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
  51 static struct buffer_head *ext4_append(handle_t *handle,
  52                                         struct inode *inode,
  53                                         ext4_lblk_t *block)
  54 {
  55         struct buffer_head *bh;
  56         int err = 0;
  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                 return ERR_PTR(-ENOSPC);
  62
  63         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  64
  65         bh = ext4_bread(handle, inode, *block, 1, &err);
  66         if (!bh)
  67                 return ERR_PTR(err);
  68         inode->i_size += inode->i_sb->s_blocksize;
  69         EXT4_I(inode)->i_disksize = inode->i_size;
  70         BUFFER_TRACE(bh, "get_write_access");
  71         err = ext4_journal_get_write_access(handle, bh);
  72         if (err) {
  73                 brelse(bh);
  74                 ext4_std_error(inode->i_sb, err);
  75                 return ERR_PTR(err);
  76         }
  77         return bh;
  78 }
  79
  80 static int ext4_dx_csum_verify(struct inode *inode,
  81                                struct ext4_dir_entry *dirent);
  82
  83 typedef enum {
  84         EITHER, INDEX, DIRENT
  85 } dirblock_type_t;
  86
  87 #define ext4_read_dirblock(inode, block, type) \
  88         __ext4_read_dirblock((inode), (block), (type), __LINE__)
  89
  90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  91                                               ext4_lblk_t block,
  92                                               dirblock_type_t type,
  93                                               unsigned int line)
  94 {
  95         struct buffer_head *bh;
  96         struct ext4_dir_entry *dirent;
  97         int err = 0, is_dx_block = 0;
  98
  99         bh = ext4_bread(NULL, inode, block, 0, &err);
 100         if (!bh) {
 101                 if (err == 0) {
 102                         ext4_error_inode(inode, __func__, line, block,
 103                                                "Directory hole found");
 104                         return ERR_PTR(-EIO);
 105                 }
 106                 __ext4_warning(inode->i_sb, __func__, line,
 107                                "error reading directory block "
 108                                "(ino %lu, block %lu)", inode->i_ino,
 109                                (unsigned long) block);
 110                 return ERR_PTR(err);
 111         }
 112         dirent = (struct ext4_dir_entry *) bh->b_data;
 113         /* Determine whether or not we have an index block */
 114         if (is_dx(inode)) {
 115                 if (block == 0)
 116                         is_dx_block = 1;
 117                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 118                                                 inode->i_sb->s_blocksize) ==
 119                          inode->i_sb->s_blocksize)
 120                         is_dx_block = 1;
 121         }
 122         if (!is_dx_block && type == INDEX) {
 123                 ext4_error_inode(inode, __func__, line, block,
 124                        "directory leaf block found instead of index block");
 125                 return ERR_PTR(-EIO);
 126         }
 127         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 128                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
 129             buffer_verified(bh))
 130                 return bh;
 131
 132         /*
 133          * An empty leaf block can get mistaken for a index block; for
 134          * this reason, we can only check the index checksum when the
 135          * caller is sure it should be an index block.
 136          */
 137         if (is_dx_block && type == INDEX) {
 138                 if (ext4_dx_csum_verify(inode, dirent))
 139                         set_buffer_verified(bh);
 140                 else {
 141                         ext4_error_inode(inode, __func__, line, block,
 142                                 "Directory index failed checksum");
 143                         brelse(bh);
 144                         return ERR_PTR(-EIO);
 145                 }
 146         }
 147         if (!is_dx_block) {
 148                 if (ext4_dirent_csum_verify(inode, dirent))
 149                         set_buffer_verified(bh);
 150                 else {
 151                         ext4_error_inode(inode, __func__, line, block,
 152                                 "Directory block failed checksum");
 153                         brelse(bh);
 154                         return ERR_PTR(-EIO);
 155                 }
 156         }
 157         return bh;
 158 }
 159
 160 #ifndef assert
 161 #define assert(test) J_ASSERT(test)
 162 #endif
 163
 164 #ifdef DX_DEBUG
 165 #define dxtrace(command) command
 166 #else
 167 #define dxtrace(command)
 168 #endif
 169
 170 struct fake_dirent
 171 {
 172         __le32 inode;
 173         __le16 rec_len;
 174         u8 name_len;
 175         u8 file_type;
 176 };
 177
 178 struct dx_countlimit
 179 {
 180         __le16 limit;
 181         __le16 count;
 182 };
 183
 184 struct dx_entry
 185 {
 186         __le32 hash;
 187         __le32 block;
 188 };
 189
 190 /*
 191  * dx_root_info is laid out so that if it should somehow get overlaid by a
 192  * dirent the two low bits of the hash version will be zero.  Therefore, the
 193  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 194  */
 195
 196 struct dx_root
 197 {
 198         struct fake_dirent dot;
 199         char dot_name[4];
 200         struct fake_dirent dotdot;
 201         char dotdot_name[4];
 202         struct dx_root_info
 203         {
 204                 __le32 reserved_zero;
 205                 u8 hash_version;
 206                 u8 info_length; /* 8 */
 207                 u8 indirect_levels;
 208                 u8 unused_flags;
 209         }
 210         info;
 211         struct dx_entry entries[0];
 212 };
 213
 214 struct dx_node
 215 {
 216         struct fake_dirent fake;
 217         struct dx_entry entries[0];
 218 };
 219
 220
 221 struct dx_frame
 222 {
 223         struct buffer_head *bh;
 224         struct dx_entry *entries;
 225         struct dx_entry *at;
 226 };
 227
 228 struct dx_map_entry
 229 {
 230         u32 hash;
 231         u16 offs;
 232         u16 size;
 233 };
 234
 235 /*
 236  * This goes at the end of each htree block.
 237  */
 238 struct dx_tail {
 239         u32 dt_reserved;
 240         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 241 };
 242
 243 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 244 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 245 static inline unsigned dx_get_hash(struct dx_entry *entry);
 246 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 247 static unsigned dx_get_count(struct dx_entry *entries);
 248 static unsigned dx_get_limit(struct dx_entry *entries);
 249 static void dx_set_count(struct dx_entry *entries, unsigned value);
 250 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 251 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 252 static unsigned dx_node_limit(struct inode *dir);
 253 static struct dx_frame *dx_probe(const struct qstr *d_name,
 254                                  struct inode *dir,
 255                                  struct dx_hash_info *hinfo,
 256                                  struct dx_frame *frame,
 257                                  int *err);
 258 static void dx_release(struct dx_frame *frames);
 259 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 260                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 261 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 262 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 263                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 264 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 265 static void dx_insert_block(struct dx_frame *frame,
 266                                         u32 hash, ext4_lblk_t block);
 267 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 268                                  struct dx_frame *frame,
 269                                  struct dx_frame *frames,
 270                                  __u32 *start_hash);
 271 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 272                 const struct qstr *d_name,
 273                 struct ext4_dir_entry_2 **res_dir,
 274                 int *err);
 275 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 276                              struct inode *inode);
 277
 278 /* checksumming functions */
 279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 280                             unsigned int blocksize)
 281 {
 282         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 283         t->det_rec_len = ext4_rec_len_to_disk(
 284                         sizeof(struct ext4_dir_entry_tail), blocksize);
 285         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 286 }
 287
 288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 290                                                    struct ext4_dir_entry *de)
 291 {
 292         struct ext4_dir_entry_tail *t;
 293
 294 #ifdef PARANOID
 295         struct ext4_dir_entry *d, *top;
 296
 297         d = de;
 298         top = (struct ext4_dir_entry *)(((void *)de) +
 299                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 300                 sizeof(struct ext4_dir_entry_tail)));
 301         while (d < top && d->rec_len)
 302                 d = (struct ext4_dir_entry *)(((void *)d) +
 303                     le16_to_cpu(d->rec_len));
 304
 305         if (d != top)
 306                 return NULL;
 307
 308         t = (struct ext4_dir_entry_tail *)d;
 309 #else
 310         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 311 #endif
 312
 313         if (t->det_reserved_zero1 ||
 314             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 315             t->det_reserved_zero2 ||
 316             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 317                 return NULL;
 318
 319         return t;
 320 }
 321
 322 static __le32 ext4_dirent_csum(struct inode *inode,
 323                                struct ext4_dir_entry *dirent, int size)
 324 {
 325         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 326         struct ext4_inode_info *ei = EXT4_I(inode);
 327         __u32 csum;
 328
 329         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 330         return cpu_to_le32(csum);
 331 }
 332
 333 static void warn_no_space_for_csum(struct inode *inode)
 334 {
 335         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
 336                      "checksum.  Please run e2fsck -D.", inode->i_ino);
 337 }
 338
 339 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 340 {
 341         struct ext4_dir_entry_tail *t;
 342
 343         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 344                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 345                 return 1;
 346
 347         t = get_dirent_tail(inode, dirent);
 348         if (!t) {
 349                 warn_no_space_for_csum(inode);
 350                 return 0;
 351         }
 352
 353         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 354                                                 (void *)t - (void *)dirent))
 355                 return 0;
 356
 357         return 1;
 358 }
 359
 360 static void ext4_dirent_csum_set(struct inode *inode,
 361                                  struct ext4_dir_entry *dirent)
 362 {
 363         struct ext4_dir_entry_tail *t;
 364
 365         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 366                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 367                 return;
 368
 369         t = get_dirent_tail(inode, dirent);
 370         if (!t) {
 371                 warn_no_space_for_csum(inode);
 372                 return;
 373         }
 374
 375         t->det_checksum = ext4_dirent_csum(inode, dirent,
 376                                            (void *)t - (void *)dirent);
 377 }
 378
 379 int ext4_handle_dirty_dirent_node(handle_t *handle,
 380                                   struct inode *inode,
 381                                   struct buffer_head *bh)
 382 {
 383         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 384         return ext4_handle_dirty_metadata(handle, inode, bh);
 385 }
 386
 387 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 388                                                struct ext4_dir_entry *dirent,
 389                                                int *offset)
 390 {
 391         struct ext4_dir_entry *dp;
 392         struct dx_root_info *root;
 393         int count_offset;
 394
 395         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 396                 count_offset = 8;
 397         else if (le16_to_cpu(dirent->rec_len) == 12) {
 398                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 399                 if (le16_to_cpu(dp->rec_len) !=
 400                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 401                         return NULL;
 402                 root = (struct dx_root_info *)(((void *)dp + 12));
 403                 if (root->reserved_zero ||
 404                     root->info_length != sizeof(struct dx_root_info))
 405                         return NULL;
 406                 count_offset = 32;
 407         } else
 408                 return NULL;
 409
 410         if (offset)
 411                 *offset = count_offset;
 412         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 413 }
 414
 415 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 416                            int count_offset, int count, struct dx_tail *t)
 417 {
 418         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 419         struct ext4_inode_info *ei = EXT4_I(inode);
 420         __u32 csum;
 421         __le32 save_csum;
 422         int size;
 423
 424         size = count_offset + (count * sizeof(struct dx_entry));
 425         save_csum = t->dt_checksum;
 426         t->dt_checksum = 0;
 427         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 428         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
 429         t->dt_checksum = save_csum;
 430
 431         return cpu_to_le32(csum);
 432 }
 433
 434 static int ext4_dx_csum_verify(struct inode *inode,
 435                                struct ext4_dir_entry *dirent)
 436 {
 437         struct dx_countlimit *c;
 438         struct dx_tail *t;
 439         int count_offset, limit, count;
 440
 441         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 442                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 443                 return 1;
 444
 445         c = get_dx_countlimit(inode, dirent, &count_offset);
 446         if (!c) {
 447                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 448                 return 1;
 449         }
 450         limit = le16_to_cpu(c->limit);
 451         count = le16_to_cpu(c->count);
 452         if (count_offset + (limit * sizeof(struct dx_entry)) >
 453             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 454                 warn_no_space_for_csum(inode);
 455                 return 1;
 456         }
 457         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 458
 459         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 460                                             count, t))
 461                 return 0;
 462         return 1;
 463 }
 464
 465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 466 {
 467         struct dx_countlimit *c;
 468         struct dx_tail *t;
 469         int count_offset, limit, count;
 470
 471         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 472                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 473                 return;
 474
 475         c = get_dx_countlimit(inode, dirent, &count_offset);
 476         if (!c) {
 477                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 478                 return;
 479         }
 480         limit = le16_to_cpu(c->limit);
 481         count = le16_to_cpu(c->count);
 482         if (count_offset + (limit * sizeof(struct dx_entry)) >
 483             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 484                 warn_no_space_for_csum(inode);
 485                 return;
 486         }
 487         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 488
 489         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 490 }
 491
 492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 493                                             struct inode *inode,
 494                                             struct buffer_head *bh)
 495 {
 496         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 497         return ext4_handle_dirty_metadata(handle, inode, bh);
 498 }
 499
 500 /*
 501  * p is at least 6 bytes before the end of page
 502  */
 503 static inline struct ext4_dir_entry_2 *
 504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 505 {
 506         return (struct ext4_dir_entry_2 *)((char *)p +
 507                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 508 }
 509
 510 /*
 511  * Future: use high four bits of block for coalesce-on-delete flags
 512  * Mask them off for now.
 513  */
 514
 515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 516 {
 517         return le32_to_cpu(entry->block) & 0x00ffffff;
 518 }
 519
 520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 521 {
 522         entry->block = cpu_to_le32(value);
 523 }
 524
 525 static inline unsigned dx_get_hash(struct dx_entry *entry)
 526 {
 527         return le32_to_cpu(entry->hash);
 528 }
 529
 530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 531 {
 532         entry->hash = cpu_to_le32(value);
 533 }
 534
 535 static inline unsigned dx_get_count(struct dx_entry *entries)
 536 {
 537         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 538 }
 539
 540 static inline unsigned dx_get_limit(struct dx_entry *entries)
 541 {
 542         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 543 }
 544
 545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 546 {
 547         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 548 }
 549
 550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 551 {
 552         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 553 }
 554
 555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 556 {
 557         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 558                 EXT4_DIR_REC_LEN(2) - infosize;
 559
 560         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 561                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 562                 entry_space -= sizeof(struct dx_tail);
 563         return entry_space / sizeof(struct dx_entry);
 564 }
 565
 566 static inline unsigned dx_node_limit(struct inode *dir)
 567 {
 568         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 569
 570         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 571                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 572                 entry_space -= sizeof(struct dx_tail);
 573         return entry_space / sizeof(struct dx_entry);
 574 }
 575
 576 /*
 577  * Debug
 578  */
 579 #ifdef DX_DEBUG
 580 static void dx_show_index(char * label, struct dx_entry *entries)
 581 {
 582         int i, n = dx_get_count (entries);
 583         printk(KERN_DEBUG "%s index ", label);
 584         for (i = 0; i < n; i++) {
 585                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 586                                 0, (unsigned long)dx_get_block(entries + i));
 587         }
 588         printk("\n");
 589 }
 590
 591 struct stats
 592 {
 593         unsigned names;
 594         unsigned space;
 595         unsigned bcount;
 596 };
 597
 598 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 599                                  int size, int show_names)
 600 {
 601         unsigned names = 0, space = 0;
 602         char *base = (char *) de;
 603         struct dx_hash_info h = *hinfo;
 604
 605         printk("names: ");
 606         while ((char *) de < base + size)
 607         {
 608                 if (de->inode)
 609                 {
 610                         if (show_names)
 611                         {
 612                                 int len = de->name_len;
 613                                 char *name = de->name;
 614                                 while (len--) printk("%c", *name++);
 615                                 ext4fs_dirhash(de->name, de->name_len, &h);
 616                                 printk(":%x.%u ", h.hash,
 617                                        (unsigned) ((char *) de - base));
 618                         }
 619                         space += EXT4_DIR_REC_LEN(de->name_len);
 620                         names++;
 621                 }
 622                 de = ext4_next_entry(de, size);
 623         }
 624         printk("(%i)\n", names);
 625         return (struct stats) { names, space, 1 };
 626 }
 627
 628 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 629                              struct dx_entry *entries, int levels)
 630 {
 631         unsigned blocksize = dir->i_sb->s_blocksize;
 632         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 633         unsigned bcount = 0;
 634         struct buffer_head *bh;
 635         int err;
 636         printk("%i indexed blocks...\n", count);
 637         for (i = 0; i < count; i++, entries++)
 638         {
 639                 ext4_lblk_t block = dx_get_block(entries);
 640                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 641                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 642                 struct stats stats;
 643                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 644                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 645                 stats = levels?
 646                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 647                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 648                 names += stats.names;
 649                 space += stats.space;
 650                 bcount += stats.bcount;
 651                 brelse(bh);
 652         }
 653         if (bcount)
 654                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 655                        levels ? "" : "   ", names, space/bcount,
 656                        (space/bcount)*100/blocksize);
 657         return (struct stats) { names, space, bcount};
 658 }
 659 #endif /* DX_DEBUG */
 660
 661 /*
 662  * Probe for a directory leaf block to search.
 663  *
 664  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 665  * error in the directory index, and the caller should fall back to
 666  * searching the directory normally.  The callers of dx_probe **MUST**
 667  * check for this error code, and make sure it never gets reflected
 668  * back to userspace.
 669  */
 670 static struct dx_frame *
 671 dx_probe(const struct qstr *d_name, struct inode *dir,
 672          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 673 {
 674         unsigned count, indirect;
 675         struct dx_entry *at, *entries, *p, *q, *m;
 676         struct dx_root *root;
 677         struct buffer_head *bh;
 678         struct dx_frame *frame = frame_in;
 679         u32 hash;
 680
 681         frame->bh = NULL;
 682         bh = ext4_read_dirblock(dir, 0, INDEX);
 683         if (IS_ERR(bh)) {
 684                 *err = PTR_ERR(bh);
 685                 goto fail;
 686         }
 687         root = (struct dx_root *) bh->b_data;
 688         if (root->info.hash_version != DX_HASH_TEA &&
 689             root->info.hash_version != DX_HASH_HALF_MD4 &&
 690             root->info.hash_version != DX_HASH_LEGACY) {
 691                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 692                              root->info.hash_version);
 693                 brelse(bh);
 694                 *err = ERR_BAD_DX_DIR;
 695                 goto fail;
 696         }
 697         hinfo->hash_version = root->info.hash_version;
 698         if (hinfo->hash_version <= DX_HASH_TEA)
 699                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 700         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 701         if (d_name)
 702                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 703         hash = hinfo->hash;
 704
 705         if (root->info.unused_flags & 1) {
 706                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 707                              root->info.unused_flags);
 708                 brelse(bh);
 709                 *err = ERR_BAD_DX_DIR;
 710                 goto fail;
 711         }
 712
 713         if ((indirect = root->info.indirect_levels) > 1) {
 714                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 715                              root->info.indirect_levels);
 716                 brelse(bh);
 717                 *err = ERR_BAD_DX_DIR;
 718                 goto fail;
 719         }
 720
 721         entries = (struct dx_entry *) (((char *)&root->info) +
 722                                        root->info.info_length);
 723
 724         if (dx_get_limit(entries) != dx_root_limit(dir,
 725                                                    root->info.info_length)) {
 726                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 727                 brelse(bh);
 728                 *err = ERR_BAD_DX_DIR;
 729                 goto fail;
 730         }
 731
 732         dxtrace(printk("Look up %x", hash));
 733         while (1)
 734         {
 735                 count = dx_get_count(entries);
 736                 if (!count || count > dx_get_limit(entries)) {
 737                         ext4_warning(dir->i_sb,
 738                                      "dx entry: no count or count > limit");
 739                         brelse(bh);
 740                         *err = ERR_BAD_DX_DIR;
 741                         goto fail2;
 742                 }
 743
 744                 p = entries + 1;
 745                 q = entries + count - 1;
 746                 while (p <= q)
 747                 {
 748                         m = p + (q - p)/2;
 749                         dxtrace(printk("."));
 750                         if (dx_get_hash(m) > hash)
 751                                 q = m - 1;
 752                         else
 753                                 p = m + 1;
 754                 }
 755
 756                 if (0) // linear search cross check
 757                 {
 758                         unsigned n = count - 1;
 759                         at = entries;
 760                         while (n--)
 761                         {
 762                                 dxtrace(printk(","));
 763                                 if (dx_get_hash(++at) > hash)
 764                                 {
 765                                         at--;
 766                                         break;
 767                                 }
 768                         }
 769                         assert (at == p - 1);
 770                 }
 771
 772                 at = p - 1;
 773                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 774                 frame->bh = bh;
 775                 frame->entries = entries;
 776                 frame->at = at;
 777                 if (!indirect--) return frame;
 778                 bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 779                 if (IS_ERR(bh)) {
 780                         *err = PTR_ERR(bh);
 781                         goto fail2;
 782                 }
 783                 entries = ((struct dx_node *) bh->b_data)->entries;
 784
 785                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 786                         ext4_warning(dir->i_sb,
 787                                      "dx entry: limit != node limit");
 788                         brelse(bh);
 789                         *err = ERR_BAD_DX_DIR;
 790                         goto fail2;
 791                 }
 792                 frame++;
 793                 frame->bh = NULL;
 794         }
 795 fail2:
 796         while (frame >= frame_in) {
 797                 brelse(frame->bh);
 798                 frame--;
 799         }
 800 fail:
 801         if (*err == ERR_BAD_DX_DIR)
 802                 ext4_warning(dir->i_sb,
 803                              "Corrupt dir inode %lu, running e2fsck is "
 804                              "recommended.", dir->i_ino);
 805         return NULL;
 806 }
 807
 808 static void dx_release (struct dx_frame *frames)
 809 {
 810         if (frames[0].bh == NULL)
 811                 return;
 812
 813         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 814                 brelse(frames[1].bh);
 815         brelse(frames[0].bh);
 816 }
 817
 818 /*
 819  * This function increments the frame pointer to search the next leaf
 820  * block, and reads in the necessary intervening nodes if the search
 821  * should be necessary.  Whether or not the search is necessary is
 822  * controlled by the hash parameter.  If the hash value is even, then
 823  * the search is only continued if the next block starts with that
 824  * hash value.  This is used if we are searching for a specific file.
 825  *
 826  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 827  *
 828  * This function returns 1 if the caller should continue to search,
 829  * or 0 if it should not.  If there is an error reading one of the
 830  * index blocks, it will a negative error code.
 831  *
 832  * If start_hash is non-null, it will be filled in with the starting
 833  * hash of the next page.
 834  */
 835 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 836                                  struct dx_frame *frame,
 837                                  struct dx_frame *frames,
 838                                  __u32 *start_hash)
 839 {
 840         struct dx_frame *p;
 841         struct buffer_head *bh;
 842         int num_frames = 0;
 843         __u32 bhash;
 844
 845         p = frame;
 846         /*
 847          * Find the next leaf page by incrementing the frame pointer.
 848          * If we run out of entries in the interior node, loop around and
 849          * increment pointer in the parent node.  When we break out of
 850          * this loop, num_frames indicates the number of interior
 851          * nodes need to be read.
 852          */
 853         while (1) {
 854                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 855                         break;
 856                 if (p == frames)
 857                         return 0;
 858                 num_frames++;
 859                 p--;
 860         }
 861
 862         /*
 863          * If the hash is 1, then continue only if the next page has a
 864          * continuation hash of any value.  This is used for readdir
 865          * handling.  Otherwise, check to see if the hash matches the
 866          * desired contiuation hash.  If it doesn't, return since
 867          * there's no point to read in the successive index pages.
 868          */
 869         bhash = dx_get_hash(p->at);
 870         if (start_hash)
 871                 *start_hash = bhash;
 872         if ((hash & 1) == 0) {
 873                 if ((bhash & ~1) != hash)
 874                         return 0;
 875         }
 876         /*
 877          * If the hash is HASH_NB_ALWAYS, we always go to the next
 878          * block so no check is necessary
 879          */
 880         while (num_frames--) {
 881                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 882                 if (IS_ERR(bh))
 883                         return PTR_ERR(bh);
 884                 p++;
 885                 brelse(p->bh);
 886                 p->bh = bh;
 887                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 888         }
 889         return 1;
 890 }
 891
 892
 893 /*
 894  * This function fills a red-black tree with information from a
 895  * directory block.  It returns the number directory entries loaded
 896  * into the tree.  If there is an error it is returned in err.
 897  */
 898 static int htree_dirblock_to_tree(struct file *dir_file,
 899                                   struct inode *dir, ext4_lblk_t block,
 900                                   struct dx_hash_info *hinfo,
 901                                   __u32 start_hash, __u32 start_minor_hash)
 902 {
 903         struct buffer_head *bh;
 904         struct ext4_dir_entry_2 *de, *top;
 905         int err = 0, count = 0;
 906
 907         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 908                                                         (unsigned long)block));
 909         bh = ext4_read_dirblock(dir, block, DIRENT);
 910         if (IS_ERR(bh))
 911                 return PTR_ERR(bh);
 912
 913         de = (struct ext4_dir_entry_2 *) bh->b_data;
 914         top = (struct ext4_dir_entry_2 *) ((char *) de +
 915                                            dir->i_sb->s_blocksize -
 916                                            EXT4_DIR_REC_LEN(0));
 917         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 918                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 919                                 bh->b_data, bh->b_size,
 920                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 921                                          + ((char *)de - bh->b_data))) {
 922                         /* silently ignore the rest of the block */
 923                         break;
 924                 }
 925                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 926                 if ((hinfo->hash < start_hash) ||
 927                     ((hinfo->hash == start_hash) &&
 928                      (hinfo->minor_hash < start_minor_hash)))
 929                         continue;
 930                 if (de->inode == 0)
 931                         continue;
 932                 if ((err = ext4_htree_store_dirent(dir_file,
 933                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 934                         brelse(bh);
 935                         return err;
 936                 }
 937                 count++;
 938         }
 939         brelse(bh);
 940         return count;
 941 }
 942
 943
 944 /*
 945  * This function fills a red-black tree with information from a
 946  * directory.  We start scanning the directory in hash order, starting
 947  * at start_hash and start_minor_hash.
 948  *
 949  * This function returns the number of entries inserted into the tree,
 950  * or a negative error code.
 951  */
 952 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 953                          __u32 start_minor_hash, __u32 *next_hash)
 954 {
 955         struct dx_hash_info hinfo;
 956         struct ext4_dir_entry_2 *de;
 957         struct dx_frame frames[2], *frame;
 958         struct inode *dir;
 959         ext4_lblk_t block;
 960         int count = 0;
 961         int ret, err;
 962         __u32 hashval;
 963
 964         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 965                        start_hash, start_minor_hash));
 966         dir = file_inode(dir_file);
 967         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 968                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 969                 if (hinfo.hash_version <= DX_HASH_TEA)
 970                         hinfo.hash_version +=
 971                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 972                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 973                 if (ext4_has_inline_data(dir)) {
 974                         int has_inline_data = 1;
 975                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
 976                                                         &hinfo, start_hash,
 977                                                         start_minor_hash,
 978                                                         &has_inline_data);
 979                         if (has_inline_data) {
 980                                 *next_hash = ~0;
 981                                 return count;
 982                         }
 983                 }
 984                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 985                                                start_hash, start_minor_hash);
 986                 *next_hash = ~0;
 987                 return count;
 988         }
 989         hinfo.hash = start_hash;
 990         hinfo.minor_hash = 0;
 991         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 992         if (!frame)
 993                 return err;
 994
 995         /* Add '.' and '..' from the htree header */
 996         if (!start_hash && !start_minor_hash) {
 997                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 998                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 999                         goto errout;
1000                 count++;
1001         }
1002         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1003                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1004                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1005                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1006                         goto errout;
1007                 count++;
1008         }
1009
1010         while (1) {
1011                 block = dx_get_block(frame->at);
1012                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1013                                              start_hash, start_minor_hash);
1014                 if (ret < 0) {
1015                         err = ret;
1016                         goto errout;
1017                 }
1018                 count += ret;
1019                 hashval = ~0;
1020                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1021                                             frame, frames, &hashval);
1022                 *next_hash = hashval;
1023                 if (ret < 0) {
1024                         err = ret;
1025                         goto errout;
1026                 }
1027                 /*
1028                  * Stop if:  (a) there are no more entries, or
1029                  * (b) we have inserted at least one entry and the
1030                  * next hash value is not a continuation
1031                  */
1032                 if ((ret == 0) ||
1033                     (count && ((hashval & 1) == 0)))
1034                         break;
1035         }
1036         dx_release(frames);
1037         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1038                        "next hash: %x\n", count, *next_hash));
1039         return count;
1040 errout:
1041         dx_release(frames);
1042         return (err);
1043 }
1044
1045 static inline int search_dirblock(struct buffer_head *bh,
1046                                   struct inode *dir,
1047                                   const struct qstr *d_name,
1048                                   unsigned int offset,
1049                                   struct ext4_dir_entry_2 **res_dir)
1050 {
1051         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1052                           d_name, offset, res_dir);
1053 }
1054
1055 /*
1056  * Directory block splitting, compacting
1057  */
1058
1059 /*
1060  * Create map of hash values, offsets, and sizes, stored at end of block.
1061  * Returns number of entries mapped.
1062  */
1063 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1064                        struct dx_hash_info *hinfo,
1065                        struct dx_map_entry *map_tail)
1066 {
1067         int count = 0;
1068         char *base = (char *) de;
1069         struct dx_hash_info h = *hinfo;
1070
1071         while ((char *) de < base + blocksize) {
1072                 if (de->name_len && de->inode) {
1073                         ext4fs_dirhash(de->name, de->name_len, &h);
1074                         map_tail--;
1075                         map_tail->hash = h.hash;
1076                         map_tail->offs = ((char *) de - base)>>2;
1077                         map_tail->size = le16_to_cpu(de->rec_len);
1078                         count++;
1079                         cond_resched();
1080                 }
1081                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1082                 de = ext4_next_entry(de, blocksize);
1083         }
1084         return count;
1085 }
1086
1087 /* Sort map by hash value */
1088 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1089 {
1090         struct dx_map_entry *p, *q, *top = map + count - 1;
1091         int more;
1092         /* Combsort until bubble sort doesn't suck */
1093         while (count > 2) {
1094                 count = count*10/13;
1095                 if (count - 9 < 2) /* 9, 10 -> 11 */
1096                         count = 11;
1097                 for (p = top, q = p - count; q >= map; p--, q--)
1098                         if (p->hash < q->hash)
1099                                 swap(*p, *q);
1100         }
1101         /* Garden variety bubble sort */
1102         do {
1103                 more = 0;
1104                 q = top;
1105                 while (q-- > map) {
1106                         if (q[1].hash >= q[0].hash)
1107                                 continue;
1108                         swap(*(q+1), *q);
1109                         more = 1;
1110                 }
1111         } while(more);
1112 }
1113
1114 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1115 {
1116         struct dx_entry *entries = frame->entries;
1117         struct dx_entry *old = frame->at, *new = old + 1;
1118         int count = dx_get_count(entries);
1119
1120         assert(count < dx_get_limit(entries));
1121         assert(old < entries + count);
1122         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1123         dx_set_hash(new, hash);
1124         dx_set_block(new, block);
1125         dx_set_count(entries, count + 1);
1126 }
1127
1128 /*
1129  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1130  *
1131  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1132  * `de != NULL' is guaranteed by caller.
1133  */
1134 static inline int ext4_match (int len, const char * const name,
1135                               struct ext4_dir_entry_2 * de)
1136 {
1137         if (len != de->name_len)
1138                 return 0;
1139         if (!de->inode)
1140                 return 0;
1141         return !memcmp(name, de->name, len);
1142 }
1143
1144 /*
1145  * Returns 0 if not found, -1 on failure, and 1 on success
1146  */
1147 int search_dir(struct buffer_head *bh,
1148                char *search_buf,
1149                int buf_size,
1150                struct inode *dir,
1151                const struct qstr *d_name,
1152                unsigned int offset,
1153                struct ext4_dir_entry_2 **res_dir)
1154 {
1155         struct ext4_dir_entry_2 * de;
1156         char * dlimit;
1157         int de_len;
1158         const char *name = d_name->name;
1159         int namelen = d_name->len;
1160
1161         de = (struct ext4_dir_entry_2 *)search_buf;
1162         dlimit = search_buf + buf_size;
1163         while ((char *) de < dlimit) {
1164                 /* this code is executed quadratically often */
1165                 /* do minimal checking `by hand' */
1166
1167                 if ((char *) de + namelen <= dlimit &&
1168                     ext4_match (namelen, name, de)) {
1169                         /* found a match - just to be sure, do a full check */
1170                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1171                                                  bh->b_size, offset))
1172                                 return -1;
1173                         *res_dir = de;
1174                         return 1;
1175                 }
1176                 /* prevent looping on a bad block */
1177                 de_len = ext4_rec_len_from_disk(de->rec_len,
1178                                                 dir->i_sb->s_blocksize);
1179                 if (de_len <= 0)
1180                         return -1;
1181                 offset += de_len;
1182                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1183         }
1184         return 0;
1185 }
1186
1187 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1188                                struct ext4_dir_entry *de)
1189 {
1190         struct super_block *sb = dir->i_sb;
1191
1192         if (!is_dx(dir))
1193                 return 0;
1194         if (block == 0)
1195                 return 1;
1196         if (de->inode == 0 &&
1197             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1198                         sb->s_blocksize)
1199                 return 1;
1200         return 0;
1201 }
1202
1203 /*
1204  *      ext4_find_entry()
1205  *
1206  * finds an entry in the specified directory with the wanted name. It
1207  * returns the cache buffer in which the entry was found, and the entry
1208  * itself (as a parameter - res_dir). It does NOT read the inode of the
1209  * entry - you'll have to do that yourself if you want to.
1210  *
1211  * The returned buffer_head has ->b_count elevated.  The caller is expected
1212  * to brelse() it when appropriate.
1213  */
1214 static struct buffer_head * ext4_find_entry (struct inode *dir,
1215                                         const struct qstr *d_name,
1216                                         struct ext4_dir_entry_2 **res_dir,
1217                                         int *inlined)
1218 {
1219         struct super_block *sb;
1220         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1221         struct buffer_head *bh, *ret = NULL;
1222         ext4_lblk_t start, block, b;
1223         const u8 *name = d_name->name;
1224         int ra_max = 0;         /* Number of bh's in the readahead
1225                                    buffer, bh_use[] */
1226         int ra_ptr = 0;         /* Current index into readahead
1227                                    buffer */
1228         int num = 0;
1229         ext4_lblk_t  nblocks;
1230         int i, err = 0;
1231         int namelen;
1232
1233         *res_dir = NULL;
1234         sb = dir->i_sb;
1235         namelen = d_name->len;
1236         if (namelen > EXT4_NAME_LEN)
1237                 return NULL;
1238
1239         if (ext4_has_inline_data(dir)) {
1240                 int has_inline_data = 1;
1241                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1242                                              &has_inline_data);
1243                 if (has_inline_data) {
1244                         if (inlined)
1245                                 *inlined = 1;
1246                         return ret;
1247                 }
1248         }
1249
1250         if ((namelen <= 2) && (name[0] == '.') &&
1251             (name[1] == '.' || name[1] == '\0')) {
1252                 /*
1253                  * "." or ".." will only be in the first block
1254                  * NFS may look up ".."; "." should be handled by the VFS
1255                  */
1256                 block = start = 0;
1257                 nblocks = 1;
1258                 goto restart;
1259         }
1260         if (is_dx(dir)) {
1261                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1262                 /*
1263                  * On success, or if the error was file not found,
1264                  * return.  Otherwise, fall back to doing a search the
1265                  * old fashioned way.
1266                  */
1267                 if (err == -ENOENT)
1268                         return NULL;
1269                 if (err && err != ERR_BAD_DX_DIR)
1270                         return ERR_PTR(err);
1271                 if (bh)
1272                         return bh;
1273                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1274                                "falling back\n"));
1275         }
1276         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1277         start = EXT4_I(dir)->i_dir_start_lookup;
1278         if (start >= nblocks)
1279                 start = 0;
1280         block = start;
1281 restart:
1282         do {
1283                 /*
1284                  * We deal with the read-ahead logic here.
1285                  */
1286                 if (ra_ptr >= ra_max) {
1287                         /* Refill the readahead buffer */
1288                         ra_ptr = 0;
1289                         b = block;
1290                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1291                                 /*
1292                                  * Terminate if we reach the end of the
1293                                  * directory and must wrap, or if our
1294                                  * search has finished at this block.
1295                                  */
1296                                 if (b >= nblocks || (num && block == start)) {
1297                                         bh_use[ra_max] = NULL;
1298                                         break;
1299                                 }
1300                                 num++;
1301                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1302                                 if (unlikely(err)) {
1303                                         if (ra_max == 0)
1304                                                 return ERR_PTR(err);
1305                                         break;
1306                                 }
1307                                 bh_use[ra_max] = bh;
1308                                 if (bh)
1309                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1310                                                     1, &bh);
1311                         }
1312                 }
1313                 if ((bh = bh_use[ra_ptr++]) == NULL)
1314                         goto next;
1315                 wait_on_buffer(bh);
1316                 if (!buffer_uptodate(bh)) {
1317                         /* read error, skip block & hope for the best */
1318                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1319                                          (unsigned long) block);
1320                         brelse(bh);
1321                         goto next;
1322                 }
1323                 if (!buffer_verified(bh) &&
1324                     !is_dx_internal_node(dir, block,
1325                                          (struct ext4_dir_entry *)bh->b_data) &&
1326                     !ext4_dirent_csum_verify(dir,
1327                                 (struct ext4_dir_entry *)bh->b_data)) {
1328                         EXT4_ERROR_INODE(dir, "checksumming directory "
1329                                          "block %lu", (unsigned long)block);
1330                         brelse(bh);
1331                         goto next;
1332                 }
1333                 set_buffer_verified(bh);
1334                 i = search_dirblock(bh, dir, d_name,
1335                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1336                 if (i == 1) {
1337                         EXT4_I(dir)->i_dir_start_lookup = block;
1338                         ret = bh;
1339                         goto cleanup_and_exit;
1340                 } else {
1341                         brelse(bh);
1342                         if (i < 0)
1343                                 goto cleanup_and_exit;
1344                 }
1345         next:
1346                 if (++block >= nblocks)
1347                         block = 0;
1348         } while (block != start);
1349
1350         /*
1351          * If the directory has grown while we were searching, then
1352          * search the last part of the directory before giving up.
1353          */
1354         block = nblocks;
1355         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1356         if (block < nblocks) {
1357                 start = 0;
1358                 goto restart;
1359         }
1360
1361 cleanup_and_exit:
1362         /* Clean up the read-ahead blocks */
1363         for (; ra_ptr < ra_max; ra_ptr++)
1364                 brelse(bh_use[ra_ptr]);
1365         return ret;
1366 }
1367
1368 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1369                        struct ext4_dir_entry_2 **res_dir, int *err)
1370 {
1371         struct super_block * sb = dir->i_sb;
1372         struct dx_hash_info     hinfo;
1373         struct dx_frame frames[2], *frame;
1374         struct buffer_head *bh;
1375         ext4_lblk_t block;
1376         int retval;
1377
1378         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1379                 return NULL;
1380         do {
1381                 block = dx_get_block(frame->at);
1382                 bh = ext4_read_dirblock(dir, block, DIRENT);
1383                 if (IS_ERR(bh)) {
1384                         *err = PTR_ERR(bh);
1385                         goto errout;
1386                 }
1387                 retval = search_dirblock(bh, dir, d_name,
1388                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1389                                          res_dir);
1390                 if (retval == 1) {      /* Success! */
1391                         dx_release(frames);
1392                         return bh;
1393                 }
1394                 brelse(bh);
1395                 if (retval == -1) {
1396                         *err = ERR_BAD_DX_DIR;
1397                         goto errout;
1398                 }
1399
1400                 /* Check to see if we should continue to search */
1401                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1402                                                frames, NULL);
1403                 if (retval < 0) {
1404                         ext4_warning(sb,
1405                              "error reading index page in directory #%lu",
1406                              dir->i_ino);
1407                         *err = retval;
1408                         goto errout;
1409                 }
1410         } while (retval == 1);
1411
1412         *err = -ENOENT;
1413 errout:
1414         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1415         dx_release (frames);
1416         return NULL;
1417 }
1418
1419 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1420 {
1421         struct inode *inode;
1422         struct ext4_dir_entry_2 *de;
1423         struct buffer_head *bh;
1424
1425         if (dentry->d_name.len > EXT4_NAME_LEN)
1426                 return ERR_PTR(-ENAMETOOLONG);
1427
1428         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1429         if (IS_ERR(bh))
1430                 return (struct dentry *) bh;
1431         inode = NULL;
1432         if (bh) {
1433                 __u32 ino = le32_to_cpu(de->inode);
1434                 brelse(bh);
1435                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1436                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1437                         return ERR_PTR(-EIO);
1438                 }
1439                 if (unlikely(ino == dir->i_ino)) {
1440                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1441                                          dentry);
1442                         return ERR_PTR(-EIO);
1443                 }
1444                 inode = ext4_iget(dir->i_sb, ino);
1445                 if (inode == ERR_PTR(-ESTALE)) {
1446                         EXT4_ERROR_INODE(dir,
1447                                          "deleted inode referenced: %u",
1448                                          ino);
1449                         return ERR_PTR(-EIO);
1450                 }
1451         }
1452         return d_splice_alias(inode, dentry);
1453 }
1454
1455
1456 struct dentry *ext4_get_parent(struct dentry *child)
1457 {
1458         __u32 ino;
1459         static const struct qstr dotdot = QSTR_INIT("..", 2);
1460         struct ext4_dir_entry_2 * de;
1461         struct buffer_head *bh;
1462
1463         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1464         if (IS_ERR(bh))
1465                 return (struct dentry *) bh;
1466         if (!bh)
1467                 return ERR_PTR(-ENOENT);
1468         ino = le32_to_cpu(de->inode);
1469         brelse(bh);
1470
1471         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1472                 EXT4_ERROR_INODE(child->d_inode,
1473                                  "bad parent inode number: %u", ino);
1474                 return ERR_PTR(-EIO);
1475         }
1476
1477         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1478 }
1479
1480 /*
1481  * Move count entries from end of map between two memory locations.
1482  * Returns pointer to last entry moved.
1483  */
1484 static struct ext4_dir_entry_2 *
1485 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1486                 unsigned blocksize)
1487 {
1488         unsigned rec_len = 0;
1489
1490         while (count--) {
1491                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1492                                                 (from + (map->offs<<2));
1493                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1494                 memcpy (to, de, rec_len);
1495                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1496                                 ext4_rec_len_to_disk(rec_len, blocksize);
1497                 de->inode = 0;
1498                 map++;
1499                 to += rec_len;
1500         }
1501         return (struct ext4_dir_entry_2 *) (to - rec_len);
1502 }
1503
1504 /*
1505  * Compact each dir entry in the range to the minimal rec_len.
1506  * Returns pointer to last entry in range.
1507  */
1508 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1509 {
1510         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1511         unsigned rec_len = 0;
1512
1513         prev = to = de;
1514         while ((char*)de < base + blocksize) {
1515                 next = ext4_next_entry(de, blocksize);
1516                 if (de->inode && de->name_len) {
1517                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1518                         if (de > to)
1519                                 memmove(to, de, rec_len);
1520                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1521                         prev = to;
1522                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1523                 }
1524                 de = next;
1525         }
1526         return prev;
1527 }
1528
1529 /*
1530  * Split a full leaf block to make room for a new dir entry.
1531  * Allocate a new block, and move entries so that they are approx. equally full.
1532  * Returns pointer to de in block into which the new entry will be inserted.
1533  */
1534 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1535                         struct buffer_head **bh,struct dx_frame *frame,
1536                         struct dx_hash_info *hinfo, int *error)
1537 {
1538         unsigned blocksize = dir->i_sb->s_blocksize;
1539         unsigned count, continued;
1540         struct buffer_head *bh2;
1541         ext4_lblk_t newblock;
1542         u32 hash2;
1543         struct dx_map_entry *map;
1544         char *data1 = (*bh)->b_data, *data2;
1545         unsigned split, move, size;
1546         struct ext4_dir_entry_2 *de = NULL, *de2;
1547         struct ext4_dir_entry_tail *t;
1548         int     csum_size = 0;
1549         int     err = 0, i;
1550
1551         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1552                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1553                 csum_size = sizeof(struct ext4_dir_entry_tail);
1554
1555         bh2 = ext4_append(handle, dir, &newblock);
1556         if (IS_ERR(bh2)) {
1557                 brelse(*bh);
1558                 *bh = NULL;
1559                 *error = PTR_ERR(bh2);
1560                 return NULL;
1561         }
1562
1563         BUFFER_TRACE(*bh, "get_write_access");
1564         err = ext4_journal_get_write_access(handle, *bh);
1565         if (err)
1566                 goto journal_error;
1567
1568         BUFFER_TRACE(frame->bh, "get_write_access");
1569         err = ext4_journal_get_write_access(handle, frame->bh);
1570         if (err)
1571                 goto journal_error;
1572
1573         data2 = bh2->b_data;
1574
1575         /* create map in the end of data2 block */
1576         map = (struct dx_map_entry *) (data2 + blocksize);
1577         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1578                              blocksize, hinfo, map);
1579         map -= count;
1580         dx_sort_map(map, count);
1581         /* Split the existing block in the middle, size-wise */
1582         size = 0;
1583         move = 0;
1584         for (i = count-1; i >= 0; i--) {
1585                 /* is more than half of this entry in 2nd half of the block? */
1586                 if (size + map[i].size/2 > blocksize/2)
1587                         break;
1588                 size += map[i].size;
1589                 move++;
1590         }
1591         /* map index at which we will split */
1592         split = count - move;
1593         hash2 = map[split].hash;
1594         continued = hash2 == map[split - 1].hash;
1595         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1596                         (unsigned long)dx_get_block(frame->at),
1597                                         hash2, split, count-split));
1598
1599         /* Fancy dance to stay within two buffers */
1600         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1601         de = dx_pack_dirents(data1, blocksize);
1602         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1603                                            (char *) de,
1604                                            blocksize);
1605         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1606                                             (char *) de2,
1607                                             blocksize);
1608         if (csum_size) {
1609                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1610                 initialize_dirent_tail(t, blocksize);
1611
1612                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1613                 initialize_dirent_tail(t, blocksize);
1614         }
1615
1616         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1617         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1618
1619         /* Which block gets the new entry? */
1620         if (hinfo->hash >= hash2)
1621         {
1622                 swap(*bh, bh2);
1623                 de = de2;
1624         }
1625         dx_insert_block(frame, hash2 + continued, newblock);
1626         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1627         if (err)
1628                 goto journal_error;
1629         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1630         if (err)
1631                 goto journal_error;
1632         brelse(bh2);
1633         dxtrace(dx_show_index("frame", frame->entries));
1634         return de;
1635
1636 journal_error:
1637         brelse(*bh);
1638         brelse(bh2);
1639         *bh = NULL;
1640         ext4_std_error(dir->i_sb, err);
1641         *error = err;
1642         return NULL;
1643 }
1644
1645 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1646                       struct buffer_head *bh,
1647                       void *buf, int buf_size,
1648                       const char *name, int namelen,
1649                       struct ext4_dir_entry_2 **dest_de)
1650 {
1651         struct ext4_dir_entry_2 *de;
1652         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1653         int nlen, rlen;
1654         unsigned int offset = 0;
1655         char *top;
1656
1657         de = (struct ext4_dir_entry_2 *)buf;
1658         top = buf + buf_size - reclen;
1659         while ((char *) de <= top) {
1660                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1661                                          buf, buf_size, offset))
1662                         return -EIO;
1663                 if (ext4_match(namelen, name, de))
1664                         return -EEXIST;
1665                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1666                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1667                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1668                         break;
1669                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1670                 offset += rlen;
1671         }
1672         if ((char *) de > top)
1673                 return -ENOSPC;
1674
1675         *dest_de = de;
1676         return 0;
1677 }
1678
1679 void ext4_insert_dentry(struct inode *inode,
1680                         struct ext4_dir_entry_2 *de,
1681                         int buf_size,
1682                         const char *name, int namelen)
1683 {
1684
1685         int nlen, rlen;
1686
1687         nlen = EXT4_DIR_REC_LEN(de->name_len);
1688         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1689         if (de->inode) {
1690                 struct ext4_dir_entry_2 *de1 =
1691                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1692                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1693                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1694                 de = de1;
1695         }
1696         de->file_type = EXT4_FT_UNKNOWN;
1697         de->inode = cpu_to_le32(inode->i_ino);
1698         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1699         de->name_len = namelen;
1700         memcpy(de->name, name, namelen);
1701 }
1702 /*
1703  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1704  * it points to a directory entry which is guaranteed to be large
1705  * enough for new directory entry.  If de is NULL, then
1706  * add_dirent_to_buf will attempt search the directory block for
1707  * space.  It will return -ENOSPC if no space is available, and -EIO
1708  * and -EEXIST if directory entry already exists.
1709  */
1710 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1711                              struct inode *inode, struct ext4_dir_entry_2 *de,
1712                              struct buffer_head *bh)
1713 {
1714         struct inode    *dir = dentry->d_parent->d_inode;
1715         const char      *name = dentry->d_name.name;
1716         int             namelen = dentry->d_name.len;
1717         unsigned int    blocksize = dir->i_sb->s_blocksize;
1718         int             csum_size = 0;
1719         int             err;
1720
1721         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1722                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1723                 csum_size = sizeof(struct ext4_dir_entry_tail);
1724
1725         if (!de) {
1726                 err = ext4_find_dest_de(dir, inode,
1727                                         bh, bh->b_data, blocksize - csum_size,
1728                                         name, namelen, &de);
1729                 if (err)
1730                         return err;
1731         }
1732         BUFFER_TRACE(bh, "get_write_access");
1733         err = ext4_journal_get_write_access(handle, bh);
1734         if (err) {
1735                 ext4_std_error(dir->i_sb, err);
1736                 return err;
1737         }
1738
1739         /* By now the buffer is marked for journaling */
1740         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1741
1742         /*
1743          * XXX shouldn't update any times until successful
1744          * completion of syscall, but too many callers depend
1745          * on this.
1746          *
1747          * XXX similarly, too many callers depend on
1748          * ext4_new_inode() setting the times, but error
1749          * recovery deletes the inode, so the worst that can
1750          * happen is that the times are slightly out of date
1751          * and/or different from the directory change time.
1752          */
1753         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1754         ext4_update_dx_flag(dir);
1755         dir->i_version++;
1756         ext4_mark_inode_dirty(handle, dir);
1757         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1758         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1759         if (err)
1760                 ext4_std_error(dir->i_sb, err);
1761         return 0;
1762 }
1763
1764 /*
1765  * This converts a one block unindexed directory to a 3 block indexed
1766  * directory, and adds the dentry to the indexed directory.
1767  */
1768 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1769                             struct inode *inode, struct buffer_head *bh)
1770 {
1771         struct inode    *dir = dentry->d_parent->d_inode;
1772         const char      *name = dentry->d_name.name;
1773         int             namelen = dentry->d_name.len;
1774         struct buffer_head *bh2;
1775         struct dx_root  *root;
1776         struct dx_frame frames[2], *frame;
1777         struct dx_entry *entries;
1778         struct ext4_dir_entry_2 *de, *de2;
1779         struct ext4_dir_entry_tail *t;
1780         char            *data1, *top;
1781         unsigned        len;
1782         int             retval;
1783         unsigned        blocksize;
1784         struct dx_hash_info hinfo;
1785         ext4_lblk_t  block;
1786         struct fake_dirent *fde;
1787         int             csum_size = 0;
1788
1789         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1790                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1791                 csum_size = sizeof(struct ext4_dir_entry_tail);
1792
1793         blocksize =  dir->i_sb->s_blocksize;
1794         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1795         BUFFER_TRACE(bh, "get_write_access");
1796         retval = ext4_journal_get_write_access(handle, bh);
1797         if (retval) {
1798                 ext4_std_error(dir->i_sb, retval);
1799                 brelse(bh);
1800                 return retval;
1801         }
1802         root = (struct dx_root *) bh->b_data;
1803
1804         /* The 0th block becomes the root, move the dirents out */
1805         fde = &root->dotdot;
1806         de = (struct ext4_dir_entry_2 *)((char *)fde +
1807                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1808         if ((char *) de >= (((char *) root) + blocksize)) {
1809                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1810                 brelse(bh);
1811                 return -EIO;
1812         }
1813         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1814
1815         /* Allocate new block for the 0th block's dirents */
1816         bh2 = ext4_append(handle, dir, &block);
1817         if (IS_ERR(bh2)) {
1818                 brelse(bh);
1819                 return PTR_ERR(bh2);
1820         }
1821         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1822         data1 = bh2->b_data;
1823
1824         memcpy (data1, de, len);
1825         de = (struct ext4_dir_entry_2 *) data1;
1826         top = data1 + len;
1827         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1828                 de = de2;
1829         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1830                                            (char *) de,
1831                                            blocksize);
1832
1833         if (csum_size) {
1834                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1835                 initialize_dirent_tail(t, blocksize);
1836         }
1837
1838         /* Initialize the root; the dot dirents already exist */
1839         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1840         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1841                                            blocksize);
1842         memset (&root->info, 0, sizeof(root->info));
1843         root->info.info_length = sizeof(root->info);
1844         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1845         entries = root->entries;
1846         dx_set_block(entries, 1);
1847         dx_set_count(entries, 1);
1848         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1849
1850         /* Initialize as for dx_probe */
1851         hinfo.hash_version = root->info.hash_version;
1852         if (hinfo.hash_version <= DX_HASH_TEA)
1853                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1854         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1855         ext4fs_dirhash(name, namelen, &hinfo);
1856         frame = frames;
1857         frame->entries = entries;
1858         frame->at = entries;
1859         frame->bh = bh;
1860         bh = bh2;
1861
1862         ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1863         ext4_handle_dirty_dirent_node(handle, dir, bh);
1864
1865         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1866         if (!de) {
1867                 /*
1868                  * Even if the block split failed, we have to properly write
1869                  * out all the changes we did so far. Otherwise we can end up
1870                  * with corrupted filesystem.
1871                  */
1872                 ext4_mark_inode_dirty(handle, dir);
1873                 dx_release(frames);
1874                 return retval;
1875         }
1876         dx_release(frames);
1877
1878         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1879         brelse(bh);
1880         return retval;
1881 }
1882
1883 /*
1884  *      ext4_add_entry()
1885  *
1886  * adds a file entry to the specified directory, using the same
1887  * semantics as ext4_find_entry(). It returns NULL if it failed.
1888  *
1889  * NOTE!! The inode part of 'de' is left at 0 - which means you
1890  * may not sleep between calling this and putting something into
1891  * the entry, as someone else might have used it while you slept.
1892  */
1893 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1894                           struct inode *inode)
1895 {
1896         struct inode *dir = dentry->d_parent->d_inode;
1897         struct buffer_head *bh;
1898         struct ext4_dir_entry_2 *de;
1899         struct ext4_dir_entry_tail *t;
1900         struct super_block *sb;
1901         int     retval;
1902         int     dx_fallback=0;
1903         unsigned blocksize;
1904         ext4_lblk_t block, blocks;
1905         int     csum_size = 0;
1906
1907         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1908                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1909                 csum_size = sizeof(struct ext4_dir_entry_tail);
1910
1911         sb = dir->i_sb;
1912         blocksize = sb->s_blocksize;
1913         if (!dentry->d_name.len)
1914                 return -EINVAL;
1915
1916         if (ext4_has_inline_data(dir)) {
1917                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1918                 if (retval < 0)
1919                         return retval;
1920                 if (retval == 1) {
1921                         retval = 0;
1922                         return retval;
1923                 }
1924         }
1925
1926         if (is_dx(dir)) {
1927                 retval = ext4_dx_add_entry(handle, dentry, inode);
1928                 if (!retval || (retval != ERR_BAD_DX_DIR))
1929                         return retval;
1930                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1931                 dx_fallback++;
1932                 ext4_mark_inode_dirty(handle, dir);
1933         }
1934         blocks = dir->i_size >> sb->s_blocksize_bits;
1935         for (block = 0; block < blocks; block++) {
1936                 bh = ext4_read_dirblock(dir, block, DIRENT);
1937                 if (IS_ERR(bh))
1938                         return PTR_ERR(bh);
1939
1940                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1941                 if (retval != -ENOSPC) {
1942                         brelse(bh);
1943                         return retval;
1944                 }
1945
1946                 if (blocks == 1 && !dx_fallback &&
1947                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1948                         return make_indexed_dir(handle, dentry, inode, bh);
1949                 brelse(bh);
1950         }
1951         bh = ext4_append(handle, dir, &block);
1952         if (IS_ERR(bh))
1953                 return PTR_ERR(bh);
1954         de = (struct ext4_dir_entry_2 *) bh->b_data;
1955         de->inode = 0;
1956         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1957
1958         if (csum_size) {
1959                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1960                 initialize_dirent_tail(t, blocksize);
1961         }
1962
1963         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1964         brelse(bh);
1965         if (retval == 0)
1966                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1967         return retval;
1968 }
1969
1970 /*
1971  * Returns 0 for success, or a negative error value
1972  */
1973 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1974                              struct inode *inode)
1975 {
1976         struct dx_frame frames[2], *frame;
1977         struct dx_entry *entries, *at;
1978         struct dx_hash_info hinfo;
1979         struct buffer_head *bh;
1980         struct inode *dir = dentry->d_parent->d_inode;
1981         struct super_block *sb = dir->i_sb;
1982         struct ext4_dir_entry_2 *de;
1983         int err;
1984
1985         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1986         if (!frame)
1987                 return err;
1988         entries = frame->entries;
1989         at = frame->at;
1990         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1991         if (IS_ERR(bh)) {
1992                 err = PTR_ERR(bh);
1993                 bh = NULL;
1994                 goto cleanup;
1995         }
1996
1997         BUFFER_TRACE(bh, "get_write_access");
1998         err = ext4_journal_get_write_access(handle, bh);
1999         if (err)
2000                 goto journal_error;
2001
2002         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2003         if (err != -ENOSPC)
2004                 goto cleanup;
2005
2006         /* Block full, should compress but for now just split */
2007         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2008                        dx_get_count(entries), dx_get_limit(entries)));
2009         /* Need to split index? */
2010         if (dx_get_count(entries) == dx_get_limit(entries)) {
2011                 ext4_lblk_t newblock;
2012                 unsigned icount = dx_get_count(entries);
2013                 int levels = frame - frames;
2014                 struct dx_entry *entries2;
2015                 struct dx_node *node2;
2016                 struct buffer_head *bh2;
2017
2018                 if (levels && (dx_get_count(frames->entries) ==
2019                                dx_get_limit(frames->entries))) {
2020                         ext4_warning(sb, "Directory index full!");
2021                         err = -ENOSPC;
2022                         goto cleanup;
2023                 }
2024                 bh2 = ext4_append(handle, dir, &newblock);
2025                 if (IS_ERR(bh2)) {
2026                         err = PTR_ERR(bh2);
2027                         goto cleanup;
2028                 }
2029                 node2 = (struct dx_node *)(bh2->b_data);
2030                 entries2 = node2->entries;
2031                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2032                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2033                                                            sb->s_blocksize);
2034                 BUFFER_TRACE(frame->bh, "get_write_access");
2035                 err = ext4_journal_get_write_access(handle, frame->bh);
2036                 if (err)
2037                         goto journal_error;
2038                 if (levels) {
2039                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2040                         unsigned hash2 = dx_get_hash(entries + icount1);
2041                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2042                                        icount1, icount2));
2043
2044                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2045                         err = ext4_journal_get_write_access(handle,
2046                                                              frames[0].bh);
2047                         if (err)
2048                                 goto journal_error;
2049
2050                         memcpy((char *) entries2, (char *) (entries + icount1),
2051                                icount2 * sizeof(struct dx_entry));
2052                         dx_set_count(entries, icount1);
2053                         dx_set_count(entries2, icount2);
2054                         dx_set_limit(entries2, dx_node_limit(dir));
2055
2056                         /* Which index block gets the new entry? */
2057                         if (at - entries >= icount1) {
2058                                 frame->at = at = at - entries - icount1 + entries2;
2059                                 frame->entries = entries = entries2;
2060                                 swap(frame->bh, bh2);
2061                         }
2062                         dx_insert_block(frames + 0, hash2, newblock);
2063                         dxtrace(dx_show_index("node", frames[1].entries));
2064                         dxtrace(dx_show_index("node",
2065                                ((struct dx_node *) bh2->b_data)->entries));
2066                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2067                         if (err)
2068                                 goto journal_error;
2069                         brelse (bh2);
2070                 } else {
2071                         dxtrace(printk(KERN_DEBUG
2072                                        "Creating second level index...\n"));
2073                         memcpy((char *) entries2, (char *) entries,
2074                                icount * sizeof(struct dx_entry));
2075                         dx_set_limit(entries2, dx_node_limit(dir));
2076
2077                         /* Set up root */
2078                         dx_set_count(entries, 1);
2079                         dx_set_block(entries + 0, newblock);
2080                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2081
2082                         /* Add new access path frame */
2083                         frame = frames + 1;
2084                         frame->at = at = at - entries + entries2;
2085                         frame->entries = entries = entries2;
2086                         frame->bh = bh2;
2087                         err = ext4_journal_get_write_access(handle,
2088                                                              frame->bh);
2089                         if (err)
2090                                 goto journal_error;
2091                 }
2092                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2093                 if (err) {
2094                         ext4_std_error(inode->i_sb, err);
2095                         goto cleanup;
2096                 }
2097         }
2098         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2099         if (!de)
2100                 goto cleanup;
2101         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2102         goto cleanup;
2103
2104 journal_error:
2105         ext4_std_error(dir->i_sb, err);
2106 cleanup:
2107         brelse(bh);
2108         dx_release(frames);
2109         return err;
2110 }
2111
2112 /*
2113  * ext4_generic_delete_entry deletes a directory entry by merging it
2114  * with the previous entry
2115  */
2116 int ext4_generic_delete_entry(handle_t *handle,
2117                               struct inode *dir,
2118                               struct ext4_dir_entry_2 *de_del,
2119                               struct buffer_head *bh,
2120                               void *entry_buf,
2121                               int buf_size,
2122                               int csum_size)
2123 {
2124         struct ext4_dir_entry_2 *de, *pde;
2125         unsigned int blocksize = dir->i_sb->s_blocksize;
2126         int i;
2127
2128         i = 0;
2129         pde = NULL;
2130         de = (struct ext4_dir_entry_2 *)entry_buf;
2131         while (i < buf_size - csum_size) {
2132                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2133                                          bh->b_data, bh->b_size, i))
2134                         return -EIO;
2135                 if (de == de_del)  {
2136                         if (pde)
2137                                 pde->rec_len = ext4_rec_len_to_disk(
2138                                         ext4_rec_len_from_disk(pde->rec_len,
2139                                                                blocksize) +
2140                                         ext4_rec_len_from_disk(de->rec_len,
2141                                                                blocksize),
2142                                         blocksize);
2143                         else
2144                                 de->inode = 0;
2145                         dir->i_version++;
2146                         return 0;
2147                 }
2148                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2149                 pde = de;
2150                 de = ext4_next_entry(de, blocksize);
2151         }
2152         return -ENOENT;
2153 }
2154
2155 static int ext4_delete_entry(handle_t *handle,
2156                              struct inode *dir,
2157                              struct ext4_dir_entry_2 *de_del,
2158                              struct buffer_head *bh)
2159 {
2160         int err, csum_size = 0;
2161
2162         if (ext4_has_inline_data(dir)) {
2163                 int has_inline_data = 1;
2164                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2165                                                &has_inline_data);
2166                 if (has_inline_data)
2167                         return err;
2168         }
2169
2170         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2171                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2172                 csum_size = sizeof(struct ext4_dir_entry_tail);
2173
2174         BUFFER_TRACE(bh, "get_write_access");
2175         err = ext4_journal_get_write_access(handle, bh);
2176         if (unlikely(err))
2177                 goto out;
2178
2179         err = ext4_generic_delete_entry(handle, dir, de_del,
2180                                         bh, bh->b_data,
2181                                         dir->i_sb->s_blocksize, csum_size);
2182         if (err)
2183                 goto out;
2184
2185         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2186         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2187         if (unlikely(err))
2188                 goto out;
2189
2190         return 0;
2191 out:
2192         if (err != -ENOENT)
2193                 ext4_std_error(dir->i_sb, err);
2194         return err;
2195 }
2196
2197 /*
2198  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2199  * since this indicates that nlinks count was previously 1.
2200  */
2201 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2202 {
2203         inc_nlink(inode);
2204         if (is_dx(inode) && inode->i_nlink > 1) {
2205                 /* limit is 16-bit i_links_count */
2206                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2207                         set_nlink(inode, 1);
2208                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2209                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2210                 }
2211         }
2212 }
2213
2214 /*
2215  * If a directory had nlink == 1, then we should let it be 1. This indicates
2216  * directory has >EXT4_LINK_MAX subdirs.
2217  */
2218 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2219 {
2220         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2221                 drop_nlink(inode);
2222 }
2223
2224
2225 static int ext4_add_nondir(handle_t *handle,
2226                 struct dentry *dentry, struct inode *inode)
2227 {
2228         int err = ext4_add_entry(handle, dentry, inode);
2229         if (!err) {
2230                 ext4_mark_inode_dirty(handle, inode);
2231                 unlock_new_inode(inode);
2232                 d_instantiate(dentry, inode);
2233                 return 0;
2234         }
2235         drop_nlink(inode);
2236         unlock_new_inode(inode);
2237         iput(inode);
2238         return err;
2239 }
2240
2241 /*
2242  * By the time this is called, we already have created
2243  * the directory cache entry for the new file, but it
2244  * is so far negative - it has no inode.
2245  *
2246  * If the create succeeds, we fill in the inode information
2247  * with d_instantiate().
2248  */
2249 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2250                        bool excl)
2251 {
2252         handle_t *handle;
2253         struct inode *inode;
2254         int err, credits, retries = 0;
2255
2256         dquot_initialize(dir);
2257
2258         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2259                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2260 retry:
2261         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2262                                             NULL, EXT4_HT_DIR, credits);
2263         handle = ext4_journal_current_handle();
2264         err = PTR_ERR(inode);
2265         if (!IS_ERR(inode)) {
2266                 inode->i_op = &ext4_file_inode_operations;
2267                 inode->i_fop = &ext4_file_operations;
2268                 ext4_set_aops(inode);
2269                 err = ext4_add_nondir(handle, dentry, inode);
2270                 if (!err && IS_DIRSYNC(dir))
2271                         ext4_handle_sync(handle);
2272         }
2273         if (handle)
2274                 ext4_journal_stop(handle);
2275         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2276                 goto retry;
2277         return err;
2278 }
2279
2280 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2281                       umode_t mode, dev_t rdev)
2282 {
2283         handle_t *handle;
2284         struct inode *inode;
2285         int err, credits, retries = 0;
2286
2287         if (!new_valid_dev(rdev))
2288                 return -EINVAL;
2289
2290         dquot_initialize(dir);
2291
2292         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2293                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2294 retry:
2295         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2296                                             NULL, EXT4_HT_DIR, credits);
2297         handle = ext4_journal_current_handle();
2298         err = PTR_ERR(inode);
2299         if (!IS_ERR(inode)) {
2300                 init_special_inode(inode, inode->i_mode, rdev);
2301                 inode->i_op = &ext4_special_inode_operations;
2302                 err = ext4_add_nondir(handle, dentry, inode);
2303                 if (!err && IS_DIRSYNC(dir))
2304                         ext4_handle_sync(handle);
2305         }
2306         if (handle)
2307                 ext4_journal_stop(handle);
2308         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2309                 goto retry;
2310         return err;
2311 }
2312
2313 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2314 {
2315         handle_t *handle;
2316         struct inode *inode;
2317         int err, retries = 0;
2318
2319         dquot_initialize(dir);
2320
2321 retry:
2322         inode = ext4_new_inode_start_handle(dir, mode,
2323                                             NULL, 0, NULL,
2324                                             EXT4_HT_DIR,
2325                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2326                           4 + EXT4_XATTR_TRANS_BLOCKS);
2327         handle = ext4_journal_current_handle();
2328         err = PTR_ERR(inode);
2329         if (!IS_ERR(inode)) {
2330                 inode->i_op = &ext4_file_inode_operations;
2331                 inode->i_fop = &ext4_file_operations;
2332                 ext4_set_aops(inode);
2333                 d_tmpfile(dentry, inode);
2334                 err = ext4_orphan_add(handle, inode);
2335                 if (err)
2336                         goto err_unlock_inode;
2337                 mark_inode_dirty(inode);
2338                 unlock_new_inode(inode);
2339         }
2340         if (handle)
2341                 ext4_journal_stop(handle);
2342         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2343                 goto retry;
2344         return err;
2345 err_unlock_inode:
2346         ext4_journal_stop(handle);
2347         unlock_new_inode(inode);
2348         return err;
2349 }
2350
2351 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2352                           struct ext4_dir_entry_2 *de,
2353                           int blocksize, int csum_size,
2354                           unsigned int parent_ino, int dotdot_real_len)
2355 {
2356         de->inode = cpu_to_le32(inode->i_ino);
2357         de->name_len = 1;
2358         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2359                                            blocksize);
2360         strcpy(de->name, ".");
2361         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2362
2363         de = ext4_next_entry(de, blocksize);
2364         de->inode = cpu_to_le32(parent_ino);
2365         de->name_len = 2;
2366         if (!dotdot_real_len)
2367                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2368                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2369                                         blocksize);
2370         else
2371                 de->rec_len = ext4_rec_len_to_disk(
2372                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2373         strcpy(de->name, "..");
2374         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2375
2376         return ext4_next_entry(de, blocksize);
2377 }
2378
2379 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2380                              struct inode *inode)
2381 {
2382         struct buffer_head *dir_block = NULL;
2383         struct ext4_dir_entry_2 *de;
2384         struct ext4_dir_entry_tail *t;
2385         ext4_lblk_t block = 0;
2386         unsigned int blocksize = dir->i_sb->s_blocksize;
2387         int csum_size = 0;
2388         int err;
2389
2390         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2391                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2392                 csum_size = sizeof(struct ext4_dir_entry_tail);
2393
2394         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2395                 err = ext4_try_create_inline_dir(handle, dir, inode);
2396                 if (err < 0 && err != -ENOSPC)
2397                         goto out;
2398                 if (!err)
2399                         goto out;
2400         }
2401
2402         inode->i_size = 0;
2403         dir_block = ext4_append(handle, inode, &block);
2404         if (IS_ERR(dir_block))
2405                 return PTR_ERR(dir_block);
2406         BUFFER_TRACE(dir_block, "get_write_access");
2407         err = ext4_journal_get_write_access(handle, dir_block);
2408         if (err)
2409                 goto out;
2410         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2411         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2412         set_nlink(inode, 2);
2413         if (csum_size) {
2414                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2415                 initialize_dirent_tail(t, blocksize);
2416         }
2417
2418         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2419         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2420         if (err)
2421                 goto out;
2422         set_buffer_verified(dir_block);
2423 out:
2424         brelse(dir_block);
2425         return err;
2426 }
2427
2428 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2429 {
2430         handle_t *handle;
2431         struct inode *inode;
2432         int err, credits, retries = 0;
2433
2434         if (EXT4_DIR_LINK_MAX(dir))
2435                 return -EMLINK;
2436
2437         dquot_initialize(dir);
2438
2439         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2440                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2441 retry:
2442         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2443                                             &dentry->d_name,
2444                                             0, NULL, EXT4_HT_DIR, credits);
2445         handle = ext4_journal_current_handle();
2446         err = PTR_ERR(inode);
2447         if (IS_ERR(inode))
2448                 goto out_stop;
2449
2450         inode->i_op = &ext4_dir_inode_operations;
2451         inode->i_fop = &ext4_dir_operations;
2452         err = ext4_init_new_dir(handle, dir, inode);
2453         if (err)
2454                 goto out_clear_inode;
2455         err = ext4_mark_inode_dirty(handle, inode);
2456         if (!err)
2457                 err = ext4_add_entry(handle, dentry, inode);
2458         if (err) {
2459 out_clear_inode:
2460                 clear_nlink(inode);
2461                 unlock_new_inode(inode);
2462                 ext4_mark_inode_dirty(handle, inode);
2463                 iput(inode);
2464                 goto out_stop;
2465         }
2466         ext4_inc_count(handle, dir);
2467         ext4_update_dx_flag(dir);
2468         err = ext4_mark_inode_dirty(handle, dir);
2469         if (err)
2470                 goto out_clear_inode;
2471         unlock_new_inode(inode);
2472         d_instantiate(dentry, inode);
2473         if (IS_DIRSYNC(dir))
2474                 ext4_handle_sync(handle);
2475
2476 out_stop:
2477         if (handle)
2478                 ext4_journal_stop(handle);
2479         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2480                 goto retry;
2481         return err;
2482 }
2483
2484 /*
2485  * routine to check that the specified directory is empty (for rmdir)
2486  */
2487 static int empty_dir(struct inode *inode)
2488 {
2489         unsigned int offset;
2490         struct buffer_head *bh;
2491         struct ext4_dir_entry_2 *de, *de1;
2492         struct super_block *sb;
2493         int err = 0;
2494
2495         if (ext4_has_inline_data(inode)) {
2496                 int has_inline_data = 1;
2497
2498                 err = empty_inline_dir(inode, &has_inline_data);
2499                 if (has_inline_data)
2500                         return err;
2501         }
2502
2503         sb = inode->i_sb;
2504         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2505                 EXT4_ERROR_INODE(inode, "invalid size");
2506                 return 1;
2507         }
2508         bh = ext4_read_dirblock(inode, 0, EITHER);
2509         if (IS_ERR(bh))
2510                 return 1;
2511
2512         de = (struct ext4_dir_entry_2 *) bh->b_data;
2513         de1 = ext4_next_entry(de, sb->s_blocksize);
2514         if (le32_to_cpu(de->inode) != inode->i_ino ||
2515                         !le32_to_cpu(de1->inode) ||
2516                         strcmp(".", de->name) ||
2517                         strcmp("..", de1->name)) {
2518                 ext4_warning(inode->i_sb,
2519                              "bad directory (dir #%lu) - no `.' or `..'",
2520                              inode->i_ino);
2521                 brelse(bh);
2522                 return 1;
2523         }
2524         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2525                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2526         de = ext4_next_entry(de1, sb->s_blocksize);
2527         while (offset < inode->i_size) {
2528                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2529                         unsigned int lblock;
2530                         err = 0;
2531                         brelse(bh);
2532                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2533                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2534                         if (IS_ERR(bh))
2535                                 return 1;
2536                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2537                 }
2538                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2539                                          bh->b_data, bh->b_size, offset)) {
2540                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2541                                                          sb->s_blocksize);
2542                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2543                         continue;
2544                 }
2545                 if (le32_to_cpu(de->inode)) {
2546                         brelse(bh);
2547                         return 0;
2548                 }
2549                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2550                 de = ext4_next_entry(de, sb->s_blocksize);
2551         }
2552         brelse(bh);
2553         return 1;
2554 }
2555
2556 /*
2557  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2558  * such inodes, starting at the superblock, in case we crash before the
2559  * file is closed/deleted, or in case the inode truncate spans multiple
2560  * transactions and the last transaction is not recovered after a crash.
2561  *
2562  * At filesystem recovery time, we walk this list deleting unlinked
2563  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2564  *
2565  * Orphan list manipulation functions must be called under i_mutex unless
2566  * we are just creating the inode or deleting it.
2567  */
2568 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2569 {
2570         struct super_block *sb = inode->i_sb;
2571         struct ext4_sb_info *sbi = EXT4_SB(sb);
2572         struct ext4_iloc iloc;
2573         int err = 0, rc;
2574         bool dirty = false;
2575
2576         if (!sbi->s_journal)
2577                 return 0;
2578
2579         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2580                      !mutex_is_locked(&inode->i_mutex));
2581         /*
2582          * Exit early if inode already is on orphan list. This is a big speedup
2583          * since we don't have to contend on the global s_orphan_lock.
2584          */
2585         if (!list_empty(&EXT4_I(inode)->i_orphan))
2586                 return 0;
2587
2588         /*
2589          * Orphan handling is only valid for files with data blocks
2590          * being truncated, or files being unlinked. Note that we either
2591          * hold i_mutex, or the inode can not be referenced from outside,
2592          * so i_nlink should not be bumped due to race
2593          */
2594         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2595                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2596
2597         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2598         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2599         if (err)
2600                 goto out;
2601
2602         err = ext4_reserve_inode_write(handle, inode, &iloc);
2603         if (err)
2604                 goto out;
2605
2606         mutex_lock(&sbi->s_orphan_lock);
2607         /*
2608          * Due to previous errors inode may be already a part of on-disk
2609          * orphan list. If so skip on-disk list modification.
2610          */
2611         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2612             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2613                 /* Insert this inode at the head of the on-disk orphan list */
2614                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2615                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2616                 dirty = true;
2617         }
2618         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2619         mutex_unlock(&sbi->s_orphan_lock);
2620
2621         if (dirty) {
2622                 err = ext4_handle_dirty_super(handle, sb);
2623                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2624                 if (!err)
2625                         err = rc;
2626                 if (err) {
2627                         /*
2628                          * We have to remove inode from in-memory list if
2629                          * addition to on disk orphan list failed. Stray orphan
2630                          * list entries can cause panics at unmount time.
2631                          */
2632                         mutex_lock(&sbi->s_orphan_lock);
2633                         list_del(&EXT4_I(inode)->i_orphan);
2634                         mutex_unlock(&sbi->s_orphan_lock);
2635                 }
2636         }
2637         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2638         jbd_debug(4, "orphan inode %lu will point to %d\n",
2639                         inode->i_ino, NEXT_ORPHAN(inode));
2640 out:
2641         ext4_std_error(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 = EXT4_SB(inode->i_sb);
2654         __u32 ino_next;
2655         struct ext4_iloc iloc;
2656         int err = 0;
2657
2658         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2659                 return 0;
2660
2661         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2662                      !mutex_is_locked(&inode->i_mutex));
2663         /* Do this quick check before taking global s_orphan_lock. */
2664         if (list_empty(&ei->i_orphan))
2665                 return 0;
2666
2667         if (handle) {
2668                 /* Grab inode buffer early before taking global s_orphan_lock */
2669                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2670         }
2671
2672         mutex_lock(&sbi->s_orphan_lock);
2673         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2674
2675         prev = ei->i_orphan.prev;
2676         list_del_init(&ei->i_orphan);
2677
2678         /* If we're on an error path, we may not have a valid
2679          * transaction handle with which to update the orphan list on
2680          * disk, but we still need to remove the inode from the linked
2681          * list in memory. */
2682         if (!handle || err) {
2683                 mutex_unlock(&sbi->s_orphan_lock);
2684                 goto out_err;
2685         }
2686
2687         ino_next = NEXT_ORPHAN(inode);
2688         if (prev == &sbi->s_orphan) {
2689                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2690                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2691                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2692                 if (err) {
2693                         mutex_unlock(&sbi->s_orphan_lock);
2694                         goto out_brelse;
2695                 }
2696                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2697                 mutex_unlock(&sbi->s_orphan_lock);
2698                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2699         } else {
2700                 struct ext4_iloc iloc2;
2701                 struct inode *i_prev =
2702                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2703
2704                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2705                           i_prev->i_ino, ino_next);
2706                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2707                 if (err) {
2708                         mutex_unlock(&sbi->s_orphan_lock);
2709                         goto out_brelse;
2710                 }
2711                 NEXT_ORPHAN(i_prev) = ino_next;
2712                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2713                 mutex_unlock(&sbi->s_orphan_lock);
2714         }
2715         if (err)
2716                 goto out_brelse;
2717         NEXT_ORPHAN(inode) = 0;
2718         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2719 out_err:
2720         ext4_std_error(inode->i_sb, err);
2721         return err;
2722
2723 out_brelse:
2724         brelse(iloc.bh);
2725         goto out_err;
2726 }
2727
2728 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2729 {
2730         int retval;
2731         struct inode *inode;
2732         struct buffer_head *bh;
2733         struct ext4_dir_entry_2 *de;
2734         handle_t *handle = NULL;
2735
2736         /* Initialize quotas before so that eventual writes go in
2737          * separate transaction */
2738         dquot_initialize(dir);
2739         dquot_initialize(dentry->d_inode);
2740
2741         retval = -ENOENT;
2742         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2743         if (IS_ERR(bh))
2744                 return PTR_ERR(bh);
2745         if (!bh)
2746                 goto end_rmdir;
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         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2759                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2760         if (IS_ERR(handle)) {
2761                 retval = PTR_ERR(handle);
2762                 handle = NULL;
2763                 goto end_rmdir;
2764         }
2765
2766         if (IS_DIRSYNC(dir))
2767                 ext4_handle_sync(handle);
2768
2769         retval = ext4_delete_entry(handle, dir, de, bh);
2770         if (retval)
2771                 goto end_rmdir;
2772         if (!EXT4_DIR_LINK_EMPTY(inode))
2773                 ext4_warning(inode->i_sb,
2774                              "empty directory has too many links (%d)",
2775                              inode->i_nlink);
2776         inode->i_version++;
2777         clear_nlink(inode);
2778         /* There's no need to set i_disksize: the fact that i_nlink is
2779          * zero will ensure that the right thing happens during any
2780          * recovery. */
2781         inode->i_size = 0;
2782         ext4_orphan_add(handle, inode);
2783         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2784         ext4_mark_inode_dirty(handle, inode);
2785         ext4_dec_count(handle, dir);
2786         ext4_update_dx_flag(dir);
2787         ext4_mark_inode_dirty(handle, dir);
2788
2789 end_rmdir:
2790         brelse(bh);
2791         if (handle)
2792                 ext4_journal_stop(handle);
2793         return retval;
2794 }
2795
2796 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2797 {
2798         int retval;
2799         struct inode *inode;
2800         struct buffer_head *bh;
2801         struct ext4_dir_entry_2 *de;
2802         handle_t *handle = NULL;
2803
2804         trace_ext4_unlink_enter(dir, dentry);
2805         /* Initialize quotas before so that eventual writes go
2806          * in separate transaction */
2807         dquot_initialize(dir);
2808         dquot_initialize(dentry->d_inode);
2809
2810         retval = -ENOENT;
2811         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2812         if (IS_ERR(bh))
2813                 return PTR_ERR(bh);
2814         if (!bh)
2815                 goto end_unlink;
2816
2817         inode = dentry->d_inode;
2818
2819         retval = -EIO;
2820         if (le32_to_cpu(de->inode) != inode->i_ino)
2821                 goto end_unlink;
2822
2823         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2824                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2825         if (IS_ERR(handle)) {
2826                 retval = PTR_ERR(handle);
2827                 handle = NULL;
2828                 goto end_unlink;
2829         }
2830
2831         if (IS_DIRSYNC(dir))
2832                 ext4_handle_sync(handle);
2833
2834         if (!inode->i_nlink) {
2835                 ext4_warning(inode->i_sb,
2836                              "Deleting nonexistent file (%lu), %d",
2837                              inode->i_ino, inode->i_nlink);
2838                 set_nlink(inode, 1);
2839         }
2840         retval = ext4_delete_entry(handle, dir, de, bh);
2841         if (retval)
2842                 goto end_unlink;
2843         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2844         ext4_update_dx_flag(dir);
2845         ext4_mark_inode_dirty(handle, dir);
2846         drop_nlink(inode);
2847         if (!inode->i_nlink)
2848                 ext4_orphan_add(handle, inode);
2849         inode->i_ctime = ext4_current_time(inode);
2850         ext4_mark_inode_dirty(handle, inode);
2851         retval = 0;
2852
2853 end_unlink:
2854         brelse(bh);
2855         if (handle)
2856                 ext4_journal_stop(handle);
2857         trace_ext4_unlink_exit(dentry, retval);
2858         return retval;
2859 }
2860
2861 static int ext4_symlink(struct inode *dir,
2862                         struct dentry *dentry, const char *symname)
2863 {
2864         handle_t *handle;
2865         struct inode *inode;
2866         int l, err, retries = 0;
2867         int credits;
2868
2869         l = strlen(symname)+1;
2870         if (l > dir->i_sb->s_blocksize)
2871                 return -ENAMETOOLONG;
2872
2873         dquot_initialize(dir);
2874
2875         if (l > EXT4_N_BLOCKS * 4) {
2876                 /*
2877                  * For non-fast symlinks, we just allocate inode and put it on
2878                  * orphan list in the first transaction => we need bitmap,
2879                  * group descriptor, sb, inode block, quota blocks, and
2880                  * possibly selinux xattr blocks.
2881                  */
2882                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2883                           EXT4_XATTR_TRANS_BLOCKS;
2884         } else {
2885                 /*
2886                  * Fast symlink. We have to add entry to directory
2887                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2888                  * allocate new inode (bitmap, group descriptor, inode block,
2889                  * quota blocks, sb is already counted in previous macros).
2890                  */
2891                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2892                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2893         }
2894 retry:
2895         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2896                                             &dentry->d_name, 0, NULL,
2897                                             EXT4_HT_DIR, credits);
2898         handle = ext4_journal_current_handle();
2899         err = PTR_ERR(inode);
2900         if (IS_ERR(inode))
2901                 goto out_stop;
2902
2903         if (l > EXT4_N_BLOCKS * 4) {
2904                 inode->i_op = &ext4_symlink_inode_operations;
2905                 ext4_set_aops(inode);
2906                 /*
2907                  * We cannot call page_symlink() with transaction started
2908                  * because it calls into ext4_write_begin() which can wait
2909                  * for transaction commit if we are running out of space
2910                  * and thus we deadlock. So we have to stop transaction now
2911                  * and restart it when symlink contents is written.
2912                  *
2913                  * To keep fs consistent in case of crash, we have to put inode
2914                  * to orphan list in the mean time.
2915                  */
2916                 drop_nlink(inode);
2917                 err = ext4_orphan_add(handle, inode);
2918                 ext4_journal_stop(handle);
2919                 if (err)
2920                         goto err_drop_inode;
2921                 err = __page_symlink(inode, symname, l, 1);
2922                 if (err)
2923                         goto err_drop_inode;
2924                 /*
2925                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2926                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2927                  */
2928                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2929                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2930                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2931                 if (IS_ERR(handle)) {
2932                         err = PTR_ERR(handle);
2933                         goto err_drop_inode;
2934                 }
2935                 set_nlink(inode, 1);
2936                 err = ext4_orphan_del(handle, inode);
2937                 if (err) {
2938                         ext4_journal_stop(handle);
2939                         clear_nlink(inode);
2940                         goto err_drop_inode;
2941                 }
2942         } else {
2943                 /* clear the extent format for fast symlink */
2944                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2945                 inode->i_op = &ext4_fast_symlink_inode_operations;
2946                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2947                 inode->i_size = l-1;
2948         }
2949         EXT4_I(inode)->i_disksize = inode->i_size;
2950         err = ext4_add_nondir(handle, dentry, inode);
2951         if (!err && IS_DIRSYNC(dir))
2952                 ext4_handle_sync(handle);
2953
2954 out_stop:
2955         if (handle)
2956                 ext4_journal_stop(handle);
2957         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2958                 goto retry;
2959         return err;
2960 err_drop_inode:
2961         unlock_new_inode(inode);
2962         iput(inode);
2963         return err;
2964 }
2965
2966 static int ext4_link(struct dentry *old_dentry,
2967                      struct inode *dir, struct dentry *dentry)
2968 {
2969         handle_t *handle;
2970         struct inode *inode = old_dentry->d_inode;
2971         int err, retries = 0;
2972
2973         if (inode->i_nlink >= EXT4_LINK_MAX)
2974                 return -EMLINK;
2975
2976         dquot_initialize(dir);
2977
2978 retry:
2979         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2980                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2981                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2982         if (IS_ERR(handle))
2983                 return PTR_ERR(handle);
2984
2985         if (IS_DIRSYNC(dir))
2986                 ext4_handle_sync(handle);
2987
2988         inode->i_ctime = ext4_current_time(inode);
2989         ext4_inc_count(handle, inode);
2990         ihold(inode);
2991
2992         err = ext4_add_entry(handle, dentry, inode);
2993         if (!err) {
2994                 ext4_mark_inode_dirty(handle, inode);
2995                 /* this can happen only for tmpfile being
2996                  * linked the first time
2997                  */
2998                 if (inode->i_nlink == 1)
2999                         ext4_orphan_del(handle, inode);
3000                 d_instantiate(dentry, inode);
3001         } else {
3002                 drop_nlink(inode);
3003                 iput(inode);
3004         }
3005         ext4_journal_stop(handle);
3006         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3007                 goto retry;
3008         return err;
3009 }
3010
3011
3012 /*
3013  * Try to find buffer head where contains the parent block.
3014  * It should be the inode block if it is inlined or the 1st block
3015  * if it is a normal dir.
3016  */
3017 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3018                                         struct inode *inode,
3019                                         int *retval,
3020                                         struct ext4_dir_entry_2 **parent_de,
3021                                         int *inlined)
3022 {
3023         struct buffer_head *bh;
3024
3025         if (!ext4_has_inline_data(inode)) {
3026                 bh = ext4_read_dirblock(inode, 0, EITHER);
3027                 if (IS_ERR(bh)) {
3028                         *retval = PTR_ERR(bh);
3029                         return NULL;
3030                 }
3031                 *parent_de = ext4_next_entry(
3032                                         (struct ext4_dir_entry_2 *)bh->b_data,
3033                                         inode->i_sb->s_blocksize);
3034                 return bh;
3035         }
3036
3037         *inlined = 1;
3038         return ext4_get_first_inline_block(inode, parent_de, retval);
3039 }
3040
3041 struct ext4_renament {
3042         struct inode *dir;
3043         struct dentry *dentry;
3044         struct inode *inode;
3045         bool is_dir;
3046         int dir_nlink_delta;
3047
3048         /* entry for "dentry" */
3049         struct buffer_head *bh;
3050         struct ext4_dir_entry_2 *de;
3051         int inlined;
3052
3053         /* entry for ".." in inode if it's a directory */
3054         struct buffer_head *dir_bh;
3055         struct ext4_dir_entry_2 *parent_de;
3056         int dir_inlined;
3057 };
3058
3059 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3060 {
3061         int retval;
3062
3063         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3064                                               &retval, &ent->parent_de,
3065                                               &ent->dir_inlined);
3066         if (!ent->dir_bh)
3067                 return retval;
3068         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3069                 return -EIO;
3070         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3071         return ext4_journal_get_write_access(handle, ent->dir_bh);
3072 }
3073
3074 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3075                                   unsigned dir_ino)
3076 {
3077         int retval;
3078
3079         ent->parent_de->inode = cpu_to_le32(dir_ino);
3080         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3081         if (!ent->dir_inlined) {
3082                 if (is_dx(ent->inode)) {
3083                         retval = ext4_handle_dirty_dx_node(handle,
3084                                                            ent->inode,
3085                                                            ent->dir_bh);
3086                 } else {
3087                         retval = ext4_handle_dirty_dirent_node(handle,
3088                                                                ent->inode,
3089                                                                ent->dir_bh);
3090                 }
3091         } else {
3092                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3093         }
3094         if (retval) {
3095                 ext4_std_error(ent->dir->i_sb, retval);
3096                 return retval;
3097         }
3098         return 0;
3099 }
3100
3101 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3102                        unsigned ino, unsigned file_type)
3103 {
3104         int retval;
3105
3106         BUFFER_TRACE(ent->bh, "get write access");
3107         retval = ext4_journal_get_write_access(handle, ent->bh);
3108         if (retval)
3109                 return retval;
3110         ent->de->inode = cpu_to_le32(ino);
3111         if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3112                                       EXT4_FEATURE_INCOMPAT_FILETYPE))
3113                 ent->de->file_type = file_type;
3114         ent->dir->i_version++;
3115         ent->dir->i_ctime = ent->dir->i_mtime =
3116                 ext4_current_time(ent->dir);
3117         ext4_mark_inode_dirty(handle, ent->dir);
3118         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3119         if (!ent->inlined) {
3120                 retval = ext4_handle_dirty_dirent_node(handle,
3121                                                        ent->dir, ent->bh);
3122                 if (unlikely(retval)) {
3123                         ext4_std_error(ent->dir->i_sb, retval);
3124                         return retval;
3125                 }
3126         }
3127         brelse(ent->bh);
3128         ent->bh = NULL;
3129
3130         return 0;
3131 }
3132
3133 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3134                                   const struct qstr *d_name)
3135 {
3136         int retval = -ENOENT;
3137         struct buffer_head *bh;
3138         struct ext4_dir_entry_2 *de;
3139
3140         bh = ext4_find_entry(dir, d_name, &de, NULL);
3141         if (IS_ERR(bh))
3142                 return PTR_ERR(bh);
3143         if (bh) {
3144                 retval = ext4_delete_entry(handle, dir, de, bh);
3145                 brelse(bh);
3146         }
3147         return retval;
3148 }
3149
3150 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3151                                int force_reread)
3152 {
3153         int retval;
3154         /*
3155          * ent->de could have moved from under us during htree split, so make
3156          * sure that we are deleting the right entry.  We might also be pointing
3157          * to a stale entry in the unused part of ent->bh so just checking inum
3158          * and the name isn't enough.
3159          */
3160         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3161             ent->de->name_len != ent->dentry->d_name.len ||
3162             strncmp(ent->de->name, ent->dentry->d_name.name,
3163                     ent->de->name_len) ||
3164             force_reread) {
3165                 retval = ext4_find_delete_entry(handle, ent->dir,
3166                                                 &ent->dentry->d_name);
3167         } else {
3168                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3169                 if (retval == -ENOENT) {
3170                         retval = ext4_find_delete_entry(handle, ent->dir,
3171                                                         &ent->dentry->d_name);
3172                 }
3173         }
3174
3175         if (retval) {
3176                 ext4_warning(ent->dir->i_sb,
3177                                 "Deleting old file (%lu), %d, error=%d",
3178                                 ent->dir->i_ino, ent->dir->i_nlink, retval);
3179         }
3180 }
3181
3182 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3183 {
3184         if (ent->dir_nlink_delta) {
3185                 if (ent->dir_nlink_delta == -1)
3186                         ext4_dec_count(handle, ent->dir);
3187                 else
3188                         ext4_inc_count(handle, ent->dir);
3189                 ext4_mark_inode_dirty(handle, ent->dir);
3190         }
3191 }
3192
3193 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3194                                               int credits, handle_t **h)
3195 {
3196         struct inode *wh;
3197         handle_t *handle;
3198         int retries = 0;
3199
3200         /*
3201          * for inode block, sb block, group summaries,
3202          * and inode bitmap
3203          */
3204         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3205                     EXT4_XATTR_TRANS_BLOCKS + 4);
3206 retry:
3207         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3208                                          &ent->dentry->d_name, 0, NULL,
3209                                          EXT4_HT_DIR, credits);
3210
3211         handle = ext4_journal_current_handle();
3212         if (IS_ERR(wh)) {
3213                 if (handle)
3214                         ext4_journal_stop(handle);
3215                 if (PTR_ERR(wh) == -ENOSPC &&
3216                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3217                         goto retry;
3218         } else {
3219                 *h = handle;
3220                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3221                 wh->i_op = &ext4_special_inode_operations;
3222         }
3223         return wh;
3224 }
3225
3226 /*
3227  * Anybody can rename anything with this: the permission checks are left to the
3228  * higher-level routines.
3229  *
3230  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3231  * while new_{dentry,inode) refers to the destination dentry/inode
3232  * This comes from rename(const char *oldpath, const char *newpath)
3233  */
3234 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3235                        struct inode *new_dir, struct dentry *new_dentry,
3236                        unsigned int flags)
3237 {
3238         handle_t *handle = NULL;
3239         struct ext4_renament old = {
3240                 .dir = old_dir,
3241                 .dentry = old_dentry,
3242                 .inode = old_dentry->d_inode,
3243         };
3244         struct ext4_renament new = {
3245                 .dir = new_dir,
3246                 .dentry = new_dentry,
3247                 .inode = new_dentry->d_inode,
3248         };
3249         int force_reread;
3250         int retval;
3251         struct inode *whiteout = NULL;
3252         int credits;
3253         u8 old_file_type;
3254
3255         dquot_initialize(old.dir);
3256         dquot_initialize(new.dir);
3257
3258         /* Initialize quotas before so that eventual writes go
3259          * in separate transaction */
3260         if (new.inode)
3261                 dquot_initialize(new.inode);
3262
3263         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3264         if (IS_ERR(old.bh))
3265                 return PTR_ERR(old.bh);
3266         /*
3267          *  Check for inode number is _not_ due to possible IO errors.
3268          *  We might rmdir the source, keep it as pwd of some process
3269          *  and merrily kill the link to whatever was created under the
3270          *  same name. Goodbye sticky bit ;-<
3271          */
3272         retval = -ENOENT;
3273         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3274                 goto end_rename;
3275
3276         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3277                                  &new.de, &new.inlined);
3278         if (IS_ERR(new.bh)) {
3279                 retval = PTR_ERR(new.bh);
3280                 new.bh = NULL;
3281                 goto end_rename;
3282         }
3283         if (new.bh) {
3284                 if (!new.inode) {
3285                         brelse(new.bh);
3286                         new.bh = NULL;
3287                 }
3288         }
3289         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3290                 ext4_alloc_da_blocks(old.inode);
3291
3292         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3293                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3294         if (!(flags & RENAME_WHITEOUT)) {
3295                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3296                 if (IS_ERR(handle))
3297                         return PTR_ERR(handle);
3298         } else {
3299                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3300                 if (IS_ERR(whiteout))
3301                         return PTR_ERR(whiteout);
3302         }
3303
3304         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3305                 ext4_handle_sync(handle);
3306
3307         if (S_ISDIR(old.inode->i_mode)) {
3308                 if (new.inode) {
3309                         retval = -ENOTEMPTY;
3310                         if (!empty_dir(new.inode))
3311                                 goto end_rename;
3312                 } else {
3313                         retval = -EMLINK;
3314                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3315                                 goto end_rename;
3316                 }
3317                 retval = ext4_rename_dir_prepare(handle, &old);
3318                 if (retval)
3319                         goto end_rename;
3320         }
3321         /*
3322          * If we're renaming a file within an inline_data dir and adding or
3323          * setting the new dirent causes a conversion from inline_data to
3324          * extents/blockmap, we need to force the dirent delete code to
3325          * re-read the directory, or else we end up trying to delete a dirent
3326          * from what is now the extent tree root (or a block map).
3327          */
3328         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3329                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3330
3331         old_file_type = old.de->file_type;
3332         if (whiteout) {
3333                 /*
3334                  * Do this before adding a new entry, so the old entry is sure
3335                  * to be still pointing to the valid old entry.
3336                  */
3337                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3338                                      EXT4_FT_CHRDEV);
3339                 if (retval)
3340                         goto end_rename;
3341                 ext4_mark_inode_dirty(handle, whiteout);
3342         }
3343         if (!new.bh) {
3344                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3345                 if (retval)
3346                         goto end_rename;
3347         } else {
3348                 retval = ext4_setent(handle, &new,
3349                                      old.inode->i_ino, old_file_type);
3350                 if (retval)
3351                         goto end_rename;
3352         }
3353         if (force_reread)
3354                 force_reread = !ext4_test_inode_flag(new.dir,
3355                                                      EXT4_INODE_INLINE_DATA);
3356
3357         /*
3358          * Like most other Unix systems, set the ctime for inodes on a
3359          * rename.
3360          */
3361         old.inode->i_ctime = ext4_current_time(old.inode);
3362         ext4_mark_inode_dirty(handle, old.inode);
3363
3364         if (!whiteout) {
3365                 /*
3366                  * ok, that's it
3367                  */
3368                 ext4_rename_delete(handle, &old, force_reread);
3369         }
3370
3371         if (new.inode) {
3372                 ext4_dec_count(handle, new.inode);
3373                 new.inode->i_ctime = ext4_current_time(new.inode);
3374         }
3375         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3376         ext4_update_dx_flag(old.dir);
3377         if (old.dir_bh) {
3378                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3379                 if (retval)
3380                         goto end_rename;
3381
3382                 ext4_dec_count(handle, old.dir);
3383                 if (new.inode) {
3384                         /* checked empty_dir above, can't have another parent,
3385                          * ext4_dec_count() won't work for many-linked dirs */
3386                         clear_nlink(new.inode);
3387                 } else {
3388                         ext4_inc_count(handle, new.dir);
3389                         ext4_update_dx_flag(new.dir);
3390                         ext4_mark_inode_dirty(handle, new.dir);
3391                 }
3392         }
3393         ext4_mark_inode_dirty(handle, old.dir);
3394         if (new.inode) {
3395                 ext4_mark_inode_dirty(handle, new.inode);
3396                 if (!new.inode->i_nlink)
3397                         ext4_orphan_add(handle, new.inode);
3398         }
3399         retval = 0;
3400
3401 end_rename:
3402         brelse(old.dir_bh);
3403         brelse(old.bh);
3404         brelse(new.bh);
3405         if (whiteout) {
3406                 if (retval)
3407                         drop_nlink(whiteout);
3408                 unlock_new_inode(whiteout);
3409                 iput(whiteout);
3410         }
3411         if (handle)
3412                 ext4_journal_stop(handle);
3413         return retval;
3414 }
3415
3416 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3417                              struct inode *new_dir, struct dentry *new_dentry)
3418 {
3419         handle_t *handle = NULL;
3420         struct ext4_renament old = {
3421                 .dir = old_dir,
3422                 .dentry = old_dentry,
3423                 .inode = old_dentry->d_inode,
3424         };
3425         struct ext4_renament new = {
3426                 .dir = new_dir,
3427                 .dentry = new_dentry,
3428                 .inode = new_dentry->d_inode,
3429         };
3430         u8 new_file_type;
3431         int retval;
3432
3433         dquot_initialize(old.dir);
3434         dquot_initialize(new.dir);
3435
3436         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3437                                  &old.de, &old.inlined);
3438         if (IS_ERR(old.bh))
3439                 return PTR_ERR(old.bh);
3440         /*
3441          *  Check for inode number is _not_ due to possible IO errors.
3442          *  We might rmdir the source, keep it as pwd of some process
3443          *  and merrily kill the link to whatever was created under the
3444          *  same name. Goodbye sticky bit ;-<
3445          */
3446         retval = -ENOENT;
3447         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3448                 goto end_rename;
3449
3450         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3451                                  &new.de, &new.inlined);
3452         if (IS_ERR(new.bh)) {
3453                 retval = PTR_ERR(new.bh);
3454                 new.bh = NULL;
3455                 goto end_rename;
3456         }
3457
3458         /* RENAME_EXCHANGE case: old *and* new must both exist */
3459         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3460                 goto end_rename;
3461
3462         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3463                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3464                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3465         if (IS_ERR(handle))
3466                 return PTR_ERR(handle);
3467
3468         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3469                 ext4_handle_sync(handle);
3470
3471         if (S_ISDIR(old.inode->i_mode)) {
3472                 old.is_dir = true;
3473                 retval = ext4_rename_dir_prepare(handle, &old);
3474                 if (retval)
3475                         goto end_rename;
3476         }
3477         if (S_ISDIR(new.inode->i_mode)) {
3478                 new.is_dir = true;
3479                 retval = ext4_rename_dir_prepare(handle, &new);
3480                 if (retval)
3481                         goto end_rename;
3482         }
3483
3484         /*
3485          * Other than the special case of overwriting a directory, parents'
3486          * nlink only needs to be modified if this is a cross directory rename.
3487          */
3488         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3489                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3490                 new.dir_nlink_delta = -old.dir_nlink_delta;
3491                 retval = -EMLINK;
3492                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3493                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3494                         goto end_rename;
3495         }
3496
3497         new_file_type = new.de->file_type;
3498         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3499         if (retval)
3500                 goto end_rename;
3501
3502         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3503         if (retval)
3504                 goto end_rename;
3505
3506         /*
3507          * Like most other Unix systems, set the ctime for inodes on a
3508          * rename.
3509          */
3510         old.inode->i_ctime = ext4_current_time(old.inode);
3511         new.inode->i_ctime = ext4_current_time(new.inode);
3512         ext4_mark_inode_dirty(handle, old.inode);
3513         ext4_mark_inode_dirty(handle, new.inode);
3514
3515         if (old.dir_bh) {
3516                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3517                 if (retval)
3518                         goto end_rename;
3519         }
3520         if (new.dir_bh) {
3521                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3522                 if (retval)
3523                         goto end_rename;
3524         }
3525         ext4_update_dir_count(handle, &old);
3526         ext4_update_dir_count(handle, &new);
3527         retval = 0;
3528
3529 end_rename:
3530         brelse(old.dir_bh);
3531         brelse(new.dir_bh);
3532         brelse(old.bh);
3533         brelse(new.bh);
3534         if (handle)
3535                 ext4_journal_stop(handle);
3536         return retval;
3537 }
3538
3539 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3540                         struct inode *new_dir, struct dentry *new_dentry,
3541                         unsigned int flags)
3542 {
3543         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3544                 return -EINVAL;
3545
3546         if (flags & RENAME_EXCHANGE) {
3547                 return ext4_cross_rename(old_dir, old_dentry,
3548                                          new_dir, new_dentry);
3549         }
3550
3551         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3552 }
3553
3554 /*
3555  * directories can handle most operations...
3556  */
3557 const struct inode_operations ext4_dir_inode_operations = {
3558         .create         = ext4_create,
3559         .lookup         = ext4_lookup,
3560         .link           = ext4_link,
3561         .unlink         = ext4_unlink,
3562         .symlink        = ext4_symlink,
3563         .mkdir          = ext4_mkdir,
3564         .rmdir          = ext4_rmdir,
3565         .mknod          = ext4_mknod,
3566         .tmpfile        = ext4_tmpfile,
3567         .rename2        = ext4_rename2,
3568         .setattr        = ext4_setattr,
3569         .setxattr       = generic_setxattr,
3570         .getxattr       = generic_getxattr,
3571         .listxattr      = ext4_listxattr,
3572         .removexattr    = generic_removexattr,
3573         .get_acl        = ext4_get_acl,
3574         .set_acl        = ext4_set_acl,
3575         .fiemap         = ext4_fiemap,
3576 };
3577
3578 const struct inode_operations ext4_special_inode_operations = {
3579         .setattr        = ext4_setattr,
3580         .setxattr       = generic_setxattr,
3581         .getxattr       = generic_getxattr,
3582         .listxattr      = ext4_listxattr,
3583         .removexattr    = generic_removexattr,
3584         .get_acl        = ext4_get_acl,
3585         .set_acl        = ext4_set_acl,
3586 };