fs/ext4/namei.c

   1 /*
   2  *  linux/fs/ext4/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *  Hash Tree Directory indexing (c)
  20  *      Daniel Phillips, 2001
  21  *  Hash Tree Directory indexing porting
  22  *      Christopher Li, 2002
  23  *  Hash Tree Directory indexing cleanup
  24  *      Theodore Ts'o, 2002
  25  */
  26
  27 #include <linux/fs.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/jbd2.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include "ext4.h"
  38 #include "ext4_jbd2.h"
  39
  40 #include "xattr.h"
  41 #include "acl.h"
  42
  43 #include <trace/events/ext4.h>
  44 /*
  45  * define how far ahead to read directories while searching them.
  46  */
  47 #define NAMEI_RA_CHUNKS  2
  48 #define NAMEI_RA_BLOCKS  4
  49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52 static struct buffer_head *ext4_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         ext4_lblk_t *block, int *err)
  55 {
  56         struct buffer_head *bh;
  57
  58         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  59                      ((inode->i_size >> 10) >=
  60                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) {
  61                 *err = -ENOSPC;
  62                 return NULL;
  63         }
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, 1, err);
  68         if (bh) {
  69                 inode->i_size += inode->i_sb->s_blocksize;
  70                 EXT4_I(inode)->i_disksize = inode->i_size;
  71                 *err = ext4_journal_get_write_access(handle, bh);
  72                 if (*err) {
  73                         brelse(bh);
  74                         bh = NULL;
  75                 }
  76         }
  77         if (!bh && !(*err)) {
  78                 *err = -EIO;
  79                 ext4_error(inode->i_sb,
  80                            "Directory hole detected on inode %lu\n",
  81                            inode->i_ino);
  82         }
  83         return bh;
  84 }
  85
  86 #ifndef assert
  87 #define assert(test) J_ASSERT(test)
  88 #endif
  89
  90 #ifdef DX_DEBUG
  91 #define dxtrace(command) command
  92 #else
  93 #define dxtrace(command)
  94 #endif
  95
  96 struct fake_dirent
  97 {
  98         __le32 inode;
  99         __le16 rec_len;
 100         u8 name_len;
 101         u8 file_type;
 102 };
 103
 104 struct dx_countlimit
 105 {
 106         __le16 limit;
 107         __le16 count;
 108 };
 109
 110 struct dx_entry
 111 {
 112         __le32 hash;
 113         __le32 block;
 114 };
 115
 116 /*
 117  * dx_root_info is laid out so that if it should somehow get overlaid by a
 118  * dirent the two low bits of the hash version will be zero.  Therefore, the
 119  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 120  */
 121
 122 struct dx_root
 123 {
 124         struct fake_dirent dot;
 125         char dot_name[4];
 126         struct fake_dirent dotdot;
 127         char dotdot_name[4];
 128         struct dx_root_info
 129         {
 130                 __le32 reserved_zero;
 131                 u8 hash_version;
 132                 u8 info_length; /* 8 */
 133                 u8 indirect_levels;
 134                 u8 unused_flags;
 135         }
 136         info;
 137         struct dx_entry entries[0];
 138 };
 139
 140 struct dx_node
 141 {
 142         struct fake_dirent fake;
 143         struct dx_entry entries[0];
 144 };
 145
 146
 147 struct dx_frame
 148 {
 149         struct buffer_head *bh;
 150         struct dx_entry *entries;
 151         struct dx_entry *at;
 152 };
 153
 154 struct dx_map_entry
 155 {
 156         u32 hash;
 157         u16 offs;
 158         u16 size;
 159 };
 160
 161 /*
 162  * This goes at the end of each htree block.
 163  */
 164 struct dx_tail {
 165         u32 dt_reserved;
 166         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 167 };
 168
 169 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 170 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 171 static inline unsigned dx_get_hash(struct dx_entry *entry);
 172 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 173 static unsigned dx_get_count(struct dx_entry *entries);
 174 static unsigned dx_get_limit(struct dx_entry *entries);
 175 static void dx_set_count(struct dx_entry *entries, unsigned value);
 176 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 177 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 178 static unsigned dx_node_limit(struct inode *dir);
 179 static struct dx_frame *dx_probe(const struct qstr *d_name,
 180                                  struct inode *dir,
 181                                  struct dx_hash_info *hinfo,
 182                                  struct dx_frame *frame,
 183                                  int *err);
 184 static void dx_release(struct dx_frame *frames);
 185 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 186                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 187 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 188 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 189                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 190 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 191 static void dx_insert_block(struct dx_frame *frame,
 192                                         u32 hash, ext4_lblk_t block);
 193 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 194                                  struct dx_frame *frame,
 195                                  struct dx_frame *frames,
 196                                  __u32 *start_hash);
 197 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 198                 const struct qstr *d_name,
 199                 struct ext4_dir_entry_2 **res_dir,
 200                 int *err);
 201 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 202                              struct inode *inode);
 203
 204 /* checksumming functions */
 205 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 206                             unsigned int blocksize)
 207 {
 208         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 209         t->det_rec_len = ext4_rec_len_to_disk(
 210                         sizeof(struct ext4_dir_entry_tail), blocksize);
 211         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 212 }
 213
 214 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 215 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 216                                                    struct ext4_dir_entry *de)
 217 {
 218         struct ext4_dir_entry_tail *t;
 219
 220 #ifdef PARANOID
 221         struct ext4_dir_entry *d, *top;
 222
 223         d = de;
 224         top = (struct ext4_dir_entry *)(((void *)de) +
 225                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 226                 sizeof(struct ext4_dir_entry_tail)));
 227         while (d < top && d->rec_len)
 228                 d = (struct ext4_dir_entry *)(((void *)d) +
 229                     le16_to_cpu(d->rec_len));
 230
 231         if (d != top)
 232                 return NULL;
 233
 234         t = (struct ext4_dir_entry_tail *)d;
 235 #else
 236         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 237 #endif
 238
 239         if (t->det_reserved_zero1 ||
 240             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 241             t->det_reserved_zero2 ||
 242             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 243                 return NULL;
 244
 245         return t;
 246 }
 247
 248 static __le32 ext4_dirent_csum(struct inode *inode,
 249                                struct ext4_dir_entry *dirent, int size)
 250 {
 251         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 252         struct ext4_inode_info *ei = EXT4_I(inode);
 253         __u32 csum;
 254
 255         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 256         return cpu_to_le32(csum);
 257 }
 258
 259 static void warn_no_space_for_csum(struct inode *inode)
 260 {
 261         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
 262                      "checksum.  Please run e2fsck -D.", inode->i_ino);
 263 }
 264
 265 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 266 {
 267         struct ext4_dir_entry_tail *t;
 268
 269         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 270                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 271                 return 1;
 272
 273         t = get_dirent_tail(inode, dirent);
 274         if (!t) {
 275                 warn_no_space_for_csum(inode);
 276                 return 0;
 277         }
 278
 279         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 280                                                 (void *)t - (void *)dirent))
 281                 return 0;
 282
 283         return 1;
 284 }
 285
 286 static void ext4_dirent_csum_set(struct inode *inode,
 287                                  struct ext4_dir_entry *dirent)
 288 {
 289         struct ext4_dir_entry_tail *t;
 290
 291         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 292                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 293                 return;
 294
 295         t = get_dirent_tail(inode, dirent);
 296         if (!t) {
 297                 warn_no_space_for_csum(inode);
 298                 return;
 299         }
 300
 301         t->det_checksum = ext4_dirent_csum(inode, dirent,
 302                                            (void *)t - (void *)dirent);
 303 }
 304
 305 int ext4_handle_dirty_dirent_node(handle_t *handle,
 306                                   struct inode *inode,
 307                                   struct buffer_head *bh)
 308 {
 309         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 310         return ext4_handle_dirty_metadata(handle, inode, bh);
 311 }
 312
 313 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 314                                                struct ext4_dir_entry *dirent,
 315                                                int *offset)
 316 {
 317         struct ext4_dir_entry *dp;
 318         struct dx_root_info *root;
 319         int count_offset;
 320
 321         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 322                 count_offset = 8;
 323         else if (le16_to_cpu(dirent->rec_len) == 12) {
 324                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 325                 if (le16_to_cpu(dp->rec_len) !=
 326                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 327                         return NULL;
 328                 root = (struct dx_root_info *)(((void *)dp + 12));
 329                 if (root->reserved_zero ||
 330                     root->info_length != sizeof(struct dx_root_info))
 331                         return NULL;
 332                 count_offset = 32;
 333         } else
 334                 return NULL;
 335
 336         if (offset)
 337                 *offset = count_offset;
 338         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 339 }
 340
 341 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 342                            int count_offset, int count, struct dx_tail *t)
 343 {
 344         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 345         struct ext4_inode_info *ei = EXT4_I(inode);
 346         __u32 csum, old_csum;
 347         int size;
 348
 349         size = count_offset + (count * sizeof(struct dx_entry));
 350         old_csum = t->dt_checksum;
 351         t->dt_checksum = 0;
 352         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 353         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
 354         t->dt_checksum = old_csum;
 355
 356         return cpu_to_le32(csum);
 357 }
 358
 359 static int ext4_dx_csum_verify(struct inode *inode,
 360                                struct ext4_dir_entry *dirent)
 361 {
 362         struct dx_countlimit *c;
 363         struct dx_tail *t;
 364         int count_offset, limit, count;
 365
 366         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 367                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 368                 return 1;
 369
 370         c = get_dx_countlimit(inode, dirent, &count_offset);
 371         if (!c) {
 372                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 373                 return 1;
 374         }
 375         limit = le16_to_cpu(c->limit);
 376         count = le16_to_cpu(c->count);
 377         if (count_offset + (limit * sizeof(struct dx_entry)) >
 378             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 379                 warn_no_space_for_csum(inode);
 380                 return 1;
 381         }
 382         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 383
 384         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 385                                             count, t))
 386                 return 0;
 387         return 1;
 388 }
 389
 390 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 391 {
 392         struct dx_countlimit *c;
 393         struct dx_tail *t;
 394         int count_offset, limit, count;
 395
 396         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 397                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 398                 return;
 399
 400         c = get_dx_countlimit(inode, dirent, &count_offset);
 401         if (!c) {
 402                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 403                 return;
 404         }
 405         limit = le16_to_cpu(c->limit);
 406         count = le16_to_cpu(c->count);
 407         if (count_offset + (limit * sizeof(struct dx_entry)) >
 408             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 409                 warn_no_space_for_csum(inode);
 410                 return;
 411         }
 412         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 413
 414         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 415 }
 416
 417 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 418                                             struct inode *inode,
 419                                             struct buffer_head *bh)
 420 {
 421         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 422         return ext4_handle_dirty_metadata(handle, inode, bh);
 423 }
 424
 425 /*
 426  * p is at least 6 bytes before the end of page
 427  */
 428 static inline struct ext4_dir_entry_2 *
 429 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 430 {
 431         return (struct ext4_dir_entry_2 *)((char *)p +
 432                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 433 }
 434
 435 /*
 436  * Future: use high four bits of block for coalesce-on-delete flags
 437  * Mask them off for now.
 438  */
 439
 440 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 441 {
 442         return le32_to_cpu(entry->block) & 0x00ffffff;
 443 }
 444
 445 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 446 {
 447         entry->block = cpu_to_le32(value);
 448 }
 449
 450 static inline unsigned dx_get_hash(struct dx_entry *entry)
 451 {
 452         return le32_to_cpu(entry->hash);
 453 }
 454
 455 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 456 {
 457         entry->hash = cpu_to_le32(value);
 458 }
 459
 460 static inline unsigned dx_get_count(struct dx_entry *entries)
 461 {
 462         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 463 }
 464
 465 static inline unsigned dx_get_limit(struct dx_entry *entries)
 466 {
 467         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 468 }
 469
 470 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 471 {
 472         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 473 }
 474
 475 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 476 {
 477         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 478 }
 479
 480 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 481 {
 482         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 483                 EXT4_DIR_REC_LEN(2) - infosize;
 484
 485         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 486                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 487                 entry_space -= sizeof(struct dx_tail);
 488         return entry_space / sizeof(struct dx_entry);
 489 }
 490
 491 static inline unsigned dx_node_limit(struct inode *dir)
 492 {
 493         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 494
 495         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 496                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 497                 entry_space -= sizeof(struct dx_tail);
 498         return entry_space / sizeof(struct dx_entry);
 499 }
 500
 501 /*
 502  * Debug
 503  */
 504 #ifdef DX_DEBUG
 505 static void dx_show_index(char * label, struct dx_entry *entries)
 506 {
 507         int i, n = dx_get_count (entries);
 508         printk(KERN_DEBUG "%s index ", label);
 509         for (i = 0; i < n; i++) {
 510                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 511                                 0, (unsigned long)dx_get_block(entries + i));
 512         }
 513         printk("\n");
 514 }
 515
 516 struct stats
 517 {
 518         unsigned names;
 519         unsigned space;
 520         unsigned bcount;
 521 };
 522
 523 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 524                                  int size, int show_names)
 525 {
 526         unsigned names = 0, space = 0;
 527         char *base = (char *) de;
 528         struct dx_hash_info h = *hinfo;
 529
 530         printk("names: ");
 531         while ((char *) de < base + size)
 532         {
 533                 if (de->inode)
 534                 {
 535                         if (show_names)
 536                         {
 537                                 int len = de->name_len;
 538                                 char *name = de->name;
 539                                 while (len--) printk("%c", *name++);
 540                                 ext4fs_dirhash(de->name, de->name_len, &h);
 541                                 printk(":%x.%u ", h.hash,
 542                                        (unsigned) ((char *) de - base));
 543                         }
 544                         space += EXT4_DIR_REC_LEN(de->name_len);
 545                         names++;
 546                 }
 547                 de = ext4_next_entry(de, size);
 548         }
 549         printk("(%i)\n", names);
 550         return (struct stats) { names, space, 1 };
 551 }
 552
 553 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 554                              struct dx_entry *entries, int levels)
 555 {
 556         unsigned blocksize = dir->i_sb->s_blocksize;
 557         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 558         unsigned bcount = 0;
 559         struct buffer_head *bh;
 560         int err;
 561         printk("%i indexed blocks...\n", count);
 562         for (i = 0; i < count; i++, entries++)
 563         {
 564                 ext4_lblk_t block = dx_get_block(entries);
 565                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 566                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 567                 struct stats stats;
 568                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 569                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 570                 stats = levels?
 571                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 572                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 573                 names += stats.names;
 574                 space += stats.space;
 575                 bcount += stats.bcount;
 576                 brelse(bh);
 577         }
 578         if (bcount)
 579                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 580                        levels ? "" : "   ", names, space/bcount,
 581                        (space/bcount)*100/blocksize);
 582         return (struct stats) { names, space, bcount};
 583 }
 584 #endif /* DX_DEBUG */
 585
 586 /*
 587  * Probe for a directory leaf block to search.
 588  *
 589  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 590  * error in the directory index, and the caller should fall back to
 591  * searching the directory normally.  The callers of dx_probe **MUST**
 592  * check for this error code, and make sure it never gets reflected
 593  * back to userspace.
 594  */
 595 static struct dx_frame *
 596 dx_probe(const struct qstr *d_name, struct inode *dir,
 597          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 598 {
 599         unsigned count, indirect;
 600         struct dx_entry *at, *entries, *p, *q, *m;
 601         struct dx_root *root;
 602         struct buffer_head *bh;
 603         struct dx_frame *frame = frame_in;
 604         u32 hash;
 605
 606         frame->bh = NULL;
 607         if (!(bh = ext4_bread(NULL, dir, 0, 0, err))) {
 608                 if (*err == 0)
 609                         *err = ERR_BAD_DX_DIR;
 610                 goto fail;
 611         }
 612         root = (struct dx_root *) bh->b_data;
 613         if (root->info.hash_version != DX_HASH_TEA &&
 614             root->info.hash_version != DX_HASH_HALF_MD4 &&
 615             root->info.hash_version != DX_HASH_LEGACY) {
 616                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 617                              root->info.hash_version);
 618                 brelse(bh);
 619                 *err = ERR_BAD_DX_DIR;
 620                 goto fail;
 621         }
 622         hinfo->hash_version = root->info.hash_version;
 623         if (hinfo->hash_version <= DX_HASH_TEA)
 624                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 625         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 626         if (d_name)
 627                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 628         hash = hinfo->hash;
 629
 630         if (root->info.unused_flags & 1) {
 631                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 632                              root->info.unused_flags);
 633                 brelse(bh);
 634                 *err = ERR_BAD_DX_DIR;
 635                 goto fail;
 636         }
 637
 638         if ((indirect = root->info.indirect_levels) > 1) {
 639                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 640                              root->info.indirect_levels);
 641                 brelse(bh);
 642                 *err = ERR_BAD_DX_DIR;
 643                 goto fail;
 644         }
 645
 646         if (!buffer_verified(bh) &&
 647             !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
 648                 ext4_warning(dir->i_sb, "Root failed checksum");
 649                 brelse(bh);
 650                 *err = ERR_BAD_DX_DIR;
 651                 goto fail;
 652         }
 653         set_buffer_verified(bh);
 654
 655         entries = (struct dx_entry *) (((char *)&root->info) +
 656                                        root->info.info_length);
 657
 658         if (dx_get_limit(entries) != dx_root_limit(dir,
 659                                                    root->info.info_length)) {
 660                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 661                 brelse(bh);
 662                 *err = ERR_BAD_DX_DIR;
 663                 goto fail;
 664         }
 665
 666         dxtrace(printk("Look up %x", hash));
 667         while (1)
 668         {
 669                 count = dx_get_count(entries);
 670                 if (!count || count > dx_get_limit(entries)) {
 671                         ext4_warning(dir->i_sb,
 672                                      "dx entry: no count or count > limit");
 673                         brelse(bh);
 674                         *err = ERR_BAD_DX_DIR;
 675                         goto fail2;
 676                 }
 677
 678                 p = entries + 1;
 679                 q = entries + count - 1;
 680                 while (p <= q)
 681                 {
 682                         m = p + (q - p)/2;
 683                         dxtrace(printk("."));
 684                         if (dx_get_hash(m) > hash)
 685                                 q = m - 1;
 686                         else
 687                                 p = m + 1;
 688                 }
 689
 690                 if (0) // linear search cross check
 691                 {
 692                         unsigned n = count - 1;
 693                         at = entries;
 694                         while (n--)
 695                         {
 696                                 dxtrace(printk(","));
 697                                 if (dx_get_hash(++at) > hash)
 698                                 {
 699                                         at--;
 700                                         break;
 701                                 }
 702                         }
 703                         assert (at == p - 1);
 704                 }
 705
 706                 at = p - 1;
 707                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 708                 frame->bh = bh;
 709                 frame->entries = entries;
 710                 frame->at = at;
 711                 if (!indirect--) return frame;
 712                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(at), 0, err))) {
 713                         if (!(*err))
 714                                 *err = ERR_BAD_DX_DIR;
 715                         goto fail2;
 716                 }
 717                 entries = ((struct dx_node *) bh->b_data)->entries;
 718
 719                 if (!buffer_verified(bh) &&
 720                     !ext4_dx_csum_verify(dir,
 721                                          (struct ext4_dir_entry *)bh->b_data)) {
 722                         ext4_warning(dir->i_sb, "Node failed checksum");
 723                         brelse(bh);
 724                         *err = ERR_BAD_DX_DIR;
 725                         goto fail2;
 726                 }
 727                 set_buffer_verified(bh);
 728
 729                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 730                         ext4_warning(dir->i_sb,
 731                                      "dx entry: limit != node limit");
 732                         brelse(bh);
 733                         *err = ERR_BAD_DX_DIR;
 734                         goto fail2;
 735                 }
 736                 frame++;
 737                 frame->bh = NULL;
 738         }
 739 fail2:
 740         while (frame >= frame_in) {
 741                 brelse(frame->bh);
 742                 frame--;
 743         }
 744 fail:
 745         if (*err == ERR_BAD_DX_DIR)
 746                 ext4_warning(dir->i_sb,
 747                              "Corrupt dir inode %lu, running e2fsck is "
 748                              "recommended.", dir->i_ino);
 749         return NULL;
 750 }
 751
 752 static void dx_release (struct dx_frame *frames)
 753 {
 754         if (frames[0].bh == NULL)
 755                 return;
 756
 757         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 758                 brelse(frames[1].bh);
 759         brelse(frames[0].bh);
 760 }
 761
 762 /*
 763  * This function increments the frame pointer to search the next leaf
 764  * block, and reads in the necessary intervening nodes if the search
 765  * should be necessary.  Whether or not the search is necessary is
 766  * controlled by the hash parameter.  If the hash value is even, then
 767  * the search is only continued if the next block starts with that
 768  * hash value.  This is used if we are searching for a specific file.
 769  *
 770  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 771  *
 772  * This function returns 1 if the caller should continue to search,
 773  * or 0 if it should not.  If there is an error reading one of the
 774  * index blocks, it will a negative error code.
 775  *
 776  * If start_hash is non-null, it will be filled in with the starting
 777  * hash of the next page.
 778  */
 779 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 780                                  struct dx_frame *frame,
 781                                  struct dx_frame *frames,
 782                                  __u32 *start_hash)
 783 {
 784         struct dx_frame *p;
 785         struct buffer_head *bh;
 786         int err, num_frames = 0;
 787         __u32 bhash;
 788
 789         p = frame;
 790         /*
 791          * Find the next leaf page by incrementing the frame pointer.
 792          * If we run out of entries in the interior node, loop around and
 793          * increment pointer in the parent node.  When we break out of
 794          * this loop, num_frames indicates the number of interior
 795          * nodes need to be read.
 796          */
 797         while (1) {
 798                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 799                         break;
 800                 if (p == frames)
 801                         return 0;
 802                 num_frames++;
 803                 p--;
 804         }
 805
 806         /*
 807          * If the hash is 1, then continue only if the next page has a
 808          * continuation hash of any value.  This is used for readdir
 809          * handling.  Otherwise, check to see if the hash matches the
 810          * desired contiuation hash.  If it doesn't, return since
 811          * there's no point to read in the successive index pages.
 812          */
 813         bhash = dx_get_hash(p->at);
 814         if (start_hash)
 815                 *start_hash = bhash;
 816         if ((hash & 1) == 0) {
 817                 if ((bhash & ~1) != hash)
 818                         return 0;
 819         }
 820         /*
 821          * If the hash is HASH_NB_ALWAYS, we always go to the next
 822          * block so no check is necessary
 823          */
 824         while (num_frames--) {
 825                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 826                                       0, &err))) {
 827                         if (!err) {
 828                                 ext4_error(dir->i_sb,
 829                                            "Directory hole detected on inode %lu\n",
 830                                            dir->i_ino);
 831                                 return -EIO;
 832                         }
 833                         return err; /* Failure */
 834                 }
 835
 836                 if (!buffer_verified(bh) &&
 837                     !ext4_dx_csum_verify(dir,
 838                                          (struct ext4_dir_entry *)bh->b_data)) {
 839                         ext4_warning(dir->i_sb, "Node failed checksum");
 840                         brelse(bh);
 841                         return -EIO;
 842                 }
 843                 set_buffer_verified(bh);
 844
 845                 p++;
 846                 brelse(p->bh);
 847                 p->bh = bh;
 848                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 849         }
 850         return 1;
 851 }
 852
 853
 854 /*
 855  * This function fills a red-black tree with information from a
 856  * directory block.  It returns the number directory entries loaded
 857  * into the tree.  If there is an error it is returned in err.
 858  */
 859 static int htree_dirblock_to_tree(struct file *dir_file,
 860                                   struct inode *dir, ext4_lblk_t block,
 861                                   struct dx_hash_info *hinfo,
 862                                   __u32 start_hash, __u32 start_minor_hash)
 863 {
 864         struct buffer_head *bh;
 865         struct ext4_dir_entry_2 *de, *top;
 866         int err = 0, count = 0;
 867
 868         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 869                                                         (unsigned long)block));
 870         if (!(bh = ext4_bread(NULL, dir, block, 0, &err))) {
 871                 if (!err) {
 872                         err = -EIO;
 873                         ext4_error(dir->i_sb,
 874                                    "Directory hole detected on inode %lu\n",
 875                                    dir->i_ino);
 876                 }
 877                 return err;
 878         }
 879
 880         if (!buffer_verified(bh) &&
 881                         !ext4_dirent_csum_verify(dir,
 882                                 (struct ext4_dir_entry *)bh->b_data)) {
 883                 brelse(bh);
 884                 return -EIO;
 885         }
 886         set_buffer_verified(bh);
 887
 888         de = (struct ext4_dir_entry_2 *) bh->b_data;
 889         top = (struct ext4_dir_entry_2 *) ((char *) de +
 890                                            dir->i_sb->s_blocksize -
 891                                            EXT4_DIR_REC_LEN(0));
 892         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 893                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 894                                 bh->b_data, bh->b_size,
 895                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 896                                          + ((char *)de - bh->b_data))) {
 897                         /* On error, skip the f_pos to the next block. */
 898                         dir_file->f_pos = (dir_file->f_pos |
 899                                         (dir->i_sb->s_blocksize - 1)) + 1;
 900                         brelse(bh);
 901                         return count;
 902                 }
 903                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 904                 if ((hinfo->hash < start_hash) ||
 905                     ((hinfo->hash == start_hash) &&
 906                      (hinfo->minor_hash < start_minor_hash)))
 907                         continue;
 908                 if (de->inode == 0)
 909                         continue;
 910                 if ((err = ext4_htree_store_dirent(dir_file,
 911                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 912                         brelse(bh);
 913                         return err;
 914                 }
 915                 count++;
 916         }
 917         brelse(bh);
 918         return count;
 919 }
 920
 921
 922 /*
 923  * This function fills a red-black tree with information from a
 924  * directory.  We start scanning the directory in hash order, starting
 925  * at start_hash and start_minor_hash.
 926  *
 927  * This function returns the number of entries inserted into the tree,
 928  * or a negative error code.
 929  */
 930 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 931                          __u32 start_minor_hash, __u32 *next_hash)
 932 {
 933         struct dx_hash_info hinfo;
 934         struct ext4_dir_entry_2 *de;
 935         struct dx_frame frames[2], *frame;
 936         struct inode *dir;
 937         ext4_lblk_t block;
 938         int count = 0;
 939         int ret, err;
 940         __u32 hashval;
 941
 942         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 943                        start_hash, start_minor_hash));
 944         dir = dir_file->f_path.dentry->d_inode;
 945         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 946                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 947                 if (hinfo.hash_version <= DX_HASH_TEA)
 948                         hinfo.hash_version +=
 949                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 950                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 951                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 952                                                start_hash, start_minor_hash);
 953                 *next_hash = ~0;
 954                 return count;
 955         }
 956         hinfo.hash = start_hash;
 957         hinfo.minor_hash = 0;
 958         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 959         if (!frame)
 960                 return err;
 961
 962         /* Add '.' and '..' from the htree header */
 963         if (!start_hash && !start_minor_hash) {
 964                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 965                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 966                         goto errout;
 967                 count++;
 968         }
 969         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 970                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 971                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 972                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 973                         goto errout;
 974                 count++;
 975         }
 976
 977         while (1) {
 978                 block = dx_get_block(frame->at);
 979                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 980                                              start_hash, start_minor_hash);
 981                 if (ret < 0) {
 982                         err = ret;
 983                         goto errout;
 984                 }
 985                 count += ret;
 986                 hashval = ~0;
 987                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 988                                             frame, frames, &hashval);
 989                 *next_hash = hashval;
 990                 if (ret < 0) {
 991                         err = ret;
 992                         goto errout;
 993                 }
 994                 /*
 995                  * Stop if:  (a) there are no more entries, or
 996                  * (b) we have inserted at least one entry and the
 997                  * next hash value is not a continuation
 998                  */
 999                 if ((ret == 0) ||
1000                     (count && ((hashval & 1) == 0)))
1001                         break;
1002         }
1003         dx_release(frames);
1004         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1005                        "next hash: %x\n", count, *next_hash));
1006         return count;
1007 errout:
1008         dx_release(frames);
1009         return (err);
1010 }
1011
1012 static inline int search_dirblock(struct buffer_head *bh,
1013                                   struct inode *dir,
1014                                   const struct qstr *d_name,
1015                                   unsigned int offset,
1016                                   struct ext4_dir_entry_2 **res_dir)
1017 {
1018         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1019                           d_name, offset, res_dir);
1020 }
1021
1022 /*
1023  * Directory block splitting, compacting
1024  */
1025
1026 /*
1027  * Create map of hash values, offsets, and sizes, stored at end of block.
1028  * Returns number of entries mapped.
1029  */
1030 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1031                        struct dx_hash_info *hinfo,
1032                        struct dx_map_entry *map_tail)
1033 {
1034         int count = 0;
1035         char *base = (char *) de;
1036         struct dx_hash_info h = *hinfo;
1037
1038         while ((char *) de < base + blocksize) {
1039                 if (de->name_len && de->inode) {
1040                         ext4fs_dirhash(de->name, de->name_len, &h);
1041                         map_tail--;
1042                         map_tail->hash = h.hash;
1043                         map_tail->offs = ((char *) de - base)>>2;
1044                         map_tail->size = le16_to_cpu(de->rec_len);
1045                         count++;
1046                         cond_resched();
1047                 }
1048                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1049                 de = ext4_next_entry(de, blocksize);
1050         }
1051         return count;
1052 }
1053
1054 /* Sort map by hash value */
1055 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1056 {
1057         struct dx_map_entry *p, *q, *top = map + count - 1;
1058         int more;
1059         /* Combsort until bubble sort doesn't suck */
1060         while (count > 2) {
1061                 count = count*10/13;
1062                 if (count - 9 < 2) /* 9, 10 -> 11 */
1063                         count = 11;
1064                 for (p = top, q = p - count; q >= map; p--, q--)
1065                         if (p->hash < q->hash)
1066                                 swap(*p, *q);
1067         }
1068         /* Garden variety bubble sort */
1069         do {
1070                 more = 0;
1071                 q = top;
1072                 while (q-- > map) {
1073                         if (q[1].hash >= q[0].hash)
1074                                 continue;
1075                         swap(*(q+1), *q);
1076                         more = 1;
1077                 }
1078         } while(more);
1079 }
1080
1081 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1082 {
1083         struct dx_entry *entries = frame->entries;
1084         struct dx_entry *old = frame->at, *new = old + 1;
1085         int count = dx_get_count(entries);
1086
1087         assert(count < dx_get_limit(entries));
1088         assert(old < entries + count);
1089         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1090         dx_set_hash(new, hash);
1091         dx_set_block(new, block);
1092         dx_set_count(entries, count + 1);
1093 }
1094
1095 /*
1096  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1097  *
1098  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1099  * `de != NULL' is guaranteed by caller.
1100  */
1101 static inline int ext4_match (int len, const char * const name,
1102                               struct ext4_dir_entry_2 * de)
1103 {
1104         if (len != de->name_len)
1105                 return 0;
1106         if (!de->inode)
1107                 return 0;
1108         return !memcmp(name, de->name, len);
1109 }
1110
1111 /*
1112  * Returns 0 if not found, -1 on failure, and 1 on success
1113  */
1114 int search_dir(struct buffer_head *bh,
1115                char *search_buf,
1116                int buf_size,
1117                struct inode *dir,
1118                const struct qstr *d_name,
1119                unsigned int offset,
1120                struct ext4_dir_entry_2 **res_dir)
1121 {
1122         struct ext4_dir_entry_2 * de;
1123         char * dlimit;
1124         int de_len;
1125         const char *name = d_name->name;
1126         int namelen = d_name->len;
1127
1128         de = (struct ext4_dir_entry_2 *)search_buf;
1129         dlimit = search_buf + buf_size;
1130         while ((char *) de < dlimit) {
1131                 /* this code is executed quadratically often */
1132                 /* do minimal checking `by hand' */
1133
1134                 if ((char *) de + namelen <= dlimit &&
1135                     ext4_match (namelen, name, de)) {
1136                         /* found a match - just to be sure, do a full check */
1137                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1138                                                  bh->b_size, offset))
1139                                 return -1;
1140                         *res_dir = de;
1141                         return 1;
1142                 }
1143                 /* prevent looping on a bad block */
1144                 de_len = ext4_rec_len_from_disk(de->rec_len,
1145                                                 dir->i_sb->s_blocksize);
1146                 if (de_len <= 0)
1147                         return -1;
1148                 offset += de_len;
1149                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1150         }
1151         return 0;
1152 }
1153
1154 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1155                                struct ext4_dir_entry *de)
1156 {
1157         struct super_block *sb = dir->i_sb;
1158
1159         if (!is_dx(dir))
1160                 return 0;
1161         if (block == 0)
1162                 return 1;
1163         if (de->inode == 0 &&
1164             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1165                         sb->s_blocksize)
1166                 return 1;
1167         return 0;
1168 }
1169
1170 /*
1171  *      ext4_find_entry()
1172  *
1173  * finds an entry in the specified directory with the wanted name. It
1174  * returns the cache buffer in which the entry was found, and the entry
1175  * itself (as a parameter - res_dir). It does NOT read the inode of the
1176  * entry - you'll have to do that yourself if you want to.
1177  *
1178  * The returned buffer_head has ->b_count elevated.  The caller is expected
1179  * to brelse() it when appropriate.
1180  */
1181 static struct buffer_head * ext4_find_entry (struct inode *dir,
1182                                         const struct qstr *d_name,
1183                                         struct ext4_dir_entry_2 **res_dir,
1184                                         int *inlined)
1185 {
1186         struct super_block *sb;
1187         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1188         struct buffer_head *bh, *ret = NULL;
1189         ext4_lblk_t start, block, b;
1190         const u8 *name = d_name->name;
1191         int ra_max = 0;         /* Number of bh's in the readahead
1192                                    buffer, bh_use[] */
1193         int ra_ptr = 0;         /* Current index into readahead
1194                                    buffer */
1195         int num = 0;
1196         ext4_lblk_t  nblocks;
1197         int i, err;
1198         int namelen;
1199
1200         *res_dir = NULL;
1201         sb = dir->i_sb;
1202         namelen = d_name->len;
1203         if (namelen > EXT4_NAME_LEN)
1204                 return NULL;
1205
1206         if (ext4_has_inline_data(dir)) {
1207                 int has_inline_data = 1;
1208                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1209                                              &has_inline_data);
1210                 if (has_inline_data) {
1211                         if (inlined)
1212                                 *inlined = 1;
1213                         return ret;
1214                 }
1215         }
1216
1217         if ((namelen <= 2) && (name[0] == '.') &&
1218             (name[1] == '.' || name[1] == '\0')) {
1219                 /*
1220                  * "." or ".." will only be in the first block
1221                  * NFS may look up ".."; "." should be handled by the VFS
1222                  */
1223                 block = start = 0;
1224                 nblocks = 1;
1225                 goto restart;
1226         }
1227         if (is_dx(dir)) {
1228                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1229                 /*
1230                  * On success, or if the error was file not found,
1231                  * return.  Otherwise, fall back to doing a search the
1232                  * old fashioned way.
1233                  */
1234                 if (bh || (err != ERR_BAD_DX_DIR))
1235                         return bh;
1236                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1237                                "falling back\n"));
1238         }
1239         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1240         start = EXT4_I(dir)->i_dir_start_lookup;
1241         if (start >= nblocks)
1242                 start = 0;
1243         block = start;
1244 restart:
1245         do {
1246                 /*
1247                  * We deal with the read-ahead logic here.
1248                  */
1249                 if (ra_ptr >= ra_max) {
1250                         /* Refill the readahead buffer */
1251                         ra_ptr = 0;
1252                         b = block;
1253                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1254                                 /*
1255                                  * Terminate if we reach the end of the
1256                                  * directory and must wrap, or if our
1257                                  * search has finished at this block.
1258                                  */
1259                                 if (b >= nblocks || (num && block == start)) {
1260                                         bh_use[ra_max] = NULL;
1261                                         break;
1262                                 }
1263                                 num++;
1264                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1265                                 bh_use[ra_max] = bh;
1266                                 if (bh)
1267                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1268                                                     1, &bh);
1269                         }
1270                 }
1271                 if ((bh = bh_use[ra_ptr++]) == NULL)
1272                         goto next;
1273                 wait_on_buffer(bh);
1274                 if (!buffer_uptodate(bh)) {
1275                         /* read error, skip block & hope for the best */
1276                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1277                                          (unsigned long) block);
1278                         brelse(bh);
1279                         goto next;
1280                 }
1281                 if (!buffer_verified(bh) &&
1282                     !is_dx_internal_node(dir, block,
1283                                          (struct ext4_dir_entry *)bh->b_data) &&
1284                     !ext4_dirent_csum_verify(dir,
1285                                 (struct ext4_dir_entry *)bh->b_data)) {
1286                         EXT4_ERROR_INODE(dir, "checksumming directory "
1287                                          "block %lu", (unsigned long)block);
1288                         brelse(bh);
1289                         goto next;
1290                 }
1291                 set_buffer_verified(bh);
1292                 i = search_dirblock(bh, dir, d_name,
1293                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1294                 if (i == 1) {
1295                         EXT4_I(dir)->i_dir_start_lookup = block;
1296                         ret = bh;
1297                         goto cleanup_and_exit;
1298                 } else {
1299                         brelse(bh);
1300                         if (i < 0)
1301                                 goto cleanup_and_exit;
1302                 }
1303         next:
1304                 if (++block >= nblocks)
1305                         block = 0;
1306         } while (block != start);
1307
1308         /*
1309          * If the directory has grown while we were searching, then
1310          * search the last part of the directory before giving up.
1311          */
1312         block = nblocks;
1313         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1314         if (block < nblocks) {
1315                 start = 0;
1316                 goto restart;
1317         }
1318
1319 cleanup_and_exit:
1320         /* Clean up the read-ahead blocks */
1321         for (; ra_ptr < ra_max; ra_ptr++)
1322                 brelse(bh_use[ra_ptr]);
1323         return ret;
1324 }
1325
1326 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1327                        struct ext4_dir_entry_2 **res_dir, int *err)
1328 {
1329         struct super_block * sb = dir->i_sb;
1330         struct dx_hash_info     hinfo;
1331         struct dx_frame frames[2], *frame;
1332         struct buffer_head *bh;
1333         ext4_lblk_t block;
1334         int retval;
1335
1336         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1337                 return NULL;
1338         do {
1339                 block = dx_get_block(frame->at);
1340                 if (!(bh = ext4_bread(NULL, dir, block, 0, err))) {
1341                         if (!(*err)) {
1342                                 *err = -EIO;
1343                                 ext4_error(dir->i_sb,
1344                                            "Directory hole detected on inode %lu\n",
1345                                            dir->i_ino);
1346                         }
1347                         goto errout;
1348                 }
1349
1350                 if (!buffer_verified(bh) &&
1351                     !ext4_dirent_csum_verify(dir,
1352                                 (struct ext4_dir_entry *)bh->b_data)) {
1353                         EXT4_ERROR_INODE(dir, "checksumming directory "
1354                                          "block %lu", (unsigned long)block);
1355                         brelse(bh);
1356                         *err = -EIO;
1357                         goto errout;
1358                 }
1359                 set_buffer_verified(bh);
1360                 retval = search_dirblock(bh, dir, d_name,
1361                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1362                                          res_dir);
1363                 if (retval == 1) {      /* Success! */
1364                         dx_release(frames);
1365                         return bh;
1366                 }
1367                 brelse(bh);
1368                 if (retval == -1) {
1369                         *err = ERR_BAD_DX_DIR;
1370                         goto errout;
1371                 }
1372
1373                 /* Check to see if we should continue to search */
1374                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1375                                                frames, NULL);
1376                 if (retval < 0) {
1377                         ext4_warning(sb,
1378                              "error reading index page in directory #%lu",
1379                              dir->i_ino);
1380                         *err = retval;
1381                         goto errout;
1382                 }
1383         } while (retval == 1);
1384
1385         *err = -ENOENT;
1386 errout:
1387         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1388         dx_release (frames);
1389         return NULL;
1390 }
1391
1392 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1393 {
1394         struct inode *inode;
1395         struct ext4_dir_entry_2 *de;
1396         struct buffer_head *bh;
1397
1398         if (dentry->d_name.len > EXT4_NAME_LEN)
1399                 return ERR_PTR(-ENAMETOOLONG);
1400
1401         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1402         inode = NULL;
1403         if (bh) {
1404                 __u32 ino = le32_to_cpu(de->inode);
1405                 brelse(bh);
1406                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1407                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1408                         return ERR_PTR(-EIO);
1409                 }
1410                 if (unlikely(ino == dir->i_ino)) {
1411                         EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1412                                          dentry->d_name.len,
1413                                          dentry->d_name.name);
1414                         return ERR_PTR(-EIO);
1415                 }
1416                 inode = ext4_iget(dir->i_sb, ino);
1417                 if (inode == ERR_PTR(-ESTALE)) {
1418                         EXT4_ERROR_INODE(dir,
1419                                          "deleted inode referenced: %u",
1420                                          ino);
1421                         return ERR_PTR(-EIO);
1422                 }
1423         }
1424         return d_splice_alias(inode, dentry);
1425 }
1426
1427
1428 struct dentry *ext4_get_parent(struct dentry *child)
1429 {
1430         __u32 ino;
1431         static const struct qstr dotdot = QSTR_INIT("..", 2);
1432         struct ext4_dir_entry_2 * de;
1433         struct buffer_head *bh;
1434
1435         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1436         if (!bh)
1437                 return ERR_PTR(-ENOENT);
1438         ino = le32_to_cpu(de->inode);
1439         brelse(bh);
1440
1441         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1442                 EXT4_ERROR_INODE(child->d_inode,
1443                                  "bad parent inode number: %u", ino);
1444                 return ERR_PTR(-EIO);
1445         }
1446
1447         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1448 }
1449
1450 #define S_SHIFT 12
1451 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1452         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1453         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1454         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1455         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1456         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1457         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1458         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1459 };
1460
1461 static inline void ext4_set_de_type(struct super_block *sb,
1462                                 struct ext4_dir_entry_2 *de,
1463                                 umode_t mode) {
1464         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1465                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1466 }
1467
1468 /*
1469  * Move count entries from end of map between two memory locations.
1470  * Returns pointer to last entry moved.
1471  */
1472 static struct ext4_dir_entry_2 *
1473 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1474                 unsigned blocksize)
1475 {
1476         unsigned rec_len = 0;
1477
1478         while (count--) {
1479                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1480                                                 (from + (map->offs<<2));
1481                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1482                 memcpy (to, de, rec_len);
1483                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1484                                 ext4_rec_len_to_disk(rec_len, blocksize);
1485                 de->inode = 0;
1486                 map++;
1487                 to += rec_len;
1488         }
1489         return (struct ext4_dir_entry_2 *) (to - rec_len);
1490 }
1491
1492 /*
1493  * Compact each dir entry in the range to the minimal rec_len.
1494  * Returns pointer to last entry in range.
1495  */
1496 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1497 {
1498         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1499         unsigned rec_len = 0;
1500
1501         prev = to = de;
1502         while ((char*)de < base + blocksize) {
1503                 next = ext4_next_entry(de, blocksize);
1504                 if (de->inode && de->name_len) {
1505                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1506                         if (de > to)
1507                                 memmove(to, de, rec_len);
1508                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1509                         prev = to;
1510                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1511                 }
1512                 de = next;
1513         }
1514         return prev;
1515 }
1516
1517 /*
1518  * Split a full leaf block to make room for a new dir entry.
1519  * Allocate a new block, and move entries so that they are approx. equally full.
1520  * Returns pointer to de in block into which the new entry will be inserted.
1521  */
1522 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1523                         struct buffer_head **bh,struct dx_frame *frame,
1524                         struct dx_hash_info *hinfo, int *error)
1525 {
1526         unsigned blocksize = dir->i_sb->s_blocksize;
1527         unsigned count, continued;
1528         struct buffer_head *bh2;
1529         ext4_lblk_t newblock;
1530         u32 hash2;
1531         struct dx_map_entry *map;
1532         char *data1 = (*bh)->b_data, *data2;
1533         unsigned split, move, size;
1534         struct ext4_dir_entry_2 *de = NULL, *de2;
1535         struct ext4_dir_entry_tail *t;
1536         int     csum_size = 0;
1537         int     err = 0, i;
1538
1539         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1540                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1541                 csum_size = sizeof(struct ext4_dir_entry_tail);
1542
1543         bh2 = ext4_append (handle, dir, &newblock, &err);
1544         if (!(bh2)) {
1545                 brelse(*bh);
1546                 *bh = NULL;
1547                 goto errout;
1548         }
1549
1550         BUFFER_TRACE(*bh, "get_write_access");
1551         err = ext4_journal_get_write_access(handle, *bh);
1552         if (err)
1553                 goto journal_error;
1554
1555         BUFFER_TRACE(frame->bh, "get_write_access");
1556         err = ext4_journal_get_write_access(handle, frame->bh);
1557         if (err)
1558                 goto journal_error;
1559
1560         data2 = bh2->b_data;
1561
1562         /* create map in the end of data2 block */
1563         map = (struct dx_map_entry *) (data2 + blocksize);
1564         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1565                              blocksize, hinfo, map);
1566         map -= count;
1567         dx_sort_map(map, count);
1568         /* Split the existing block in the middle, size-wise */
1569         size = 0;
1570         move = 0;
1571         for (i = count-1; i >= 0; i--) {
1572                 /* is more than half of this entry in 2nd half of the block? */
1573                 if (size + map[i].size/2 > blocksize/2)
1574                         break;
1575                 size += map[i].size;
1576                 move++;
1577         }
1578         /* map index at which we will split */
1579         split = count - move;
1580         hash2 = map[split].hash;
1581         continued = hash2 == map[split - 1].hash;
1582         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1583                         (unsigned long)dx_get_block(frame->at),
1584                                         hash2, split, count-split));
1585
1586         /* Fancy dance to stay within two buffers */
1587         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1588         de = dx_pack_dirents(data1, blocksize);
1589         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1590                                            (char *) de,
1591                                            blocksize);
1592         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1593                                             (char *) de2,
1594                                             blocksize);
1595         if (csum_size) {
1596                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1597                 initialize_dirent_tail(t, blocksize);
1598
1599                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1600                 initialize_dirent_tail(t, blocksize);
1601         }
1602
1603         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1604         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1605
1606         /* Which block gets the new entry? */
1607         if (hinfo->hash >= hash2)
1608         {
1609                 swap(*bh, bh2);
1610                 de = de2;
1611         }
1612         dx_insert_block(frame, hash2 + continued, newblock);
1613         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1614         if (err)
1615                 goto journal_error;
1616         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1617         if (err)
1618                 goto journal_error;
1619         brelse(bh2);
1620         dxtrace(dx_show_index("frame", frame->entries));
1621         return de;
1622
1623 journal_error:
1624         brelse(*bh);
1625         brelse(bh2);
1626         *bh = NULL;
1627         ext4_std_error(dir->i_sb, err);
1628 errout:
1629         *error = err;
1630         return NULL;
1631 }
1632
1633 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1634                       struct buffer_head *bh,
1635                       void *buf, int buf_size,
1636                       const char *name, int namelen,
1637                       struct ext4_dir_entry_2 **dest_de)
1638 {
1639         struct ext4_dir_entry_2 *de;
1640         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1641         int nlen, rlen;
1642         unsigned int offset = 0;
1643         char *top;
1644
1645         de = (struct ext4_dir_entry_2 *)buf;
1646         top = buf + buf_size - reclen;
1647         while ((char *) de <= top) {
1648                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1649                                          buf, buf_size, offset))
1650                         return -EIO;
1651                 if (ext4_match(namelen, name, de))
1652                         return -EEXIST;
1653                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1654                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1655                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1656                         break;
1657                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1658                 offset += rlen;
1659         }
1660         if ((char *) de > top)
1661                 return -ENOSPC;
1662
1663         *dest_de = de;
1664         return 0;
1665 }
1666
1667 void ext4_insert_dentry(struct inode *inode,
1668                         struct ext4_dir_entry_2 *de,
1669                         int buf_size,
1670                         const char *name, int namelen)
1671 {
1672
1673         int nlen, rlen;
1674
1675         nlen = EXT4_DIR_REC_LEN(de->name_len);
1676         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1677         if (de->inode) {
1678                 struct ext4_dir_entry_2 *de1 =
1679                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1680                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1681                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1682                 de = de1;
1683         }
1684         de->file_type = EXT4_FT_UNKNOWN;
1685         de->inode = cpu_to_le32(inode->i_ino);
1686         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1687         de->name_len = namelen;
1688         memcpy(de->name, name, namelen);
1689 }
1690 /*
1691  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1692  * it points to a directory entry which is guaranteed to be large
1693  * enough for new directory entry.  If de is NULL, then
1694  * add_dirent_to_buf will attempt search the directory block for
1695  * space.  It will return -ENOSPC if no space is available, and -EIO
1696  * and -EEXIST if directory entry already exists.
1697  */
1698 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1699                              struct inode *inode, struct ext4_dir_entry_2 *de,
1700                              struct buffer_head *bh)
1701 {
1702         struct inode    *dir = dentry->d_parent->d_inode;
1703         const char      *name = dentry->d_name.name;
1704         int             namelen = dentry->d_name.len;
1705         unsigned int    blocksize = dir->i_sb->s_blocksize;
1706         int             csum_size = 0;
1707         int             err;
1708
1709         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1710                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1711                 csum_size = sizeof(struct ext4_dir_entry_tail);
1712
1713         if (!de) {
1714                 err = ext4_find_dest_de(dir, inode,
1715                                         bh, bh->b_data, blocksize - csum_size,
1716                                         name, namelen, &de);
1717                 if (err)
1718                         return err;
1719         }
1720         BUFFER_TRACE(bh, "get_write_access");
1721         err = ext4_journal_get_write_access(handle, bh);
1722         if (err) {
1723                 ext4_std_error(dir->i_sb, err);
1724                 return err;
1725         }
1726
1727         /* By now the buffer is marked for journaling */
1728         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1729
1730         /*
1731          * XXX shouldn't update any times until successful
1732          * completion of syscall, but too many callers depend
1733          * on this.
1734          *
1735          * XXX similarly, too many callers depend on
1736          * ext4_new_inode() setting the times, but error
1737          * recovery deletes the inode, so the worst that can
1738          * happen is that the times are slightly out of date
1739          * and/or different from the directory change time.
1740          */
1741         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1742         ext4_update_dx_flag(dir);
1743         dir->i_version++;
1744         ext4_mark_inode_dirty(handle, dir);
1745         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1746         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1747         if (err)
1748                 ext4_std_error(dir->i_sb, err);
1749         return 0;
1750 }
1751
1752 /*
1753  * This converts a one block unindexed directory to a 3 block indexed
1754  * directory, and adds the dentry to the indexed directory.
1755  */
1756 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1757                             struct inode *inode, struct buffer_head *bh)
1758 {
1759         struct inode    *dir = dentry->d_parent->d_inode;
1760         const char      *name = dentry->d_name.name;
1761         int             namelen = dentry->d_name.len;
1762         struct buffer_head *bh2;
1763         struct dx_root  *root;
1764         struct dx_frame frames[2], *frame;
1765         struct dx_entry *entries;
1766         struct ext4_dir_entry_2 *de, *de2;
1767         struct ext4_dir_entry_tail *t;
1768         char            *data1, *top;
1769         unsigned        len;
1770         int             retval;
1771         unsigned        blocksize;
1772         struct dx_hash_info hinfo;
1773         ext4_lblk_t  block;
1774         struct fake_dirent *fde;
1775         int             csum_size = 0;
1776
1777         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1778                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1779                 csum_size = sizeof(struct ext4_dir_entry_tail);
1780
1781         blocksize =  dir->i_sb->s_blocksize;
1782         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1783         retval = ext4_journal_get_write_access(handle, bh);
1784         if (retval) {
1785                 ext4_std_error(dir->i_sb, retval);
1786                 brelse(bh);
1787                 return retval;
1788         }
1789         root = (struct dx_root *) bh->b_data;
1790
1791         /* The 0th block becomes the root, move the dirents out */
1792         fde = &root->dotdot;
1793         de = (struct ext4_dir_entry_2 *)((char *)fde +
1794                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1795         if ((char *) de >= (((char *) root) + blocksize)) {
1796                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1797                 brelse(bh);
1798                 return -EIO;
1799         }
1800         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1801
1802         /* Allocate new block for the 0th block's dirents */
1803         bh2 = ext4_append(handle, dir, &block, &retval);
1804         if (!(bh2)) {
1805                 brelse(bh);
1806                 return retval;
1807         }
1808         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1809         data1 = bh2->b_data;
1810
1811         memcpy (data1, de, len);
1812         de = (struct ext4_dir_entry_2 *) data1;
1813         top = data1 + len;
1814         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1815                 de = de2;
1816         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1817                                            (char *) de,
1818                                            blocksize);
1819
1820         if (csum_size) {
1821                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1822                 initialize_dirent_tail(t, blocksize);
1823         }
1824
1825         /* Initialize the root; the dot dirents already exist */
1826         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1827         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1828                                            blocksize);
1829         memset (&root->info, 0, sizeof(root->info));
1830         root->info.info_length = sizeof(root->info);
1831         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1832         entries = root->entries;
1833         dx_set_block(entries, 1);
1834         dx_set_count(entries, 1);
1835         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1836
1837         /* Initialize as for dx_probe */
1838         hinfo.hash_version = root->info.hash_version;
1839         if (hinfo.hash_version <= DX_HASH_TEA)
1840                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1841         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1842         ext4fs_dirhash(name, namelen, &hinfo);
1843         frame = frames;
1844         frame->entries = entries;
1845         frame->at = entries;
1846         frame->bh = bh;
1847         bh = bh2;
1848
1849         ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1850         ext4_handle_dirty_dirent_node(handle, dir, bh);
1851
1852         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1853         if (!de) {
1854                 /*
1855                  * Even if the block split failed, we have to properly write
1856                  * out all the changes we did so far. Otherwise we can end up
1857                  * with corrupted filesystem.
1858                  */
1859                 ext4_mark_inode_dirty(handle, dir);
1860                 dx_release(frames);
1861                 return retval;
1862         }
1863         dx_release(frames);
1864
1865         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1866         brelse(bh);
1867         return retval;
1868 }
1869
1870 /*
1871  *      ext4_add_entry()
1872  *
1873  * adds a file entry to the specified directory, using the same
1874  * semantics as ext4_find_entry(). It returns NULL if it failed.
1875  *
1876  * NOTE!! The inode part of 'de' is left at 0 - which means you
1877  * may not sleep between calling this and putting something into
1878  * the entry, as someone else might have used it while you slept.
1879  */
1880 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1881                           struct inode *inode)
1882 {
1883         struct inode *dir = dentry->d_parent->d_inode;
1884         struct buffer_head *bh;
1885         struct ext4_dir_entry_2 *de;
1886         struct ext4_dir_entry_tail *t;
1887         struct super_block *sb;
1888         int     retval;
1889         int     dx_fallback=0;
1890         unsigned blocksize;
1891         ext4_lblk_t block, blocks;
1892         int     csum_size = 0;
1893
1894         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1895                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1896                 csum_size = sizeof(struct ext4_dir_entry_tail);
1897
1898         sb = dir->i_sb;
1899         blocksize = sb->s_blocksize;
1900         if (!dentry->d_name.len)
1901                 return -EINVAL;
1902
1903         if (ext4_has_inline_data(dir)) {
1904                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1905                 if (retval < 0)
1906                         return retval;
1907                 if (retval == 1) {
1908                         retval = 0;
1909                         return retval;
1910                 }
1911         }
1912
1913         if (is_dx(dir)) {
1914                 retval = ext4_dx_add_entry(handle, dentry, inode);
1915                 if (!retval || (retval != ERR_BAD_DX_DIR))
1916                         return retval;
1917                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1918                 dx_fallback++;
1919                 ext4_mark_inode_dirty(handle, dir);
1920         }
1921         blocks = dir->i_size >> sb->s_blocksize_bits;
1922         for (block = 0; block < blocks; block++) {
1923                 if (!(bh = ext4_bread(handle, dir, block, 0, &retval))) {
1924                         if (!retval) {
1925                                 retval = -EIO;
1926                                 ext4_error(inode->i_sb,
1927                                            "Directory hole detected on inode %lu\n",
1928                                            inode->i_ino);
1929                         }
1930                         return retval;
1931                 }
1932                 if (!buffer_verified(bh) &&
1933                     !ext4_dirent_csum_verify(dir,
1934                                 (struct ext4_dir_entry *)bh->b_data)) {
1935                         brelse(bh);
1936                         return -EIO;
1937                 }
1938                 set_buffer_verified(bh);
1939                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1940                 if (retval != -ENOSPC) {
1941                         brelse(bh);
1942                         return retval;
1943                 }
1944
1945                 if (blocks == 1 && !dx_fallback &&
1946                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1947                         return make_indexed_dir(handle, dentry, inode, bh);
1948                 brelse(bh);
1949         }
1950         bh = ext4_append(handle, dir, &block, &retval);
1951         if (!bh)
1952                 return retval;
1953         de = (struct ext4_dir_entry_2 *) bh->b_data;
1954         de->inode = 0;
1955         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1956
1957         if (csum_size) {
1958                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1959                 initialize_dirent_tail(t, blocksize);
1960         }
1961
1962         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1963         brelse(bh);
1964         if (retval == 0)
1965                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1966         return retval;
1967 }
1968
1969 /*
1970  * Returns 0 for success, or a negative error value
1971  */
1972 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1973                              struct inode *inode)
1974 {
1975         struct dx_frame frames[2], *frame;
1976         struct dx_entry *entries, *at;
1977         struct dx_hash_info hinfo;
1978         struct buffer_head *bh;
1979         struct inode *dir = dentry->d_parent->d_inode;
1980         struct super_block *sb = dir->i_sb;
1981         struct ext4_dir_entry_2 *de;
1982         int err;
1983
1984         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1985         if (!frame)
1986                 return err;
1987         entries = frame->entries;
1988         at = frame->at;
1989
1990         if (!(bh = ext4_bread(handle, dir, dx_get_block(frame->at), 0, &err))) {
1991                 if (!err) {
1992                         err = -EIO;
1993                         ext4_error(dir->i_sb,
1994                                    "Directory hole detected on inode %lu\n",
1995                                    dir->i_ino);
1996                 }
1997                 goto cleanup;
1998         }
1999
2000         if (!buffer_verified(bh) &&
2001             !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
2002                 goto journal_error;
2003         set_buffer_verified(bh);
2004
2005         BUFFER_TRACE(bh, "get_write_access");
2006         err = ext4_journal_get_write_access(handle, bh);
2007         if (err)
2008                 goto journal_error;
2009
2010         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2011         if (err != -ENOSPC)
2012                 goto cleanup;
2013
2014         /* Block full, should compress but for now just split */
2015         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2016                        dx_get_count(entries), dx_get_limit(entries)));
2017         /* Need to split index? */
2018         if (dx_get_count(entries) == dx_get_limit(entries)) {
2019                 ext4_lblk_t newblock;
2020                 unsigned icount = dx_get_count(entries);
2021                 int levels = frame - frames;
2022                 struct dx_entry *entries2;
2023                 struct dx_node *node2;
2024                 struct buffer_head *bh2;
2025
2026                 if (levels && (dx_get_count(frames->entries) ==
2027                                dx_get_limit(frames->entries))) {
2028                         ext4_warning(sb, "Directory index full!");
2029                         err = -ENOSPC;
2030                         goto cleanup;
2031                 }
2032                 bh2 = ext4_append (handle, dir, &newblock, &err);
2033                 if (!(bh2))
2034                         goto cleanup;
2035                 node2 = (struct dx_node *)(bh2->b_data);
2036                 entries2 = node2->entries;
2037                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2038                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2039                                                            sb->s_blocksize);
2040                 BUFFER_TRACE(frame->bh, "get_write_access");
2041                 err = ext4_journal_get_write_access(handle, frame->bh);
2042                 if (err)
2043                         goto journal_error;
2044                 if (levels) {
2045                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2046                         unsigned hash2 = dx_get_hash(entries + icount1);
2047                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2048                                        icount1, icount2));
2049
2050                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2051                         err = ext4_journal_get_write_access(handle,
2052                                                              frames[0].bh);
2053                         if (err)
2054                                 goto journal_error;
2055
2056                         memcpy((char *) entries2, (char *) (entries + icount1),
2057                                icount2 * sizeof(struct dx_entry));
2058                         dx_set_count(entries, icount1);
2059                         dx_set_count(entries2, icount2);
2060                         dx_set_limit(entries2, dx_node_limit(dir));
2061
2062                         /* Which index block gets the new entry? */
2063                         if (at - entries >= icount1) {
2064                                 frame->at = at = at - entries - icount1 + entries2;
2065                                 frame->entries = entries = entries2;
2066                                 swap(frame->bh, bh2);
2067                         }
2068                         dx_insert_block(frames + 0, hash2, newblock);
2069                         dxtrace(dx_show_index("node", frames[1].entries));
2070                         dxtrace(dx_show_index("node",
2071                                ((struct dx_node *) bh2->b_data)->entries));
2072                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2073                         if (err)
2074                                 goto journal_error;
2075                         brelse (bh2);
2076                 } else {
2077                         dxtrace(printk(KERN_DEBUG
2078                                        "Creating second level index...\n"));
2079                         memcpy((char *) entries2, (char *) entries,
2080                                icount * sizeof(struct dx_entry));
2081                         dx_set_limit(entries2, dx_node_limit(dir));
2082
2083                         /* Set up root */
2084                         dx_set_count(entries, 1);
2085                         dx_set_block(entries + 0, newblock);
2086                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2087
2088                         /* Add new access path frame */
2089                         frame = frames + 1;
2090                         frame->at = at = at - entries + entries2;
2091                         frame->entries = entries = entries2;
2092                         frame->bh = bh2;
2093                         err = ext4_journal_get_write_access(handle,
2094                                                              frame->bh);
2095                         if (err)
2096                                 goto journal_error;
2097                 }
2098                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2099                 if (err) {
2100                         ext4_std_error(inode->i_sb, err);
2101                         goto cleanup;
2102                 }
2103         }
2104         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2105         if (!de)
2106                 goto cleanup;
2107         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2108         goto cleanup;
2109
2110 journal_error:
2111         ext4_std_error(dir->i_sb, err);
2112 cleanup:
2113         brelse(bh);
2114         dx_release(frames);
2115         return err;
2116 }
2117
2118 /*
2119  * ext4_generic_delete_entry deletes a directory entry by merging it
2120  * with the previous entry
2121  */
2122 int ext4_generic_delete_entry(handle_t *handle,
2123                               struct inode *dir,
2124                               struct ext4_dir_entry_2 *de_del,
2125                               struct buffer_head *bh,
2126                               void *entry_buf,
2127                               int buf_size,
2128                               int csum_size)
2129 {
2130         struct ext4_dir_entry_2 *de, *pde;
2131         unsigned int blocksize = dir->i_sb->s_blocksize;
2132         int i;
2133
2134         i = 0;
2135         pde = NULL;
2136         de = (struct ext4_dir_entry_2 *)entry_buf;
2137         while (i < buf_size - csum_size) {
2138                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2139                                          bh->b_data, bh->b_size, i))
2140                         return -EIO;
2141                 if (de == de_del)  {
2142                         if (pde)
2143                                 pde->rec_len = ext4_rec_len_to_disk(
2144                                         ext4_rec_len_from_disk(pde->rec_len,
2145                                                                blocksize) +
2146                                         ext4_rec_len_from_disk(de->rec_len,
2147                                                                blocksize),
2148                                         blocksize);
2149                         else
2150                                 de->inode = 0;
2151                         dir->i_version++;
2152                         return 0;
2153                 }
2154                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2155                 pde = de;
2156                 de = ext4_next_entry(de, blocksize);
2157         }
2158         return -ENOENT;
2159 }
2160
2161 static int ext4_delete_entry(handle_t *handle,
2162                              struct inode *dir,
2163                              struct ext4_dir_entry_2 *de_del,
2164                              struct buffer_head *bh)
2165 {
2166         int err, csum_size = 0;
2167
2168         if (ext4_has_inline_data(dir)) {
2169                 int has_inline_data = 1;
2170                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2171                                                &has_inline_data);
2172                 if (has_inline_data)
2173                         return err;
2174         }
2175
2176         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2177                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2178                 csum_size = sizeof(struct ext4_dir_entry_tail);
2179
2180         BUFFER_TRACE(bh, "get_write_access");
2181         err = ext4_journal_get_write_access(handle, bh);
2182         if (unlikely(err))
2183                 goto out;
2184
2185         err = ext4_generic_delete_entry(handle, dir, de_del,
2186                                         bh, bh->b_data,
2187                                         dir->i_sb->s_blocksize, csum_size);
2188         if (err)
2189                 goto out;
2190
2191         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2192         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2193         if (unlikely(err))
2194                 goto out;
2195
2196         return 0;
2197 out:
2198         if (err != -ENOENT)
2199                 ext4_std_error(dir->i_sb, err);
2200         return err;
2201 }
2202
2203 /*
2204  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2205  * since this indicates that nlinks count was previously 1.
2206  */
2207 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2208 {
2209         inc_nlink(inode);
2210         if (is_dx(inode) && inode->i_nlink > 1) {
2211                 /* limit is 16-bit i_links_count */
2212                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2213                         set_nlink(inode, 1);
2214                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2215                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2216                 }
2217         }
2218 }
2219
2220 /*
2221  * If a directory had nlink == 1, then we should let it be 1. This indicates
2222  * directory has >EXT4_LINK_MAX subdirs.
2223  */
2224 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2225 {
2226         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2227                 drop_nlink(inode);
2228 }
2229
2230
2231 static int ext4_add_nondir(handle_t *handle,
2232                 struct dentry *dentry, struct inode *inode)
2233 {
2234         int err = ext4_add_entry(handle, dentry, inode);
2235         if (!err) {
2236                 ext4_mark_inode_dirty(handle, inode);
2237                 unlock_new_inode(inode);
2238                 d_instantiate(dentry, inode);
2239                 return 0;
2240         }
2241         drop_nlink(inode);
2242         unlock_new_inode(inode);
2243         iput(inode);
2244         return err;
2245 }
2246
2247 /*
2248  * By the time this is called, we already have created
2249  * the directory cache entry for the new file, but it
2250  * is so far negative - it has no inode.
2251  *
2252  * If the create succeeds, we fill in the inode information
2253  * with d_instantiate().
2254  */
2255 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2256                        bool excl)
2257 {
2258         handle_t *handle;
2259         struct inode *inode;
2260         int err, retries = 0;
2261
2262         dquot_initialize(dir);
2263
2264 retry:
2265         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2266                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2267                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2268                  EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)));
2269         if (IS_ERR(handle))
2270                 return PTR_ERR(handle);
2271
2272         if (IS_DIRSYNC(dir))
2273                 ext4_handle_sync(handle);
2274
2275         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2276         err = PTR_ERR(inode);
2277         if (!IS_ERR(inode)) {
2278                 inode->i_op = &ext4_file_inode_operations;
2279                 inode->i_fop = &ext4_file_operations;
2280                 ext4_set_aops(inode);
2281                 err = ext4_add_nondir(handle, dentry, inode);
2282         }
2283         ext4_journal_stop(handle);
2284         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2285                 goto retry;
2286         return err;
2287 }
2288
2289 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2290                       umode_t mode, dev_t rdev)
2291 {
2292         handle_t *handle;
2293         struct inode *inode;
2294         int err, retries = 0;
2295
2296         if (!new_valid_dev(rdev))
2297                 return -EINVAL;
2298
2299         dquot_initialize(dir);
2300
2301 retry:
2302         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2303                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2304                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2305                  EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)));
2306         if (IS_ERR(handle))
2307                 return PTR_ERR(handle);
2308
2309         if (IS_DIRSYNC(dir))
2310                 ext4_handle_sync(handle);
2311
2312         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2313         err = PTR_ERR(inode);
2314         if (!IS_ERR(inode)) {
2315                 init_special_inode(inode, inode->i_mode, rdev);
2316                 inode->i_op = &ext4_special_inode_operations;
2317                 err = ext4_add_nondir(handle, dentry, inode);
2318         }
2319         ext4_journal_stop(handle);
2320         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2321                 goto retry;
2322         return err;
2323 }
2324
2325 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2326                           struct ext4_dir_entry_2 *de,
2327                           int blocksize, int csum_size,
2328                           unsigned int parent_ino, int dotdot_real_len)
2329 {
2330         de->inode = cpu_to_le32(inode->i_ino);
2331         de->name_len = 1;
2332         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2333                                            blocksize);
2334         strcpy(de->name, ".");
2335         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2336
2337         de = ext4_next_entry(de, blocksize);
2338         de->inode = cpu_to_le32(parent_ino);
2339         de->name_len = 2;
2340         if (!dotdot_real_len)
2341                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2342                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2343                                         blocksize);
2344         else
2345                 de->rec_len = ext4_rec_len_to_disk(
2346                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2347         strcpy(de->name, "..");
2348         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2349
2350         return ext4_next_entry(de, blocksize);
2351 }
2352
2353 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2354                              struct inode *inode)
2355 {
2356         struct buffer_head *dir_block = NULL;
2357         struct ext4_dir_entry_2 *de;
2358         struct ext4_dir_entry_tail *t;
2359         unsigned int blocksize = dir->i_sb->s_blocksize;
2360         int csum_size = 0;
2361         int err;
2362
2363         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2364                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2365                 csum_size = sizeof(struct ext4_dir_entry_tail);
2366
2367         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2368                 err = ext4_try_create_inline_dir(handle, dir, inode);
2369                 if (err < 0 && err != -ENOSPC)
2370                         goto out;
2371                 if (!err)
2372                         goto out;
2373         }
2374
2375         inode->i_size = EXT4_I(inode)->i_disksize = blocksize;
2376         if (!(dir_block = ext4_bread(handle, inode, 0, 1, &err))) {
2377                 if (!err) {
2378                         err = -EIO;
2379                         ext4_error(inode->i_sb,
2380                                    "Directory hole detected on inode %lu\n",
2381                                    inode->i_ino);
2382                 }
2383                 goto out;
2384         }
2385         BUFFER_TRACE(dir_block, "get_write_access");
2386         err = ext4_journal_get_write_access(handle, dir_block);
2387         if (err)
2388                 goto out;
2389         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2390         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2391         set_nlink(inode, 2);
2392         if (csum_size) {
2393                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2394                 initialize_dirent_tail(t, blocksize);
2395         }
2396
2397         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2398         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2399         if (err)
2400                 goto out;
2401         set_buffer_verified(dir_block);
2402 out:
2403         brelse(dir_block);
2404         return err;
2405 }
2406
2407 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2408 {
2409         handle_t *handle;
2410         struct inode *inode;
2411         int err, retries = 0;
2412
2413         if (EXT4_DIR_LINK_MAX(dir))
2414                 return -EMLINK;
2415
2416         dquot_initialize(dir);
2417
2418 retry:
2419         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2420                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2421                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2422                  EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)));
2423         if (IS_ERR(handle))
2424                 return PTR_ERR(handle);
2425
2426         if (IS_DIRSYNC(dir))
2427                 ext4_handle_sync(handle);
2428
2429         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
2430                                &dentry->d_name, 0, NULL);
2431         err = PTR_ERR(inode);
2432         if (IS_ERR(inode))
2433                 goto out_stop;
2434
2435         inode->i_op = &ext4_dir_inode_operations;
2436         inode->i_fop = &ext4_dir_operations;
2437         err = ext4_init_new_dir(handle, dir, inode);
2438         if (err)
2439                 goto out_clear_inode;
2440         err = ext4_mark_inode_dirty(handle, inode);
2441         if (!err)
2442                 err = ext4_add_entry(handle, dentry, inode);
2443         if (err) {
2444 out_clear_inode:
2445                 clear_nlink(inode);
2446                 unlock_new_inode(inode);
2447                 ext4_mark_inode_dirty(handle, inode);
2448                 iput(inode);
2449                 goto out_stop;
2450         }
2451         ext4_inc_count(handle, dir);
2452         ext4_update_dx_flag(dir);
2453         err = ext4_mark_inode_dirty(handle, dir);
2454         if (err)
2455                 goto out_clear_inode;
2456         unlock_new_inode(inode);
2457         d_instantiate(dentry, inode);
2458 out_stop:
2459         ext4_journal_stop(handle);
2460         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2461                 goto retry;
2462         return err;
2463 }
2464
2465 /*
2466  * routine to check that the specified directory is empty (for rmdir)
2467  */
2468 static int empty_dir(struct inode *inode)
2469 {
2470         unsigned int offset;
2471         struct buffer_head *bh;
2472         struct ext4_dir_entry_2 *de, *de1;
2473         struct super_block *sb;
2474         int err = 0;
2475
2476         if (ext4_has_inline_data(inode)) {
2477                 int has_inline_data = 1;
2478
2479                 err = empty_inline_dir(inode, &has_inline_data);
2480                 if (has_inline_data)
2481                         return err;
2482         }
2483
2484         sb = inode->i_sb;
2485         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2486             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
2487                 if (err)
2488                         EXT4_ERROR_INODE(inode,
2489                                 "error %d reading directory lblock 0", err);
2490                 else
2491                         ext4_warning(inode->i_sb,
2492                                      "bad directory (dir #%lu) - no data block",
2493                                      inode->i_ino);
2494                 return 1;
2495         }
2496         if (!buffer_verified(bh) &&
2497             !ext4_dirent_csum_verify(inode,
2498                         (struct ext4_dir_entry *)bh->b_data)) {
2499                 EXT4_ERROR_INODE(inode, "checksum error reading directory "
2500                                  "lblock 0");
2501                 brelse(bh);
2502                 return -EIO;
2503         }
2504         set_buffer_verified(bh);
2505         de = (struct ext4_dir_entry_2 *) bh->b_data;
2506         de1 = ext4_next_entry(de, sb->s_blocksize);
2507         if (le32_to_cpu(de->inode) != inode->i_ino ||
2508                         !le32_to_cpu(de1->inode) ||
2509                         strcmp(".", de->name) ||
2510                         strcmp("..", de1->name)) {
2511                 ext4_warning(inode->i_sb,
2512                              "bad directory (dir #%lu) - no `.' or `..'",
2513                              inode->i_ino);
2514                 brelse(bh);
2515                 return 1;
2516         }
2517         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2518                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2519         de = ext4_next_entry(de1, sb->s_blocksize);
2520         while (offset < inode->i_size) {
2521                 if (!bh ||
2522                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2523                         unsigned int lblock;
2524                         err = 0;
2525                         brelse(bh);
2526                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2527                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
2528                         if (!bh) {
2529                                 if (err)
2530                                         EXT4_ERROR_INODE(inode,
2531                                                 "error %d reading directory "
2532                                                 "lblock %u", err, lblock);
2533                                 else
2534                                         ext4_warning(inode->i_sb,
2535                                                 "bad directory (dir #%lu) - no data block",
2536                                                 inode->i_ino);
2537
2538                                 offset += sb->s_blocksize;
2539                                 continue;
2540                         }
2541                         if (!buffer_verified(bh) &&
2542                             !ext4_dirent_csum_verify(inode,
2543                                         (struct ext4_dir_entry *)bh->b_data)) {
2544                                 EXT4_ERROR_INODE(inode, "checksum error "
2545                                                  "reading directory lblock 0");
2546                                 brelse(bh);
2547                                 return -EIO;
2548                         }
2549                         set_buffer_verified(bh);
2550                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2551                 }
2552                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2553                                          bh->b_data, bh->b_size, offset)) {
2554                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2555                                                          sb->s_blocksize);
2556                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2557                         continue;
2558                 }
2559                 if (le32_to_cpu(de->inode)) {
2560                         brelse(bh);
2561                         return 0;
2562                 }
2563                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2564                 de = ext4_next_entry(de, sb->s_blocksize);
2565         }
2566         brelse(bh);
2567         return 1;
2568 }
2569
2570 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2571  * such inodes, starting at the superblock, in case we crash before the
2572  * file is closed/deleted, or in case the inode truncate spans multiple
2573  * transactions and the last transaction is not recovered after a crash.
2574  *
2575  * At filesystem recovery time, we walk this list deleting unlinked
2576  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2577  */
2578 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2579 {
2580         struct super_block *sb = inode->i_sb;
2581         struct ext4_iloc iloc;
2582         int err = 0, rc;
2583
2584         if (!EXT4_SB(sb)->s_journal)
2585                 return 0;
2586
2587         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2588         if (!list_empty(&EXT4_I(inode)->i_orphan))
2589                 goto out_unlock;
2590
2591         /*
2592          * Orphan handling is only valid for files with data blocks
2593          * being truncated, or files being unlinked. Note that we either
2594          * hold i_mutex, or the inode can not be referenced from outside,
2595          * so i_nlink should not be bumped due to race
2596          */
2597         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2598                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2599
2600         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2601         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2602         if (err)
2603                 goto out_unlock;
2604
2605         err = ext4_reserve_inode_write(handle, inode, &iloc);
2606         if (err)
2607                 goto out_unlock;
2608         /*
2609          * Due to previous errors inode may be already a part of on-disk
2610          * orphan list. If so skip on-disk list modification.
2611          */
2612         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2613                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2614                         goto mem_insert;
2615
2616         /* Insert this inode at the head of the on-disk orphan list... */
2617         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2618         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2619         err = ext4_handle_dirty_super(handle, sb);
2620         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2621         if (!err)
2622                 err = rc;
2623
2624         /* Only add to the head of the in-memory list if all the
2625          * previous operations succeeded.  If the orphan_add is going to
2626          * fail (possibly taking the journal offline), we can't risk
2627          * leaving the inode on the orphan list: stray orphan-list
2628          * entries can cause panics at unmount time.
2629          *
2630          * This is safe: on error we're going to ignore the orphan list
2631          * anyway on the next recovery. */
2632 mem_insert:
2633         if (!err)
2634                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2635
2636         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2637         jbd_debug(4, "orphan inode %lu will point to %d\n",
2638                         inode->i_ino, NEXT_ORPHAN(inode));
2639 out_unlock:
2640         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2641         ext4_std_error(inode->i_sb, err);
2642         return err;
2643 }
2644
2645 /*
2646  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2647  * of such inodes stored on disk, because it is finally being cleaned up.
2648  */
2649 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2650 {
2651         struct list_head *prev;
2652         struct ext4_inode_info *ei = EXT4_I(inode);
2653         struct ext4_sb_info *sbi;
2654         __u32 ino_next;
2655         struct ext4_iloc iloc;
2656         int err = 0;
2657
2658         if ((!EXT4_SB(inode->i_sb)->s_journal) &&
2659             !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2660                 return 0;
2661
2662         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2663         if (list_empty(&ei->i_orphan))
2664                 goto out;
2665
2666         ino_next = NEXT_ORPHAN(inode);
2667         prev = ei->i_orphan.prev;
2668         sbi = EXT4_SB(inode->i_sb);
2669
2670         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2671
2672         list_del_init(&ei->i_orphan);
2673
2674         /* If we're on an error path, we may not have a valid
2675          * transaction handle with which to update the orphan list on
2676          * disk, but we still need to remove the inode from the linked
2677          * list in memory. */
2678         if (!handle)
2679                 goto out;
2680
2681         err = ext4_reserve_inode_write(handle, inode, &iloc);
2682         if (err)
2683                 goto out_err;
2684
2685         if (prev == &sbi->s_orphan) {
2686                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2687                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2688                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2689                 if (err)
2690                         goto out_brelse;
2691                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2692                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2693         } else {
2694                 struct ext4_iloc iloc2;
2695                 struct inode *i_prev =
2696                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2697
2698                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2699                           i_prev->i_ino, ino_next);
2700                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2701                 if (err)
2702                         goto out_brelse;
2703                 NEXT_ORPHAN(i_prev) = ino_next;
2704                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2705         }
2706         if (err)
2707                 goto out_brelse;
2708         NEXT_ORPHAN(inode) = 0;
2709         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2710
2711 out_err:
2712         ext4_std_error(inode->i_sb, err);
2713 out:
2714         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2715         return err;
2716
2717 out_brelse:
2718         brelse(iloc.bh);
2719         goto out_err;
2720 }
2721
2722 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2723 {
2724         int retval;
2725         struct inode *inode;
2726         struct buffer_head *bh;
2727         struct ext4_dir_entry_2 *de;
2728         handle_t *handle;
2729
2730         /* Initialize quotas before so that eventual writes go in
2731          * separate transaction */
2732         dquot_initialize(dir);
2733         dquot_initialize(dentry->d_inode);
2734
2735         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2736                                     EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2737         if (IS_ERR(handle))
2738                 return PTR_ERR(handle);
2739
2740         retval = -ENOENT;
2741         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2742         if (!bh)
2743                 goto end_rmdir;
2744
2745         if (IS_DIRSYNC(dir))
2746                 ext4_handle_sync(handle);
2747
2748         inode = dentry->d_inode;
2749
2750         retval = -EIO;
2751         if (le32_to_cpu(de->inode) != inode->i_ino)
2752                 goto end_rmdir;
2753
2754         retval = -ENOTEMPTY;
2755         if (!empty_dir(inode))
2756                 goto end_rmdir;
2757
2758         retval = ext4_delete_entry(handle, dir, de, bh);
2759         if (retval)
2760                 goto end_rmdir;
2761         if (!EXT4_DIR_LINK_EMPTY(inode))
2762                 ext4_warning(inode->i_sb,
2763                              "empty directory has too many links (%d)",
2764                              inode->i_nlink);
2765         inode->i_version++;
2766         clear_nlink(inode);
2767         /* There's no need to set i_disksize: the fact that i_nlink is
2768          * zero will ensure that the right thing happens during any
2769          * recovery. */
2770         inode->i_size = 0;
2771         ext4_orphan_add(handle, inode);
2772         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2773         ext4_mark_inode_dirty(handle, inode);
2774         ext4_dec_count(handle, dir);
2775         ext4_update_dx_flag(dir);
2776         ext4_mark_inode_dirty(handle, dir);
2777
2778 end_rmdir:
2779         ext4_journal_stop(handle);
2780         brelse(bh);
2781         return retval;
2782 }
2783
2784 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2785 {
2786         int retval;
2787         struct inode *inode;
2788         struct buffer_head *bh;
2789         struct ext4_dir_entry_2 *de;
2790         handle_t *handle = NULL;
2791
2792         trace_ext4_unlink_enter(dir, dentry);
2793         /* Initialize quotas before so that eventual writes go
2794          * in separate transaction */
2795         dquot_initialize(dir);
2796         dquot_initialize(dentry->d_inode);
2797
2798         retval = -ENOENT;
2799         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2800         if (!bh)
2801                 goto end_unlink;
2802
2803         inode = dentry->d_inode;
2804
2805         retval = -EIO;
2806         if (le32_to_cpu(de->inode) != inode->i_ino)
2807                 goto end_unlink;
2808
2809         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2810                                     EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2811         if (IS_ERR(handle)) {
2812                 retval = PTR_ERR(handle);
2813                 handle = NULL;
2814                 goto end_unlink;
2815         }
2816
2817         if (IS_DIRSYNC(dir))
2818                 ext4_handle_sync(handle);
2819
2820         if (!inode->i_nlink) {
2821                 ext4_warning(inode->i_sb,
2822                              "Deleting nonexistent file (%lu), %d",
2823                              inode->i_ino, inode->i_nlink);
2824                 set_nlink(inode, 1);
2825         }
2826         retval = ext4_delete_entry(handle, dir, de, bh);
2827         if (retval)
2828                 goto end_unlink;
2829         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2830         ext4_update_dx_flag(dir);
2831         ext4_mark_inode_dirty(handle, dir);
2832         drop_nlink(inode);
2833         if (!inode->i_nlink)
2834                 ext4_orphan_add(handle, inode);
2835         inode->i_ctime = ext4_current_time(inode);
2836         ext4_mark_inode_dirty(handle, inode);
2837         retval = 0;
2838
2839 end_unlink:
2840         brelse(bh);
2841         if (handle)
2842                 ext4_journal_stop(handle);
2843         trace_ext4_unlink_exit(dentry, retval);
2844         return retval;
2845 }
2846
2847 static int ext4_symlink(struct inode *dir,
2848                         struct dentry *dentry, const char *symname)
2849 {
2850         handle_t *handle;
2851         struct inode *inode;
2852         int l, err, retries = 0;
2853         int credits;
2854
2855         l = strlen(symname)+1;
2856         if (l > dir->i_sb->s_blocksize)
2857                 return -ENAMETOOLONG;
2858
2859         dquot_initialize(dir);
2860
2861         if (l > EXT4_N_BLOCKS * 4) {
2862                 /*
2863                  * For non-fast symlinks, we just allocate inode and put it on
2864                  * orphan list in the first transaction => we need bitmap,
2865                  * group descriptor, sb, inode block, quota blocks, and
2866                  * possibly selinux xattr blocks.
2867                  */
2868                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2869                           EXT4_XATTR_TRANS_BLOCKS;
2870         } else {
2871                 /*
2872                  * Fast symlink. We have to add entry to directory
2873                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2874                  * allocate new inode (bitmap, group descriptor, inode block,
2875                  * quota blocks, sb is already counted in previous macros).
2876                  */
2877                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2878                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2879                           EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2880         }
2881 retry:
2882         handle = ext4_journal_start(dir, EXT4_HT_DIR, credits);
2883         if (IS_ERR(handle))
2884                 return PTR_ERR(handle);
2885
2886         if (IS_DIRSYNC(dir))
2887                 ext4_handle_sync(handle);
2888
2889         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2890                                &dentry->d_name, 0, NULL);
2891         err = PTR_ERR(inode);
2892         if (IS_ERR(inode))
2893                 goto out_stop;
2894
2895         if (l > EXT4_N_BLOCKS * 4) {
2896                 inode->i_op = &ext4_symlink_inode_operations;
2897                 ext4_set_aops(inode);
2898                 /*
2899                  * We cannot call page_symlink() with transaction started
2900                  * because it calls into ext4_write_begin() which can wait
2901                  * for transaction commit if we are running out of space
2902                  * and thus we deadlock. So we have to stop transaction now
2903                  * and restart it when symlink contents is written.
2904                  *
2905                  * To keep fs consistent in case of crash, we have to put inode
2906                  * to orphan list in the mean time.
2907                  */
2908                 drop_nlink(inode);
2909                 err = ext4_orphan_add(handle, inode);
2910                 ext4_journal_stop(handle);
2911                 if (err)
2912                         goto err_drop_inode;
2913                 err = __page_symlink(inode, symname, l, 1);
2914                 if (err)
2915                         goto err_drop_inode;
2916                 /*
2917                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2918                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2919                  */
2920                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2921                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2922                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2923                 if (IS_ERR(handle)) {
2924                         err = PTR_ERR(handle);
2925                         goto err_drop_inode;
2926                 }
2927                 set_nlink(inode, 1);
2928                 err = ext4_orphan_del(handle, inode);
2929                 if (err) {
2930                         ext4_journal_stop(handle);
2931                         clear_nlink(inode);
2932                         goto err_drop_inode;
2933                 }
2934         } else {
2935                 /* clear the extent format for fast symlink */
2936                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2937                 inode->i_op = &ext4_fast_symlink_inode_operations;
2938                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2939                 inode->i_size = l-1;
2940         }
2941         EXT4_I(inode)->i_disksize = inode->i_size;
2942         err = ext4_add_nondir(handle, dentry, inode);
2943 out_stop:
2944         ext4_journal_stop(handle);
2945         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2946                 goto retry;
2947         return err;
2948 err_drop_inode:
2949         unlock_new_inode(inode);
2950         iput(inode);
2951         return err;
2952 }
2953
2954 static int ext4_link(struct dentry *old_dentry,
2955                      struct inode *dir, struct dentry *dentry)
2956 {
2957         handle_t *handle;
2958         struct inode *inode = old_dentry->d_inode;
2959         int err, retries = 0;
2960
2961         if (inode->i_nlink >= EXT4_LINK_MAX)
2962                 return -EMLINK;
2963
2964         dquot_initialize(dir);
2965
2966 retry:
2967         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2968                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2969                  EXT4_INDEX_EXTRA_TRANS_BLOCKS));
2970         if (IS_ERR(handle))
2971                 return PTR_ERR(handle);
2972
2973         if (IS_DIRSYNC(dir))
2974                 ext4_handle_sync(handle);
2975
2976         inode->i_ctime = ext4_current_time(inode);
2977         ext4_inc_count(handle, inode);
2978         ihold(inode);
2979
2980         err = ext4_add_entry(handle, dentry, inode);
2981         if (!err) {
2982                 ext4_mark_inode_dirty(handle, inode);
2983                 d_instantiate(dentry, inode);
2984         } else {
2985                 drop_nlink(inode);
2986                 iput(inode);
2987         }
2988         ext4_journal_stop(handle);
2989         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2990                 goto retry;
2991         return err;
2992 }
2993
2994
2995 /*
2996  * Try to find buffer head where contains the parent block.
2997  * It should be the inode block if it is inlined or the 1st block
2998  * if it is a normal dir.
2999  */
3000 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3001                                         struct inode *inode,
3002                                         int *retval,
3003                                         struct ext4_dir_entry_2 **parent_de,
3004                                         int *inlined)
3005 {
3006         struct buffer_head *bh;
3007
3008         if (!ext4_has_inline_data(inode)) {
3009                 if (!(bh = ext4_bread(handle, inode, 0, 0, retval))) {
3010                         if (!*retval) {
3011                                 *retval = -EIO;
3012                                 ext4_error(inode->i_sb,
3013                                            "Directory hole detected on inode %lu\n",
3014                                            inode->i_ino);
3015                         }
3016                         return NULL;
3017                 }
3018                 *parent_de = ext4_next_entry(
3019                                         (struct ext4_dir_entry_2 *)bh->b_data,
3020                                         inode->i_sb->s_blocksize);
3021                 return bh;
3022         }
3023
3024         *inlined = 1;
3025         return ext4_get_first_inline_block(inode, parent_de, retval);
3026 }
3027
3028 /*
3029  * Anybody can rename anything with this: the permission checks are left to the
3030  * higher-level routines.
3031  */
3032 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3033                        struct inode *new_dir, struct dentry *new_dentry)
3034 {
3035         handle_t *handle;
3036         struct inode *old_inode, *new_inode;
3037         struct buffer_head *old_bh, *new_bh, *dir_bh;
3038         struct ext4_dir_entry_2 *old_de, *new_de;
3039         int retval, force_da_alloc = 0;
3040         int inlined = 0, new_inlined = 0;
3041         struct ext4_dir_entry_2 *parent_de;
3042
3043         dquot_initialize(old_dir);
3044         dquot_initialize(new_dir);
3045
3046         old_bh = new_bh = dir_bh = NULL;
3047
3048         /* Initialize quotas before so that eventual writes go
3049          * in separate transaction */
3050         if (new_dentry->d_inode)
3051                 dquot_initialize(new_dentry->d_inode);
3052         handle = ext4_journal_start(old_dir, EXT4_HT_DIR,
3053                 (2 * EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
3054                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3055         if (IS_ERR(handle))
3056                 return PTR_ERR(handle);
3057
3058         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
3059                 ext4_handle_sync(handle);
3060
3061         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
3062         /*
3063          *  Check for inode number is _not_ due to possible IO errors.
3064          *  We might rmdir the source, keep it as pwd of some process
3065          *  and merrily kill the link to whatever was created under the
3066          *  same name. Goodbye sticky bit ;-<
3067          */
3068         old_inode = old_dentry->d_inode;
3069         retval = -ENOENT;
3070         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
3071                 goto end_rename;
3072
3073         new_inode = new_dentry->d_inode;
3074         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name,
3075                                  &new_de, &new_inlined);
3076         if (new_bh) {
3077                 if (!new_inode) {
3078                         brelse(new_bh);
3079                         new_bh = NULL;
3080                 }
3081         }
3082         if (S_ISDIR(old_inode->i_mode)) {
3083                 if (new_inode) {
3084                         retval = -ENOTEMPTY;
3085                         if (!empty_dir(new_inode))
3086                                 goto end_rename;
3087                 }
3088                 retval = -EIO;
3089                 dir_bh = ext4_get_first_dir_block(handle, old_inode,
3090                                                   &retval, &parent_de,
3091                                                   &inlined);
3092                 if (!dir_bh)
3093                         goto end_rename;
3094                 if (!inlined && !buffer_verified(dir_bh) &&
3095                     !ext4_dirent_csum_verify(old_inode,
3096                                 (struct ext4_dir_entry *)dir_bh->b_data))
3097                         goto end_rename;
3098                 set_buffer_verified(dir_bh);
3099                 if (le32_to_cpu(parent_de->inode) != old_dir->i_ino)
3100                         goto end_rename;
3101                 retval = -EMLINK;
3102                 if (!new_inode && new_dir != old_dir &&
3103                     EXT4_DIR_LINK_MAX(new_dir))
3104                         goto end_rename;
3105                 BUFFER_TRACE(dir_bh, "get_write_access");
3106                 retval = ext4_journal_get_write_access(handle, dir_bh);
3107                 if (retval)
3108                         goto end_rename;
3109         }
3110         if (!new_bh) {
3111                 retval = ext4_add_entry(handle, new_dentry, old_inode);
3112                 if (retval)
3113                         goto end_rename;
3114         } else {
3115                 BUFFER_TRACE(new_bh, "get write access");
3116                 retval = ext4_journal_get_write_access(handle, new_bh);
3117                 if (retval)
3118                         goto end_rename;
3119                 new_de->inode = cpu_to_le32(old_inode->i_ino);
3120                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3121                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
3122                         new_de->file_type = old_de->file_type;
3123                 new_dir->i_version++;
3124                 new_dir->i_ctime = new_dir->i_mtime =
3125                                         ext4_current_time(new_dir);
3126                 ext4_mark_inode_dirty(handle, new_dir);
3127                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3128                 if (!new_inlined) {
3129                         retval = ext4_handle_dirty_dirent_node(handle,
3130                                                                new_dir, new_bh);
3131                         if (unlikely(retval)) {
3132                                 ext4_std_error(new_dir->i_sb, retval);
3133                                 goto end_rename;
3134                         }
3135                 }
3136                 brelse(new_bh);
3137                 new_bh = NULL;
3138         }
3139
3140         /*
3141          * Like most other Unix systems, set the ctime for inodes on a
3142          * rename.
3143          */
3144         old_inode->i_ctime = ext4_current_time(old_inode);
3145         ext4_mark_inode_dirty(handle, old_inode);
3146
3147         /*
3148          * ok, that's it
3149          */
3150         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3151             old_de->name_len != old_dentry->d_name.len ||
3152             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3153             (retval = ext4_delete_entry(handle, old_dir,
3154                                         old_de, old_bh)) == -ENOENT) {
3155                 /* old_de could have moved from under us during htree split, so
3156                  * make sure that we are deleting the right entry.  We might
3157                  * also be pointing to a stale entry in the unused part of
3158                  * old_bh so just checking inum and the name isn't enough. */
3159                 struct buffer_head *old_bh2;
3160                 struct ext4_dir_entry_2 *old_de2;
3161
3162                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
3163                                           &old_de2, NULL);
3164                 if (old_bh2) {
3165                         retval = ext4_delete_entry(handle, old_dir,
3166                                                    old_de2, old_bh2);
3167                         brelse(old_bh2);
3168                 }
3169         }
3170         if (retval) {
3171                 ext4_warning(old_dir->i_sb,
3172                                 "Deleting old file (%lu), %d, error=%d",
3173                                 old_dir->i_ino, old_dir->i_nlink, retval);
3174         }
3175
3176         if (new_inode) {
3177                 ext4_dec_count(handle, new_inode);
3178                 new_inode->i_ctime = ext4_current_time(new_inode);
3179         }
3180         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3181         ext4_update_dx_flag(old_dir);
3182         if (dir_bh) {
3183                 parent_de->inode = cpu_to_le32(new_dir->i_ino);
3184                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3185                 if (!inlined) {
3186                         if (is_dx(old_inode)) {
3187                                 retval = ext4_handle_dirty_dx_node(handle,
3188                                                                    old_inode,
3189                                                                    dir_bh);
3190                         } else {
3191                                 retval = ext4_handle_dirty_dirent_node(handle,
3192                                                         old_inode, dir_bh);
3193                         }
3194                 } else {
3195                         retval = ext4_mark_inode_dirty(handle, old_inode);
3196                 }
3197                 if (retval) {
3198                         ext4_std_error(old_dir->i_sb, retval);
3199                         goto end_rename;
3200                 }
3201                 ext4_dec_count(handle, old_dir);
3202                 if (new_inode) {
3203                         /* checked empty_dir above, can't have another parent,
3204                          * ext4_dec_count() won't work for many-linked dirs */
3205                         clear_nlink(new_inode);
3206                 } else {
3207                         ext4_inc_count(handle, new_dir);
3208                         ext4_update_dx_flag(new_dir);
3209                         ext4_mark_inode_dirty(handle, new_dir);
3210                 }
3211         }
3212         ext4_mark_inode_dirty(handle, old_dir);
3213         if (new_inode) {
3214                 ext4_mark_inode_dirty(handle, new_inode);
3215                 if (!new_inode->i_nlink)
3216                         ext4_orphan_add(handle, new_inode);
3217                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3218                         force_da_alloc = 1;
3219         }
3220         retval = 0;
3221
3222 end_rename:
3223         brelse(dir_bh);
3224         brelse(old_bh);
3225         brelse(new_bh);
3226         ext4_journal_stop(handle);
3227         if (retval == 0 && force_da_alloc)
3228                 ext4_alloc_da_blocks(old_inode);
3229         return retval;
3230 }
3231
3232 /*
3233  * directories can handle most operations...
3234  */
3235 const struct inode_operations ext4_dir_inode_operations = {
3236         .create         = ext4_create,
3237         .lookup         = ext4_lookup,
3238         .link           = ext4_link,
3239         .unlink         = ext4_unlink,
3240         .symlink        = ext4_symlink,
3241         .mkdir          = ext4_mkdir,
3242         .rmdir          = ext4_rmdir,
3243         .mknod          = ext4_mknod,
3244         .rename         = ext4_rename,
3245         .setattr        = ext4_setattr,
3246         .setxattr       = generic_setxattr,
3247         .getxattr       = generic_getxattr,
3248         .listxattr      = ext4_listxattr,
3249         .removexattr    = generic_removexattr,
3250         .get_acl        = ext4_get_acl,
3251         .fiemap         = ext4_fiemap,
3252 };
3253
3254 const struct inode_operations ext4_special_inode_operations = {
3255         .setattr        = ext4_setattr,
3256         .setxattr       = generic_setxattr,
3257         .getxattr       = generic_getxattr,
3258         .listxattr      = ext4_listxattr,
3259         .removexattr    = generic_removexattr,
3260         .get_acl        = ext4_get_acl,
3261 };