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, __func__,
 387                              "Unrecognised inode hash code %d",
 388                              root->info.hash_version);
 389                 brelse(bh);
 390                 *err = ERR_BAD_DX_DIR;
 391                 goto fail;
 392         }
 393         hinfo->hash_version = root->info.hash_version;
 394         if (hinfo->hash_version <= DX_HASH_TEA)
 395                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 396         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 397         if (d_name)
 398                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 399         hash = hinfo->hash;
 400
 401         if (root->info.unused_flags & 1) {
 402                 ext4_warning(dir->i_sb, __func__,
 403                              "Unimplemented inode hash flags: %#06x",
 404                              root->info.unused_flags);
 405                 brelse(bh);
 406                 *err = ERR_BAD_DX_DIR;
 407                 goto fail;
 408         }
 409
 410         if ((indirect = root->info.indirect_levels) > 1) {
 411                 ext4_warning(dir->i_sb, __func__,
 412                              "Unimplemented inode hash depth: %#06x",
 413                              root->info.indirect_levels);
 414                 brelse(bh);
 415                 *err = ERR_BAD_DX_DIR;
 416                 goto fail;
 417         }
 418
 419         entries = (struct dx_entry *) (((char *)&root->info) +
 420                                        root->info.info_length);
 421
 422         if (dx_get_limit(entries) != dx_root_limit(dir,
 423                                                    root->info.info_length)) {
 424                 ext4_warning(dir->i_sb, __func__,
 425                              "dx entry: limit != root limit");
 426                 brelse(bh);
 427                 *err = ERR_BAD_DX_DIR;
 428                 goto fail;
 429         }
 430
 431         dxtrace(printk("Look up %x", hash));
 432         while (1)
 433         {
 434                 count = dx_get_count(entries);
 435                 if (!count || count > dx_get_limit(entries)) {
 436                         ext4_warning(dir->i_sb, __func__,
 437                                      "dx entry: no count or count > limit");
 438                         brelse(bh);
 439                         *err = ERR_BAD_DX_DIR;
 440                         goto fail2;
 441                 }
 442
 443                 p = entries + 1;
 444                 q = entries + count - 1;
 445                 while (p <= q)
 446                 {
 447                         m = p + (q - p)/2;
 448                         dxtrace(printk("."));
 449                         if (dx_get_hash(m) > hash)
 450                                 q = m - 1;
 451                         else
 452                                 p = m + 1;
 453                 }
 454
 455                 if (0) // linear search cross check
 456                 {
 457                         unsigned n = count - 1;
 458                         at = entries;
 459                         while (n--)
 460                         {
 461                                 dxtrace(printk(","));
 462                                 if (dx_get_hash(++at) > hash)
 463                                 {
 464                                         at--;
 465                                         break;
 466                                 }
 467                         }
 468                         assert (at == p - 1);
 469                 }
 470
 471                 at = p - 1;
 472                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 473                 frame->bh = bh;
 474                 frame->entries = entries;
 475                 frame->at = at;
 476                 if (!indirect--) return frame;
 477                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 478                         goto fail2;
 479                 at = entries = ((struct dx_node *) bh->b_data)->entries;
 480                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 481                         ext4_warning(dir->i_sb, __func__,
 482                                      "dx entry: limit != node limit");
 483                         brelse(bh);
 484                         *err = ERR_BAD_DX_DIR;
 485                         goto fail2;
 486                 }
 487                 frame++;
 488                 frame->bh = NULL;
 489         }
 490 fail2:
 491         while (frame >= frame_in) {
 492                 brelse(frame->bh);
 493                 frame--;
 494         }
 495 fail:
 496         if (*err == ERR_BAD_DX_DIR)
 497                 ext4_warning(dir->i_sb, __func__,
 498                              "Corrupt dir inode %ld, running e2fsck is "
 499                              "recommended.", dir->i_ino);
 500         return NULL;
 501 }
 502
 503 static void dx_release (struct dx_frame *frames)
 504 {
 505         if (frames[0].bh == NULL)
 506                 return;
 507
 508         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 509                 brelse(frames[1].bh);
 510         brelse(frames[0].bh);
 511 }
 512
 513 /*
 514  * This function increments the frame pointer to search the next leaf
 515  * block, and reads in the necessary intervening nodes if the search
 516  * should be necessary.  Whether or not the search is necessary is
 517  * controlled by the hash parameter.  If the hash value is even, then
 518  * the search is only continued if the next block starts with that
 519  * hash value.  This is used if we are searching for a specific file.
 520  *
 521  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 522  *
 523  * This function returns 1 if the caller should continue to search,
 524  * or 0 if it should not.  If there is an error reading one of the
 525  * index blocks, it will a negative error code.
 526  *
 527  * If start_hash is non-null, it will be filled in with the starting
 528  * hash of the next page.
 529  */
 530 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 531                                  struct dx_frame *frame,
 532                                  struct dx_frame *frames,
 533                                  __u32 *start_hash)
 534 {
 535         struct dx_frame *p;
 536         struct buffer_head *bh;
 537         int err, num_frames = 0;
 538         __u32 bhash;
 539
 540         p = frame;
 541         /*
 542          * Find the next leaf page by incrementing the frame pointer.
 543          * If we run out of entries in the interior node, loop around and
 544          * increment pointer in the parent node.  When we break out of
 545          * this loop, num_frames indicates the number of interior
 546          * nodes need to be read.
 547          */
 548         while (1) {
 549                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 550                         break;
 551                 if (p == frames)
 552                         return 0;
 553                 num_frames++;
 554                 p--;
 555         }
 556
 557         /*
 558          * If the hash is 1, then continue only if the next page has a
 559          * continuation hash of any value.  This is used for readdir
 560          * handling.  Otherwise, check to see if the hash matches the
 561          * desired contiuation hash.  If it doesn't, return since
 562          * there's no point to read in the successive index pages.
 563          */
 564         bhash = dx_get_hash(p->at);
 565         if (start_hash)
 566                 *start_hash = bhash;
 567         if ((hash & 1) == 0) {
 568                 if ((bhash & ~1) != hash)
 569                         return 0;
 570         }
 571         /*
 572          * If the hash is HASH_NB_ALWAYS, we always go to the next
 573          * block so no check is necessary
 574          */
 575         while (num_frames--) {
 576                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 577                                       0, &err)))
 578                         return err; /* Failure */
 579                 p++;
 580                 brelse(p->bh);
 581                 p->bh = bh;
 582                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 583         }
 584         return 1;
 585 }
 586
 587
 588 /*
 589  * This function fills a red-black tree with information from a
 590  * directory block.  It returns the number directory entries loaded
 591  * into the tree.  If there is an error it is returned in err.
 592  */
 593 static int htree_dirblock_to_tree(struct file *dir_file,
 594                                   struct inode *dir, ext4_lblk_t block,
 595                                   struct dx_hash_info *hinfo,
 596                                   __u32 start_hash, __u32 start_minor_hash)
 597 {
 598         struct buffer_head *bh;
 599         struct ext4_dir_entry_2 *de, *top;
 600         int err, count = 0;
 601
 602         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 603                                                         (unsigned long)block));
 604         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 605                 return err;
 606
 607         de = (struct ext4_dir_entry_2 *) bh->b_data;
 608         top = (struct ext4_dir_entry_2 *) ((char *) de +
 609                                            dir->i_sb->s_blocksize -
 610                                            EXT4_DIR_REC_LEN(0));
 611         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 612                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
 613                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 614                                                 +((char *)de - bh->b_data))) {
 615                         /* On error, skip the f_pos to the next block. */
 616                         dir_file->f_pos = (dir_file->f_pos |
 617                                         (dir->i_sb->s_blocksize - 1)) + 1;
 618                         brelse(bh);
 619                         return count;
 620                 }
 621                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 622                 if ((hinfo->hash < start_hash) ||
 623                     ((hinfo->hash == start_hash) &&
 624                      (hinfo->minor_hash < start_minor_hash)))
 625                         continue;
 626                 if (de->inode == 0)
 627                         continue;
 628                 if ((err = ext4_htree_store_dirent(dir_file,
 629                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 630                         brelse(bh);
 631                         return err;
 632                 }
 633                 count++;
 634         }
 635         brelse(bh);
 636         return count;
 637 }
 638
 639
 640 /*
 641  * This function fills a red-black tree with information from a
 642  * directory.  We start scanning the directory in hash order, starting
 643  * at start_hash and start_minor_hash.
 644  *
 645  * This function returns the number of entries inserted into the tree,
 646  * or a negative error code.
 647  */
 648 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 649                          __u32 start_minor_hash, __u32 *next_hash)
 650 {
 651         struct dx_hash_info hinfo;
 652         struct ext4_dir_entry_2 *de;
 653         struct dx_frame frames[2], *frame;
 654         struct inode *dir;
 655         ext4_lblk_t block;
 656         int count = 0;
 657         int ret, err;
 658         __u32 hashval;
 659
 660         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 661                        start_hash, start_minor_hash));
 662         dir = dir_file->f_path.dentry->d_inode;
 663         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
 664                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 665                 if (hinfo.hash_version <= DX_HASH_TEA)
 666                         hinfo.hash_version +=
 667                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 668                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 669                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 670                                                start_hash, start_minor_hash);
 671                 *next_hash = ~0;
 672                 return count;
 673         }
 674         hinfo.hash = start_hash;
 675         hinfo.minor_hash = 0;
 676         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 677         if (!frame)
 678                 return err;
 679
 680         /* Add '.' and '..' from the htree header */
 681         if (!start_hash && !start_minor_hash) {
 682                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 683                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 684                         goto errout;
 685                 count++;
 686         }
 687         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 688                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 689                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 690                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 691                         goto errout;
 692                 count++;
 693         }
 694
 695         while (1) {
 696                 block = dx_get_block(frame->at);
 697                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 698                                              start_hash, start_minor_hash);
 699                 if (ret < 0) {
 700                         err = ret;
 701                         goto errout;
 702                 }
 703                 count += ret;
 704                 hashval = ~0;
 705                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 706                                             frame, frames, &hashval);
 707                 *next_hash = hashval;
 708                 if (ret < 0) {
 709                         err = ret;
 710                         goto errout;
 711                 }
 712                 /*
 713                  * Stop if:  (a) there are no more entries, or
 714                  * (b) we have inserted at least one entry and the
 715                  * next hash value is not a continuation
 716                  */
 717                 if ((ret == 0) ||
 718                     (count && ((hashval & 1) == 0)))
 719                         break;
 720         }
 721         dx_release(frames);
 722         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 723                        "next hash: %x\n", count, *next_hash));
 724         return count;
 725 errout:
 726         dx_release(frames);
 727         return (err);
 728 }
 729
 730
 731 /*
 732  * Directory block splitting, compacting
 733  */
 734
 735 /*
 736  * Create map of hash values, offsets, and sizes, stored at end of block.
 737  * Returns number of entries mapped.
 738  */
 739 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 740                        struct dx_hash_info *hinfo,
 741                        struct dx_map_entry *map_tail)
 742 {
 743         int count = 0;
 744         char *base = (char *) de;
 745         struct dx_hash_info h = *hinfo;
 746
 747         while ((char *) de < base + blocksize) {
 748                 if (de->name_len && de->inode) {
 749                         ext4fs_dirhash(de->name, de->name_len, &h);
 750                         map_tail--;
 751                         map_tail->hash = h.hash;
 752                         map_tail->offs = ((char *) de - base)>>2;
 753                         map_tail->size = le16_to_cpu(de->rec_len);
 754                         count++;
 755                         cond_resched();
 756                 }
 757                 /* XXX: do we need to check rec_len == 0 case? -Chris */
 758                 de = ext4_next_entry(de, blocksize);
 759         }
 760         return count;
 761 }
 762
 763 /* Sort map by hash value */
 764 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 765 {
 766         struct dx_map_entry *p, *q, *top = map + count - 1;
 767         int more;
 768         /* Combsort until bubble sort doesn't suck */
 769         while (count > 2) {
 770                 count = count*10/13;
 771                 if (count - 9 < 2) /* 9, 10 -> 11 */
 772                         count = 11;
 773                 for (p = top, q = p - count; q >= map; p--, q--)
 774                         if (p->hash < q->hash)
 775                                 swap(*p, *q);
 776         }
 777         /* Garden variety bubble sort */
 778         do {
 779                 more = 0;
 780                 q = top;
 781                 while (q-- > map) {
 782                         if (q[1].hash >= q[0].hash)
 783                                 continue;
 784                         swap(*(q+1), *q);
 785                         more = 1;
 786                 }
 787         } while(more);
 788 }
 789
 790 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
 791 {
 792         struct dx_entry *entries = frame->entries;
 793         struct dx_entry *old = frame->at, *new = old + 1;
 794         int count = dx_get_count(entries);
 795
 796         assert(count < dx_get_limit(entries));
 797         assert(old < entries + count);
 798         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 799         dx_set_hash(new, hash);
 800         dx_set_block(new, block);
 801         dx_set_count(entries, count + 1);
 802 }
 803
 804 static void ext4_update_dx_flag(struct inode *inode)
 805 {
 806         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 807                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
 808                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
 809 }
 810
 811 /*
 812  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
 813  *
 814  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
 815  * `de != NULL' is guaranteed by caller.
 816  */
 817 static inline int ext4_match (int len, const char * const name,
 818                               struct ext4_dir_entry_2 * de)
 819 {
 820         if (len != de->name_len)
 821                 return 0;
 822         if (!de->inode)
 823                 return 0;
 824         return !memcmp(name, de->name, len);
 825 }
 826
 827 /*
 828  * Returns 0 if not found, -1 on failure, and 1 on success
 829  */
 830 static inline int search_dirblock(struct buffer_head *bh,
 831                                   struct inode *dir,
 832                                   const struct qstr *d_name,
 833                                   unsigned int offset,
 834                                   struct ext4_dir_entry_2 ** res_dir)
 835 {
 836         struct ext4_dir_entry_2 * de;
 837         char * dlimit;
 838         int de_len;
 839         const char *name = d_name->name;
 840         int namelen = d_name->len;
 841
 842         de = (struct ext4_dir_entry_2 *) bh->b_data;
 843         dlimit = bh->b_data + dir->i_sb->s_blocksize;
 844         while ((char *) de < dlimit) {
 845                 /* this code is executed quadratically often */
 846                 /* do minimal checking `by hand' */
 847
 848                 if ((char *) de + namelen <= dlimit &&
 849                     ext4_match (namelen, name, de)) {
 850                         /* found a match - just to be sure, do a full check */
 851                         if (!ext4_check_dir_entry("ext4_find_entry",
 852                                                   dir, de, bh, offset))
 853                                 return -1;
 854                         *res_dir = de;
 855                         return 1;
 856                 }
 857                 /* prevent looping on a bad block */
 858                 de_len = ext4_rec_len_from_disk(de->rec_len,
 859                                                 dir->i_sb->s_blocksize);
 860                 if (de_len <= 0)
 861                         return -1;
 862                 offset += de_len;
 863                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
 864         }
 865         return 0;
 866 }
 867
 868
 869 /*
 870  *      ext4_find_entry()
 871  *
 872  * finds an entry in the specified directory with the wanted name. It
 873  * returns the cache buffer in which the entry was found, and the entry
 874  * itself (as a parameter - res_dir). It does NOT read the inode of the
 875  * entry - you'll have to do that yourself if you want to.
 876  *
 877  * The returned buffer_head has ->b_count elevated.  The caller is expected
 878  * to brelse() it when appropriate.
 879  */
 880 static struct buffer_head * ext4_find_entry (struct inode *dir,
 881                                         const struct qstr *d_name,
 882                                         struct ext4_dir_entry_2 ** res_dir)
 883 {
 884         struct super_block *sb;
 885         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 886         struct buffer_head *bh, *ret = NULL;
 887         ext4_lblk_t start, block, b;
 888         int ra_max = 0;         /* Number of bh's in the readahead
 889                                    buffer, bh_use[] */
 890         int ra_ptr = 0;         /* Current index into readahead
 891                                    buffer */
 892         int num = 0;
 893         ext4_lblk_t  nblocks;
 894         int i, err;
 895         int namelen;
 896
 897         *res_dir = NULL;
 898         sb = dir->i_sb;
 899         namelen = d_name->len;
 900         if (namelen > EXT4_NAME_LEN)
 901                 return NULL;
 902         if (is_dx(dir)) {
 903                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
 904                 /*
 905                  * On success, or if the error was file not found,
 906                  * return.  Otherwise, fall back to doing a search the
 907                  * old fashioned way.
 908                  */
 909                 if (bh || (err != ERR_BAD_DX_DIR))
 910                         return bh;
 911                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
 912                                "falling back\n"));
 913         }
 914         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 915         start = EXT4_I(dir)->i_dir_start_lookup;
 916         if (start >= nblocks)
 917                 start = 0;
 918         block = start;
 919 restart:
 920         do {
 921                 /*
 922                  * We deal with the read-ahead logic here.
 923                  */
 924                 if (ra_ptr >= ra_max) {
 925                         /* Refill the readahead buffer */
 926                         ra_ptr = 0;
 927                         b = block;
 928                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 929                                 /*
 930                                  * Terminate if we reach the end of the
 931                                  * directory and must wrap, or if our
 932                                  * search has finished at this block.
 933                                  */
 934                                 if (b >= nblocks || (num && block == start)) {
 935                                         bh_use[ra_max] = NULL;
 936                                         break;
 937                                 }
 938                                 num++;
 939                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
 940                                 bh_use[ra_max] = bh;
 941                                 if (bh)
 942                                         ll_rw_block(READ_META, 1, &bh);
 943                         }
 944                 }
 945                 if ((bh = bh_use[ra_ptr++]) == NULL)
 946                         goto next;
 947                 wait_on_buffer(bh);
 948                 if (!buffer_uptodate(bh)) {
 949                         /* read error, skip block & hope for the best */
 950                         ext4_error(sb, __func__, "reading directory #%lu "
 951                                    "offset %lu", dir->i_ino,
 952                                    (unsigned long)block);
 953                         brelse(bh);
 954                         goto next;
 955                 }
 956                 i = search_dirblock(bh, dir, d_name,
 957                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
 958                 if (i == 1) {
 959                         EXT4_I(dir)->i_dir_start_lookup = block;
 960                         ret = bh;
 961                         goto cleanup_and_exit;
 962                 } else {
 963                         brelse(bh);
 964                         if (i < 0)
 965                                 goto cleanup_and_exit;
 966                 }
 967         next:
 968                 if (++block >= nblocks)
 969                         block = 0;
 970         } while (block != start);
 971
 972         /*
 973          * If the directory has grown while we were searching, then
 974          * search the last part of the directory before giving up.
 975          */
 976         block = nblocks;
 977         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
 978         if (block < nblocks) {
 979                 start = 0;
 980                 goto restart;
 981         }
 982
 983 cleanup_and_exit:
 984         /* Clean up the read-ahead blocks */
 985         for (; ra_ptr < ra_max; ra_ptr++)
 986                 brelse(bh_use[ra_ptr]);
 987         return ret;
 988 }
 989
 990 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
 991                        struct ext4_dir_entry_2 **res_dir, int *err)
 992 {
 993         struct super_block * sb;
 994         struct dx_hash_info     hinfo;
 995         u32 hash;
 996         struct dx_frame frames[2], *frame;
 997         struct ext4_dir_entry_2 *de, *top;
 998         struct buffer_head *bh;
 999         ext4_lblk_t block;
1000         int retval;
1001         int namelen = d_name->len;
1002         const u8 *name = d_name->name;
1003
1004         sb = dir->i_sb;
1005         /* NFS may look up ".." - look at dx_root directory block */
1006         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1007                 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1008                         return NULL;
1009         } else {
1010                 frame = frames;
1011                 frame->bh = NULL;                       /* for dx_release() */
1012                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
1013                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
1014         }
1015         hash = hinfo.hash;
1016         do {
1017                 block = dx_get_block(frame->at);
1018                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1019                         goto errout;
1020                 de = (struct ext4_dir_entry_2 *) bh->b_data;
1021                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1022                                        EXT4_DIR_REC_LEN(0));
1023                 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1024                         int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1025                                   + ((char *) de - bh->b_data);
1026
1027                         if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1028                                 brelse(bh);
1029                                 *err = ERR_BAD_DX_DIR;
1030                                 goto errout;
1031                         }
1032
1033                         if (ext4_match(namelen, name, de)) {
1034                                 *res_dir = de;
1035                                 dx_release(frames);
1036                                 return bh;
1037                         }
1038                 }
1039                 brelse(bh);
1040                 /* Check to see if we should continue to search */
1041                 retval = ext4_htree_next_block(dir, hash, frame,
1042                                                frames, NULL);
1043                 if (retval < 0) {
1044                         ext4_warning(sb, __func__,
1045                              "error reading index page in directory #%lu",
1046                              dir->i_ino);
1047                         *err = retval;
1048                         goto errout;
1049                 }
1050         } while (retval == 1);
1051
1052         *err = -ENOENT;
1053 errout:
1054         dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1055         dx_release (frames);
1056         return NULL;
1057 }
1058
1059 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1060 {
1061         struct inode *inode;
1062         struct ext4_dir_entry_2 *de;
1063         struct buffer_head *bh;
1064
1065         if (dentry->d_name.len > EXT4_NAME_LEN)
1066                 return ERR_PTR(-ENAMETOOLONG);
1067
1068         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1069         inode = NULL;
1070         if (bh) {
1071                 __u32 ino = le32_to_cpu(de->inode);
1072                 brelse(bh);
1073                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1074                         ext4_error(dir->i_sb, "ext4_lookup",
1075                                    "bad inode number: %u", ino);
1076                         return ERR_PTR(-EIO);
1077                 }
1078                 inode = ext4_iget(dir->i_sb, ino);
1079                 if (unlikely(IS_ERR(inode))) {
1080                         if (PTR_ERR(inode) == -ESTALE) {
1081                                 ext4_error(dir->i_sb, __func__,
1082                                                 "deleted inode referenced: %u",
1083                                                 ino);
1084                                 return ERR_PTR(-EIO);
1085                         } else {
1086                                 return ERR_CAST(inode);
1087                         }
1088                 }
1089         }
1090         return d_splice_alias(inode, dentry);
1091 }
1092
1093
1094 struct dentry *ext4_get_parent(struct dentry *child)
1095 {
1096         __u32 ino;
1097         struct inode *inode;
1098         static const struct qstr dotdot = {
1099                 .name = "..",
1100                 .len = 2,
1101         };
1102         struct ext4_dir_entry_2 * de;
1103         struct buffer_head *bh;
1104
1105         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1106         inode = NULL;
1107         if (!bh)
1108                 return ERR_PTR(-ENOENT);
1109         ino = le32_to_cpu(de->inode);
1110         brelse(bh);
1111
1112         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1113                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1114                            "bad inode number: %u", ino);
1115                 return ERR_PTR(-EIO);
1116         }
1117
1118         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1119 }
1120
1121 #define S_SHIFT 12
1122 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1123         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1124         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1125         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1126         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1127         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1128         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1129         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1130 };
1131
1132 static inline void ext4_set_de_type(struct super_block *sb,
1133                                 struct ext4_dir_entry_2 *de,
1134                                 umode_t mode) {
1135         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1136                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1137 }
1138
1139 /*
1140  * Move count entries from end of map between two memory locations.
1141  * Returns pointer to last entry moved.
1142  */
1143 static struct ext4_dir_entry_2 *
1144 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1145                 unsigned blocksize)
1146 {
1147         unsigned rec_len = 0;
1148
1149         while (count--) {
1150                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1151                                                 (from + (map->offs<<2));
1152                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1153                 memcpy (to, de, rec_len);
1154                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1155                                 ext4_rec_len_to_disk(rec_len, blocksize);
1156                 de->inode = 0;
1157                 map++;
1158                 to += rec_len;
1159         }
1160         return (struct ext4_dir_entry_2 *) (to - rec_len);
1161 }
1162
1163 /*
1164  * Compact each dir entry in the range to the minimal rec_len.
1165  * Returns pointer to last entry in range.
1166  */
1167 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1168 {
1169         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1170         unsigned rec_len = 0;
1171
1172         prev = to = de;
1173         while ((char*)de < base + blocksize) {
1174                 next = ext4_next_entry(de, blocksize);
1175                 if (de->inode && de->name_len) {
1176                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1177                         if (de > to)
1178                                 memmove(to, de, rec_len);
1179                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1180                         prev = to;
1181                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1182                 }
1183                 de = next;
1184         }
1185         return prev;
1186 }
1187
1188 /*
1189  * Split a full leaf block to make room for a new dir entry.
1190  * Allocate a new block, and move entries so that they are approx. equally full.
1191  * Returns pointer to de in block into which the new entry will be inserted.
1192  */
1193 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1194                         struct buffer_head **bh,struct dx_frame *frame,
1195                         struct dx_hash_info *hinfo, int *error)
1196 {
1197         unsigned blocksize = dir->i_sb->s_blocksize;
1198         unsigned count, continued;
1199         struct buffer_head *bh2;
1200         ext4_lblk_t newblock;
1201         u32 hash2;
1202         struct dx_map_entry *map;
1203         char *data1 = (*bh)->b_data, *data2;
1204         unsigned split, move, size;
1205         struct ext4_dir_entry_2 *de = NULL, *de2;
1206         int     err = 0, i;
1207
1208         bh2 = ext4_append (handle, dir, &newblock, &err);
1209         if (!(bh2)) {
1210                 brelse(*bh);
1211                 *bh = NULL;
1212                 goto errout;
1213         }
1214
1215         BUFFER_TRACE(*bh, "get_write_access");
1216         err = ext4_journal_get_write_access(handle, *bh);
1217         if (err)
1218                 goto journal_error;
1219
1220         BUFFER_TRACE(frame->bh, "get_write_access");
1221         err = ext4_journal_get_write_access(handle, frame->bh);
1222         if (err)
1223                 goto journal_error;
1224
1225         data2 = bh2->b_data;
1226
1227         /* create map in the end of data2 block */
1228         map = (struct dx_map_entry *) (data2 + blocksize);
1229         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1230                              blocksize, hinfo, map);
1231         map -= count;
1232         dx_sort_map(map, count);
1233         /* Split the existing block in the middle, size-wise */
1234         size = 0;
1235         move = 0;
1236         for (i = count-1; i >= 0; i--) {
1237                 /* is more than half of this entry in 2nd half of the block? */
1238                 if (size + map[i].size/2 > blocksize/2)
1239                         break;
1240                 size += map[i].size;
1241                 move++;
1242         }
1243         /* map index at which we will split */
1244         split = count - move;
1245         hash2 = map[split].hash;
1246         continued = hash2 == map[split - 1].hash;
1247         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1248                         (unsigned long)dx_get_block(frame->at),
1249                                         hash2, split, count-split));
1250
1251         /* Fancy dance to stay within two buffers */
1252         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1253         de = dx_pack_dirents(data1, blocksize);
1254         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1255                                            blocksize);
1256         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1257                                             blocksize);
1258         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1259         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1260
1261         /* Which block gets the new entry? */
1262         if (hinfo->hash >= hash2)
1263         {
1264                 swap(*bh, bh2);
1265                 de = de2;
1266         }
1267         dx_insert_block(frame, hash2 + continued, newblock);
1268         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1269         if (err)
1270                 goto journal_error;
1271         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1272         if (err)
1273                 goto journal_error;
1274         brelse(bh2);
1275         dxtrace(dx_show_index("frame", frame->entries));
1276         return de;
1277
1278 journal_error:
1279         brelse(*bh);
1280         brelse(bh2);
1281         *bh = NULL;
1282         ext4_std_error(dir->i_sb, err);
1283 errout:
1284         *error = err;
1285         return NULL;
1286 }
1287
1288 /*
1289  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1290  * it points to a directory entry which is guaranteed to be large
1291  * enough for new directory entry.  If de is NULL, then
1292  * add_dirent_to_buf will attempt search the directory block for
1293  * space.  It will return -ENOSPC if no space is available, and -EIO
1294  * and -EEXIST if directory entry already exists.
1295  *
1296  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1297  * all other cases bh is released.
1298  */
1299 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1300                              struct inode *inode, struct ext4_dir_entry_2 *de,
1301                              struct buffer_head *bh)
1302 {
1303         struct inode    *dir = dentry->d_parent->d_inode;
1304         const char      *name = dentry->d_name.name;
1305         int             namelen = dentry->d_name.len;
1306         unsigned int    offset = 0;
1307         unsigned int    blocksize = dir->i_sb->s_blocksize;
1308         unsigned short  reclen;
1309         int             nlen, rlen, err;
1310         char            *top;
1311
1312         reclen = EXT4_DIR_REC_LEN(namelen);
1313         if (!de) {
1314                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1315                 top = bh->b_data + blocksize - reclen;
1316                 while ((char *) de <= top) {
1317                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1318                                                   bh, offset)) {
1319                                 brelse(bh);
1320                                 return -EIO;
1321                         }
1322                         if (ext4_match(namelen, name, de)) {
1323                                 brelse(bh);
1324                                 return -EEXIST;
1325                         }
1326                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1327                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1328                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1329                                 break;
1330                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1331                         offset += rlen;
1332                 }
1333                 if ((char *) de > top)
1334                         return -ENOSPC;
1335         }
1336         BUFFER_TRACE(bh, "get_write_access");
1337         err = ext4_journal_get_write_access(handle, bh);
1338         if (err) {
1339                 ext4_std_error(dir->i_sb, err);
1340                 brelse(bh);
1341                 return err;
1342         }
1343
1344         /* By now the buffer is marked for journaling */
1345         nlen = EXT4_DIR_REC_LEN(de->name_len);
1346         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1347         if (de->inode) {
1348                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1349                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1350                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1351                 de = de1;
1352         }
1353         de->file_type = EXT4_FT_UNKNOWN;
1354         if (inode) {
1355                 de->inode = cpu_to_le32(inode->i_ino);
1356                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1357         } else
1358                 de->inode = 0;
1359         de->name_len = namelen;
1360         memcpy(de->name, name, namelen);
1361         /*
1362          * XXX shouldn't update any times until successful
1363          * completion of syscall, but too many callers depend
1364          * on this.
1365          *
1366          * XXX similarly, too many callers depend on
1367          * ext4_new_inode() setting the times, but error
1368          * recovery deletes the inode, so the worst that can
1369          * happen is that the times are slightly out of date
1370          * and/or different from the directory change time.
1371          */
1372         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1373         ext4_update_dx_flag(dir);
1374         dir->i_version++;
1375         ext4_mark_inode_dirty(handle, dir);
1376         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1377         err = ext4_handle_dirty_metadata(handle, dir, bh);
1378         if (err)
1379                 ext4_std_error(dir->i_sb, err);
1380         brelse(bh);
1381         return 0;
1382 }
1383
1384 /*
1385  * This converts a one block unindexed directory to a 3 block indexed
1386  * directory, and adds the dentry to the indexed directory.
1387  */
1388 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1389                             struct inode *inode, struct buffer_head *bh)
1390 {
1391         struct inode    *dir = dentry->d_parent->d_inode;
1392         const char      *name = dentry->d_name.name;
1393         int             namelen = dentry->d_name.len;
1394         struct buffer_head *bh2;
1395         struct dx_root  *root;
1396         struct dx_frame frames[2], *frame;
1397         struct dx_entry *entries;
1398         struct ext4_dir_entry_2 *de, *de2;
1399         char            *data1, *top;
1400         unsigned        len;
1401         int             retval;
1402         unsigned        blocksize;
1403         struct dx_hash_info hinfo;
1404         ext4_lblk_t  block;
1405         struct fake_dirent *fde;
1406
1407         blocksize =  dir->i_sb->s_blocksize;
1408         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1409         retval = ext4_journal_get_write_access(handle, bh);
1410         if (retval) {
1411                 ext4_std_error(dir->i_sb, retval);
1412                 brelse(bh);
1413                 return retval;
1414         }
1415         root = (struct dx_root *) bh->b_data;
1416
1417         /* The 0th block becomes the root, move the dirents out */
1418         fde = &root->dotdot;
1419         de = (struct ext4_dir_entry_2 *)((char *)fde +
1420                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1421         if ((char *) de >= (((char *) root) + blocksize)) {
1422                 ext4_error(dir->i_sb, __func__,
1423                            "invalid rec_len for '..' in inode %lu",
1424                            dir->i_ino);
1425                 brelse(bh);
1426                 return -EIO;
1427         }
1428         len = ((char *) root) + blocksize - (char *) de;
1429
1430         /* Allocate new block for the 0th block's dirents */
1431         bh2 = ext4_append(handle, dir, &block, &retval);
1432         if (!(bh2)) {
1433                 brelse(bh);
1434                 return retval;
1435         }
1436         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1437         data1 = bh2->b_data;
1438
1439         memcpy (data1, de, len);
1440         de = (struct ext4_dir_entry_2 *) data1;
1441         top = data1 + len;
1442         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1443                 de = de2;
1444         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1445                                            blocksize);
1446         /* Initialize the root; the dot dirents already exist */
1447         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1448         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1449                                            blocksize);
1450         memset (&root->info, 0, sizeof(root->info));
1451         root->info.info_length = sizeof(root->info);
1452         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1453         entries = root->entries;
1454         dx_set_block(entries, 1);
1455         dx_set_count(entries, 1);
1456         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1457
1458         /* Initialize as for dx_probe */
1459         hinfo.hash_version = root->info.hash_version;
1460         if (hinfo.hash_version <= DX_HASH_TEA)
1461                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1462         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1463         ext4fs_dirhash(name, namelen, &hinfo);
1464         frame = frames;
1465         frame->entries = entries;
1466         frame->at = entries;
1467         frame->bh = bh;
1468         bh = bh2;
1469         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1470         dx_release (frames);
1471         if (!(de))
1472                 return retval;
1473
1474         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1475 }
1476
1477 /*
1478  *      ext4_add_entry()
1479  *
1480  * adds a file entry to the specified directory, using the same
1481  * semantics as ext4_find_entry(). It returns NULL if it failed.
1482  *
1483  * NOTE!! The inode part of 'de' is left at 0 - which means you
1484  * may not sleep between calling this and putting something into
1485  * the entry, as someone else might have used it while you slept.
1486  */
1487 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1488                           struct inode *inode)
1489 {
1490         struct inode *dir = dentry->d_parent->d_inode;
1491         struct buffer_head *bh;
1492         struct ext4_dir_entry_2 *de;
1493         struct super_block *sb;
1494         int     retval;
1495         int     dx_fallback=0;
1496         unsigned blocksize;
1497         ext4_lblk_t block, blocks;
1498
1499         sb = dir->i_sb;
1500         blocksize = sb->s_blocksize;
1501         if (!dentry->d_name.len)
1502                 return -EINVAL;
1503         if (is_dx(dir)) {
1504                 retval = ext4_dx_add_entry(handle, dentry, inode);
1505                 if (!retval || (retval != ERR_BAD_DX_DIR))
1506                         return retval;
1507                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1508                 dx_fallback++;
1509                 ext4_mark_inode_dirty(handle, dir);
1510         }
1511         blocks = dir->i_size >> sb->s_blocksize_bits;
1512         for (block = 0; block < blocks; block++) {
1513                 bh = ext4_bread(handle, dir, block, 0, &retval);
1514                 if(!bh)
1515                         return retval;
1516                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1517                 if (retval != -ENOSPC)
1518                         return retval;
1519
1520                 if (blocks == 1 && !dx_fallback &&
1521                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
1522                         retval = make_indexed_dir(handle, dentry, inode, bh);
1523                         if (retval == -ENOSPC)
1524                                 brelse(bh);
1525                         return retval;
1526                 }
1527                 brelse(bh);
1528         }
1529         bh = ext4_append(handle, dir, &block, &retval);
1530         if (!bh)
1531                 return retval;
1532         de = (struct ext4_dir_entry_2 *) bh->b_data;
1533         de->inode = 0;
1534         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1535         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1536         if (retval == -ENOSPC)
1537                 brelse(bh);
1538         return retval;
1539 }
1540
1541 /*
1542  * Returns 0 for success, or a negative error value
1543  */
1544 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1545                              struct inode *inode)
1546 {
1547         struct dx_frame frames[2], *frame;
1548         struct dx_entry *entries, *at;
1549         struct dx_hash_info hinfo;
1550         struct buffer_head *bh;
1551         struct inode *dir = dentry->d_parent->d_inode;
1552         struct super_block *sb = dir->i_sb;
1553         struct ext4_dir_entry_2 *de;
1554         int err;
1555
1556         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1557         if (!frame)
1558                 return err;
1559         entries = frame->entries;
1560         at = frame->at;
1561
1562         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1563                 goto cleanup;
1564
1565         BUFFER_TRACE(bh, "get_write_access");
1566         err = ext4_journal_get_write_access(handle, bh);
1567         if (err)
1568                 goto journal_error;
1569
1570         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1571         if (err != -ENOSPC) {
1572                 bh = NULL;
1573                 goto cleanup;
1574         }
1575
1576         /* Block full, should compress but for now just split */
1577         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1578                        dx_get_count(entries), dx_get_limit(entries)));
1579         /* Need to split index? */
1580         if (dx_get_count(entries) == dx_get_limit(entries)) {
1581                 ext4_lblk_t newblock;
1582                 unsigned icount = dx_get_count(entries);
1583                 int levels = frame - frames;
1584                 struct dx_entry *entries2;
1585                 struct dx_node *node2;
1586                 struct buffer_head *bh2;
1587
1588                 if (levels && (dx_get_count(frames->entries) ==
1589                                dx_get_limit(frames->entries))) {
1590                         ext4_warning(sb, __func__,
1591                                      "Directory index full!");
1592                         err = -ENOSPC;
1593                         goto cleanup;
1594                 }
1595                 bh2 = ext4_append (handle, dir, &newblock, &err);
1596                 if (!(bh2))
1597                         goto cleanup;
1598                 node2 = (struct dx_node *)(bh2->b_data);
1599                 entries2 = node2->entries;
1600                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1601                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1602                                                            sb->s_blocksize);
1603                 BUFFER_TRACE(frame->bh, "get_write_access");
1604                 err = ext4_journal_get_write_access(handle, frame->bh);
1605                 if (err)
1606                         goto journal_error;
1607                 if (levels) {
1608                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1609                         unsigned hash2 = dx_get_hash(entries + icount1);
1610                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1611                                        icount1, icount2));
1612
1613                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1614                         err = ext4_journal_get_write_access(handle,
1615                                                              frames[0].bh);
1616                         if (err)
1617                                 goto journal_error;
1618
1619                         memcpy((char *) entries2, (char *) (entries + icount1),
1620                                icount2 * sizeof(struct dx_entry));
1621                         dx_set_count(entries, icount1);
1622                         dx_set_count(entries2, icount2);
1623                         dx_set_limit(entries2, dx_node_limit(dir));
1624
1625                         /* Which index block gets the new entry? */
1626                         if (at - entries >= icount1) {
1627                                 frame->at = at = at - entries - icount1 + entries2;
1628                                 frame->entries = entries = entries2;
1629                                 swap(frame->bh, bh2);
1630                         }
1631                         dx_insert_block(frames + 0, hash2, newblock);
1632                         dxtrace(dx_show_index("node", frames[1].entries));
1633                         dxtrace(dx_show_index("node",
1634                                ((struct dx_node *) bh2->b_data)->entries));
1635                         err = ext4_handle_dirty_metadata(handle, inode, bh2);
1636                         if (err)
1637                                 goto journal_error;
1638                         brelse (bh2);
1639                 } else {
1640                         dxtrace(printk(KERN_DEBUG
1641                                        "Creating second level index...\n"));
1642                         memcpy((char *) entries2, (char *) entries,
1643                                icount * sizeof(struct dx_entry));
1644                         dx_set_limit(entries2, dx_node_limit(dir));
1645
1646                         /* Set up root */
1647                         dx_set_count(entries, 1);
1648                         dx_set_block(entries + 0, newblock);
1649                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1650
1651                         /* Add new access path frame */
1652                         frame = frames + 1;
1653                         frame->at = at = at - entries + entries2;
1654                         frame->entries = entries = entries2;
1655                         frame->bh = bh2;
1656                         err = ext4_journal_get_write_access(handle,
1657                                                              frame->bh);
1658                         if (err)
1659                                 goto journal_error;
1660                 }
1661                 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1662         }
1663         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1664         if (!de)
1665                 goto cleanup;
1666         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1667         if (err != -ENOSPC)
1668                 bh = NULL;
1669         goto cleanup;
1670
1671 journal_error:
1672         ext4_std_error(dir->i_sb, err);
1673 cleanup:
1674         if (bh)
1675                 brelse(bh);
1676         dx_release(frames);
1677         return err;
1678 }
1679
1680 /*
1681  * ext4_delete_entry deletes a directory entry by merging it with the
1682  * previous entry
1683  */
1684 static int ext4_delete_entry(handle_t *handle,
1685                              struct inode *dir,
1686                              struct ext4_dir_entry_2 *de_del,
1687                              struct buffer_head *bh)
1688 {
1689         struct ext4_dir_entry_2 *de, *pde;
1690         unsigned int blocksize = dir->i_sb->s_blocksize;
1691         int i;
1692
1693         i = 0;
1694         pde = NULL;
1695         de = (struct ext4_dir_entry_2 *) bh->b_data;
1696         while (i < bh->b_size) {
1697                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1698                         return -EIO;
1699                 if (de == de_del)  {
1700                         BUFFER_TRACE(bh, "get_write_access");
1701                         ext4_journal_get_write_access(handle, bh);
1702                         if (pde)
1703                                 pde->rec_len = ext4_rec_len_to_disk(
1704                                         ext4_rec_len_from_disk(pde->rec_len,
1705                                                                blocksize) +
1706                                         ext4_rec_len_from_disk(de->rec_len,
1707                                                                blocksize),
1708                                         blocksize);
1709                         else
1710                                 de->inode = 0;
1711                         dir->i_version++;
1712                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1713                         ext4_handle_dirty_metadata(handle, dir, bh);
1714                         return 0;
1715                 }
1716                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1717                 pde = de;
1718                 de = ext4_next_entry(de, blocksize);
1719         }
1720         return -ENOENT;
1721 }
1722
1723 /*
1724  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1725  * since this indicates that nlinks count was previously 1.
1726  */
1727 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1728 {
1729         inc_nlink(inode);
1730         if (is_dx(inode) && inode->i_nlink > 1) {
1731                 /* limit is 16-bit i_links_count */
1732                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1733                         inode->i_nlink = 1;
1734                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1735                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1736                 }
1737         }
1738 }
1739
1740 /*
1741  * If a directory had nlink == 1, then we should let it be 1. This indicates
1742  * directory has >EXT4_LINK_MAX subdirs.
1743  */
1744 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1745 {
1746         drop_nlink(inode);
1747         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1748                 inc_nlink(inode);
1749 }
1750
1751
1752 static int ext4_add_nondir(handle_t *handle,
1753                 struct dentry *dentry, struct inode *inode)
1754 {
1755         int err = ext4_add_entry(handle, dentry, inode);
1756         if (!err) {
1757                 ext4_mark_inode_dirty(handle, inode);
1758                 d_instantiate(dentry, inode);
1759                 unlock_new_inode(inode);
1760                 return 0;
1761         }
1762         drop_nlink(inode);
1763         unlock_new_inode(inode);
1764         iput(inode);
1765         return err;
1766 }
1767
1768 /*
1769  * By the time this is called, we already have created
1770  * the directory cache entry for the new file, but it
1771  * is so far negative - it has no inode.
1772  *
1773  * If the create succeeds, we fill in the inode information
1774  * with d_instantiate().
1775  */
1776 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1777                        struct nameidata *nd)
1778 {
1779         handle_t *handle;
1780         struct inode *inode;
1781         int err, retries = 0;
1782
1783 retry:
1784         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1785                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1786                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1787         if (IS_ERR(handle))
1788                 return PTR_ERR(handle);
1789
1790         if (IS_DIRSYNC(dir))
1791                 ext4_handle_sync(handle);
1792
1793         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1794         err = PTR_ERR(inode);
1795         if (!IS_ERR(inode)) {
1796                 inode->i_op = &ext4_file_inode_operations;
1797                 inode->i_fop = &ext4_file_operations;
1798                 ext4_set_aops(inode);
1799                 err = ext4_add_nondir(handle, dentry, inode);
1800         }
1801         ext4_journal_stop(handle);
1802         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1803                 goto retry;
1804         return err;
1805 }
1806
1807 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1808                       int mode, dev_t rdev)
1809 {
1810         handle_t *handle;
1811         struct inode *inode;
1812         int err, retries = 0;
1813
1814         if (!new_valid_dev(rdev))
1815                 return -EINVAL;
1816
1817 retry:
1818         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1819                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1820                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1821         if (IS_ERR(handle))
1822                 return PTR_ERR(handle);
1823
1824         if (IS_DIRSYNC(dir))
1825                 ext4_handle_sync(handle);
1826
1827         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1828         err = PTR_ERR(inode);
1829         if (!IS_ERR(inode)) {
1830                 init_special_inode(inode, inode->i_mode, rdev);
1831 #ifdef CONFIG_EXT4_FS_XATTR
1832                 inode->i_op = &ext4_special_inode_operations;
1833 #endif
1834                 err = ext4_add_nondir(handle, dentry, inode);
1835         }
1836         ext4_journal_stop(handle);
1837         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1838                 goto retry;
1839         return err;
1840 }
1841
1842 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1843 {
1844         handle_t *handle;
1845         struct inode *inode;
1846         struct buffer_head *dir_block;
1847         struct ext4_dir_entry_2 *de;
1848         unsigned int blocksize = dir->i_sb->s_blocksize;
1849         int err, retries = 0;
1850
1851         if (EXT4_DIR_LINK_MAX(dir))
1852                 return -EMLINK;
1853
1854 retry:
1855         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1856                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1857                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1858         if (IS_ERR(handle))
1859                 return PTR_ERR(handle);
1860
1861         if (IS_DIRSYNC(dir))
1862                 ext4_handle_sync(handle);
1863
1864         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1865                                &dentry->d_name, 0);
1866         err = PTR_ERR(inode);
1867         if (IS_ERR(inode))
1868                 goto out_stop;
1869
1870         inode->i_op = &ext4_dir_inode_operations;
1871         inode->i_fop = &ext4_dir_operations;
1872         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1873         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1874         if (!dir_block)
1875                 goto out_clear_inode;
1876         BUFFER_TRACE(dir_block, "get_write_access");
1877         ext4_journal_get_write_access(handle, dir_block);
1878         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1879         de->inode = cpu_to_le32(inode->i_ino);
1880         de->name_len = 1;
1881         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1882                                            blocksize);
1883         strcpy(de->name, ".");
1884         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1885         de = ext4_next_entry(de, blocksize);
1886         de->inode = cpu_to_le32(dir->i_ino);
1887         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1888                                            blocksize);
1889         de->name_len = 2;
1890         strcpy(de->name, "..");
1891         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1892         inode->i_nlink = 2;
1893         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1894         ext4_handle_dirty_metadata(handle, dir, dir_block);
1895         brelse(dir_block);
1896         ext4_mark_inode_dirty(handle, inode);
1897         err = ext4_add_entry(handle, dentry, inode);
1898         if (err) {
1899 out_clear_inode:
1900                 clear_nlink(inode);
1901                 unlock_new_inode(inode);
1902                 ext4_mark_inode_dirty(handle, inode);
1903                 iput(inode);
1904                 goto out_stop;
1905         }
1906         ext4_inc_count(handle, dir);
1907         ext4_update_dx_flag(dir);
1908         ext4_mark_inode_dirty(handle, dir);
1909         d_instantiate(dentry, inode);
1910         unlock_new_inode(inode);
1911 out_stop:
1912         ext4_journal_stop(handle);
1913         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1914                 goto retry;
1915         return err;
1916 }
1917
1918 /*
1919  * routine to check that the specified directory is empty (for rmdir)
1920  */
1921 static int empty_dir(struct inode *inode)
1922 {
1923         unsigned int offset;
1924         struct buffer_head *bh;
1925         struct ext4_dir_entry_2 *de, *de1;
1926         struct super_block *sb;
1927         int err = 0;
1928
1929         sb = inode->i_sb;
1930         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1931             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1932                 if (err)
1933                         ext4_error(inode->i_sb, __func__,
1934                                    "error %d reading directory #%lu offset 0",
1935                                    err, inode->i_ino);
1936                 else
1937                         ext4_warning(inode->i_sb, __func__,
1938                                      "bad directory (dir #%lu) - no data block",
1939                                      inode->i_ino);
1940                 return 1;
1941         }
1942         de = (struct ext4_dir_entry_2 *) bh->b_data;
1943         de1 = ext4_next_entry(de, sb->s_blocksize);
1944         if (le32_to_cpu(de->inode) != inode->i_ino ||
1945                         !le32_to_cpu(de1->inode) ||
1946                         strcmp(".", de->name) ||
1947                         strcmp("..", de1->name)) {
1948                 ext4_warning(inode->i_sb, "empty_dir",
1949                              "bad directory (dir #%lu) - no `.' or `..'",
1950                              inode->i_ino);
1951                 brelse(bh);
1952                 return 1;
1953         }
1954         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1955                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1956         de = ext4_next_entry(de1, sb->s_blocksize);
1957         while (offset < inode->i_size) {
1958                 if (!bh ||
1959                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1960                         err = 0;
1961                         brelse(bh);
1962                         bh = ext4_bread(NULL, inode,
1963                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1964                         if (!bh) {
1965                                 if (err)
1966                                         ext4_error(sb, __func__,
1967                                                    "error %d reading directory"
1968                                                    " #%lu offset %u",
1969                                                    err, inode->i_ino, offset);
1970                                 offset += sb->s_blocksize;
1971                                 continue;
1972                         }
1973                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1974                 }
1975                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1976                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1977                                                          sb->s_blocksize);
1978                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1979                         continue;
1980                 }
1981                 if (le32_to_cpu(de->inode)) {
1982                         brelse(bh);
1983                         return 0;
1984                 }
1985                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1986                 de = ext4_next_entry(de, sb->s_blocksize);
1987         }
1988         brelse(bh);
1989         return 1;
1990 }
1991
1992 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1993  * such inodes, starting at the superblock, in case we crash before the
1994  * file is closed/deleted, or in case the inode truncate spans multiple
1995  * transactions and the last transaction is not recovered after a crash.
1996  *
1997  * At filesystem recovery time, we walk this list deleting unlinked
1998  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1999  */
2000 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2001 {
2002         struct super_block *sb = inode->i_sb;
2003         struct ext4_iloc iloc;
2004         int err = 0, rc;
2005
2006         if (!ext4_handle_valid(handle))
2007                 return 0;
2008
2009         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2010         if (!list_empty(&EXT4_I(inode)->i_orphan))
2011                 goto out_unlock;
2012
2013         /* Orphan handling is only valid for files with data blocks
2014          * being truncated, or files being unlinked. */
2015
2016         /* @@@ FIXME: Observation from aviro:
2017          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
2018          * here (on s_orphan_lock), so race with ext4_link() which might bump
2019          * ->i_nlink. For, say it, character device. Not a regular file,
2020          * not a directory, not a symlink and ->i_nlink > 0.
2021          *
2022          * tytso, 4/25/2009: I'm not sure how that could happen;
2023          * shouldn't the fs core protect us from these sort of
2024          * unlink()/link() races?
2025          */
2026         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2027                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2028
2029         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2030         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2031         if (err)
2032                 goto out_unlock;
2033
2034         err = ext4_reserve_inode_write(handle, inode, &iloc);
2035         if (err)
2036                 goto out_unlock;
2037
2038         /* Insert this inode at the head of the on-disk orphan list... */
2039         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2040         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2041         err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
2042         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2043         if (!err)
2044                 err = rc;
2045
2046         /* Only add to the head of the in-memory list if all the
2047          * previous operations succeeded.  If the orphan_add is going to
2048          * fail (possibly taking the journal offline), we can't risk
2049          * leaving the inode on the orphan list: stray orphan-list
2050          * entries can cause panics at unmount time.
2051          *
2052          * This is safe: on error we're going to ignore the orphan list
2053          * anyway on the next recovery. */
2054         if (!err)
2055                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2056
2057         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2058         jbd_debug(4, "orphan inode %lu will point to %d\n",
2059                         inode->i_ino, NEXT_ORPHAN(inode));
2060 out_unlock:
2061         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2062         ext4_std_error(inode->i_sb, err);
2063         return err;
2064 }
2065
2066 /*
2067  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2068  * of such inodes stored on disk, because it is finally being cleaned up.
2069  */
2070 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2071 {
2072         struct list_head *prev;
2073         struct ext4_inode_info *ei = EXT4_I(inode);
2074         struct ext4_sb_info *sbi;
2075         __u32 ino_next;
2076         struct ext4_iloc iloc;
2077         int err = 0;
2078
2079         if (!ext4_handle_valid(handle))
2080                 return 0;
2081
2082         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2083         if (list_empty(&ei->i_orphan))
2084                 goto out;
2085
2086         ino_next = NEXT_ORPHAN(inode);
2087         prev = ei->i_orphan.prev;
2088         sbi = EXT4_SB(inode->i_sb);
2089
2090         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2091
2092         list_del_init(&ei->i_orphan);
2093
2094         /* If we're on an error path, we may not have a valid
2095          * transaction handle with which to update the orphan list on
2096          * disk, but we still need to remove the inode from the linked
2097          * list in memory. */
2098         if (sbi->s_journal && !handle)
2099                 goto out;
2100
2101         err = ext4_reserve_inode_write(handle, inode, &iloc);
2102         if (err)
2103                 goto out_err;
2104
2105         if (prev == &sbi->s_orphan) {
2106                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2107                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2108                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2109                 if (err)
2110                         goto out_brelse;
2111                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2112                 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
2113         } else {
2114                 struct ext4_iloc iloc2;
2115                 struct inode *i_prev =
2116                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2117
2118                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2119                           i_prev->i_ino, ino_next);
2120                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2121                 if (err)
2122                         goto out_brelse;
2123                 NEXT_ORPHAN(i_prev) = ino_next;
2124                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2125         }
2126         if (err)
2127                 goto out_brelse;
2128         NEXT_ORPHAN(inode) = 0;
2129         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2130
2131 out_err:
2132         ext4_std_error(inode->i_sb, err);
2133 out:
2134         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2135         return err;
2136
2137 out_brelse:
2138         brelse(iloc.bh);
2139         goto out_err;
2140 }
2141
2142 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2143 {
2144         int retval;
2145         struct inode *inode;
2146         struct buffer_head *bh;
2147         struct ext4_dir_entry_2 *de;
2148         handle_t *handle;
2149
2150         /* Initialize quotas before so that eventual writes go in
2151          * separate transaction */
2152         vfs_dq_init(dentry->d_inode);
2153         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2154         if (IS_ERR(handle))
2155                 return PTR_ERR(handle);
2156
2157         retval = -ENOENT;
2158         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2159         if (!bh)
2160                 goto end_rmdir;
2161
2162         if (IS_DIRSYNC(dir))
2163                 ext4_handle_sync(handle);
2164
2165         inode = dentry->d_inode;
2166
2167         retval = -EIO;
2168         if (le32_to_cpu(de->inode) != inode->i_ino)
2169                 goto end_rmdir;
2170
2171         retval = -ENOTEMPTY;
2172         if (!empty_dir(inode))
2173                 goto end_rmdir;
2174
2175         retval = ext4_delete_entry(handle, dir, de, bh);
2176         if (retval)
2177                 goto end_rmdir;
2178         if (!EXT4_DIR_LINK_EMPTY(inode))
2179                 ext4_warning(inode->i_sb, "ext4_rmdir",
2180                              "empty directory has too many links (%d)",
2181                              inode->i_nlink);
2182         inode->i_version++;
2183         clear_nlink(inode);
2184         /* There's no need to set i_disksize: the fact that i_nlink is
2185          * zero will ensure that the right thing happens during any
2186          * recovery. */
2187         inode->i_size = 0;
2188         ext4_orphan_add(handle, inode);
2189         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2190         ext4_mark_inode_dirty(handle, inode);
2191         ext4_dec_count(handle, dir);
2192         ext4_update_dx_flag(dir);
2193         ext4_mark_inode_dirty(handle, dir);
2194
2195 end_rmdir:
2196         ext4_journal_stop(handle);
2197         brelse(bh);
2198         return retval;
2199 }
2200
2201 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2202 {
2203         int retval;
2204         struct inode *inode;
2205         struct buffer_head *bh;
2206         struct ext4_dir_entry_2 *de;
2207         handle_t *handle;
2208
2209         /* Initialize quotas before so that eventual writes go
2210          * in separate transaction */
2211         vfs_dq_init(dentry->d_inode);
2212         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2213         if (IS_ERR(handle))
2214                 return PTR_ERR(handle);
2215
2216         if (IS_DIRSYNC(dir))
2217                 ext4_handle_sync(handle);
2218
2219         retval = -ENOENT;
2220         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2221         if (!bh)
2222                 goto end_unlink;
2223
2224         inode = dentry->d_inode;
2225
2226         retval = -EIO;
2227         if (le32_to_cpu(de->inode) != inode->i_ino)
2228                 goto end_unlink;
2229
2230         if (!inode->i_nlink) {
2231                 ext4_warning(inode->i_sb, "ext4_unlink",
2232                              "Deleting nonexistent file (%lu), %d",
2233                              inode->i_ino, inode->i_nlink);
2234                 inode->i_nlink = 1;
2235         }
2236         retval = ext4_delete_entry(handle, dir, de, bh);
2237         if (retval)
2238                 goto end_unlink;
2239         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2240         ext4_update_dx_flag(dir);
2241         ext4_mark_inode_dirty(handle, dir);
2242         drop_nlink(inode);
2243         if (!inode->i_nlink)
2244                 ext4_orphan_add(handle, inode);
2245         inode->i_ctime = ext4_current_time(inode);
2246         ext4_mark_inode_dirty(handle, inode);
2247         retval = 0;
2248
2249 end_unlink:
2250         ext4_journal_stop(handle);
2251         brelse(bh);
2252         return retval;
2253 }
2254
2255 static int ext4_symlink(struct inode *dir,
2256                         struct dentry *dentry, const char *symname)
2257 {
2258         handle_t *handle;
2259         struct inode *inode;
2260         int l, err, retries = 0;
2261
2262         l = strlen(symname)+1;
2263         if (l > dir->i_sb->s_blocksize)
2264                 return -ENAMETOOLONG;
2265
2266 retry:
2267         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2268                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2269                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2270         if (IS_ERR(handle))
2271                 return PTR_ERR(handle);
2272
2273         if (IS_DIRSYNC(dir))
2274                 ext4_handle_sync(handle);
2275
2276         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2277                                &dentry->d_name, 0);
2278         err = PTR_ERR(inode);
2279         if (IS_ERR(inode))
2280                 goto out_stop;
2281
2282         if (l > sizeof(EXT4_I(inode)->i_data)) {
2283                 inode->i_op = &ext4_symlink_inode_operations;
2284                 ext4_set_aops(inode);
2285                 /*
2286                  * page_symlink() calls into ext4_prepare/commit_write.
2287                  * We have a transaction open.  All is sweetness.  It also sets
2288                  * i_size in generic_commit_write().
2289                  */
2290                 err = __page_symlink(inode, symname, l, 1);
2291                 if (err) {
2292                         clear_nlink(inode);
2293                         unlock_new_inode(inode);
2294                         ext4_mark_inode_dirty(handle, inode);
2295                         iput(inode);
2296                         goto out_stop;
2297                 }
2298         } else {
2299                 /* clear the extent format for fast symlink */
2300                 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2301                 inode->i_op = &ext4_fast_symlink_inode_operations;
2302                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2303                 inode->i_size = l-1;
2304         }
2305         EXT4_I(inode)->i_disksize = inode->i_size;
2306         err = ext4_add_nondir(handle, dentry, inode);
2307 out_stop:
2308         ext4_journal_stop(handle);
2309         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2310                 goto retry;
2311         return err;
2312 }
2313
2314 static int ext4_link(struct dentry *old_dentry,
2315                      struct inode *dir, struct dentry *dentry)
2316 {
2317         handle_t *handle;
2318         struct inode *inode = old_dentry->d_inode;
2319         int err, retries = 0;
2320
2321         if (inode->i_nlink >= EXT4_LINK_MAX)
2322                 return -EMLINK;
2323
2324         /*
2325          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2326          * otherwise has the potential to corrupt the orphan inode list.
2327          */
2328         if (inode->i_nlink == 0)
2329                 return -ENOENT;
2330
2331 retry:
2332         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2333                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2334         if (IS_ERR(handle))
2335                 return PTR_ERR(handle);
2336
2337         if (IS_DIRSYNC(dir))
2338                 ext4_handle_sync(handle);
2339
2340         inode->i_ctime = ext4_current_time(inode);
2341         ext4_inc_count(handle, inode);
2342         atomic_inc(&inode->i_count);
2343
2344         err = ext4_add_entry(handle, dentry, inode);
2345         if (!err) {
2346                 ext4_mark_inode_dirty(handle, inode);
2347                 d_instantiate(dentry, inode);
2348         } else {
2349                 drop_nlink(inode);
2350                 iput(inode);
2351         }
2352         ext4_journal_stop(handle);
2353         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2354                 goto retry;
2355         return err;
2356 }
2357
2358 #define PARENT_INO(buffer, size) \
2359         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2360
2361 /*
2362  * Anybody can rename anything with this: the permission checks are left to the
2363  * higher-level routines.
2364  */
2365 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2366                        struct inode *new_dir, struct dentry *new_dentry)
2367 {
2368         handle_t *handle;
2369         struct inode *old_inode, *new_inode;
2370         struct buffer_head *old_bh, *new_bh, *dir_bh;
2371         struct ext4_dir_entry_2 *old_de, *new_de;
2372         int retval, force_da_alloc = 0;
2373
2374         old_bh = new_bh = dir_bh = NULL;
2375
2376         /* Initialize quotas before so that eventual writes go
2377          * in separate transaction */
2378         if (new_dentry->d_inode)
2379                 vfs_dq_init(new_dentry->d_inode);
2380         handle = ext4_journal_start(old_dir, 2 *
2381                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2382                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2383         if (IS_ERR(handle))
2384                 return PTR_ERR(handle);
2385
2386         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2387                 ext4_handle_sync(handle);
2388
2389         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2390         /*
2391          *  Check for inode number is _not_ due to possible IO errors.
2392          *  We might rmdir the source, keep it as pwd of some process
2393          *  and merrily kill the link to whatever was created under the
2394          *  same name. Goodbye sticky bit ;-<
2395          */
2396         old_inode = old_dentry->d_inode;
2397         retval = -ENOENT;
2398         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2399                 goto end_rename;
2400
2401         new_inode = new_dentry->d_inode;
2402         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2403         if (new_bh) {
2404                 if (!new_inode) {
2405                         brelse(new_bh);
2406                         new_bh = NULL;
2407                 }
2408         }
2409         if (S_ISDIR(old_inode->i_mode)) {
2410                 if (new_inode) {
2411                         retval = -ENOTEMPTY;
2412                         if (!empty_dir(new_inode))
2413                                 goto end_rename;
2414                 }
2415                 retval = -EIO;
2416                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2417                 if (!dir_bh)
2418                         goto end_rename;
2419                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2420                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2421                         goto end_rename;
2422                 retval = -EMLINK;
2423                 if (!new_inode && new_dir != old_dir &&
2424                     EXT4_DIR_LINK_MAX(new_dir))
2425                         goto end_rename;
2426         }
2427         if (!new_bh) {
2428                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2429                 if (retval)
2430                         goto end_rename;
2431         } else {
2432                 BUFFER_TRACE(new_bh, "get write access");
2433                 ext4_journal_get_write_access(handle, new_bh);
2434                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2435                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2436                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2437                         new_de->file_type = old_de->file_type;
2438                 new_dir->i_version++;
2439                 new_dir->i_ctime = new_dir->i_mtime =
2440                                         ext4_current_time(new_dir);
2441                 ext4_mark_inode_dirty(handle, new_dir);
2442                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2443                 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2444                 brelse(new_bh);
2445                 new_bh = NULL;
2446         }
2447
2448         /*
2449          * Like most other Unix systems, set the ctime for inodes on a
2450          * rename.
2451          */
2452         old_inode->i_ctime = ext4_current_time(old_inode);
2453         ext4_mark_inode_dirty(handle, old_inode);
2454
2455         /*
2456          * ok, that's it
2457          */
2458         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2459             old_de->name_len != old_dentry->d_name.len ||
2460             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2461             (retval = ext4_delete_entry(handle, old_dir,
2462                                         old_de, old_bh)) == -ENOENT) {
2463                 /* old_de could have moved from under us during htree split, so
2464                  * make sure that we are deleting the right entry.  We might
2465                  * also be pointing to a stale entry in the unused part of
2466                  * old_bh so just checking inum and the name isn't enough. */
2467                 struct buffer_head *old_bh2;
2468                 struct ext4_dir_entry_2 *old_de2;
2469
2470                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2471                 if (old_bh2) {
2472                         retval = ext4_delete_entry(handle, old_dir,
2473                                                    old_de2, old_bh2);
2474                         brelse(old_bh2);
2475                 }
2476         }
2477         if (retval) {
2478                 ext4_warning(old_dir->i_sb, "ext4_rename",
2479                                 "Deleting old file (%lu), %d, error=%d",
2480                                 old_dir->i_ino, old_dir->i_nlink, retval);
2481         }
2482
2483         if (new_inode) {
2484                 ext4_dec_count(handle, new_inode);
2485                 new_inode->i_ctime = ext4_current_time(new_inode);
2486         }
2487         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2488         ext4_update_dx_flag(old_dir);
2489         if (dir_bh) {
2490                 BUFFER_TRACE(dir_bh, "get_write_access");
2491                 ext4_journal_get_write_access(handle, dir_bh);
2492                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2493                                                 cpu_to_le32(new_dir->i_ino);
2494                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2495                 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2496                 ext4_dec_count(handle, old_dir);
2497                 if (new_inode) {
2498                         /* checked empty_dir above, can't have another parent,
2499                          * ext4_dec_count() won't work for many-linked dirs */
2500                         new_inode->i_nlink = 0;
2501                 } else {
2502                         ext4_inc_count(handle, new_dir);
2503                         ext4_update_dx_flag(new_dir);
2504                         ext4_mark_inode_dirty(handle, new_dir);
2505                 }
2506         }
2507         ext4_mark_inode_dirty(handle, old_dir);
2508         if (new_inode) {
2509                 ext4_mark_inode_dirty(handle, new_inode);
2510                 if (!new_inode->i_nlink)
2511                         ext4_orphan_add(handle, new_inode);
2512                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2513                         force_da_alloc = 1;
2514         }
2515         retval = 0;
2516
2517 end_rename:
2518         brelse(dir_bh);
2519         brelse(old_bh);
2520         brelse(new_bh);
2521         ext4_journal_stop(handle);
2522         if (retval == 0 && force_da_alloc)
2523                 ext4_alloc_da_blocks(old_inode);
2524         return retval;
2525 }
2526
2527 /*
2528  * directories can handle most operations...
2529  */
2530 const struct inode_operations ext4_dir_inode_operations = {
2531         .create         = ext4_create,
2532         .lookup         = ext4_lookup,
2533         .link           = ext4_link,
2534         .unlink         = ext4_unlink,
2535         .symlink        = ext4_symlink,
2536         .mkdir          = ext4_mkdir,
2537         .rmdir          = ext4_rmdir,
2538         .mknod          = ext4_mknod,
2539         .rename         = ext4_rename,
2540         .setattr        = ext4_setattr,
2541 #ifdef CONFIG_EXT4_FS_XATTR
2542         .setxattr       = generic_setxattr,
2543         .getxattr       = generic_getxattr,
2544         .listxattr      = ext4_listxattr,
2545         .removexattr    = generic_removexattr,
2546 #endif
2547         .check_acl      = ext4_check_acl,
2548         .fiemap         = ext4_fiemap,
2549 };
2550
2551 const struct inode_operations ext4_special_inode_operations = {
2552         .setattr        = ext4_setattr,
2553 #ifdef CONFIG_EXT4_FS_XATTR
2554         .setxattr       = generic_setxattr,
2555         .getxattr       = generic_getxattr,
2556         .listxattr      = ext4_listxattr,
2557         .removexattr    = generic_removexattr,
2558 #endif
2559         .check_acl      = ext4_check_acl,
2560 };