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;
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 (bh || (err != ERR_BAD_DX_DIR))
1268                         return bh;
1269                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1270                                "falling back\n"));
1271         }
1272         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1273         start = EXT4_I(dir)->i_dir_start_lookup;
1274         if (start >= nblocks)
1275                 start = 0;
1276         block = start;
1277 restart:
1278         do {
1279                 /*
1280                  * We deal with the read-ahead logic here.
1281                  */
1282                 if (ra_ptr >= ra_max) {
1283                         /* Refill the readahead buffer */
1284                         ra_ptr = 0;
1285                         b = block;
1286                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1287                                 /*
1288                                  * Terminate if we reach the end of the
1289                                  * directory and must wrap, or if our
1290                                  * search has finished at this block.
1291                                  */
1292                                 if (b >= nblocks || (num && block == start)) {
1293                                         bh_use[ra_max] = NULL;
1294                                         break;
1295                                 }
1296                                 num++;
1297                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1298                                 bh_use[ra_max] = bh;
1299                                 if (bh)
1300                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1301                                                     1, &bh);
1302                         }
1303                 }
1304                 if ((bh = bh_use[ra_ptr++]) == NULL)
1305                         goto next;
1306                 wait_on_buffer(bh);
1307                 if (!buffer_uptodate(bh)) {
1308                         /* read error, skip block & hope for the best */
1309                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1310                                          (unsigned long) block);
1311                         brelse(bh);
1312                         goto next;
1313                 }
1314                 if (!buffer_verified(bh) &&
1315                     !is_dx_internal_node(dir, block,
1316                                          (struct ext4_dir_entry *)bh->b_data) &&
1317                     !ext4_dirent_csum_verify(dir,
1318                                 (struct ext4_dir_entry *)bh->b_data)) {
1319                         EXT4_ERROR_INODE(dir, "checksumming directory "
1320                                          "block %lu", (unsigned long)block);
1321                         brelse(bh);
1322                         goto next;
1323                 }
1324                 set_buffer_verified(bh);
1325                 i = search_dirblock(bh, dir, d_name,
1326                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1327                 if (i == 1) {
1328                         EXT4_I(dir)->i_dir_start_lookup = block;
1329                         ret = bh;
1330                         goto cleanup_and_exit;
1331                 } else {
1332                         brelse(bh);
1333                         if (i < 0)
1334                                 goto cleanup_and_exit;
1335                 }
1336         next:
1337                 if (++block >= nblocks)
1338                         block = 0;
1339         } while (block != start);
1340
1341         /*
1342          * If the directory has grown while we were searching, then
1343          * search the last part of the directory before giving up.
1344          */
1345         block = nblocks;
1346         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1347         if (block < nblocks) {
1348                 start = 0;
1349                 goto restart;
1350         }
1351
1352 cleanup_and_exit:
1353         /* Clean up the read-ahead blocks */
1354         for (; ra_ptr < ra_max; ra_ptr++)
1355                 brelse(bh_use[ra_ptr]);
1356         return ret;
1357 }
1358
1359 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1360                        struct ext4_dir_entry_2 **res_dir, int *err)
1361 {
1362         struct super_block * sb = dir->i_sb;
1363         struct dx_hash_info     hinfo;
1364         struct dx_frame frames[2], *frame;
1365         struct buffer_head *bh;
1366         ext4_lblk_t block;
1367         int retval;
1368
1369         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1370                 return NULL;
1371         do {
1372                 block = dx_get_block(frame->at);
1373                 bh = ext4_read_dirblock(dir, block, DIRENT);
1374                 if (IS_ERR(bh)) {
1375                         *err = PTR_ERR(bh);
1376                         goto errout;
1377                 }
1378                 retval = search_dirblock(bh, dir, d_name,
1379                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1380                                          res_dir);
1381                 if (retval == 1) {      /* Success! */
1382                         dx_release(frames);
1383                         return bh;
1384                 }
1385                 brelse(bh);
1386                 if (retval == -1) {
1387                         *err = ERR_BAD_DX_DIR;
1388                         goto errout;
1389                 }
1390
1391                 /* Check to see if we should continue to search */
1392                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1393                                                frames, NULL);
1394                 if (retval < 0) {
1395                         ext4_warning(sb,
1396                              "error reading index page in directory #%lu",
1397                              dir->i_ino);
1398                         *err = retval;
1399                         goto errout;
1400                 }
1401         } while (retval == 1);
1402
1403         *err = -ENOENT;
1404 errout:
1405         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1406         dx_release (frames);
1407         return NULL;
1408 }
1409
1410 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1411 {
1412         struct inode *inode;
1413         struct ext4_dir_entry_2 *de;
1414         struct buffer_head *bh;
1415
1416         if (dentry->d_name.len > EXT4_NAME_LEN)
1417                 return ERR_PTR(-ENAMETOOLONG);
1418
1419         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1420         inode = NULL;
1421         if (bh) {
1422                 __u32 ino = le32_to_cpu(de->inode);
1423                 brelse(bh);
1424                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1425                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1426                         return ERR_PTR(-EIO);
1427                 }
1428                 if (unlikely(ino == dir->i_ino)) {
1429                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1430                                          dentry);
1431                         return ERR_PTR(-EIO);
1432                 }
1433                 inode = ext4_iget(dir->i_sb, ino);
1434                 if (inode == ERR_PTR(-ESTALE)) {
1435                         EXT4_ERROR_INODE(dir,
1436                                          "deleted inode referenced: %u",
1437                                          ino);
1438                         return ERR_PTR(-EIO);
1439                 }
1440         }
1441         return d_splice_alias(inode, dentry);
1442 }
1443
1444
1445 struct dentry *ext4_get_parent(struct dentry *child)
1446 {
1447         __u32 ino;
1448         static const struct qstr dotdot = QSTR_INIT("..", 2);
1449         struct ext4_dir_entry_2 * de;
1450         struct buffer_head *bh;
1451
1452         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1453         if (!bh)
1454                 return ERR_PTR(-ENOENT);
1455         ino = le32_to_cpu(de->inode);
1456         brelse(bh);
1457
1458         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1459                 EXT4_ERROR_INODE(child->d_inode,
1460                                  "bad parent inode number: %u", ino);
1461                 return ERR_PTR(-EIO);
1462         }
1463
1464         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1465 }
1466
1467 /*
1468  * Move count entries from end of map between two memory locations.
1469  * Returns pointer to last entry moved.
1470  */
1471 static struct ext4_dir_entry_2 *
1472 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1473                 unsigned blocksize)
1474 {
1475         unsigned rec_len = 0;
1476
1477         while (count--) {
1478                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1479                                                 (from + (map->offs<<2));
1480                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1481                 memcpy (to, de, rec_len);
1482                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1483                                 ext4_rec_len_to_disk(rec_len, blocksize);
1484                 de->inode = 0;
1485                 map++;
1486                 to += rec_len;
1487         }
1488         return (struct ext4_dir_entry_2 *) (to - rec_len);
1489 }
1490
1491 /*
1492  * Compact each dir entry in the range to the minimal rec_len.
1493  * Returns pointer to last entry in range.
1494  */
1495 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1496 {
1497         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1498         unsigned rec_len = 0;
1499
1500         prev = to = de;
1501         while ((char*)de < base + blocksize) {
1502                 next = ext4_next_entry(de, blocksize);
1503                 if (de->inode && de->name_len) {
1504                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1505                         if (de > to)
1506                                 memmove(to, de, rec_len);
1507                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1508                         prev = to;
1509                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1510                 }
1511                 de = next;
1512         }
1513         return prev;
1514 }
1515
1516 /*
1517  * Split a full leaf block to make room for a new dir entry.
1518  * Allocate a new block, and move entries so that they are approx. equally full.
1519  * Returns pointer to de in block into which the new entry will be inserted.
1520  */
1521 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1522                         struct buffer_head **bh,struct dx_frame *frame,
1523                         struct dx_hash_info *hinfo, int *error)
1524 {
1525         unsigned blocksize = dir->i_sb->s_blocksize;
1526         unsigned count, continued;
1527         struct buffer_head *bh2;
1528         ext4_lblk_t newblock;
1529         u32 hash2;
1530         struct dx_map_entry *map;
1531         char *data1 = (*bh)->b_data, *data2;
1532         unsigned split, move, size;
1533         struct ext4_dir_entry_2 *de = NULL, *de2;
1534         struct ext4_dir_entry_tail *t;
1535         int     csum_size = 0;
1536         int     err = 0, i;
1537
1538         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1539                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1540                 csum_size = sizeof(struct ext4_dir_entry_tail);
1541
1542         bh2 = ext4_append(handle, dir, &newblock);
1543         if (IS_ERR(bh2)) {
1544                 brelse(*bh);
1545                 *bh = NULL;
1546                 *error = PTR_ERR(bh2);
1547                 return NULL;
1548         }
1549
1550         BUFFER_TRACE(*bh, "get_write_access");
1551         err = ext4_journal_get_write_access(handle, *bh);
1552         if (err)
1553                 goto journal_error;
1554
1555         BUFFER_TRACE(frame->bh, "get_write_access");
1556         err = ext4_journal_get_write_access(handle, frame->bh);
1557         if (err)
1558                 goto journal_error;
1559
1560         data2 = bh2->b_data;
1561
1562         /* create map in the end of data2 block */
1563         map = (struct dx_map_entry *) (data2 + blocksize);
1564         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1565                              blocksize, hinfo, map);
1566         map -= count;
1567         dx_sort_map(map, count);
1568         /* Split the existing block in the middle, size-wise */
1569         size = 0;
1570         move = 0;
1571         for (i = count-1; i >= 0; i--) {
1572                 /* is more than half of this entry in 2nd half of the block? */
1573                 if (size + map[i].size/2 > blocksize/2)
1574                         break;
1575                 size += map[i].size;
1576                 move++;
1577         }
1578         /* map index at which we will split */
1579         split = count - move;
1580         hash2 = map[split].hash;
1581         continued = hash2 == map[split - 1].hash;
1582         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1583                         (unsigned long)dx_get_block(frame->at),
1584                                         hash2, split, count-split));
1585
1586         /* Fancy dance to stay within two buffers */
1587         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1588         de = dx_pack_dirents(data1, blocksize);
1589         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1590                                            (char *) de,
1591                                            blocksize);
1592         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1593                                             (char *) de2,
1594                                             blocksize);
1595         if (csum_size) {
1596                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1597                 initialize_dirent_tail(t, blocksize);
1598
1599                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1600                 initialize_dirent_tail(t, blocksize);
1601         }
1602
1603         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1604         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1605
1606         /* Which block gets the new entry? */
1607         if (hinfo->hash >= hash2)
1608         {
1609                 swap(*bh, bh2);
1610                 de = de2;
1611         }
1612         dx_insert_block(frame, hash2 + continued, newblock);
1613         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1614         if (err)
1615                 goto journal_error;
1616         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1617         if (err)
1618                 goto journal_error;
1619         brelse(bh2);
1620         dxtrace(dx_show_index("frame", frame->entries));
1621         return de;
1622
1623 journal_error:
1624         brelse(*bh);
1625         brelse(bh2);
1626         *bh = NULL;
1627         ext4_std_error(dir->i_sb, err);
1628         *error = err;
1629         return NULL;
1630 }
1631
1632 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1633                       struct buffer_head *bh,
1634                       void *buf, int buf_size,
1635                       const char *name, int namelen,
1636                       struct ext4_dir_entry_2 **dest_de)
1637 {
1638         struct ext4_dir_entry_2 *de;
1639         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1640         int nlen, rlen;
1641         unsigned int offset = 0;
1642         char *top;
1643
1644         de = (struct ext4_dir_entry_2 *)buf;
1645         top = buf + buf_size - reclen;
1646         while ((char *) de <= top) {
1647                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1648                                          buf, buf_size, offset))
1649                         return -EIO;
1650                 if (ext4_match(namelen, name, de))
1651                         return -EEXIST;
1652                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1653                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1654                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1655                         break;
1656                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1657                 offset += rlen;
1658         }
1659         if ((char *) de > top)
1660                 return -ENOSPC;
1661
1662         *dest_de = de;
1663         return 0;
1664 }
1665
1666 void ext4_insert_dentry(struct inode *inode,
1667                         struct ext4_dir_entry_2 *de,
1668                         int buf_size,
1669                         const char *name, int namelen)
1670 {
1671
1672         int nlen, rlen;
1673
1674         nlen = EXT4_DIR_REC_LEN(de->name_len);
1675         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1676         if (de->inode) {
1677                 struct ext4_dir_entry_2 *de1 =
1678                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1679                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1680                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1681                 de = de1;
1682         }
1683         de->file_type = EXT4_FT_UNKNOWN;
1684         de->inode = cpu_to_le32(inode->i_ino);
1685         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1686         de->name_len = namelen;
1687         memcpy(de->name, name, namelen);
1688 }
1689 /*
1690  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1691  * it points to a directory entry which is guaranteed to be large
1692  * enough for new directory entry.  If de is NULL, then
1693  * add_dirent_to_buf will attempt search the directory block for
1694  * space.  It will return -ENOSPC if no space is available, and -EIO
1695  * and -EEXIST if directory entry already exists.
1696  */
1697 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1698                              struct inode *inode, struct ext4_dir_entry_2 *de,
1699                              struct buffer_head *bh)
1700 {
1701         struct inode    *dir = dentry->d_parent->d_inode;
1702         const char      *name = dentry->d_name.name;
1703         int             namelen = dentry->d_name.len;
1704         unsigned int    blocksize = dir->i_sb->s_blocksize;
1705         int             csum_size = 0;
1706         int             err;
1707
1708         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1709                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1710                 csum_size = sizeof(struct ext4_dir_entry_tail);
1711
1712         if (!de) {
1713                 err = ext4_find_dest_de(dir, inode,
1714                                         bh, bh->b_data, blocksize - csum_size,
1715                                         name, namelen, &de);
1716                 if (err)
1717                         return err;
1718         }
1719         BUFFER_TRACE(bh, "get_write_access");
1720         err = ext4_journal_get_write_access(handle, bh);
1721         if (err) {
1722                 ext4_std_error(dir->i_sb, err);
1723                 return err;
1724         }
1725
1726         /* By now the buffer is marked for journaling */
1727         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1728
1729         /*
1730          * XXX shouldn't update any times until successful
1731          * completion of syscall, but too many callers depend
1732          * on this.
1733          *
1734          * XXX similarly, too many callers depend on
1735          * ext4_new_inode() setting the times, but error
1736          * recovery deletes the inode, so the worst that can
1737          * happen is that the times are slightly out of date
1738          * and/or different from the directory change time.
1739          */
1740         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1741         ext4_update_dx_flag(dir);
1742         dir->i_version++;
1743         ext4_mark_inode_dirty(handle, dir);
1744         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1745         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1746         if (err)
1747                 ext4_std_error(dir->i_sb, err);
1748         return 0;
1749 }
1750
1751 /*
1752  * This converts a one block unindexed directory to a 3 block indexed
1753  * directory, and adds the dentry to the indexed directory.
1754  */
1755 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1756                             struct inode *inode, struct buffer_head *bh)
1757 {
1758         struct inode    *dir = dentry->d_parent->d_inode;
1759         const char      *name = dentry->d_name.name;
1760         int             namelen = dentry->d_name.len;
1761         struct buffer_head *bh2;
1762         struct dx_root  *root;
1763         struct dx_frame frames[2], *frame;
1764         struct dx_entry *entries;
1765         struct ext4_dir_entry_2 *de, *de2;
1766         struct ext4_dir_entry_tail *t;
1767         char            *data1, *top;
1768         unsigned        len;
1769         int             retval;
1770         unsigned        blocksize;
1771         struct dx_hash_info hinfo;
1772         ext4_lblk_t  block;
1773         struct fake_dirent *fde;
1774         int             csum_size = 0;
1775
1776         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1777                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1778                 csum_size = sizeof(struct ext4_dir_entry_tail);
1779
1780         blocksize =  dir->i_sb->s_blocksize;
1781         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1782         BUFFER_TRACE(bh, "get_write_access");
1783         retval = ext4_journal_get_write_access(handle, bh);
1784         if (retval) {
1785                 ext4_std_error(dir->i_sb, retval);
1786                 brelse(bh);
1787                 return retval;
1788         }
1789         root = (struct dx_root *) bh->b_data;
1790
1791         /* The 0th block becomes the root, move the dirents out */
1792         fde = &root->dotdot;
1793         de = (struct ext4_dir_entry_2 *)((char *)fde +
1794                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1795         if ((char *) de >= (((char *) root) + blocksize)) {
1796                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1797                 brelse(bh);
1798                 return -EIO;
1799         }
1800         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1801
1802         /* Allocate new block for the 0th block's dirents */
1803         bh2 = ext4_append(handle, dir, &block);
1804         if (IS_ERR(bh2)) {
1805                 brelse(bh);
1806                 return PTR_ERR(bh2);
1807         }
1808         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1809         data1 = bh2->b_data;
1810
1811         memcpy (data1, de, len);
1812         de = (struct ext4_dir_entry_2 *) data1;
1813         top = data1 + len;
1814         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1815                 de = de2;
1816         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1817                                            (char *) de,
1818                                            blocksize);
1819
1820         if (csum_size) {
1821                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1822                 initialize_dirent_tail(t, blocksize);
1823         }
1824
1825         /* Initialize the root; the dot dirents already exist */
1826         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1827         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1828                                            blocksize);
1829         memset (&root->info, 0, sizeof(root->info));
1830         root->info.info_length = sizeof(root->info);
1831         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1832         entries = root->entries;
1833         dx_set_block(entries, 1);
1834         dx_set_count(entries, 1);
1835         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1836
1837         /* Initialize as for dx_probe */
1838         hinfo.hash_version = root->info.hash_version;
1839         if (hinfo.hash_version <= DX_HASH_TEA)
1840                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1841         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1842         ext4fs_dirhash(name, namelen, &hinfo);
1843         frame = frames;
1844         frame->entries = entries;
1845         frame->at = entries;
1846         frame->bh = bh;
1847         bh = bh2;
1848
1849         ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1850         ext4_handle_dirty_dirent_node(handle, dir, bh);
1851
1852         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1853         if (!de) {
1854                 /*
1855                  * Even if the block split failed, we have to properly write
1856                  * out all the changes we did so far. Otherwise we can end up
1857                  * with corrupted filesystem.
1858                  */
1859                 ext4_mark_inode_dirty(handle, dir);
1860                 dx_release(frames);
1861                 return retval;
1862         }
1863         dx_release(frames);
1864
1865         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1866         brelse(bh);
1867         return retval;
1868 }
1869
1870 /*
1871  *      ext4_add_entry()
1872  *
1873  * adds a file entry to the specified directory, using the same
1874  * semantics as ext4_find_entry(). It returns NULL if it failed.
1875  *
1876  * NOTE!! The inode part of 'de' is left at 0 - which means you
1877  * may not sleep between calling this and putting something into
1878  * the entry, as someone else might have used it while you slept.
1879  */
1880 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1881                           struct inode *inode)
1882 {
1883         struct inode *dir = dentry->d_parent->d_inode;
1884         struct buffer_head *bh;
1885         struct ext4_dir_entry_2 *de;
1886         struct ext4_dir_entry_tail *t;
1887         struct super_block *sb;
1888         int     retval;
1889         int     dx_fallback=0;
1890         unsigned blocksize;
1891         ext4_lblk_t block, blocks;
1892         int     csum_size = 0;
1893
1894         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1895                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1896                 csum_size = sizeof(struct ext4_dir_entry_tail);
1897
1898         sb = dir->i_sb;
1899         blocksize = sb->s_blocksize;
1900         if (!dentry->d_name.len)
1901                 return -EINVAL;
1902
1903         if (ext4_has_inline_data(dir)) {
1904                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1905                 if (retval < 0)
1906                         return retval;
1907                 if (retval == 1) {
1908                         retval = 0;
1909                         return retval;
1910                 }
1911         }
1912
1913         if (is_dx(dir)) {
1914                 retval = ext4_dx_add_entry(handle, dentry, inode);
1915                 if (!retval || (retval != ERR_BAD_DX_DIR))
1916                         return retval;
1917                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1918                 dx_fallback++;
1919                 ext4_mark_inode_dirty(handle, dir);
1920         }
1921         blocks = dir->i_size >> sb->s_blocksize_bits;
1922         for (block = 0; block < blocks; block++) {
1923                 bh = ext4_read_dirblock(dir, block, DIRENT);
1924                 if (IS_ERR(bh))
1925                         return PTR_ERR(bh);
1926
1927                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1928                 if (retval != -ENOSPC) {
1929                         brelse(bh);
1930                         return retval;
1931                 }
1932
1933                 if (blocks == 1 && !dx_fallback &&
1934                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1935                         return make_indexed_dir(handle, dentry, inode, bh);
1936                 brelse(bh);
1937         }
1938         bh = ext4_append(handle, dir, &block);
1939         if (IS_ERR(bh))
1940                 return PTR_ERR(bh);
1941         de = (struct ext4_dir_entry_2 *) bh->b_data;
1942         de->inode = 0;
1943         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1944
1945         if (csum_size) {
1946                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1947                 initialize_dirent_tail(t, blocksize);
1948         }
1949
1950         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1951         brelse(bh);
1952         if (retval == 0)
1953                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1954         return retval;
1955 }
1956
1957 /*
1958  * Returns 0 for success, or a negative error value
1959  */
1960 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1961                              struct inode *inode)
1962 {
1963         struct dx_frame frames[2], *frame;
1964         struct dx_entry *entries, *at;
1965         struct dx_hash_info hinfo;
1966         struct buffer_head *bh;
1967         struct inode *dir = dentry->d_parent->d_inode;
1968         struct super_block *sb = dir->i_sb;
1969         struct ext4_dir_entry_2 *de;
1970         int err;
1971
1972         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1973         if (!frame)
1974                 return err;
1975         entries = frame->entries;
1976         at = frame->at;
1977         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1978         if (IS_ERR(bh)) {
1979                 err = PTR_ERR(bh);
1980                 bh = NULL;
1981                 goto cleanup;
1982         }
1983
1984         BUFFER_TRACE(bh, "get_write_access");
1985         err = ext4_journal_get_write_access(handle, bh);
1986         if (err)
1987                 goto journal_error;
1988
1989         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1990         if (err != -ENOSPC)
1991                 goto cleanup;
1992
1993         /* Block full, should compress but for now just split */
1994         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1995                        dx_get_count(entries), dx_get_limit(entries)));
1996         /* Need to split index? */
1997         if (dx_get_count(entries) == dx_get_limit(entries)) {
1998                 ext4_lblk_t newblock;
1999                 unsigned icount = dx_get_count(entries);
2000                 int levels = frame - frames;
2001                 struct dx_entry *entries2;
2002                 struct dx_node *node2;
2003                 struct buffer_head *bh2;
2004
2005                 if (levels && (dx_get_count(frames->entries) ==
2006                                dx_get_limit(frames->entries))) {
2007                         ext4_warning(sb, "Directory index full!");
2008                         err = -ENOSPC;
2009                         goto cleanup;
2010                 }
2011                 bh2 = ext4_append(handle, dir, &newblock);
2012                 if (IS_ERR(bh2)) {
2013                         err = PTR_ERR(bh2);
2014                         goto cleanup;
2015                 }
2016                 node2 = (struct dx_node *)(bh2->b_data);
2017                 entries2 = node2->entries;
2018                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2019                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2020                                                            sb->s_blocksize);
2021                 BUFFER_TRACE(frame->bh, "get_write_access");
2022                 err = ext4_journal_get_write_access(handle, frame->bh);
2023                 if (err)
2024                         goto journal_error;
2025                 if (levels) {
2026                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2027                         unsigned hash2 = dx_get_hash(entries + icount1);
2028                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2029                                        icount1, icount2));
2030
2031                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2032                         err = ext4_journal_get_write_access(handle,
2033                                                              frames[0].bh);
2034                         if (err)
2035                                 goto journal_error;
2036
2037                         memcpy((char *) entries2, (char *) (entries + icount1),
2038                                icount2 * sizeof(struct dx_entry));
2039                         dx_set_count(entries, icount1);
2040                         dx_set_count(entries2, icount2);
2041                         dx_set_limit(entries2, dx_node_limit(dir));
2042
2043                         /* Which index block gets the new entry? */
2044                         if (at - entries >= icount1) {
2045                                 frame->at = at = at - entries - icount1 + entries2;
2046                                 frame->entries = entries = entries2;
2047                                 swap(frame->bh, bh2);
2048                         }
2049                         dx_insert_block(frames + 0, hash2, newblock);
2050                         dxtrace(dx_show_index("node", frames[1].entries));
2051                         dxtrace(dx_show_index("node",
2052                                ((struct dx_node *) bh2->b_data)->entries));
2053                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2054                         if (err)
2055                                 goto journal_error;
2056                         brelse (bh2);
2057                 } else {
2058                         dxtrace(printk(KERN_DEBUG
2059                                        "Creating second level index...\n"));
2060                         memcpy((char *) entries2, (char *) entries,
2061                                icount * sizeof(struct dx_entry));
2062                         dx_set_limit(entries2, dx_node_limit(dir));
2063
2064                         /* Set up root */
2065                         dx_set_count(entries, 1);
2066                         dx_set_block(entries + 0, newblock);
2067                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2068
2069                         /* Add new access path frame */
2070                         frame = frames + 1;
2071                         frame->at = at = at - entries + entries2;
2072                         frame->entries = entries = entries2;
2073                         frame->bh = bh2;
2074                         err = ext4_journal_get_write_access(handle,
2075                                                              frame->bh);
2076                         if (err)
2077                                 goto journal_error;
2078                 }
2079                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2080                 if (err) {
2081                         ext4_std_error(inode->i_sb, err);
2082                         goto cleanup;
2083                 }
2084         }
2085         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2086         if (!de)
2087                 goto cleanup;
2088         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2089         goto cleanup;
2090
2091 journal_error:
2092         ext4_std_error(dir->i_sb, err);
2093 cleanup:
2094         brelse(bh);
2095         dx_release(frames);
2096         return err;
2097 }
2098
2099 /*
2100  * ext4_generic_delete_entry deletes a directory entry by merging it
2101  * with the previous entry
2102  */
2103 int ext4_generic_delete_entry(handle_t *handle,
2104                               struct inode *dir,
2105                               struct ext4_dir_entry_2 *de_del,
2106                               struct buffer_head *bh,
2107                               void *entry_buf,
2108                               int buf_size,
2109                               int csum_size)
2110 {
2111         struct ext4_dir_entry_2 *de, *pde;
2112         unsigned int blocksize = dir->i_sb->s_blocksize;
2113         int i;
2114
2115         i = 0;
2116         pde = NULL;
2117         de = (struct ext4_dir_entry_2 *)entry_buf;
2118         while (i < buf_size - csum_size) {
2119                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2120                                          bh->b_data, bh->b_size, i))
2121                         return -EIO;
2122                 if (de == de_del)  {
2123                         if (pde)
2124                                 pde->rec_len = ext4_rec_len_to_disk(
2125                                         ext4_rec_len_from_disk(pde->rec_len,
2126                                                                blocksize) +
2127                                         ext4_rec_len_from_disk(de->rec_len,
2128                                                                blocksize),
2129                                         blocksize);
2130                         else
2131                                 de->inode = 0;
2132                         dir->i_version++;
2133                         return 0;
2134                 }
2135                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2136                 pde = de;
2137                 de = ext4_next_entry(de, blocksize);
2138         }
2139         return -ENOENT;
2140 }
2141
2142 static int ext4_delete_entry(handle_t *handle,
2143                              struct inode *dir,
2144                              struct ext4_dir_entry_2 *de_del,
2145                              struct buffer_head *bh)
2146 {
2147         int err, csum_size = 0;
2148
2149         if (ext4_has_inline_data(dir)) {
2150                 int has_inline_data = 1;
2151                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2152                                                &has_inline_data);
2153                 if (has_inline_data)
2154                         return err;
2155         }
2156
2157         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2158                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2159                 csum_size = sizeof(struct ext4_dir_entry_tail);
2160
2161         BUFFER_TRACE(bh, "get_write_access");
2162         err = ext4_journal_get_write_access(handle, bh);
2163         if (unlikely(err))
2164                 goto out;
2165
2166         err = ext4_generic_delete_entry(handle, dir, de_del,
2167                                         bh, bh->b_data,
2168                                         dir->i_sb->s_blocksize, csum_size);
2169         if (err)
2170                 goto out;
2171
2172         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2173         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2174         if (unlikely(err))
2175                 goto out;
2176
2177         return 0;
2178 out:
2179         if (err != -ENOENT)
2180                 ext4_std_error(dir->i_sb, err);
2181         return err;
2182 }
2183
2184 /*
2185  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2186  * since this indicates that nlinks count was previously 1.
2187  */
2188 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2189 {
2190         inc_nlink(inode);
2191         if (is_dx(inode) && inode->i_nlink > 1) {
2192                 /* limit is 16-bit i_links_count */
2193                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2194                         set_nlink(inode, 1);
2195                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2196                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2197                 }
2198         }
2199 }
2200
2201 /*
2202  * If a directory had nlink == 1, then we should let it be 1. This indicates
2203  * directory has >EXT4_LINK_MAX subdirs.
2204  */
2205 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2206 {
2207         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2208                 drop_nlink(inode);
2209 }
2210
2211
2212 static int ext4_add_nondir(handle_t *handle,
2213                 struct dentry *dentry, struct inode *inode)
2214 {
2215         int err = ext4_add_entry(handle, dentry, inode);
2216         if (!err) {
2217                 ext4_mark_inode_dirty(handle, inode);
2218                 unlock_new_inode(inode);
2219                 d_instantiate(dentry, inode);
2220                 return 0;
2221         }
2222         drop_nlink(inode);
2223         unlock_new_inode(inode);
2224         iput(inode);
2225         return err;
2226 }
2227
2228 /*
2229  * By the time this is called, we already have created
2230  * the directory cache entry for the new file, but it
2231  * is so far negative - it has no inode.
2232  *
2233  * If the create succeeds, we fill in the inode information
2234  * with d_instantiate().
2235  */
2236 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2237                        bool excl)
2238 {
2239         handle_t *handle;
2240         struct inode *inode;
2241         int err, credits, retries = 0;
2242
2243         dquot_initialize(dir);
2244
2245         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2246                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2247 retry:
2248         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2249                                             NULL, EXT4_HT_DIR, credits);
2250         handle = ext4_journal_current_handle();
2251         err = PTR_ERR(inode);
2252         if (!IS_ERR(inode)) {
2253                 inode->i_op = &ext4_file_inode_operations;
2254                 inode->i_fop = &ext4_file_operations;
2255                 ext4_set_aops(inode);
2256                 err = ext4_add_nondir(handle, dentry, inode);
2257                 if (!err && IS_DIRSYNC(dir))
2258                         ext4_handle_sync(handle);
2259         }
2260         if (handle)
2261                 ext4_journal_stop(handle);
2262         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2263                 goto retry;
2264         return err;
2265 }
2266
2267 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2268                       umode_t mode, dev_t rdev)
2269 {
2270         handle_t *handle;
2271         struct inode *inode;
2272         int err, credits, retries = 0;
2273
2274         if (!new_valid_dev(rdev))
2275                 return -EINVAL;
2276
2277         dquot_initialize(dir);
2278
2279         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2280                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2281 retry:
2282         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2283                                             NULL, EXT4_HT_DIR, credits);
2284         handle = ext4_journal_current_handle();
2285         err = PTR_ERR(inode);
2286         if (!IS_ERR(inode)) {
2287                 init_special_inode(inode, inode->i_mode, rdev);
2288                 inode->i_op = &ext4_special_inode_operations;
2289                 err = ext4_add_nondir(handle, dentry, inode);
2290                 if (!err && IS_DIRSYNC(dir))
2291                         ext4_handle_sync(handle);
2292         }
2293         if (handle)
2294                 ext4_journal_stop(handle);
2295         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2296                 goto retry;
2297         return err;
2298 }
2299
2300 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2301 {
2302         handle_t *handle;
2303         struct inode *inode;
2304         int err, retries = 0;
2305
2306         dquot_initialize(dir);
2307
2308 retry:
2309         inode = ext4_new_inode_start_handle(dir, mode,
2310                                             NULL, 0, NULL,
2311                                             EXT4_HT_DIR,
2312                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2313                           4 + EXT4_XATTR_TRANS_BLOCKS);
2314         handle = ext4_journal_current_handle();
2315         err = PTR_ERR(inode);
2316         if (!IS_ERR(inode)) {
2317                 inode->i_op = &ext4_file_inode_operations;
2318                 inode->i_fop = &ext4_file_operations;
2319                 ext4_set_aops(inode);
2320                 d_tmpfile(dentry, inode);
2321                 err = ext4_orphan_add(handle, inode);
2322                 if (err)
2323                         goto err_unlock_inode;
2324                 mark_inode_dirty(inode);
2325                 unlock_new_inode(inode);
2326         }
2327         if (handle)
2328                 ext4_journal_stop(handle);
2329         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2330                 goto retry;
2331         return err;
2332 err_unlock_inode:
2333         ext4_journal_stop(handle);
2334         unlock_new_inode(inode);
2335         return err;
2336 }
2337
2338 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2339                           struct ext4_dir_entry_2 *de,
2340                           int blocksize, int csum_size,
2341                           unsigned int parent_ino, int dotdot_real_len)
2342 {
2343         de->inode = cpu_to_le32(inode->i_ino);
2344         de->name_len = 1;
2345         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2346                                            blocksize);
2347         strcpy(de->name, ".");
2348         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2349
2350         de = ext4_next_entry(de, blocksize);
2351         de->inode = cpu_to_le32(parent_ino);
2352         de->name_len = 2;
2353         if (!dotdot_real_len)
2354                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2355                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2356                                         blocksize);
2357         else
2358                 de->rec_len = ext4_rec_len_to_disk(
2359                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2360         strcpy(de->name, "..");
2361         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2362
2363         return ext4_next_entry(de, blocksize);
2364 }
2365
2366 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2367                              struct inode *inode)
2368 {
2369         struct buffer_head *dir_block = NULL;
2370         struct ext4_dir_entry_2 *de;
2371         struct ext4_dir_entry_tail *t;
2372         ext4_lblk_t block = 0;
2373         unsigned int blocksize = dir->i_sb->s_blocksize;
2374         int csum_size = 0;
2375         int err;
2376
2377         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2378                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2379                 csum_size = sizeof(struct ext4_dir_entry_tail);
2380
2381         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2382                 err = ext4_try_create_inline_dir(handle, dir, inode);
2383                 if (err < 0 && err != -ENOSPC)
2384                         goto out;
2385                 if (!err)
2386                         goto out;
2387         }
2388
2389         inode->i_size = 0;
2390         dir_block = ext4_append(handle, inode, &block);
2391         if (IS_ERR(dir_block))
2392                 return PTR_ERR(dir_block);
2393         BUFFER_TRACE(dir_block, "get_write_access");
2394         err = ext4_journal_get_write_access(handle, dir_block);
2395         if (err)
2396                 goto out;
2397         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2398         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2399         set_nlink(inode, 2);
2400         if (csum_size) {
2401                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2402                 initialize_dirent_tail(t, blocksize);
2403         }
2404
2405         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2406         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2407         if (err)
2408                 goto out;
2409         set_buffer_verified(dir_block);
2410 out:
2411         brelse(dir_block);
2412         return err;
2413 }
2414
2415 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2416 {
2417         handle_t *handle;
2418         struct inode *inode;
2419         int err, credits, retries = 0;
2420
2421         if (EXT4_DIR_LINK_MAX(dir))
2422                 return -EMLINK;
2423
2424         dquot_initialize(dir);
2425
2426         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2427                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2428 retry:
2429         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2430                                             &dentry->d_name,
2431                                             0, NULL, EXT4_HT_DIR, credits);
2432         handle = ext4_journal_current_handle();
2433         err = PTR_ERR(inode);
2434         if (IS_ERR(inode))
2435                 goto out_stop;
2436
2437         inode->i_op = &ext4_dir_inode_operations;
2438         inode->i_fop = &ext4_dir_operations;
2439         err = ext4_init_new_dir(handle, dir, inode);
2440         if (err)
2441                 goto out_clear_inode;
2442         err = ext4_mark_inode_dirty(handle, inode);
2443         if (!err)
2444                 err = ext4_add_entry(handle, dentry, inode);
2445         if (err) {
2446 out_clear_inode:
2447                 clear_nlink(inode);
2448                 unlock_new_inode(inode);
2449                 ext4_mark_inode_dirty(handle, inode);
2450                 iput(inode);
2451                 goto out_stop;
2452         }
2453         ext4_inc_count(handle, dir);
2454         ext4_update_dx_flag(dir);
2455         err = ext4_mark_inode_dirty(handle, dir);
2456         if (err)
2457                 goto out_clear_inode;
2458         unlock_new_inode(inode);
2459         d_instantiate(dentry, inode);
2460         if (IS_DIRSYNC(dir))
2461                 ext4_handle_sync(handle);
2462
2463 out_stop:
2464         if (handle)
2465                 ext4_journal_stop(handle);
2466         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2467                 goto retry;
2468         return err;
2469 }
2470
2471 /*
2472  * routine to check that the specified directory is empty (for rmdir)
2473  */
2474 static int empty_dir(struct inode *inode)
2475 {
2476         unsigned int offset;
2477         struct buffer_head *bh;
2478         struct ext4_dir_entry_2 *de, *de1;
2479         struct super_block *sb;
2480         int err = 0;
2481
2482         if (ext4_has_inline_data(inode)) {
2483                 int has_inline_data = 1;
2484
2485                 err = empty_inline_dir(inode, &has_inline_data);
2486                 if (has_inline_data)
2487                         return err;
2488         }
2489
2490         sb = inode->i_sb;
2491         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2492                 EXT4_ERROR_INODE(inode, "invalid size");
2493                 return 1;
2494         }
2495         bh = ext4_read_dirblock(inode, 0, EITHER);
2496         if (IS_ERR(bh))
2497                 return 1;
2498
2499         de = (struct ext4_dir_entry_2 *) bh->b_data;
2500         de1 = ext4_next_entry(de, sb->s_blocksize);
2501         if (le32_to_cpu(de->inode) != inode->i_ino ||
2502                         !le32_to_cpu(de1->inode) ||
2503                         strcmp(".", de->name) ||
2504                         strcmp("..", de1->name)) {
2505                 ext4_warning(inode->i_sb,
2506                              "bad directory (dir #%lu) - no `.' or `..'",
2507                              inode->i_ino);
2508                 brelse(bh);
2509                 return 1;
2510         }
2511         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2512                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2513         de = ext4_next_entry(de1, sb->s_blocksize);
2514         while (offset < inode->i_size) {
2515                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2516                         unsigned int lblock;
2517                         err = 0;
2518                         brelse(bh);
2519                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2520                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2521                         if (IS_ERR(bh))
2522                                 return 1;
2523                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2524                 }
2525                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2526                                          bh->b_data, bh->b_size, offset)) {
2527                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2528                                                          sb->s_blocksize);
2529                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2530                         continue;
2531                 }
2532                 if (le32_to_cpu(de->inode)) {
2533                         brelse(bh);
2534                         return 0;
2535                 }
2536                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2537                 de = ext4_next_entry(de, sb->s_blocksize);
2538         }
2539         brelse(bh);
2540         return 1;
2541 }
2542
2543 /*
2544  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2545  * such inodes, starting at the superblock, in case we crash before the
2546  * file is closed/deleted, or in case the inode truncate spans multiple
2547  * transactions and the last transaction is not recovered after a crash.
2548  *
2549  * At filesystem recovery time, we walk this list deleting unlinked
2550  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2551  *
2552  * Orphan list manipulation functions must be called under i_mutex unless
2553  * we are just creating the inode or deleting it.
2554  */
2555 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2556 {
2557         struct super_block *sb = inode->i_sb;
2558         struct ext4_sb_info *sbi = EXT4_SB(sb);
2559         struct ext4_iloc iloc;
2560         int err = 0, rc;
2561         bool dirty = false;
2562
2563         if (!sbi->s_journal)
2564                 return 0;
2565
2566         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2567                      !mutex_is_locked(&inode->i_mutex));
2568         /*
2569          * Exit early if inode already is on orphan list. This is a big speedup
2570          * since we don't have to contend on the global s_orphan_lock.
2571          */
2572         if (!list_empty(&EXT4_I(inode)->i_orphan))
2573                 return 0;
2574
2575         /*
2576          * Orphan handling is only valid for files with data blocks
2577          * being truncated, or files being unlinked. Note that we either
2578          * hold i_mutex, or the inode can not be referenced from outside,
2579          * so i_nlink should not be bumped due to race
2580          */
2581         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2582                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2583
2584         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2585         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2586         if (err)
2587                 goto out;
2588
2589         err = ext4_reserve_inode_write(handle, inode, &iloc);
2590         if (err)
2591                 goto out;
2592
2593         mutex_lock(&sbi->s_orphan_lock);
2594         /*
2595          * Due to previous errors inode may be already a part of on-disk
2596          * orphan list. If so skip on-disk list modification.
2597          */
2598         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2599             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2600                 /* Insert this inode at the head of the on-disk orphan list */
2601                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2602                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2603                 dirty = true;
2604         }
2605         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2606         mutex_unlock(&sbi->s_orphan_lock);
2607
2608         if (dirty) {
2609                 err = ext4_handle_dirty_super(handle, sb);
2610                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2611                 if (!err)
2612                         err = rc;
2613                 if (err) {
2614                         /*
2615                          * We have to remove inode from in-memory list if
2616                          * addition to on disk orphan list failed. Stray orphan
2617                          * list entries can cause panics at unmount time.
2618                          */
2619                         mutex_lock(&sbi->s_orphan_lock);
2620                         list_del(&EXT4_I(inode)->i_orphan);
2621                         mutex_unlock(&sbi->s_orphan_lock);
2622                 }
2623         }
2624         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2625         jbd_debug(4, "orphan inode %lu will point to %d\n",
2626                         inode->i_ino, NEXT_ORPHAN(inode));
2627 out:
2628         ext4_std_error(sb, err);
2629         return err;
2630 }
2631
2632 /*
2633  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2634  * of such inodes stored on disk, because it is finally being cleaned up.
2635  */
2636 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2637 {
2638         struct list_head *prev;
2639         struct ext4_inode_info *ei = EXT4_I(inode);
2640         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2641         __u32 ino_next;
2642         struct ext4_iloc iloc;
2643         int err = 0;
2644
2645         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2646                 return 0;
2647
2648         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2649                      !mutex_is_locked(&inode->i_mutex));
2650         /* Do this quick check before taking global s_orphan_lock. */
2651         if (list_empty(&ei->i_orphan))
2652                 return 0;
2653
2654         if (handle) {
2655                 /* Grab inode buffer early before taking global s_orphan_lock */
2656                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2657         }
2658
2659         mutex_lock(&sbi->s_orphan_lock);
2660         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2661
2662         prev = ei->i_orphan.prev;
2663         list_del_init(&ei->i_orphan);
2664
2665         /* If we're on an error path, we may not have a valid
2666          * transaction handle with which to update the orphan list on
2667          * disk, but we still need to remove the inode from the linked
2668          * list in memory. */
2669         if (!handle || err) {
2670                 mutex_unlock(&sbi->s_orphan_lock);
2671                 goto out_err;
2672         }
2673
2674         ino_next = NEXT_ORPHAN(inode);
2675         if (prev == &sbi->s_orphan) {
2676                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2677                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2678                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2679                 if (err) {
2680                         mutex_unlock(&sbi->s_orphan_lock);
2681                         goto out_brelse;
2682                 }
2683                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2684                 mutex_unlock(&sbi->s_orphan_lock);
2685                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2686         } else {
2687                 struct ext4_iloc iloc2;
2688                 struct inode *i_prev =
2689                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2690
2691                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2692                           i_prev->i_ino, ino_next);
2693                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2694                 if (err) {
2695                         mutex_unlock(&sbi->s_orphan_lock);
2696                         goto out_brelse;
2697                 }
2698                 NEXT_ORPHAN(i_prev) = ino_next;
2699                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2700                 mutex_unlock(&sbi->s_orphan_lock);
2701         }
2702         if (err)
2703                 goto out_brelse;
2704         NEXT_ORPHAN(inode) = 0;
2705         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2706 out_err:
2707         ext4_std_error(inode->i_sb, err);
2708         return err;
2709
2710 out_brelse:
2711         brelse(iloc.bh);
2712         goto out_err;
2713 }
2714
2715 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2716 {
2717         int retval;
2718         struct inode *inode;
2719         struct buffer_head *bh;
2720         struct ext4_dir_entry_2 *de;
2721         handle_t *handle = NULL;
2722
2723         /* Initialize quotas before so that eventual writes go in
2724          * separate transaction */
2725         dquot_initialize(dir);
2726         dquot_initialize(dentry->d_inode);
2727
2728         retval = -ENOENT;
2729         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2730         if (!bh)
2731                 goto end_rmdir;
2732
2733         inode = dentry->d_inode;
2734
2735         retval = -EIO;
2736         if (le32_to_cpu(de->inode) != inode->i_ino)
2737                 goto end_rmdir;
2738
2739         retval = -ENOTEMPTY;
2740         if (!empty_dir(inode))
2741                 goto end_rmdir;
2742
2743         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2744                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2745         if (IS_ERR(handle)) {
2746                 retval = PTR_ERR(handle);
2747                 handle = NULL;
2748                 goto end_rmdir;
2749         }
2750
2751         if (IS_DIRSYNC(dir))
2752                 ext4_handle_sync(handle);
2753
2754         retval = ext4_delete_entry(handle, dir, de, bh);
2755         if (retval)
2756                 goto end_rmdir;
2757         if (!EXT4_DIR_LINK_EMPTY(inode))
2758                 ext4_warning(inode->i_sb,
2759                              "empty directory has too many links (%d)",
2760                              inode->i_nlink);
2761         inode->i_version++;
2762         clear_nlink(inode);
2763         /* There's no need to set i_disksize: the fact that i_nlink is
2764          * zero will ensure that the right thing happens during any
2765          * recovery. */
2766         inode->i_size = 0;
2767         ext4_orphan_add(handle, inode);
2768         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2769         ext4_mark_inode_dirty(handle, inode);
2770         ext4_dec_count(handle, dir);
2771         ext4_update_dx_flag(dir);
2772         ext4_mark_inode_dirty(handle, dir);
2773
2774 end_rmdir:
2775         brelse(bh);
2776         if (handle)
2777                 ext4_journal_stop(handle);
2778         return retval;
2779 }
2780
2781 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2782 {
2783         int retval;
2784         struct inode *inode;
2785         struct buffer_head *bh;
2786         struct ext4_dir_entry_2 *de;
2787         handle_t *handle = NULL;
2788
2789         trace_ext4_unlink_enter(dir, dentry);
2790         /* Initialize quotas before so that eventual writes go
2791          * in separate transaction */
2792         dquot_initialize(dir);
2793         dquot_initialize(dentry->d_inode);
2794
2795         retval = -ENOENT;
2796         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2797         if (!bh)
2798                 goto end_unlink;
2799
2800         inode = dentry->d_inode;
2801
2802         retval = -EIO;
2803         if (le32_to_cpu(de->inode) != inode->i_ino)
2804                 goto end_unlink;
2805
2806         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2807                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2808         if (IS_ERR(handle)) {
2809                 retval = PTR_ERR(handle);
2810                 handle = NULL;
2811                 goto end_unlink;
2812         }
2813
2814         if (IS_DIRSYNC(dir))
2815                 ext4_handle_sync(handle);
2816
2817         if (!inode->i_nlink) {
2818                 ext4_warning(inode->i_sb,
2819                              "Deleting nonexistent file (%lu), %d",
2820                              inode->i_ino, inode->i_nlink);
2821                 set_nlink(inode, 1);
2822         }
2823         retval = ext4_delete_entry(handle, dir, de, bh);
2824         if (retval)
2825                 goto end_unlink;
2826         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2827         ext4_update_dx_flag(dir);
2828         ext4_mark_inode_dirty(handle, dir);
2829         drop_nlink(inode);
2830         if (!inode->i_nlink)
2831                 ext4_orphan_add(handle, inode);
2832         inode->i_ctime = ext4_current_time(inode);
2833         ext4_mark_inode_dirty(handle, inode);
2834         retval = 0;
2835
2836 end_unlink:
2837         brelse(bh);
2838         if (handle)
2839                 ext4_journal_stop(handle);
2840         trace_ext4_unlink_exit(dentry, retval);
2841         return retval;
2842 }
2843
2844 static int ext4_symlink(struct inode *dir,
2845                         struct dentry *dentry, const char *symname)
2846 {
2847         handle_t *handle;
2848         struct inode *inode;
2849         int l, err, retries = 0;
2850         int credits;
2851
2852         l = strlen(symname)+1;
2853         if (l > dir->i_sb->s_blocksize)
2854                 return -ENAMETOOLONG;
2855
2856         dquot_initialize(dir);
2857
2858         if (l > EXT4_N_BLOCKS * 4) {
2859                 /*
2860                  * For non-fast symlinks, we just allocate inode and put it on
2861                  * orphan list in the first transaction => we need bitmap,
2862                  * group descriptor, sb, inode block, quota blocks, and
2863                  * possibly selinux xattr blocks.
2864                  */
2865                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2866                           EXT4_XATTR_TRANS_BLOCKS;
2867         } else {
2868                 /*
2869                  * Fast symlink. We have to add entry to directory
2870                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2871                  * allocate new inode (bitmap, group descriptor, inode block,
2872                  * quota blocks, sb is already counted in previous macros).
2873                  */
2874                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2875                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2876         }
2877 retry:
2878         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2879                                             &dentry->d_name, 0, NULL,
2880                                             EXT4_HT_DIR, credits);
2881         handle = ext4_journal_current_handle();
2882         err = PTR_ERR(inode);
2883         if (IS_ERR(inode))
2884                 goto out_stop;
2885
2886         if (l > EXT4_N_BLOCKS * 4) {
2887                 inode->i_op = &ext4_symlink_inode_operations;
2888                 ext4_set_aops(inode);
2889                 /*
2890                  * We cannot call page_symlink() with transaction started
2891                  * because it calls into ext4_write_begin() which can wait
2892                  * for transaction commit if we are running out of space
2893                  * and thus we deadlock. So we have to stop transaction now
2894                  * and restart it when symlink contents is written.
2895                  *
2896                  * To keep fs consistent in case of crash, we have to put inode
2897                  * to orphan list in the mean time.
2898                  */
2899                 drop_nlink(inode);
2900                 err = ext4_orphan_add(handle, inode);
2901                 ext4_journal_stop(handle);
2902                 if (err)
2903                         goto err_drop_inode;
2904                 err = __page_symlink(inode, symname, l, 1);
2905                 if (err)
2906                         goto err_drop_inode;
2907                 /*
2908                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2909                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2910                  */
2911                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2912                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2913                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2914                 if (IS_ERR(handle)) {
2915                         err = PTR_ERR(handle);
2916                         goto err_drop_inode;
2917                 }
2918                 set_nlink(inode, 1);
2919                 err = ext4_orphan_del(handle, inode);
2920                 if (err) {
2921                         ext4_journal_stop(handle);
2922                         clear_nlink(inode);
2923                         goto err_drop_inode;
2924                 }
2925         } else {
2926                 /* clear the extent format for fast symlink */
2927                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2928                 inode->i_op = &ext4_fast_symlink_inode_operations;
2929                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2930                 inode->i_size = l-1;
2931         }
2932         EXT4_I(inode)->i_disksize = inode->i_size;
2933         err = ext4_add_nondir(handle, dentry, inode);
2934         if (!err && IS_DIRSYNC(dir))
2935                 ext4_handle_sync(handle);
2936
2937 out_stop:
2938         if (handle)
2939                 ext4_journal_stop(handle);
2940         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2941                 goto retry;
2942         return err;
2943 err_drop_inode:
2944         unlock_new_inode(inode);
2945         iput(inode);
2946         return err;
2947 }
2948
2949 static int ext4_link(struct dentry *old_dentry,
2950                      struct inode *dir, struct dentry *dentry)
2951 {
2952         handle_t *handle;
2953         struct inode *inode = old_dentry->d_inode;
2954         int err, retries = 0;
2955
2956         if (inode->i_nlink >= EXT4_LINK_MAX)
2957                 return -EMLINK;
2958
2959         dquot_initialize(dir);
2960
2961 retry:
2962         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2963                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2964                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2965         if (IS_ERR(handle))
2966                 return PTR_ERR(handle);
2967
2968         if (IS_DIRSYNC(dir))
2969                 ext4_handle_sync(handle);
2970
2971         inode->i_ctime = ext4_current_time(inode);
2972         ext4_inc_count(handle, inode);
2973         ihold(inode);
2974
2975         err = ext4_add_entry(handle, dentry, inode);
2976         if (!err) {
2977                 ext4_mark_inode_dirty(handle, inode);
2978                 /* this can happen only for tmpfile being
2979                  * linked the first time
2980                  */
2981                 if (inode->i_nlink == 1)
2982                         ext4_orphan_del(handle, inode);
2983                 d_instantiate(dentry, inode);
2984         } else {
2985                 drop_nlink(inode);
2986                 iput(inode);
2987         }
2988         ext4_journal_stop(handle);
2989         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2990                 goto retry;
2991         return err;
2992 }
2993
2994
2995 /*
2996  * Try to find buffer head where contains the parent block.
2997  * It should be the inode block if it is inlined or the 1st block
2998  * if it is a normal dir.
2999  */
3000 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3001                                         struct inode *inode,
3002                                         int *retval,
3003                                         struct ext4_dir_entry_2 **parent_de,
3004                                         int *inlined)
3005 {
3006         struct buffer_head *bh;
3007
3008         if (!ext4_has_inline_data(inode)) {
3009                 bh = ext4_read_dirblock(inode, 0, EITHER);
3010                 if (IS_ERR(bh)) {
3011                         *retval = PTR_ERR(bh);
3012                         return NULL;
3013                 }
3014                 *parent_de = ext4_next_entry(
3015                                         (struct ext4_dir_entry_2 *)bh->b_data,
3016                                         inode->i_sb->s_blocksize);
3017                 return bh;
3018         }
3019
3020         *inlined = 1;
3021         return ext4_get_first_inline_block(inode, parent_de, retval);
3022 }
3023
3024 struct ext4_renament {
3025         struct inode *dir;
3026         struct dentry *dentry;
3027         struct inode *inode;
3028         bool is_dir;
3029         int dir_nlink_delta;
3030
3031         /* entry for "dentry" */
3032         struct buffer_head *bh;
3033         struct ext4_dir_entry_2 *de;
3034         int inlined;
3035
3036         /* entry for ".." in inode if it's a directory */
3037         struct buffer_head *dir_bh;
3038         struct ext4_dir_entry_2 *parent_de;
3039         int dir_inlined;
3040 };
3041
3042 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3043 {
3044         int retval;
3045
3046         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3047                                               &retval, &ent->parent_de,
3048                                               &ent->dir_inlined);
3049         if (!ent->dir_bh)
3050                 return retval;
3051         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3052                 return -EIO;
3053         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3054         return ext4_journal_get_write_access(handle, ent->dir_bh);
3055 }
3056
3057 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3058                                   unsigned dir_ino)
3059 {
3060         int retval;
3061
3062         ent->parent_de->inode = cpu_to_le32(dir_ino);
3063         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3064         if (!ent->dir_inlined) {
3065                 if (is_dx(ent->inode)) {
3066                         retval = ext4_handle_dirty_dx_node(handle,
3067                                                            ent->inode,
3068                                                            ent->dir_bh);
3069                 } else {
3070                         retval = ext4_handle_dirty_dirent_node(handle,
3071                                                                ent->inode,
3072                                                                ent->dir_bh);
3073                 }
3074         } else {
3075                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3076         }
3077         if (retval) {
3078                 ext4_std_error(ent->dir->i_sb, retval);
3079                 return retval;
3080         }
3081         return 0;
3082 }
3083
3084 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3085                        unsigned ino, unsigned file_type)
3086 {
3087         int retval;
3088
3089         BUFFER_TRACE(ent->bh, "get write access");
3090         retval = ext4_journal_get_write_access(handle, ent->bh);
3091         if (retval)
3092                 return retval;
3093         ent->de->inode = cpu_to_le32(ino);
3094         if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3095                                       EXT4_FEATURE_INCOMPAT_FILETYPE))
3096                 ent->de->file_type = file_type;
3097         ent->dir->i_version++;
3098         ent->dir->i_ctime = ent->dir->i_mtime =
3099                 ext4_current_time(ent->dir);
3100         ext4_mark_inode_dirty(handle, ent->dir);
3101         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3102         if (!ent->inlined) {
3103                 retval = ext4_handle_dirty_dirent_node(handle,
3104                                                        ent->dir, ent->bh);
3105                 if (unlikely(retval)) {
3106                         ext4_std_error(ent->dir->i_sb, retval);
3107                         return retval;
3108                 }
3109         }
3110         brelse(ent->bh);
3111         ent->bh = NULL;
3112
3113         return 0;
3114 }
3115
3116 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3117                                   const struct qstr *d_name)
3118 {
3119         int retval = -ENOENT;
3120         struct buffer_head *bh;
3121         struct ext4_dir_entry_2 *de;
3122
3123         bh = ext4_find_entry(dir, d_name, &de, NULL);
3124         if (bh) {
3125                 retval = ext4_delete_entry(handle, dir, de, bh);
3126                 brelse(bh);
3127         }
3128         return retval;
3129 }
3130
3131 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent)
3132 {
3133         int retval;
3134         /*
3135          * ent->de could have moved from under us during htree split, so make
3136          * sure that we are deleting the right entry.  We might also be pointing
3137          * to a stale entry in the unused part of ent->bh so just checking inum
3138          * and the name isn't enough.
3139          */
3140         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3141             ent->de->name_len != ent->dentry->d_name.len ||
3142             strncmp(ent->de->name, ent->dentry->d_name.name,
3143                     ent->de->name_len)) {
3144                 retval = ext4_find_delete_entry(handle, ent->dir,
3145                                                 &ent->dentry->d_name);
3146         } else {
3147                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3148                 if (retval == -ENOENT) {
3149                         retval = ext4_find_delete_entry(handle, ent->dir,
3150                                                         &ent->dentry->d_name);
3151                 }
3152         }
3153
3154         if (retval) {
3155                 ext4_warning(ent->dir->i_sb,
3156                                 "Deleting old file (%lu), %d, error=%d",
3157                                 ent->dir->i_ino, ent->dir->i_nlink, retval);
3158         }
3159 }
3160
3161 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3162 {
3163         if (ent->dir_nlink_delta) {
3164                 if (ent->dir_nlink_delta == -1)
3165                         ext4_dec_count(handle, ent->dir);
3166                 else
3167                         ext4_inc_count(handle, ent->dir);
3168                 ext4_mark_inode_dirty(handle, ent->dir);
3169         }
3170 }
3171
3172 /*
3173  * Anybody can rename anything with this: the permission checks are left to the
3174  * higher-level routines.
3175  *
3176  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3177  * while new_{dentry,inode) refers to the destination dentry/inode
3178  * This comes from rename(const char *oldpath, const char *newpath)
3179  */
3180 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3181                        struct inode *new_dir, struct dentry *new_dentry)
3182 {
3183         handle_t *handle = NULL;
3184         struct ext4_renament old = {
3185                 .dir = old_dir,
3186                 .dentry = old_dentry,
3187                 .inode = old_dentry->d_inode,
3188         };
3189         struct ext4_renament new = {
3190                 .dir = new_dir,
3191                 .dentry = new_dentry,
3192                 .inode = new_dentry->d_inode,
3193         };
3194         int retval;
3195
3196         dquot_initialize(old.dir);
3197         dquot_initialize(new.dir);
3198
3199         /* Initialize quotas before so that eventual writes go
3200          * in separate transaction */
3201         if (new.inode)
3202                 dquot_initialize(new.inode);
3203
3204         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3205         /*
3206          *  Check for inode number is _not_ due to possible IO errors.
3207          *  We might rmdir the source, keep it as pwd of some process
3208          *  and merrily kill the link to whatever was created under the
3209          *  same name. Goodbye sticky bit ;-<
3210          */
3211         retval = -ENOENT;
3212         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3213                 goto end_rename;
3214
3215         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3216                                  &new.de, &new.inlined);
3217         if (new.bh) {
3218                 if (!new.inode) {
3219                         brelse(new.bh);
3220                         new.bh = NULL;
3221                 }
3222         }
3223         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3224                 ext4_alloc_da_blocks(old.inode);
3225
3226         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3227                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3228                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3229         if (IS_ERR(handle))
3230                 return PTR_ERR(handle);
3231
3232         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3233                 ext4_handle_sync(handle);
3234
3235         if (S_ISDIR(old.inode->i_mode)) {
3236                 if (new.inode) {
3237                         retval = -ENOTEMPTY;
3238                         if (!empty_dir(new.inode))
3239                                 goto end_rename;
3240                 } else {
3241                         retval = -EMLINK;
3242                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3243                                 goto end_rename;
3244                 }
3245                 retval = ext4_rename_dir_prepare(handle, &old);
3246                 if (retval)
3247                         goto end_rename;
3248         }
3249         if (!new.bh) {
3250                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3251                 if (retval)
3252                         goto end_rename;
3253         } else {
3254                 retval = ext4_setent(handle, &new,
3255                                      old.inode->i_ino, old.de->file_type);
3256                 if (retval)
3257                         goto end_rename;
3258         }
3259
3260         /*
3261          * Like most other Unix systems, set the ctime for inodes on a
3262          * rename.
3263          */
3264         old.inode->i_ctime = ext4_current_time(old.inode);
3265         ext4_mark_inode_dirty(handle, old.inode);
3266
3267         /*
3268          * ok, that's it
3269          */
3270         ext4_rename_delete(handle, &old);
3271
3272         if (new.inode) {
3273                 ext4_dec_count(handle, new.inode);
3274                 new.inode->i_ctime = ext4_current_time(new.inode);
3275         }
3276         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3277         ext4_update_dx_flag(old.dir);
3278         if (old.dir_bh) {
3279                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3280                 if (retval)
3281                         goto end_rename;
3282
3283                 ext4_dec_count(handle, old.dir);
3284                 if (new.inode) {
3285                         /* checked empty_dir above, can't have another parent,
3286                          * ext4_dec_count() won't work for many-linked dirs */
3287                         clear_nlink(new.inode);
3288                 } else {
3289                         ext4_inc_count(handle, new.dir);
3290                         ext4_update_dx_flag(new.dir);
3291                         ext4_mark_inode_dirty(handle, new.dir);
3292                 }
3293         }
3294         ext4_mark_inode_dirty(handle, old.dir);
3295         if (new.inode) {
3296                 ext4_mark_inode_dirty(handle, new.inode);
3297                 if (!new.inode->i_nlink)
3298                         ext4_orphan_add(handle, new.inode);
3299         }
3300         retval = 0;
3301
3302 end_rename:
3303         brelse(old.dir_bh);
3304         brelse(old.bh);
3305         brelse(new.bh);
3306         if (handle)
3307                 ext4_journal_stop(handle);
3308         return retval;
3309 }
3310
3311 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3312                              struct inode *new_dir, struct dentry *new_dentry)
3313 {
3314         handle_t *handle = NULL;
3315         struct ext4_renament old = {
3316                 .dir = old_dir,
3317                 .dentry = old_dentry,
3318                 .inode = old_dentry->d_inode,
3319         };
3320         struct ext4_renament new = {
3321                 .dir = new_dir,
3322                 .dentry = new_dentry,
3323                 .inode = new_dentry->d_inode,
3324         };
3325         u8 new_file_type;
3326         int retval;
3327
3328         dquot_initialize(old.dir);
3329         dquot_initialize(new.dir);
3330
3331         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3332                                  &old.de, &old.inlined);
3333         /*
3334          *  Check for inode number is _not_ due to possible IO errors.
3335          *  We might rmdir the source, keep it as pwd of some process
3336          *  and merrily kill the link to whatever was created under the
3337          *  same name. Goodbye sticky bit ;-<
3338          */
3339         retval = -ENOENT;
3340         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3341                 goto end_rename;
3342
3343         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3344                                  &new.de, &new.inlined);
3345
3346         /* RENAME_EXCHANGE case: old *and* new must both exist */
3347         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3348                 goto end_rename;
3349
3350         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3351                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3352                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3353         if (IS_ERR(handle))
3354                 return PTR_ERR(handle);
3355
3356         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3357                 ext4_handle_sync(handle);
3358
3359         if (S_ISDIR(old.inode->i_mode)) {
3360                 old.is_dir = true;
3361                 retval = ext4_rename_dir_prepare(handle, &old);
3362                 if (retval)
3363                         goto end_rename;
3364         }
3365         if (S_ISDIR(new.inode->i_mode)) {
3366                 new.is_dir = true;
3367                 retval = ext4_rename_dir_prepare(handle, &new);
3368                 if (retval)
3369                         goto end_rename;
3370         }
3371
3372         /*
3373          * Other than the special case of overwriting a directory, parents'
3374          * nlink only needs to be modified if this is a cross directory rename.
3375          */
3376         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3377                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3378                 new.dir_nlink_delta = -old.dir_nlink_delta;
3379                 retval = -EMLINK;
3380                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3381                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3382                         goto end_rename;
3383         }
3384
3385         new_file_type = new.de->file_type;
3386         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3387         if (retval)
3388                 goto end_rename;
3389
3390         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3391         if (retval)
3392                 goto end_rename;
3393
3394         /*
3395          * Like most other Unix systems, set the ctime for inodes on a
3396          * rename.
3397          */
3398         old.inode->i_ctime = ext4_current_time(old.inode);
3399         new.inode->i_ctime = ext4_current_time(new.inode);
3400         ext4_mark_inode_dirty(handle, old.inode);
3401         ext4_mark_inode_dirty(handle, new.inode);
3402
3403         if (old.dir_bh) {
3404                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3405                 if (retval)
3406                         goto end_rename;
3407         }
3408         if (new.dir_bh) {
3409                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3410                 if (retval)
3411                         goto end_rename;
3412         }
3413         ext4_update_dir_count(handle, &old);
3414         ext4_update_dir_count(handle, &new);
3415         retval = 0;
3416
3417 end_rename:
3418         brelse(old.dir_bh);
3419         brelse(new.dir_bh);
3420         brelse(old.bh);
3421         brelse(new.bh);
3422         if (handle)
3423                 ext4_journal_stop(handle);
3424         return retval;
3425 }
3426
3427 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3428                         struct inode *new_dir, struct dentry *new_dentry,
3429                         unsigned int flags)
3430 {
3431         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
3432                 return -EINVAL;
3433
3434         if (flags & RENAME_EXCHANGE) {
3435                 return ext4_cross_rename(old_dir, old_dentry,
3436                                          new_dir, new_dentry);
3437         }
3438         /*
3439          * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3440          * is equivalent to regular rename.
3441          */
3442         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry);
3443 }
3444
3445 /*
3446  * directories can handle most operations...
3447  */
3448 const struct inode_operations ext4_dir_inode_operations = {
3449         .create         = ext4_create,
3450         .lookup         = ext4_lookup,
3451         .link           = ext4_link,
3452         .unlink         = ext4_unlink,
3453         .symlink        = ext4_symlink,
3454         .mkdir          = ext4_mkdir,
3455         .rmdir          = ext4_rmdir,
3456         .mknod          = ext4_mknod,
3457         .tmpfile        = ext4_tmpfile,
3458         .rename         = ext4_rename,
3459         .rename2        = ext4_rename2,
3460         .setattr        = ext4_setattr,
3461         .setxattr       = generic_setxattr,
3462         .getxattr       = generic_getxattr,
3463         .listxattr      = ext4_listxattr,
3464         .removexattr    = generic_removexattr,
3465         .get_acl        = ext4_get_acl,
3466         .set_acl        = ext4_set_acl,
3467         .fiemap         = ext4_fiemap,
3468 };
3469
3470 const struct inode_operations ext4_special_inode_operations = {
3471         .setattr        = ext4_setattr,
3472         .setxattr       = generic_setxattr,
3473         .getxattr       = generic_getxattr,
3474         .listxattr      = ext4_listxattr,
3475         .removexattr    = generic_removexattr,
3476         .get_acl        = ext4_get_acl,
3477         .set_acl        = ext4_set_acl,
3478 };