kernel/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, EXT4_GET_BLOCKS_CREATE);
  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), __func__, __LINE__)
  88
  89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  90                                                 ext4_lblk_t block,
  91                                                 dirblock_type_t type,
  92                                                 const char *func,
  93                                                 unsigned int line)
  94 {
  95         struct buffer_head *bh;
  96         struct ext4_dir_entry *dirent;
  97         int is_dx_block = 0;
  98
  99         bh = ext4_bread(NULL, inode, block, 0);
 100         if (IS_ERR(bh)) {
 101                 __ext4_warning(inode->i_sb, func, line,
 102                                "inode #%lu: lblock %lu: comm %s: "
 103                                "error %ld reading directory block",
 104                                inode->i_ino, (unsigned long)block,
 105                                current->comm, PTR_ERR(bh));
 106
 107                 return bh;
 108         }
 109         if (!bh) {
 110                 ext4_error_inode(inode, func, line, block,
 111                                  "Directory hole found");
 112                 return ERR_PTR(-EFSCORRUPTED);
 113         }
 114         dirent = (struct ext4_dir_entry *) bh->b_data;
 115         /* Determine whether or not we have an index block */
 116         if (is_dx(inode)) {
 117                 if (block == 0)
 118                         is_dx_block = 1;
 119                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 120                                                 inode->i_sb->s_blocksize) ==
 121                          inode->i_sb->s_blocksize)
 122                         is_dx_block = 1;
 123         }
 124         if (!is_dx_block && type == INDEX) {
 125                 ext4_error_inode(inode, func, line, block,
 126                        "directory leaf block found instead of index block");
 127                 return ERR_PTR(-EFSCORRUPTED);
 128         }
 129         if (!ext4_has_metadata_csum(inode->i_sb) ||
 130             buffer_verified(bh))
 131                 return bh;
 132
 133         /*
 134          * An empty leaf block can get mistaken for a index block; for
 135          * this reason, we can only check the index checksum when the
 136          * caller is sure it should be an index block.
 137          */
 138         if (is_dx_block && type == INDEX) {
 139                 if (ext4_dx_csum_verify(inode, dirent))
 140                         set_buffer_verified(bh);
 141                 else {
 142                         ext4_error_inode(inode, func, line, block,
 143                                          "Directory index failed checksum");
 144                         brelse(bh);
 145                         return ERR_PTR(-EFSBADCRC);
 146                 }
 147         }
 148         if (!is_dx_block) {
 149                 if (ext4_dirent_csum_verify(inode, dirent))
 150                         set_buffer_verified(bh);
 151                 else {
 152                         ext4_error_inode(inode, func, line, block,
 153                                          "Directory block failed checksum");
 154                         brelse(bh);
 155                         return ERR_PTR(-EFSBADCRC);
 156                 }
 157         }
 158         return bh;
 159 }
 160
 161 #ifndef assert
 162 #define assert(test) J_ASSERT(test)
 163 #endif
 164
 165 #ifdef DX_DEBUG
 166 #define dxtrace(command) command
 167 #else
 168 #define dxtrace(command)
 169 #endif
 170
 171 struct fake_dirent
 172 {
 173         __le32 inode;
 174         __le16 rec_len;
 175         u8 name_len;
 176         u8 file_type;
 177 };
 178
 179 struct dx_countlimit
 180 {
 181         __le16 limit;
 182         __le16 count;
 183 };
 184
 185 struct dx_entry
 186 {
 187         __le32 hash;
 188         __le32 block;
 189 };
 190
 191 /*
 192  * dx_root_info is laid out so that if it should somehow get overlaid by a
 193  * dirent the two low bits of the hash version will be zero.  Therefore, the
 194  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 195  */
 196
 197 struct dx_root
 198 {
 199         struct fake_dirent dot;
 200         char dot_name[4];
 201         struct fake_dirent dotdot;
 202         char dotdot_name[4];
 203         struct dx_root_info
 204         {
 205                 __le32 reserved_zero;
 206                 u8 hash_version;
 207                 u8 info_length; /* 8 */
 208                 u8 indirect_levels;
 209                 u8 unused_flags;
 210         }
 211         info;
 212         struct dx_entry entries[0];
 213 };
 214
 215 struct dx_node
 216 {
 217         struct fake_dirent fake;
 218         struct dx_entry entries[0];
 219 };
 220
 221
 222 struct dx_frame
 223 {
 224         struct buffer_head *bh;
 225         struct dx_entry *entries;
 226         struct dx_entry *at;
 227 };
 228
 229 struct dx_map_entry
 230 {
 231         u32 hash;
 232         u16 offs;
 233         u16 size;
 234 };
 235
 236 /*
 237  * This goes at the end of each htree block.
 238  */
 239 struct dx_tail {
 240         u32 dt_reserved;
 241         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 242 };
 243
 244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 246 static inline unsigned dx_get_hash(struct dx_entry *entry);
 247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 248 static unsigned dx_get_count(struct dx_entry *entries);
 249 static unsigned dx_get_limit(struct dx_entry *entries);
 250 static void dx_set_count(struct dx_entry *entries, unsigned value);
 251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 253 static unsigned dx_node_limit(struct inode *dir);
 254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 255                                  struct inode *dir,
 256                                  struct dx_hash_info *hinfo,
 257                                  struct dx_frame *frame);
 258 static void dx_release(struct dx_frame *frames);
 259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 260                        unsigned blocksize, struct dx_hash_info *hinfo,
 261                        struct dx_map_entry map[]);
 262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 264                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 266 static void dx_insert_block(struct dx_frame *frame,
 267                                         u32 hash, ext4_lblk_t block);
 268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 269                                  struct dx_frame *frame,
 270                                  struct dx_frame *frames,
 271                                  __u32 *start_hash);
 272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 273                 struct ext4_filename *fname,
 274                 struct ext4_dir_entry_2 **res_dir);
 275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 276                              struct dentry *dentry, struct inode *inode);
 277
 278 /* checksumming functions */
 279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 280                             unsigned int blocksize)
 281 {
 282         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 283         t->det_rec_len = ext4_rec_len_to_disk(
 284                         sizeof(struct ext4_dir_entry_tail), blocksize);
 285         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 286 }
 287
 288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 290                                                    struct ext4_dir_entry *de)
 291 {
 292         struct ext4_dir_entry_tail *t;
 293
 294 #ifdef PARANOID
 295         struct ext4_dir_entry *d, *top;
 296
 297         d = de;
 298         top = (struct ext4_dir_entry *)(((void *)de) +
 299                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 300                 sizeof(struct ext4_dir_entry_tail)));
 301         while (d < top && d->rec_len)
 302                 d = (struct ext4_dir_entry *)(((void *)d) +
 303                     le16_to_cpu(d->rec_len));
 304
 305         if (d != top)
 306                 return NULL;
 307
 308         t = (struct ext4_dir_entry_tail *)d;
 309 #else
 310         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 311 #endif
 312
 313         if (t->det_reserved_zero1 ||
 314             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 315             t->det_reserved_zero2 ||
 316             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 317                 return NULL;
 318
 319         return t;
 320 }
 321
 322 static __le32 ext4_dirent_csum(struct inode *inode,
 323                                struct ext4_dir_entry *dirent, int size)
 324 {
 325         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 326         struct ext4_inode_info *ei = EXT4_I(inode);
 327         __u32 csum;
 328
 329         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 330         return cpu_to_le32(csum);
 331 }
 332
 333 #define warn_no_space_for_csum(inode)                                   \
 334         __warn_no_space_for_csum((inode), __func__, __LINE__)
 335
 336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 337                                      unsigned int line)
 338 {
 339         __ext4_warning_inode(inode, func, line,
 340                 "No space for directory leaf checksum. Please run e2fsck -D.");
 341 }
 342
 343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 344 {
 345         struct ext4_dir_entry_tail *t;
 346
 347         if (!ext4_has_metadata_csum(inode->i_sb))
 348                 return 1;
 349
 350         t = get_dirent_tail(inode, dirent);
 351         if (!t) {
 352                 warn_no_space_for_csum(inode);
 353                 return 0;
 354         }
 355
 356         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 357                                                 (void *)t - (void *)dirent))
 358                 return 0;
 359
 360         return 1;
 361 }
 362
 363 static void ext4_dirent_csum_set(struct inode *inode,
 364                                  struct ext4_dir_entry *dirent)
 365 {
 366         struct ext4_dir_entry_tail *t;
 367
 368         if (!ext4_has_metadata_csum(inode->i_sb))
 369                 return;
 370
 371         t = get_dirent_tail(inode, dirent);
 372         if (!t) {
 373                 warn_no_space_for_csum(inode);
 374                 return;
 375         }
 376
 377         t->det_checksum = ext4_dirent_csum(inode, dirent,
 378                                            (void *)t - (void *)dirent);
 379 }
 380
 381 int ext4_handle_dirty_dirent_node(handle_t *handle,
 382                                   struct inode *inode,
 383                                   struct buffer_head *bh)
 384 {
 385         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 386         return ext4_handle_dirty_metadata(handle, inode, bh);
 387 }
 388
 389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 390                                                struct ext4_dir_entry *dirent,
 391                                                int *offset)
 392 {
 393         struct ext4_dir_entry *dp;
 394         struct dx_root_info *root;
 395         int count_offset;
 396
 397         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 398                 count_offset = 8;
 399         else if (le16_to_cpu(dirent->rec_len) == 12) {
 400                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 401                 if (le16_to_cpu(dp->rec_len) !=
 402                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 403                         return NULL;
 404                 root = (struct dx_root_info *)(((void *)dp + 12));
 405                 if (root->reserved_zero ||
 406                     root->info_length != sizeof(struct dx_root_info))
 407                         return NULL;
 408                 count_offset = 32;
 409         } else
 410                 return NULL;
 411
 412         if (offset)
 413                 *offset = count_offset;
 414         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 415 }
 416
 417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 418                            int count_offset, int count, struct dx_tail *t)
 419 {
 420         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 421         struct ext4_inode_info *ei = EXT4_I(inode);
 422         __u32 csum;
 423         int size;
 424         __u32 dummy_csum = 0;
 425         int offset = offsetof(struct dx_tail, dt_checksum);
 426
 427         size = count_offset + (count * sizeof(struct dx_entry));
 428         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 429         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 430         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 431
 432         return cpu_to_le32(csum);
 433 }
 434
 435 static int ext4_dx_csum_verify(struct inode *inode,
 436                                struct ext4_dir_entry *dirent)
 437 {
 438         struct dx_countlimit *c;
 439         struct dx_tail *t;
 440         int count_offset, limit, count;
 441
 442         if (!ext4_has_metadata_csum(inode->i_sb))
 443                 return 1;
 444
 445         c = get_dx_countlimit(inode, dirent, &count_offset);
 446         if (!c) {
 447                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 448                 return 1;
 449         }
 450         limit = le16_to_cpu(c->limit);
 451         count = le16_to_cpu(c->count);
 452         if (count_offset + (limit * sizeof(struct dx_entry)) >
 453             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 454                 warn_no_space_for_csum(inode);
 455                 return 1;
 456         }
 457         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 458
 459         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 460                                             count, t))
 461                 return 0;
 462         return 1;
 463 }
 464
 465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 466 {
 467         struct dx_countlimit *c;
 468         struct dx_tail *t;
 469         int count_offset, limit, count;
 470
 471         if (!ext4_has_metadata_csum(inode->i_sb))
 472                 return;
 473
 474         c = get_dx_countlimit(inode, dirent, &count_offset);
 475         if (!c) {
 476                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 477                 return;
 478         }
 479         limit = le16_to_cpu(c->limit);
 480         count = le16_to_cpu(c->count);
 481         if (count_offset + (limit * sizeof(struct dx_entry)) >
 482             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 483                 warn_no_space_for_csum(inode);
 484                 return;
 485         }
 486         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 487
 488         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 489 }
 490
 491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 492                                             struct inode *inode,
 493                                             struct buffer_head *bh)
 494 {
 495         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 496         return ext4_handle_dirty_metadata(handle, inode, bh);
 497 }
 498
 499 /*
 500  * p is at least 6 bytes before the end of page
 501  */
 502 static inline struct ext4_dir_entry_2 *
 503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 504 {
 505         return (struct ext4_dir_entry_2 *)((char *)p +
 506                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 507 }
 508
 509 /*
 510  * Future: use high four bits of block for coalesce-on-delete flags
 511  * Mask them off for now.
 512  */
 513
 514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 515 {
 516         return le32_to_cpu(entry->block) & 0x00ffffff;
 517 }
 518
 519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 520 {
 521         entry->block = cpu_to_le32(value);
 522 }
 523
 524 static inline unsigned dx_get_hash(struct dx_entry *entry)
 525 {
 526         return le32_to_cpu(entry->hash);
 527 }
 528
 529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 530 {
 531         entry->hash = cpu_to_le32(value);
 532 }
 533
 534 static inline unsigned dx_get_count(struct dx_entry *entries)
 535 {
 536         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 537 }
 538
 539 static inline unsigned dx_get_limit(struct dx_entry *entries)
 540 {
 541         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 542 }
 543
 544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 545 {
 546         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 547 }
 548
 549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 550 {
 551         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 552 }
 553
 554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 555 {
 556         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 557                 EXT4_DIR_REC_LEN(2) - infosize;
 558
 559         if (ext4_has_metadata_csum(dir->i_sb))
 560                 entry_space -= sizeof(struct dx_tail);
 561         return entry_space / sizeof(struct dx_entry);
 562 }
 563
 564 static inline unsigned dx_node_limit(struct inode *dir)
 565 {
 566         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 567
 568         if (ext4_has_metadata_csum(dir->i_sb))
 569                 entry_space -= sizeof(struct dx_tail);
 570         return entry_space / sizeof(struct dx_entry);
 571 }
 572
 573 /*
 574  * Debug
 575  */
 576 #ifdef DX_DEBUG
 577 static void dx_show_index(char * label, struct dx_entry *entries)
 578 {
 579         int i, n = dx_get_count (entries);
 580         printk(KERN_DEBUG "%s index ", label);
 581         for (i = 0; i < n; i++) {
 582                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 583                                 0, (unsigned long)dx_get_block(entries + i));
 584         }
 585         printk("\n");
 586 }
 587
 588 struct stats
 589 {
 590         unsigned names;
 591         unsigned space;
 592         unsigned bcount;
 593 };
 594
 595 static struct stats dx_show_leaf(struct inode *dir,
 596                                 struct dx_hash_info *hinfo,
 597                                 struct ext4_dir_entry_2 *de,
 598                                 int size, int show_names)
 599 {
 600         unsigned names = 0, space = 0;
 601         char *base = (char *) de;
 602         struct dx_hash_info h = *hinfo;
 603
 604         printk("names: ");
 605         while ((char *) de < base + size)
 606         {
 607                 if (de->inode)
 608                 {
 609                         if (show_names)
 610                         {
 611 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 612                                 int len;
 613                                 char *name;
 614                                 struct ext4_str fname_crypto_str
 615                                         = {.name = NULL, .len = 0};
 616                                 int res = 0;
 617
 618                                 name  = de->name;
 619                                 len = de->name_len;
 620                                 if (ext4_encrypted_inode(inode))
 621                                         res = ext4_get_encryption_info(dir);
 622                                 if (res) {
 623                                         printk(KERN_WARNING "Error setting up"
 624                                                " fname crypto: %d\n", res);
 625                                 }
 626                                 if (ctx == NULL) {
 627                                         /* Directory is not encrypted */
 628                                         ext4fs_dirhash(de->name,
 629                                                 de->name_len, &h);
 630                                         printk("%*.s:(U)%x.%u ", len,
 631                                                name, h.hash,
 632                                                (unsigned) ((char *) de
 633                                                            - base));
 634                                 } else {
 635                                         /* Directory is encrypted */
 636                                         res = ext4_fname_crypto_alloc_buffer(
 637                                                 ctx, de->name_len,
 638                                                 &fname_crypto_str);
 639                                         if (res < 0) {
 640                                                 printk(KERN_WARNING "Error "
 641                                                         "allocating crypto "
 642                                                         "buffer--skipping "
 643                                                         "crypto\n");
 644                                                 ctx = NULL;
 645                                         }
 646                                         res = ext4_fname_disk_to_usr(ctx, NULL, de,
 647                                                         &fname_crypto_str);
 648                                         if (res < 0) {
 649                                                 printk(KERN_WARNING "Error "
 650                                                         "converting filename "
 651                                                         "from disk to usr"
 652                                                         "\n");
 653                                                 name = "??";
 654                                                 len = 2;
 655                                         } else {
 656                                                 name = fname_crypto_str.name;
 657                                                 len = fname_crypto_str.len;
 658                                         }
 659                                         ext4fs_dirhash(de->name, de->name_len,
 660                                                        &h);
 661                                         printk("%*.s:(E)%x.%u ", len, name,
 662                                                h.hash, (unsigned) ((char *) de
 663                                                                    - base));
 664                                         ext4_fname_crypto_free_buffer(
 665                                                 &fname_crypto_str);
 666                                 }
 667 #else
 668                                 int len = de->name_len;
 669                                 char *name = de->name;
 670                                 ext4fs_dirhash(de->name, de->name_len, &h);
 671                                 printk("%*.s:%x.%u ", len, name, h.hash,
 672                                        (unsigned) ((char *) de - base));
 673 #endif
 674                         }
 675                         space += EXT4_DIR_REC_LEN(de->name_len);
 676                         names++;
 677                 }
 678                 de = ext4_next_entry(de, size);
 679         }
 680         printk("(%i)\n", names);
 681         return (struct stats) { names, space, 1 };
 682 }
 683
 684 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 685                              struct dx_entry *entries, int levels)
 686 {
 687         unsigned blocksize = dir->i_sb->s_blocksize;
 688         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 689         unsigned bcount = 0;
 690         struct buffer_head *bh;
 691         printk("%i indexed blocks...\n", count);
 692         for (i = 0; i < count; i++, entries++)
 693         {
 694                 ext4_lblk_t block = dx_get_block(entries);
 695                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 696                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 697                 struct stats stats;
 698                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 699                 bh = ext4_bread(NULL,dir, block, 0);
 700                 if (!bh || IS_ERR(bh))
 701                         continue;
 702                 stats = levels?
 703                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 704                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 705                         bh->b_data, blocksize, 0);
 706                 names += stats.names;
 707                 space += stats.space;
 708                 bcount += stats.bcount;
 709                 brelse(bh);
 710         }
 711         if (bcount)
 712                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 713                        levels ? "" : "   ", names, space/bcount,
 714                        (space/bcount)*100/blocksize);
 715         return (struct stats) { names, space, bcount};
 716 }
 717 #endif /* DX_DEBUG */
 718
 719 /*
 720  * Probe for a directory leaf block to search.
 721  *
 722  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 723  * error in the directory index, and the caller should fall back to
 724  * searching the directory normally.  The callers of dx_probe **MUST**
 725  * check for this error code, and make sure it never gets reflected
 726  * back to userspace.
 727  */
 728 static struct dx_frame *
 729 dx_probe(struct ext4_filename *fname, struct inode *dir,
 730          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 731 {
 732         unsigned count, indirect;
 733         struct dx_entry *at, *entries, *p, *q, *m;
 734         struct dx_root *root;
 735         struct dx_frame *frame = frame_in;
 736         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 737         u32 hash;
 738
 739         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 740         if (IS_ERR(frame->bh))
 741                 return (struct dx_frame *) frame->bh;
 742
 743         root = (struct dx_root *) frame->bh->b_data;
 744         if (root->info.hash_version != DX_HASH_TEA &&
 745             root->info.hash_version != DX_HASH_HALF_MD4 &&
 746             root->info.hash_version != DX_HASH_LEGACY) {
 747                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 748                                    root->info.hash_version);
 749                 goto fail;
 750         }
 751         if (fname)
 752                 hinfo = &fname->hinfo;
 753         hinfo->hash_version = root->info.hash_version;
 754         if (hinfo->hash_version <= DX_HASH_TEA)
 755                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 756         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 757         if (fname && fname_name(fname))
 758                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 759         hash = hinfo->hash;
 760
 761         if (root->info.unused_flags & 1) {
 762                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 763                                    root->info.unused_flags);
 764                 goto fail;
 765         }
 766
 767         indirect = root->info.indirect_levels;
 768         if (indirect > 1) {
 769                 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
 770                                    root->info.indirect_levels);
 771                 goto fail;
 772         }
 773
 774         entries = (struct dx_entry *)(((char *)&root->info) +
 775                                       root->info.info_length);
 776
 777         if (dx_get_limit(entries) != dx_root_limit(dir,
 778                                                    root->info.info_length)) {
 779                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 780                                    dx_get_limit(entries),
 781                                    dx_root_limit(dir, root->info.info_length));
 782                 goto fail;
 783         }
 784
 785         dxtrace(printk("Look up %x", hash));
 786         while (1) {
 787                 count = dx_get_count(entries);
 788                 if (!count || count > dx_get_limit(entries)) {
 789                         ext4_warning_inode(dir,
 790                                            "dx entry: count %u beyond limit %u",
 791                                            count, dx_get_limit(entries));
 792                         goto fail;
 793                 }
 794
 795                 p = entries + 1;
 796                 q = entries + count - 1;
 797                 while (p <= q) {
 798                         m = p + (q - p) / 2;
 799                         dxtrace(printk("."));
 800                         if (dx_get_hash(m) > hash)
 801                                 q = m - 1;
 802                         else
 803                                 p = m + 1;
 804                 }
 805
 806                 if (0) { // linear search cross check
 807                         unsigned n = count - 1;
 808                         at = entries;
 809                         while (n--)
 810                         {
 811                                 dxtrace(printk(","));
 812                                 if (dx_get_hash(++at) > hash)
 813                                 {
 814                                         at--;
 815                                         break;
 816                                 }
 817                         }
 818                         assert (at == p - 1);
 819                 }
 820
 821                 at = p - 1;
 822                 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
 823                                dx_get_block(at)));
 824                 frame->entries = entries;
 825                 frame->at = at;
 826                 if (!indirect--)
 827                         return frame;
 828                 frame++;
 829                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 830                 if (IS_ERR(frame->bh)) {
 831                         ret_err = (struct dx_frame *) frame->bh;
 832                         frame->bh = NULL;
 833                         goto fail;
 834                 }
 835                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 836
 837                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 838                         ext4_warning_inode(dir,
 839                                 "dx entry: limit %u != node limit %u",
 840                                 dx_get_limit(entries), dx_node_limit(dir));
 841                         goto fail;
 842                 }
 843         }
 844 fail:
 845         while (frame >= frame_in) {
 846                 brelse(frame->bh);
 847                 frame--;
 848         }
 849
 850         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 851                 ext4_warning_inode(dir,
 852                         "Corrupt directory, running e2fsck is recommended");
 853         return ret_err;
 854 }
 855
 856 static void dx_release(struct dx_frame *frames)
 857 {
 858         if (frames[0].bh == NULL)
 859                 return;
 860
 861         if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
 862                 brelse(frames[1].bh);
 863         brelse(frames[0].bh);
 864 }
 865
 866 /*
 867  * This function increments the frame pointer to search the next leaf
 868  * block, and reads in the necessary intervening nodes if the search
 869  * should be necessary.  Whether or not the search is necessary is
 870  * controlled by the hash parameter.  If the hash value is even, then
 871  * the search is only continued if the next block starts with that
 872  * hash value.  This is used if we are searching for a specific file.
 873  *
 874  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 875  *
 876  * This function returns 1 if the caller should continue to search,
 877  * or 0 if it should not.  If there is an error reading one of the
 878  * index blocks, it will a negative error code.
 879  *
 880  * If start_hash is non-null, it will be filled in with the starting
 881  * hash of the next page.
 882  */
 883 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 884                                  struct dx_frame *frame,
 885                                  struct dx_frame *frames,
 886                                  __u32 *start_hash)
 887 {
 888         struct dx_frame *p;
 889         struct buffer_head *bh;
 890         int num_frames = 0;
 891         __u32 bhash;
 892
 893         p = frame;
 894         /*
 895          * Find the next leaf page by incrementing the frame pointer.
 896          * If we run out of entries in the interior node, loop around and
 897          * increment pointer in the parent node.  When we break out of
 898          * this loop, num_frames indicates the number of interior
 899          * nodes need to be read.
 900          */
 901         while (1) {
 902                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 903                         break;
 904                 if (p == frames)
 905                         return 0;
 906                 num_frames++;
 907                 p--;
 908         }
 909
 910         /*
 911          * If the hash is 1, then continue only if the next page has a
 912          * continuation hash of any value.  This is used for readdir
 913          * handling.  Otherwise, check to see if the hash matches the
 914          * desired contiuation hash.  If it doesn't, return since
 915          * there's no point to read in the successive index pages.
 916          */
 917         bhash = dx_get_hash(p->at);
 918         if (start_hash)
 919                 *start_hash = bhash;
 920         if ((hash & 1) == 0) {
 921                 if ((bhash & ~1) != hash)
 922                         return 0;
 923         }
 924         /*
 925          * If the hash is HASH_NB_ALWAYS, we always go to the next
 926          * block so no check is necessary
 927          */
 928         while (num_frames--) {
 929                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 930                 if (IS_ERR(bh))
 931                         return PTR_ERR(bh);
 932                 p++;
 933                 brelse(p->bh);
 934                 p->bh = bh;
 935                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 936         }
 937         return 1;
 938 }
 939
 940
 941 /*
 942  * This function fills a red-black tree with information from a
 943  * directory block.  It returns the number directory entries loaded
 944  * into the tree.  If there is an error it is returned in err.
 945  */
 946 static int htree_dirblock_to_tree(struct file *dir_file,
 947                                   struct inode *dir, ext4_lblk_t block,
 948                                   struct dx_hash_info *hinfo,
 949                                   __u32 start_hash, __u32 start_minor_hash)
 950 {
 951         struct buffer_head *bh;
 952         struct ext4_dir_entry_2 *de, *top;
 953         int err = 0, count = 0;
 954         struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
 955
 956         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 957                                                         (unsigned long)block));
 958         bh = ext4_read_dirblock(dir, block, DIRENT);
 959         if (IS_ERR(bh))
 960                 return PTR_ERR(bh);
 961
 962         de = (struct ext4_dir_entry_2 *) bh->b_data;
 963         top = (struct ext4_dir_entry_2 *) ((char *) de +
 964                                            dir->i_sb->s_blocksize -
 965                                            EXT4_DIR_REC_LEN(0));
 966 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 967         /* Check if the directory is encrypted */
 968         if (ext4_encrypted_inode(dir)) {
 969                 err = ext4_get_encryption_info(dir);
 970                 if (err < 0) {
 971                         brelse(bh);
 972                         return err;
 973                 }
 974                 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
 975                                                      &fname_crypto_str);
 976                 if (err < 0) {
 977                         brelse(bh);
 978                         return err;
 979                 }
 980         }
 981 #endif
 982         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 983                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 984                                 bh->b_data, bh->b_size,
 985                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 986                                          + ((char *)de - bh->b_data))) {
 987                         /* silently ignore the rest of the block */
 988                         break;
 989                 }
 990                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 991                 if ((hinfo->hash < start_hash) ||
 992                     ((hinfo->hash == start_hash) &&
 993                      (hinfo->minor_hash < start_minor_hash)))
 994                         continue;
 995                 if (de->inode == 0)
 996                         continue;
 997                 if (!ext4_encrypted_inode(dir)) {
 998                         tmp_str.name = de->name;
 999                         tmp_str.len = de->name_len;
1000                         err = ext4_htree_store_dirent(dir_file,
1001                                    hinfo->hash, hinfo->minor_hash, de,
1002                                    &tmp_str);
1003                 } else {
1004                         int save_len = fname_crypto_str.len;
1005
1006                         /* Directory is encrypted */
1007                         err = ext4_fname_disk_to_usr(dir, hinfo, de,
1008                                                      &fname_crypto_str);
1009                         if (err < 0) {
1010                                 count = err;
1011                                 goto errout;
1012                         }
1013                         err = ext4_htree_store_dirent(dir_file,
1014                                    hinfo->hash, hinfo->minor_hash, de,
1015                                         &fname_crypto_str);
1016                         fname_crypto_str.len = save_len;
1017                 }
1018                 if (err != 0) {
1019                         count = err;
1020                         goto errout;
1021                 }
1022                 count++;
1023         }
1024 errout:
1025         brelse(bh);
1026 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1027         ext4_fname_crypto_free_buffer(&fname_crypto_str);
1028 #endif
1029         return count;
1030 }
1031
1032
1033 /*
1034  * This function fills a red-black tree with information from a
1035  * directory.  We start scanning the directory in hash order, starting
1036  * at start_hash and start_minor_hash.
1037  *
1038  * This function returns the number of entries inserted into the tree,
1039  * or a negative error code.
1040  */
1041 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1042                          __u32 start_minor_hash, __u32 *next_hash)
1043 {
1044         struct dx_hash_info hinfo;
1045         struct ext4_dir_entry_2 *de;
1046         struct dx_frame frames[2], *frame;
1047         struct inode *dir;
1048         ext4_lblk_t block;
1049         int count = 0;
1050         int ret, err;
1051         __u32 hashval;
1052         struct ext4_str tmp_str;
1053
1054         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1055                        start_hash, start_minor_hash));
1056         dir = file_inode(dir_file);
1057         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1058                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1059                 if (hinfo.hash_version <= DX_HASH_TEA)
1060                         hinfo.hash_version +=
1061                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1062                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1063                 if (ext4_has_inline_data(dir)) {
1064                         int has_inline_data = 1;
1065                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1066                                                         &hinfo, start_hash,
1067                                                         start_minor_hash,
1068                                                         &has_inline_data);
1069                         if (has_inline_data) {
1070                                 *next_hash = ~0;
1071                                 return count;
1072                         }
1073                 }
1074                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1075                                                start_hash, start_minor_hash);
1076                 *next_hash = ~0;
1077                 return count;
1078         }
1079         hinfo.hash = start_hash;
1080         hinfo.minor_hash = 0;
1081         frame = dx_probe(NULL, dir, &hinfo, frames);
1082         if (IS_ERR(frame))
1083                 return PTR_ERR(frame);
1084
1085         /* Add '.' and '..' from the htree header */
1086         if (!start_hash && !start_minor_hash) {
1087                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1088                 tmp_str.name = de->name;
1089                 tmp_str.len = de->name_len;
1090                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1091                                               de, &tmp_str);
1092                 if (err != 0)
1093                         goto errout;
1094                 count++;
1095         }
1096         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1097                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1098                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1099                 tmp_str.name = de->name;
1100                 tmp_str.len = de->name_len;
1101                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1102                                               de, &tmp_str);
1103                 if (err != 0)
1104                         goto errout;
1105                 count++;
1106         }
1107
1108         while (1) {
1109                 block = dx_get_block(frame->at);
1110                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1111                                              start_hash, start_minor_hash);
1112                 if (ret < 0) {
1113                         err = ret;
1114                         goto errout;
1115                 }
1116                 count += ret;
1117                 hashval = ~0;
1118                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1119                                             frame, frames, &hashval);
1120                 *next_hash = hashval;
1121                 if (ret < 0) {
1122                         err = ret;
1123                         goto errout;
1124                 }
1125                 /*
1126                  * Stop if:  (a) there are no more entries, or
1127                  * (b) we have inserted at least one entry and the
1128                  * next hash value is not a continuation
1129                  */
1130                 if ((ret == 0) ||
1131                     (count && ((hashval & 1) == 0)))
1132                         break;
1133         }
1134         dx_release(frames);
1135         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1136                        "next hash: %x\n", count, *next_hash));
1137         return count;
1138 errout:
1139         dx_release(frames);
1140         return (err);
1141 }
1142
1143 static inline int search_dirblock(struct buffer_head *bh,
1144                                   struct inode *dir,
1145                                   struct ext4_filename *fname,
1146                                   const struct qstr *d_name,
1147                                   unsigned int offset,
1148                                   struct ext4_dir_entry_2 **res_dir)
1149 {
1150         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1151                                fname, d_name, offset, res_dir);
1152 }
1153
1154 /*
1155  * Directory block splitting, compacting
1156  */
1157
1158 /*
1159  * Create map of hash values, offsets, and sizes, stored at end of block.
1160  * Returns number of entries mapped.
1161  */
1162 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1163                        unsigned blocksize, struct dx_hash_info *hinfo,
1164                        struct dx_map_entry *map_tail)
1165 {
1166         int count = 0;
1167         char *base = (char *) de;
1168         struct dx_hash_info h = *hinfo;
1169
1170         while ((char *) de < base + blocksize) {
1171                 if (de->name_len && de->inode) {
1172                         ext4fs_dirhash(de->name, de->name_len, &h);
1173                         map_tail--;
1174                         map_tail->hash = h.hash;
1175                         map_tail->offs = ((char *) de - base)>>2;
1176                         map_tail->size = le16_to_cpu(de->rec_len);
1177                         count++;
1178                         cond_resched();
1179                 }
1180                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1181                 de = ext4_next_entry(de, blocksize);
1182         }
1183         return count;
1184 }
1185
1186 /* Sort map by hash value */
1187 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1188 {
1189         struct dx_map_entry *p, *q, *top = map + count - 1;
1190         int more;
1191         /* Combsort until bubble sort doesn't suck */
1192         while (count > 2) {
1193                 count = count*10/13;
1194                 if (count - 9 < 2) /* 9, 10 -> 11 */
1195                         count = 11;
1196                 for (p = top, q = p - count; q >= map; p--, q--)
1197                         if (p->hash < q->hash)
1198                                 swap(*p, *q);
1199         }
1200         /* Garden variety bubble sort */
1201         do {
1202                 more = 0;
1203                 q = top;
1204                 while (q-- > map) {
1205                         if (q[1].hash >= q[0].hash)
1206                                 continue;
1207                         swap(*(q+1), *q);
1208                         more = 1;
1209                 }
1210         } while(more);
1211 }
1212
1213 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1214 {
1215         struct dx_entry *entries = frame->entries;
1216         struct dx_entry *old = frame->at, *new = old + 1;
1217         int count = dx_get_count(entries);
1218
1219         assert(count < dx_get_limit(entries));
1220         assert(old < entries + count);
1221         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1222         dx_set_hash(new, hash);
1223         dx_set_block(new, block);
1224         dx_set_count(entries, count + 1);
1225 }
1226
1227 /*
1228  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1229  *
1230  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1231  * `de != NULL' is guaranteed by caller.
1232  */
1233 static inline int ext4_match(struct ext4_filename *fname,
1234                              struct ext4_dir_entry_2 *de)
1235 {
1236         const void *name = fname_name(fname);
1237         u32 len = fname_len(fname);
1238
1239         if (!de->inode)
1240                 return 0;
1241
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243         if (unlikely(!name)) {
1244                 if (fname->usr_fname->name[0] == '_') {
1245                         int ret;
1246                         if (de->name_len < 16)
1247                                 return 0;
1248                         ret = memcmp(de->name + de->name_len - 16,
1249                                      fname->crypto_buf.name + 8, 16);
1250                         return (ret == 0) ? 1 : 0;
1251                 }
1252                 name = fname->crypto_buf.name;
1253                 len = fname->crypto_buf.len;
1254         }
1255 #endif
1256         if (de->name_len != len)
1257                 return 0;
1258         return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1259 }
1260
1261 /*
1262  * Returns 0 if not found, -1 on failure, and 1 on success
1263  */
1264 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1265                     struct inode *dir, struct ext4_filename *fname,
1266                     const struct qstr *d_name,
1267                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1268 {
1269         struct ext4_dir_entry_2 * de;
1270         char * dlimit;
1271         int de_len;
1272         int res;
1273
1274         de = (struct ext4_dir_entry_2 *)search_buf;
1275         dlimit = search_buf + buf_size;
1276         while ((char *) de < dlimit) {
1277                 /* this code is executed quadratically often */
1278                 /* do minimal checking `by hand' */
1279                 if ((char *) de + de->name_len <= dlimit) {
1280                         res = ext4_match(fname, de);
1281                         if (res < 0) {
1282                                 res = -1;
1283                                 goto return_result;
1284                         }
1285                         if (res > 0) {
1286                                 /* found a match - just to be sure, do
1287                                  * a full check */
1288                                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1289                                                 bh->b_data,
1290                                                  bh->b_size, offset)) {
1291                                         res = -1;
1292                                         goto return_result;
1293                                 }
1294                                 *res_dir = de;
1295                                 res = 1;
1296                                 goto return_result;
1297                         }
1298
1299                 }
1300                 /* prevent looping on a bad block */
1301                 de_len = ext4_rec_len_from_disk(de->rec_len,
1302                                                 dir->i_sb->s_blocksize);
1303                 if (de_len <= 0) {
1304                         res = -1;
1305                         goto return_result;
1306                 }
1307                 offset += de_len;
1308                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1309         }
1310
1311         res = 0;
1312 return_result:
1313         return res;
1314 }
1315
1316 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1317                                struct ext4_dir_entry *de)
1318 {
1319         struct super_block *sb = dir->i_sb;
1320
1321         if (!is_dx(dir))
1322                 return 0;
1323         if (block == 0)
1324                 return 1;
1325         if (de->inode == 0 &&
1326             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1327                         sb->s_blocksize)
1328                 return 1;
1329         return 0;
1330 }
1331
1332 /*
1333  *      ext4_find_entry()
1334  *
1335  * finds an entry in the specified directory with the wanted name. It
1336  * returns the cache buffer in which the entry was found, and the entry
1337  * itself (as a parameter - res_dir). It does NOT read the inode of the
1338  * entry - you'll have to do that yourself if you want to.
1339  *
1340  * The returned buffer_head has ->b_count elevated.  The caller is expected
1341  * to brelse() it when appropriate.
1342  */
1343 static struct buffer_head * ext4_find_entry (struct inode *dir,
1344                                         const struct qstr *d_name,
1345                                         struct ext4_dir_entry_2 **res_dir,
1346                                         int *inlined)
1347 {
1348         struct super_block *sb;
1349         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1350         struct buffer_head *bh, *ret = NULL;
1351         ext4_lblk_t start, block, b;
1352         const u8 *name = d_name->name;
1353         int ra_max = 0;         /* Number of bh's in the readahead
1354                                    buffer, bh_use[] */
1355         int ra_ptr = 0;         /* Current index into readahead
1356                                    buffer */
1357         int num = 0;
1358         ext4_lblk_t  nblocks;
1359         int i, namelen, retval;
1360         struct ext4_filename fname;
1361
1362         *res_dir = NULL;
1363         sb = dir->i_sb;
1364         namelen = d_name->len;
1365         if (namelen > EXT4_NAME_LEN)
1366                 return NULL;
1367
1368         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1369         if (retval)
1370                 return ERR_PTR(retval);
1371
1372         if (ext4_has_inline_data(dir)) {
1373                 int has_inline_data = 1;
1374                 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1375                                              &has_inline_data);
1376                 if (has_inline_data) {
1377                         if (inlined)
1378                                 *inlined = 1;
1379                         goto cleanup_and_exit;
1380                 }
1381         }
1382
1383         if ((namelen <= 2) && (name[0] == '.') &&
1384             (name[1] == '.' || name[1] == '\0')) {
1385                 /*
1386                  * "." or ".." will only be in the first block
1387                  * NFS may look up ".."; "." should be handled by the VFS
1388                  */
1389                 block = start = 0;
1390                 nblocks = 1;
1391                 goto restart;
1392         }
1393         if (is_dx(dir)) {
1394                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1395                 /*
1396                  * On success, or if the error was file not found,
1397                  * return.  Otherwise, fall back to doing a search the
1398                  * old fashioned way.
1399                  */
1400                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1401                         goto cleanup_and_exit;
1402                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1403                                "falling back\n"));
1404         }
1405         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1406         start = EXT4_I(dir)->i_dir_start_lookup;
1407         if (start >= nblocks)
1408                 start = 0;
1409         block = start;
1410 restart:
1411         do {
1412                 /*
1413                  * We deal with the read-ahead logic here.
1414                  */
1415                 if (ra_ptr >= ra_max) {
1416                         /* Refill the readahead buffer */
1417                         ra_ptr = 0;
1418                         b = block;
1419                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1420                                 /*
1421                                  * Terminate if we reach the end of the
1422                                  * directory and must wrap, or if our
1423                                  * search has finished at this block.
1424                                  */
1425                                 if (b >= nblocks || (num && block == start)) {
1426                                         bh_use[ra_max] = NULL;
1427                                         break;
1428                                 }
1429                                 num++;
1430                                 bh = ext4_getblk(NULL, dir, b++, 0);
1431                                 if (IS_ERR(bh)) {
1432                                         if (ra_max == 0) {
1433                                                 ret = bh;
1434                                                 goto cleanup_and_exit;
1435                                         }
1436                                         break;
1437                                 }
1438                                 bh_use[ra_max] = bh;
1439                                 if (bh)
1440                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1441                                                     1, &bh);
1442                         }
1443                 }
1444                 if ((bh = bh_use[ra_ptr++]) == NULL)
1445                         goto next;
1446                 wait_on_buffer(bh);
1447                 if (!buffer_uptodate(bh)) {
1448                         /* read error, skip block & hope for the best */
1449                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1450                                          (unsigned long) block);
1451                         brelse(bh);
1452                         goto next;
1453                 }
1454                 if (!buffer_verified(bh) &&
1455                     !is_dx_internal_node(dir, block,
1456                                          (struct ext4_dir_entry *)bh->b_data) &&
1457                     !ext4_dirent_csum_verify(dir,
1458                                 (struct ext4_dir_entry *)bh->b_data)) {
1459                         EXT4_ERROR_INODE(dir, "checksumming directory "
1460                                          "block %lu", (unsigned long)block);
1461                         brelse(bh);
1462                         goto next;
1463                 }
1464                 set_buffer_verified(bh);
1465                 i = search_dirblock(bh, dir, &fname, d_name,
1466                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1467                 if (i == 1) {
1468                         EXT4_I(dir)->i_dir_start_lookup = block;
1469                         ret = bh;
1470                         goto cleanup_and_exit;
1471                 } else {
1472                         brelse(bh);
1473                         if (i < 0)
1474                                 goto cleanup_and_exit;
1475                 }
1476         next:
1477                 if (++block >= nblocks)
1478                         block = 0;
1479         } while (block != start);
1480
1481         /*
1482          * If the directory has grown while we were searching, then
1483          * search the last part of the directory before giving up.
1484          */
1485         block = nblocks;
1486         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1487         if (block < nblocks) {
1488                 start = 0;
1489                 goto restart;
1490         }
1491
1492 cleanup_and_exit:
1493         /* Clean up the read-ahead blocks */
1494         for (; ra_ptr < ra_max; ra_ptr++)
1495                 brelse(bh_use[ra_ptr]);
1496         ext4_fname_free_filename(&fname);
1497         return ret;
1498 }
1499
1500 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1501                         struct ext4_filename *fname,
1502                         struct ext4_dir_entry_2 **res_dir)
1503 {
1504         struct super_block * sb = dir->i_sb;
1505         struct dx_frame frames[2], *frame;
1506         const struct qstr *d_name = fname->usr_fname;
1507         struct buffer_head *bh;
1508         ext4_lblk_t block;
1509         int retval;
1510
1511 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1512         *res_dir = NULL;
1513 #endif
1514         frame = dx_probe(fname, dir, NULL, frames);
1515         if (IS_ERR(frame))
1516                 return (struct buffer_head *) frame;
1517         do {
1518                 block = dx_get_block(frame->at);
1519                 bh = ext4_read_dirblock(dir, block, DIRENT);
1520                 if (IS_ERR(bh))
1521                         goto errout;
1522
1523                 retval = search_dirblock(bh, dir, fname, d_name,
1524                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1525                                          res_dir);
1526                 if (retval == 1)
1527                         goto success;
1528                 brelse(bh);
1529                 if (retval == -1) {
1530                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1531                         goto errout;
1532                 }
1533
1534                 /* Check to see if we should continue to search */
1535                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1536                                                frames, NULL);
1537                 if (retval < 0) {
1538                         ext4_warning_inode(dir,
1539                                 "error %d reading directory index block",
1540                                 retval);
1541                         bh = ERR_PTR(retval);
1542                         goto errout;
1543                 }
1544         } while (retval == 1);
1545
1546         bh = NULL;
1547 errout:
1548         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1549 success:
1550         dx_release(frames);
1551         return bh;
1552 }
1553
1554 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1555 {
1556         struct inode *inode;
1557         struct ext4_dir_entry_2 *de;
1558         struct buffer_head *bh;
1559
1560         if (dentry->d_name.len > EXT4_NAME_LEN)
1561                 return ERR_PTR(-ENAMETOOLONG);
1562
1563         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1564         if (IS_ERR(bh))
1565                 return (struct dentry *) bh;
1566         inode = NULL;
1567         if (bh) {
1568                 __u32 ino = le32_to_cpu(de->inode);
1569                 brelse(bh);
1570                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1571                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1572                         return ERR_PTR(-EFSCORRUPTED);
1573                 }
1574                 if (unlikely(ino == dir->i_ino)) {
1575                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1576                                          dentry);
1577                         return ERR_PTR(-EFSCORRUPTED);
1578                 }
1579                 inode = ext4_iget_normal(dir->i_sb, ino);
1580                 if (inode == ERR_PTR(-ESTALE)) {
1581                         EXT4_ERROR_INODE(dir,
1582                                          "deleted inode referenced: %u",
1583                                          ino);
1584                         return ERR_PTR(-EFSCORRUPTED);
1585                 }
1586                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1587                     (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1588                      S_ISLNK(inode->i_mode)) &&
1589                     !ext4_is_child_context_consistent_with_parent(dir,
1590                                                                   inode)) {
1591                         iput(inode);
1592                         ext4_warning(inode->i_sb,
1593                                      "Inconsistent encryption contexts: %lu/%lu\n",
1594                                      (unsigned long) dir->i_ino,
1595                                      (unsigned long) inode->i_ino);
1596                         return ERR_PTR(-EPERM);
1597                 }
1598         }
1599         return d_splice_alias(inode, dentry);
1600 }
1601
1602
1603 struct dentry *ext4_get_parent(struct dentry *child)
1604 {
1605         __u32 ino;
1606         static const struct qstr dotdot = QSTR_INIT("..", 2);
1607         struct ext4_dir_entry_2 * de;
1608         struct buffer_head *bh;
1609
1610         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1611         if (IS_ERR(bh))
1612                 return (struct dentry *) bh;
1613         if (!bh)
1614                 return ERR_PTR(-ENOENT);
1615         ino = le32_to_cpu(de->inode);
1616         brelse(bh);
1617
1618         if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1619                 EXT4_ERROR_INODE(d_inode(child),
1620                                  "bad parent inode number: %u", ino);
1621                 return ERR_PTR(-EFSCORRUPTED);
1622         }
1623
1624         return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1625 }
1626
1627 /*
1628  * Move count entries from end of map between two memory locations.
1629  * Returns pointer to last entry moved.
1630  */
1631 static struct ext4_dir_entry_2 *
1632 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1633                 unsigned blocksize)
1634 {
1635         unsigned rec_len = 0;
1636
1637         while (count--) {
1638                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1639                                                 (from + (map->offs<<2));
1640                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1641                 memcpy (to, de, rec_len);
1642                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1643                                 ext4_rec_len_to_disk(rec_len, blocksize);
1644                 de->inode = 0;
1645                 map++;
1646                 to += rec_len;
1647         }
1648         return (struct ext4_dir_entry_2 *) (to - rec_len);
1649 }
1650
1651 /*
1652  * Compact each dir entry in the range to the minimal rec_len.
1653  * Returns pointer to last entry in range.
1654  */
1655 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1656 {
1657         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1658         unsigned rec_len = 0;
1659
1660         prev = to = de;
1661         while ((char*)de < base + blocksize) {
1662                 next = ext4_next_entry(de, blocksize);
1663                 if (de->inode && de->name_len) {
1664                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1665                         if (de > to)
1666                                 memmove(to, de, rec_len);
1667                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1668                         prev = to;
1669                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1670                 }
1671                 de = next;
1672         }
1673         return prev;
1674 }
1675
1676 /*
1677  * Split a full leaf block to make room for a new dir entry.
1678  * Allocate a new block, and move entries so that they are approx. equally full.
1679  * Returns pointer to de in block into which the new entry will be inserted.
1680  */
1681 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1682                         struct buffer_head **bh,struct dx_frame *frame,
1683                         struct dx_hash_info *hinfo)
1684 {
1685         unsigned blocksize = dir->i_sb->s_blocksize;
1686         unsigned count, continued;
1687         struct buffer_head *bh2;
1688         ext4_lblk_t newblock;
1689         u32 hash2;
1690         struct dx_map_entry *map;
1691         char *data1 = (*bh)->b_data, *data2;
1692         unsigned split, move, size;
1693         struct ext4_dir_entry_2 *de = NULL, *de2;
1694         struct ext4_dir_entry_tail *t;
1695         int     csum_size = 0;
1696         int     err = 0, i;
1697
1698         if (ext4_has_metadata_csum(dir->i_sb))
1699                 csum_size = sizeof(struct ext4_dir_entry_tail);
1700
1701         bh2 = ext4_append(handle, dir, &newblock);
1702         if (IS_ERR(bh2)) {
1703                 brelse(*bh);
1704                 *bh = NULL;
1705                 return (struct ext4_dir_entry_2 *) bh2;
1706         }
1707
1708         BUFFER_TRACE(*bh, "get_write_access");
1709         err = ext4_journal_get_write_access(handle, *bh);
1710         if (err)
1711                 goto journal_error;
1712
1713         BUFFER_TRACE(frame->bh, "get_write_access");
1714         err = ext4_journal_get_write_access(handle, frame->bh);
1715         if (err)
1716                 goto journal_error;
1717
1718         data2 = bh2->b_data;
1719
1720         /* create map in the end of data2 block */
1721         map = (struct dx_map_entry *) (data2 + blocksize);
1722         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1723                              blocksize, hinfo, map);
1724         map -= count;
1725         dx_sort_map(map, count);
1726         /* Split the existing block in the middle, size-wise */
1727         size = 0;
1728         move = 0;
1729         for (i = count-1; i >= 0; i--) {
1730                 /* is more than half of this entry in 2nd half of the block? */
1731                 if (size + map[i].size/2 > blocksize/2)
1732                         break;
1733                 size += map[i].size;
1734                 move++;
1735         }
1736         /* map index at which we will split */
1737         split = count - move;
1738         hash2 = map[split].hash;
1739         continued = hash2 == map[split - 1].hash;
1740         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1741                         (unsigned long)dx_get_block(frame->at),
1742                                         hash2, split, count-split));
1743
1744         /* Fancy dance to stay within two buffers */
1745         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1746                               blocksize);
1747         de = dx_pack_dirents(data1, blocksize);
1748         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1749                                            (char *) de,
1750                                            blocksize);
1751         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1752                                             (char *) de2,
1753                                             blocksize);
1754         if (csum_size) {
1755                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1756                 initialize_dirent_tail(t, blocksize);
1757
1758                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1759                 initialize_dirent_tail(t, blocksize);
1760         }
1761
1762         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1763                         blocksize, 1));
1764         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1765                         blocksize, 1));
1766
1767         /* Which block gets the new entry? */
1768         if (hinfo->hash >= hash2) {
1769                 swap(*bh, bh2);
1770                 de = de2;
1771         }
1772         dx_insert_block(frame, hash2 + continued, newblock);
1773         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1774         if (err)
1775                 goto journal_error;
1776         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1777         if (err)
1778                 goto journal_error;
1779         brelse(bh2);
1780         dxtrace(dx_show_index("frame", frame->entries));
1781         return de;
1782
1783 journal_error:
1784         brelse(*bh);
1785         brelse(bh2);
1786         *bh = NULL;
1787         ext4_std_error(dir->i_sb, err);
1788         return ERR_PTR(err);
1789 }
1790
1791 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1792                       struct buffer_head *bh,
1793                       void *buf, int buf_size,
1794                       struct ext4_filename *fname,
1795                       struct ext4_dir_entry_2 **dest_de)
1796 {
1797         struct ext4_dir_entry_2 *de;
1798         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1799         int nlen, rlen;
1800         unsigned int offset = 0;
1801         char *top;
1802         int res;
1803
1804         de = (struct ext4_dir_entry_2 *)buf;
1805         top = buf + buf_size - reclen;
1806         while ((char *) de <= top) {
1807                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1808                                          buf, buf_size, offset)) {
1809                         res = -EFSCORRUPTED;
1810                         goto return_result;
1811                 }
1812                 /* Provide crypto context and crypto buffer to ext4 match */
1813                 res = ext4_match(fname, de);
1814                 if (res < 0)
1815                         goto return_result;
1816                 if (res > 0) {
1817                         res = -EEXIST;
1818                         goto return_result;
1819                 }
1820                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1821                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1822                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1823                         break;
1824                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1825                 offset += rlen;
1826         }
1827
1828         if ((char *) de > top)
1829                 res = -ENOSPC;
1830         else {
1831                 *dest_de = de;
1832                 res = 0;
1833         }
1834 return_result:
1835         return res;
1836 }
1837
1838 int ext4_insert_dentry(struct inode *dir,
1839                        struct inode *inode,
1840                        struct ext4_dir_entry_2 *de,
1841                        int buf_size,
1842                        struct ext4_filename *fname)
1843 {
1844
1845         int nlen, rlen;
1846
1847         nlen = EXT4_DIR_REC_LEN(de->name_len);
1848         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1849         if (de->inode) {
1850                 struct ext4_dir_entry_2 *de1 =
1851                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1852                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1853                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1854                 de = de1;
1855         }
1856         de->file_type = EXT4_FT_UNKNOWN;
1857         de->inode = cpu_to_le32(inode->i_ino);
1858         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1859         de->name_len = fname_len(fname);
1860         memcpy(de->name, fname_name(fname), fname_len(fname));
1861         return 0;
1862 }
1863
1864 /*
1865  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1866  * it points to a directory entry which is guaranteed to be large
1867  * enough for new directory entry.  If de is NULL, then
1868  * add_dirent_to_buf will attempt search the directory block for
1869  * space.  It will return -ENOSPC if no space is available, and -EIO
1870  * and -EEXIST if directory entry already exists.
1871  */
1872 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1873                              struct inode *dir,
1874                              struct inode *inode, struct ext4_dir_entry_2 *de,
1875                              struct buffer_head *bh)
1876 {
1877         unsigned int    blocksize = dir->i_sb->s_blocksize;
1878         int             csum_size = 0;
1879         int             err;
1880
1881         if (ext4_has_metadata_csum(inode->i_sb))
1882                 csum_size = sizeof(struct ext4_dir_entry_tail);
1883
1884         if (!de) {
1885                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1886                                         blocksize - csum_size, fname, &de);
1887                 if (err)
1888                         return err;
1889         }
1890         BUFFER_TRACE(bh, "get_write_access");
1891         err = ext4_journal_get_write_access(handle, bh);
1892         if (err) {
1893                 ext4_std_error(dir->i_sb, err);
1894                 return err;
1895         }
1896
1897         /* By now the buffer is marked for journaling. Due to crypto operations,
1898          * the following function call may fail */
1899         err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1900         if (err < 0)
1901                 return err;
1902
1903         /*
1904          * XXX shouldn't update any times until successful
1905          * completion of syscall, but too many callers depend
1906          * on this.
1907          *
1908          * XXX similarly, too many callers depend on
1909          * ext4_new_inode() setting the times, but error
1910          * recovery deletes the inode, so the worst that can
1911          * happen is that the times are slightly out of date
1912          * and/or different from the directory change time.
1913          */
1914         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1915         ext4_update_dx_flag(dir);
1916         dir->i_version++;
1917         ext4_mark_inode_dirty(handle, dir);
1918         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1919         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1920         if (err)
1921                 ext4_std_error(dir->i_sb, err);
1922         return 0;
1923 }
1924
1925 /*
1926  * This converts a one block unindexed directory to a 3 block indexed
1927  * directory, and adds the dentry to the indexed directory.
1928  */
1929 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1930                             struct dentry *dentry,
1931                             struct inode *inode, struct buffer_head *bh)
1932 {
1933         struct inode    *dir = d_inode(dentry->d_parent);
1934         struct buffer_head *bh2;
1935         struct dx_root  *root;
1936         struct dx_frame frames[2], *frame;
1937         struct dx_entry *entries;
1938         struct ext4_dir_entry_2 *de, *de2;
1939         struct ext4_dir_entry_tail *t;
1940         char            *data1, *top;
1941         unsigned        len;
1942         int             retval;
1943         unsigned        blocksize;
1944         ext4_lblk_t  block;
1945         struct fake_dirent *fde;
1946         int csum_size = 0;
1947
1948         if (ext4_has_metadata_csum(inode->i_sb))
1949                 csum_size = sizeof(struct ext4_dir_entry_tail);
1950
1951         blocksize =  dir->i_sb->s_blocksize;
1952         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1953         BUFFER_TRACE(bh, "get_write_access");
1954         retval = ext4_journal_get_write_access(handle, bh);
1955         if (retval) {
1956                 ext4_std_error(dir->i_sb, retval);
1957                 brelse(bh);
1958                 return retval;
1959         }
1960         root = (struct dx_root *) bh->b_data;
1961
1962         /* The 0th block becomes the root, move the dirents out */
1963         fde = &root->dotdot;
1964         de = (struct ext4_dir_entry_2 *)((char *)fde +
1965                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1966         if ((char *) de >= (((char *) root) + blocksize)) {
1967                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1968                 brelse(bh);
1969                 return -EFSCORRUPTED;
1970         }
1971         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1972
1973         /* Allocate new block for the 0th block's dirents */
1974         bh2 = ext4_append(handle, dir, &block);
1975         if (IS_ERR(bh2)) {
1976                 brelse(bh);
1977                 return PTR_ERR(bh2);
1978         }
1979         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1980         data1 = bh2->b_data;
1981
1982         memcpy (data1, de, len);
1983         de = (struct ext4_dir_entry_2 *) data1;
1984         top = data1 + len;
1985         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1986                 de = de2;
1987         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1988                                            (char *) de,
1989                                            blocksize);
1990
1991         if (csum_size) {
1992                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1993                 initialize_dirent_tail(t, blocksize);
1994         }
1995
1996         /* Initialize the root; the dot dirents already exist */
1997         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1998         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1999                                            blocksize);
2000         memset (&root->info, 0, sizeof(root->info));
2001         root->info.info_length = sizeof(root->info);
2002         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2003         entries = root->entries;
2004         dx_set_block(entries, 1);
2005         dx_set_count(entries, 1);
2006         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2007
2008         /* Initialize as for dx_probe */
2009         fname->hinfo.hash_version = root->info.hash_version;
2010         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2011                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2012         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2013         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2014
2015         memset(frames, 0, sizeof(frames));
2016         frame = frames;
2017         frame->entries = entries;
2018         frame->at = entries;
2019         frame->bh = bh;
2020
2021         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2022         if (retval)
2023                 goto out_frames;
2024         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2025         if (retval)
2026                 goto out_frames;
2027
2028         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2029         if (IS_ERR(de)) {
2030                 retval = PTR_ERR(de);
2031                 goto out_frames;
2032         }
2033
2034         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2035 out_frames:
2036         /*
2037          * Even if the block split failed, we have to properly write
2038          * out all the changes we did so far. Otherwise we can end up
2039          * with corrupted filesystem.
2040          */
2041         if (retval)
2042                 ext4_mark_inode_dirty(handle, dir);
2043         dx_release(frames);
2044         brelse(bh2);
2045         return retval;
2046 }
2047
2048 /*
2049  *      ext4_add_entry()
2050  *
2051  * adds a file entry to the specified directory, using the same
2052  * semantics as ext4_find_entry(). It returns NULL if it failed.
2053  *
2054  * NOTE!! The inode part of 'de' is left at 0 - which means you
2055  * may not sleep between calling this and putting something into
2056  * the entry, as someone else might have used it while you slept.
2057  */
2058 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2059                           struct inode *inode)
2060 {
2061         struct inode *dir = d_inode(dentry->d_parent);
2062         struct buffer_head *bh = NULL;
2063         struct ext4_dir_entry_2 *de;
2064         struct ext4_dir_entry_tail *t;
2065         struct super_block *sb;
2066         struct ext4_filename fname;
2067         int     retval;
2068         int     dx_fallback=0;
2069         unsigned blocksize;
2070         ext4_lblk_t block, blocks;
2071         int     csum_size = 0;
2072
2073         if (ext4_has_metadata_csum(inode->i_sb))
2074                 csum_size = sizeof(struct ext4_dir_entry_tail);
2075
2076         sb = dir->i_sb;
2077         blocksize = sb->s_blocksize;
2078         if (!dentry->d_name.len)
2079                 return -EINVAL;
2080
2081         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2082         if (retval)
2083                 return retval;
2084
2085         if (ext4_has_inline_data(dir)) {
2086                 retval = ext4_try_add_inline_entry(handle, &fname,
2087                                                    dentry, inode);
2088                 if (retval < 0)
2089                         goto out;
2090                 if (retval == 1) {
2091                         retval = 0;
2092                         goto out;
2093                 }
2094         }
2095
2096         if (is_dx(dir)) {
2097                 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2098                 if (!retval || (retval != ERR_BAD_DX_DIR))
2099                         goto out;
2100                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2101                 dx_fallback++;
2102                 ext4_mark_inode_dirty(handle, dir);
2103         }
2104         blocks = dir->i_size >> sb->s_blocksize_bits;
2105         for (block = 0; block < blocks; block++) {
2106                 bh = ext4_read_dirblock(dir, block, DIRENT);
2107                 if (IS_ERR(bh)) {
2108                         retval = PTR_ERR(bh);
2109                         bh = NULL;
2110                         goto out;
2111                 }
2112                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2113                                            NULL, bh);
2114                 if (retval != -ENOSPC)
2115                         goto out;
2116
2117                 if (blocks == 1 && !dx_fallback &&
2118                     ext4_has_feature_dir_index(sb)) {
2119                         retval = make_indexed_dir(handle, &fname, dentry,
2120                                                   inode, bh);
2121                         bh = NULL; /* make_indexed_dir releases bh */
2122                         goto out;
2123                 }
2124                 brelse(bh);
2125         }
2126         bh = ext4_append(handle, dir, &block);
2127         if (IS_ERR(bh)) {
2128                 retval = PTR_ERR(bh);
2129                 bh = NULL;
2130                 goto out;
2131         }
2132         de = (struct ext4_dir_entry_2 *) bh->b_data;
2133         de->inode = 0;
2134         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2135
2136         if (csum_size) {
2137                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2138                 initialize_dirent_tail(t, blocksize);
2139         }
2140
2141         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2142 out:
2143         ext4_fname_free_filename(&fname);
2144         brelse(bh);
2145         if (retval == 0)
2146                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2147         return retval;
2148 }
2149
2150 /*
2151  * Returns 0 for success, or a negative error value
2152  */
2153 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2154                              struct dentry *dentry, struct inode *inode)
2155 {
2156         struct dx_frame frames[2], *frame;
2157         struct dx_entry *entries, *at;
2158         struct buffer_head *bh;
2159         struct inode *dir = d_inode(dentry->d_parent);
2160         struct super_block *sb = dir->i_sb;
2161         struct ext4_dir_entry_2 *de;
2162         int err;
2163
2164         frame = dx_probe(fname, dir, NULL, frames);
2165         if (IS_ERR(frame))
2166                 return PTR_ERR(frame);
2167         entries = frame->entries;
2168         at = frame->at;
2169         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2170         if (IS_ERR(bh)) {
2171                 err = PTR_ERR(bh);
2172                 bh = NULL;
2173                 goto cleanup;
2174         }
2175
2176         BUFFER_TRACE(bh, "get_write_access");
2177         err = ext4_journal_get_write_access(handle, bh);
2178         if (err)
2179                 goto journal_error;
2180
2181         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2182         if (err != -ENOSPC)
2183                 goto cleanup;
2184
2185         /* Block full, should compress but for now just split */
2186         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2187                        dx_get_count(entries), dx_get_limit(entries)));
2188         /* Need to split index? */
2189         if (dx_get_count(entries) == dx_get_limit(entries)) {
2190                 ext4_lblk_t newblock;
2191                 unsigned icount = dx_get_count(entries);
2192                 int levels = frame - frames;
2193                 struct dx_entry *entries2;
2194                 struct dx_node *node2;
2195                 struct buffer_head *bh2;
2196
2197                 if (levels && (dx_get_count(frames->entries) ==
2198                                dx_get_limit(frames->entries))) {
2199                         ext4_warning_inode(dir, "Directory index full!");
2200                         err = -ENOSPC;
2201                         goto cleanup;
2202                 }
2203                 bh2 = ext4_append(handle, dir, &newblock);
2204                 if (IS_ERR(bh2)) {
2205                         err = PTR_ERR(bh2);
2206                         goto cleanup;
2207                 }
2208                 node2 = (struct dx_node *)(bh2->b_data);
2209                 entries2 = node2->entries;
2210                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2211                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2212                                                            sb->s_blocksize);
2213                 BUFFER_TRACE(frame->bh, "get_write_access");
2214                 err = ext4_journal_get_write_access(handle, frame->bh);
2215                 if (err)
2216                         goto journal_error;
2217                 if (levels) {
2218                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2219                         unsigned hash2 = dx_get_hash(entries + icount1);
2220                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2221                                        icount1, icount2));
2222
2223                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2224                         err = ext4_journal_get_write_access(handle,
2225                                                              frames[0].bh);
2226                         if (err)
2227                                 goto journal_error;
2228
2229                         memcpy((char *) entries2, (char *) (entries + icount1),
2230                                icount2 * sizeof(struct dx_entry));
2231                         dx_set_count(entries, icount1);
2232                         dx_set_count(entries2, icount2);
2233                         dx_set_limit(entries2, dx_node_limit(dir));
2234
2235                         /* Which index block gets the new entry? */
2236                         if (at - entries >= icount1) {
2237                                 frame->at = at = at - entries - icount1 + entries2;
2238                                 frame->entries = entries = entries2;
2239                                 swap(frame->bh, bh2);
2240                         }
2241                         dx_insert_block(frames + 0, hash2, newblock);
2242                         dxtrace(dx_show_index("node", frames[1].entries));
2243                         dxtrace(dx_show_index("node",
2244                                ((struct dx_node *) bh2->b_data)->entries));
2245                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2246                         if (err)
2247                                 goto journal_error;
2248                         brelse (bh2);
2249                 } else {
2250                         dxtrace(printk(KERN_DEBUG
2251                                        "Creating second level index...\n"));
2252                         memcpy((char *) entries2, (char *) entries,
2253                                icount * sizeof(struct dx_entry));
2254                         dx_set_limit(entries2, dx_node_limit(dir));
2255
2256                         /* Set up root */
2257                         dx_set_count(entries, 1);
2258                         dx_set_block(entries + 0, newblock);
2259                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2260
2261                         /* Add new access path frame */
2262                         frame = frames + 1;
2263                         frame->at = at = at - entries + entries2;
2264                         frame->entries = entries = entries2;
2265                         frame->bh = bh2;
2266                         err = ext4_journal_get_write_access(handle,
2267                                                              frame->bh);
2268                         if (err)
2269                                 goto journal_error;
2270                 }
2271                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2272                 if (err) {
2273                         ext4_std_error(inode->i_sb, err);
2274                         goto cleanup;
2275                 }
2276         }
2277         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2278         if (IS_ERR(de)) {
2279                 err = PTR_ERR(de);
2280                 goto cleanup;
2281         }
2282         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2283         goto cleanup;
2284
2285 journal_error:
2286         ext4_std_error(dir->i_sb, err);
2287 cleanup:
2288         brelse(bh);
2289         dx_release(frames);
2290         return err;
2291 }
2292
2293 /*
2294  * ext4_generic_delete_entry deletes a directory entry by merging it
2295  * with the previous entry
2296  */
2297 int ext4_generic_delete_entry(handle_t *handle,
2298                               struct inode *dir,
2299                               struct ext4_dir_entry_2 *de_del,
2300                               struct buffer_head *bh,
2301                               void *entry_buf,
2302                               int buf_size,
2303                               int csum_size)
2304 {
2305         struct ext4_dir_entry_2 *de, *pde;
2306         unsigned int blocksize = dir->i_sb->s_blocksize;
2307         int i;
2308
2309         i = 0;
2310         pde = NULL;
2311         de = (struct ext4_dir_entry_2 *)entry_buf;
2312         while (i < buf_size - csum_size) {
2313                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2314                                          bh->b_data, bh->b_size, i))
2315                         return -EFSCORRUPTED;
2316                 if (de == de_del)  {
2317                         if (pde)
2318                                 pde->rec_len = ext4_rec_len_to_disk(
2319                                         ext4_rec_len_from_disk(pde->rec_len,
2320                                                                blocksize) +
2321                                         ext4_rec_len_from_disk(de->rec_len,
2322                                                                blocksize),
2323                                         blocksize);
2324                         else
2325                                 de->inode = 0;
2326                         dir->i_version++;
2327                         return 0;
2328                 }
2329                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2330                 pde = de;
2331                 de = ext4_next_entry(de, blocksize);
2332         }
2333         return -ENOENT;
2334 }
2335
2336 static int ext4_delete_entry(handle_t *handle,
2337                              struct inode *dir,
2338                              struct ext4_dir_entry_2 *de_del,
2339                              struct buffer_head *bh)
2340 {
2341         int err, csum_size = 0;
2342
2343         if (ext4_has_inline_data(dir)) {
2344                 int has_inline_data = 1;
2345                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2346                                                &has_inline_data);
2347                 if (has_inline_data)
2348                         return err;
2349         }
2350
2351         if (ext4_has_metadata_csum(dir->i_sb))
2352                 csum_size = sizeof(struct ext4_dir_entry_tail);
2353
2354         BUFFER_TRACE(bh, "get_write_access");
2355         err = ext4_journal_get_write_access(handle, bh);
2356         if (unlikely(err))
2357                 goto out;
2358
2359         err = ext4_generic_delete_entry(handle, dir, de_del,
2360                                         bh, bh->b_data,
2361                                         dir->i_sb->s_blocksize, csum_size);
2362         if (err)
2363                 goto out;
2364
2365         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2366         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2367         if (unlikely(err))
2368                 goto out;
2369
2370         return 0;
2371 out:
2372         if (err != -ENOENT)
2373                 ext4_std_error(dir->i_sb, err);
2374         return err;
2375 }
2376
2377 /*
2378  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2379  * since this indicates that nlinks count was previously 1.
2380  */
2381 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2382 {
2383         inc_nlink(inode);
2384         if (is_dx(inode) && inode->i_nlink > 1) {
2385                 /* limit is 16-bit i_links_count */
2386                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2387                         set_nlink(inode, 1);
2388                         ext4_set_feature_dir_nlink(inode->i_sb);
2389                 }
2390         }
2391 }
2392
2393 /*
2394  * If a directory had nlink == 1, then we should let it be 1. This indicates
2395  * directory has >EXT4_LINK_MAX subdirs.
2396  */
2397 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2398 {
2399         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2400                 drop_nlink(inode);
2401 }
2402
2403
2404 static int ext4_add_nondir(handle_t *handle,
2405                 struct dentry *dentry, struct inode *inode)
2406 {
2407         int err = ext4_add_entry(handle, dentry, inode);
2408         if (!err) {
2409                 ext4_mark_inode_dirty(handle, inode);
2410                 unlock_new_inode(inode);
2411                 d_instantiate(dentry, inode);
2412                 return 0;
2413         }
2414         drop_nlink(inode);
2415         unlock_new_inode(inode);
2416         iput(inode);
2417         return err;
2418 }
2419
2420 /*
2421  * By the time this is called, we already have created
2422  * the directory cache entry for the new file, but it
2423  * is so far negative - it has no inode.
2424  *
2425  * If the create succeeds, we fill in the inode information
2426  * with d_instantiate().
2427  */
2428 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2429                        bool excl)
2430 {
2431         handle_t *handle;
2432         struct inode *inode;
2433         int err, credits, retries = 0;
2434
2435         err = dquot_initialize(dir);
2436         if (err)
2437                 return err;
2438
2439         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2440                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2441 retry:
2442         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2443                                             NULL, EXT4_HT_DIR, credits);
2444         handle = ext4_journal_current_handle();
2445         err = PTR_ERR(inode);
2446         if (!IS_ERR(inode)) {
2447                 inode->i_op = &ext4_file_inode_operations;
2448                 inode->i_fop = &ext4_file_operations;
2449                 ext4_set_aops(inode);
2450                 err = ext4_add_nondir(handle, dentry, inode);
2451                 if (!err && IS_DIRSYNC(dir))
2452                         ext4_handle_sync(handle);
2453         }
2454         if (handle)
2455                 ext4_journal_stop(handle);
2456         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2457                 goto retry;
2458         return err;
2459 }
2460
2461 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2462                       umode_t mode, dev_t rdev)
2463 {
2464         handle_t *handle;
2465         struct inode *inode;
2466         int err, credits, retries = 0;
2467
2468         err = dquot_initialize(dir);
2469         if (err)
2470                 return err;
2471
2472         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2473                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2474 retry:
2475         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2476                                             NULL, EXT4_HT_DIR, credits);
2477         handle = ext4_journal_current_handle();
2478         err = PTR_ERR(inode);
2479         if (!IS_ERR(inode)) {
2480                 init_special_inode(inode, inode->i_mode, rdev);
2481                 inode->i_op = &ext4_special_inode_operations;
2482                 err = ext4_add_nondir(handle, dentry, inode);
2483                 if (!err && IS_DIRSYNC(dir))
2484                         ext4_handle_sync(handle);
2485         }
2486         if (handle)
2487                 ext4_journal_stop(handle);
2488         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2489                 goto retry;
2490         return err;
2491 }
2492
2493 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2494 {
2495         handle_t *handle;
2496         struct inode *inode;
2497         int err, retries = 0;
2498
2499         err = dquot_initialize(dir);
2500         if (err)
2501                 return err;
2502
2503 retry:
2504         inode = ext4_new_inode_start_handle(dir, mode,
2505                                             NULL, 0, NULL,
2506                                             EXT4_HT_DIR,
2507                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2508                           4 + EXT4_XATTR_TRANS_BLOCKS);
2509         handle = ext4_journal_current_handle();
2510         err = PTR_ERR(inode);
2511         if (!IS_ERR(inode)) {
2512                 inode->i_op = &ext4_file_inode_operations;
2513                 inode->i_fop = &ext4_file_operations;
2514                 ext4_set_aops(inode);
2515                 d_tmpfile(dentry, inode);
2516                 err = ext4_orphan_add(handle, inode);
2517                 if (err)
2518                         goto err_unlock_inode;
2519                 mark_inode_dirty(inode);
2520                 unlock_new_inode(inode);
2521         }
2522         if (handle)
2523                 ext4_journal_stop(handle);
2524         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2525                 goto retry;
2526         return err;
2527 err_unlock_inode:
2528         ext4_journal_stop(handle);
2529         unlock_new_inode(inode);
2530         return err;
2531 }
2532
2533 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2534                           struct ext4_dir_entry_2 *de,
2535                           int blocksize, int csum_size,
2536                           unsigned int parent_ino, int dotdot_real_len)
2537 {
2538         de->inode = cpu_to_le32(inode->i_ino);
2539         de->name_len = 1;
2540         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2541                                            blocksize);
2542         strcpy(de->name, ".");
2543         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2544
2545         de = ext4_next_entry(de, blocksize);
2546         de->inode = cpu_to_le32(parent_ino);
2547         de->name_len = 2;
2548         if (!dotdot_real_len)
2549                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2550                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2551                                         blocksize);
2552         else
2553                 de->rec_len = ext4_rec_len_to_disk(
2554                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2555         strcpy(de->name, "..");
2556         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2557
2558         return ext4_next_entry(de, blocksize);
2559 }
2560
2561 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2562                              struct inode *inode)
2563 {
2564         struct buffer_head *dir_block = NULL;
2565         struct ext4_dir_entry_2 *de;
2566         struct ext4_dir_entry_tail *t;
2567         ext4_lblk_t block = 0;
2568         unsigned int blocksize = dir->i_sb->s_blocksize;
2569         int csum_size = 0;
2570         int err;
2571
2572         if (ext4_has_metadata_csum(dir->i_sb))
2573                 csum_size = sizeof(struct ext4_dir_entry_tail);
2574
2575         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2576                 err = ext4_try_create_inline_dir(handle, dir, inode);
2577                 if (err < 0 && err != -ENOSPC)
2578                         goto out;
2579                 if (!err)
2580                         goto out;
2581         }
2582
2583         inode->i_size = 0;
2584         dir_block = ext4_append(handle, inode, &block);
2585         if (IS_ERR(dir_block))
2586                 return PTR_ERR(dir_block);
2587         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2588         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2589         set_nlink(inode, 2);
2590         if (csum_size) {
2591                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2592                 initialize_dirent_tail(t, blocksize);
2593         }
2594
2595         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2596         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2597         if (err)
2598                 goto out;
2599         set_buffer_verified(dir_block);
2600 out:
2601         brelse(dir_block);
2602         return err;
2603 }
2604
2605 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2606 {
2607         handle_t *handle;
2608         struct inode *inode;
2609         int err, credits, retries = 0;
2610
2611         if (EXT4_DIR_LINK_MAX(dir))
2612                 return -EMLINK;
2613
2614         err = dquot_initialize(dir);
2615         if (err)
2616                 return err;
2617
2618         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2619                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2620 retry:
2621         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2622                                             &dentry->d_name,
2623                                             0, NULL, EXT4_HT_DIR, credits);
2624         handle = ext4_journal_current_handle();
2625         err = PTR_ERR(inode);
2626         if (IS_ERR(inode))
2627                 goto out_stop;
2628
2629         inode->i_op = &ext4_dir_inode_operations;
2630         inode->i_fop = &ext4_dir_operations;
2631         err = ext4_init_new_dir(handle, dir, inode);
2632         if (err)
2633                 goto out_clear_inode;
2634         err = ext4_mark_inode_dirty(handle, inode);
2635         if (!err)
2636                 err = ext4_add_entry(handle, dentry, inode);
2637         if (err) {
2638 out_clear_inode:
2639                 clear_nlink(inode);
2640                 unlock_new_inode(inode);
2641                 ext4_mark_inode_dirty(handle, inode);
2642                 iput(inode);
2643                 goto out_stop;
2644         }
2645         ext4_inc_count(handle, dir);
2646         ext4_update_dx_flag(dir);
2647         err = ext4_mark_inode_dirty(handle, dir);
2648         if (err)
2649                 goto out_clear_inode;
2650         unlock_new_inode(inode);
2651         d_instantiate(dentry, inode);
2652         if (IS_DIRSYNC(dir))
2653                 ext4_handle_sync(handle);
2654
2655 out_stop:
2656         if (handle)
2657                 ext4_journal_stop(handle);
2658         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2659                 goto retry;
2660         return err;
2661 }
2662
2663 /*
2664  * routine to check that the specified directory is empty (for rmdir)
2665  */
2666 int ext4_empty_dir(struct inode *inode)
2667 {
2668         unsigned int offset;
2669         struct buffer_head *bh;
2670         struct ext4_dir_entry_2 *de, *de1;
2671         struct super_block *sb;
2672         int err = 0;
2673
2674         if (ext4_has_inline_data(inode)) {
2675                 int has_inline_data = 1;
2676
2677                 err = empty_inline_dir(inode, &has_inline_data);
2678                 if (has_inline_data)
2679                         return err;
2680         }
2681
2682         sb = inode->i_sb;
2683         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2684                 EXT4_ERROR_INODE(inode, "invalid size");
2685                 return 1;
2686         }
2687         bh = ext4_read_dirblock(inode, 0, EITHER);
2688         if (IS_ERR(bh))
2689                 return 1;
2690
2691         de = (struct ext4_dir_entry_2 *) bh->b_data;
2692         de1 = ext4_next_entry(de, sb->s_blocksize);
2693         if (le32_to_cpu(de->inode) != inode->i_ino ||
2694                         le32_to_cpu(de1->inode) == 0 ||
2695                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2696                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2697                 brelse(bh);
2698                 return 1;
2699         }
2700         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2701                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2702         de = ext4_next_entry(de1, sb->s_blocksize);
2703         while (offset < inode->i_size) {
2704                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2705                         unsigned int lblock;
2706                         err = 0;
2707                         brelse(bh);
2708                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2709                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2710                         if (IS_ERR(bh))
2711                                 return 1;
2712                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2713                 }
2714                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2715                                          bh->b_data, bh->b_size, offset)) {
2716                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2717                                                          sb->s_blocksize);
2718                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2719                         continue;
2720                 }
2721                 if (le32_to_cpu(de->inode)) {
2722                         brelse(bh);
2723                         return 0;
2724                 }
2725                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2726                 de = ext4_next_entry(de, sb->s_blocksize);
2727         }
2728         brelse(bh);
2729         return 1;
2730 }
2731
2732 /*
2733  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2734  * such inodes, starting at the superblock, in case we crash before the
2735  * file is closed/deleted, or in case the inode truncate spans multiple
2736  * transactions and the last transaction is not recovered after a crash.
2737  *
2738  * At filesystem recovery time, we walk this list deleting unlinked
2739  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2740  *
2741  * Orphan list manipulation functions must be called under i_mutex unless
2742  * we are just creating the inode or deleting it.
2743  */
2744 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2745 {
2746         struct super_block *sb = inode->i_sb;
2747         struct ext4_sb_info *sbi = EXT4_SB(sb);
2748         struct ext4_iloc iloc;
2749         int err = 0, rc;
2750         bool dirty = false;
2751
2752         if (!sbi->s_journal || is_bad_inode(inode))
2753                 return 0;
2754
2755         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2756                      !mutex_is_locked(&inode->i_mutex));
2757         /*
2758          * Exit early if inode already is on orphan list. This is a big speedup
2759          * since we don't have to contend on the global s_orphan_lock.
2760          */
2761         if (!list_empty(&EXT4_I(inode)->i_orphan))
2762                 return 0;
2763
2764         /*
2765          * Orphan handling is only valid for files with data blocks
2766          * being truncated, or files being unlinked. Note that we either
2767          * hold i_mutex, or the inode can not be referenced from outside,
2768          * so i_nlink should not be bumped due to race
2769          */
2770         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2771                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2772
2773         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2774         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2775         if (err)
2776                 goto out;
2777
2778         err = ext4_reserve_inode_write(handle, inode, &iloc);
2779         if (err)
2780                 goto out;
2781
2782         mutex_lock(&sbi->s_orphan_lock);
2783         /*
2784          * Due to previous errors inode may be already a part of on-disk
2785          * orphan list. If so skip on-disk list modification.
2786          */
2787         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2788             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2789                 /* Insert this inode at the head of the on-disk orphan list */
2790                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2791                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2792                 dirty = true;
2793         }
2794         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2795         mutex_unlock(&sbi->s_orphan_lock);
2796
2797         if (dirty) {
2798                 err = ext4_handle_dirty_super(handle, sb);
2799                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2800                 if (!err)
2801                         err = rc;
2802                 if (err) {
2803                         /*
2804                          * We have to remove inode from in-memory list if
2805                          * addition to on disk orphan list failed. Stray orphan
2806                          * list entries can cause panics at unmount time.
2807                          */
2808                         mutex_lock(&sbi->s_orphan_lock);
2809                         list_del_init(&EXT4_I(inode)->i_orphan);
2810                         mutex_unlock(&sbi->s_orphan_lock);
2811                 }
2812         }
2813         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2814         jbd_debug(4, "orphan inode %lu will point to %d\n",
2815                         inode->i_ino, NEXT_ORPHAN(inode));
2816 out:
2817         ext4_std_error(sb, err);
2818         return err;
2819 }
2820
2821 /*
2822  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2823  * of such inodes stored on disk, because it is finally being cleaned up.
2824  */
2825 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2826 {
2827         struct list_head *prev;
2828         struct ext4_inode_info *ei = EXT4_I(inode);
2829         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2830         __u32 ino_next;
2831         struct ext4_iloc iloc;
2832         int err = 0;
2833
2834         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2835                 return 0;
2836
2837         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2838                      !mutex_is_locked(&inode->i_mutex));
2839         /* Do this quick check before taking global s_orphan_lock. */
2840         if (list_empty(&ei->i_orphan))
2841                 return 0;
2842
2843         if (handle) {
2844                 /* Grab inode buffer early before taking global s_orphan_lock */
2845                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2846         }
2847
2848         mutex_lock(&sbi->s_orphan_lock);
2849         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2850
2851         prev = ei->i_orphan.prev;
2852         list_del_init(&ei->i_orphan);
2853
2854         /* If we're on an error path, we may not have a valid
2855          * transaction handle with which to update the orphan list on
2856          * disk, but we still need to remove the inode from the linked
2857          * list in memory. */
2858         if (!handle || err) {
2859                 mutex_unlock(&sbi->s_orphan_lock);
2860                 goto out_err;
2861         }
2862
2863         ino_next = NEXT_ORPHAN(inode);
2864         if (prev == &sbi->s_orphan) {
2865                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2866                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2867                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2868                 if (err) {
2869                         mutex_unlock(&sbi->s_orphan_lock);
2870                         goto out_brelse;
2871                 }
2872                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2873                 mutex_unlock(&sbi->s_orphan_lock);
2874                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2875         } else {
2876                 struct ext4_iloc iloc2;
2877                 struct inode *i_prev =
2878                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2879
2880                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2881                           i_prev->i_ino, ino_next);
2882                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2883                 if (err) {
2884                         mutex_unlock(&sbi->s_orphan_lock);
2885                         goto out_brelse;
2886                 }
2887                 NEXT_ORPHAN(i_prev) = ino_next;
2888                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2889                 mutex_unlock(&sbi->s_orphan_lock);
2890         }
2891         if (err)
2892                 goto out_brelse;
2893         NEXT_ORPHAN(inode) = 0;
2894         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2895 out_err:
2896         ext4_std_error(inode->i_sb, err);
2897         return err;
2898
2899 out_brelse:
2900         brelse(iloc.bh);
2901         goto out_err;
2902 }
2903
2904 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2905 {
2906         int retval;
2907         struct inode *inode;
2908         struct buffer_head *bh;
2909         struct ext4_dir_entry_2 *de;
2910         handle_t *handle = NULL;
2911
2912         /* Initialize quotas before so that eventual writes go in
2913          * separate transaction */
2914         retval = dquot_initialize(dir);
2915         if (retval)
2916                 return retval;
2917         retval = dquot_initialize(d_inode(dentry));
2918         if (retval)
2919                 return retval;
2920
2921         retval = -ENOENT;
2922         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2923         if (IS_ERR(bh))
2924                 return PTR_ERR(bh);
2925         if (!bh)
2926                 goto end_rmdir;
2927
2928         inode = d_inode(dentry);
2929
2930         retval = -EFSCORRUPTED;
2931         if (le32_to_cpu(de->inode) != inode->i_ino)
2932                 goto end_rmdir;
2933
2934         retval = -ENOTEMPTY;
2935         if (!ext4_empty_dir(inode))
2936                 goto end_rmdir;
2937
2938         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2939                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2940         if (IS_ERR(handle)) {
2941                 retval = PTR_ERR(handle);
2942                 handle = NULL;
2943                 goto end_rmdir;
2944         }
2945
2946         if (IS_DIRSYNC(dir))
2947                 ext4_handle_sync(handle);
2948
2949         retval = ext4_delete_entry(handle, dir, de, bh);
2950         if (retval)
2951                 goto end_rmdir;
2952         if (!EXT4_DIR_LINK_EMPTY(inode))
2953                 ext4_warning_inode(inode,
2954                              "empty directory '%.*s' has too many links (%u)",
2955                              dentry->d_name.len, dentry->d_name.name,
2956                              inode->i_nlink);
2957         inode->i_version++;
2958         clear_nlink(inode);
2959         /* There's no need to set i_disksize: the fact that i_nlink is
2960          * zero will ensure that the right thing happens during any
2961          * recovery. */
2962         inode->i_size = 0;
2963         ext4_orphan_add(handle, inode);
2964         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2965         ext4_mark_inode_dirty(handle, inode);
2966         ext4_dec_count(handle, dir);
2967         ext4_update_dx_flag(dir);
2968         ext4_mark_inode_dirty(handle, dir);
2969
2970 end_rmdir:
2971         brelse(bh);
2972         if (handle)
2973                 ext4_journal_stop(handle);
2974         return retval;
2975 }
2976
2977 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2978 {
2979         int retval;
2980         struct inode *inode;
2981         struct buffer_head *bh;
2982         struct ext4_dir_entry_2 *de;
2983         handle_t *handle = NULL;
2984
2985         trace_ext4_unlink_enter(dir, dentry);
2986         /* Initialize quotas before so that eventual writes go
2987          * in separate transaction */
2988         retval = dquot_initialize(dir);
2989         if (retval)
2990                 return retval;
2991         retval = dquot_initialize(d_inode(dentry));
2992         if (retval)
2993                 return retval;
2994
2995         retval = -ENOENT;
2996         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2997         if (IS_ERR(bh))
2998                 return PTR_ERR(bh);
2999         if (!bh)
3000                 goto end_unlink;
3001
3002         inode = d_inode(dentry);
3003
3004         retval = -EFSCORRUPTED;
3005         if (le32_to_cpu(de->inode) != inode->i_ino)
3006                 goto end_unlink;
3007
3008         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3009                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3010         if (IS_ERR(handle)) {
3011                 retval = PTR_ERR(handle);
3012                 handle = NULL;
3013                 goto end_unlink;
3014         }
3015
3016         if (IS_DIRSYNC(dir))
3017                 ext4_handle_sync(handle);
3018
3019         if (inode->i_nlink == 0) {
3020                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3021                                    dentry->d_name.len, dentry->d_name.name);
3022                 set_nlink(inode, 1);
3023         }
3024         retval = ext4_delete_entry(handle, dir, de, bh);
3025         if (retval)
3026                 goto end_unlink;
3027         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3028         ext4_update_dx_flag(dir);
3029         ext4_mark_inode_dirty(handle, dir);
3030         drop_nlink(inode);
3031         if (!inode->i_nlink)
3032                 ext4_orphan_add(handle, inode);
3033         inode->i_ctime = ext4_current_time(inode);
3034         ext4_mark_inode_dirty(handle, inode);
3035
3036 end_unlink:
3037         brelse(bh);
3038         if (handle)
3039                 ext4_journal_stop(handle);
3040         trace_ext4_unlink_exit(dentry, retval);
3041         return retval;
3042 }
3043
3044 static int ext4_symlink(struct inode *dir,
3045                         struct dentry *dentry, const char *symname)
3046 {
3047         handle_t *handle;
3048         struct inode *inode;
3049         int err, len = strlen(symname);
3050         int credits;
3051         bool encryption_required;
3052         struct ext4_str disk_link;
3053         struct ext4_encrypted_symlink_data *sd = NULL;
3054
3055         disk_link.len = len + 1;
3056         disk_link.name = (char *) symname;
3057
3058         encryption_required = (ext4_encrypted_inode(dir) ||
3059                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3060         if (encryption_required) {
3061                 err = ext4_get_encryption_info(dir);
3062                 if (err)
3063                         return err;
3064                 if (ext4_encryption_info(dir) == NULL)
3065                         return -EPERM;
3066                 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3067                                  sizeof(struct ext4_encrypted_symlink_data));
3068                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3069                 if (!sd)
3070                         return -ENOMEM;
3071         }
3072
3073         if (disk_link.len > dir->i_sb->s_blocksize) {
3074                 err = -ENAMETOOLONG;
3075                 goto err_free_sd;
3076         }
3077
3078         err = dquot_initialize(dir);
3079         if (err)
3080                 goto err_free_sd;
3081
3082         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3083                 /*
3084                  * For non-fast symlinks, we just allocate inode and put it on
3085                  * orphan list in the first transaction => we need bitmap,
3086                  * group descriptor, sb, inode block, quota blocks, and
3087                  * possibly selinux xattr blocks.
3088                  */
3089                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3090                           EXT4_XATTR_TRANS_BLOCKS;
3091         } else {
3092                 /*
3093                  * Fast symlink. We have to add entry to directory
3094                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3095                  * allocate new inode (bitmap, group descriptor, inode block,
3096                  * quota blocks, sb is already counted in previous macros).
3097                  */
3098                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3099                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3100         }
3101
3102         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3103                                             &dentry->d_name, 0, NULL,
3104                                             EXT4_HT_DIR, credits);
3105         handle = ext4_journal_current_handle();
3106         if (IS_ERR(inode)) {
3107                 if (handle)
3108                         ext4_journal_stop(handle);
3109                 err = PTR_ERR(inode);
3110                 goto err_free_sd;
3111         }
3112
3113         if (encryption_required) {
3114                 struct qstr istr;
3115                 struct ext4_str ostr;
3116
3117                 istr.name = (const unsigned char *) symname;
3118                 istr.len = len;
3119                 ostr.name = sd->encrypted_path;
3120                 ostr.len = disk_link.len;
3121                 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3122                 if (err < 0)
3123                         goto err_drop_inode;
3124                 sd->len = cpu_to_le16(ostr.len);
3125                 disk_link.name = (char *) sd;
3126                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3127         }
3128
3129         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3130                 if (!encryption_required)
3131                         inode->i_op = &ext4_symlink_inode_operations;
3132                 ext4_set_aops(inode);
3133                 /*
3134                  * We cannot call page_symlink() with transaction started
3135                  * because it calls into ext4_write_begin() which can wait
3136                  * for transaction commit if we are running out of space
3137                  * and thus we deadlock. So we have to stop transaction now
3138                  * and restart it when symlink contents is written.
3139                  *
3140                  * To keep fs consistent in case of crash, we have to put inode
3141                  * to orphan list in the mean time.
3142                  */
3143                 drop_nlink(inode);
3144                 err = ext4_orphan_add(handle, inode);
3145                 ext4_journal_stop(handle);
3146                 handle = NULL;
3147                 if (err)
3148                         goto err_drop_inode;
3149                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3150                 if (err)
3151                         goto err_drop_inode;
3152                 /*
3153                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3154                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3155                  */
3156                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3157                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3158                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3159                 if (IS_ERR(handle)) {
3160                         err = PTR_ERR(handle);
3161                         handle = NULL;
3162                         goto err_drop_inode;
3163                 }
3164                 set_nlink(inode, 1);
3165                 err = ext4_orphan_del(handle, inode);
3166                 if (err)
3167                         goto err_drop_inode;
3168         } else {
3169                 /* clear the extent format for fast symlink */
3170                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3171                 if (!encryption_required) {
3172                         inode->i_op = &ext4_fast_symlink_inode_operations;
3173                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3174                 }
3175                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3176                        disk_link.len);
3177                 inode->i_size = disk_link.len - 1;
3178         }
3179         EXT4_I(inode)->i_disksize = inode->i_size;
3180         err = ext4_add_nondir(handle, dentry, inode);
3181         if (!err && IS_DIRSYNC(dir))
3182                 ext4_handle_sync(handle);
3183
3184         if (handle)
3185                 ext4_journal_stop(handle);
3186         kfree(sd);
3187         return err;
3188 err_drop_inode:
3189         if (handle)
3190                 ext4_journal_stop(handle);
3191         clear_nlink(inode);
3192         unlock_new_inode(inode);
3193         iput(inode);
3194 err_free_sd:
3195         kfree(sd);
3196         return err;
3197 }
3198
3199 static int ext4_link(struct dentry *old_dentry,
3200                      struct inode *dir, struct dentry *dentry)
3201 {
3202         handle_t *handle;
3203         struct inode *inode = d_inode(old_dentry);
3204         int err, retries = 0;
3205
3206         if (inode->i_nlink >= EXT4_LINK_MAX)
3207                 return -EMLINK;
3208         if (ext4_encrypted_inode(dir) &&
3209             !ext4_is_child_context_consistent_with_parent(dir, inode))
3210                 return -EPERM;
3211         err = dquot_initialize(dir);
3212         if (err)
3213                 return err;
3214
3215 retry:
3216         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3217                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3218                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3219         if (IS_ERR(handle))
3220                 return PTR_ERR(handle);
3221
3222         if (IS_DIRSYNC(dir))
3223                 ext4_handle_sync(handle);
3224
3225         inode->i_ctime = ext4_current_time(inode);
3226         ext4_inc_count(handle, inode);
3227         ihold(inode);
3228
3229         err = ext4_add_entry(handle, dentry, inode);
3230         if (!err) {
3231                 ext4_mark_inode_dirty(handle, inode);
3232                 /* this can happen only for tmpfile being
3233                  * linked the first time
3234                  */
3235                 if (inode->i_nlink == 1)
3236                         ext4_orphan_del(handle, inode);
3237                 d_instantiate(dentry, inode);
3238         } else {
3239                 drop_nlink(inode);
3240                 iput(inode);
3241         }
3242         ext4_journal_stop(handle);
3243         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3244                 goto retry;
3245         return err;
3246 }
3247
3248
3249 /*
3250  * Try to find buffer head where contains the parent block.
3251  * It should be the inode block if it is inlined or the 1st block
3252  * if it is a normal dir.
3253  */
3254 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3255                                         struct inode *inode,
3256                                         int *retval,
3257                                         struct ext4_dir_entry_2 **parent_de,
3258                                         int *inlined)
3259 {
3260         struct buffer_head *bh;
3261
3262         if (!ext4_has_inline_data(inode)) {
3263                 bh = ext4_read_dirblock(inode, 0, EITHER);
3264                 if (IS_ERR(bh)) {
3265                         *retval = PTR_ERR(bh);
3266                         return NULL;
3267                 }
3268                 *parent_de = ext4_next_entry(
3269                                         (struct ext4_dir_entry_2 *)bh->b_data,
3270                                         inode->i_sb->s_blocksize);
3271                 return bh;
3272         }
3273
3274         *inlined = 1;
3275         return ext4_get_first_inline_block(inode, parent_de, retval);
3276 }
3277
3278 struct ext4_renament {
3279         struct inode *dir;
3280         struct dentry *dentry;
3281         struct inode *inode;
3282         bool is_dir;
3283         int dir_nlink_delta;
3284
3285         /* entry for "dentry" */
3286         struct buffer_head *bh;
3287         struct ext4_dir_entry_2 *de;
3288         int inlined;
3289
3290         /* entry for ".." in inode if it's a directory */
3291         struct buffer_head *dir_bh;
3292         struct ext4_dir_entry_2 *parent_de;
3293         int dir_inlined;
3294 };
3295
3296 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3297 {
3298         int retval;
3299
3300         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3301                                               &retval, &ent->parent_de,
3302                                               &ent->dir_inlined);
3303         if (!ent->dir_bh)
3304                 return retval;
3305         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3306                 return -EFSCORRUPTED;
3307         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3308         return ext4_journal_get_write_access(handle, ent->dir_bh);
3309 }
3310
3311 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3312                                   unsigned dir_ino)
3313 {
3314         int retval;
3315
3316         ent->parent_de->inode = cpu_to_le32(dir_ino);
3317         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3318         if (!ent->dir_inlined) {
3319                 if (is_dx(ent->inode)) {
3320                         retval = ext4_handle_dirty_dx_node(handle,
3321                                                            ent->inode,
3322                                                            ent->dir_bh);
3323                 } else {
3324                         retval = ext4_handle_dirty_dirent_node(handle,
3325                                                                ent->inode,
3326                                                                ent->dir_bh);
3327                 }
3328         } else {
3329                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3330         }
3331         if (retval) {
3332                 ext4_std_error(ent->dir->i_sb, retval);
3333                 return retval;
3334         }
3335         return 0;
3336 }
3337
3338 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3339                        unsigned ino, unsigned file_type)
3340 {
3341         int retval;
3342
3343         BUFFER_TRACE(ent->bh, "get write access");
3344         retval = ext4_journal_get_write_access(handle, ent->bh);
3345         if (retval)
3346                 return retval;
3347         ent->de->inode = cpu_to_le32(ino);
3348         if (ext4_has_feature_filetype(ent->dir->i_sb))
3349                 ent->de->file_type = file_type;
3350         ent->dir->i_version++;
3351         ent->dir->i_ctime = ent->dir->i_mtime =
3352                 ext4_current_time(ent->dir);
3353         ext4_mark_inode_dirty(handle, ent->dir);
3354         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3355         if (!ent->inlined) {
3356                 retval = ext4_handle_dirty_dirent_node(handle,
3357                                                        ent->dir, ent->bh);
3358                 if (unlikely(retval)) {
3359                         ext4_std_error(ent->dir->i_sb, retval);
3360                         return retval;
3361                 }
3362         }
3363         brelse(ent->bh);
3364         ent->bh = NULL;
3365
3366         return 0;
3367 }
3368
3369 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3370                                   const struct qstr *d_name)
3371 {
3372         int retval = -ENOENT;
3373         struct buffer_head *bh;
3374         struct ext4_dir_entry_2 *de;
3375
3376         bh = ext4_find_entry(dir, d_name, &de, NULL);
3377         if (IS_ERR(bh))
3378                 return PTR_ERR(bh);
3379         if (bh) {
3380                 retval = ext4_delete_entry(handle, dir, de, bh);
3381                 brelse(bh);
3382         }
3383         return retval;
3384 }
3385
3386 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3387                                int force_reread)
3388 {
3389         int retval;
3390         /*
3391          * ent->de could have moved from under us during htree split, so make
3392          * sure that we are deleting the right entry.  We might also be pointing
3393          * to a stale entry in the unused part of ent->bh so just checking inum
3394          * and the name isn't enough.
3395          */
3396         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3397             ent->de->name_len != ent->dentry->d_name.len ||
3398             strncmp(ent->de->name, ent->dentry->d_name.name,
3399                     ent->de->name_len) ||
3400             force_reread) {
3401                 retval = ext4_find_delete_entry(handle, ent->dir,
3402                                                 &ent->dentry->d_name);
3403         } else {
3404                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3405                 if (retval == -ENOENT) {
3406                         retval = ext4_find_delete_entry(handle, ent->dir,
3407                                                         &ent->dentry->d_name);
3408                 }
3409         }
3410
3411         if (retval) {
3412                 ext4_warning_inode(ent->dir,
3413                                    "Deleting old file: nlink %d, error=%d",
3414                                    ent->dir->i_nlink, retval);
3415         }
3416 }
3417
3418 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3419 {
3420         if (ent->dir_nlink_delta) {
3421                 if (ent->dir_nlink_delta == -1)
3422                         ext4_dec_count(handle, ent->dir);
3423                 else
3424                         ext4_inc_count(handle, ent->dir);
3425                 ext4_mark_inode_dirty(handle, ent->dir);
3426         }
3427 }
3428
3429 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3430                                               int credits, handle_t **h)
3431 {
3432         struct inode *wh;
3433         handle_t *handle;
3434         int retries = 0;
3435
3436         /*
3437          * for inode block, sb block, group summaries,
3438          * and inode bitmap
3439          */
3440         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3441                     EXT4_XATTR_TRANS_BLOCKS + 4);
3442 retry:
3443         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3444                                          &ent->dentry->d_name, 0, NULL,
3445                                          EXT4_HT_DIR, credits);
3446
3447         handle = ext4_journal_current_handle();
3448         if (IS_ERR(wh)) {
3449                 if (handle)
3450                         ext4_journal_stop(handle);
3451                 if (PTR_ERR(wh) == -ENOSPC &&
3452                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3453                         goto retry;
3454         } else {
3455                 *h = handle;
3456                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3457                 wh->i_op = &ext4_special_inode_operations;
3458         }
3459         return wh;
3460 }
3461
3462 /*
3463  * Anybody can rename anything with this: the permission checks are left to the
3464  * higher-level routines.
3465  *
3466  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3467  * while new_{dentry,inode) refers to the destination dentry/inode
3468  * This comes from rename(const char *oldpath, const char *newpath)
3469  */
3470 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3471                        struct inode *new_dir, struct dentry *new_dentry,
3472                        unsigned int flags)
3473 {
3474         handle_t *handle = NULL;
3475         struct ext4_renament old = {
3476                 .dir = old_dir,
3477                 .dentry = old_dentry,
3478                 .inode = d_inode(old_dentry),
3479         };
3480         struct ext4_renament new = {
3481                 .dir = new_dir,
3482                 .dentry = new_dentry,
3483                 .inode = d_inode(new_dentry),
3484         };
3485         int force_reread;
3486         int retval;
3487         struct inode *whiteout = NULL;
3488         int credits;
3489         u8 old_file_type;
3490
3491         retval = dquot_initialize(old.dir);
3492         if (retval)
3493                 return retval;
3494         retval = dquot_initialize(new.dir);
3495         if (retval)
3496                 return retval;
3497
3498         /* Initialize quotas before so that eventual writes go
3499          * in separate transaction */
3500         if (new.inode) {
3501                 retval = dquot_initialize(new.inode);
3502                 if (retval)
3503                         return retval;
3504         }
3505
3506         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3507         if (IS_ERR(old.bh))
3508                 return PTR_ERR(old.bh);
3509         /*
3510          *  Check for inode number is _not_ due to possible IO errors.
3511          *  We might rmdir the source, keep it as pwd of some process
3512          *  and merrily kill the link to whatever was created under the
3513          *  same name. Goodbye sticky bit ;-<
3514          */
3515         retval = -ENOENT;
3516         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3517                 goto end_rename;
3518
3519         if ((old.dir != new.dir) &&
3520             ext4_encrypted_inode(new.dir) &&
3521             !ext4_is_child_context_consistent_with_parent(new.dir,
3522                                                           old.inode)) {
3523                 retval = -EPERM;
3524                 goto end_rename;
3525         }
3526
3527         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3528                                  &new.de, &new.inlined);
3529         if (IS_ERR(new.bh)) {
3530                 retval = PTR_ERR(new.bh);
3531                 new.bh = NULL;
3532                 goto end_rename;
3533         }
3534         if (new.bh) {
3535                 if (!new.inode) {
3536                         brelse(new.bh);
3537                         new.bh = NULL;
3538                 }
3539         }
3540         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3541                 ext4_alloc_da_blocks(old.inode);
3542
3543         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3544                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3545         if (!(flags & RENAME_WHITEOUT)) {
3546                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3547                 if (IS_ERR(handle)) {
3548                         retval = PTR_ERR(handle);
3549                         handle = NULL;
3550                         goto end_rename;
3551                 }
3552         } else {
3553                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3554                 if (IS_ERR(whiteout)) {
3555                         retval = PTR_ERR(whiteout);
3556                         whiteout = NULL;
3557                         goto end_rename;
3558                 }
3559         }
3560
3561         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3562                 ext4_handle_sync(handle);
3563
3564         if (S_ISDIR(old.inode->i_mode)) {
3565                 if (new.inode) {
3566                         retval = -ENOTEMPTY;
3567                         if (!ext4_empty_dir(new.inode))
3568                                 goto end_rename;
3569                 } else {
3570                         retval = -EMLINK;
3571                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3572                                 goto end_rename;
3573                 }
3574                 retval = ext4_rename_dir_prepare(handle, &old);
3575                 if (retval)
3576                         goto end_rename;
3577         }
3578         /*
3579          * If we're renaming a file within an inline_data dir and adding or
3580          * setting the new dirent causes a conversion from inline_data to
3581          * extents/blockmap, we need to force the dirent delete code to
3582          * re-read the directory, or else we end up trying to delete a dirent
3583          * from what is now the extent tree root (or a block map).
3584          */
3585         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3586                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3587
3588         old_file_type = old.de->file_type;
3589         if (whiteout) {
3590                 /*
3591                  * Do this before adding a new entry, so the old entry is sure
3592                  * to be still pointing to the valid old entry.
3593                  */
3594                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3595                                      EXT4_FT_CHRDEV);
3596                 if (retval)
3597                         goto end_rename;
3598                 ext4_mark_inode_dirty(handle, whiteout);
3599         }
3600         if (!new.bh) {
3601                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3602                 if (retval)
3603                         goto end_rename;
3604         } else {
3605                 retval = ext4_setent(handle, &new,
3606                                      old.inode->i_ino, old_file_type);
3607                 if (retval)
3608                         goto end_rename;
3609         }
3610         if (force_reread)
3611                 force_reread = !ext4_test_inode_flag(new.dir,
3612                                                      EXT4_INODE_INLINE_DATA);
3613
3614         /*
3615          * Like most other Unix systems, set the ctime for inodes on a
3616          * rename.
3617          */
3618         old.inode->i_ctime = ext4_current_time(old.inode);
3619         ext4_mark_inode_dirty(handle, old.inode);
3620
3621         if (!whiteout) {
3622                 /*
3623                  * ok, that's it
3624                  */
3625                 ext4_rename_delete(handle, &old, force_reread);
3626         }
3627
3628         if (new.inode) {
3629                 ext4_dec_count(handle, new.inode);
3630                 new.inode->i_ctime = ext4_current_time(new.inode);
3631         }
3632         old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3633         ext4_update_dx_flag(old.dir);
3634         if (old.dir_bh) {
3635                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3636                 if (retval)
3637                         goto end_rename;
3638
3639                 ext4_dec_count(handle, old.dir);
3640                 if (new.inode) {
3641                         /* checked ext4_empty_dir above, can't have another
3642                          * parent, ext4_dec_count() won't work for many-linked
3643                          * dirs */
3644                         clear_nlink(new.inode);
3645                 } else {
3646                         ext4_inc_count(handle, new.dir);
3647                         ext4_update_dx_flag(new.dir);
3648                         ext4_mark_inode_dirty(handle, new.dir);
3649                 }
3650         }
3651         ext4_mark_inode_dirty(handle, old.dir);
3652         if (new.inode) {
3653                 ext4_mark_inode_dirty(handle, new.inode);
3654                 if (!new.inode->i_nlink)
3655                         ext4_orphan_add(handle, new.inode);
3656         }
3657         retval = 0;
3658
3659 end_rename:
3660         brelse(old.dir_bh);
3661         brelse(old.bh);
3662         brelse(new.bh);
3663         if (whiteout) {
3664                 if (retval)
3665                         drop_nlink(whiteout);
3666                 unlock_new_inode(whiteout);
3667                 iput(whiteout);
3668         }
3669         if (handle)
3670                 ext4_journal_stop(handle);
3671         return retval;
3672 }
3673
3674 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3675                              struct inode *new_dir, struct dentry *new_dentry)
3676 {
3677         handle_t *handle = NULL;
3678         struct ext4_renament old = {
3679                 .dir = old_dir,
3680                 .dentry = old_dentry,
3681                 .inode = d_inode(old_dentry),
3682         };
3683         struct ext4_renament new = {
3684                 .dir = new_dir,
3685                 .dentry = new_dentry,
3686                 .inode = d_inode(new_dentry),
3687         };
3688         u8 new_file_type;
3689         int retval;
3690
3691         if ((ext4_encrypted_inode(old_dir) ||
3692              ext4_encrypted_inode(new_dir)) &&
3693             (old_dir != new_dir) &&
3694             (!ext4_is_child_context_consistent_with_parent(new_dir,
3695                                                            old.inode) ||
3696              !ext4_is_child_context_consistent_with_parent(old_dir,
3697                                                            new.inode)))
3698                 return -EPERM;
3699
3700         retval = dquot_initialize(old.dir);
3701         if (retval)
3702                 return retval;
3703         retval = dquot_initialize(new.dir);
3704         if (retval)
3705                 return retval;
3706
3707         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3708                                  &old.de, &old.inlined);
3709         if (IS_ERR(old.bh))
3710                 return PTR_ERR(old.bh);
3711         /*
3712          *  Check for inode number is _not_ due to possible IO errors.
3713          *  We might rmdir the source, keep it as pwd of some process
3714          *  and merrily kill the link to whatever was created under the
3715          *  same name. Goodbye sticky bit ;-<
3716          */
3717         retval = -ENOENT;
3718         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3719                 goto end_rename;
3720
3721         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3722                                  &new.de, &new.inlined);
3723         if (IS_ERR(new.bh)) {
3724                 retval = PTR_ERR(new.bh);
3725                 new.bh = NULL;
3726                 goto end_rename;
3727         }
3728
3729         /* RENAME_EXCHANGE case: old *and* new must both exist */
3730         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3731                 goto end_rename;
3732
3733         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3734                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3735                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3736         if (IS_ERR(handle)) {
3737                 retval = PTR_ERR(handle);
3738                 handle = NULL;
3739                 goto end_rename;
3740         }
3741
3742         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3743                 ext4_handle_sync(handle);
3744
3745         if (S_ISDIR(old.inode->i_mode)) {
3746                 old.is_dir = true;
3747                 retval = ext4_rename_dir_prepare(handle, &old);
3748                 if (retval)
3749                         goto end_rename;
3750         }
3751         if (S_ISDIR(new.inode->i_mode)) {
3752                 new.is_dir = true;
3753                 retval = ext4_rename_dir_prepare(handle, &new);
3754                 if (retval)
3755                         goto end_rename;
3756         }
3757
3758         /*
3759          * Other than the special case of overwriting a directory, parents'
3760          * nlink only needs to be modified if this is a cross directory rename.
3761          */
3762         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3763                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3764                 new.dir_nlink_delta = -old.dir_nlink_delta;
3765                 retval = -EMLINK;
3766                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3767                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3768                         goto end_rename;
3769         }
3770
3771         new_file_type = new.de->file_type;
3772         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3773         if (retval)
3774                 goto end_rename;
3775
3776         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3777         if (retval)
3778                 goto end_rename;
3779
3780         /*
3781          * Like most other Unix systems, set the ctime for inodes on a
3782          * rename.
3783          */
3784         old.inode->i_ctime = ext4_current_time(old.inode);
3785         new.inode->i_ctime = ext4_current_time(new.inode);
3786         ext4_mark_inode_dirty(handle, old.inode);
3787         ext4_mark_inode_dirty(handle, new.inode);
3788
3789         if (old.dir_bh) {
3790                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3791                 if (retval)
3792                         goto end_rename;
3793         }
3794         if (new.dir_bh) {
3795                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3796                 if (retval)
3797                         goto end_rename;
3798         }
3799         ext4_update_dir_count(handle, &old);
3800         ext4_update_dir_count(handle, &new);
3801         retval = 0;
3802
3803 end_rename:
3804         brelse(old.dir_bh);
3805         brelse(new.dir_bh);
3806         brelse(old.bh);
3807         brelse(new.bh);
3808         if (handle)
3809                 ext4_journal_stop(handle);
3810         return retval;
3811 }
3812
3813 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3814                         struct inode *new_dir, struct dentry *new_dentry,
3815                         unsigned int flags)
3816 {
3817         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3818                 return -EINVAL;
3819
3820         if (flags & RENAME_EXCHANGE) {
3821                 return ext4_cross_rename(old_dir, old_dentry,
3822                                          new_dir, new_dentry);
3823         }
3824
3825         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3826 }
3827
3828 /*
3829  * directories can handle most operations...
3830  */
3831 const struct inode_operations ext4_dir_inode_operations = {
3832         .create         = ext4_create,
3833         .lookup         = ext4_lookup,
3834         .link           = ext4_link,
3835         .unlink         = ext4_unlink,
3836         .symlink        = ext4_symlink,
3837         .mkdir          = ext4_mkdir,
3838         .rmdir          = ext4_rmdir,
3839         .mknod          = ext4_mknod,
3840         .tmpfile        = ext4_tmpfile,
3841         .rename2        = ext4_rename2,
3842         .setattr        = ext4_setattr,
3843         .setxattr       = generic_setxattr,
3844         .getxattr       = generic_getxattr,
3845         .listxattr      = ext4_listxattr,
3846         .removexattr    = generic_removexattr,
3847         .get_acl        = ext4_get_acl,
3848         .set_acl        = ext4_set_acl,
3849         .fiemap         = ext4_fiemap,
3850 };
3851
3852 const struct inode_operations ext4_special_inode_operations = {
3853         .setattr        = ext4_setattr,
3854         .setxattr       = generic_setxattr,
3855         .getxattr       = generic_getxattr,
3856         .listxattr      = ext4_listxattr,
3857         .removexattr    = generic_removexattr,
3858         .get_acl        = ext4_get_acl,
3859         .set_acl        = ext4_set_acl,
3860 };