames[0].bh == NULL)
 502                 return;
 503
 504         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 505                 brelse(frames[1].bh);
 506         brelse(frames[0].bh);
 507 }
 508
 509 /*
 510  * This function increments the frame pointer to search the next leaf
 511  * block, and reads in the necessary intervening nodes if the search
 512  * should be necessary.  Whether or not the search is necessary is
 513  * controlled by the hash parameter.  If the hash value is even, then
 514  * the search is only continued if the next block starts with that
 515  * hash value.  This is used if we are searching for a specific file.
 516  *
 517  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 518  *
 519  * This function returns 1 if the caller should continue to search,
 520  * or 0 if it should not.  If there is an error reading one of the
 521  * index blocks, it will a negative error code.
 522  *
 523  * If start_hash is non-null, it will be filled in with the starting
 524  * hash of the next page.
 525  */
 526 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 527                                  struct dx_frame *frame,
 528                                  struct dx_frame *frames,
 529                                  __u32 *start_hash)
 530 {
 531         struct dx_frame *p;
 532         struct buffer_head *bh;
 533         int err, num_frames = 0;
 534         __u32 bhash;
 535
 536         p = frame;
 537         /*
 538          * Find the next leaf page by incrementing the frame pointer.
 539          * If we run out of entries in the interior node, loop around and
 540          * increment pointer in the parent node.  When we break out of
 541          * this loop, num_frames indicates the number of interior
 542          * nodes need to be read.
 543          */
 544         while (1) {
 545                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 546                         break;
 547                 if (p == frames)
 548                         return 0;
 549                 num_frames++;
 550                 p--;
 551         }
 552
 553         /*
 554          * If the hash is 1, then continue only if the next page has a
 555          * continuation hash of any value.  This is used for readdir
 556          * handling.  Otherwise, check to see if the hash matches the
 557          * desired contiuation hash.  If it doesn't, return since
 558          * there's no point to read in the successive index pages.
 559          */
 560         bhash = dx_get_hash(p->at);
 561         if (start_hash)
 562                 *start_hash = bhash;
 563         if ((hash & 1) == 0) {
 564                 if ((bhash & ~1) != hash)
 565                         return 0;
 566         }
 567         /*
 568          * If the hash is HASH_NB_ALWAYS, we always go to the next
 569          * block so no check is necessary
 570          */
 571         while (num_frames--) {
 572                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 573                                       0, &err)))
 574                         return err; /* Failure */
 575                 p++;
 576                 brelse(p->bh);
 577                 p->bh = bh;
 578                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 579         }
 580         return 1;
 581 }
 582
 583
 584 /*
 585  * This function fills a red-black tree with information from a
 586  * directory block.  It returns the number directory entries loaded
 587  * into the tree.  If there is an error it is returned in err.
 588  */
 589 static int htree_dirblock_to_tree(struct file *dir_file,
 590                                   struct inode *dir, ext4_lblk_t block,
 591                                   struct dx_hash_info *hinfo,
 592                                   __u32 start_hash, __u32 start_minor_hash)
 593 {
 594         struct buffer_head *bh;
 595         struct ext4_dir_entry_2 *de, *top;
 596         int err, count = 0;
 597
 598         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 599                                                         (unsigned long)block));
 600         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 601                 return err;
 602
 603         de = (struct ext4_dir_entry_2 *) bh->b_data;
 604         top = (struct ext4_dir_entry_2 *) ((char *) de +
 605                                            dir->i_sb->s_blocksize -
 606                                            EXT4_DIR_REC_LEN(0));
 607         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 608                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 609                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 610                                                 +((char *)de - bh->b_data))) {
 611                         /* On error, skip the f_pos to the next block. */
 612                         dir_file->f_pos = (dir_file->f_pos |
 613                                         (dir->i_sb->s_blocksize - 1)) + 1;
 614                         brelse(bh);
 615                         return count;
 616                 }
 617                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 618                 if ((hinfo->hash < start_hash) ||
 619                     ((hinfo->hash == start_hash) &&
 620                      (hinfo->minor_hash < start_minor_hash)))
 621                         continue;
 622                 if (de->inode == 0)
 623                         continue;
 624                 if ((err = ext4_htree_store_dirent(dir_file,
 625                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 626                         brelse(bh);
 627                         return err;
 628                 }
 629                 count++;
 630         }
 631         brelse(bh);
 632         return count;
 633 }
 634
 635
 636 /*
 637  * This function fills a red-black tree with information from a
 638  * directory.  We start scanning the directory in hash order, starting
 639  * at start_hash and start_minor_hash.
 640  *
 641  * This function returns the number of entries inserted into the tree,
 642  * or a negative error code.
 643  */
 644 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 645                          __u32 start_minor_hash, __u32 *next_hash)
 646 {
 647         struct dx_hash_info hinfo;
 648         struct ext4_dir_entry_2 *de;
 649         struct dx_frame frames[2], *frame;
 650         struct inode *dir;
 651         ext4_lblk_t block;
 652         int count = 0;
 653         int ret, err;
 654         __u32 hashval;
 655
 656         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 657                        start_hash, start_minor_hash));
 658         dir = dir_file->f_path.dentry->d_inode;
 659         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 660                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 661                 if (hinfo.hash_version <= DX_HASH_TEA)
 662                         hinfo.hash_version +=
 663                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 664                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 665                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 666                                                start_hash, start_minor_hash);
 667                 *next_hash = ~0;
 668                 return count;
 669         }
 670         hinfo.hash = start_hash;
 671         hinfo.minor_hash = 0;
 672         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 673         if (!frame)
 674                 return err;
 675
 676         /* Add '.' and '..' from the htree header */
 677         if (!start_hash && !start_minor_hash) {
 678                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 679                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 680                         goto errout;
 681                 count++;
 682         }
 683         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 684                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 685                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 686                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 687                         goto errout;
 688                 count++;
 689         }
 690
 691         while (1) {
 692                 block = dx_get_block(frame->at);
 693                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 694                                              start_hash, start_minor_hash);
 695                 if (ret < 0) {
 696                         err = ret;
 697                         goto errout;
 698                 }
 699                 count += ret;
 700                 hashval = ~0;
 701                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 702                                             frame, frames, &hashval);
 703                 *next_hash = hashval;
 704                 if (ret < 0) {
 705                         err = ret;
 706                         goto errout;
 707                 }
 708                 /*
 709                  * Stop if:  (a) there are no more entries, or
 710                  * (b) we have inserted at least one entry and the
 711                  * next hash value is not a continuation
 712                  */
 713                 if ((ret == 0) ||
 714                     (count && ((hashval & 1) == 0)))
 715                         break;
 716         }
 717         dx_release(frames);
 718         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 719                        "next hash: %x\n", count, *next_hash));
 720         return count;
 721 errout:
 722         dx_release(frames);
 723         return (err);
 724 }
 725
 726
 727 /*
 728  * Directory block splitting, compacting
 729  */
 730
 731 /*
 732  * Create map of hash values, offsets, and sizes, stored at end of block.
 733  * Returns number of entries mapped.
 734  */
 735 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 736                        struct dx_hash_info *hinfo,
 737                        struct dx_map_entry *map_tail)
 738 {
 739         int count = 0;
 740         char *base = (char *) de;
 741         struct dx_hash_info h = *hinfo;
 742
 743         while ((char *) de < base + blocksize) {
 744                 if (de->name_len && de->inode) {
 745                         ext4fs_dirhash(de->name, de->name_len, &h);
 746                         map_tail--;
 747                         map_tail->hash = h.hash;
 748                         map_tail->offs = ((char *) de - base)>>2;
 749                         map_tail->size = le16_to_cpu(de->rec_len);
 750                         count++;
 751                         cond_resched();
 752                 }
 753                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 754                 de = ext4_next_entry(de, blocksize);
 755         }
 756         return count;
 757 }
 758
 759 /* Sort map by hash value */
 760 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 761 {
 762         struct dx_map_entry *p, *q, *top = map + count - 1;
 763         int more;
 764         /* Combsort until bubble sort doesn't suck */
 765         while (count > 2) {
 766                 count = count*10/13;
 767                 if (count - 9 < 2) /* 9, 10 -> 11 */
 768                         count = 11;
 769                 for (p = top, q = p - count; q >= map; p--, q--)
 770                         if (p->hash < q->hash)
 771                                 swap(*p, *q);
 772         }
 773         /* Garden variety bubble sort */
 774         do {
 775                 more = 0;
 776                 q = top;
 777                 while (q-- > map) {
 778                         if (q[1].hash >= q[0].hash)
 779                                 continue;
 780                         swap(*(q+1), *q);
 781                         more = 1;
 782                 }
 783         } while(more);
 784 }
 785
 786 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
 787 {
 788         struct dx_entry *entries = frame->entries;
 789         struct dx_entry *old = frame->at, *new = old + 1;
 790         int count = dx_get_count(entries);
 791
 792         assert(count < dx_get_limit(entries));
 793         assert(old < entries + count);
 794         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 795         dx_set_hash(new, hash);
 796         dx_set_block(new, block);
 797         dx_set_count(entries, count + 1);
 798 }
 799
 800 static void ext4_update_dx_flag(struct inode *inode)
 801 {
 802         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 803                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 804                 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
 805 }
 806
 807 /*
 808  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 809  *
 810  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 811  * `de != NULL' is guaranteed by caller.
 812  */
 813 static inline int ext4_match (int len, const char * const name,
 814                               struct ext4_dir_entry_2 * de)
 815 {
 816         if (len != de->name_len)
 817                 return 0;
 818         if (!de->inode)
 819                 return 0;
 820         return !memcmp(name, de->name, len);
 821 }
 822
 823 /*
 824  * Returns 0 if not found, -1 on failure, and 1 on success
 825  */
 826 static inline int search_dirblock(struct buffer_head *bh,
 827                                   struct inode *dir,
 828                                   const struct qstr *d_name,
 829                                   unsigned int offset,
 830                                   struct ext4_dir_entry_2 ** res_dir)
 831 {
 832         struct ext4_dir_entry_2 * de;
 833         char * dlimit;
 834         int de_len;
 835         const char *name = d_name->name;
 836         int namelen = d_name->len;
 837
 838         de = (struct ext4_dir_entry_2 *) bh->b_data;
 839         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 840         while ((char *) de < dlimit) {
 841                 /* this code is executed quadratically often */
 842                 /* do minimal checking `by hand' */
 843
 844                 if ((char *) de + namelen <= dlimit &&
 845                     ext4_match (namelen, name, de)) {
 846                         /* found a match - just to be sure, do a full check */
 847                         if (!ext4_check_dir_entry("ext4_find_entry",
 848                                                   dir, de, bh, offset))
 849                                 return -1;
 850                         *res_dir = de;
 851                         return 1;
 852                 }
 853                 /* prevent looping on a bad block */
 854                 de_len = ext4_rec_len_from_disk(de->rec_len,
 855                                                 dir->i_sb->s_blocksize);
 856                 if (de_len <= 0)
 857                         return -1;
 858                 offset += de_len;
 859                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 860         }
 861         return 0;
 862 }
 863
 864
 865 /*
 866  *      ext4_find_entry()
 867  *
 868  * finds an entry in the specified directory with the wanted name. It
 869  * returns the cache buffer in which the entry was found, and the entry
 870  * itself (as a parameter - res_dir). It does NOT read the inode of the
 871  * entry - you'll have to do that yourself if you want to.
 872  *
 873  * The returned buffer_head has ->b_count elevated.  The caller is expected
 874  * to brelse() it when appropriate.
 875  */
 876 static struct buffer_head * ext4_find_entry (struct inode *dir,
 877                                         const struct qstr *d_name,
 878                                         struct ext4_dir_entry_2 ** res_dir)
 879 {
 880         struct super_block *sb;
 881         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 882         struct buffer_head *bh, *ret = NULL;
 883         ext4_lblk_t start, block, b;
 884         int ra_max = 0;         /* Number of bh's in the readahead
 885                                    buffer, bh_use[] */
 886         int ra_ptr = 0;         /* Current index into readahead
 887                                    buffer */
 888         int num = 0;
 889         ext4_lblk_t  nblocks;
 890         int i, err;
 891         int namelen;
 892
 893         *res_dir = NULL;
 894         sb = dir->i_sb;
 895         namelen = d_name->len;
 896         if (namelen > EXT4_NAME_LEN)
 897                 return NULL;
 898         if (is_dx(dir)) {
 899                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
 900                 /*
 901                  * On success, or if the error was file not found,
 902                  * return.  Otherwise, fall back to doing a search the
 903                  * old fashioned way.
 904                  */
 905                 if (bh || (err != ERR_BAD_DX_DIR))
 906                         return bh;
 907                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
 908                                "falling back\n"));
 909         }
 910         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 911         start = EXT4_I(dir)->i_dir_start_lookup;
 912         if (start >= nblocks)
 913                 start = 0;
 914         block = start;
 915 restart:
 916         do {
 917                 /*
 918                  * We deal with the read-ahead logic here.
 919                  */
 920                 if (ra_ptr >= ra_max) {
 921                         /* Refill the readahead buffer */
 922                         ra_ptr = 0;
 923                         b = block;
 924                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 925                                 /*
 926                                  * Terminate if we reach the end of the
 927                                  * directory and must wrap, or if our
 928                                  * search has finished at this block.
 929                                  */
 930                                 if (b >= nblocks || (num && block == start)) {
 931                                         bh_use[ra_max] = NULL;
 932                                         break;
 933                                 }
 934                                 num++;
 935                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
 936                                 bh_use[ra_max] = bh;
 937                                 if (bh)
 938                                         ll_rw_block(READ_META, 1, &bh);
 939                         }
 940                 }
 941                 if ((bh = bh_use[ra_ptr++]) == NULL)
 942                         goto next;
 943                 wait_on_buffer(bh);
 944                 if (!buffer_uptodate(bh)) {
 945                         /* read error, skip block & hope for the best */
 946                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
 947                                          (unsigned long) block);
 948                         brelse(bh);
 949                         goto next;
 950                 }
 951                 i = search_dirblock(bh, dir, d_name,
 952                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
 953                 if (i == 1) {
 954                         EXT4_I(dir)->i_dir_start_lookup = block;
 955                         ret = bh;
 956                         goto cleanup_and_exit;
 957                 } else {
 958                         brelse(bh);
 959                         if (i < 0)
 960                                 goto cleanup_and_exit;
 961                 }
 962         next:
 963                 if (++block >= nblocks)
 964                         block = 0;
 965         } while (block != start);
 966
 967         /*
 968          * If the directory has grown while we were searching, then
 969          * search the last part of the directory before giving up.
 970          */
 971         block = nblocks;
 972         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 973         if (block < nblocks) {
 974                 start = 0;
 975                 goto restart;
 976         }
 977
 978 cleanup_and_exit:
 979         /* Clean up the read-ahead blocks */
 980         for (; ra_ptr < ra_max; ra_ptr++)
 981                 brelse(bh_use[ra_ptr]);
 982         return ret;
 983 }
 984
 985 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
 986                        struct ext4_dir_entry_2 **res_dir, int *err)
 987 {
 988         struct super_block * sb;
 989         struct dx_hash_info     hinfo;
 990         u32 hash;
 991         struct dx_frame frames[2], *frame;
 992         struct ext4_dir_entry_2 *de, *top;
 993         struct buffer_head *bh;
 994         ext4_lblk_t block;
 995         int retval;
 996         int namelen = d_name->len;
 997         const u8 *name = d_name->name;
 998
 999         sb = dir->i_sb;
1000         /* NFS may look up ".." - look at dx_root directory block */
1001         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1002                 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1003                         return NULL;
1004         } else {
1005                 frame = frames;
1006                 frame->bh = NULL;                       /* for dx_release() */
1007                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
1008                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
1009         }
1010         hash = hinfo.hash;
1011         do {
1012                 block = dx_get_block(frame->at);
1013                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1014                         goto errout;
1015                 de = (struct ext4_dir_entry_2 *) bh->b_data;
1016                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1017                                        EXT4_DIR_REC_LEN(0));
1018                 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1019                         int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1020                                   + ((char *) de - bh->b_data);
1021
1022                         if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1023                                 brelse(bh);
1024                                 *err = ERR_BAD_DX_DIR;
1025                                 goto errout;
1026                         }
1027
1028                         if (ext4_match(namelen, name, de)) {
1029                                 *res_dir = de;
1030                                 dx_release(frames);
1031                                 return bh;
1032                         }
1033                 }
1034                 brelse(bh);
1035                 /* Check to see if we should continue to search */
1036                 retval = ext4_htree_next_block(dir, hash, frame,
1037                                                frames, NULL);
1038                 if (retval < 0) {
1039                         ext4_warning(sb,
1040                              "error reading index page in directory #%lu",
1041                              dir->i_ino);
1042                         *err = retval;
1043                         goto errout;
1044                 }
1045         } while (retval == 1);
1046
1047         *err = -ENOENT;
1048 errout:
1049         dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1050         dx_release (frames);
1051         return NULL;
1052 }
1053
1054 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1055 {
1056         struct inode *inode;
1057         struct ext4_dir_entry_2 *de;
1058         struct buffer_head *bh;
1059
1060         if (dentry->d_name.len > EXT4_NAME_LEN)
1061                 return ERR_PTR(-ENAMETOOLONG);
1062
1063         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1064         inode = NULL;
1065         if (bh) {
1066                 __u32 ino = le32_to_cpu(de->inode);
1067                 brelse(bh);
1068                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1069                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1070                         return ERR_PTR(-EIO);
1071                 }
1072                 inode = ext4_iget(dir->i_sb, ino);
1073                 if (unlikely(IS_ERR(inode))) {
1074                         if (PTR_ERR(inode) == -ESTALE) {
1075                                 EXT4_ERROR_INODE(dir,
1076                                                  "deleted inode referenced: %u",
1077                                                  ino);
1078                                 return ERR_PTR(-EIO);
1079                         } else {
1080                                 return ERR_CAST(inode);
1081                         }
1082                 }
1083         }
1084         return d_splice_alias(inode, dentry);
1085 }
1086
1087
1088 struct dentry *ext4_get_parent(struct dentry *child)
1089 {
1090         __u32 ino;
1091         struct inode *inode;
1092         static const struct qstr dotdot = {
1093                 .name = "..",
1094                 .len = 2,
1095         };
1096         struct ext4_dir_entry_2 * de;
1097         struct buffer_head *bh;
1098
1099         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1100         inode = NULL;
1101         if (!bh)
1102                 return ERR_PTR(-ENOENT);
1103         ino = le32_to_cpu(de->inode);
1104         brelse(bh);
1105
1106         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1107                 EXT4_ERROR_INODE(child->d_inode,
1108                                  "bad parent inode number: %u", ino);
1109                 return ERR_PTR(-EIO);
1110         }
1111
1112         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1113 }
1114
1115 #define S_SHIFT 12
1116 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1117         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1118         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1119         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1120         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1121         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1122         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1123         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1124 };
1125
1126 static inline void ext4_set_de_type(struct super_block *sb,
1127                                 struct ext4_dir_entry_2 *de,
1128                                 umode_t mode) {
1129         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1130                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1131 }
1132
1133 /*
1134  * Move count entries from end of map between two memory locations.
1135  * Returns pointer to last entry moved.
1136  */
1137 static struct ext4_dir_entry_2 *
1138 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1139                 unsigned blocksize)
1140 {
1141         unsigned rec_len = 0;
1142
1143         while (count--) {
1144                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1145                                                 (from + (map->offs<<2));
1146                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1147                 memcpy (to, de, rec_len);
1148                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1149                                 ext4_rec_len_to_disk(rec_len, blocksize);
1150                 de->inode = 0;
1151                 map++;
1152                 to += rec_len;
1153         }
1154         return (struct ext4_dir_entry_2 *) (to - rec_len);
1155 }
1156
1157 /*
1158  * Compact each dir entry in the range to the minimal rec_len.
1159  * Returns pointer to last entry in range.
1160  */
1161 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1162 {
1163         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1164         unsigned rec_len = 0;
1165
1166         prev = to = de;
1167         while ((char*)de < base + blocksize) {
1168                 next = ext4_next_entry(de, blocksize);
1169                 if (de->inode && de->name_len) {
1170                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1171                         if (de > to)
1172                                 memmove(to, de, rec_len);
1173                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1174                         prev = to;
1175                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1176                 }
1177                 de = next;
1178         }
1179         return prev;
1180 }
1181
1182 /*
1183  * Split a full leaf block to make room for a new dir entry.
1184  * Allocate a new block, and move entries so that they are approx. equally full.
1185  * Returns pointer to de in block into which the new entry will be inserted.
1186  */
1187 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1188                         struct buffer_head **bh,struct dx_frame *frame,
1189                         struct dx_hash_info *hinfo, int *error)
1190 {
1191         unsigned blocksize = dir->i_sb->s_blocksize;
1192         unsigned count, continued;
1193         struct buffer_head *bh2;
1194         ext4_lblk_t newblock;
1195         u32 hash2;
1196         struct dx_map_entry *map;
1197         char *data1 = (*bh)->b_data, *data2;
1198         unsigned split, move, size;
1199         struct ext4_dir_entry_2 *de = NULL, *de2;
1200         int     err = 0, i;
1201
1202         bh2 = ext4_append (handle, dir, &newblock, &err);
1203         if (!(bh2)) {
1204                 brelse(*bh);
1205                 *bh = NULL;
1206                 goto errout;
1207         }
1208
1209         BUFFER_TRACE(*bh, "get_write_access");
1210         err = ext4_journal_get_write_access(handle, *bh);
1211         if (err)
1212                 goto journal_error;
1213
1214         BUFFER_TRACE(frame->bh, "get_write_access");
1215         err = ext4_journal_get_write_access(handle, frame->bh);
1216         if (err)
1217                 goto journal_error;
1218
1219         data2 = bh2->b_data;
1220
1221         /* create map in the end of data2 block */
1222         map = (struct dx_map_entry *) (data2 + blocksize);
1223         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1224                              blocksize, hinfo, map);
1225         map -= count;
1226         dx_sort_map(map, count);
1227         /* Split the existing block in the middle, size-wise */
1228         size = 0;
1229         move = 0;
1230         for (i = count-1; i >= 0; i--) {
1231                 /* is more than half of this entry in 2nd half of the block? */
1232                 if (size + map[i].size/2 > blocksize/2)
1233                         break;
1234                 size += map[i].size;
1235                 move++;
1236         }
1237         /* map index at which we will split */
1238         split = count - move;
1239         hash2 = map[split].hash;
1240         continued = hash2 == map[split - 1].hash;
1241         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1242                         (unsigned long)dx_get_block(frame->at),
1243                                         hash2, split, count-split));
1244
1245         /* Fancy dance to stay within two buffers */
1246         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1247         de = dx_pack_dirents(data1, blocksize);
1248         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1249                                            blocksize);
1250         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1251                                             blocksize);
1252         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1253         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1254
1255         /* Which block gets the new entry? */
1256         if (hinfo->hash >= hash2)
1257         {
1258                 swap(*bh, bh2);
1259                 de = de2;
1260         }
1261         dx_insert_block(frame, hash2 + continued, newblock);
1262         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1263         if (err)
1264                 goto journal_error;
1265         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1266         if (err)
1267                 goto journal_error;
1268         brelse(bh2);
1269         dxtrace(dx_show_index("frame", frame->entries));
1270         return de;
1271
1272 journal_error:
1273         brelse(*bh);
1274         brelse(bh2);
1275         *bh = NULL;
1276         ext4_std_error(dir->i_sb, err);
1277 errout:
1278         *error = err;
1279         return NULL;
1280 }
1281
1282 /*
1283  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1284  * it points to a directory entry which is guaranteed to be large
1285  * enough for new directory entry.  If de is NULL, then
1286  * add_dirent_to_buf will attempt search the directory block for
1287  * space.  It will return -ENOSPC if no space is available, and -EIO
1288  * and -EEXIST if directory entry already exists.
1289  */
1290 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1291                              struct inode *inode, struct ext4_dir_entry_2 *de,
1292                              struct buffer_head *bh)
1293 {
1294         struct inode    *dir = dentry->d_parent->d_inode;
1295         const char      *name = dentry->d_name.name;
1296         int             namelen = dentry->d_name.len;
1297         unsigned int    offset = 0;
1298         unsigned int    blocksize = dir->i_sb->s_blocksize;
1299         unsigned short  reclen;
1300         int             nlen, rlen, err;
1301         char            *top;
1302
1303         reclen = EXT4_DIR_REC_LEN(namelen);
1304         if (!de) {
1305                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1306                 top = bh->b_data + blocksize - reclen;
1307                 while ((char *) de <= top) {
1308                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1309                                                   bh, offset))
1310                                 return -EIO;
1311                         if (ext4_match(namelen, name, de))
1312                                 return -EEXIST;
1313                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1314                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1315                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1316                                 break;
1317                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1318                         offset += rlen;
1319                 }
1320                 if ((char *) de > top)
1321                         return -ENOSPC;
1322         }
1323         BUFFER_TRACE(bh, "get_write_access");
1324         err = ext4_journal_get_write_access(handle, bh);
1325         if (err) {
1326                 ext4_std_error(dir->i_sb, err);
1327                 return err;
1328         }
1329
1330         /* By now the buffer is marked for journaling */
1331         nlen = EXT4_DIR_REC_LEN(de->name_len);
1332         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1333         if (de->inode) {
1334                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1335                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1336                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1337                 de = de1;
1338         }
1339         de->file_type = EXT4_FT_UNKNOWN;
1340         if (inode) {
1341                 de->inode = cpu_to_le32(inode->i_ino);
1342                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1343         } else
1344                 de->inode = 0;
1345         de->name_len = namelen;
1346         memcpy(de->name, name, namelen);
1347         /*
1348          * XXX shouldn't update any times until successful
1349          * completion of syscall, but too many callers depend
1350          * on this.
1351          *
1352          * XXX similarly, too many callers depend on
1353          * ext4_new_inode() setting the times, but error
1354          * recovery deletes the inode, so the worst that can
1355          * happen is that the times are slightly out of date
1356          * and/or different from the directory change time.
1357          */
1358         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1359         ext4_update_dx_flag(dir);
1360         dir->i_version++;
1361         ext4_mark_inode_dirty(handle, dir);
1362         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1363         err = ext4_handle_dirty_metadata(handle, dir, bh);
1364         if (err)
1365                 ext4_std_error(dir->i_sb, err);
1366         return 0;
1367 }
1368
1369 /*
1370  * This converts a one block unindexed directory to a 3 block indexed
1371  * directory, and adds the dentry to the indexed directory.
1372  */
1373 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1374                             struct inode *inode, struct buffer_head *bh)
1375 {
1376         struct inode    *dir = dentry->d_parent->d_inode;
1377         const char      *name = dentry->d_name.name;
1378         int             namelen = dentry->d_name.len;
1379         struct buffer_head *bh2;
1380         struct dx_root  *root;
1381         struct dx_frame frames[2], *frame;
1382         struct dx_entry *entries;
1383         struct ext4_dir_entry_2 *de, *de2;
1384         char            *data1, *top;
1385         unsigned        len;
1386         int             retval;
1387         unsigned        blocksize;
1388         struct dx_hash_info hinfo;
1389         ext4_lblk_t  block;
1390         struct fake_dirent *fde;
1391
1392         blocksize =  dir->i_sb->s_blocksize;
1393         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1394         retval = ext4_journal_get_write_access(handle, bh);
1395         if (retval) {
1396                 ext4_std_error(dir->i_sb, retval);
1397                 brelse(bh);
1398                 return retval;
1399         }
1400         root = (struct dx_root *) bh->b_data;
1401
1402         /* The 0th block becomes the root, move the dirents out */
1403         fde = &root->dotdot;
1404         de = (struct ext4_dir_entry_2 *)((char *)fde +
1405                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1406         if ((char *) de >= (((char *) root) + blocksize)) {
1407                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1408                 brelse(bh);
1409                 return -EIO;
1410         }
1411         len = ((char *) root) + blocksize - (char *) de;
1412
1413         /* Allocate new block for the 0th block's dirents */
1414         bh2 = ext4_append(handle, dir, &block, &retval);
1415         if (!(bh2)) {
1416                 brelse(bh);
1417                 return retval;
1418         }
1419         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1420         data1 = bh2->b_data;
1421
1422         memcpy (data1, de, len);
1423         de = (struct ext4_dir_entry_2 *) data1;
1424         top = data1 + len;
1425         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1426                 de = de2;
1427         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1428                                            blocksize);
1429         /* Initialize the root; the dot dirents already exist */
1430         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1431         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1432                                            blocksize);
1433         memset (&root->info, 0, sizeof(root->info));
1434         root->info.info_length = sizeof(root->info);
1435         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1436         entries = root->entries;
1437         dx_set_block(entries, 1);
1438         dx_set_count(entries, 1);
1439         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1440
1441         /* Initialize as for dx_probe */
1442         hinfo.hash_version = root->info.hash_version;
1443         if (hinfo.hash_version <= DX_HASH_TEA)
1444                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1445         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1446         ext4fs_dirhash(name, namelen, &hinfo);
1447         frame = frames;
1448         frame->entries = entries;
1449         frame->at = entries;
1450         frame->bh = bh;
1451         bh = bh2;
1452         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1453         dx_release (frames);
1454         if (!(de))
1455                 return retval;
1456
1457         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1458         brelse(bh);
1459         return retval;
1460 }
1461
1462 /*
1463  *      ext4_add_entry()
1464  *
1465  * adds a file entry to the specified directory, using the same
1466  * semantics as ext4_find_entry(). It returns NULL if it failed.
1467  *
1468  * NOTE!! The inode part of 'de' is left at 0 - which means you
1469  * may not sleep between calling this and putting something into
1470  * the entry, as someone else might have used it while you slept.
1471  */
1472 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1473                           struct inode *inode)
1474 {
1475         struct inode *dir = dentry->d_parent->d_inode;
1476         struct buffer_head *bh;
1477         struct ext4_dir_entry_2 *de;
1478         struct super_block *sb;
1479         int     retval;
1480         int     dx_fallback=0;
1481         unsigned blocksize;
1482         ext4_lblk_t block, blocks;
1483
1484         sb = dir->i_sb;
1485         blocksize = sb->s_blocksize;
1486         if (!dentry->d_name.len)
1487                 return -EINVAL;
1488         if (is_dx(dir)) {
1489                 retval = ext4_dx_add_entry(handle, dentry, inode);
1490                 if (!retval || (retval != ERR_BAD_DX_DIR))
1491                         return retval;
1492                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1493                 dx_fallback++;
1494                 ext4_mark_inode_dirty(handle, dir);
1495         }
1496         blocks = dir->i_size >> sb->s_blocksize_bits;
1497         for (block = 0; block < blocks; block++) {
1498                 bh = ext4_bread(handle, dir, block, 0, &retval);
1499                 if(!bh)
1500                         return retval;
1501                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1502                 if (retval != -ENOSPC) {
1503                         brelse(bh);
1504                         return retval;
1505                 }
1506
1507                 if (blocks == 1 && !dx_fallback &&
1508                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1509                         return make_indexed_dir(handle, dentry, inode, bh);
1510                 brelse(bh);
1511         }
1512         bh = ext4_append(handle, dir, &block, &retval);
1513         if (!bh)
1514                 return retval;
1515         de = (struct ext4_dir_entry_2 *) bh->b_data;
1516         de->inode = 0;
1517         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1518         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1519         brelse(bh);
1520         if (retval == 0)
1521                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1522         return retval;
1523 }
1524
1525 /*
1526  * Returns 0 for success, or a negative error value
1527  */
1528 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1529                              struct inode *inode)
1530 {
1531         struct dx_frame frames[2], *frame;
1532         struct dx_entry *entries, *at;
1533         struct dx_hash_info hinfo;
1534         struct buffer_head *bh;
1535         struct inode *dir = dentry->d_parent->d_inode;
1536         struct super_block *sb = dir->i_sb;
1537         struct ext4_dir_entry_2 *de;
1538         int err;
1539
1540         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1541         if (!frame)
1542                 return err;
1543         entries = frame->entries;
1544         at = frame->at;
1545
1546         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1547                 goto cleanup;
1548
1549         BUFFER_TRACE(bh, "get_write_access");
1550         err = ext4_journal_get_write_access(handle, bh);
1551         if (err)
1552                 goto journal_error;
1553
1554         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1555         if (err != -ENOSPC)
1556                 goto cleanup;
1557
1558         /* Block full, should compress but for now just split */
1559         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1560                        dx_get_count(entries), dx_get_limit(entries)));
1561         /* Need to split index? */
1562         if (dx_get_count(entries) == dx_get_limit(entries)) {
1563                 ext4_lblk_t newblock;
1564                 unsigned icount = dx_get_count(entries);
1565                 int levels = frame - frames;
1566                 struct dx_entry *entries2;
1567                 struct dx_node *node2;
1568                 struct buffer_head *bh2;
1569
1570                 if (levels && (dx_get_count(frames->entries) ==
1571                                dx_get_limit(frames->entries))) {
1572                         ext4_warning(sb, "Directory index full!");
1573                         err = -ENOSPC;
1574                         goto cleanup;
1575                 }
1576                 bh2 = ext4_append (handle, dir, &newblock, &err);
1577                 if (!(bh2))
1578                         goto cleanup;
1579                 node2 = (struct dx_node *)(bh2->b_data);
1580                 entries2 = node2->entries;
1581                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1582                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1583                                                            sb->s_blocksize);
1584                 BUFFER_TRACE(frame->bh, "get_write_access");
1585                 err = ext4_journal_get_write_access(handle, frame->bh);
1586                 if (err)
1587                         goto journal_error;
1588                 if (levels) {
1589                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1590                         unsigned hash2 = dx_get_hash(entries + icount1);
1591                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1592                                        icount1, icount2));
1593
1594                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1595                         err = ext4_journal_get_write_access(handle,
1596                                                              frames[0].bh);
1597                         if (err)
1598                                 goto journal_error;
1599
1600                         memcpy((char *) entries2, (char *) (entries + icount1),
1601                                icount2 * sizeof(struct dx_entry));
1602                         dx_set_count(entries, icount1);
1603                         dx_set_count(entries2, icount2);
1604                         dx_set_limit(entries2, dx_node_limit(dir));
1605
1606                         /* Which index block gets the new entry? */
1607                         if (at - entries >= icount1) {
1608                                 frame->at = at = at - entries - icount1 + entries2;
1609                                 frame->entries = entries = entries2;
1610                                 swap(frame->bh, bh2);
1611                         }
1612                         dx_insert_block(frames + 0, hash2, newblock);
1613                         dxtrace(dx_show_index("node", frames[1].entries));
1614                         dxtrace(dx_show_index("node",
1615                                ((struct dx_node *) bh2->b_data)->entries));
1616                         err = ext4_handle_dirty_metadata(handle, inode, bh2);
1617                         if (err)
1618                                 goto journal_error;
1619                         brelse (bh2);
1620                 } else {
1621                         dxtrace(printk(KERN_DEBUG
1622                                        "Creating second level index...\n"));
1623                         memcpy((char *) entries2, (char *) entries,
1624                                icount * sizeof(struct dx_entry));
1625                         dx_set_limit(entries2, dx_node_limit(dir));
1626
1627                         /* Set up root */
1628                         dx_set_count(entries, 1);
1629                         dx_set_block(entries + 0, newblock);
1630                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1631
1632                         /* Add new access path frame */
1633                         frame = frames + 1;
1634                         frame->at = at = at - entries + entries2;
1635                         frame->entries = entries = entries2;
1636                         frame->bh = bh2;
1637                         err = ext4_journal_get_write_access(handle,
1638                                                              frame->bh);
1639                         if (err)
1640                                 goto journal_error;
1641                 }
1642                 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1643         }
1644         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1645         if (!de)
1646                 goto cleanup;
1647         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1648         goto cleanup;
1649
1650 journal_error:
1651         ext4_std_error(dir->i_sb, err);
1652 cleanup:
1653         if (bh)
1654                 brelse(bh);
1655         dx_release(frames);
1656         return err;
1657 }
1658
1659 /*
1660  * ext4_delete_entry deletes a directory entry by merging it with the
1661  * previous entry
1662  */
1663 static int ext4_delete_entry(handle_t *handle,
1664                              struct inode *dir,
1665                              struct ext4_dir_entry_2 *de_del,
1666                              struct buffer_head *bh)
1667 {
1668         struct ext4_dir_entry_2 *de, *pde;
1669         unsigned int blocksize = dir->i_sb->s_blocksize;
1670         int i;
1671
1672         i = 0;
1673         pde = NULL;
1674         de = (struct ext4_dir_entry_2 *) bh->b_data;
1675         while (i < bh->b_size) {
1676                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1677                         return -EIO;
1678                 if (de == de_del)  {
1679                         BUFFER_TRACE(bh, "get_write_access");
1680                         ext4_journal_get_write_access(handle, bh);
1681                         if (pde)
1682                                 pde->rec_len = ext4_rec_len_to_disk(
1683                                         ext4_rec_len_from_disk(pde->rec_len,
1684                                                                blocksize) +
1685                                         ext4_rec_len_from_disk(de->rec_len,
1686                                                                blocksize),
1687                                         blocksize);
1688                         else
1689                                 de->inode = 0;
1690                         dir->i_version++;
1691                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1692                         ext4_handle_dirty_metadata(handle, dir, bh);
1693                         return 0;
1694                 }
1695                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1696                 pde = de;
1697                 de = ext4_next_entry(de, blocksize);
1698         }
1699         return -ENOENT;
1700 }
1701
1702 /*
1703  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1704  * since this indicates that nlinks count was previously 1.
1705  */
1706 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1707 {
1708         inc_nlink(inode);
1709         if (is_dx(inode) && inode->i_nlink > 1) {
1710                 /* limit is 16-bit i_links_count */
1711                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1712                         inode->i_nlink = 1;
1713                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1714                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1715                 }
1716         }
1717 }
1718
1719 /*
1720  * If a directory had nlink == 1, then we should let it be 1. This indicates
1721  * directory has >EXT4_LINK_MAX subdirs.
1722  */
1723 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1724 {
1725         drop_nlink(inode);
1726         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1727                 inc_nlink(inode);
1728 }
1729
1730
1731 static int ext4_add_nondir(handle_t *handle,
1732                 struct dentry *dentry, struct inode *inode)
1733 {
1734         int err = ext4_add_entry(handle, dentry, inode);
1735         if (!err) {
1736                 ext4_mark_inode_dirty(handle, inode);
1737                 d_instantiate(dentry, inode);
1738                 unlock_new_inode(inode);
1739                 return 0;
1740         }
1741         drop_nlink(inode);
1742         unlock_new_inode(inode);
1743         iput(inode);
1744         return err;
1745 }
1746
1747 /*
1748  * By the time this is called, we already have created
1749  * the directory cache entry for the new file, but it
1750  * is so far negative - it has no inode.
1751  *
1752  * If the create succeeds, we fill in the inode information
1753  * with d_instantiate().
1754  */
1755 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1756                        struct nameidata *nd)
1757 {
1758         handle_t *handle;
1759         struct inode *inode;
1760         int err, retries = 0;
1761
1762         dquot_initialize(dir);
1763
1764 retry:
1765         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1766                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1767                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1768         if (IS_ERR(handle))
1769                 return PTR_ERR(handle);
1770
1771         if (IS_DIRSYNC(dir))
1772                 ext4_handle_sync(handle);
1773
1774         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1775         err = PTR_ERR(inode);
1776         if (!IS_ERR(inode)) {
1777                 inode->i_op = &ext4_file_inode_operations;
1778                 inode->i_fop = &ext4_file_operations;
1779                 ext4_set_aops(inode);
1780                 err = ext4_add_nondir(handle, dentry, inode);
1781         }
1782         ext4_journal_stop(handle);
1783         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1784                 goto retry;
1785         return err;
1786 }
1787
1788 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1789                       int mode, dev_t rdev)
1790 {
1791         handle_t *handle;
1792         struct inode *inode;
1793         int err, retries = 0;
1794
1795         if (!new_valid_dev(rdev))
1796                 return -EINVAL;
1797
1798         dquot_initialize(dir);
1799
1800 retry:
1801         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1802                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1803                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1804         if (IS_ERR(handle))
1805                 return PTR_ERR(handle);
1806
1807         if (IS_DIRSYNC(dir))
1808                 ext4_handle_sync(handle);
1809
1810         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1811         err = PTR_ERR(inode);
1812         if (!IS_ERR(inode)) {
1813                 init_special_inode(inode, inode->i_mode, rdev);
1814 #ifdef CONFIG_EXT4_FS_XATTR
1815                 inode->i_op = &ext4_special_inode_operations;
1816 #endif
1817                 err = ext4_add_nondir(handle, dentry, inode);
1818         }
1819         ext4_journal_stop(handle);
1820         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1821                 goto retry;
1822         return err;
1823 }
1824
1825 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1826 {
1827         handle_t *handle;
1828         struct inode *inode;
1829         struct buffer_head *dir_block;
1830         struct ext4_dir_entry_2 *de;
1831         unsigned int blocksize = dir->i_sb->s_blocksize;
1832         int err, retries = 0;
1833
1834         if (EXT4_DIR_LINK_MAX(dir))
1835                 return -EMLINK;
1836
1837         dquot_initialize(dir);
1838
1839 retry:
1840         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1841                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1842                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1843         if (IS_ERR(handle))
1844                 return PTR_ERR(handle);
1845
1846         if (IS_DIRSYNC(dir))
1847                 ext4_handle_sync(handle);
1848
1849         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1850                                &dentry->d_name, 0);
1851         err = PTR_ERR(inode);
1852         if (IS_ERR(inode))
1853                 goto out_stop;
1854
1855         inode->i_op = &ext4_dir_inode_operations;
1856         inode->i_fop = &ext4_dir_operations;
1857         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1858         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1859         if (!dir_block)
1860                 goto out_clear_inode;
1861         BUFFER_TRACE(dir_block, "get_write_access");
1862         ext4_journal_get_write_access(handle, dir_block);
1863         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1864         de->inode = cpu_to_le32(inode->i_ino);
1865         de->name_len = 1;
1866         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1867                                            blocksize);
1868         strcpy(de->name, ".");
1869         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1870         de = ext4_next_entry(de, blocksize);
1871         de->inode = cpu_to_le32(dir->i_ino);
1872         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1873                                            blocksize);
1874         de->name_len = 2;
1875         strcpy(de->name, "..");
1876         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1877         inode->i_nlink = 2;
1878         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1879         ext4_handle_dirty_metadata(handle, dir, dir_block);
1880         brelse(dir_block);
1881         ext4_mark_inode_dirty(handle, inode);
1882         err = ext4_add_entry(handle, dentry, inode);
1883         if (err) {
1884 out_clear_inode:
1885                 clear_nlink(inode);
1886                 unlock_new_inode(inode);
1887                 ext4_mark_inode_dirty(handle, inode);
1888                 iput(inode);
1889                 goto out_stop;
1890         }
1891         ext4_inc_count(handle, dir);
1892         ext4_update_dx_flag(dir);
1893         ext4_mark_inode_dirty(handle, dir);
1894         d_instantiate(dentry, inode);
1895         unlock_new_inode(inode);
1896 out_stop:
1897         ext4_journal_stop(handle);
1898         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1899                 goto retry;
1900         return err;
1901 }
1902
1903 /*
1904  * routine to check that the specified directory is empty (for rmdir)
1905  */
1906 static int empty_dir(struct inode *inode)
1907 {
1908         unsigned int offset;
1909         struct buffer_head *bh;
1910         struct ext4_dir_entry_2 *de, *de1;
1911         struct super_block *sb;
1912         int err = 0;
1913
1914         sb = inode->i_sb;
1915         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1916             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1917                 if (err)
1918                         EXT4_ERROR_INODE(inode,
1919                                 "error %d reading directory lblock 0", err);
1920                 else
1921                         ext4_warning(inode->i_sb,
1922                                      "bad directory (dir #%lu) - no data block",
1923                                      inode->i_ino);
1924                 return 1;
1925         }
1926         de = (struct ext4_dir_entry_2 *) bh->b_data;
1927         de1 = ext4_next_entry(de, sb->s_blocksize);
1928         if (le32_to_cpu(de->inode) != inode->i_ino ||
1929                         !le32_to_cpu(de1->inode) ||
1930                         strcmp(".", de->name) ||
1931                         strcmp("..", de1->name)) {
1932                 ext4_warning(inode->i_sb,
1933                              "bad directory (dir #%lu) - no `.' or `..'",
1934                              inode->i_ino);
1935                 brelse(bh);
1936                 return 1;
1937         }
1938         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1939                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1940         de = ext4_next_entry(de1, sb->s_blocksize);
1941         while (offset < inode->i_size) {
1942                 if (!bh ||
1943                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1944                         unsigned int lblock;
1945                         err = 0;
1946                         brelse(bh);
1947                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1948                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
1949                         if (!bh) {
1950                                 if (err)
1951                                         EXT4_ERROR_INODE(inode,
1952                                                 "error %d reading directory "
1953                                                 "lblock %u", err, lblock);
1954                                 offset += sb->s_blocksize;
1955                                 continue;
1956                         }
1957                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1958                 }
1959                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1960                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1961                                                          sb->s_blocksize);
1962                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1963                         continue;
1964                 }
1965                 if (le32_to_cpu(de->inode)) {
1966                         brelse(bh);
1967                         return 0;
1968                 }
1969                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1970                 de = ext4_next_entry(de, sb->s_blocksize);
1971         }
1972         brelse(bh);
1973         return 1;
1974 }
1975
1976 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1977  * such inodes, starting at the superblock, in case we crash before the
1978  * file is closed/deleted, or in case the inode truncate spans multiple
1979  * transactions and the last transaction is not recovered after a crash.
1980  *
1981  * At filesystem recovery time, we walk this list deleting unlinked
1982  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1983  */
1984 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1985 {
1986         struct super_block *sb = inode->i_sb;
1987         struct ext4_iloc iloc;
1988         int err = 0, rc;
1989
1990         if (!ext4_handle_valid(handle))
1991                 return 0;
1992
1993         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1994         if (!list_empty(&EXT4_I(inode)->i_orphan))
1995                 goto out_unlock;
1996
1997         /* Orphan handling is only valid for files with data blocks
1998          * being truncated, or files being unlinked. */
1999
2000         /* @@@ FIXME: Observation from aviro:
2001          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
2002          * here (on s_orphan_lock), so race with ext4_link() which might bump
2003          * ->i_nlink. For, say it, character device. Not a regular file,
2004          * not a directory, not a symlink and ->i_nlink > 0.
2005          *
2006          * tytso, 4/25/2009: I'm not sure how that could happen;
2007          * shouldn't the fs core protect us from these sort of
2008          * unlink()/link() races?
2009          */
2010         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2011                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2012
2013         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2014         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2015         if (err)
2016                 goto out_unlock;
2017
2018         err = ext4_reserve_inode_write(handle, inode, &iloc);
2019         if (err)
2020                 goto out_unlock;
2021         /*
2022          * Due to previous errors inode may be already a part of on-disk
2023          * orphan list. If so skip on-disk list modification.
2024          */
2025         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2026                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2027                         goto mem_insert;
2028
2029         /* Insert this inode at the head of the on-disk orphan list... */
2030         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2031         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2032         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2033         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2034         if (!err)
2035                 err = rc;
2036
2037         /* Only add to the head of the in-memory list if all the
2038          * previous operations succeeded.  If the orphan_add is going to
2039          * fail (possibly taking the journal offline), we can't risk
2040          * leaving the inode on the orphan list: stray orphan-list
2041          * entries can cause panics at unmount time.
2042          *
2043          * This is safe: on error we're going to ignore the orphan list
2044          * anyway on the next recovery. */
2045 mem_insert:
2046         if (!err)
2047                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2048
2049         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2050         jbd_debug(4, "orphan inode %lu will point to %d\n",
2051                         inode->i_ino, NEXT_ORPHAN(inode));
2052 out_unlock:
2053         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2054         ext4_std_error(inode->i_sb, err);
2055         return err;
2056 }
2057
2058 /*
2059  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2060  * of such inodes stored on disk, because it is finally being cleaned up.
2061  */
2062 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2063 {
2064         struct list_head *prev;
2065         struct ext4_inode_info *ei = EXT4_I(inode);
2066         struct ext4_sb_info *sbi;
2067         __u32 ino_next;
2068         struct ext4_iloc iloc;
2069         int err = 0;
2070
2071         /* ext4_handle_valid() assumes a valid handle_t pointer */
2072         if (handle && !ext4_handle_valid(handle))
2073                 return 0;
2074
2075         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2076         if (list_empty(&ei->i_orphan))
2077                 goto out;
2078
2079         ino_next = NEXT_ORPHAN(inode);
2080         prev = ei->i_orphan.prev;
2081         sbi = EXT4_SB(inode->i_sb);
2082
2083         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2084
2085         list_del_init(&ei->i_orphan);
2086
2087         /* If we're on an error path, we may not have a valid
2088          * transaction handle with which to update the orphan list on
2089          * disk, but we still need to remove the inode from the linked
2090          * list in memory. */
2091         if (sbi->s_journal && !handle)
2092                 goto out;
2093
2094         err = ext4_reserve_inode_write(handle, inode, &iloc);
2095         if (err)
2096                 goto out_err;
2097
2098         if (prev == &sbi->s_orphan) {
2099                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2100                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2101                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2102                 if (err)
2103                         goto out_brelse;
2104                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2105                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2106         } else {
2107                 struct ext4_iloc iloc2;
2108                 struct inode *i_prev =
2109                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2110
2111                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2112                           i_prev->i_ino, ino_next);
2113                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2114                 if (err)
2115                         goto out_brelse;
2116                 NEXT_ORPHAN(i_prev) = ino_next;
2117                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2118         }
2119         if (err)
2120                 goto out_brelse;
2121         NEXT_ORPHAN(inode) = 0;
2122         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2123
2124 out_err:
2125         ext4_std_error(inode->i_sb, err);
2126 out:
2127         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2128         return err;
2129
2130 out_brelse:
2131         brelse(iloc.bh);
2132         goto out_err;
2133 }
2134
2135 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2136 {
2137         int retval;
2138         struct inode *inode;
2139         struct buffer_head *bh;
2140         struct ext4_dir_entry_2 *de;
2141         handle_t *handle;
2142
2143         /* Initialize quotas before so that eventual writes go in
2144          * separate transaction */
2145         dquot_initialize(dir);
2146         dquot_initialize(dentry->d_inode);
2147
2148         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2149         if (IS_ERR(handle))
2150                 return PTR_ERR(handle);
2151
2152         retval = -ENOENT;
2153         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2154         if (!bh)
2155                 goto end_rmdir;
2156
2157         if (IS_DIRSYNC(dir))
2158                 ext4_handle_sync(handle);
2159
2160         inode = dentry->d_inode;
2161
2162         retval = -EIO;
2163         if (le32_to_cpu(de->inode) != inode->i_ino)
2164                 goto end_rmdir;
2165
2166         retval = -ENOTEMPTY;
2167         if (!empty_dir(inode))
2168                 goto end_rmdir;
2169
2170         retval = ext4_delete_entry(handle, dir, de, bh);
2171         if (retval)
2172                 goto end_rmdir;
2173         if (!EXT4_DIR_LINK_EMPTY(inode))
2174                 ext4_warning(inode->i_sb,
2175                              "empty directory has too many links (%d)",
2176                              inode->i_nlink);
2177         inode->i_version++;
2178         clear_nlink(inode);
2179         /* There's no need to set i_disksize: the fact that i_nlink is
2180          * zero will ensure that the right thing happens during any
2181          * recovery. */
2182         inode->i_size = 0;
2183         ext4_orphan_add(handle, inode);
2184         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2185         ext4_mark_inode_dirty(handle, inode);
2186         ext4_dec_count(handle, dir);
2187         ext4_update_dx_flag(dir);
2188         ext4_mark_inode_dirty(handle, dir);
2189
2190 end_rmdir:
2191         ext4_journal_stop(handle);
2192         brelse(bh);
2193         return retval;
2194 }
2195
2196 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2197 {
2198         int retval;
2199         struct inode *inode;
2200         struct buffer_head *bh;
2201         struct ext4_dir_entry_2 *de;
2202         handle_t *handle;
2203
2204         /* Initialize quotas before so that eventual writes go
2205          * in separate transaction */
2206         dquot_initialize(dir);
2207         dquot_initialize(dentry->d_inode);
2208
2209         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2210         if (IS_ERR(handle))
2211                 return PTR_ERR(handle);
2212
2213         if (IS_DIRSYNC(dir))
2214                 ext4_handle_sync(handle);
2215
2216         retval = -ENOENT;
2217         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2218         if (!bh)
2219                 goto end_unlink;
2220
2221         inode = dentry->d_inode;
2222
2223         retval = -EIO;
2224         if (le32_to_cpu(de->inode) != inode->i_ino)
2225                 goto end_unlink;
2226
2227         if (!inode->i_nlink) {
2228                 ext4_warning(inode->i_sb,
2229                              "Deleting nonexistent file (%lu), %d",
2230                              inode->i_ino, inode->i_nlink);
2231                 inode->i_nlink = 1;
2232         }
2233         retval = ext4_delete_entry(handle, dir, de, bh);
2234         if (retval)
2235                 goto end_unlink;
2236         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2237         ext4_update_dx_flag(dir);
2238         ext4_mark_inode_dirty(handle, dir);
2239         drop_nlink(inode);
2240         if (!inode->i_nlink)
2241                 ext4_orphan_add(handle, inode);
2242         inode->i_ctime = ext4_current_time(inode);
2243         ext4_mark_inode_dirty(handle, inode);
2244         retval = 0;
2245
2246 end_unlink:
2247         ext4_journal_stop(handle);
2248         brelse(bh);
2249         return retval;
2250 }
2251
2252 static int ext4_symlink(struct inode *dir,
2253                         struct dentry *dentry, const char *symname)
2254 {
2255         handle_t *handle;
2256         struct inode *inode;
2257         int l, err, retries = 0;
2258
2259         l = strlen(symname)+1;
2260         if (l > dir->i_sb->s_blocksize)
2261                 return -ENAMETOOLONG;
2262
2263         dquot_initialize(dir);
2264
2265 retry:
2266         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2267                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2268                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2269         if (IS_ERR(handle))
2270                 return PTR_ERR(handle);
2271
2272         if (IS_DIRSYNC(dir))
2273                 ext4_handle_sync(handle);
2274
2275         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2276                                &dentry->d_name, 0);
2277         err = PTR_ERR(inode);
2278         if (IS_ERR(inode))
2279                 goto out_stop;
2280
2281         if (l > sizeof(EXT4_I(inode)->i_data)) {
2282                 inode->i_op = &ext4_symlink_inode_operations;
2283                 ext4_set_aops(inode);
2284                 /*
2285                  * page_symlink() calls into ext4_prepare/commit_write.
2286                  * We have a transaction open.  All is sweetness.  It also sets
2287                  * i_size in generic_commit_write().
2288                  */
2289                 err = __page_symlink(inode, symname, l, 1);
2290                 if (err) {
2291                         clear_nlink(inode);
2292                         unlock_new_inode(inode);
2293                         ext4_mark_inode_dirty(handle, inode);
2294                         iput(inode);
2295                         goto out_stop;
2296                 }
2297         } else {
2298                 /* clear the extent format for fast symlink */
2299                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2300                 inode->i_op = &ext4_fast_symlink_inode_operations;
2301                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2302                 inode->i_size = l-1;
2303         }
2304         EXT4_I(inode)->i_disksize = inode->i_size;
2305         err = ext4_add_nondir(handle, dentry, inode);
2306 out_stop:
2307         ext4_journal_stop(handle);
2308         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2309                 goto retry;
2310         return err;
2311 }
2312
2313 static int ext4_link(struct dentry *old_dentry,
2314                      struct inode *dir, struct dentry *dentry)
2315 {
2316         handle_t *handle;
2317         struct inode *inode = old_dentry->d_inode;
2318         int err, retries = 0;
2319
2320         if (inode->i_nlink >= EXT4_LINK_MAX)
2321                 return -EMLINK;
2322
2323         dquot_initialize(dir);
2324
2325         /*
2326          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2327          * otherwise has the potential to corrupt the orphan inode list.
2328          */
2329         if (inode->i_nlink == 0)
2330                 return -ENOENT;
2331
2332 retry:
2333         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2334                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2335         if (IS_ERR(handle))
2336                 return PTR_ERR(handle);
2337
2338         if (IS_DIRSYNC(dir))
2339                 ext4_handle_sync(handle);
2340
2341         inode->i_ctime = ext4_current_time(inode);
2342         ext4_inc_count(handle, inode);
2343         atomic_inc(&inode->i_count);
2344
2345         err = ext4_add_entry(handle, dentry, inode);
2346         if (!err) {
2347                 ext4_mark_inode_dirty(handle, inode);
2348                 d_instantiate(dentry, inode);
2349         } else {
2350                 drop_nlink(inode);
2351                 iput(inode);
2352         }
2353         ext4_journal_stop(handle);
2354         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2355                 goto retry;
2356         return err;
2357 }
2358
2359 #define PARENT_INO(buffer, size) \
2360         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2361
2362 /*
2363  * Anybody can rename anything with this: the permission checks are left to the
2364  * higher-level routines.
2365  */
2366 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2367                        struct inode *new_dir, struct dentry *new_dentry)
2368 {
2369         handle_t *handle;
2370         struct inode *old_inode, *new_inode;
2371         struct buffer_head *old_bh, *new_bh, *dir_bh;
2372         struct ext4_dir_entry_2 *old_de, *new_de;
2373         int retval, force_da_alloc = 0;
2374
2375         dquot_initialize(old_dir);
2376         dquot_initialize(new_dir);
2377
2378         old_bh = new_bh = dir_bh = NULL;
2379
2380         /* Initialize quotas before so that eventual writes go
2381          * in separate transaction */
2382         if (new_dentry->d_inode)
2383                 dquot_initialize(new_dentry->d_inode);
2384         handle = ext4_journal_start(old_dir, 2 *
2385                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2386                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2387         if (IS_ERR(handle))
2388                 return PTR_ERR(handle);
2389
2390         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2391                 ext4_handle_sync(handle);
2392
2393         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2394         /*
2395          *  Check for inode number is _not_ due to possible IO errors.
2396          *  We might rmdir the source, keep it as pwd of some process
2397          *  and merrily kill the link to whatever was created under the
2398          *  same name. Goodbye sticky bit ;-<
2399          */
2400         old_inode = old_dentry->d_inode;
2401         retval = -ENOENT;
2402         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2403                 goto end_rename;
2404
2405         new_inode = new_dentry->d_inode;
2406         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2407         if (new_bh) {
2408                 if (!new_inode) {
2409                         brelse(new_bh);
2410                         new_bh = NULL;
2411                 }
2412         }
2413         if (S_ISDIR(old_inode->i_mode)) {
2414                 if (new_inode) {
2415                         retval = -ENOTEMPTY;
2416                         if (!empty_dir(new_inode))
2417                                 goto end_rename;
2418                 }
2419                 retval = -EIO;
2420                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2421                 if (!dir_bh)
2422                         goto end_rename;
2423                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2424                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2425                         goto end_rename;
2426                 retval = -EMLINK;
2427                 if (!new_inode && new_dir != old_dir &&
2428                     EXT4_DIR_LINK_MAX(new_dir))
2429                         goto end_rename;
2430         }
2431         if (!new_bh) {
2432                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2433                 if (retval)
2434                         goto end_rename;
2435         } else {
2436                 BUFFER_TRACE(new_bh, "get write access");
2437                 ext4_journal_get_write_access(handle, new_bh);
2438                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2439                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2440                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2441                         new_de->file_type = old_de->file_type;
2442                 new_dir->i_version++;
2443                 new_dir->i_ctime = new_dir->i_mtime =
2444                                         ext4_current_time(new_dir);
2445                 ext4_mark_inode_dirty(handle, new_dir);
2446                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2447                 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2448                 brelse(new_bh);
2449                 new_bh = NULL;
2450         }
2451
2452         /*
2453          * Like most other Unix systems, set the ctime for inodes on a
2454          * rename.
2455          */
2456         old_inode->i_ctime = ext4_current_time(old_inode);
2457         ext4_mark_inode_dirty(handle, old_inode);
2458
2459         /*
2460          * ok, that's it
2461          */
2462         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2463             old_de->name_len != old_dentry->d_name.len ||
2464             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2465             (retval = ext4_delete_entry(handle, old_dir,
2466                                         old_de, old_bh)) == -ENOENT) {
2467                 /* old_de could have moved from under us during htree split, so
2468                  * make sure that we are deleting the right entry.  We might
2469                  * also be pointing to a stale entry in the unused part of
2470                  * old_bh so just checking inum and the name isn't enough. */
2471                 struct buffer_head *old_bh2;
2472                 struct ext4_dir_entry_2 *old_de2;
2473
2474                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2475                 if (old_bh2) {
2476                         retval = ext4_delete_entry(handle, old_dir,
2477                                                    old_de2, old_bh2);
2478                         brelse(old_bh2);
2479                 }
2480         }
2481         if (retval) {
2482                 ext4_warning(old_dir->i_sb,
2483                                 "Deleting old file (%lu), %d, error=%d",
2484                                 old_dir->i_ino, old_dir->i_nlink, retval);
2485         }
2486
2487         if (new_inode) {
2488                 ext4_dec_count(handle, new_inode);
2489                 new_inode->i_ctime = ext4_current_time(new_inode);
2490         }
2491         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2492         ext4_update_dx_flag(old_dir);
2493         if (dir_bh) {
2494                 BUFFER_TRACE(dir_bh, "get_write_access");
2495                 ext4_journal_get_write_access(handle, dir_bh);
2496                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2497                                                 cpu_to_le32(new_dir->i_ino);
2498                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2499                 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2500                 ext4_dec_count(handle, old_dir);
2501                 if (new_inode) {
2502                         /* checked empty_dir above, can't have another parent,
2503                          * ext4_dec_count() won't work for many-linked dirs */
2504                         new_inode->i_nlink = 0;
2505                 } else {
2506                         ext4_inc_count(handle, new_dir);
2507                         ext4_update_dx_flag(new_dir);
2508                         ext4_mark_inode_dirty(handle, new_dir);
2509                 }
2510         }
2511         ext4_mark_inode_dirty(handle, old_dir);
2512         if (new_inode) {
2513                 ext4_mark_inode_dirty(handle, new_inode);
2514                 if (!new_inode->i_nlink)
2515                         ext4_orphan_add(handle, new_inode);
2516                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2517                         force_da_alloc = 1;
2518         }
2519         retval = 0;
2520
2521 end_rename:
2522         brelse(dir_bh);
2523         brelse(old_bh);
2524         brelse(new_bh);
2525         ext4_journal_stop(handle);
2526         if (retval == 0 && force_da_alloc)
2527                 ext4_alloc_da_blocks(old_inode);
2528         return retval;
2529 }
2530
2531 /*
2532  * directories can handle most operations...
2533  */
2534 const struct inode_operations ext4_dir_inode_operations = {
2535         .create         = ext4_create,
2536         .lookup         = ext4_lookup,
2537         .link           = ext4_link,
2538         .unlink         = ext4_unlink,
2539         .symlink        = ext4_symlink,
2540         .mkdir          = ext4_mkdir,
2541         .rmdir          = ext4_rmdir,
2542         .mknod          = ext4_mknod,
2543         .rename         = ext4_rename,
2544         .setattr        = ext4_setattr,
2545 #ifdef CONFIG_EXT4_FS_XATTR
2546         .setxattr       = generic_setxattr,
2547         .getxattr       = generic_getxattr,
2548         .listxattr      = ext4_listxattr,
2549         .removexattr    = generic_removexattr,
2550 #endif
2551         .check_acl      = ext4_check_acl,
2552         .fiemap         = ext4_fiemap,
2553 };
2554
2555 const struct inode_operations ext4_special_inode_operations = {
2556         .setattr        = ext4_setattr,
2557 #ifdef CONFIG_EXT4_FS_XATTR
2558         .setxattr       = generic_setxattr,
2559         .getxattr       = generic_getxattr,
2560         .listxattr      = ext4_listxattr,
2561         .removexattr    = generic_removexattr,
2562 #endif
2563         .check_acl      = ext4_check_acl,
2564 };