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/time.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/stat.h>
  32 #include <linux/string.h>
  33 #include <linux/quotaops.h>
  34 #include <linux/buffer_head.h>
  35 #include <linux/bio.h>
  36 #include "ext4.h"
  37 #include "ext4_jbd2.h"
  38
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 #include <trace/events/ext4.h>
  43 /*
  44  * define how far ahead to read directories while searching them.
  45  */
  46 #define NAMEI_RA_CHUNKS  2
  47 #define NAMEI_RA_BLOCKS  4
  48 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  49
  50 static struct buffer_head *ext4_append(handle_t *handle,
  51                                         struct inode *inode,
  52                                         ext4_lblk_t *block)
  53 {
  54         struct buffer_head *bh;
  55         int err;
  56
  57         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  58                      ((inode->i_size >> 10) >=
  59                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  60                 return ERR_PTR(-ENOSPC);
  61
  62         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  63
  64         bh = ext4_bread(handle, inode, *block, 1);
  65         if (IS_ERR(bh))
  66                 return bh;
  67         inode->i_size += inode->i_sb->s_blocksize;
  68         EXT4_I(inode)->i_disksize = inode->i_size;
  69         BUFFER_TRACE(bh, "get_write_access");
  70         err = ext4_journal_get_write_access(handle, bh);
  71         if (err) {
  72                 brelse(bh);
  73                 ext4_std_error(inode->i_sb, err);
  74                 return ERR_PTR(err);
  75         }
  76         return bh;
  77 }
  78
  79 static int ext4_dx_csum_verify(struct inode *inode,
  80                                struct ext4_dir_entry *dirent);
  81
  82 typedef enum {
  83         EITHER, INDEX, DIRENT
  84 } dirblock_type_t;
  85
  86 #define ext4_read_dirblock(inode, block, type) \
  87         __ext4_read_dirblock((inode), (block), (type), __LINE__)
  88
  89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  90                                               ext4_lblk_t block,
  91                                               dirblock_type_t type,
  92                                               unsigned int line)
  93 {
  94         struct buffer_head *bh;
  95         struct ext4_dir_entry *dirent;
  96         int is_dx_block = 0;
  97
  98         bh = ext4_bread(NULL, inode, block, 0);
  99         if (IS_ERR(bh)) {
 100                 __ext4_warning(inode->i_sb, __func__, line,
 101                                "error %ld reading directory block "
 102                                "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
 103                                (unsigned long) block);
 104
 105                 return bh;
 106         }
 107         if (!bh) {
 108                 ext4_error_inode(inode, __func__, line, block, "Directory hole found");
 109                 return ERR_PTR(-EIO);
 110         }
 111         dirent = (struct ext4_dir_entry *) bh->b_data;
 112         /* Determine whether or not we have an index block */
 113         if (is_dx(inode)) {
 114                 if (block == 0)
 115                         is_dx_block = 1;
 116                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 117                                                 inode->i_sb->s_blocksize) ==
 118                          inode->i_sb->s_blocksize)
 119                         is_dx_block = 1;
 120         }
 121         if (!is_dx_block && type == INDEX) {
 122                 ext4_error_inode(inode, __func__, line, block,
 123                        "directory leaf block found instead of index block");
 124                 return ERR_PTR(-EIO);
 125         }
 126         if (!ext4_has_metadata_csum(inode->i_sb) ||
 127             buffer_verified(bh))
 128                 return bh;
 129
 130         /*
 131          * An empty leaf block can get mistaken for a index block; for
 132          * this reason, we can only check the index checksum when the
 133          * caller is sure it should be an index block.
 134          */
 135         if (is_dx_block && type == INDEX) {
 136                 if (ext4_dx_csum_verify(inode, dirent))
 137                         set_buffer_verified(bh);
 138                 else {
 139                         ext4_error_inode(inode, __func__, line, block,
 140                                 "Directory index failed checksum");
 141                         brelse(bh);
 142                         return ERR_PTR(-EIO);
 143                 }
 144         }
 145         if (!is_dx_block) {
 146                 if (ext4_dirent_csum_verify(inode, dirent))
 147                         set_buffer_verified(bh);
 148                 else {
 149                         ext4_error_inode(inode, __func__, line, block,
 150                                 "Directory block failed checksum");
 151                         brelse(bh);
 152                         return ERR_PTR(-EIO);
 153                 }
 154         }
 155         return bh;
 156 }
 157
 158 #ifndef assert
 159 #define assert(test) J_ASSERT(test)
 160 #endif
 161
 162 #ifdef DX_DEBUG
 163 #define dxtrace(command) command
 164 #else
 165 #define dxtrace(command)
 166 #endif
 167
 168 struct fake_dirent
 169 {
 170         __le32 inode;
 171         __le16 rec_len;
 172         u8 name_len;
 173         u8 file_type;
 174 };
 175
 176 struct dx_countlimit
 177 {
 178         __le16 limit;
 179         __le16 count;
 180 };
 181
 182 struct dx_entry
 183 {
 184         __le32 hash;
 185         __le32 block;
 186 };
 187
 188 /*
 189  * dx_root_info is laid out so that if it should somehow get overlaid by a
 190  * dirent the two low bits of the hash version will be zero.  Therefore, the
 191  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 192  */
 193
 194 struct dx_root
 195 {
 196         struct fake_dirent dot;
 197         char dot_name[4];
 198         struct fake_dirent dotdot;
 199         char dotdot_name[4];
 200         struct dx_root_info
 201         {
 202                 __le32 reserved_zero;
 203                 u8 hash_version;
 204                 u8 info_length; /* 8 */
 205                 u8 indirect_levels;
 206                 u8 unused_flags;
 207         }
 208         info;
 209         struct dx_entry entries[0];
 210 };
 211
 212 struct dx_node
 213 {
 214         struct fake_dirent fake;
 215         struct dx_entry entries[0];
 216 };
 217
 218
 219 struct dx_frame
 220 {
 221         struct buffer_head *bh;
 222         struct dx_entry *entries;
 223         struct dx_entry *at;
 224 };
 225
 226 struct dx_map_entry
 227 {
 228         u32 hash;
 229         u16 offs;
 230         u16 size;
 231 };
 232
 233 /*
 234  * This goes at the end of each htree block.
 235  */
 236 struct dx_tail {
 237         u32 dt_reserved;
 238         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 239 };
 240
 241 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 242 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 243 static inline unsigned dx_get_hash(struct dx_entry *entry);
 244 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 245 static unsigned dx_get_count(struct dx_entry *entries);
 246 static unsigned dx_get_limit(struct dx_entry *entries);
 247 static void dx_set_count(struct dx_entry *entries, unsigned value);
 248 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 249 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 250 static unsigned dx_node_limit(struct inode *dir);
 251 static struct dx_frame *dx_probe(const struct qstr *d_name,
 252                                  struct inode *dir,
 253                                  struct dx_hash_info *hinfo,
 254                                  struct dx_frame *frame);
 255 static void dx_release(struct dx_frame *frames);
 256 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 257                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 258 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 259 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 260                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 261 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 262 static void dx_insert_block(struct dx_frame *frame,
 263                                         u32 hash, ext4_lblk_t block);
 264 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 265                                  struct dx_frame *frame,
 266                                  struct dx_frame *frames,
 267                                  __u32 *start_hash);
 268 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 269                 const struct qstr *d_name,
 270                 struct ext4_dir_entry_2 **res_dir);
 271 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 272                              struct inode *inode);
 273
 274 /* checksumming functions */
 275 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 276                             unsigned int blocksize)
 277 {
 278         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 279         t->det_rec_len = ext4_rec_len_to_disk(
 280                         sizeof(struct ext4_dir_entry_tail), blocksize);
 281         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 282 }
 283
 284 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 285 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 286                                                    struct ext4_dir_entry *de)
 287 {
 288         struct ext4_dir_entry_tail *t;
 289
 290 #ifdef PARANOID
 291         struct ext4_dir_entry *d, *top;
 292
 293         d = de;
 294         top = (struct ext4_dir_entry *)(((void *)de) +
 295                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 296                 sizeof(struct ext4_dir_entry_tail)));
 297         while (d < top && d->rec_len)
 298                 d = (struct ext4_dir_entry *)(((void *)d) +
 299                     le16_to_cpu(d->rec_len));
 300
 301         if (d != top)
 302                 return NULL;
 303
 304         t = (struct ext4_dir_entry_tail *)d;
 305 #else
 306         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 307 #endif
 308
 309         if (t->det_reserved_zero1 ||
 310             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 311             t->det_reserved_zero2 ||
 312             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 313                 return NULL;
 314
 315         return t;
 316 }
 317
 318 static __le32 ext4_dirent_csum(struct inode *inode,
 319                                struct ext4_dir_entry *dirent, int size)
 320 {
 321         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 322         struct ext4_inode_info *ei = EXT4_I(inode);
 323         __u32 csum;
 324
 325         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 326         return cpu_to_le32(csum);
 327 }
 328
 329 static void warn_no_space_for_csum(struct inode *inode)
 330 {
 331         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
 332                      "checksum.  Please run e2fsck -D.", inode->i_ino);
 333 }
 334
 335 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 336 {
 337         struct ext4_dir_entry_tail *t;
 338
 339         if (!ext4_has_metadata_csum(inode->i_sb))
 340                 return 1;
 341
 342         t = get_dirent_tail(inode, dirent);
 343         if (!t) {
 344                 warn_no_space_for_csum(inode);
 345                 return 0;
 346         }
 347
 348         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 349                                                 (void *)t - (void *)dirent))
 350                 return 0;
 351
 352         return 1;
 353 }
 354
 355 static void ext4_dirent_csum_set(struct inode *inode,
 356                                  struct ext4_dir_entry *dirent)
 357 {
 358         struct ext4_dir_entry_tail *t;
 359
 360         if (!ext4_has_metadata_csum(inode->i_sb))
 361                 return;
 362
 363         t = get_dirent_tail(inode, dirent);
 364         if (!t) {
 365                 warn_no_space_for_csum(inode);
 366                 return;
 367         }
 368
 369         t->det_checksum = ext4_dirent_csum(inode, dirent,
 370                                            (void *)t - (void *)dirent);
 371 }
 372
 373 int ext4_handle_dirty_dirent_node(handle_t *handle,
 374                                   struct inode *inode,
 375                                   struct buffer_head *bh)
 376 {
 377         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 378         return ext4_handle_dirty_metadata(handle, inode, bh);
 379 }
 380
 381 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 382                                                struct ext4_dir_entry *dirent,
 383                                                int *offset)
 384 {
 385         struct ext4_dir_entry *dp;
 386         struct dx_root_info *root;
 387         int count_offset;
 388
 389         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 390                 count_offset = 8;
 391         else if (le16_to_cpu(dirent->rec_len) == 12) {
 392                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 393                 if (le16_to_cpu(dp->rec_len) !=
 394                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 395                         return NULL;
 396                 root = (struct dx_root_info *)(((void *)dp + 12));
 397                 if (root->reserved_zero ||
 398                     root->info_length != sizeof(struct dx_root_info))
 399                         return NULL;
 400                 count_offset = 32;
 401         } else
 402                 return NULL;
 403
 404         if (offset)
 405                 *offset = count_offset;
 406         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 407 }
 408
 409 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 410                            int count_offset, int count, struct dx_tail *t)
 411 {
 412         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 413         struct ext4_inode_info *ei = EXT4_I(inode);
 414         __u32 csum;
 415         __le32 save_csum;
 416         int size;
 417
 418         size = count_offset + (count * sizeof(struct dx_entry));
 419         save_csum = t->dt_checksum;
 420         t->dt_checksum = 0;
 421         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 422         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
 423         t->dt_checksum = save_csum;
 424
 425         return cpu_to_le32(csum);
 426 }
 427
 428 static int ext4_dx_csum_verify(struct inode *inode,
 429                                struct ext4_dir_entry *dirent)
 430 {
 431         struct dx_countlimit *c;
 432         struct dx_tail *t;
 433         int count_offset, limit, count;
 434
 435         if (!ext4_has_metadata_csum(inode->i_sb))
 436                 return 1;
 437
 438         c = get_dx_countlimit(inode, dirent, &count_offset);
 439         if (!c) {
 440                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 441                 return 1;
 442         }
 443         limit = le16_to_cpu(c->limit);
 444         count = le16_to_cpu(c->count);
 445         if (count_offset + (limit * sizeof(struct dx_entry)) >
 446             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 447                 warn_no_space_for_csum(inode);
 448                 return 1;
 449         }
 450         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 451
 452         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 453                                             count, t))
 454                 return 0;
 455         return 1;
 456 }
 457
 458 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 459 {
 460         struct dx_countlimit *c;
 461         struct dx_tail *t;
 462         int count_offset, limit, count;
 463
 464         if (!ext4_has_metadata_csum(inode->i_sb))
 465                 return;
 466
 467         c = get_dx_countlimit(inode, dirent, &count_offset);
 468         if (!c) {
 469                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 470                 return;
 471         }
 472         limit = le16_to_cpu(c->limit);
 473         count = le16_to_cpu(c->count);
 474         if (count_offset + (limit * sizeof(struct dx_entry)) >
 475             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 476                 warn_no_space_for_csum(inode);
 477                 return;
 478         }
 479         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 480
 481         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 482 }
 483
 484 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 485                                             struct inode *inode,
 486                                             struct buffer_head *bh)
 487 {
 488         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 489         return ext4_handle_dirty_metadata(handle, inode, bh);
 490 }
 491
 492 /*
 493  * p is at least 6 bytes before the end of page
 494  */
 495 static inline struct ext4_dir_entry_2 *
 496 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 497 {
 498         return (struct ext4_dir_entry_2 *)((char *)p +
 499                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 500 }
 501
 502 /*
 503  * Future: use high four bits of block for coalesce-on-delete flags
 504  * Mask them off for now.
 505  */
 506
 507 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 508 {
 509         return le32_to_cpu(entry->block) & 0x00ffffff;
 510 }
 511
 512 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 513 {
 514         entry->block = cpu_to_le32(value);
 515 }
 516
 517 static inline unsigned dx_get_hash(struct dx_entry *entry)
 518 {
 519         return le32_to_cpu(entry->hash);
 520 }
 521
 522 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 523 {
 524         entry->hash = cpu_to_le32(value);
 525 }
 526
 527 static inline unsigned dx_get_count(struct dx_entry *entries)
 528 {
 529         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 530 }
 531
 532 static inline unsigned dx_get_limit(struct dx_entry *entries)
 533 {
 534         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 535 }
 536
 537 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 538 {
 539         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 540 }
 541
 542 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 543 {
 544         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 545 }
 546
 547 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 548 {
 549         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 550                 EXT4_DIR_REC_LEN(2) - infosize;
 551
 552         if (ext4_has_metadata_csum(dir->i_sb))
 553                 entry_space -= sizeof(struct dx_tail);
 554         return entry_space / sizeof(struct dx_entry);
 555 }
 556
 557 static inline unsigned dx_node_limit(struct inode *dir)
 558 {
 559         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 560
 561         if (ext4_has_metadata_csum(dir->i_sb))
 562                 entry_space -= sizeof(struct dx_tail);
 563         return entry_space / sizeof(struct dx_entry);
 564 }
 565
 566 /*
 567  * Debug
 568  */
 569 #ifdef DX_DEBUG
 570 static void dx_show_index(char * label, struct dx_entry *entries)
 571 {
 572         int i, n = dx_get_count (entries);
 573         printk(KERN_DEBUG "%s index ", label);
 574         for (i = 0; i < n; i++) {
 575                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 576                                 0, (unsigned long)dx_get_block(entries + i));
 577         }
 578         printk("\n");
 579 }
 580
 581 struct stats
 582 {
 583         unsigned names;
 584         unsigned space;
 585         unsigned bcount;
 586 };
 587
 588 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 589                                  int size, int show_names)
 590 {
 591         unsigned names = 0, space = 0;
 592         char *base = (char *) de;
 593         struct dx_hash_info h = *hinfo;
 594
 595         printk("names: ");
 596         while ((char *) de < base + size)
 597         {
 598                 if (de->inode)
 599                 {
 600                         if (show_names)
 601                         {
 602                                 int len = de->name_len;
 603                                 char *name = de->name;
 604                                 while (len--) printk("%c", *name++);
 605                                 ext4fs_dirhash(de->name, de->name_len, &h);
 606                                 printk(":%x.%u ", h.hash,
 607                                        (unsigned) ((char *) de - base));
 608                         }
 609                         space += EXT4_DIR_REC_LEN(de->name_len);
 610                         names++;
 611                 }
 612                 de = ext4_next_entry(de, size);
 613         }
 614         printk("(%i)\n", names);
 615         return (struct stats) { names, space, 1 };
 616 }
 617
 618 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 619                              struct dx_entry *entries, int levels)
 620 {
 621         unsigned blocksize = dir->i_sb->s_blocksize;
 622         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 623         unsigned bcount = 0;
 624         struct buffer_head *bh;
 625         int err;
 626         printk("%i indexed blocks...\n", count);
 627         for (i = 0; i < count; i++, entries++)
 628         {
 629                 ext4_lblk_t block = dx_get_block(entries);
 630                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 631                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 632                 struct stats stats;
 633                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 634                 bh = ext4_bread(NULL,dir, block, 0);
 635                 if (!bh || IS_ERR(bh))
 636                         continue;
 637                 stats = levels?
 638                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 639                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 640                 names += stats.names;
 641                 space += stats.space;
 642                 bcount += stats.bcount;
 643                 brelse(bh);
 644         }
 645         if (bcount)
 646                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 647                        levels ? "" : "   ", names, space/bcount,
 648                        (space/bcount)*100/blocksize);
 649         return (struct stats) { names, space, bcount};
 650 }
 651 #endif /* DX_DEBUG */
 652
 653 /*
 654  * Probe for a directory leaf block to search.
 655  *
 656  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 657  * error in the directory index, and the caller should fall back to
 658  * searching the directory normally.  The callers of dx_probe **MUST**
 659  * check for this error code, and make sure it never gets reflected
 660  * back to userspace.
 661  */
 662 static struct dx_frame *
 663 dx_probe(const struct qstr *d_name, struct inode *dir,
 664          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 665 {
 666         unsigned count, indirect;
 667         struct dx_entry *at, *entries, *p, *q, *m;
 668         struct dx_root *root;
 669         struct dx_frame *frame = frame_in;
 670         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 671         u32 hash;
 672
 673         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 674         if (IS_ERR(frame->bh))
 675                 return (struct dx_frame *) frame->bh;
 676
 677         root = (struct dx_root *) frame->bh->b_data;
 678         if (root->info.hash_version != DX_HASH_TEA &&
 679             root->info.hash_version != DX_HASH_HALF_MD4 &&
 680             root->info.hash_version != DX_HASH_LEGACY) {
 681                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 682                              root->info.hash_version);
 683                 goto fail;
 684         }
 685         hinfo->hash_version = root->info.hash_version;
 686         if (hinfo->hash_version <= DX_HASH_TEA)
 687                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 688         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 689         if (d_name)
 690                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 691         hash = hinfo->hash;
 692
 693         if (root->info.unused_flags & 1) {
 694                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 695                              root->info.unused_flags);
 696                 goto fail;
 697         }
 698
 699         if ((indirect = root->info.indirect_levels) > 1) {
 700                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 701                              root->info.indirect_levels);
 702                 goto fail;
 703         }
 704
 705         entries = (struct dx_entry *) (((char *)&root->info) +
 706                                        root->info.info_length);
 707
 708         if (dx_get_limit(entries) != dx_root_limit(dir,
 709                                                    root->info.info_length)) {
 710                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 711                 goto fail;
 712         }
 713
 714         dxtrace(printk("Look up %x", hash));
 715         while (1) {
 716                 count = dx_get_count(entries);
 717                 if (!count || count > dx_get_limit(entries)) {
 718                         ext4_warning(dir->i_sb,
 719                                      "dx entry: no count or count > limit");
 720                         goto fail;
 721                 }
 722
 723                 p = entries + 1;
 724                 q = entries + count - 1;
 725                 while (p <= q) {
 726                         m = p + (q - p)/2;
 727                         dxtrace(printk("."));
 728                         if (dx_get_hash(m) > hash)
 729                                 q = m - 1;
 730                         else
 731                                 p = m + 1;
 732                 }
 733
 734                 if (0) { // linear search cross check
 735                         unsigned n = count - 1;
 736                         at = entries;
 737                         while (n--)
 738                         {
 739                                 dxtrace(printk(","));
 740                                 if (dx_get_hash(++at) > hash)
 741                                 {
 742                                         at--;
 743                                         break;
 744                                 }
 745                         }
 746                         assert (at == p - 1);
 747                 }
 748
 749                 at = p - 1;
 750                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 751                 frame->entries = entries;
 752                 frame->at = at;
 753                 if (!indirect--)
 754                         return frame;
 755                 frame++;
 756                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 757                 if (IS_ERR(frame->bh)) {
 758                         ret_err = (struct dx_frame *) frame->bh;
 759                         frame->bh = NULL;
 760                         goto fail;
 761                 }
 762                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 763
 764                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
 765                         ext4_warning(dir->i_sb,
 766                                      "dx entry: limit != node limit");
 767                         goto fail;
 768                 }
 769         }
 770 fail:
 771         while (frame >= frame_in) {
 772                 brelse(frame->bh);
 773                 frame--;
 774         }
 775         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 776                 ext4_warning(dir->i_sb,
 777                              "Corrupt dir inode %lu, running e2fsck is "
 778                              "recommended.", dir->i_ino);
 779         return ret_err;
 780 }
 781
 782 static void dx_release (struct dx_frame *frames)
 783 {
 784         if (frames[0].bh == NULL)
 785                 return;
 786
 787         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 788                 brelse(frames[1].bh);
 789         brelse(frames[0].bh);
 790 }
 791
 792 /*
 793  * This function increments the frame pointer to search the next leaf
 794  * block, and reads in the necessary intervening nodes if the search
 795  * should be necessary.  Whether or not the search is necessary is
 796  * controlled by the hash parameter.  If the hash value is even, then
 797  * the search is only continued if the next block starts with that
 798  * hash value.  This is used if we are searching for a specific file.
 799  *
 800  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 801  *
 802  * This function returns 1 if the caller should continue to search,
 803  * or 0 if it should not.  If there is an error reading one of the
 804  * index blocks, it will a negative error code.
 805  *
 806  * If start_hash is non-null, it will be filled in with the starting
 807  * hash of the next page.
 808  */
 809 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 810                                  struct dx_frame *frame,
 811                                  struct dx_frame *frames,
 812                                  __u32 *start_hash)
 813 {
 814         struct dx_frame *p;
 815         struct buffer_head *bh;
 816         int num_frames = 0;
 817         __u32 bhash;
 818
 819         p = frame;
 820         /*
 821          * Find the next leaf page by incrementing the frame pointer.
 822          * If we run out of entries in the interior node, loop around and
 823          * increment pointer in the parent node.  When we break out of
 824          * this loop, num_frames indicates the number of interior
 825          * nodes need to be read.
 826          */
 827         while (1) {
 828                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 829                         break;
 830                 if (p == frames)
 831                         return 0;
 832                 num_frames++;
 833                 p--;
 834         }
 835
 836         /*
 837          * If the hash is 1, then continue only if the next page has a
 838          * continuation hash of any value.  This is used for readdir
 839          * handling.  Otherwise, check to see if the hash matches the
 840          * desired contiuation hash.  If it doesn't, return since
 841          * there's no point to read in the successive index pages.
 842          */
 843         bhash = dx_get_hash(p->at);
 844         if (start_hash)
 845                 *start_hash = bhash;
 846         if ((hash & 1) == 0) {
 847                 if ((bhash & ~1) != hash)
 848                         return 0;
 849         }
 850         /*
 851          * If the hash is HASH_NB_ALWAYS, we always go to the next
 852          * block so no check is necessary
 853          */
 854         while (num_frames--) {
 855                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 856                 if (IS_ERR(bh))
 857                         return PTR_ERR(bh);
 858                 p++;
 859                 brelse(p->bh);
 860                 p->bh = bh;
 861                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 862         }
 863         return 1;
 864 }
 865
 866
 867 /*
 868  * This function fills a red-black tree with information from a
 869  * directory block.  It returns the number directory entries loaded
 870  * into the tree.  If there is an error it is returned in err.
 871  */
 872 static int htree_dirblock_to_tree(struct file *dir_file,
 873                                   struct inode *dir, ext4_lblk_t block,
 874                                   struct dx_hash_info *hinfo,
 875                                   __u32 start_hash, __u32 start_minor_hash)
 876 {
 877         struct buffer_head *bh;
 878         struct ext4_dir_entry_2 *de, *top;
 879         int err = 0, count = 0;
 880
 881         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 882                                                         (unsigned long)block));
 883         bh = ext4_read_dirblock(dir, block, DIRENT);
 884         if (IS_ERR(bh))
 885                 return PTR_ERR(bh);
 886
 887         de = (struct ext4_dir_entry_2 *) bh->b_data;
 888         top = (struct ext4_dir_entry_2 *) ((char *) de +
 889                                            dir->i_sb->s_blocksize -
 890                                            EXT4_DIR_REC_LEN(0));
 891         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 892                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 893                                 bh->b_data, bh->b_size,
 894                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 895                                          + ((char *)de - bh->b_data))) {
 896                         /* silently ignore the rest of the block */
 897                         break;
 898                 }
 899                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 900                 if ((hinfo->hash < start_hash) ||
 901                     ((hinfo->hash == start_hash) &&
 902                      (hinfo->minor_hash < start_minor_hash)))
 903                         continue;
 904                 if (de->inode == 0)
 905                         continue;
 906                 if ((err = ext4_htree_store_dirent(dir_file,
 907                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
 908                         brelse(bh);
 909                         return err;
 910                 }
 911                 count++;
 912         }
 913         brelse(bh);
 914         return count;
 915 }
 916
 917
 918 /*
 919  * This function fills a red-black tree with information from a
 920  * directory.  We start scanning the directory in hash order, starting
 921  * at start_hash and start_minor_hash.
 922  *
 923  * This function returns the number of entries inserted into the tree,
 924  * or a negative error code.
 925  */
 926 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 927                          __u32 start_minor_hash, __u32 *next_hash)
 928 {
 929         struct dx_hash_info hinfo;
 930         struct ext4_dir_entry_2 *de;
 931         struct dx_frame frames[2], *frame;
 932         struct inode *dir;
 933         ext4_lblk_t block;
 934         int count = 0;
 935         int ret, err;
 936         __u32 hashval;
 937
 938         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 939                        start_hash, start_minor_hash));
 940         dir = file_inode(dir_file);
 941         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 942                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 943                 if (hinfo.hash_version <= DX_HASH_TEA)
 944                         hinfo.hash_version +=
 945                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
 946                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 947                 if (ext4_has_inline_data(dir)) {
 948                         int has_inline_data = 1;
 949                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
 950                                                         &hinfo, start_hash,
 951                                                         start_minor_hash,
 952                                                         &has_inline_data);
 953                         if (has_inline_data) {
 954                                 *next_hash = ~0;
 955                                 return count;
 956                         }
 957                 }
 958                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 959                                                start_hash, start_minor_hash);
 960                 *next_hash = ~0;
 961                 return count;
 962         }
 963         hinfo.hash = start_hash;
 964         hinfo.minor_hash = 0;
 965         frame = dx_probe(NULL, dir, &hinfo, frames);
 966         if (IS_ERR(frame))
 967                 return PTR_ERR(frame);
 968
 969         /* Add '.' and '..' from the htree header */
 970         if (!start_hash && !start_minor_hash) {
 971                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 972                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 973                         goto errout;
 974                 count++;
 975         }
 976         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 977                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 978                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 979                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 980                         goto errout;
 981                 count++;
 982         }
 983
 984         while (1) {
 985                 block = dx_get_block(frame->at);
 986                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 987                                              start_hash, start_minor_hash);
 988                 if (ret < 0) {
 989                         err = ret;
 990                         goto errout;
 991                 }
 992                 count += ret;
 993                 hashval = ~0;
 994                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 995                                             frame, frames, &hashval);
 996                 *next_hash = hashval;
 997                 if (ret < 0) {
 998                         err = ret;
 999                         goto errout;
1000                 }
1001                 /*
1002                  * Stop if:  (a) there are no more entries, or
1003                  * (b) we have inserted at least one entry and the
1004                  * next hash value is not a continuation
1005                  */
1006                 if ((ret == 0) ||
1007                     (count && ((hashval & 1) == 0)))
1008                         break;
1009         }
1010         dx_release(frames);
1011         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1012                        "next hash: %x\n", count, *next_hash));
1013         return count;
1014 errout:
1015         dx_release(frames);
1016         return (err);
1017 }
1018
1019 static inline int search_dirblock(struct buffer_head *bh,
1020                                   struct inode *dir,
1021                                   const struct qstr *d_name,
1022                                   unsigned int offset,
1023                                   struct ext4_dir_entry_2 **res_dir)
1024 {
1025         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1026                           d_name, offset, res_dir);
1027 }
1028
1029 /*
1030  * Directory block splitting, compacting
1031  */
1032
1033 /*
1034  * Create map of hash values, offsets, and sizes, stored at end of block.
1035  * Returns number of entries mapped.
1036  */
1037 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1038                        struct dx_hash_info *hinfo,
1039                        struct dx_map_entry *map_tail)
1040 {
1041         int count = 0;
1042         char *base = (char *) de;
1043         struct dx_hash_info h = *hinfo;
1044
1045         while ((char *) de < base + blocksize) {
1046                 if (de->name_len && de->inode) {
1047                         ext4fs_dirhash(de->name, de->name_len, &h);
1048                         map_tail--;
1049                         map_tail->hash = h.hash;
1050                         map_tail->offs = ((char *) de - base)>>2;
1051                         map_tail->size = le16_to_cpu(de->rec_len);
1052                         count++;
1053                         cond_resched();
1054                 }
1055                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1056                 de = ext4_next_entry(de, blocksize);
1057         }
1058         return count;
1059 }
1060
1061 /* Sort map by hash value */
1062 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1063 {
1064         struct dx_map_entry *p, *q, *top = map + count - 1;
1065         int more;
1066         /* Combsort until bubble sort doesn't suck */
1067         while (count > 2) {
1068                 count = count*10/13;
1069                 if (count - 9 < 2) /* 9, 10 -> 11 */
1070                         count = 11;
1071                 for (p = top, q = p - count; q >= map; p--, q--)
1072                         if (p->hash < q->hash)
1073                                 swap(*p, *q);
1074         }
1075         /* Garden variety bubble sort */
1076         do {
1077                 more = 0;
1078                 q = top;
1079                 while (q-- > map) {
1080                         if (q[1].hash >= q[0].hash)
1081                                 continue;
1082                         swap(*(q+1), *q);
1083                         more = 1;
1084                 }
1085         } while(more);
1086 }
1087
1088 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1089 {
1090         struct dx_entry *entries = frame->entries;
1091         struct dx_entry *old = frame->at, *new = old + 1;
1092         int count = dx_get_count(entries);
1093
1094         assert(count < dx_get_limit(entries));
1095         assert(old < entries + count);
1096         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1097         dx_set_hash(new, hash);
1098         dx_set_block(new, block);
1099         dx_set_count(entries, count + 1);
1100 }
1101
1102 /*
1103  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1104  *
1105  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1106  * `de != NULL' is guaranteed by caller.
1107  */
1108 static inline int ext4_match (int len, const char * const name,
1109                               struct ext4_dir_entry_2 * de)
1110 {
1111         if (len != de->name_len)
1112                 return 0;
1113         if (!de->inode)
1114                 return 0;
1115         return !memcmp(name, de->name, len);
1116 }
1117
1118 /*
1119  * Returns 0 if not found, -1 on failure, and 1 on success
1120  */
1121 int search_dir(struct buffer_head *bh,
1122                char *search_buf,
1123                int buf_size,
1124                struct inode *dir,
1125                const struct qstr *d_name,
1126                unsigned int offset,
1127                struct ext4_dir_entry_2 **res_dir)
1128 {
1129         struct ext4_dir_entry_2 * de;
1130         char * dlimit;
1131         int de_len;
1132         const char *name = d_name->name;
1133         int namelen = d_name->len;
1134
1135         de = (struct ext4_dir_entry_2 *)search_buf;
1136         dlimit = search_buf + buf_size;
1137         while ((char *) de < dlimit) {
1138                 /* this code is executed quadratically often */
1139                 /* do minimal checking `by hand' */
1140
1141                 if ((char *) de + namelen <= dlimit &&
1142                     ext4_match (namelen, name, de)) {
1143                         /* found a match - just to be sure, do a full check */
1144                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1145                                                  bh->b_size, offset))
1146                                 return -1;
1147                         *res_dir = de;
1148                         return 1;
1149                 }
1150                 /* prevent looping on a bad block */
1151                 de_len = ext4_rec_len_from_disk(de->rec_len,
1152                                                 dir->i_sb->s_blocksize);
1153                 if (de_len <= 0)
1154                         return -1;
1155                 offset += de_len;
1156                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1157         }
1158         return 0;
1159 }
1160
1161 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1162                                struct ext4_dir_entry *de)
1163 {
1164         struct super_block *sb = dir->i_sb;
1165
1166         if (!is_dx(dir))
1167                 return 0;
1168         if (block == 0)
1169                 return 1;
1170         if (de->inode == 0 &&
1171             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1172                         sb->s_blocksize)
1173                 return 1;
1174         return 0;
1175 }
1176
1177 /*
1178  *      ext4_find_entry()
1179  *
1180  * finds an entry in the specified directory with the wanted name. It
1181  * returns the cache buffer in which the entry was found, and the entry
1182  * itself (as a parameter - res_dir). It does NOT read the inode of the
1183  * entry - you'll have to do that yourself if you want to.
1184  *
1185  * The returned buffer_head has ->b_count elevated.  The caller is expected
1186  * to brelse() it when appropriate.
1187  */
1188 static struct buffer_head * ext4_find_entry (struct inode *dir,
1189                                         const struct qstr *d_name,
1190                                         struct ext4_dir_entry_2 **res_dir,
1191                                         int *inlined)
1192 {
1193         struct super_block *sb;
1194         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1195         struct buffer_head *bh, *ret = NULL;
1196         ext4_lblk_t start, block, b;
1197         const u8 *name = d_name->name;
1198         int ra_max = 0;         /* Number of bh's in the readahead
1199                                    buffer, bh_use[] */
1200         int ra_ptr = 0;         /* Current index into readahead
1201                                    buffer */
1202         int num = 0;
1203         ext4_lblk_t  nblocks;
1204         int i, namelen;
1205
1206         *res_dir = NULL;
1207         sb = dir->i_sb;
1208         namelen = d_name->len;
1209         if (namelen > EXT4_NAME_LEN)
1210                 return NULL;
1211
1212         if (ext4_has_inline_data(dir)) {
1213                 int has_inline_data = 1;
1214                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1215                                              &has_inline_data);
1216                 if (has_inline_data) {
1217                         if (inlined)
1218                                 *inlined = 1;
1219                         return ret;
1220                 }
1221         }
1222
1223         if ((namelen <= 2) && (name[0] == '.') &&
1224             (name[1] == '.' || name[1] == '\0')) {
1225                 /*
1226                  * "." or ".." will only be in the first block
1227                  * NFS may look up ".."; "." should be handled by the VFS
1228                  */
1229                 block = start = 0;
1230                 nblocks = 1;
1231                 goto restart;
1232         }
1233         if (is_dx(dir)) {
1234                 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1235                 /*
1236                  * On success, or if the error was file not found,
1237                  * return.  Otherwise, fall back to doing a search the
1238                  * old fashioned way.
1239                  */
1240                 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1241                         return bh;
1242                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1243                                "falling back\n"));
1244         }
1245         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1246         start = EXT4_I(dir)->i_dir_start_lookup;
1247         if (start >= nblocks)
1248                 start = 0;
1249         block = start;
1250 restart:
1251         do {
1252                 /*
1253                  * We deal with the read-ahead logic here.
1254                  */
1255                 if (ra_ptr >= ra_max) {
1256                         /* Refill the readahead buffer */
1257                         ra_ptr = 0;
1258                         b = block;
1259                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1260                                 /*
1261                                  * Terminate if we reach the end of the
1262                                  * directory and must wrap, or if our
1263                                  * search has finished at this block.
1264                                  */
1265                                 if (b >= nblocks || (num && block == start)) {
1266                                         bh_use[ra_max] = NULL;
1267                                         break;
1268                                 }
1269                                 num++;
1270                                 bh = ext4_getblk(NULL, dir, b++, 0);
1271                                 if (unlikely(IS_ERR(bh))) {
1272                                         if (ra_max == 0)
1273                                                 return bh;
1274                                         break;
1275                                 }
1276                                 bh_use[ra_max] = bh;
1277                                 if (bh)
1278                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1279                                                     1, &bh);
1280                         }
1281                 }
1282                 if ((bh = bh_use[ra_ptr++]) == NULL)
1283                         goto next;
1284                 wait_on_buffer(bh);
1285                 if (!buffer_uptodate(bh)) {
1286                         /* read error, skip block & hope for the best */
1287                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1288                                          (unsigned long) block);
1289                         brelse(bh);
1290                         goto next;
1291                 }
1292                 if (!buffer_verified(bh) &&
1293                     !is_dx_internal_node(dir, block,
1294                                          (struct ext4_dir_entry *)bh->b_data) &&
1295                     !ext4_dirent_csum_verify(dir,
1296                                 (struct ext4_dir_entry *)bh->b_data)) {
1297                         EXT4_ERROR_INODE(dir, "checksumming directory "
1298                                          "block %lu", (unsigned long)block);
1299                         brelse(bh);
1300                         goto next;
1301                 }
1302                 set_buffer_verified(bh);
1303                 i = search_dirblock(bh, dir, d_name,
1304                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1305                 if (i == 1) {
1306                         EXT4_I(dir)->i_dir_start_lookup = block;
1307                         ret = bh;
1308                         goto cleanup_and_exit;
1309                 } else {
1310                         brelse(bh);
1311                         if (i < 0)
1312                                 goto cleanup_and_exit;
1313                 }
1314         next:
1315                 if (++block >= nblocks)
1316                         block = 0;
1317         } while (block != start);
1318
1319         /*
1320          * If the directory has grown while we were searching, then
1321          * search the last part of the directory before giving up.
1322          */
1323         block = nblocks;
1324         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1325         if (block < nblocks) {
1326                 start = 0;
1327                 goto restart;
1328         }
1329
1330 cleanup_and_exit:
1331         /* Clean up the read-ahead blocks */
1332         for (; ra_ptr < ra_max; ra_ptr++)
1333                 brelse(bh_use[ra_ptr]);
1334         return ret;
1335 }
1336
1337 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1338                        struct ext4_dir_entry_2 **res_dir)
1339 {
1340         struct super_block * sb = dir->i_sb;
1341         struct dx_hash_info     hinfo;
1342         struct dx_frame frames[2], *frame;
1343         struct buffer_head *bh;
1344         ext4_lblk_t block;
1345         int retval;
1346
1347         frame = dx_probe(d_name, dir, &hinfo, frames);
1348         if (IS_ERR(frame))
1349                 return (struct buffer_head *) frame;
1350         do {
1351                 block = dx_get_block(frame->at);
1352                 bh = ext4_read_dirblock(dir, block, DIRENT);
1353                 if (IS_ERR(bh))
1354                         goto errout;
1355
1356                 retval = search_dirblock(bh, dir, d_name,
1357                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1358                                          res_dir);
1359                 if (retval == 1)
1360                         goto success;
1361                 brelse(bh);
1362                 if (retval == -1) {
1363                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1364                         goto errout;
1365                 }
1366
1367                 /* Check to see if we should continue to search */
1368                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1369                                                frames, NULL);
1370                 if (retval < 0) {
1371                         ext4_warning(sb,
1372                              "error %d reading index page in directory #%lu",
1373                              retval, dir->i_ino);
1374                         bh = ERR_PTR(retval);
1375                         goto errout;
1376                 }
1377         } while (retval == 1);
1378
1379         bh = NULL;
1380 errout:
1381         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1382 success:
1383         dx_release(frames);
1384         return bh;
1385 }
1386
1387 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1388 {
1389         struct inode *inode;
1390         struct ext4_dir_entry_2 *de;
1391         struct buffer_head *bh;
1392
1393         if (dentry->d_name.len > EXT4_NAME_LEN)
1394                 return ERR_PTR(-ENAMETOOLONG);
1395
1396         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1397         if (IS_ERR(bh))
1398                 return (struct dentry *) bh;
1399         inode = NULL;
1400         if (bh) {
1401                 __u32 ino = le32_to_cpu(de->inode);
1402                 brelse(bh);
1403                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1404                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1405                         return ERR_PTR(-EIO);
1406                 }
1407                 if (unlikely(ino == dir->i_ino)) {
1408                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1409                                          dentry);
1410                         return ERR_PTR(-EIO);
1411                 }
1412                 inode = ext4_iget_normal(dir->i_sb, ino);
1413                 if (inode == ERR_PTR(-ESTALE)) {
1414                         EXT4_ERROR_INODE(dir,
1415                                          "deleted inode referenced: %u",
1416                                          ino);
1417                         return ERR_PTR(-EIO);
1418                 }
1419                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1420                     (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1421                      S_ISLNK(inode->i_mode)) &&
1422                     !ext4_is_child_context_consistent_with_parent(dir,
1423                                                                   inode)) {
1424                         iput(inode);
1425                         ext4_warning(inode->i_sb,
1426                                      "Inconsistent encryption contexts: %lu/%lu\n",
1427                                      (unsigned long) dir->i_ino,
1428                                      (unsigned long) inode->i_ino);
1429                         return ERR_PTR(-EPERM);
1430                 }
1431         }
1432         return d_splice_alias(inode, dentry);
1433 }
1434
1435
1436 struct dentry *ext4_get_parent(struct dentry *child)
1437 {
1438         __u32 ino;
1439         static const struct qstr dotdot = QSTR_INIT("..", 2);
1440         struct ext4_dir_entry_2 * de;
1441         struct buffer_head *bh;
1442
1443         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1444         if (IS_ERR(bh))
1445                 return (struct dentry *) bh;
1446         if (!bh)
1447                 return ERR_PTR(-ENOENT);
1448         ino = le32_to_cpu(de->inode);
1449         brelse(bh);
1450
1451         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1452                 EXT4_ERROR_INODE(child->d_inode,
1453                                  "bad parent inode number: %u", ino);
1454                 return ERR_PTR(-EIO);
1455         }
1456
1457         return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1458 }
1459
1460 /*
1461  * Move count entries from end of map between two memory locations.
1462  * Returns pointer to last entry moved.
1463  */
1464 static struct ext4_dir_entry_2 *
1465 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1466                 unsigned blocksize)
1467 {
1468         unsigned rec_len = 0;
1469
1470         while (count--) {
1471                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1472                                                 (from + (map->offs<<2));
1473                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1474                 memcpy (to, de, rec_len);
1475                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1476                                 ext4_rec_len_to_disk(rec_len, blocksize);
1477                 de->inode = 0;
1478                 map++;
1479                 to += rec_len;
1480         }
1481         return (struct ext4_dir_entry_2 *) (to - rec_len);
1482 }
1483
1484 /*
1485  * Compact each dir entry in the range to the minimal rec_len.
1486  * Returns pointer to last entry in range.
1487  */
1488 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1489 {
1490         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1491         unsigned rec_len = 0;
1492
1493         prev = to = de;
1494         while ((char*)de < base + blocksize) {
1495                 next = ext4_next_entry(de, blocksize);
1496                 if (de->inode && de->name_len) {
1497                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1498                         if (de > to)
1499                                 memmove(to, de, rec_len);
1500                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1501                         prev = to;
1502                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1503                 }
1504                 de = next;
1505         }
1506         return prev;
1507 }
1508
1509 /*
1510  * Split a full leaf block to make room for a new dir entry.
1511  * Allocate a new block, and move entries so that they are approx. equally full.
1512  * Returns pointer to de in block into which the new entry will be inserted.
1513  */
1514 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1515                         struct buffer_head **bh,struct dx_frame *frame,
1516                         struct dx_hash_info *hinfo)
1517 {
1518         unsigned blocksize = dir->i_sb->s_blocksize;
1519         unsigned count, continued;
1520         struct buffer_head *bh2;
1521         ext4_lblk_t newblock;
1522         u32 hash2;
1523         struct dx_map_entry *map;
1524         char *data1 = (*bh)->b_data, *data2;
1525         unsigned split, move, size;
1526         struct ext4_dir_entry_2 *de = NULL, *de2;
1527         struct ext4_dir_entry_tail *t;
1528         int     csum_size = 0;
1529         int     err = 0, i;
1530
1531         if (ext4_has_metadata_csum(dir->i_sb))
1532                 csum_size = sizeof(struct ext4_dir_entry_tail);
1533
1534         bh2 = ext4_append(handle, dir, &newblock);
1535         if (IS_ERR(bh2)) {
1536                 brelse(*bh);
1537                 *bh = NULL;
1538                 return (struct ext4_dir_entry_2 *) bh2;
1539         }
1540
1541         BUFFER_TRACE(*bh, "get_write_access");
1542         err = ext4_journal_get_write_access(handle, *bh);
1543         if (err)
1544                 goto journal_error;
1545
1546         BUFFER_TRACE(frame->bh, "get_write_access");
1547         err = ext4_journal_get_write_access(handle, frame->bh);
1548         if (err)
1549                 goto journal_error;
1550
1551         data2 = bh2->b_data;
1552
1553         /* create map in the end of data2 block */
1554         map = (struct dx_map_entry *) (data2 + blocksize);
1555         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1556                              blocksize, hinfo, map);
1557         map -= count;
1558         dx_sort_map(map, count);
1559         /* Split the existing block in the middle, size-wise */
1560         size = 0;
1561         move = 0;
1562         for (i = count-1; i >= 0; i--) {
1563                 /* is more than half of this entry in 2nd half of the block? */
1564                 if (size + map[i].size/2 > blocksize/2)
1565                         break;
1566                 size += map[i].size;
1567                 move++;
1568         }
1569         /* map index at which we will split */
1570         split = count - move;
1571         hash2 = map[split].hash;
1572         continued = hash2 == map[split - 1].hash;
1573         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1574                         (unsigned long)dx_get_block(frame->at),
1575                                         hash2, split, count-split));
1576
1577         /* Fancy dance to stay within two buffers */
1578         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1579         de = dx_pack_dirents(data1, blocksize);
1580         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1581                                            (char *) de,
1582                                            blocksize);
1583         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1584                                             (char *) de2,
1585                                             blocksize);
1586         if (csum_size) {
1587                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1588                 initialize_dirent_tail(t, blocksize);
1589
1590                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1591                 initialize_dirent_tail(t, blocksize);
1592         }
1593
1594         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1595         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1596
1597         /* Which block gets the new entry? */
1598         if (hinfo->hash >= hash2) {
1599                 swap(*bh, bh2);
1600                 de = de2;
1601         }
1602         dx_insert_block(frame, hash2 + continued, newblock);
1603         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1604         if (err)
1605                 goto journal_error;
1606         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1607         if (err)
1608                 goto journal_error;
1609         brelse(bh2);
1610         dxtrace(dx_show_index("frame", frame->entries));
1611         return de;
1612
1613 journal_error:
1614         brelse(*bh);
1615         brelse(bh2);
1616         *bh = NULL;
1617         ext4_std_error(dir->i_sb, err);
1618         return ERR_PTR(err);
1619 }
1620
1621 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1622                       struct buffer_head *bh,
1623                       void *buf, int buf_size,
1624                       const char *name, int namelen,
1625                       struct ext4_dir_entry_2 **dest_de)
1626 {
1627         struct ext4_dir_entry_2 *de;
1628         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1629         int nlen, rlen;
1630         unsigned int offset = 0;
1631         char *top;
1632
1633         de = (struct ext4_dir_entry_2 *)buf;
1634         top = buf + buf_size - reclen;
1635         while ((char *) de <= top) {
1636                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1637                                          buf, buf_size, offset))
1638                         return -EIO;
1639                 if (ext4_match(namelen, name, de))
1640                         return -EEXIST;
1641                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1642                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1643                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1644                         break;
1645                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1646                 offset += rlen;
1647         }
1648         if ((char *) de > top)
1649                 return -ENOSPC;
1650
1651         *dest_de = de;
1652         return 0;
1653 }
1654
1655 void ext4_insert_dentry(struct inode *inode,
1656                         struct ext4_dir_entry_2 *de,
1657                         int buf_size,
1658                         const char *name, int namelen)
1659 {
1660
1661         int nlen, rlen;
1662
1663         nlen = EXT4_DIR_REC_LEN(de->name_len);
1664         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1665         if (de->inode) {
1666                 struct ext4_dir_entry_2 *de1 =
1667                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1668                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1669                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1670                 de = de1;
1671         }
1672         de->file_type = EXT4_FT_UNKNOWN;
1673         de->inode = cpu_to_le32(inode->i_ino);
1674         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1675         de->name_len = namelen;
1676         memcpy(de->name, name, namelen);
1677 }
1678 /*
1679  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1680  * it points to a directory entry which is guaranteed to be large
1681  * enough for new directory entry.  If de is NULL, then
1682  * add_dirent_to_buf will attempt search the directory block for
1683  * space.  It will return -ENOSPC if no space is available, and -EIO
1684  * and -EEXIST if directory entry already exists.
1685  */
1686 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1687                              struct inode *inode, struct ext4_dir_entry_2 *de,
1688                              struct buffer_head *bh)
1689 {
1690         struct inode    *dir = dentry->d_parent->d_inode;
1691         const char      *name = dentry->d_name.name;
1692         int             namelen = dentry->d_name.len;
1693         unsigned int    blocksize = dir->i_sb->s_blocksize;
1694         int             csum_size = 0;
1695         int             err;
1696
1697         if (ext4_has_metadata_csum(inode->i_sb))
1698                 csum_size = sizeof(struct ext4_dir_entry_tail);
1699
1700         if (!de) {
1701                 err = ext4_find_dest_de(dir, inode,
1702                                         bh, bh->b_data, blocksize - csum_size,
1703                                         name, namelen, &de);
1704                 if (err)
1705                         return err;
1706         }
1707         BUFFER_TRACE(bh, "get_write_access");
1708         err = ext4_journal_get_write_access(handle, bh);
1709         if (err) {
1710                 ext4_std_error(dir->i_sb, err);
1711                 return err;
1712         }
1713
1714         /* By now the buffer is marked for journaling */
1715         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1716
1717         /*
1718          * XXX shouldn't update any times until successful
1719          * completion of syscall, but too many callers depend
1720          * on this.
1721          *
1722          * XXX similarly, too many callers depend on
1723          * ext4_new_inode() setting the times, but error
1724          * recovery deletes the inode, so the worst that can
1725          * happen is that the times are slightly out of date
1726          * and/or different from the directory change time.
1727          */
1728         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1729         ext4_update_dx_flag(dir);
1730         dir->i_version++;
1731         ext4_mark_inode_dirty(handle, dir);
1732         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1733         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1734         if (err)
1735                 ext4_std_error(dir->i_sb, err);
1736         return 0;
1737 }
1738
1739 /*
1740  * This converts a one block unindexed directory to a 3 block indexed
1741  * directory, and adds the dentry to the indexed directory.
1742  */
1743 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1744                             struct inode *inode, struct buffer_head *bh)
1745 {
1746         struct inode    *dir = dentry->d_parent->d_inode;
1747         const char      *name = dentry->d_name.name;
1748         int             namelen = dentry->d_name.len;
1749         struct buffer_head *bh2;
1750         struct dx_root  *root;
1751         struct dx_frame frames[2], *frame;
1752         struct dx_entry *entries;
1753         struct ext4_dir_entry_2 *de, *de2;
1754         struct ext4_dir_entry_tail *t;
1755         char            *data1, *top;
1756         unsigned        len;
1757         int             retval;
1758         unsigned        blocksize;
1759         struct dx_hash_info hinfo;
1760         ext4_lblk_t  block;
1761         struct fake_dirent *fde;
1762         int             csum_size = 0;
1763
1764         if (ext4_has_metadata_csum(inode->i_sb))
1765                 csum_size = sizeof(struct ext4_dir_entry_tail);
1766
1767         blocksize =  dir->i_sb->s_blocksize;
1768         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1769         BUFFER_TRACE(bh, "get_write_access");
1770         retval = ext4_journal_get_write_access(handle, bh);
1771         if (retval) {
1772                 ext4_std_error(dir->i_sb, retval);
1773                 brelse(bh);
1774                 return retval;
1775         }
1776         root = (struct dx_root *) bh->b_data;
1777
1778         /* The 0th block becomes the root, move the dirents out */
1779         fde = &root->dotdot;
1780         de = (struct ext4_dir_entry_2 *)((char *)fde +
1781                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1782         if ((char *) de >= (((char *) root) + blocksize)) {
1783                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1784                 brelse(bh);
1785                 return -EIO;
1786         }
1787         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1788
1789         /* Allocate new block for the 0th block's dirents */
1790         bh2 = ext4_append(handle, dir, &block);
1791         if (IS_ERR(bh2)) {
1792                 brelse(bh);
1793                 return PTR_ERR(bh2);
1794         }
1795         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1796         data1 = bh2->b_data;
1797
1798         memcpy (data1, de, len);
1799         de = (struct ext4_dir_entry_2 *) data1;
1800         top = data1 + len;
1801         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1802                 de = de2;
1803         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1804                                            (char *) de,
1805                                            blocksize);
1806
1807         if (csum_size) {
1808                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1809                 initialize_dirent_tail(t, blocksize);
1810         }
1811
1812         /* Initialize the root; the dot dirents already exist */
1813         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1814         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1815                                            blocksize);
1816         memset (&root->info, 0, sizeof(root->info));
1817         root->info.info_length = sizeof(root->info);
1818         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1819         entries = root->entries;
1820         dx_set_block(entries, 1);
1821         dx_set_count(entries, 1);
1822         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1823
1824         /* Initialize as for dx_probe */
1825         hinfo.hash_version = root->info.hash_version;
1826         if (hinfo.hash_version <= DX_HASH_TEA)
1827                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1828         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1829         ext4fs_dirhash(name, namelen, &hinfo);
1830         memset(frames, 0, sizeof(frames));
1831         frame = frames;
1832         frame->entries = entries;
1833         frame->at = entries;
1834         frame->bh = bh;
1835         bh = bh2;
1836
1837         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1838         if (retval)
1839                 goto out_frames;
1840         retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
1841         if (retval)
1842                 goto out_frames;
1843
1844         de = do_split(handle,dir, &bh, frame, &hinfo);
1845         if (IS_ERR(de)) {
1846                 retval = PTR_ERR(de);
1847                 goto out_frames;
1848         }
1849         dx_release(frames);
1850
1851         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1852         brelse(bh);
1853         return retval;
1854 out_frames:
1855         /*
1856          * Even if the block split failed, we have to properly write
1857          * out all the changes we did so far. Otherwise we can end up
1858          * with corrupted filesystem.
1859          */
1860         ext4_mark_inode_dirty(handle, dir);
1861         dx_release(frames);
1862         return retval;
1863 }
1864
1865 /*
1866  *      ext4_add_entry()
1867  *
1868  * adds a file entry to the specified directory, using the same
1869  * semantics as ext4_find_entry(). It returns NULL if it failed.
1870  *
1871  * NOTE!! The inode part of 'de' is left at 0 - which means you
1872  * may not sleep between calling this and putting something into
1873  * the entry, as someone else might have used it while you slept.
1874  */
1875 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1876                           struct inode *inode)
1877 {
1878         struct inode *dir = dentry->d_parent->d_inode;
1879         struct buffer_head *bh = NULL;
1880         struct ext4_dir_entry_2 *de;
1881         struct ext4_dir_entry_tail *t;
1882         struct super_block *sb;
1883         int     retval;
1884         int     dx_fallback=0;
1885         unsigned blocksize;
1886         ext4_lblk_t block, blocks;
1887         int     csum_size = 0;
1888
1889         if (ext4_has_metadata_csum(inode->i_sb))
1890                 csum_size = sizeof(struct ext4_dir_entry_tail);
1891
1892         sb = dir->i_sb;
1893         blocksize = sb->s_blocksize;
1894         if (!dentry->d_name.len)
1895                 return -EINVAL;
1896
1897         if (ext4_has_inline_data(dir)) {
1898                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1899                 if (retval < 0)
1900                         return retval;
1901                 if (retval == 1) {
1902                         retval = 0;
1903                         goto out;
1904                 }
1905         }
1906
1907         if (is_dx(dir)) {
1908                 retval = ext4_dx_add_entry(handle, dentry, inode);
1909                 if (!retval || (retval != ERR_BAD_DX_DIR))
1910                         goto out;
1911                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1912                 dx_fallback++;
1913                 ext4_mark_inode_dirty(handle, dir);
1914         }
1915         blocks = dir->i_size >> sb->s_blocksize_bits;
1916         for (block = 0; block < blocks; block++) {
1917                 bh = ext4_read_dirblock(dir, block, DIRENT);
1918                 if (IS_ERR(bh))
1919                         return PTR_ERR(bh);
1920
1921                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1922                 if (retval != -ENOSPC)
1923                         goto out;
1924
1925                 if (blocks == 1 && !dx_fallback &&
1926                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
1927                         retval = make_indexed_dir(handle, dentry, inode, bh);
1928                         bh = NULL; /* make_indexed_dir releases bh */
1929                         goto out;
1930                 }
1931                 brelse(bh);
1932         }
1933         bh = ext4_append(handle, dir, &block);
1934         if (IS_ERR(bh))
1935                 return PTR_ERR(bh);
1936         de = (struct ext4_dir_entry_2 *) bh->b_data;
1937         de->inode = 0;
1938         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1939
1940         if (csum_size) {
1941                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1942                 initialize_dirent_tail(t, blocksize);
1943         }
1944
1945         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1946 out:
1947         brelse(bh);
1948         if (retval == 0)
1949                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1950         return retval;
1951 }
1952
1953 /*
1954  * Returns 0 for success, or a negative error value
1955  */
1956 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1957                              struct inode *inode)
1958 {
1959         struct dx_frame frames[2], *frame;
1960         struct dx_entry *entries, *at;
1961         struct dx_hash_info hinfo;
1962         struct buffer_head *bh;
1963         struct inode *dir = dentry->d_parent->d_inode;
1964         struct super_block *sb = dir->i_sb;
1965         struct ext4_dir_entry_2 *de;
1966         int err;
1967
1968         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
1969         if (IS_ERR(frame))
1970                 return PTR_ERR(frame);
1971         entries = frame->entries;
1972         at = frame->at;
1973         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1974         if (IS_ERR(bh)) {
1975                 err = PTR_ERR(bh);
1976                 bh = NULL;
1977                 goto cleanup;
1978         }
1979
1980         BUFFER_TRACE(bh, "get_write_access");
1981         err = ext4_journal_get_write_access(handle, bh);
1982         if (err)
1983                 goto journal_error;
1984
1985         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1986         if (err != -ENOSPC)
1987                 goto cleanup;
1988
1989         /* Block full, should compress but for now just split */
1990         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1991                        dx_get_count(entries), dx_get_limit(entries)));
1992         /* Need to split index? */
1993         if (dx_get_count(entries) == dx_get_limit(entries)) {
1994                 ext4_lblk_t newblock;
1995                 unsigned icount = dx_get_count(entries);
1996                 int levels = frame - frames;
1997                 struct dx_entry *entries2;
1998                 struct dx_node *node2;
1999                 struct buffer_head *bh2;
2000
2001                 if (levels && (dx_get_count(frames->entries) ==
2002                                dx_get_limit(frames->entries))) {
2003                         ext4_warning(sb, "Directory index full!");
2004                         err = -ENOSPC;
2005                         goto cleanup;
2006                 }
2007                 bh2 = ext4_append(handle, dir, &newblock);
2008                 if (IS_ERR(bh2)) {
2009                         err = PTR_ERR(bh2);
2010                         goto cleanup;
2011                 }
2012                 node2 = (struct dx_node *)(bh2->b_data);
2013                 entries2 = node2->entries;
2014                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2015                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2016                                                            sb->s_blocksize);
2017                 BUFFER_TRACE(frame->bh, "get_write_access");
2018                 err = ext4_journal_get_write_access(handle, frame->bh);
2019                 if (err)
2020                         goto journal_error;
2021                 if (levels) {
2022                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2023                         unsigned hash2 = dx_get_hash(entries + icount1);
2024                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2025                                        icount1, icount2));
2026
2027                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2028                         err = ext4_journal_get_write_access(handle,
2029                                                              frames[0].bh);
2030                         if (err)
2031                                 goto journal_error;
2032
2033                         memcpy((char *) entries2, (char *) (entries + icount1),
2034                                icount2 * sizeof(struct dx_entry));
2035                         dx_set_count(entries, icount1);
2036                         dx_set_count(entries2, icount2);
2037                         dx_set_limit(entries2, dx_node_limit(dir));
2038
2039                         /* Which index block gets the new entry? */
2040                         if (at - entries >= icount1) {
2041                                 frame->at = at = at - entries - icount1 + entries2;
2042                                 frame->entries = entries = entries2;
2043                                 swap(frame->bh, bh2);
2044                         }
2045                         dx_insert_block(frames + 0, hash2, newblock);
2046                         dxtrace(dx_show_index("node", frames[1].entries));
2047                         dxtrace(dx_show_index("node",
2048                                ((struct dx_node *) bh2->b_data)->entries));
2049                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2050                         if (err)
2051                                 goto journal_error;
2052                         brelse (bh2);
2053                 } else {
2054                         dxtrace(printk(KERN_DEBUG
2055                                        "Creating second level index...\n"));
2056                         memcpy((char *) entries2, (char *) entries,
2057                                icount * sizeof(struct dx_entry));
2058                         dx_set_limit(entries2, dx_node_limit(dir));
2059
2060                         /* Set up root */
2061                         dx_set_count(entries, 1);
2062                         dx_set_block(entries + 0, newblock);
2063                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2064
2065                         /* Add new access path frame */
2066                         frame = frames + 1;
2067                         frame->at = at = at - entries + entries2;
2068                         frame->entries = entries = entries2;
2069                         frame->bh = bh2;
2070                         err = ext4_journal_get_write_access(handle,
2071                                                              frame->bh);
2072                         if (err)
2073                                 goto journal_error;
2074                 }
2075                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2076                 if (err) {
2077                         ext4_std_error(inode->i_sb, err);
2078                         goto cleanup;
2079                 }
2080         }
2081         de = do_split(handle, dir, &bh, frame, &hinfo);
2082         if (IS_ERR(de)) {
2083                 err = PTR_ERR(de);
2084                 goto cleanup;
2085         }
2086         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2087         goto cleanup;
2088
2089 journal_error:
2090         ext4_std_error(dir->i_sb, err);
2091 cleanup:
2092         brelse(bh);
2093         dx_release(frames);
2094         return err;
2095 }
2096
2097 /*
2098  * ext4_generic_delete_entry deletes a directory entry by merging it
2099  * with the previous entry
2100  */
2101 int ext4_generic_delete_entry(handle_t *handle,
2102                               struct inode *dir,
2103                               struct ext4_dir_entry_2 *de_del,
2104                               struct buffer_head *bh,
2105                               void *entry_buf,
2106                               int buf_size,
2107                               int csum_size)
2108 {
2109         struct ext4_dir_entry_2 *de, *pde;
2110         unsigned int blocksize = dir->i_sb->s_blocksize;
2111         int i;
2112
2113         i = 0;
2114         pde = NULL;
2115         de = (struct ext4_dir_entry_2 *)entry_buf;
2116         while (i < buf_size - csum_size) {
2117                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2118                                          bh->b_data, bh->b_size, i))
2119                         return -EIO;
2120                 if (de == de_del)  {
2121                         if (pde)
2122                                 pde->rec_len = ext4_rec_len_to_disk(
2123                                         ext4_rec_len_from_disk(pde->rec_len,
2124                                                                blocksize) +
2125                                         ext4_rec_len_from_disk(de->rec_len,
2126                                                                blocksize),
2127                                         blocksize);
2128                         else
2129                                 de->inode = 0;
2130                         dir->i_version++;
2131                         return 0;
2132                 }
2133                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2134                 pde = de;
2135                 de = ext4_next_entry(de, blocksize);
2136         }
2137         return -ENOENT;
2138 }
2139
2140 static int ext4_delete_entry(handle_t *handle,
2141                              struct inode *dir,
2142                              struct ext4_dir_entry_2 *de_del,
2143                              struct buffer_head *bh)
2144 {
2145         int err, csum_size = 0;
2146
2147         if (ext4_has_inline_data(dir)) {
2148                 int has_inline_data = 1;
2149                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2150                                                &has_inline_data);
2151                 if (has_inline_data)
2152                         return err;
2153         }
2154
2155         if (ext4_has_metadata_csum(dir->i_sb))
2156                 csum_size = sizeof(struct ext4_dir_entry_tail);
2157
2158         BUFFER_TRACE(bh, "get_write_access");
2159         err = ext4_journal_get_write_access(handle, bh);
2160         if (unlikely(err))
2161                 goto out;
2162
2163         err = ext4_generic_delete_entry(handle, dir, de_del,
2164                                         bh, bh->b_data,
2165                                         dir->i_sb->s_blocksize, csum_size);
2166         if (err)
2167                 goto out;
2168
2169         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2170         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2171         if (unlikely(err))
2172                 goto out;
2173
2174         return 0;
2175 out:
2176         if (err != -ENOENT)
2177                 ext4_std_error(dir->i_sb, err);
2178         return err;
2179 }
2180
2181 /*
2182  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2183  * since this indicates that nlinks count was previously 1.
2184  */
2185 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2186 {
2187         inc_nlink(inode);
2188         if (is_dx(inode) && inode->i_nlink > 1) {
2189                 /* limit is 16-bit i_links_count */
2190                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2191                         set_nlink(inode, 1);
2192                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2193                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2194                 }
2195         }
2196 }
2197
2198 /*
2199  * If a directory had nlink == 1, then we should let it be 1. This indicates
2200  * directory has >EXT4_LINK_MAX subdirs.
2201  */
2202 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2203 {
2204         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2205                 drop_nlink(inode);
2206 }
2207
2208
2209 static int ext4_add_nondir(handle_t *handle,
2210                 struct dentry *dentry, struct inode *inode)
2211 {
2212         int err = ext4_add_entry(handle, dentry, inode);
2213         if (!err) {
2214                 ext4_mark_inode_dirty(handle, inode);
2215                 unlock_new_inode(inode);
2216                 d_instantiate(dentry, inode);
2217                 return 0;
2218         }
2219         drop_nlink(inode);
2220         unlock_new_inode(inode);
2221         iput(inode);
2222         return err;
2223 }
2224
2225 /*
2226  * By the time this is called, we already have created
2227  * the directory cache entry for the new file, but it
2228  * is so far negative - it has no inode.
2229  *
2230  * If the create succeeds, we fill in the inode information
2231  * with d_instantiate().
2232  */
2233 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2234                        bool excl)
2235 {
2236         handle_t *handle;
2237         struct inode *inode;
2238         int err, credits, retries = 0;
2239
2240         dquot_initialize(dir);
2241
2242         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2243                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2244 retry:
2245         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2246                                             NULL, EXT4_HT_DIR, credits);
2247         handle = ext4_journal_current_handle();
2248         err = PTR_ERR(inode);
2249         if (!IS_ERR(inode)) {
2250                 inode->i_op = &ext4_file_inode_operations;
2251                 if (test_opt(inode->i_sb, DAX))
2252                         inode->i_fop = &ext4_dax_file_operations;
2253                 else
2254                         inode->i_fop = &ext4_file_operations;
2255                 ext4_set_aops(inode);
2256                 err = 0;
2257 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2258                 if (!err && ext4_encrypted_inode(dir)) {
2259                         err = ext4_inherit_context(dir, inode);
2260                         if (err) {
2261                                 clear_nlink(inode);
2262                                 unlock_new_inode(inode);
2263                                 iput(inode);
2264                         }
2265                 }
2266 #endif
2267                 if (!err)
2268                         err = ext4_add_nondir(handle, dentry, inode);
2269                 if (!err && IS_DIRSYNC(dir))
2270                         ext4_handle_sync(handle);
2271         }
2272         if (handle)
2273                 ext4_journal_stop(handle);
2274         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2275                 goto retry;
2276         return err;
2277 }
2278
2279 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2280                       umode_t mode, dev_t rdev)
2281 {
2282         handle_t *handle;
2283         struct inode *inode;
2284         int err, credits, retries = 0;
2285
2286         if (!new_valid_dev(rdev))
2287                 return -EINVAL;
2288
2289         dquot_initialize(dir);
2290
2291         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2292                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2293 retry:
2294         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2295                                             NULL, EXT4_HT_DIR, credits);
2296         handle = ext4_journal_current_handle();
2297         err = PTR_ERR(inode);
2298         if (!IS_ERR(inode)) {
2299                 init_special_inode(inode, inode->i_mode, rdev);
2300                 inode->i_op = &ext4_special_inode_operations;
2301                 err = ext4_add_nondir(handle, dentry, inode);
2302                 if (!err && IS_DIRSYNC(dir))
2303                         ext4_handle_sync(handle);
2304         }
2305         if (handle)
2306                 ext4_journal_stop(handle);
2307         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2308                 goto retry;
2309         return err;
2310 }
2311
2312 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2313 {
2314         handle_t *handle;
2315         struct inode *inode;
2316         int err, retries = 0;
2317
2318         dquot_initialize(dir);
2319
2320 retry:
2321         inode = ext4_new_inode_start_handle(dir, mode,
2322                                             NULL, 0, NULL,
2323                                             EXT4_HT_DIR,
2324                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2325                           4 + EXT4_XATTR_TRANS_BLOCKS);
2326         handle = ext4_journal_current_handle();
2327         err = PTR_ERR(inode);
2328         if (!IS_ERR(inode)) {
2329                 inode->i_op = &ext4_file_inode_operations;
2330                 if (test_opt(inode->i_sb, DAX))
2331                         inode->i_fop = &ext4_dax_file_operations;
2332                 else
2333                         inode->i_fop = &ext4_file_operations;
2334                 ext4_set_aops(inode);
2335                 d_tmpfile(dentry, inode);
2336                 err = ext4_orphan_add(handle, inode);
2337                 if (err)
2338                         goto err_unlock_inode;
2339                 mark_inode_dirty(inode);
2340                 unlock_new_inode(inode);
2341         }
2342         if (handle)
2343                 ext4_journal_stop(handle);
2344         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2345                 goto retry;
2346         return err;
2347 err_unlock_inode:
2348         ext4_journal_stop(handle);
2349         unlock_new_inode(inode);
2350         return err;
2351 }
2352
2353 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2354                           struct ext4_dir_entry_2 *de,
2355                           int blocksize, int csum_size,
2356                           unsigned int parent_ino, int dotdot_real_len)
2357 {
2358         de->inode = cpu_to_le32(inode->i_ino);
2359         de->name_len = 1;
2360         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2361                                            blocksize);
2362         strcpy(de->name, ".");
2363         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2364
2365         de = ext4_next_entry(de, blocksize);
2366         de->inode = cpu_to_le32(parent_ino);
2367         de->name_len = 2;
2368         if (!dotdot_real_len)
2369                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2370                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2371                                         blocksize);
2372         else
2373                 de->rec_len = ext4_rec_len_to_disk(
2374                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2375         strcpy(de->name, "..");
2376         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2377
2378         return ext4_next_entry(de, blocksize);
2379 }
2380
2381 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2382                              struct inode *inode)
2383 {
2384         struct buffer_head *dir_block = NULL;
2385         struct ext4_dir_entry_2 *de;
2386         struct ext4_dir_entry_tail *t;
2387         ext4_lblk_t block = 0;
2388         unsigned int blocksize = dir->i_sb->s_blocksize;
2389         int csum_size = 0;
2390         int err;
2391
2392         if (ext4_has_metadata_csum(dir->i_sb))
2393                 csum_size = sizeof(struct ext4_dir_entry_tail);
2394
2395         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2396                 err = ext4_try_create_inline_dir(handle, dir, inode);
2397                 if (err < 0 && err != -ENOSPC)
2398                         goto out;
2399                 if (!err)
2400                         goto out;
2401         }
2402
2403         inode->i_size = 0;
2404         dir_block = ext4_append(handle, inode, &block);
2405         if (IS_ERR(dir_block))
2406                 return PTR_ERR(dir_block);
2407         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2408         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2409         set_nlink(inode, 2);
2410         if (csum_size) {
2411                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2412                 initialize_dirent_tail(t, blocksize);
2413         }
2414
2415         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2416         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2417         if (err)
2418                 goto out;
2419         set_buffer_verified(dir_block);
2420 out:
2421         brelse(dir_block);
2422         return err;
2423 }
2424
2425 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2426 {
2427         handle_t *handle;
2428         struct inode *inode;
2429         int err, credits, retries = 0;
2430
2431         if (EXT4_DIR_LINK_MAX(dir))
2432                 return -EMLINK;
2433
2434         dquot_initialize(dir);
2435
2436         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2437                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2438 retry:
2439         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2440                                             &dentry->d_name,
2441                                             0, NULL, EXT4_HT_DIR, credits);
2442         handle = ext4_journal_current_handle();
2443         err = PTR_ERR(inode);
2444         if (IS_ERR(inode))
2445                 goto out_stop;
2446
2447         inode->i_op = &ext4_dir_inode_operations;
2448         inode->i_fop = &ext4_dir_operations;
2449         err = ext4_init_new_dir(handle, dir, inode);
2450         if (err)
2451                 goto out_clear_inode;
2452 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2453         if (ext4_encrypted_inode(dir)) {
2454                 err = ext4_inherit_context(dir, inode);
2455                 if (err)
2456                         goto out_clear_inode;
2457         }
2458 #endif
2459         err = ext4_mark_inode_dirty(handle, inode);
2460         if (!err)
2461                 err = ext4_add_entry(handle, dentry, inode);
2462         if (err) {
2463 out_clear_inode:
2464                 clear_nlink(inode);
2465                 unlock_new_inode(inode);
2466                 ext4_mark_inode_dirty(handle, inode);
2467                 iput(inode);
2468                 goto out_stop;
2469         }
2470         ext4_inc_count(handle, dir);
2471         ext4_update_dx_flag(dir);
2472         err = ext4_mark_inode_dirty(handle, dir);
2473         if (err)
2474                 goto out_clear_inode;
2475         unlock_new_inode(inode);
2476         d_instantiate(dentry, inode);
2477         if (IS_DIRSYNC(dir))
2478                 ext4_handle_sync(handle);
2479
2480 out_stop:
2481         if (handle)
2482                 ext4_journal_stop(handle);
2483         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2484                 goto retry;
2485         return err;
2486 }
2487
2488 /*
2489  * routine to check that the specified directory is empty (for rmdir)
2490  */
2491 int ext4_empty_dir(struct inode *inode)
2492 {
2493         unsigned int offset;
2494         struct buffer_head *bh;
2495         struct ext4_dir_entry_2 *de, *de1;
2496         struct super_block *sb;
2497         int err = 0;
2498
2499         if (ext4_has_inline_data(inode)) {
2500                 int has_inline_data = 1;
2501
2502                 err = empty_inline_dir(inode, &has_inline_data);
2503                 if (has_inline_data)
2504                         return err;
2505         }
2506
2507         sb = inode->i_sb;
2508         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2509                 EXT4_ERROR_INODE(inode, "invalid size");
2510                 return 1;
2511         }
2512         bh = ext4_read_dirblock(inode, 0, EITHER);
2513         if (IS_ERR(bh))
2514                 return 1;
2515
2516         de = (struct ext4_dir_entry_2 *) bh->b_data;
2517         de1 = ext4_next_entry(de, sb->s_blocksize);
2518         if (le32_to_cpu(de->inode) != inode->i_ino ||
2519                         !le32_to_cpu(de1->inode) ||
2520                         strcmp(".", de->name) ||
2521                         strcmp("..", de1->name)) {
2522                 ext4_warning(inode->i_sb,
2523                              "bad directory (dir #%lu) - no `.' or `..'",
2524                              inode->i_ino);
2525                 brelse(bh);
2526                 return 1;
2527         }
2528         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2529                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2530         de = ext4_next_entry(de1, sb->s_blocksize);
2531         while (offset < inode->i_size) {
2532                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2533                         unsigned int lblock;
2534                         err = 0;
2535                         brelse(bh);
2536                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2537                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2538                         if (IS_ERR(bh))
2539                                 return 1;
2540                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2541                 }
2542                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2543                                          bh->b_data, bh->b_size, offset)) {
2544                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2545                                                          sb->s_blocksize);
2546                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2547                         continue;
2548                 }
2549                 if (le32_to_cpu(de->inode)) {
2550                         brelse(bh);
2551                         return 0;
2552                 }
2553                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2554                 de = ext4_next_entry(de, sb->s_blocksize);
2555         }
2556         brelse(bh);
2557         return 1;
2558 }
2559
2560 /*
2561  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2562  * such inodes, starting at the superblock, in case we crash before the
2563  * file is closed/deleted, or in case the inode truncate spans multiple
2564  * transactions and the last transaction is not recovered after a crash.
2565  *
2566  * At filesystem recovery time, we walk this list deleting unlinked
2567  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2568  *
2569  * Orphan list manipulation functions must be called under i_mutex unless
2570  * we are just creating the inode or deleting it.
2571  */
2572 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2573 {
2574         struct super_block *sb = inode->i_sb;
2575         struct ext4_sb_info *sbi = EXT4_SB(sb);
2576         struct ext4_iloc iloc;
2577         int err = 0, rc;
2578         bool dirty = false;
2579
2580         if (!sbi->s_journal || is_bad_inode(inode))
2581                 return 0;
2582
2583         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2584                      !mutex_is_locked(&inode->i_mutex));
2585         /*
2586          * Exit early if inode already is on orphan list. This is a big speedup
2587          * since we don't have to contend on the global s_orphan_lock.
2588          */
2589         if (!list_empty(&EXT4_I(inode)->i_orphan))
2590                 return 0;
2591
2592         /*
2593          * Orphan handling is only valid for files with data blocks
2594          * being truncated, or files being unlinked. Note that we either
2595          * hold i_mutex, or the inode can not be referenced from outside,
2596          * so i_nlink should not be bumped due to race
2597          */
2598         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2599                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2600
2601         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2602         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2603         if (err)
2604                 goto out;
2605
2606         err = ext4_reserve_inode_write(handle, inode, &iloc);
2607         if (err)
2608                 goto out;
2609
2610         mutex_lock(&sbi->s_orphan_lock);
2611         /*
2612          * Due to previous errors inode may be already a part of on-disk
2613          * orphan list. If so skip on-disk list modification.
2614          */
2615         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2616             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2617                 /* Insert this inode at the head of the on-disk orphan list */
2618                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2619                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2620                 dirty = true;
2621         }
2622         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2623         mutex_unlock(&sbi->s_orphan_lock);
2624
2625         if (dirty) {
2626                 err = ext4_handle_dirty_super(handle, sb);
2627                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2628                 if (!err)
2629                         err = rc;
2630                 if (err) {
2631                         /*
2632                          * We have to remove inode from in-memory list if
2633                          * addition to on disk orphan list failed. Stray orphan
2634                          * list entries can cause panics at unmount time.
2635                          */
2636                         mutex_lock(&sbi->s_orphan_lock);
2637                         list_del(&EXT4_I(inode)->i_orphan);
2638                         mutex_unlock(&sbi->s_orphan_lock);
2639                 }
2640         }
2641         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2642         jbd_debug(4, "orphan inode %lu will point to %d\n",
2643                         inode->i_ino, NEXT_ORPHAN(inode));
2644 out:
2645         ext4_std_error(sb, err);
2646         return err;
2647 }
2648
2649 /*
2650  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2651  * of such inodes stored on disk, because it is finally being cleaned up.
2652  */
2653 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2654 {
2655         struct list_head *prev;
2656         struct ext4_inode_info *ei = EXT4_I(inode);
2657         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2658         __u32 ino_next;
2659         struct ext4_iloc iloc;
2660         int err = 0;
2661
2662         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2663                 return 0;
2664
2665         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2666                      !mutex_is_locked(&inode->i_mutex));
2667         /* Do this quick check before taking global s_orphan_lock. */
2668         if (list_empty(&ei->i_orphan))
2669                 return 0;
2670
2671         if (handle) {
2672                 /* Grab inode buffer early before taking global s_orphan_lock */
2673                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2674         }
2675
2676         mutex_lock(&sbi->s_orphan_lock);
2677         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2678
2679         prev = ei->i_orphan.prev;
2680         list_del_init(&ei->i_orphan);
2681
2682         /* If we're on an error path, we may not have a valid
2683          * transaction handle with which to update the orphan list on
2684          * disk, but we still need to remove the inode from the linked
2685          * list in memory. */
2686         if (!handle || err) {
2687                 mutex_unlock(&sbi->s_orphan_lock);
2688                 goto out_err;
2689         }
2690
2691         ino_next = NEXT_ORPHAN(inode);
2692         if (prev == &sbi->s_orphan) {
2693                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2694                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2695                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2696                 if (err) {
2697                         mutex_unlock(&sbi->s_orphan_lock);
2698                         goto out_brelse;
2699                 }
2700                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2701                 mutex_unlock(&sbi->s_orphan_lock);
2702                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2703         } else {
2704                 struct ext4_iloc iloc2;
2705                 struct inode *i_prev =
2706                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2707
2708                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2709                           i_prev->i_ino, ino_next);
2710                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2711                 if (err) {
2712                         mutex_unlock(&sbi->s_orphan_lock);
2713                         goto out_brelse;
2714                 }
2715                 NEXT_ORPHAN(i_prev) = ino_next;
2716                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2717                 mutex_unlock(&sbi->s_orphan_lock);
2718         }
2719         if (err)
2720                 goto out_brelse;
2721         NEXT_ORPHAN(inode) = 0;
2722         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2723 out_err:
2724         ext4_std_error(inode->i_sb, err);
2725         return err;
2726
2727 out_brelse:
2728         brelse(iloc.bh);
2729         goto out_err;
2730 }
2731
2732 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2733 {
2734         int retval;
2735         struct inode *inode;
2736         struct buffer_head *bh;
2737         struct ext4_dir_entry_2 *de;
2738         handle_t *handle = NULL;
2739
2740         /* Initialize quotas before so that eventual writes go in
2741          * separate transaction */
2742         dquot_initialize(dir);
2743         dquot_initialize(dentry->d_inode);
2744
2745         retval = -ENOENT;
2746         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2747         if (IS_ERR(bh))
2748                 return PTR_ERR(bh);
2749         if (!bh)
2750                 goto end_rmdir;
2751
2752         inode = dentry->d_inode;
2753
2754         retval = -EIO;
2755         if (le32_to_cpu(de->inode) != inode->i_ino)
2756                 goto end_rmdir;
2757
2758         retval = -ENOTEMPTY;
2759         if (!ext4_empty_dir(inode))
2760                 goto end_rmdir;
2761
2762         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2763                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2764         if (IS_ERR(handle)) {
2765                 retval = PTR_ERR(handle);
2766                 handle = NULL;
2767                 goto end_rmdir;
2768         }
2769
2770         if (IS_DIRSYNC(dir))
2771                 ext4_handle_sync(handle);
2772
2773         retval = ext4_delete_entry(handle, dir, de, bh);
2774         if (retval)
2775                 goto end_rmdir;
2776         if (!EXT4_DIR_LINK_EMPTY(inode))
2777                 ext4_warning(inode->i_sb,
2778                              "empty directory has too many links (%d)",
2779                              inode->i_nlink);
2780         inode->i_version++;
2781         clear_nlink(inode);
2782         /* There's no need to set i_disksize: the fact that i_nlink is
2783          * zero will ensure that the right thing happens during any
2784          * recovery. */
2785         inode->i_size = 0;
2786         ext4_orphan_add(handle, inode);
2787         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2788         ext4_mark_inode_dirty(handle, inode);
2789         ext4_dec_count(handle, dir);
2790         ext4_update_dx_flag(dir);
2791         ext4_mark_inode_dirty(handle, dir);
2792
2793 end_rmdir:
2794         brelse(bh);
2795         if (handle)
2796                 ext4_journal_stop(handle);
2797         return retval;
2798 }
2799
2800 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2801 {
2802         int retval;
2803         struct inode *inode;
2804         struct buffer_head *bh;
2805         struct ext4_dir_entry_2 *de;
2806         handle_t *handle = NULL;
2807
2808         trace_ext4_unlink_enter(dir, dentry);
2809         /* Initialize quotas before so that eventual writes go
2810          * in separate transaction */
2811         dquot_initialize(dir);
2812         dquot_initialize(dentry->d_inode);
2813
2814         retval = -ENOENT;
2815         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2816         if (IS_ERR(bh))
2817                 return PTR_ERR(bh);
2818         if (!bh)
2819                 goto end_unlink;
2820
2821         inode = dentry->d_inode;
2822
2823         retval = -EIO;
2824         if (le32_to_cpu(de->inode) != inode->i_ino)
2825                 goto end_unlink;
2826
2827         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2828                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2829         if (IS_ERR(handle)) {
2830                 retval = PTR_ERR(handle);
2831                 handle = NULL;
2832                 goto end_unlink;
2833         }
2834
2835         if (IS_DIRSYNC(dir))
2836                 ext4_handle_sync(handle);
2837
2838         if (!inode->i_nlink) {
2839                 ext4_warning(inode->i_sb,
2840                              "Deleting nonexistent file (%lu), %d",
2841                              inode->i_ino, inode->i_nlink);
2842                 set_nlink(inode, 1);
2843         }
2844         retval = ext4_delete_entry(handle, dir, de, bh);
2845         if (retval)
2846                 goto end_unlink;
2847         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2848         ext4_update_dx_flag(dir);
2849         ext4_mark_inode_dirty(handle, dir);
2850         drop_nlink(inode);
2851         if (!inode->i_nlink)
2852                 ext4_orphan_add(handle, inode);
2853         inode->i_ctime = ext4_current_time(inode);
2854         ext4_mark_inode_dirty(handle, inode);
2855
2856 end_unlink:
2857         brelse(bh);
2858         if (handle)
2859                 ext4_journal_stop(handle);
2860         trace_ext4_unlink_exit(dentry, retval);
2861         return retval;
2862 }
2863
2864 static int ext4_symlink(struct inode *dir,
2865                         struct dentry *dentry, const char *symname)
2866 {
2867         handle_t *handle;
2868         struct inode *inode;
2869         int l, err, retries = 0;
2870         int credits;
2871
2872         l = strlen(symname)+1;
2873         if (l > dir->i_sb->s_blocksize)
2874                 return -ENAMETOOLONG;
2875
2876         dquot_initialize(dir);
2877
2878         if (l > EXT4_N_BLOCKS * 4) {
2879                 /*
2880                  * For non-fast symlinks, we just allocate inode and put it on
2881                  * orphan list in the first transaction => we need bitmap,
2882                  * group descriptor, sb, inode block, quota blocks, and
2883                  * possibly selinux xattr blocks.
2884                  */
2885                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2886                           EXT4_XATTR_TRANS_BLOCKS;
2887         } else {
2888                 /*
2889                  * Fast symlink. We have to add entry to directory
2890                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2891                  * allocate new inode (bitmap, group descriptor, inode block,
2892                  * quota blocks, sb is already counted in previous macros).
2893                  */
2894                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2895                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2896         }
2897 retry:
2898         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2899                                             &dentry->d_name, 0, NULL,
2900                                             EXT4_HT_DIR, credits);
2901         handle = ext4_journal_current_handle();
2902         err = PTR_ERR(inode);
2903         if (IS_ERR(inode))
2904                 goto out_stop;
2905
2906         if (l > EXT4_N_BLOCKS * 4) {
2907                 inode->i_op = &ext4_symlink_inode_operations;
2908                 ext4_set_aops(inode);
2909                 /*
2910                  * We cannot call page_symlink() with transaction started
2911                  * because it calls into ext4_write_begin() which can wait
2912                  * for transaction commit if we are running out of space
2913                  * and thus we deadlock. So we have to stop transaction now
2914                  * and restart it when symlink contents is written.
2915                  *
2916                  * To keep fs consistent in case of crash, we have to put inode
2917                  * to orphan list in the mean time.
2918                  */
2919                 drop_nlink(inode);
2920                 err = ext4_orphan_add(handle, inode);
2921                 ext4_journal_stop(handle);
2922                 if (err)
2923                         goto err_drop_inode;
2924                 err = __page_symlink(inode, symname, l, 1);
2925                 if (err)
2926                         goto err_drop_inode;
2927                 /*
2928                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2929                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2930                  */
2931                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2932                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2933                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2934                 if (IS_ERR(handle)) {
2935                         err = PTR_ERR(handle);
2936                         goto err_drop_inode;
2937                 }
2938                 set_nlink(inode, 1);
2939                 err = ext4_orphan_del(handle, inode);
2940                 if (err) {
2941                         ext4_journal_stop(handle);
2942                         clear_nlink(inode);
2943                         goto err_drop_inode;
2944                 }
2945         } else {
2946                 /* clear the extent format for fast symlink */
2947                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2948                 inode->i_op = &ext4_fast_symlink_inode_operations;
2949                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2950                 inode->i_size = l-1;
2951         }
2952         EXT4_I(inode)->i_disksize = inode->i_size;
2953         err = ext4_add_nondir(handle, dentry, inode);
2954         if (!err && IS_DIRSYNC(dir))
2955                 ext4_handle_sync(handle);
2956
2957 out_stop:
2958         if (handle)
2959                 ext4_journal_stop(handle);
2960         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2961                 goto retry;
2962         return err;
2963 err_drop_inode:
2964         unlock_new_inode(inode);
2965         iput(inode);
2966         return err;
2967 }
2968
2969 static int ext4_link(struct dentry *old_dentry,
2970                      struct inode *dir, struct dentry *dentry)
2971 {
2972         handle_t *handle;
2973         struct inode *inode = old_dentry->d_inode;
2974         int err, retries = 0;
2975
2976         if (inode->i_nlink >= EXT4_LINK_MAX)
2977                 return -EMLINK;
2978         if (ext4_encrypted_inode(dir) &&
2979             !ext4_is_child_context_consistent_with_parent(dir, inode))
2980                 return -EPERM;
2981         dquot_initialize(dir);
2982
2983 retry:
2984         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2985                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2986                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2987         if (IS_ERR(handle))
2988                 return PTR_ERR(handle);
2989
2990         if (IS_DIRSYNC(dir))
2991                 ext4_handle_sync(handle);
2992
2993         inode->i_ctime = ext4_current_time(inode);
2994         ext4_inc_count(handle, inode);
2995         ihold(inode);
2996
2997         err = ext4_add_entry(handle, dentry, inode);
2998         if (!err) {
2999                 ext4_mark_inode_dirty(handle, inode);
3000                 /* this can happen only for tmpfile being
3001                  * linked the first time
3002                  */
3003                 if (inode->i_nlink == 1)
3004                         ext4_orphan_del(handle, inode);
3005                 d_instantiate(dentry, inode);
3006         } else {
3007                 drop_nlink(inode);
3008                 iput(inode);
3009         }
3010         ext4_journal_stop(handle);
3011         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3012                 goto retry;
3013         return err;
3014 }
3015
3016
3017 /*
3018  * Try to find buffer head where contains the parent block.
3019  * It should be the inode block if it is inlined or the 1st block
3020  * if it is a normal dir.
3021  */
3022 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3023                                         struct inode *inode,
3024                                         int *retval,
3025                                         struct ext4_dir_entry_2 **parent_de,
3026                                         int *inlined)
3027 {
3028         struct buffer_head *bh;
3029
3030         if (!ext4_has_inline_data(inode)) {
3031                 bh = ext4_read_dirblock(inode, 0, EITHER);
3032                 if (IS_ERR(bh)) {
3033                         *retval = PTR_ERR(bh);
3034                         return NULL;
3035                 }
3036                 *parent_de = ext4_next_entry(
3037                                         (struct ext4_dir_entry_2 *)bh->b_data,
3038                                         inode->i_sb->s_blocksize);
3039                 return bh;
3040         }
3041
3042         *inlined = 1;
3043         return ext4_get_first_inline_block(inode, parent_de, retval);
3044 }
3045
3046 struct ext4_renament {
3047         struct inode *dir;
3048         struct dentry *dentry;
3049         struct inode *inode;
3050         bool is_dir;
3051         int dir_nlink_delta;
3052
3053         /* entry for "dentry" */
3054         struct buffer_head *bh;
3055         struct ext4_dir_entry_2 *de;
3056         int inlined;
3057
3058         /* entry for ".." in inode if it's a directory */
3059         struct buffer_head *dir_bh;
3060         struct ext4_dir_entry_2 *parent_de;
3061         int dir_inlined;
3062 };
3063
3064 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3065 {
3066         int retval;
3067
3068         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3069                                               &retval, &ent->parent_de,
3070                                               &ent->dir_inlined);
3071         if (!ent->dir_bh)
3072                 return retval;
3073         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3074                 return -EIO;
3075         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3076         return ext4_journal_get_write_access(handle, ent->dir_bh);
3077 }
3078
3079 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3080                                   unsigned dir_ino)
3081 {
3082         int retval;
3083
3084         ent->parent_de->inode = cpu_to_le32(dir_ino);
3085         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3086         if (!ent->dir_inlined) {
3087                 if (is_dx(ent->inode)) {
3088                         retval = ext4_handle_dirty_dx_node(handle,
3089                                                            ent->inode,
3090                                                            ent->dir_bh);
3091                 } else {
3092                         retval = ext4_handle_dirty_dirent_node(handle,
3093                                                                ent->inode,
3094                                                                ent->dir_bh);
3095                 }
3096         } else {
3097                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3098         }
3099         if (retval) {
3100                 ext4_std_error(ent->dir->i_sb, retval);
3101                 return retval;
3102         }
3103         return 0;
3104 }
3105
3106 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3107                        unsigned ino, unsigned file_type)
3108 {
3109         int retval;
3110
3111         BUFFER_TRACE(ent->bh, "get write access");
3112         retval = ext4_journal_get_write_access(handle, ent->bh);
3113         if (retval)
3114                 return retval;
3115         ent->de->inode = cpu_to_le32(ino);
3116         if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3117                                       EXT4_FEATURE_INCOMPAT_FILETYPE))
3118                 ent->de->file_type = file_type;
3119         ent->dir->i_version++;
3120         ent->dir->i_ctime = ent->dir->i_mtime =
3121                 ext4_current_time(ent->dir);
3122         ext4_mark_inode_dirty(handle, ent->dir);
3123         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3124         if (!ent->inlined) {
3125                 retval = ext4_handle_dirty_dirent_node(handle,
3126                                                        ent->dir, ent->bh);
3127                 if (unlikely(retval)) {
3128                         ext4_std_error(ent->dir->i_sb, retval);
3129                         return retval;
3130                 }
3131         }
3132         brelse(ent->bh);
3133         ent->bh = NULL;
3134
3135         return 0;
3136 }
3137
3138 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3139                                   const struct qstr *d_name)
3140 {
3141         int retval = -ENOENT;
3142         struct buffer_head *bh;
3143         struct ext4_dir_entry_2 *de;
3144
3145         bh = ext4_find_entry(dir, d_name, &de, NULL);
3146         if (IS_ERR(bh))
3147                 return PTR_ERR(bh);
3148         if (bh) {
3149                 retval = ext4_delete_entry(handle, dir, de, bh);
3150                 brelse(bh);
3151         }
3152         return retval;
3153 }
3154
3155 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3156                                int force_reread)
3157 {
3158         int retval;
3159         /*
3160          * ent->de could have moved from under us during htree split, so make
3161          * sure that we are deleting the right entry.  We might also be pointing
3162          * to a stale entry in the unused part of ent->bh so just checking inum
3163          * and the name isn't enough.
3164          */
3165         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3166             ent->de->name_len != ent->dentry->d_name.len ||
3167             strncmp(ent->de->name, ent->dentry->d_name.name,
3168                     ent->de->name_len) ||
3169             force_reread) {
3170                 retval = ext4_find_delete_entry(handle, ent->dir,
3171                                                 &ent->dentry->d_name);
3172         } else {
3173                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3174                 if (retval == -ENOENT) {
3175                         retval = ext4_find_delete_entry(handle, ent->dir,
3176                                                         &ent->dentry->d_name);
3177                 }
3178         }
3179
3180         if (retval) {
3181                 ext4_warning(ent->dir->i_sb,
3182                                 "Deleting old file (%lu), %d, error=%d",
3183                                 ent->dir->i_ino, ent->dir->i_nlink, retval);
3184         }
3185 }
3186
3187 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3188 {
3189         if (ent->dir_nlink_delta) {
3190                 if (ent->dir_nlink_delta == -1)
3191                         ext4_dec_count(handle, ent->dir);
3192                 else
3193                         ext4_inc_count(handle, ent->dir);
3194                 ext4_mark_inode_dirty(handle, ent->dir);
3195         }
3196 }
3197
3198 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3199                                               int credits, handle_t **h)
3200 {
3201         struct inode *wh;
3202         handle_t *handle;
3203         int retries = 0;
3204
3205         /*
3206          * for inode block, sb block, group summaries,
3207          * and inode bitmap
3208          */
3209         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3210                     EXT4_XATTR_TRANS_BLOCKS + 4);
3211 retry:
3212         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3213                                          &ent->dentry->d_name, 0, NULL,
3214                                          EXT4_HT_DIR, credits);
3215
3216         handle = ext4_journal_current_handle();
3217         if (IS_ERR(wh)) {
3218                 if (handle)
3219                         ext4_journal_stop(handle);
3220                 if (PTR_ERR(wh) == -ENOSPC &&
3221                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3222                         goto retry;
3223         } else {
3224                 *h = handle;
3225                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3226                 wh->i_op = &ext4_special_inode_operations;
3227         }
3228         return wh;
3229 }
3230
3231 /*
3232  * Anybody can rename anything with this: the permission checks are left to the
3233  * higher-level routines.
3234  *
3235  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3236  * while new_{dentry,inode) refers to the destination dentry/inode
3237  * This comes from rename(const char *oldpath, const char *newpath)
3238  */
3239 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3240                        struct inode *new_dir, struct dentry *new_dentry,
3241                        unsigned int flags)
3242 {
3243         handle_t *handle = NULL;
3244         struct ext4_renament old = {
3245                 .dir = old_dir,
3246                 .dentry = old_dentry,
3247                 .inode = old_dentry->d_inode,
3248         };
3249         struct ext4_renament new = {
3250                 .dir = new_dir,
3251                 .dentry = new_dentry,
3252                 .inode = new_dentry->d_inode,
3253         };
3254         int force_reread;
3255         int retval;
3256         struct inode *whiteout = NULL;
3257         int credits;
3258         u8 old_file_type;
3259
3260         dquot_initialize(old.dir);
3261         dquot_initialize(new.dir);
3262
3263         /* Initialize quotas before so that eventual writes go
3264          * in separate transaction */
3265         if (new.inode)
3266                 dquot_initialize(new.inode);
3267
3268         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3269         if (IS_ERR(old.bh))
3270                 return PTR_ERR(old.bh);
3271         /*
3272          *  Check for inode number is _not_ due to possible IO errors.
3273          *  We might rmdir the source, keep it as pwd of some process
3274          *  and merrily kill the link to whatever was created under the
3275          *  same name. Goodbye sticky bit ;-<
3276          */
3277         retval = -ENOENT;
3278         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3279                 goto end_rename;
3280
3281         if ((old.dir != new.dir) &&
3282             ext4_encrypted_inode(new.dir) &&
3283             !ext4_is_child_context_consistent_with_parent(new.dir,
3284                                                           old.inode)) {
3285                 retval = -EPERM;
3286                 goto end_rename;
3287         }
3288
3289         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3290                                  &new.de, &new.inlined);
3291         if (IS_ERR(new.bh)) {
3292                 retval = PTR_ERR(new.bh);
3293                 new.bh = NULL;
3294                 goto end_rename;
3295         }
3296         if (new.bh) {
3297                 if (!new.inode) {
3298                         brelse(new.bh);
3299                         new.bh = NULL;
3300                 }
3301         }
3302         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3303                 ext4_alloc_da_blocks(old.inode);
3304
3305         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3306                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3307         if (!(flags & RENAME_WHITEOUT)) {
3308                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3309                 if (IS_ERR(handle)) {
3310                         retval = PTR_ERR(handle);
3311                         handle = NULL;
3312                         goto end_rename;
3313                 }
3314         } else {
3315                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3316                 if (IS_ERR(whiteout)) {
3317                         retval = PTR_ERR(whiteout);
3318                         whiteout = NULL;
3319                         goto end_rename;
3320                 }
3321         }
3322
3323         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3324                 ext4_handle_sync(handle);
3325
3326         if (S_ISDIR(old.inode->i_mode)) {
3327                 if (new.inode) {
3328                         retval = -ENOTEMPTY;
3329                         if (!ext4_empty_dir(new.inode))
3330                                 goto end_rename;
3331                 } else {
3332                         retval = -EMLINK;
3333                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3334                                 goto end_rename;
3335                 }
3336                 retval = ext4_rename_dir_prepare(handle, &old);
3337                 if (retval)
3338                         goto end_rename;
3339         }
3340         /*
3341          * If we're renaming a file within an inline_data dir and adding or
3342          * setting the new dirent causes a conversion from inline_data to
3343          * extents/blockmap, we need to force the dirent delete code to
3344          * re-read the directory, or else we end up trying to delete a dirent
3345          * from what is now the extent tree root (or a block map).
3346          */
3347         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3348                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3349
3350         old_file_type = old.de->file_type;
3351         if (whiteout) {
3352                 /*
3353                  * Do this before adding a new entry, so the old entry is sure
3354                  * to be still pointing to the valid old entry.
3355                  */
3356                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3357                                      EXT4_FT_CHRDEV);
3358                 if (retval)
3359                         goto end_rename;
3360                 ext4_mark_inode_dirty(handle, whiteout);
3361         }
3362         if (!new.bh) {
3363                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3364                 if (retval)
3365                         goto end_rename;
3366         } else {
3367                 retval = ext4_setent(handle, &new,
3368                                      old.inode->i_ino, old_file_type);
3369                 if (retval)
3370                         goto end_rename;
3371         }
3372         if (force_reread)
3373                 force_reread = !ext4_test_inode_flag(new.dir,
3374                                                      EXT4_INODE_INLINE_DATA);
3375
3376         /*
3377          * Like most other Unix systems, set the ctime for inodes on a
3378          * rename.
3379          */
3380         old.inode->i_ctime = ext4_current_time(old.inode);
3381         ext4_mark_inode_dirty(handle, old.inode);
3382
3383         if (!whiteout) {
3384                 /*
3385                  * ok, that's it
3386                  */
3387                 ext4_rename_delete(handle, &old, force_reread);
3388         }
3389
3390         if (new.inode) {
3391                 ext4_dec_count(handle, new.inode);
3392                 new.inode->i_ctime = ext4_current_time(new.inode);
3393         }
3394         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3395         ext4_update_dx_flag(old.dir);
3396         if (old.dir_bh) {
3397                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3398                 if (retval)
3399                         goto end_rename;
3400
3401                 ext4_dec_count(handle, old.dir);
3402                 if (new.inode) {
3403                         /* checked ext4_empty_dir above, can't have another
3404                          * parent, ext4_dec_count() won't work for many-linked
3405                          * dirs */
3406                         clear_nlink(new.inode);
3407                 } else {
3408                         ext4_inc_count(handle, new.dir);
3409                         ext4_update_dx_flag(new.dir);
3410                         ext4_mark_inode_dirty(handle, new.dir);
3411                 }
3412         }
3413         ext4_mark_inode_dirty(handle, old.dir);
3414         if (new.inode) {
3415                 ext4_mark_inode_dirty(handle, new.inode);
3416                 if (!new.inode->i_nlink)
3417                         ext4_orphan_add(handle, new.inode);
3418         }
3419         retval = 0;
3420
3421 end_rename:
3422         brelse(old.dir_bh);
3423         brelse(old.bh);
3424         brelse(new.bh);
3425         if (whiteout) {
3426                 if (retval)
3427                         drop_nlink(whiteout);
3428                 unlock_new_inode(whiteout);
3429                 iput(whiteout);
3430         }
3431         if (handle)
3432                 ext4_journal_stop(handle);
3433         return retval;
3434 }
3435
3436 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3437                              struct inode *new_dir, struct dentry *new_dentry)
3438 {
3439         handle_t *handle = NULL;
3440         struct ext4_renament old = {
3441                 .dir = old_dir,
3442                 .dentry = old_dentry,
3443                 .inode = old_dentry->d_inode,
3444         };
3445         struct ext4_renament new = {
3446                 .dir = new_dir,
3447                 .dentry = new_dentry,
3448                 .inode = new_dentry->d_inode,
3449         };
3450         u8 new_file_type;
3451         int retval;
3452
3453         dquot_initialize(old.dir);
3454         dquot_initialize(new.dir);
3455
3456         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3457                                  &old.de, &old.inlined);
3458         if (IS_ERR(old.bh))
3459                 return PTR_ERR(old.bh);
3460         /*
3461          *  Check for inode number is _not_ due to possible IO errors.
3462          *  We might rmdir the source, keep it as pwd of some process
3463          *  and merrily kill the link to whatever was created under the
3464          *  same name. Goodbye sticky bit ;-<
3465          */
3466         retval = -ENOENT;
3467         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3468                 goto end_rename;
3469
3470         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3471                                  &new.de, &new.inlined);
3472         if (IS_ERR(new.bh)) {
3473                 retval = PTR_ERR(new.bh);
3474                 new.bh = NULL;
3475                 goto end_rename;
3476         }
3477
3478         /* RENAME_EXCHANGE case: old *and* new must both exist */
3479         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3480                 goto end_rename;
3481
3482         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3483                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3484                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3485         if (IS_ERR(handle)) {
3486                 retval = PTR_ERR(handle);
3487                 handle = NULL;
3488                 goto end_rename;
3489         }
3490
3491         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3492                 ext4_handle_sync(handle);
3493
3494         if (S_ISDIR(old.inode->i_mode)) {
3495                 old.is_dir = true;
3496                 retval = ext4_rename_dir_prepare(handle, &old);
3497                 if (retval)
3498                         goto end_rename;
3499         }
3500         if (S_ISDIR(new.inode->i_mode)) {
3501                 new.is_dir = true;
3502                 retval = ext4_rename_dir_prepare(handle, &new);
3503                 if (retval)
3504                         goto end_rename;
3505         }
3506
3507         /*
3508          * Other than the special case of overwriting a directory, parents'
3509          * nlink only needs to be modified if this is a cross directory rename.
3510          */
3511         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3512                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3513                 new.dir_nlink_delta = -old.dir_nlink_delta;
3514                 retval = -EMLINK;
3515                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3516                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3517                         goto end_rename;
3518         }
3519
3520         new_file_type = new.de->file_type;
3521         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3522         if (retval)
3523                 goto end_rename;
3524
3525         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3526         if (retval)
3527                 goto end_rename;
3528
3529         /*
3530          * Like most other Unix systems, set the ctime for inodes on a
3531          * rename.
3532          */
3533         old.inode->i_ctime = ext4_current_time(old.inode);
3534         new.inode->i_ctime = ext4_current_time(new.inode);
3535         ext4_mark_inode_dirty(handle, old.inode);
3536         ext4_mark_inode_dirty(handle, new.inode);
3537
3538         if (old.dir_bh) {
3539                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3540                 if (retval)
3541                         goto end_rename;
3542         }
3543         if (new.dir_bh) {
3544                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3545                 if (retval)
3546                         goto end_rename;
3547         }
3548         ext4_update_dir_count(handle, &old);
3549         ext4_update_dir_count(handle, &new);
3550         retval = 0;
3551
3552 end_rename:
3553         brelse(old.dir_bh);
3554         brelse(new.dir_bh);
3555         brelse(old.bh);
3556         brelse(new.bh);
3557         if (handle)
3558                 ext4_journal_stop(handle);
3559         return retval;
3560 }
3561
3562 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3563                         struct inode *new_dir, struct dentry *new_dentry,
3564                         unsigned int flags)
3565 {
3566         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3567                 return -EINVAL;
3568
3569         if (flags & RENAME_EXCHANGE) {
3570                 return ext4_cross_rename(old_dir, old_dentry,
3571                                          new_dir, new_dentry);
3572         }
3573
3574         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3575 }
3576
3577 /*
3578  * directories can handle most operations...
3579  */
3580 const struct inode_operations ext4_dir_inode_operations = {
3581         .create         = ext4_create,
3582         .lookup         = ext4_lookup,
3583         .link           = ext4_link,
3584         .unlink         = ext4_unlink,
3585         .symlink        = ext4_symlink,
3586         .mkdir          = ext4_mkdir,
3587         .rmdir          = ext4_rmdir,
3588         .mknod          = ext4_mknod,
3589         .tmpfile        = ext4_tmpfile,
3590         .rename2        = ext4_rename2,
3591         .setattr        = ext4_setattr,
3592         .setxattr       = generic_setxattr,
3593         .getxattr       = generic_getxattr,
3594         .listxattr      = ext4_listxattr,
3595         .removexattr    = generic_removexattr,
3596         .get_acl        = ext4_get_acl,
3597         .set_acl        = ext4_set_acl,
3598         .fiemap         = ext4_fiemap,
3599 };
3600
3601 const struct inode_operations ext4_special_inode_operations = {
3602         .setattr        = ext4_setattr,
3603         .setxattr       = generic_setxattr,
3604         .getxattr       = generic_getxattr,
3605         .listxattr      = ext4_listxattr,
3606         .removexattr    = generic_removexattr,
3607         .get_acl        = ext4_get_acl,
3608         .set_acl        = ext4_set_acl,
3609 };