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 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 275                              struct inode *inode);
 276
 277 /* checksumming functions */
 278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 279                             unsigned int blocksize)
 280 {
 281         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 282         t->det_rec_len = ext4_rec_len_to_disk(
 283                         sizeof(struct ext4_dir_entry_tail), blocksize);
 284         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 285 }
 286
 287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 289                                                    struct ext4_dir_entry *de)
 290 {
 291         struct ext4_dir_entry_tail *t;
 292
 293 #ifdef PARANOID
 294         struct ext4_dir_entry *d, *top;
 295
 296         d = de;
 297         top = (struct ext4_dir_entry *)(((void *)de) +
 298                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 299                 sizeof(struct ext4_dir_entry_tail)));
 300         while (d < top && d->rec_len)
 301                 d = (struct ext4_dir_entry *)(((void *)d) +
 302                     le16_to_cpu(d->rec_len));
 303
 304         if (d != top)
 305                 return NULL;
 306
 307         t = (struct ext4_dir_entry_tail *)d;
 308 #else
 309         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 310 #endif
 311
 312         if (t->det_reserved_zero1 ||
 313             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 314             t->det_reserved_zero2 ||
 315             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 316                 return NULL;
 317
 318         return t;
 319 }
 320
 321 static __le32 ext4_dirent_csum(struct inode *inode,
 322                                struct ext4_dir_entry *dirent, int size)
 323 {
 324         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 325         struct ext4_inode_info *ei = EXT4_I(inode);
 326         __u32 csum;
 327
 328         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 329         return cpu_to_le32(csum);
 330 }
 331
 332 static void warn_no_space_for_csum(struct inode *inode)
 333 {
 334         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
 335                      "checksum.  Please run e2fsck -D.", inode->i_ino);
 336 }
 337
 338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 339 {
 340         struct ext4_dir_entry_tail *t;
 341
 342         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 343                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 344                 return 1;
 345
 346         t = get_dirent_tail(inode, dirent);
 347         if (!t) {
 348                 warn_no_space_for_csum(inode);
 349                 return 0;
 350         }
 351
 352         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 353                                                 (void *)t - (void *)dirent))
 354                 return 0;
 355
 356         return 1;
 357 }
 358
 359 static void ext4_dirent_csum_set(struct inode *inode,
 360                                  struct ext4_dir_entry *dirent)
 361 {
 362         struct ext4_dir_entry_tail *t;
 363
 364         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 365                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 366                 return;
 367
 368         t = get_dirent_tail(inode, dirent);
 369         if (!t) {
 370                 warn_no_space_for_csum(inode);
 371                 return;
 372         }
 373
 374         t->det_checksum = ext4_dirent_csum(inode, dirent,
 375                                            (void *)t - (void *)dirent);
 376 }
 377
 378 int ext4_handle_dirty_dirent_node(handle_t *handle,
 379                                   struct inode *inode,
 380                                   struct buffer_head *bh)
 381 {
 382         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 383         return ext4_handle_dirty_metadata(handle, inode, bh);
 384 }
 385
 386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 387                                                struct ext4_dir_entry *dirent,
 388                                                int *offset)
 389 {
 390         struct ext4_dir_entry *dp;
 391         struct dx_root_info *root;
 392         int count_offset;
 393
 394         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 395                 count_offset = 8;
 396         else if (le16_to_cpu(dirent->rec_len) == 12) {
 397                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 398                 if (le16_to_cpu(dp->rec_len) !=
 399                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 400                         return NULL;
 401                 root = (struct dx_root_info *)(((void *)dp + 12));
 402                 if (root->reserved_zero ||
 403                     root->info_length != sizeof(struct dx_root_info))
 404                         return NULL;
 405                 count_offset = 32;
 406         } else
 407                 return NULL;
 408
 409         if (offset)
 410                 *offset = count_offset;
 411         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 412 }
 413
 414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 415                            int count_offset, int count, struct dx_tail *t)
 416 {
 417         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 418         struct ext4_inode_info *ei = EXT4_I(inode);
 419         __u32 csum;
 420         __le32 save_csum;
 421         int size;
 422
 423         size = count_offset + (count * sizeof(struct dx_entry));
 424         save_csum = t->dt_checksum;
 425         t->dt_checksum = 0;
 426         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 427         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
 428         t->dt_checksum = save_csum;
 429
 430         return cpu_to_le32(csum);
 431 }
 432
 433 static int ext4_dx_csum_verify(struct inode *inode,
 434                                struct ext4_dir_entry *dirent)
 435 {
 436         struct dx_countlimit *c;
 437         struct dx_tail *t;
 438         int count_offset, limit, count;
 439
 440         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 441                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 442                 return 1;
 443
 444         c = get_dx_countlimit(inode, dirent, &count_offset);
 445         if (!c) {
 446                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 447                 return 1;
 448         }
 449         limit = le16_to_cpu(c->limit);
 450         count = le16_to_cpu(c->count);
 451         if (count_offset + (limit * sizeof(struct dx_entry)) >
 452             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 453                 warn_no_space_for_csum(inode);
 454                 return 1;
 455         }
 456         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 457
 458         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 459                                             count, t))
 460                 return 0;
 461         return 1;
 462 }
 463
 464 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 465 {
 466         struct dx_countlimit *c;
 467         struct dx_tail *t;
 468         int count_offset, limit, count;
 469
 470         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 471                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 472                 return;
 473
 474         c = get_dx_countlimit(inode, dirent, &count_offset);
 475         if (!c) {
 476                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 477                 return;
 478         }
 479         limit = le16_to_cpu(c->limit);
 480         count = le16_to_cpu(c->count);
 481         if (count_offset + (limit * sizeof(struct dx_entry)) >
 482             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 483                 warn_no_space_for_csum(inode);
 484                 return;
 485         }
 486         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 487
 488         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 489 }
 490
 491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 492                                             struct inode *inode,
 493                                             struct buffer_head *bh)
 494 {
 495         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 496         return ext4_handle_dirty_metadata(handle, inode, bh);
 497 }
 498
 499 /*
 500  * p is at least 6 bytes before the end of page
 501  */
 502 static inline struct ext4_dir_entry_2 *
 503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 504 {
 505         return (struct ext4_dir_entry_2 *)((char *)p +
 506                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 507 }
 508
 509 /*
 510  * Future: use high four bits of block for coalesce-on-delete flags
 511  * Mask them off for now.
 512  */
 513
 514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 515 {
 516         return le32_to_cpu(entry->block) & 0x00ffffff;
 517 }
 518
 519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 520 {
 521         entry->block = cpu_to_le32(value);
 522 }
 523
 524 static inline unsigned dx_get_hash(struct dx_entry *entry)
 525 {
 526         return le32_to_cpu(entry->hash);
 527 }
 528
 529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 530 {
 531         entry->hash = cpu_to_le32(value);
 532 }
 533
 534 static inline unsigned dx_get_count(struct dx_entry *entries)
 535 {
 536         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 537 }
 538
 539 static inline unsigned dx_get_limit(struct dx_entry *entries)
 540 {
 541         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 542 }
 543
 544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 545 {
 546         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 547 }
 548
 549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 550 {
 551         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 552 }
 553
 554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 555 {
 556         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 557                 EXT4_DIR_REC_LEN(2) - infosize;
 558
 559         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 560                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 561                 entry_space -= sizeof(struct dx_tail);
 562         return entry_space / sizeof(struct dx_entry);
 563 }
 564
 565 static inline unsigned dx_node_limit(struct inode *dir)
 566 {
 567         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 568
 569         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 570                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 571                 entry_space -= sizeof(struct dx_tail);
 572         return entry_space / sizeof(struct dx_entry);
 573 }
 574
 575 /*
 576  * Debug
 577  */
 578 #ifdef DX_DEBUG
 579 static void dx_show_index(char * label, struct dx_entry *entries)
 580 {
 581         int i, n = dx_get_count (entries);
 582         printk(KERN_DEBUG "%s index ", label);
 583         for (i = 0; i < n; i++) {
 584                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 585                                 0, (unsigned long)dx_get_block(entries + i));
 586         }
 587         printk("\n");
 588 }
 589
 590 struct stats
 591 {
 592         unsigned names;
 593         unsigned space;
 594         unsigned bcount;
 595 };
 596
 597 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 598                                  int size, int show_names)
 599 {
 600         unsigned names = 0, space = 0;
 601         char *base = (char *) de;
 602         struct dx_hash_info h = *hinfo;
 603
 604         printk("names: ");
 605         while ((char *) de < base + size)
 606         {
 607                 if (de->inode)
 608                 {
 609                         if (show_names)
 610                         {
 611                                 int len = de->name_len;
 612                                 char *name = de->name;
 613                                 while (len--) printk("%c", *name++);
 614                                 ext4fs_dirhash(de->name, de->name_len, &h);
 615                                 printk(":%x.%u ", h.hash,
 616                                        (unsigned) ((char *) de - base));
 617                         }
 618                         space += EXT4_DIR_REC_LEN(de->name_len);
 619                         names++;
 620                 }
 621                 de = ext4_next_entry(de, size);
 622         }
 623         printk("(%i)\n", names);
 624         return (struct stats) { names, space, 1 };
 625 }
 626
 627 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 628                              struct dx_entry *entries, int levels)
 629 {
 630         unsigned blocksize = dir->i_sb->s_blocksize;
 631         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 632         unsigned bcount = 0;
 633         struct buffer_head *bh;
 634         int err;
 635         printk("%i indexed blocks...\n", count);
 636         for (i = 0; i < count; i++, entries++)
 637         {
 638                 ext4_lblk_t block = dx_get_block(entries);
 639                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 640                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 641                 struct stats stats;
 642                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 643                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 644                 stats = levels?
 645                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 646                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 647                 names += stats.names;
 648                 space += stats.space;
 649                 bcount += stats.bcount;
 650                 brelse(bh);
 651         }
 652         if (bcount)
 653                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 654                        levels ? "" : "   ", names, space/bcount,
 655                        (space/bcount)*100/blocksize);
 656         return (struct stats) { names, space, bcount};
 657 }
 658 #endif /* DX_DEBUG */
 659
 660 /*
 661  * Probe for a directory leaf block to search.
 662  *
 663  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 664  * error in the directory index, and the caller should fall back to
 665  * searching the directory normally.  The callers of dx_probe **MUST**
 666  * check for this error code, and make sure it never gets reflected
 667  * back to userspace.
 668  */
 669 static struct dx_frame *
 670 dx_probe(const struct qstr *d_name, struct inode *dir,
 671          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 672 {
 673         unsigned count, indirect;
 674         struct dx_entry *at, *entries, *p, *q, *m;
 675         struct dx_root *root;
 676         struct buffer_head *bh;
 677         struct dx_frame *frame = frame_in;
 678         u32 hash;
 679
 680         frame->bh = NULL;
 681         bh = ext4_read_dirblock(dir, 0, INDEX);
 682         if (IS_ERR(bh)) {
 683                 *err = PTR_ERR(bh);
 684                 goto fail;
 685         }
 686         root = (struct dx_root *) bh->b_data;
 687         if (root->info.hash_version != DX_HASH_TEA &&
 688             root->info.hash_version != DX_HASH_HALF_MD4 &&
 689             root->info.hash_version != DX_HASH_LEGACY) {
 690                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 691                              root->info.hash_version);
 692                 brelse(bh);
 693                 *err = ERR_BAD_DX_DIR;
 694                 goto fail;
 695         }
 696         hinfo->hash_version = root->info.hash_version;
 697         if (hinfo->hash_version <= DX_HASH_TEA)
 698                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 699         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 700         if (d_name)
 701                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 702         hash = hinfo->hash;
 703
 704         if (root->info.unused_flags & 1) {
 705                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 706                              root->info.unused_flags);
 707                 brelse(bh);
 708                 *err = ERR_BAD_DX_DIR;
 709                 goto fail;
 710         }
 711
 712         if ((indirect = root->info.indirect_levels) > 1) {
 713                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 714                              root->info.indirect_levels);
 715                 brelse(bh);
 716                 *err = ERR_BAD_DX_DIR;
 717                 goto fail;
 718         }
 719
 720         entries = (struct dx_entry *) (((char *)&root->info) +
 721                                        root->info.info_length);
 722
 723         if (dx_get_limit(entries) != dx_root_limit(dir,
 724                                                    root->info.info_length)) {
 725                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 726                 brelse(bh);
 727                 *err = ERR_BAD_DX_DIR;
 728                 goto fail;
 729         }
 730
 731         dxtrace(printk("Look up %x", hash));
 732         while (1)
 733         {
 734                 count = dx_get_count(entries);
 735                 if (!count || count > dx_get_limit(entries)) {
 736                         ext4_warning(dir->i_sb,
 737                                      "dx entry: no count or count > limit");
 738                         brelse(bh);
 739                         *err = ERR_BAD_DX_DIR;
 740                         goto fail2;
 741                 }
 742
 743                 p = entries + 1;
 744                 q = entries + count - 1;
 745                 while (p <= q)
 746                 {
 747                         m = p + (q - p)/2;
 748                         dxtrace(printk("."));
 749                         if (dx_get_hash(m) > hash)
 750                                 q = m - 1;
 751                         else
 752                                 p = m + 1;
 753                 }
 754
 755                 if (0) // linear search cross check
 756                 {
 757                         unsigned n = count - 1;
 758                         at = entries;
 759                         while (n--)
 760                         {
 761                                 dxtrace(printk(","));
 762                                 if (dx_get_hash(++at) > hash)
 763                                 {
 764                                         at--;
 765                                         break;
 766                                 }
 767                         }
 768                         assert (at == p - 1);
 769                 }
 770
 771                 at = p - 1;
 772                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 773                 frame->bh = bh;
 774                 frame->entries = entries;
 775                 frame->at = at;
 776                 if (!indirect--) return frame;
 777                 bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 778                 if (IS_ERR(bh)) {
 779                         *err = PTR_ERR(bh);
 780                         goto fail2;
 781                 }
 782                 entries = ((struct dx_node *) bh->b_data)->entries;
 783
 784                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 785                         ext4_warning(dir->i_sb,
 786                                      "dx entry: limit != node limit");
 787                         brelse(bh);
 788                         *err = ERR_BAD_DX_DIR;
 789                         goto fail2;
 790                 }
 791                 frame++;
 792                 frame->bh = NULL;
 793         }
 794 fail2:
 795         while (frame >= frame_in) {
 796                 brelse(frame->bh);
 797                 frame--;
 798         }
 799 fail:
 800         if (*err == ERR_BAD_DX_DIR)
 801                 ext4_warning(dir->i_sb,
 802                              "Corrupt dir inode %lu, running e2fsck is "
 803                              "recommended.", dir->i_ino);
 804         return NULL;
 805 }
 806
 807 static void dx_release (struct dx_frame *frames)
 808 {
 809         if (frames[0].bh == NULL)
 810                 return;
 811
 812         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 813                 brelse(frames[1].bh);
 814         brelse(frames[0].bh);
 815 }
 816
 817 /*
 818  * This function increments the frame pointer to search the next leaf
 819  * block, and reads in the necessary intervening nodes if the search
 820  * should be necessary.  Whether or not the search is necessary is
 821  * controlled by the hash parameter.  If the hash value is even, then
 822  * the search is only continued if the next block starts with that
 823  * hash value.  This is used if we are searching for a specific file.
 824  *
 825  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 826  *
 827  * This function returns 1 if the caller should continue to search,
 828  * or 0 if it should not.  If there is an error reading one of the
 829  * index blocks, it will a negative error code.
 830  *
 831  * If start_hash is non-null, it will be filled in with the starting
 832  * hash of the next page.
 833  */
 834 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 835                                  struct dx_frame *frame,
 836                                  struct dx_frame *frames,
 837                                  __u32 *start_hash)
 838 {
 839         struct dx_frame *p;
 840         struct buffer_head *bh;
 841         int num_frames = 0;
 842         __u32 bhash;
 843
 844         p = frame;
 845         /*
 846          * Find the next leaf page by incrementing the frame pointer.
 847          * If we run out of entries in the interior node, loop around and
 848          * increment pointer in the parent node.  When we break out of
 849          * this loop, num_frames indicates the number of interior
 850          * nodes need to be read.
 851          */
 852         while (1) {
 853                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 854                         break;
 855                 if (p == frames)
 856                         return 0;
 857                 num_frames++;
 858                 p--;
 859         }
 860
 861         /*
 862          * If the hash is 1, then continue only if the next page has a
 863          * continuation hash of any value.  This is used for readdir
 864          * handling.  Otherwise, check to see if the hash matches the
 865          * desired contiuation hash.  If it doesn't, return since
 866          * there's no point to read in the successive index pages.
 867          */
 868         bhash = dx_get_hash(p->at);
 869         if (start_hash)
 870                 *start_hash = bhash;
 871         if ((hash & 1) == 0) {
 872                 if ((bhash & ~1) != hash)
 873                         return 0;
 874         }
 875         /*
 876          * If the hash is HASH_NB_ALWAYS, we always go to the next
 877          * block so no check is necessary
 878          */
 879         while (num_frames--) {
 880                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 881                 if (IS_ERR(bh))
 882                         return PTR_ERR(bh);
 883                 p++;
 884                 brelse(p->bh);
 885                 p->bh = bh;
 886                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 887         }
 888         return 1;
 889 }
 890
 891
 892 /*
 893  * This function fills a red-black tree with information from a
 894  * directory block.  It returns the number directory entries loaded
 895  * into the tree.  If there is an error it is returned in err.
 896  */
 897 static int htree_dirblock_to_tree(struct file *dir_file,
 898                                   struct inode *dir, ext4_lblk_t block,
 899                                   struct dx_hash_info *hinfo,
 900                                   __u32 start_hash, __u32 start_minor_hash)
 901 {
 902         struct buffer_head *bh;
 903         struct ext4_dir_entry_2 *de, *top;
 904         int err = 0, count = 0;
 905
 906         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 907                                                         (unsigned long)block));
 908         bh = ext4_read_dirblock(dir, block, DIRENT);
 909         if (IS_ERR(bh))
 910                 return PTR_ERR(bh);
 911
 912         de = (struct ext4_dir_entry_2 *) bh->b_data;
 913         top = (struct ext4_dir_entry_2 *) ((char *) de +
 914                                            dir->i_sb->s_blocksize -
 915                                            EXT4_DIR_REC_LEN(0));
 916         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 917                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 918                                 bh->b_data, bh->b_size,
 919                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 920                                          + ((char *)de - bh->b_data))) {
 921                         /* silently ignore the rest of the block */
 922                         break;
 923                 }
 924                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 925                 if ((hinfo->hash < start_hash) ||
 926                     ((hinfo->hash == start_hash) &&
 927                      (hinfo->minor_hash < start_minor_hash)))
 928                         continue;
 929                 if (de->inode == 0)
 930                         continue;
 931                 if ((err = ext4_htree_store_dirent(dir_file,
 932                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 933                         brelse(bh);
 934                         return err;
 935                 }
 936                 count++;
 937         }
 938         brelse(bh);
 939         return count;
 940 }
 941
 942
 943 /*
 944  * This function fills a red-black tree with information from a
 945  * directory.  We start scanning the directory in hash order, starting
 946  * at start_hash and start_minor_hash.
 947  *
 948  * This function returns the number of entries inserted into the tree,
 949  * or a negative error code.
 950  */
 951 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 952                          __u32 start_minor_hash, __u32 *next_hash)
 953 {
 954         struct dx_hash_info hinfo;
 955         struct ext4_dir_entry_2 *de;
 956         struct dx_frame frames[2], *frame;
 957         struct inode *dir;
 958         ext4_lblk_t block;
 959         int count = 0;
 960         int ret, err;
 961         __u32 hashval;
 962
 963         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 964                        start_hash, start_minor_hash));
 965         dir = file_inode(dir_file);
 966         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 967                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 968                 if (hinfo.hash_version <= DX_HASH_TEA)
 969                         hinfo.hash_version +=
 970                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 971                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 972                 if (ext4_has_inline_data(dir)) {
 973                         int has_inline_data = 1;
 974                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
 975                                                         &hinfo, start_hash,
 976                                                         start_minor_hash,
 977                                                         &has_inline_data);
 978                         if (has_inline_data) {
 979                                 *next_hash = ~0;
 980                                 return count;
 981                         }
 982                 }
 983                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 984                                                start_hash, start_minor_hash);
 985                 *next_hash = ~0;
 986                 return count;
 987         }
 988         hinfo.hash = start_hash;
 989         hinfo.minor_hash = 0;
 990         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 991         if (!frame)
 992                 return err;
 993
 994         /* Add '.' and '..' from the htree header */
 995         if (!start_hash && !start_minor_hash) {
 996                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 997                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 998                         goto errout;
 999                 count++;
1000         }
1001         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1002                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1003                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1004                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1005                         goto errout;
1006                 count++;
1007         }
1008
1009         while (1) {
1010                 block = dx_get_block(frame->at);
1011                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1012                                              start_hash, start_minor_hash);
1013                 if (ret < 0) {
1014                         err = ret;
1015                         goto errout;
1016                 }
1017                 count += ret;
1018                 hashval = ~0;
1019                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1020                                             frame, frames, &hashval);
1021                 *next_hash = hashval;
1022                 if (ret < 0) {
1023                         err = ret;
1024                         goto errout;
1025                 }
1026                 /*
1027                  * Stop if:  (a) there are no more entries, or
1028                  * (b) we have inserted at least one entry and the
1029                  * next hash value is not a continuation
1030                  */
1031                 if ((ret == 0) ||
1032                     (count && ((hashval & 1) == 0)))
1033                         break;
1034         }
1035         dx_release(frames);
1036         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1037                        "next hash: %x\n", count, *next_hash));
1038         return count;
1039 errout:
1040         dx_release(frames);
1041         return (err);
1042 }
1043
1044 static inline int search_dirblock(struct buffer_head *bh,
1045                                   struct inode *dir,
1046                                   const struct qstr *d_name,
1047                                   unsigned int offset,
1048                                   struct ext4_dir_entry_2 **res_dir)
1049 {
1050         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1051                           d_name, offset, res_dir);
1052 }
1053
1054 /*
1055  * Directory block splitting, compacting
1056  */
1057
1058 /*
1059  * Create map of hash values, offsets, and sizes, stored at end of block.
1060  * Returns number of entries mapped.
1061  */
1062 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1063                        struct dx_hash_info *hinfo,
1064                        struct dx_map_entry *map_tail)
1065 {
1066         int count = 0;
1067         char *base = (char *) de;
1068         struct dx_hash_info h = *hinfo;
1069
1070         while ((char *) de < base + blocksize) {
1071                 if (de->name_len && de->inode) {
1072                         ext4fs_dirhash(de->name, de->name_len, &h);
1073                         map_tail--;
1074                         map_tail->hash = h.hash;
1075                         map_tail->offs = ((char *) de - base)>>2;
1076                         map_tail->size = le16_to_cpu(de->rec_len);
1077                         count++;
1078                         cond_resched();
1079                 }
1080                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1081                 de = ext4_next_entry(de, blocksize);
1082         }
1083         return count;
1084 }
1085
1086 /* Sort map by hash value */
1087 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1088 {
1089         struct dx_map_entry *p, *q, *top = map + count - 1;
1090         int more;
1091         /* Combsort until bubble sort doesn't suck */
1092         while (count > 2) {
1093                 count = count*10/13;
1094                 if (count - 9 < 2) /* 9, 10 -> 11 */
1095                         count = 11;
1096                 for (p = top, q = p - count; q >= map; p--, q--)
1097                         if (p->hash < q->hash)
1098                                 swap(*p, *q);
1099         }
1100         /* Garden variety bubble sort */
1101         do {
1102                 more = 0;
1103                 q = top;
1104                 while (q-- > map) {
1105                         if (q[1].hash >= q[0].hash)
1106                                 continue;
1107                         swap(*(q+1), *q);
1108                         more = 1;
1109                 }
1110         } while(more);
1111 }
1112
1113 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1114 {
1115         struct dx_entry *entries = frame->entries;
1116         struct dx_entry *old = frame->at, *new = old + 1;
1117         int count = dx_get_count(entries);
1118
1119         assert(count < dx_get_limit(entries));
1120         assert(old < entries + count);
1121         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1122         dx_set_hash(new, hash);
1123         dx_set_block(new, block);
1124         dx_set_count(entries, count + 1);
1125 }
1126
1127 /*
1128  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1129  *
1130  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1131  * `de != NULL' is guaranteed by caller.
1132  */
1133 static inline int ext4_match (int len, const char * const name,
1134                               struct ext4_dir_entry_2 * de)
1135 {
1136         if (len != de->name_len)
1137                 return 0;
1138         if (!de->inode)
1139                 return 0;
1140         return !memcmp(name, de->name, len);
1141 }
1142
1143 /*
1144  * Returns 0 if not found, -1 on failure, and 1 on success
1145  */
1146 int search_dir(struct buffer_head *bh,
1147                char *search_buf,
1148                int buf_size,
1149                struct inode *dir,
1150                const struct qstr *d_name,
1151                unsigned int offset,
1152                struct ext4_dir_entry_2 **res_dir)
1153 {
1154         struct ext4_dir_entry_2 * de;
1155         char * dlimit;
1156         int de_len;
1157         const char *name = d_name->name;
1158         int namelen = d_name->len;
1159
1160         de = (struct ext4_dir_entry_2 *)search_buf;
1161         dlimit = search_buf + buf_size;
1162         while ((char *) de < dlimit) {
1163                 /* this code is executed quadratically often */
1164                 /* do minimal checking `by hand' */
1165
1166                 if ((char *) de + namelen <= dlimit &&
1167                     ext4_match (namelen, name, de)) {
1168                         /* found a match - just to be sure, do a full check */
1169                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1170                                                  bh->b_size, offset))
1171                                 return -1;
1172                         *res_dir = de;
1173                         return 1;
1174                 }
1175                 /* prevent looping on a bad block */
1176                 de_len = ext4_rec_len_from_disk(de->rec_len,
1177                                                 dir->i_sb->s_blocksize);
1178                 if (de_len <= 0)
1179                         return -1;
1180                 offset += de_len;
1181                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1182         }
1183         return 0;
1184 }
1185
1186 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1187                                struct ext4_dir_entry *de)
1188 {
1189         struct super_block *sb = dir->i_sb;
1190
1191         if (!is_dx(dir))
1192                 return 0;
1193         if (block == 0)
1194                 return 1;
1195         if (de->inode == 0 &&
1196             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1197                         sb->s_blocksize)
1198                 return 1;
1199         return 0;
1200 }
1201
1202 /*
1203  *      ext4_find_entry()
1204  *
1205  * finds an entry in the specified directory with the wanted name. It
1206  * returns the cache buffer in which the entry was found, and the entry
1207  * itself (as a parameter - res_dir). It does NOT read the inode of the
1208  * entry - you'll have to do that yourself if you want to.
1209  *
1210  * The returned buffer_head has ->b_count elevated.  The caller is expected
1211  * to brelse() it when appropriate.
1212  */
1213 static struct buffer_head * ext4_find_entry (struct inode *dir,
1214                                         const struct qstr *d_name,
1215                                         struct ext4_dir_entry_2 **res_dir,
1216                                         int *inlined)
1217 {
1218         struct super_block *sb;
1219         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1220         struct buffer_head *bh, *ret = NULL;
1221         ext4_lblk_t start, block, b;
1222         const u8 *name = d_name->name;
1223         int ra_max = 0;         /* Number of bh's in the readahead
1224                                    buffer, bh_use[] */
1225         int ra_ptr = 0;         /* Current index into readahead
1226                                    buffer */
1227         int num = 0;
1228         ext4_lblk_t  nblocks;
1229         int i, err = 0;
1230         int namelen;
1231
1232         *res_dir = NULL;
1233         sb = dir->i_sb;
1234         namelen = d_name->len;
1235         if (namelen > EXT4_NAME_LEN)
1236                 return NULL;
1237
1238         if (ext4_has_inline_data(dir)) {
1239                 int has_inline_data = 1;
1240                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1241                                              &has_inline_data);
1242                 if (has_inline_data) {
1243                         if (inlined)
1244                                 *inlined = 1;
1245                         return ret;
1246                 }
1247         }
1248
1249         if ((namelen <= 2) && (name[0] == '.') &&
1250             (name[1] == '.' || name[1] == '\0')) {
1251                 /*
1252                  * "." or ".." will only be in the first block
1253                  * NFS may look up ".."; "." should be handled by the VFS
1254                  */
1255                 block = start = 0;
1256                 nblocks = 1;
1257                 goto restart;
1258         }
1259         if (is_dx(dir)) {
1260                 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1261                 /*
1262                  * On success, or if the error was file not found,
1263                  * return.  Otherwise, fall back to doing a search the
1264                  * old fashioned way.
1265                  */
1266                 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1267                         return bh;
1268                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1269                                "falling back\n"));
1270         }
1271         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1272         start = EXT4_I(dir)->i_dir_start_lookup;
1273         if (start >= nblocks)
1274                 start = 0;
1275         block = start;
1276 restart:
1277         do {
1278                 /*
1279                  * We deal with the read-ahead logic here.
1280                  */
1281                 if (ra_ptr >= ra_max) {
1282                         /* Refill the readahead buffer */
1283                         ra_ptr = 0;
1284                         b = block;
1285                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1286                                 /*
1287                                  * Terminate if we reach the end of the
1288                                  * directory and must wrap, or if our
1289                                  * search has finished at this block.
1290                                  */
1291                                 if (b >= nblocks || (num && block == start)) {
1292                                         bh_use[ra_max] = NULL;
1293                                         break;
1294                                 }
1295                                 num++;
1296                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1297                                 if (unlikely(err)) {
1298                                         if (ra_max == 0)
1299                                                 return ERR_PTR(err);
1300                                         break;
1301                                 }
1302                                 bh_use[ra_max] = bh;
1303                                 if (bh)
1304                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1305                                                     1, &bh);
1306                         }
1307                 }
1308                 if ((bh = bh_use[ra_ptr++]) == NULL)
1309                         goto next;
1310                 wait_on_buffer(bh);
1311                 if (!buffer_uptodate(bh)) {
1312                         /* read error, skip block & hope for the best */
1313                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1314                                          (unsigned long) block);
1315                         brelse(bh);
1316                         goto next;
1317                 }
1318                 if (!buffer_verified(bh) &&
1319                     !is_dx_internal_node(dir, block,
1320                                          (struct ext4_dir_entry *)bh->b_data) &&
1321                     !ext4_dirent_csum_verify(dir,
1322                                 (struct ext4_dir_entry *)bh->b_data)) {
1323                         EXT4_ERROR_INODE(dir, "checksumming directory "
1324                                          "block %lu", (unsigned long)block);
1325                         brelse(bh);
1326                         goto next;
1327                 }
1328                 set_buffer_verified(bh);
1329                 i = search_dirblock(bh, dir, d_name,
1330                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1331                 if (i == 1) {
1332                         EXT4_I(dir)->i_dir_start_lookup = block;
1333                         ret = bh;
1334                         goto cleanup_and_exit;
1335                 } else {
1336                         brelse(bh);
1337                         if (i < 0)
1338                                 goto cleanup_and_exit;
1339                 }
1340         next:
1341                 if (++block >= nblocks)
1342                         block = 0;
1343         } while (block != start);
1344
1345         /*
1346          * If the directory has grown while we were searching, then
1347          * search the last part of the directory before giving up.
1348          */
1349         block = nblocks;
1350         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1351         if (block < nblocks) {
1352                 start = 0;
1353                 goto restart;
1354         }
1355
1356 cleanup_and_exit:
1357         /* Clean up the read-ahead blocks */
1358         for (; ra_ptr < ra_max; ra_ptr++)
1359                 brelse(bh_use[ra_ptr]);
1360         return ret;
1361 }
1362
1363 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1364                        struct ext4_dir_entry_2 **res_dir)
1365 {
1366         struct super_block * sb = dir->i_sb;
1367         struct dx_hash_info     hinfo;
1368         struct dx_frame frames[2], *frame;
1369         struct buffer_head *bh;
1370         ext4_lblk_t block;
1371         int err = 0, retval;
1372
1373         frame = dx_probe(d_name, dir, &hinfo, frames, &err);
1374         if (err)
1375                 return ERR_PTR(err);
1376         do {
1377                 block = dx_get_block(frame->at);
1378                 bh = ext4_read_dirblock(dir, block, DIRENT);
1379                 if (IS_ERR(bh))
1380                         goto errout;
1381
1382                 retval = search_dirblock(bh, dir, d_name,
1383                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1384                                          res_dir);
1385                 if (retval == 1)
1386                         goto success;
1387                 brelse(bh);
1388                 if (retval == -1) {
1389                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1390                         goto errout;
1391                 }
1392
1393                 /* Check to see if we should continue to search */
1394                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1395                                                frames, NULL);
1396                 if (retval < 0) {
1397                         ext4_warning(sb,
1398                              "error %d reading index page in directory #%lu",
1399                              retval, dir->i_ino);
1400                         bh = ERR_PTR(retval);
1401                         goto errout;
1402                 }
1403         } while (retval == 1);
1404
1405         bh = NULL;
1406 errout:
1407         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1408 success:
1409         dx_release(frames);
1410         return bh;
1411 }
1412
1413 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1414 {
1415         struct inode *inode;
1416         struct ext4_dir_entry_2 *de;
1417         struct buffer_head *bh;
1418
1419         if (dentry->d_name.len > EXT4_NAME_LEN)
1420                 return ERR_PTR(-ENAMETOOLONG);
1421
1422         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1423         if (IS_ERR(bh))
1424                 return (struct dentry *) bh;
1425         inode = NULL;
1426         if (bh) {
1427                 __u32 ino = le32_to_cpu(de->inode);
1428                 brelse(bh);
1429                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1430                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1431                         return ERR_PTR(-EIO);
1432                 }
1433                 if (unlikely(ino == dir->i_ino)) {
1434                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1435                                          dentry);
1436                         return ERR_PTR(-EIO);
1437                 }
1438                 inode = ext4_iget(dir->i_sb, ino);
1439                 if (inode == ERR_PTR(-ESTALE)) {
1440                         EXT4_ERROR_INODE(dir,
1441                                          "deleted inode referenced: %u",
1442                                          ino);
1443                         return ERR_PTR(-EIO);
1444                 }
1445         }
1446         return d_splice_alias(inode, dentry);
1447 }
1448
1449
1450 struct dentry *ext4_get_parent(struct dentry *child)
1451 {
1452         __u32 ino;
1453         static const struct qstr dotdot = QSTR_INIT("..", 2);
1454         struct ext4_dir_entry_2 * de;
1455         struct buffer_head *bh;
1456
1457         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1458         if (IS_ERR(bh))
1459                 return (struct dentry *) bh;
1460         if (!bh)
1461                 return ERR_PTR(-ENOENT);
1462         ino = le32_to_cpu(de->inode);
1463         brelse(bh);
1464
1465         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1466                 EXT4_ERROR_INODE(child->d_inode,
1467                                  "bad parent inode number: %u", ino);
1468                 return ERR_PTR(-EIO);
1469         }
1470
1471         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1472 }
1473
1474 /*
1475  * Move count entries from end of map between two memory locations.
1476  * Returns pointer to last entry moved.
1477  */
1478 static struct ext4_dir_entry_2 *
1479 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1480                 unsigned blocksize)
1481 {
1482         unsigned rec_len = 0;
1483
1484         while (count--) {
1485                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1486                                                 (from + (map->offs<<2));
1487                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1488                 memcpy (to, de, rec_len);
1489                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1490                                 ext4_rec_len_to_disk(rec_len, blocksize);
1491                 de->inode = 0;
1492                 map++;
1493                 to += rec_len;
1494         }
1495         return (struct ext4_dir_entry_2 *) (to - rec_len);
1496 }
1497
1498 /*
1499  * Compact each dir entry in the range to the minimal rec_len.
1500  * Returns pointer to last entry in range.
1501  */
1502 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1503 {
1504         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1505         unsigned rec_len = 0;
1506
1507         prev = to = de;
1508         while ((char*)de < base + blocksize) {
1509                 next = ext4_next_entry(de, blocksize);
1510                 if (de->inode && de->name_len) {
1511                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1512                         if (de > to)
1513                                 memmove(to, de, rec_len);
1514                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1515                         prev = to;
1516                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1517                 }
1518                 de = next;
1519         }
1520         return prev;
1521 }
1522
1523 /*
1524  * Split a full leaf block to make room for a new dir entry.
1525  * Allocate a new block, and move entries so that they are approx. equally full.
1526  * Returns pointer to de in block into which the new entry will be inserted.
1527  */
1528 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1529                         struct buffer_head **bh,struct dx_frame *frame,
1530                         struct dx_hash_info *hinfo, int *error)
1531 {
1532         unsigned blocksize = dir->i_sb->s_blocksize;
1533         unsigned count, continued;
1534         struct buffer_head *bh2;
1535         ext4_lblk_t newblock;
1536         u32 hash2;
1537         struct dx_map_entry *map;
1538         char *data1 = (*bh)->b_data, *data2;
1539         unsigned split, move, size;
1540         struct ext4_dir_entry_2 *de = NULL, *de2;
1541         struct ext4_dir_entry_tail *t;
1542         int     csum_size = 0;
1543         int     err = 0, i;
1544
1545         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1546                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1547                 csum_size = sizeof(struct ext4_dir_entry_tail);
1548
1549         bh2 = ext4_append(handle, dir, &newblock);
1550         if (IS_ERR(bh2)) {
1551                 brelse(*bh);
1552                 *bh = NULL;
1553                 *error = PTR_ERR(bh2);
1554                 return NULL;
1555         }
1556
1557         BUFFER_TRACE(*bh, "get_write_access");
1558         err = ext4_journal_get_write_access(handle, *bh);
1559         if (err)
1560                 goto journal_error;
1561
1562         BUFFER_TRACE(frame->bh, "get_write_access");
1563         err = ext4_journal_get_write_access(handle, frame->bh);
1564         if (err)
1565                 goto journal_error;
1566
1567         data2 = bh2->b_data;
1568
1569         /* create map in the end of data2 block */
1570         map = (struct dx_map_entry *) (data2 + blocksize);
1571         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1572                              blocksize, hinfo, map);
1573         map -= count;
1574         dx_sort_map(map, count);
1575         /* Split the existing block in the middle, size-wise */
1576         size = 0;
1577         move = 0;
1578         for (i = count-1; i >= 0; i--) {
1579                 /* is more than half of this entry in 2nd half of the block? */
1580                 if (size + map[i].size/2 > blocksize/2)
1581                         break;
1582                 size += map[i].size;
1583                 move++;
1584         }
1585         /* map index at which we will split */
1586         split = count - move;
1587         hash2 = map[split].hash;
1588         continued = hash2 == map[split - 1].hash;
1589         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1590                         (unsigned long)dx_get_block(frame->at),
1591                                         hash2, split, count-split));
1592
1593         /* Fancy dance to stay within two buffers */
1594         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1595         de = dx_pack_dirents(data1, blocksize);
1596         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1597                                            (char *) de,
1598                                            blocksize);
1599         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1600                                             (char *) de2,
1601                                             blocksize);
1602         if (csum_size) {
1603                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1604                 initialize_dirent_tail(t, blocksize);
1605
1606                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1607                 initialize_dirent_tail(t, blocksize);
1608         }
1609
1610         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1611         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1612
1613         /* Which block gets the new entry? */
1614         if (hinfo->hash >= hash2)
1615         {
1616                 swap(*bh, bh2);
1617                 de = de2;
1618         }
1619         dx_insert_block(frame, hash2 + continued, newblock);
1620         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1621         if (err)
1622                 goto journal_error;
1623         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1624         if (err)
1625                 goto journal_error;
1626         brelse(bh2);
1627         dxtrace(dx_show_index("frame", frame->entries));
1628         return de;
1629
1630 journal_error:
1631         brelse(*bh);
1632         brelse(bh2);
1633         *bh = NULL;
1634         ext4_std_error(dir->i_sb, err);
1635         *error = err;
1636         return NULL;
1637 }
1638
1639 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1640                       struct buffer_head *bh,
1641                       void *buf, int buf_size,
1642                       const char *name, int namelen,
1643                       struct ext4_dir_entry_2 **dest_de)
1644 {
1645         struct ext4_dir_entry_2 *de;
1646         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1647         int nlen, rlen;
1648         unsigned int offset = 0;
1649         char *top;
1650
1651         de = (struct ext4_dir_entry_2 *)buf;
1652         top = buf + buf_size - reclen;
1653         while ((char *) de <= top) {
1654                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1655                                          buf, buf_size, offset))
1656                         return -EIO;
1657                 if (ext4_match(namelen, name, de))
1658                         return -EEXIST;
1659                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1660                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1661                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1662                         break;
1663                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1664                 offset += rlen;
1665         }
1666         if ((char *) de > top)
1667                 return -ENOSPC;
1668
1669         *dest_de = de;
1670         return 0;
1671 }
1672
1673 void ext4_insert_dentry(struct inode *inode,
1674                         struct ext4_dir_entry_2 *de,
1675                         int buf_size,
1676                         const char *name, int namelen)
1677 {
1678
1679         int nlen, rlen;
1680
1681         nlen = EXT4_DIR_REC_LEN(de->name_len);
1682         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1683         if (de->inode) {
1684                 struct ext4_dir_entry_2 *de1 =
1685                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1686                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1687                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1688                 de = de1;
1689         }
1690         de->file_type = EXT4_FT_UNKNOWN;
1691         de->inode = cpu_to_le32(inode->i_ino);
1692         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1693         de->name_len = namelen;
1694         memcpy(de->name, name, namelen);
1695 }
1696 /*
1697  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1698  * it points to a directory entry which is guaranteed to be large
1699  * enough for new directory entry.  If de is NULL, then
1700  * add_dirent_to_buf will attempt search the directory block for
1701  * space.  It will return -ENOSPC if no space is available, and -EIO
1702  * and -EEXIST if directory entry already exists.
1703  */
1704 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1705                              struct inode *inode, struct ext4_dir_entry_2 *de,
1706                              struct buffer_head *bh)
1707 {
1708         struct inode    *dir = dentry->d_parent->d_inode;
1709         const char      *name = dentry->d_name.name;
1710         int             namelen = dentry->d_name.len;
1711         unsigned int    blocksize = dir->i_sb->s_blocksize;
1712         int             csum_size = 0;
1713         int             err;
1714
1715         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1716                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1717                 csum_size = sizeof(struct ext4_dir_entry_tail);
1718
1719         if (!de) {
1720                 err = ext4_find_dest_de(dir, inode,
1721                                         bh, bh->b_data, blocksize - csum_size,
1722                                         name, namelen, &de);
1723                 if (err)
1724                         return err;
1725         }
1726         BUFFER_TRACE(bh, "get_write_access");
1727         err = ext4_journal_get_write_access(handle, bh);
1728         if (err) {
1729                 ext4_std_error(dir->i_sb, err);
1730                 return err;
1731         }
1732
1733         /* By now the buffer is marked for journaling */
1734         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1735
1736         /*
1737          * XXX shouldn't update any times until successful
1738          * completion of syscall, but too many callers depend
1739          * on this.
1740          *
1741          * XXX similarly, too many callers depend on
1742          * ext4_new_inode() setting the times, but error
1743          * recovery deletes the inode, so the worst that can
1744          * happen is that the times are slightly out of date
1745          * and/or different from the directory change time.
1746          */
1747         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1748         ext4_update_dx_flag(dir);
1749         dir->i_version++;
1750         ext4_mark_inode_dirty(handle, dir);
1751         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1752         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1753         if (err)
1754                 ext4_std_error(dir->i_sb, err);
1755         return 0;
1756 }
1757
1758 /*
1759  * This converts a one block unindexed directory to a 3 block indexed
1760  * directory, and adds the dentry to the indexed directory.
1761  */
1762 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1763                             struct inode *inode, struct buffer_head *bh)
1764 {
1765         struct inode    *dir = dentry->d_parent->d_inode;
1766         const char      *name = dentry->d_name.name;
1767         int             namelen = dentry->d_name.len;
1768         struct buffer_head *bh2;
1769         struct dx_root  *root;
1770         struct dx_frame frames[2], *frame;
1771         struct dx_entry *entries;
1772         struct ext4_dir_entry_2 *de, *de2;
1773         struct ext4_dir_entry_tail *t;
1774         char            *data1, *top;
1775         unsigned        len;
1776         int             retval;
1777         unsigned        blocksize;
1778         struct dx_hash_info hinfo;
1779         ext4_lblk_t  block;
1780         struct fake_dirent *fde;
1781         int             csum_size = 0;
1782
1783         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1784                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1785                 csum_size = sizeof(struct ext4_dir_entry_tail);
1786
1787         blocksize =  dir->i_sb->s_blocksize;
1788         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1789         BUFFER_TRACE(bh, "get_write_access");
1790         retval = ext4_journal_get_write_access(handle, bh);
1791         if (retval) {
1792                 ext4_std_error(dir->i_sb, retval);
1793                 brelse(bh);
1794                 return retval;
1795         }
1796         root = (struct dx_root *) bh->b_data;
1797
1798         /* The 0th block becomes the root, move the dirents out */
1799         fde = &root->dotdot;
1800         de = (struct ext4_dir_entry_2 *)((char *)fde +
1801                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1802         if ((char *) de >= (((char *) root) + blocksize)) {
1803                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1804                 brelse(bh);
1805                 return -EIO;
1806         }
1807         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1808
1809         /* Allocate new block for the 0th block's dirents */
1810         bh2 = ext4_append(handle, dir, &block);
1811         if (IS_ERR(bh2)) {
1812                 brelse(bh);
1813                 return PTR_ERR(bh2);
1814         }
1815         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1816         data1 = bh2->b_data;
1817
1818         memcpy (data1, de, len);
1819         de = (struct ext4_dir_entry_2 *) data1;
1820         top = data1 + len;
1821         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1822                 de = de2;
1823         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1824                                            (char *) de,
1825                                            blocksize);
1826
1827         if (csum_size) {
1828                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1829                 initialize_dirent_tail(t, blocksize);
1830         }
1831
1832         /* Initialize the root; the dot dirents already exist */
1833         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1834         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1835                                            blocksize);
1836         memset (&root->info, 0, sizeof(root->info));
1837         root->info.info_length = sizeof(root->info);
1838         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1839         entries = root->entries;
1840         dx_set_block(entries, 1);
1841         dx_set_count(entries, 1);
1842         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1843
1844         /* Initialize as for dx_probe */
1845         hinfo.hash_version = root->info.hash_version;
1846         if (hinfo.hash_version <= DX_HASH_TEA)
1847                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1848         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1849         ext4fs_dirhash(name, namelen, &hinfo);
1850         frame = frames;
1851         frame->entries = entries;
1852         frame->at = entries;
1853         frame->bh = bh;
1854         bh = bh2;
1855
1856         ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1857         ext4_handle_dirty_dirent_node(handle, dir, bh);
1858
1859         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1860         if (!de) {
1861                 /*
1862                  * Even if the block split failed, we have to properly write
1863                  * out all the changes we did so far. Otherwise we can end up
1864                  * with corrupted filesystem.
1865                  */
1866                 ext4_mark_inode_dirty(handle, dir);
1867                 dx_release(frames);
1868                 return retval;
1869         }
1870         dx_release(frames);
1871
1872         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1873         brelse(bh);
1874         return retval;
1875 }
1876
1877 /*
1878  *      ext4_add_entry()
1879  *
1880  * adds a file entry to the specified directory, using the same
1881  * semantics as ext4_find_entry(). It returns NULL if it failed.
1882  *
1883  * NOTE!! The inode part of 'de' is left at 0 - which means you
1884  * may not sleep between calling this and putting something into
1885  * the entry, as someone else might have used it while you slept.
1886  */
1887 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1888                           struct inode *inode)
1889 {
1890         struct inode *dir = dentry->d_parent->d_inode;
1891         struct buffer_head *bh;
1892         struct ext4_dir_entry_2 *de;
1893         struct ext4_dir_entry_tail *t;
1894         struct super_block *sb;
1895         int     retval;
1896         int     dx_fallback=0;
1897         unsigned blocksize;
1898         ext4_lblk_t block, blocks;
1899         int     csum_size = 0;
1900
1901         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1902                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1903                 csum_size = sizeof(struct ext4_dir_entry_tail);
1904
1905         sb = dir->i_sb;
1906         blocksize = sb->s_blocksize;
1907         if (!dentry->d_name.len)
1908                 return -EINVAL;
1909
1910         if (ext4_has_inline_data(dir)) {
1911                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1912                 if (retval < 0)
1913                         return retval;
1914                 if (retval == 1) {
1915                         retval = 0;
1916                         return retval;
1917                 }
1918         }
1919
1920         if (is_dx(dir)) {
1921                 retval = ext4_dx_add_entry(handle, dentry, inode);
1922                 if (!retval || (retval != ERR_BAD_DX_DIR))
1923                         return retval;
1924                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1925                 dx_fallback++;
1926                 ext4_mark_inode_dirty(handle, dir);
1927         }
1928         blocks = dir->i_size >> sb->s_blocksize_bits;
1929         for (block = 0; block < blocks; block++) {
1930                 bh = ext4_read_dirblock(dir, block, DIRENT);
1931                 if (IS_ERR(bh))
1932                         return PTR_ERR(bh);
1933
1934                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1935                 if (retval != -ENOSPC) {
1936                         brelse(bh);
1937                         return retval;
1938                 }
1939
1940                 if (blocks == 1 && !dx_fallback &&
1941                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1942                         return make_indexed_dir(handle, dentry, inode, bh);
1943                 brelse(bh);
1944         }
1945         bh = ext4_append(handle, dir, &block);
1946         if (IS_ERR(bh))
1947                 return PTR_ERR(bh);
1948         de = (struct ext4_dir_entry_2 *) bh->b_data;
1949         de->inode = 0;
1950         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1951
1952         if (csum_size) {
1953                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1954                 initialize_dirent_tail(t, blocksize);
1955         }
1956
1957         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1958         brelse(bh);
1959         if (retval == 0)
1960                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1961         return retval;
1962 }
1963
1964 /*
1965  * Returns 0 for success, or a negative error value
1966  */
1967 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1968                              struct inode *inode)
1969 {
1970         struct dx_frame frames[2], *frame;
1971         struct dx_entry *entries, *at;
1972         struct dx_hash_info hinfo;
1973         struct buffer_head *bh;
1974         struct inode *dir = dentry->d_parent->d_inode;
1975         struct super_block *sb = dir->i_sb;
1976         struct ext4_dir_entry_2 *de;
1977         int err;
1978
1979         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1980         if (!frame)
1981                 return err;
1982         entries = frame->entries;
1983         at = frame->at;
1984         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1985         if (IS_ERR(bh)) {
1986                 err = PTR_ERR(bh);
1987                 bh = NULL;
1988                 goto cleanup;
1989         }
1990
1991         BUFFER_TRACE(bh, "get_write_access");
1992         err = ext4_journal_get_write_access(handle, bh);
1993         if (err)
1994                 goto journal_error;
1995
1996         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1997         if (err != -ENOSPC)
1998                 goto cleanup;
1999
2000         /* Block full, should compress but for now just split */
2001         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2002                        dx_get_count(entries), dx_get_limit(entries)));
2003         /* Need to split index? */
2004         if (dx_get_count(entries) == dx_get_limit(entries)) {
2005                 ext4_lblk_t newblock;
2006                 unsigned icount = dx_get_count(entries);
2007                 int levels = frame - frames;
2008                 struct dx_entry *entries2;
2009                 struct dx_node *node2;
2010                 struct buffer_head *bh2;
2011
2012                 if (levels && (dx_get_count(frames->entries) ==
2013                                dx_get_limit(frames->entries))) {
2014                         ext4_warning(sb, "Directory index full!");
2015                         err = -ENOSPC;
2016                         goto cleanup;
2017                 }
2018                 bh2 = ext4_append(handle, dir, &newblock);
2019                 if (IS_ERR(bh2)) {
2020                         err = PTR_ERR(bh2);
2021                         goto cleanup;
2022                 }
2023                 node2 = (struct dx_node *)(bh2->b_data);
2024                 entries2 = node2->entries;
2025                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2026                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2027                                                            sb->s_blocksize);
2028                 BUFFER_TRACE(frame->bh, "get_write_access");
2029                 err = ext4_journal_get_write_access(handle, frame->bh);
2030                 if (err)
2031                         goto journal_error;
2032                 if (levels) {
2033                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2034                         unsigned hash2 = dx_get_hash(entries + icount1);
2035                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2036                                        icount1, icount2));
2037
2038                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2039                         err = ext4_journal_get_write_access(handle,
2040                                                              frames[0].bh);
2041                         if (err)
2042                                 goto journal_error;
2043
2044                         memcpy((char *) entries2, (char *) (entries + icount1),
2045                                icount2 * sizeof(struct dx_entry));
2046                         dx_set_count(entries, icount1);
2047                         dx_set_count(entries2, icount2);
2048                         dx_set_limit(entries2, dx_node_limit(dir));
2049
2050                         /* Which index block gets the new entry? */
2051                         if (at - entries >= icount1) {
2052                                 frame->at = at = at - entries - icount1 + entries2;
2053                                 frame->entries = entries = entries2;
2054                                 swap(frame->bh, bh2);
2055                         }
2056                         dx_insert_block(frames + 0, hash2, newblock);
2057                         dxtrace(dx_show_index("node", frames[1].entries));
2058                         dxtrace(dx_show_index("node",
2059                                ((struct dx_node *) bh2->b_data)->entries));
2060                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2061                         if (err)
2062                                 goto journal_error;
2063                         brelse (bh2);
2064                 } else {
2065                         dxtrace(printk(KERN_DEBUG
2066                                        "Creating second level index...\n"));
2067                         memcpy((char *) entries2, (char *) entries,
2068                                icount * sizeof(struct dx_entry));
2069                         dx_set_limit(entries2, dx_node_limit(dir));
2070
2071                         /* Set up root */
2072                         dx_set_count(entries, 1);
2073                         dx_set_block(entries + 0, newblock);
2074                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2075
2076                         /* Add new access path frame */
2077                         frame = frames + 1;
2078                         frame->at = at = at - entries + entries2;
2079                         frame->entries = entries = entries2;
2080                         frame->bh = bh2;
2081                         err = ext4_journal_get_write_access(handle,
2082                                                              frame->bh);
2083                         if (err)
2084                                 goto journal_error;
2085                 }
2086                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2087                 if (err) {
2088                         ext4_std_error(inode->i_sb, err);
2089                         goto cleanup;
2090                 }
2091         }
2092         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2093         if (!de)
2094                 goto cleanup;
2095         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2096         goto cleanup;
2097
2098 journal_error:
2099         ext4_std_error(dir->i_sb, err);
2100 cleanup:
2101         brelse(bh);
2102         dx_release(frames);
2103         return err;
2104 }
2105
2106 /*
2107  * ext4_generic_delete_entry deletes a directory entry by merging it
2108  * with the previous entry
2109  */
2110 int ext4_generic_delete_entry(handle_t *handle,
2111                               struct inode *dir,
2112                               struct ext4_dir_entry_2 *de_del,
2113                               struct buffer_head *bh,
2114                               void *entry_buf,
2115                               int buf_size,
2116                               int csum_size)
2117 {
2118         struct ext4_dir_entry_2 *de, *pde;
2119         unsigned int blocksize = dir->i_sb->s_blocksize;
2120         int i;
2121
2122         i = 0;
2123         pde = NULL;
2124         de = (struct ext4_dir_entry_2 *)entry_buf;
2125         while (i < buf_size - csum_size) {
2126                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2127                                          bh->b_data, bh->b_size, i))
2128                         return -EIO;
2129                 if (de == de_del)  {
2130                         if (pde)
2131                                 pde->rec_len = ext4_rec_len_to_disk(
2132                                         ext4_rec_len_from_disk(pde->rec_len,
2133                                                                blocksize) +
2134                                         ext4_rec_len_from_disk(de->rec_len,
2135                                                                blocksize),
2136                                         blocksize);
2137                         else
2138                                 de->inode = 0;
2139                         dir->i_version++;
2140                         return 0;
2141                 }
2142                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2143                 pde = de;
2144                 de = ext4_next_entry(de, blocksize);
2145         }
2146         return -ENOENT;
2147 }
2148
2149 static int ext4_delete_entry(handle_t *handle,
2150                              struct inode *dir,
2151                              struct ext4_dir_entry_2 *de_del,
2152                              struct buffer_head *bh)
2153 {
2154         int err, csum_size = 0;
2155
2156         if (ext4_has_inline_data(dir)) {
2157                 int has_inline_data = 1;
2158                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2159                                                &has_inline_data);
2160                 if (has_inline_data)
2161                         return err;
2162         }
2163
2164         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2165                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2166                 csum_size = sizeof(struct ext4_dir_entry_tail);
2167
2168         BUFFER_TRACE(bh, "get_write_access");
2169         err = ext4_journal_get_write_access(handle, bh);
2170         if (unlikely(err))
2171                 goto out;
2172
2173         err = ext4_generic_delete_entry(handle, dir, de_del,
2174                                         bh, bh->b_data,
2175                                         dir->i_sb->s_blocksize, csum_size);
2176         if (err)
2177                 goto out;
2178
2179         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2180         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2181         if (unlikely(err))
2182                 goto out;
2183
2184         return 0;
2185 out:
2186         if (err != -ENOENT)
2187                 ext4_std_error(dir->i_sb, err);
2188         return err;
2189 }
2190
2191 /*
2192  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2193  * since this indicates that nlinks count was previously 1.
2194  */
2195 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2196 {
2197         inc_nlink(inode);
2198         if (is_dx(inode) && inode->i_nlink > 1) {
2199                 /* limit is 16-bit i_links_count */
2200                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2201                         set_nlink(inode, 1);
2202                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2203                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2204                 }
2205         }
2206 }
2207
2208 /*
2209  * If a directory had nlink == 1, then we should let it be 1. This indicates
2210  * directory has >EXT4_LINK_MAX subdirs.
2211  */
2212 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2213 {
2214         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2215                 drop_nlink(inode);
2216 }
2217
2218
2219 static int ext4_add_nondir(handle_t *handle,
2220                 struct dentry *dentry, struct inode *inode)
2221 {
2222         int err = ext4_add_entry(handle, dentry, inode);
2223         if (!err) {
2224                 ext4_mark_inode_dirty(handle, inode);
2225                 unlock_new_inode(inode);
2226                 d_instantiate(dentry, inode);
2227                 return 0;
2228         }
2229         drop_nlink(inode);
2230         unlock_new_inode(inode);
2231         iput(inode);
2232         return err;
2233 }
2234
2235 /*
2236  * By the time this is called, we already have created
2237  * the directory cache entry for the new file, but it
2238  * is so far negative - it has no inode.
2239  *
2240  * If the create succeeds, we fill in the inode information
2241  * with d_instantiate().
2242  */
2243 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2244                        bool excl)
2245 {
2246         handle_t *handle;
2247         struct inode *inode;
2248         int err, credits, retries = 0;
2249
2250         dquot_initialize(dir);
2251
2252         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2253                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2254 retry:
2255         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2256                                             NULL, EXT4_HT_DIR, credits);
2257         handle = ext4_journal_current_handle();
2258         err = PTR_ERR(inode);
2259         if (!IS_ERR(inode)) {
2260                 inode->i_op = &ext4_file_inode_operations;
2261                 inode->i_fop = &ext4_file_operations;
2262                 ext4_set_aops(inode);
2263                 err = ext4_add_nondir(handle, dentry, inode);
2264                 if (!err && IS_DIRSYNC(dir))
2265                         ext4_handle_sync(handle);
2266         }
2267         if (handle)
2268                 ext4_journal_stop(handle);
2269         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2270                 goto retry;
2271         return err;
2272 }
2273
2274 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2275                       umode_t mode, dev_t rdev)
2276 {
2277         handle_t *handle;
2278         struct inode *inode;
2279         int err, credits, retries = 0;
2280
2281         if (!new_valid_dev(rdev))
2282                 return -EINVAL;
2283
2284         dquot_initialize(dir);
2285
2286         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2287                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2288 retry:
2289         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2290                                             NULL, EXT4_HT_DIR, credits);
2291         handle = ext4_journal_current_handle();
2292         err = PTR_ERR(inode);
2293         if (!IS_ERR(inode)) {
2294                 init_special_inode(inode, inode->i_mode, rdev);
2295                 inode->i_op = &ext4_special_inode_operations;
2296                 err = ext4_add_nondir(handle, dentry, inode);
2297                 if (!err && IS_DIRSYNC(dir))
2298                         ext4_handle_sync(handle);
2299         }
2300         if (handle)
2301                 ext4_journal_stop(handle);
2302         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2303                 goto retry;
2304         return err;
2305 }
2306
2307 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2308 {
2309         handle_t *handle;
2310         struct inode *inode;
2311         int err, retries = 0;
2312
2313         dquot_initialize(dir);
2314
2315 retry:
2316         inode = ext4_new_inode_start_handle(dir, mode,
2317                                             NULL, 0, NULL,
2318                                             EXT4_HT_DIR,
2319                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2320                           4 + EXT4_XATTR_TRANS_BLOCKS);
2321         handle = ext4_journal_current_handle();
2322         err = PTR_ERR(inode);
2323         if (!IS_ERR(inode)) {
2324                 inode->i_op = &ext4_file_inode_operations;
2325                 inode->i_fop = &ext4_file_operations;
2326                 ext4_set_aops(inode);
2327                 d_tmpfile(dentry, inode);
2328                 err = ext4_orphan_add(handle, inode);
2329                 if (err)
2330                         goto err_unlock_inode;
2331                 mark_inode_dirty(inode);
2332                 unlock_new_inode(inode);
2333         }
2334         if (handle)
2335                 ext4_journal_stop(handle);
2336         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2337                 goto retry;
2338         return err;
2339 err_unlock_inode:
2340         ext4_journal_stop(handle);
2341         unlock_new_inode(inode);
2342         return err;
2343 }
2344
2345 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2346                           struct ext4_dir_entry_2 *de,
2347                           int blocksize, int csum_size,
2348                           unsigned int parent_ino, int dotdot_real_len)
2349 {
2350         de->inode = cpu_to_le32(inode->i_ino);
2351         de->name_len = 1;
2352         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2353                                            blocksize);
2354         strcpy(de->name, ".");
2355         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2356
2357         de = ext4_next_entry(de, blocksize);
2358         de->inode = cpu_to_le32(parent_ino);
2359         de->name_len = 2;
2360         if (!dotdot_real_len)
2361                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2362                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2363                                         blocksize);
2364         else
2365                 de->rec_len = ext4_rec_len_to_disk(
2366                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2367         strcpy(de->name, "..");
2368         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2369
2370         return ext4_next_entry(de, blocksize);
2371 }
2372
2373 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2374                              struct inode *inode)
2375 {
2376         struct buffer_head *dir_block = NULL;
2377         struct ext4_dir_entry_2 *de;
2378         struct ext4_dir_entry_tail *t;
2379         ext4_lblk_t block = 0;
2380         unsigned int blocksize = dir->i_sb->s_blocksize;
2381         int csum_size = 0;
2382         int err;
2383
2384         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2385                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2386                 csum_size = sizeof(struct ext4_dir_entry_tail);
2387
2388         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2389                 err = ext4_try_create_inline_dir(handle, dir, inode);
2390                 if (err < 0 && err != -ENOSPC)
2391                         goto out;
2392                 if (!err)
2393                         goto out;
2394         }
2395
2396         inode->i_size = 0;
2397         dir_block = ext4_append(handle, inode, &block);
2398         if (IS_ERR(dir_block))
2399                 return PTR_ERR(dir_block);
2400         BUFFER_TRACE(dir_block, "get_write_access");
2401         err = ext4_journal_get_write_access(handle, dir_block);
2402         if (err)
2403                 goto out;
2404         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2405         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2406         set_nlink(inode, 2);
2407         if (csum_size) {
2408                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2409                 initialize_dirent_tail(t, blocksize);
2410         }
2411
2412         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2413         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2414         if (err)
2415                 goto out;
2416         set_buffer_verified(dir_block);
2417 out:
2418         brelse(dir_block);
2419         return err;
2420 }
2421
2422 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2423 {
2424         handle_t *handle;
2425         struct inode *inode;
2426         int err, credits, retries = 0;
2427
2428         if (EXT4_DIR_LINK_MAX(dir))
2429                 return -EMLINK;
2430
2431         dquot_initialize(dir);
2432
2433         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2434                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2435 retry:
2436         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2437                                             &dentry->d_name,
2438                                             0, NULL, EXT4_HT_DIR, credits);
2439         handle = ext4_journal_current_handle();
2440         err = PTR_ERR(inode);
2441         if (IS_ERR(inode))
2442                 goto out_stop;
2443
2444         inode->i_op = &ext4_dir_inode_operations;
2445         inode->i_fop = &ext4_dir_operations;
2446         err = ext4_init_new_dir(handle, dir, inode);
2447         if (err)
2448                 goto out_clear_inode;
2449         err = ext4_mark_inode_dirty(handle, inode);
2450         if (!err)
2451                 err = ext4_add_entry(handle, dentry, inode);
2452         if (err) {
2453 out_clear_inode:
2454                 clear_nlink(inode);
2455                 unlock_new_inode(inode);
2456                 ext4_mark_inode_dirty(handle, inode);
2457                 iput(inode);
2458                 goto out_stop;
2459         }
2460         ext4_inc_count(handle, dir);
2461         ext4_update_dx_flag(dir);
2462         err = ext4_mark_inode_dirty(handle, dir);
2463         if (err)
2464                 goto out_clear_inode;
2465         unlock_new_inode(inode);
2466         d_instantiate(dentry, inode);
2467         if (IS_DIRSYNC(dir))
2468                 ext4_handle_sync(handle);
2469
2470 out_stop:
2471         if (handle)
2472                 ext4_journal_stop(handle);
2473         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2474                 goto retry;
2475         return err;
2476 }
2477
2478 /*
2479  * routine to check that the specified directory is empty (for rmdir)
2480  */
2481 static int empty_dir(struct inode *inode)
2482 {
2483         unsigned int offset;
2484         struct buffer_head *bh;
2485         struct ext4_dir_entry_2 *de, *de1;
2486         struct super_block *sb;
2487         int err = 0;
2488
2489         if (ext4_has_inline_data(inode)) {
2490                 int has_inline_data = 1;
2491
2492                 err = empty_inline_dir(inode, &has_inline_data);
2493                 if (has_inline_data)
2494                         return err;
2495         }
2496
2497         sb = inode->i_sb;
2498         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2499                 EXT4_ERROR_INODE(inode, "invalid size");
2500                 return 1;
2501         }
2502         bh = ext4_read_dirblock(inode, 0, EITHER);
2503         if (IS_ERR(bh))
2504                 return 1;
2505
2506         de = (struct ext4_dir_entry_2 *) bh->b_data;
2507         de1 = ext4_next_entry(de, sb->s_blocksize);
2508         if (le32_to_cpu(de->inode) != inode->i_ino ||
2509                         !le32_to_cpu(de1->inode) ||
2510                         strcmp(".", de->name) ||
2511                         strcmp("..", de1->name)) {
2512                 ext4_warning(inode->i_sb,
2513                              "bad directory (dir #%lu) - no `.' or `..'",
2514                              inode->i_ino);
2515                 brelse(bh);
2516                 return 1;
2517         }
2518         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2519                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2520         de = ext4_next_entry(de1, sb->s_blocksize);
2521         while (offset < inode->i_size) {
2522                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2523                         unsigned int lblock;
2524                         err = 0;
2525                         brelse(bh);
2526                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2527                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2528                         if (IS_ERR(bh))
2529                                 return 1;
2530                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2531                 }
2532                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2533                                          bh->b_data, bh->b_size, offset)) {
2534                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2535                                                          sb->s_blocksize);
2536                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2537                         continue;
2538                 }
2539                 if (le32_to_cpu(de->inode)) {
2540                         brelse(bh);
2541                         return 0;
2542                 }
2543                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2544                 de = ext4_next_entry(de, sb->s_blocksize);
2545         }
2546         brelse(bh);
2547         return 1;
2548 }
2549
2550 /*
2551  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2552  * such inodes, starting at the superblock, in case we crash before the
2553  * file is closed/deleted, or in case the inode truncate spans multiple
2554  * transactions and the last transaction is not recovered after a crash.
2555  *
2556  * At filesystem recovery time, we walk this list deleting unlinked
2557  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2558  *
2559  * Orphan list manipulation functions must be called under i_mutex unless
2560  * we are just creating the inode or deleting it.
2561  */
2562 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2563 {
2564         struct super_block *sb = inode->i_sb;
2565         struct ext4_sb_info *sbi = EXT4_SB(sb);
2566         struct ext4_iloc iloc;
2567         int err = 0, rc;
2568         bool dirty = false;
2569
2570         if (!sbi->s_journal)
2571                 return 0;
2572
2573         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2574                      !mutex_is_locked(&inode->i_mutex));
2575         /*
2576          * Exit early if inode already is on orphan list. This is a big speedup
2577          * since we don't have to contend on the global s_orphan_lock.
2578          */
2579         if (!list_empty(&EXT4_I(inode)->i_orphan))
2580                 return 0;
2581
2582         /*
2583          * Orphan handling is only valid for files with data blocks
2584          * being truncated, or files being unlinked. Note that we either
2585          * hold i_mutex, or the inode can not be referenced from outside,
2586          * so i_nlink should not be bumped due to race
2587          */
2588         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2589                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2590
2591         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2592         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2593         if (err)
2594                 goto out;
2595
2596         err = ext4_reserve_inode_write(handle, inode, &iloc);
2597         if (err)
2598                 goto out;
2599
2600         mutex_lock(&sbi->s_orphan_lock);
2601         /*
2602          * Due to previous errors inode may be already a part of on-disk
2603          * orphan list. If so skip on-disk list modification.
2604          */
2605         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2606             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2607                 /* Insert this inode at the head of the on-disk orphan list */
2608                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2609                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2610                 dirty = true;
2611         }
2612         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2613         mutex_unlock(&sbi->s_orphan_lock);
2614
2615         if (dirty) {
2616                 err = ext4_handle_dirty_super(handle, sb);
2617                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2618                 if (!err)
2619                         err = rc;
2620                 if (err) {
2621                         /*
2622                          * We have to remove inode from in-memory list if
2623                          * addition to on disk orphan list failed. Stray orphan
2624                          * list entries can cause panics at unmount time.
2625                          */
2626                         mutex_lock(&sbi->s_orphan_lock);
2627                         list_del(&EXT4_I(inode)->i_orphan);
2628                         mutex_unlock(&sbi->s_orphan_lock);
2629                 }
2630         }
2631         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2632         jbd_debug(4, "orphan inode %lu will point to %d\n",
2633                         inode->i_ino, NEXT_ORPHAN(inode));
2634 out:
2635         ext4_std_error(sb, err);
2636         return err;
2637 }
2638
2639 /*
2640  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2641  * of such inodes stored on disk, because it is finally being cleaned up.
2642  */
2643 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2644 {
2645         struct list_head *prev;
2646         struct ext4_inode_info *ei = EXT4_I(inode);
2647         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2648         __u32 ino_next;
2649         struct ext4_iloc iloc;
2650         int err = 0;
2651
2652         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2653                 return 0;
2654
2655         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2656                      !mutex_is_locked(&inode->i_mutex));
2657         /* Do this quick check before taking global s_orphan_lock. */
2658         if (list_empty(&ei->i_orphan))
2659                 return 0;
2660
2661         if (handle) {
2662                 /* Grab inode buffer early before taking global s_orphan_lock */
2663                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2664         }
2665
2666         mutex_lock(&sbi->s_orphan_lock);
2667         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2668
2669         prev = ei->i_orphan.prev;
2670         list_del_init(&ei->i_orphan);
2671
2672         /* If we're on an error path, we may not have a valid
2673          * transaction handle with which to update the orphan list on
2674          * disk, but we still need to remove the inode from the linked
2675          * list in memory. */
2676         if (!handle || err) {
2677                 mutex_unlock(&sbi->s_orphan_lock);
2678                 goto out_err;
2679         }
2680
2681         ino_next = NEXT_ORPHAN(inode);
2682         if (prev == &sbi->s_orphan) {
2683                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2684                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2685                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2686                 if (err) {
2687                         mutex_unlock(&sbi->s_orphan_lock);
2688                         goto out_brelse;
2689                 }
2690                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2691                 mutex_unlock(&sbi->s_orphan_lock);
2692                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2693         } else {
2694                 struct ext4_iloc iloc2;
2695                 struct inode *i_prev =
2696                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2697
2698                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2699                           i_prev->i_ino, ino_next);
2700                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2701                 if (err) {
2702                         mutex_unlock(&sbi->s_orphan_lock);
2703                         goto out_brelse;
2704                 }
2705                 NEXT_ORPHAN(i_prev) = ino_next;
2706                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2707                 mutex_unlock(&sbi->s_orphan_lock);
2708         }
2709         if (err)
2710                 goto out_brelse;
2711         NEXT_ORPHAN(inode) = 0;
2712         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2713 out_err:
2714         ext4_std_error(inode->i_sb, err);
2715         return err;
2716
2717 out_brelse:
2718         brelse(iloc.bh);
2719         goto out_err;
2720 }
2721
2722 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2723 {
2724         int retval;
2725         struct inode *inode;
2726         struct buffer_head *bh;
2727         struct ext4_dir_entry_2 *de;
2728         handle_t *handle = NULL;
2729
2730         /* Initialize quotas before so that eventual writes go in
2731          * separate transaction */
2732         dquot_initialize(dir);
2733         dquot_initialize(dentry->d_inode);
2734
2735         retval = -ENOENT;
2736         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2737         if (IS_ERR(bh))
2738                 return PTR_ERR(bh);
2739         if (!bh)
2740                 goto end_rmdir;
2741
2742         inode = dentry->d_inode;
2743
2744         retval = -EIO;
2745         if (le32_to_cpu(de->inode) != inode->i_ino)
2746                 goto end_rmdir;
2747
2748         retval = -ENOTEMPTY;
2749         if (!empty_dir(inode))
2750                 goto end_rmdir;
2751
2752         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2753                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2754         if (IS_ERR(handle)) {
2755                 retval = PTR_ERR(handle);
2756                 handle = NULL;
2757                 goto end_rmdir;
2758         }
2759
2760         if (IS_DIRSYNC(dir))
2761                 ext4_handle_sync(handle);
2762
2763         retval = ext4_delete_entry(handle, dir, de, bh);
2764         if (retval)
2765                 goto end_rmdir;
2766         if (!EXT4_DIR_LINK_EMPTY(inode))
2767                 ext4_warning(inode->i_sb,
2768                              "empty directory has too many links (%d)",
2769                              inode->i_nlink);
2770         inode->i_version++;
2771         clear_nlink(inode);
2772         /* There's no need to set i_disksize: the fact that i_nlink is
2773          * zero will ensure that the right thing happens during any
2774          * recovery. */
2775         inode->i_size = 0;
2776         ext4_orphan_add(handle, inode);
2777         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2778         ext4_mark_inode_dirty(handle, inode);
2779         ext4_dec_count(handle, dir);
2780         ext4_update_dx_flag(dir);
2781         ext4_mark_inode_dirty(handle, dir);
2782
2783 end_rmdir:
2784         brelse(bh);
2785         if (handle)
2786                 ext4_journal_stop(handle);
2787         return retval;
2788 }
2789
2790 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2791 {
2792         int retval;
2793         struct inode *inode;
2794         struct buffer_head *bh;
2795         struct ext4_dir_entry_2 *de;
2796         handle_t *handle = NULL;
2797
2798         trace_ext4_unlink_enter(dir, dentry);
2799         /* Initialize quotas before so that eventual writes go
2800          * in separate transaction */
2801         dquot_initialize(dir);
2802         dquot_initialize(dentry->d_inode);
2803
2804         retval = -ENOENT;
2805         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2806         if (IS_ERR(bh))
2807                 return PTR_ERR(bh);
2808         if (!bh)
2809                 goto end_unlink;
2810
2811         inode = dentry->d_inode;
2812
2813         retval = -EIO;
2814         if (le32_to_cpu(de->inode) != inode->i_ino)
2815                 goto end_unlink;
2816
2817         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2818                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2819         if (IS_ERR(handle)) {
2820                 retval = PTR_ERR(handle);
2821                 handle = NULL;
2822                 goto end_unlink;
2823         }
2824
2825         if (IS_DIRSYNC(dir))
2826                 ext4_handle_sync(handle);
2827
2828         if (!inode->i_nlink) {
2829                 ext4_warning(inode->i_sb,
2830                              "Deleting nonexistent file (%lu), %d",
2831                              inode->i_ino, inode->i_nlink);
2832                 set_nlink(inode, 1);
2833         }
2834         retval = ext4_delete_entry(handle, dir, de, bh);
2835         if (retval)
2836                 goto end_unlink;
2837         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2838         ext4_update_dx_flag(dir);
2839         ext4_mark_inode_dirty(handle, dir);
2840         drop_nlink(inode);
2841         if (!inode->i_nlink)
2842                 ext4_orphan_add(handle, inode);
2843         inode->i_ctime = ext4_current_time(inode);
2844         ext4_mark_inode_dirty(handle, inode);
2845         retval = 0;
2846
2847 end_unlink:
2848         brelse(bh);
2849         if (handle)
2850                 ext4_journal_stop(handle);
2851         trace_ext4_unlink_exit(dentry, retval);
2852         return retval;
2853 }
2854
2855 static int ext4_symlink(struct inode *dir,
2856                         struct dentry *dentry, const char *symname)
2857 {
2858         handle_t *handle;
2859         struct inode *inode;
2860         int l, err, retries = 0;
2861         int credits;
2862
2863         l = strlen(symname)+1;
2864         if (l > dir->i_sb->s_blocksize)
2865                 return -ENAMETOOLONG;
2866
2867         dquot_initialize(dir);
2868
2869         if (l > EXT4_N_BLOCKS * 4) {
2870                 /*
2871                  * For non-fast symlinks, we just allocate inode and put it on
2872                  * orphan list in the first transaction => we need bitmap,
2873                  * group descriptor, sb, inode block, quota blocks, and
2874                  * possibly selinux xattr blocks.
2875                  */
2876                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2877                           EXT4_XATTR_TRANS_BLOCKS;
2878         } else {
2879                 /*
2880                  * Fast symlink. We have to add entry to directory
2881                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2882                  * allocate new inode (bitmap, group descriptor, inode block,
2883                  * quota blocks, sb is already counted in previous macros).
2884                  */
2885                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2886                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2887         }
2888 retry:
2889         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2890                                             &dentry->d_name, 0, NULL,
2891                                             EXT4_HT_DIR, credits);
2892         handle = ext4_journal_current_handle();
2893         err = PTR_ERR(inode);
2894         if (IS_ERR(inode))
2895                 goto out_stop;
2896
2897         if (l > EXT4_N_BLOCKS * 4) {
2898                 inode->i_op = &ext4_symlink_inode_operations;
2899                 ext4_set_aops(inode);
2900                 /*
2901                  * We cannot call page_symlink() with transaction started
2902                  * because it calls into ext4_write_begin() which can wait
2903                  * for transaction commit if we are running out of space
2904                  * and thus we deadlock. So we have to stop transaction now
2905                  * and restart it when symlink contents is written.
2906                  *
2907                  * To keep fs consistent in case of crash, we have to put inode
2908                  * to orphan list in the mean time.
2909                  */
2910                 drop_nlink(inode);
2911                 err = ext4_orphan_add(handle, inode);
2912                 ext4_journal_stop(handle);
2913                 if (err)
2914                         goto err_drop_inode;
2915                 err = __page_symlink(inode, symname, l, 1);
2916                 if (err)
2917                         goto err_drop_inode;
2918                 /*
2919                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2920                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2921                  */
2922                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2923                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2924                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2925                 if (IS_ERR(handle)) {
2926                         err = PTR_ERR(handle);
2927                         goto err_drop_inode;
2928                 }
2929                 set_nlink(inode, 1);
2930                 err = ext4_orphan_del(handle, inode);
2931                 if (err) {
2932                         ext4_journal_stop(handle);
2933                         clear_nlink(inode);
2934                         goto err_drop_inode;
2935                 }
2936         } else {
2937                 /* clear the extent format for fast symlink */
2938                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2939                 inode->i_op = &ext4_fast_symlink_inode_operations;
2940                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2941                 inode->i_size = l-1;
2942         }
2943         EXT4_I(inode)->i_disksize = inode->i_size;
2944         err = ext4_add_nondir(handle, dentry, inode);
2945         if (!err && IS_DIRSYNC(dir))
2946                 ext4_handle_sync(handle);
2947
2948 out_stop:
2949         if (handle)
2950                 ext4_journal_stop(handle);
2951         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2952                 goto retry;
2953         return err;
2954 err_drop_inode:
2955         unlock_new_inode(inode);
2956         iput(inode);
2957         return err;
2958 }
2959
2960 static int ext4_link(struct dentry *old_dentry,
2961                      struct inode *dir, struct dentry *dentry)
2962 {
2963         handle_t *handle;
2964         struct inode *inode = old_dentry->d_inode;
2965         int err, retries = 0;
2966
2967         if (inode->i_nlink >= EXT4_LINK_MAX)
2968                 return -EMLINK;
2969
2970         dquot_initialize(dir);
2971
2972 retry:
2973         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2974                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2975                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2976         if (IS_ERR(handle))
2977                 return PTR_ERR(handle);
2978
2979         if (IS_DIRSYNC(dir))
2980                 ext4_handle_sync(handle);
2981
2982         inode->i_ctime = ext4_current_time(inode);
2983         ext4_inc_count(handle, inode);
2984         ihold(inode);
2985
2986         err = ext4_add_entry(handle, dentry, inode);
2987         if (!err) {
2988                 ext4_mark_inode_dirty(handle, inode);
2989                 /* this can happen only for tmpfile being
2990                  * linked the first time
2991                  */
2992                 if (inode->i_nlink == 1)
2993                         ext4_orphan_del(handle, inode);
2994                 d_instantiate(dentry, inode);
2995         } else {
2996                 drop_nlink(inode);
2997                 iput(inode);
2998         }
2999         ext4_journal_stop(handle);
3000         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3001                 goto retry;
3002         return err;
3003 }
3004
3005
3006 /*
3007  * Try to find buffer head where contains the parent block.
3008  * It should be the inode block if it is inlined or the 1st block
3009  * if it is a normal dir.
3010  */
3011 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3012                                         struct inode *inode,
3013                                         int *retval,
3014                                         struct ext4_dir_entry_2 **parent_de,
3015                                         int *inlined)
3016 {
3017         struct buffer_head *bh;
3018
3019         if (!ext4_has_inline_data(inode)) {
3020                 bh = ext4_read_dirblock(inode, 0, EITHER);
3021                 if (IS_ERR(bh)) {
3022                         *retval = PTR_ERR(bh);
3023                         return NULL;
3024                 }
3025                 *parent_de = ext4_next_entry(
3026                                         (struct ext4_dir_entry_2 *)bh->b_data,
3027                                         inode->i_sb->s_blocksize);
3028                 return bh;
3029         }
3030
3031         *inlined = 1;
3032         return ext4_get_first_inline_block(inode, parent_de, retval);
3033 }
3034
3035 struct ext4_renament {
3036         struct inode *dir;
3037         struct dentry *dentry;
3038         struct inode *inode;
3039         bool is_dir;
3040         int dir_nlink_delta;
3041
3042         /* entry for "dentry" */
3043         struct buffer_head *bh;
3044         struct ext4_dir_entry_2 *de;
3045         int inlined;
3046
3047         /* entry for ".." in inode if it's a directory */
3048         struct buffer_head *dir_bh;
3049         struct ext4_dir_entry_2 *parent_de;
3050         int dir_inlined;
3051 };
3052
3053 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3054 {
3055         int retval;
3056
3057         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3058                                               &retval, &ent->parent_de,
3059                                               &ent->dir_inlined);
3060         if (!ent->dir_bh)
3061                 return retval;
3062         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3063                 return -EIO;
3064         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3065         return ext4_journal_get_write_access(handle, ent->dir_bh);
3066 }
3067
3068 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3069                                   unsigned dir_ino)
3070 {
3071         int retval;
3072
3073         ent->parent_de->inode = cpu_to_le32(dir_ino);
3074         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3075         if (!ent->dir_inlined) {
3076                 if (is_dx(ent->inode)) {
3077                         retval = ext4_handle_dirty_dx_node(handle,
3078                                                            ent->inode,
3079                                                            ent->dir_bh);
3080                 } else {
3081                         retval = ext4_handle_dirty_dirent_node(handle,
3082                                                                ent->inode,
3083                                                                ent->dir_bh);
3084                 }
3085         } else {
3086                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3087         }
3088         if (retval) {
3089                 ext4_std_error(ent->dir->i_sb, retval);
3090                 return retval;
3091         }
3092         return 0;
3093 }
3094
3095 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3096                        unsigned ino, unsigned file_type)
3097 {
3098         int retval;
3099
3100         BUFFER_TRACE(ent->bh, "get write access");
3101         retval = ext4_journal_get_write_access(handle, ent->bh);
3102         if (retval)
3103                 return retval;
3104         ent->de->inode = cpu_to_le32(ino);
3105         if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3106                                       EXT4_FEATURE_INCOMPAT_FILETYPE))
3107                 ent->de->file_type = file_type;
3108         ent->dir->i_version++;
3109         ent->dir->i_ctime = ent->dir->i_mtime =
3110                 ext4_current_time(ent->dir);
3111         ext4_mark_inode_dirty(handle, ent->dir);
3112         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3113         if (!ent->inlined) {
3114                 retval = ext4_handle_dirty_dirent_node(handle,
3115                                                        ent->dir, ent->bh);
3116                 if (unlikely(retval)) {
3117                         ext4_std_error(ent->dir->i_sb, retval);
3118                         return retval;
3119                 }
3120         }
3121         brelse(ent->bh);
3122         ent->bh = NULL;
3123
3124         return 0;
3125 }
3126
3127 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3128                                   const struct qstr *d_name)
3129 {
3130         int retval = -ENOENT;
3131         struct buffer_head *bh;
3132         struct ext4_dir_entry_2 *de;
3133
3134         bh = ext4_find_entry(dir, d_name, &de, NULL);
3135         if (IS_ERR(bh))
3136                 return PTR_ERR(bh);
3137         if (bh) {
3138                 retval = ext4_delete_entry(handle, dir, de, bh);
3139                 brelse(bh);
3140         }
3141         return retval;
3142 }
3143
3144 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3145                                int force_reread)
3146 {
3147         int retval;
3148         /*
3149          * ent->de could have moved from under us during htree split, so make
3150          * sure that we are deleting the right entry.  We might also be pointing
3151          * to a stale entry in the unused part of ent->bh so just checking inum
3152          * and the name isn't enough.
3153          */
3154         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3155             ent->de->name_len != ent->dentry->d_name.len ||
3156             strncmp(ent->de->name, ent->dentry->d_name.name,
3157                     ent->de->name_len) ||
3158             force_reread) {
3159                 retval = ext4_find_delete_entry(handle, ent->dir,
3160                                                 &ent->dentry->d_name);
3161         } else {
3162                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3163                 if (retval == -ENOENT) {
3164                         retval = ext4_find_delete_entry(handle, ent->dir,
3165                                                         &ent->dentry->d_name);
3166                 }
3167         }
3168
3169         if (retval) {
3170                 ext4_warning(ent->dir->i_sb,
3171                                 "Deleting old file (%lu), %d, error=%d",
3172                                 ent->dir->i_ino, ent->dir->i_nlink, retval);
3173         }
3174 }
3175
3176 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3177 {
3178         if (ent->dir_nlink_delta) {
3179                 if (ent->dir_nlink_delta == -1)
3180                         ext4_dec_count(handle, ent->dir);
3181                 else
3182                         ext4_inc_count(handle, ent->dir);
3183                 ext4_mark_inode_dirty(handle, ent->dir);
3184         }
3185 }
3186
3187 /*
3188  * Anybody can rename anything with this: the permission checks are left to the
3189  * higher-level routines.
3190  *
3191  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3192  * while new_{dentry,inode) refers to the destination dentry/inode
3193  * This comes from rename(const char *oldpath, const char *newpath)
3194  */
3195 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3196                        struct inode *new_dir, struct dentry *new_dentry)
3197 {
3198         handle_t *handle = NULL;
3199         struct ext4_renament old = {
3200                 .dir = old_dir,
3201                 .dentry = old_dentry,
3202                 .inode = old_dentry->d_inode,
3203         };
3204         struct ext4_renament new = {
3205                 .dir = new_dir,
3206                 .dentry = new_dentry,
3207                 .inode = new_dentry->d_inode,
3208         };
3209         int force_reread;
3210         int retval;
3211
3212         dquot_initialize(old.dir);
3213         dquot_initialize(new.dir);
3214
3215         /* Initialize quotas before so that eventual writes go
3216          * in separate transaction */
3217         if (new.inode)
3218                 dquot_initialize(new.inode);
3219
3220         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3221         if (IS_ERR(old.bh))
3222                 return PTR_ERR(old.bh);
3223         /*
3224          *  Check for inode number is _not_ due to possible IO errors.
3225          *  We might rmdir the source, keep it as pwd of some process
3226          *  and merrily kill the link to whatever was created under the
3227          *  same name. Goodbye sticky bit ;-<
3228          */
3229         retval = -ENOENT;
3230         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3231                 goto end_rename;
3232
3233         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3234                                  &new.de, &new.inlined);
3235         if (IS_ERR(new.bh)) {
3236                 retval = PTR_ERR(new.bh);
3237                 goto end_rename;
3238         }
3239         if (new.bh) {
3240                 if (!new.inode) {
3241                         brelse(new.bh);
3242                         new.bh = NULL;
3243                 }
3244         }
3245         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3246                 ext4_alloc_da_blocks(old.inode);
3247
3248         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3249                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3250                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3251         if (IS_ERR(handle))
3252                 return PTR_ERR(handle);
3253
3254         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3255                 ext4_handle_sync(handle);
3256
3257         if (S_ISDIR(old.inode->i_mode)) {
3258                 if (new.inode) {
3259                         retval = -ENOTEMPTY;
3260                         if (!empty_dir(new.inode))
3261                                 goto end_rename;
3262                 } else {
3263                         retval = -EMLINK;
3264                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3265                                 goto end_rename;
3266                 }
3267                 retval = ext4_rename_dir_prepare(handle, &old);
3268                 if (retval)
3269                         goto end_rename;
3270         }
3271         /*
3272          * If we're renaming a file within an inline_data dir and adding or
3273          * setting the new dirent causes a conversion from inline_data to
3274          * extents/blockmap, we need to force the dirent delete code to
3275          * re-read the directory, or else we end up trying to delete a dirent
3276          * from what is now the extent tree root (or a block map).
3277          */
3278         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3279                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3280         if (!new.bh) {
3281                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3282                 if (retval)
3283                         goto end_rename;
3284         } else {
3285                 retval = ext4_setent(handle, &new,
3286                                      old.inode->i_ino, old.de->file_type);
3287                 if (retval)
3288                         goto end_rename;
3289         }
3290         if (force_reread)
3291                 force_reread = !ext4_test_inode_flag(new.dir,
3292                                                      EXT4_INODE_INLINE_DATA);
3293
3294         /*
3295          * Like most other Unix systems, set the ctime for inodes on a
3296          * rename.
3297          */
3298         old.inode->i_ctime = ext4_current_time(old.inode);
3299         ext4_mark_inode_dirty(handle, old.inode);
3300
3301         /*
3302          * ok, that's it
3303          */
3304         ext4_rename_delete(handle, &old, force_reread);
3305
3306         if (new.inode) {
3307                 ext4_dec_count(handle, new.inode);
3308                 new.inode->i_ctime = ext4_current_time(new.inode);
3309         }
3310         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3311         ext4_update_dx_flag(old.dir);
3312         if (old.dir_bh) {
3313                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3314                 if (retval)
3315                         goto end_rename;
3316
3317                 ext4_dec_count(handle, old.dir);
3318                 if (new.inode) {
3319                         /* checked empty_dir above, can't have another parent,
3320                          * ext4_dec_count() won't work for many-linked dirs */
3321                         clear_nlink(new.inode);
3322                 } else {
3323                         ext4_inc_count(handle, new.dir);
3324                         ext4_update_dx_flag(new.dir);
3325                         ext4_mark_inode_dirty(handle, new.dir);
3326                 }
3327         }
3328         ext4_mark_inode_dirty(handle, old.dir);
3329         if (new.inode) {
3330                 ext4_mark_inode_dirty(handle, new.inode);
3331                 if (!new.inode->i_nlink)
3332                         ext4_orphan_add(handle, new.inode);
3333         }
3334         retval = 0;
3335
3336 end_rename:
3337         brelse(old.dir_bh);
3338         brelse(old.bh);
3339         brelse(new.bh);
3340         if (handle)
3341                 ext4_journal_stop(handle);
3342         return retval;
3343 }
3344
3345 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3346                              struct inode *new_dir, struct dentry *new_dentry)
3347 {
3348         handle_t *handle = NULL;
3349         struct ext4_renament old = {
3350                 .dir = old_dir,
3351                 .dentry = old_dentry,
3352                 .inode = old_dentry->d_inode,
3353         };
3354         struct ext4_renament new = {
3355                 .dir = new_dir,
3356                 .dentry = new_dentry,
3357                 .inode = new_dentry->d_inode,
3358         };
3359         u8 new_file_type;
3360         int retval;
3361
3362         dquot_initialize(old.dir);
3363         dquot_initialize(new.dir);
3364
3365         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3366                                  &old.de, &old.inlined);
3367         if (IS_ERR(old.bh))
3368                 return PTR_ERR(old.bh);
3369         /*
3370          *  Check for inode number is _not_ due to possible IO errors.
3371          *  We might rmdir the source, keep it as pwd of some process
3372          *  and merrily kill the link to whatever was created under the
3373          *  same name. Goodbye sticky bit ;-<
3374          */
3375         retval = -ENOENT;
3376         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3377                 goto end_rename;
3378
3379         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3380                                  &new.de, &new.inlined);
3381         if (IS_ERR(new.bh)) {
3382                 retval = PTR_ERR(new.bh);
3383                 goto end_rename;
3384         }
3385
3386         /* RENAME_EXCHANGE case: old *and* new must both exist */
3387         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3388                 goto end_rename;
3389
3390         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3391                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3392                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3393         if (IS_ERR(handle))
3394                 return PTR_ERR(handle);
3395
3396         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3397                 ext4_handle_sync(handle);
3398
3399         if (S_ISDIR(old.inode->i_mode)) {
3400                 old.is_dir = true;
3401                 retval = ext4_rename_dir_prepare(handle, &old);
3402                 if (retval)
3403                         goto end_rename;
3404         }
3405         if (S_ISDIR(new.inode->i_mode)) {
3406                 new.is_dir = true;
3407                 retval = ext4_rename_dir_prepare(handle, &new);
3408                 if (retval)
3409                         goto end_rename;
3410         }
3411
3412         /*
3413          * Other than the special case of overwriting a directory, parents'
3414          * nlink only needs to be modified if this is a cross directory rename.
3415          */
3416         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3417                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3418                 new.dir_nlink_delta = -old.dir_nlink_delta;
3419                 retval = -EMLINK;
3420                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3421                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3422                         goto end_rename;
3423         }
3424
3425         new_file_type = new.de->file_type;
3426         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3427         if (retval)
3428                 goto end_rename;
3429
3430         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3431         if (retval)
3432                 goto end_rename;
3433
3434         /*
3435          * Like most other Unix systems, set the ctime for inodes on a
3436          * rename.
3437          */
3438         old.inode->i_ctime = ext4_current_time(old.inode);
3439         new.inode->i_ctime = ext4_current_time(new.inode);
3440         ext4_mark_inode_dirty(handle, old.inode);
3441         ext4_mark_inode_dirty(handle, new.inode);
3442
3443         if (old.dir_bh) {
3444                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3445                 if (retval)
3446                         goto end_rename;
3447         }
3448         if (new.dir_bh) {
3449                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3450                 if (retval)
3451                         goto end_rename;
3452         }
3453         ext4_update_dir_count(handle, &old);
3454         ext4_update_dir_count(handle, &new);
3455         retval = 0;
3456
3457 end_rename:
3458         brelse(old.dir_bh);
3459         brelse(new.dir_bh);
3460         brelse(old.bh);
3461         brelse(new.bh);
3462         if (handle)
3463                 ext4_journal_stop(handle);
3464         return retval;
3465 }
3466
3467 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3468                         struct inode *new_dir, struct dentry *new_dentry,
3469                         unsigned int flags)
3470 {
3471         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
3472                 return -EINVAL;
3473
3474         if (flags & RENAME_EXCHANGE) {
3475                 return ext4_cross_rename(old_dir, old_dentry,
3476                                          new_dir, new_dentry);
3477         }
3478         /*
3479          * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3480          * is equivalent to regular rename.
3481          */
3482         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry);
3483 }
3484
3485 /*
3486  * directories can handle most operations...
3487  */
3488 const struct inode_operations ext4_dir_inode_operations = {
3489         .create         = ext4_create,
3490         .lookup         = ext4_lookup,
3491         .link           = ext4_link,
3492         .unlink         = ext4_unlink,
3493         .symlink        = ext4_symlink,
3494         .mkdir          = ext4_mkdir,
3495         .rmdir          = ext4_rmdir,
3496         .mknod          = ext4_mknod,
3497         .tmpfile        = ext4_tmpfile,
3498         .rename2        = ext4_rename2,
3499         .setattr        = ext4_setattr,
3500         .setxattr       = generic_setxattr,
3501         .getxattr       = generic_getxattr,
3502         .listxattr      = ext4_listxattr,
3503         .removexattr    = generic_removexattr,
3504         .get_acl        = ext4_get_acl,
3505         .set_acl        = ext4_set_acl,
3506         .fiemap         = ext4_fiemap,
3507 };
3508
3509 const struct inode_operations ext4_special_inode_operations = {
3510         .setattr        = ext4_setattr,
3511         .setxattr       = generic_setxattr,
3512         .getxattr       = generic_getxattr,
3513         .listxattr      = ext4_listxattr,
3514         .removexattr    = generic_removexattr,
3515         .get_acl        = ext4_get_acl,
3516         .set_acl        = ext4_set_acl,
3517 };