fs/ext4/namei.c

   1 /*
   2  *  linux/fs/ext4/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o ([email protected]), 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 inode *dir, 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 0;
 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 0;
 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) & 0x0fffffff;
 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(KERN_CONT " %x->%lu",
 583                        i ? dx_get_hash(entries + i) : 0,
 584                        (unsigned long)dx_get_block(entries + i));
 585         }
 586         printk(KERN_CONT "\n");
 587 }
 588
 589 struct stats
 590 {
 591         unsigned names;
 592         unsigned space;
 593         unsigned bcount;
 594 };
 595
 596 static struct stats dx_show_leaf(struct inode *dir,
 597                                 struct dx_hash_info *hinfo,
 598                                 struct ext4_dir_entry_2 *de,
 599                                 int size, int show_names)
 600 {
 601         unsigned names = 0, space = 0;
 602         char *base = (char *) de;
 603         struct dx_hash_info h = *hinfo;
 604
 605         printk("names: ");
 606         while ((char *) de < base + size)
 607         {
 608                 if (de->inode)
 609                 {
 610                         if (show_names)
 611                         {
 612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 613                                 int len;
 614                                 char *name;
 615                                 struct fscrypt_str fname_crypto_str =
 616                                         FSTR_INIT(NULL, 0);
 617                                 int res = 0;
 618
 619                                 name  = de->name;
 620                                 len = de->name_len;
 621                                 if (ext4_encrypted_inode(dir))
 622                                         res = fscrypt_get_encryption_info(dir);
 623                                 if (res) {
 624                                         printk(KERN_WARNING "Error setting up"
 625                                                " fname crypto: %d\n", res);
 626                                 }
 627                                 if (!fscrypt_has_encryption_key(dir)) {
 628                                         /* Directory is not encrypted */
 629                                         ext4fs_dirhash(de->name,
 630                                                 de->name_len, &h);
 631                                         printk("%*.s:(U)%x.%u ", len,
 632                                                name, h.hash,
 633                                                (unsigned) ((char *) de
 634                                                            - base));
 635                                 } else {
 636                                         struct fscrypt_str de_name =
 637                                                 FSTR_INIT(name, len);
 638
 639                                         /* Directory is encrypted */
 640                                         res = fscrypt_fname_alloc_buffer(
 641                                                 dir, len,
 642                                                 &fname_crypto_str);
 643                                         if (res)
 644                                                 printk(KERN_WARNING "Error "
 645                                                         "allocating crypto "
 646                                                         "buffer--skipping "
 647                                                         "crypto\n");
 648                                         res = fscrypt_fname_disk_to_usr(dir,
 649                                                 0, 0, &de_name,
 650                                                 &fname_crypto_str);
 651                                         if (res) {
 652                                                 printk(KERN_WARNING "Error "
 653                                                         "converting filename "
 654                                                         "from disk to usr"
 655                                                         "\n");
 656                                                 name = "??";
 657                                                 len = 2;
 658                                         } else {
 659                                                 name = fname_crypto_str.name;
 660                                                 len = fname_crypto_str.len;
 661                                         }
 662                                         ext4fs_dirhash(de->name, de->name_len,
 663                                                        &h);
 664                                         printk("%*.s:(E)%x.%u ", len, name,
 665                                                h.hash, (unsigned) ((char *) de
 666                                                                    - base));
 667                                         fscrypt_fname_free_buffer(
 668                                                         &fname_crypto_str);
 669                                 }
 670 #else
 671                                 int len = de->name_len;
 672                                 char *name = de->name;
 673                                 ext4fs_dirhash(de->name, de->name_len, &h);
 674                                 printk("%*.s:%x.%u ", len, name, h.hash,
 675                                        (unsigned) ((char *) de - base));
 676 #endif
 677                         }
 678                         space += EXT4_DIR_REC_LEN(de->name_len);
 679                         names++;
 680                 }
 681                 de = ext4_next_entry(de, size);
 682         }
 683         printk(KERN_CONT "(%i)\n", names);
 684         return (struct stats) { names, space, 1 };
 685 }
 686
 687 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 688                              struct dx_entry *entries, int levels)
 689 {
 690         unsigned blocksize = dir->i_sb->s_blocksize;
 691         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 692         unsigned bcount = 0;
 693         struct buffer_head *bh;
 694         printk("%i indexed blocks...\n", count);
 695         for (i = 0; i < count; i++, entries++)
 696         {
 697                 ext4_lblk_t block = dx_get_block(entries);
 698                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 699                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 700                 struct stats stats;
 701                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 702                 bh = ext4_bread(NULL,dir, block, 0);
 703                 if (!bh || IS_ERR(bh))
 704                         continue;
 705                 stats = levels?
 706                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 707                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 708                         bh->b_data, blocksize, 0);
 709                 names += stats.names;
 710                 space += stats.space;
 711                 bcount += stats.bcount;
 712                 brelse(bh);
 713         }
 714         if (bcount)
 715                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 716                        levels ? "" : "   ", names, space/bcount,
 717                        (space/bcount)*100/blocksize);
 718         return (struct stats) { names, space, bcount};
 719 }
 720 #endif /* DX_DEBUG */
 721
 722 /*
 723  * Probe for a directory leaf block to search.
 724  *
 725  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 726  * error in the directory index, and the caller should fall back to
 727  * searching the directory normally.  The callers of dx_probe **MUST**
 728  * check for this error code, and make sure it never gets reflected
 729  * back to userspace.
 730  */
 731 static struct dx_frame *
 732 dx_probe(struct ext4_filename *fname, struct inode *dir,
 733          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 734 {
 735         unsigned count, indirect;
 736         struct dx_entry *at, *entries, *p, *q, *m;
 737         struct dx_root *root;
 738         struct dx_frame *frame = frame_in;
 739         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 740         u32 hash;
 741
 742         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 743         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 744         if (IS_ERR(frame->bh))
 745                 return (struct dx_frame *) frame->bh;
 746
 747         root = (struct dx_root *) frame->bh->b_data;
 748         if (root->info.hash_version != DX_HASH_TEA &&
 749             root->info.hash_version != DX_HASH_HALF_MD4 &&
 750             root->info.hash_version != DX_HASH_LEGACY) {
 751                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 752                                    root->info.hash_version);
 753                 goto fail;
 754         }
 755         if (fname)
 756                 hinfo = &fname->hinfo;
 757         hinfo->hash_version = root->info.hash_version;
 758         if (hinfo->hash_version <= DX_HASH_TEA)
 759                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 760         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 761         if (fname && fname_name(fname))
 762                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 763         hash = hinfo->hash;
 764
 765         if (root->info.unused_flags & 1) {
 766                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 767                                    root->info.unused_flags);
 768                 goto fail;
 769         }
 770
 771         indirect = root->info.indirect_levels;
 772         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 773                 ext4_warning(dir->i_sb,
 774                              "Directory (ino: %lu) htree depth %#06x exceed"
 775                              "supported value", dir->i_ino,
 776                              ext4_dir_htree_level(dir->i_sb));
 777                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 778                         ext4_warning(dir->i_sb, "Enable large directory "
 779                                                 "feature to access it");
 780                 }
 781                 goto fail;
 782         }
 783
 784         entries = (struct dx_entry *)(((char *)&root->info) +
 785                                       root->info.info_length);
 786
 787         if (dx_get_limit(entries) != dx_root_limit(dir,
 788                                                    root->info.info_length)) {
 789                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 790                                    dx_get_limit(entries),
 791                                    dx_root_limit(dir, root->info.info_length));
 792                 goto fail;
 793         }
 794
 795         dxtrace(printk("Look up %x", hash));
 796         while (1) {
 797                 count = dx_get_count(entries);
 798                 if (!count || count > dx_get_limit(entries)) {
 799                         ext4_warning_inode(dir,
 800                                            "dx entry: count %u beyond limit %u",
 801                                            count, dx_get_limit(entries));
 802                         goto fail;
 803                 }
 804
 805                 p = entries + 1;
 806                 q = entries + count - 1;
 807                 while (p <= q) {
 808                         m = p + (q - p) / 2;
 809                         dxtrace(printk(KERN_CONT "."));
 810                         if (dx_get_hash(m) > hash)
 811                                 q = m - 1;
 812                         else
 813                                 p = m + 1;
 814                 }
 815
 816                 if (0) { // linear search cross check
 817                         unsigned n = count - 1;
 818                         at = entries;
 819                         while (n--)
 820                         {
 821                                 dxtrace(printk(KERN_CONT ","));
 822                                 if (dx_get_hash(++at) > hash)
 823                                 {
 824                                         at--;
 825                                         break;
 826                                 }
 827                         }
 828                         assert (at == p - 1);
 829                 }
 830
 831                 at = p - 1;
 832                 dxtrace(printk(KERN_CONT " %x->%u\n",
 833                                at == entries ? 0 : dx_get_hash(at),
 834                                dx_get_block(at)));
 835                 frame->entries = entries;
 836                 frame->at = at;
 837                 if (!indirect--)
 838                         return frame;
 839                 frame++;
 840                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 841                 if (IS_ERR(frame->bh)) {
 842                         ret_err = (struct dx_frame *) frame->bh;
 843                         frame->bh = NULL;
 844                         goto fail;
 845                 }
 846                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 847
 848                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 849                         ext4_warning_inode(dir,
 850                                 "dx entry: limit %u != node limit %u",
 851                                 dx_get_limit(entries), dx_node_limit(dir));
 852                         goto fail;
 853                 }
 854         }
 855 fail:
 856         while (frame >= frame_in) {
 857                 brelse(frame->bh);
 858                 frame--;
 859         }
 860
 861         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 862                 ext4_warning_inode(dir,
 863                         "Corrupt directory, running e2fsck is recommended");
 864         return ret_err;
 865 }
 866
 867 static void dx_release(struct dx_frame *frames)
 868 {
 869         struct dx_root_info *info;
 870         int i;
 871
 872         if (frames[0].bh == NULL)
 873                 return;
 874
 875         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 876         for (i = 0; i <= info->indirect_levels; i++) {
 877                 if (frames[i].bh == NULL)
 878                         break;
 879                 brelse(frames[i].bh);
 880                 frames[i].bh = NULL;
 881         }
 882 }
 883
 884 /*
 885  * This function increments the frame pointer to search the next leaf
 886  * block, and reads in the necessary intervening nodes if the search
 887  * should be necessary.  Whether or not the search is necessary is
 888  * controlled by the hash parameter.  If the hash value is even, then
 889  * the search is only continued if the next block starts with that
 890  * hash value.  This is used if we are searching for a specific file.
 891  *
 892  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 893  *
 894  * This function returns 1 if the caller should continue to search,
 895  * or 0 if it should not.  If there is an error reading one of the
 896  * index blocks, it will a negative error code.
 897  *
 898  * If start_hash is non-null, it will be filled in with the starting
 899  * hash of the next page.
 900  */
 901 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 902                                  struct dx_frame *frame,
 903                                  struct dx_frame *frames,
 904                                  __u32 *start_hash)
 905 {
 906         struct dx_frame *p;
 907         struct buffer_head *bh;
 908         int num_frames = 0;
 909         __u32 bhash;
 910
 911         p = frame;
 912         /*
 913          * Find the next leaf page by incrementing the frame pointer.
 914          * If we run out of entries in the interior node, loop around and
 915          * increment pointer in the parent node.  When we break out of
 916          * this loop, num_frames indicates the number of interior
 917          * nodes need to be read.
 918          */
 919         while (1) {
 920                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 921                         break;
 922                 if (p == frames)
 923                         return 0;
 924                 num_frames++;
 925                 p--;
 926         }
 927
 928         /*
 929          * If the hash is 1, then continue only if the next page has a
 930          * continuation hash of any value.  This is used for readdir
 931          * handling.  Otherwise, check to see if the hash matches the
 932          * desired contiuation hash.  If it doesn't, return since
 933          * there's no point to read in the successive index pages.
 934          */
 935         bhash = dx_get_hash(p->at);
 936         if (start_hash)
 937                 *start_hash = bhash;
 938         if ((hash & 1) == 0) {
 939                 if ((bhash & ~1) != hash)
 940                         return 0;
 941         }
 942         /*
 943          * If the hash is HASH_NB_ALWAYS, we always go to the next
 944          * block so no check is necessary
 945          */
 946         while (num_frames--) {
 947                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 948                 if (IS_ERR(bh))
 949                         return PTR_ERR(bh);
 950                 p++;
 951                 brelse(p->bh);
 952                 p->bh = bh;
 953                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 954         }
 955         return 1;
 956 }
 957
 958
 959 /*
 960  * This function fills a red-black tree with information from a
 961  * directory block.  It returns the number directory entries loaded
 962  * into the tree.  If there is an error it is returned in err.
 963  */
 964 static int htree_dirblock_to_tree(struct file *dir_file,
 965                                   struct inode *dir, ext4_lblk_t block,
 966                                   struct dx_hash_info *hinfo,
 967                                   __u32 start_hash, __u32 start_minor_hash)
 968 {
 969         struct buffer_head *bh;
 970         struct ext4_dir_entry_2 *de, *top;
 971         int err = 0, count = 0;
 972         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 973
 974         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 975                                                         (unsigned long)block));
 976         bh = ext4_read_dirblock(dir, block, DIRENT);
 977         if (IS_ERR(bh))
 978                 return PTR_ERR(bh);
 979
 980         de = (struct ext4_dir_entry_2 *) bh->b_data;
 981         top = (struct ext4_dir_entry_2 *) ((char *) de +
 982                                            dir->i_sb->s_blocksize -
 983                                            EXT4_DIR_REC_LEN(0));
 984 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 985         /* Check if the directory is encrypted */
 986         if (ext4_encrypted_inode(dir)) {
 987                 err = fscrypt_get_encryption_info(dir);
 988                 if (err < 0) {
 989                         brelse(bh);
 990                         return err;
 991                 }
 992                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
 993                                                      &fname_crypto_str);
 994                 if (err < 0) {
 995                         brelse(bh);
 996                         return err;
 997                 }
 998         }
 999 #endif
1000         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1001                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1002                                 bh->b_data, bh->b_size,
1003                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1004                                          + ((char *)de - bh->b_data))) {
1005                         /* silently ignore the rest of the block */
1006                         break;
1007                 }
1008                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1009                 if ((hinfo->hash < start_hash) ||
1010                     ((hinfo->hash == start_hash) &&
1011                      (hinfo->minor_hash < start_minor_hash)))
1012                         continue;
1013                 if (de->inode == 0)
1014                         continue;
1015                 if (!ext4_encrypted_inode(dir)) {
1016                         tmp_str.name = de->name;
1017                         tmp_str.len = de->name_len;
1018                         err = ext4_htree_store_dirent(dir_file,
1019                                    hinfo->hash, hinfo->minor_hash, de,
1020                                    &tmp_str);
1021                 } else {
1022                         int save_len = fname_crypto_str.len;
1023                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1024                                                                 de->name_len);
1025
1026                         /* Directory is encrypted */
1027                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1028                                         hinfo->minor_hash, &de_name,
1029                                         &fname_crypto_str);
1030                         if (err) {
1031                                 count = err;
1032                                 goto errout;
1033                         }
1034                         err = ext4_htree_store_dirent(dir_file,
1035                                    hinfo->hash, hinfo->minor_hash, de,
1036                                         &fname_crypto_str);
1037                         fname_crypto_str.len = save_len;
1038                 }
1039                 if (err != 0) {
1040                         count = err;
1041                         goto errout;
1042                 }
1043                 count++;
1044         }
1045 errout:
1046         brelse(bh);
1047 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1048         fscrypt_fname_free_buffer(&fname_crypto_str);
1049 #endif
1050         return count;
1051 }
1052
1053
1054 /*
1055  * This function fills a red-black tree with information from a
1056  * directory.  We start scanning the directory in hash order, starting
1057  * at start_hash and start_minor_hash.
1058  *
1059  * This function returns the number of entries inserted into the tree,
1060  * or a negative error code.
1061  */
1062 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1063                          __u32 start_minor_hash, __u32 *next_hash)
1064 {
1065         struct dx_hash_info hinfo;
1066         struct ext4_dir_entry_2 *de;
1067         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1068         struct inode *dir;
1069         ext4_lblk_t block;
1070         int count = 0;
1071         int ret, err;
1072         __u32 hashval;
1073         struct fscrypt_str tmp_str;
1074
1075         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1076                        start_hash, start_minor_hash));
1077         dir = file_inode(dir_file);
1078         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1079                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1080                 if (hinfo.hash_version <= DX_HASH_TEA)
1081                         hinfo.hash_version +=
1082                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1083                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1084                 if (ext4_has_inline_data(dir)) {
1085                         int has_inline_data = 1;
1086                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1087                                                         &hinfo, start_hash,
1088                                                         start_minor_hash,
1089                                                         &has_inline_data);
1090                         if (has_inline_data) {
1091                                 *next_hash = ~0;
1092                                 return count;
1093                         }
1094                 }
1095                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1096                                                start_hash, start_minor_hash);
1097                 *next_hash = ~0;
1098                 return count;
1099         }
1100         hinfo.hash = start_hash;
1101         hinfo.minor_hash = 0;
1102         frame = dx_probe(NULL, dir, &hinfo, frames);
1103         if (IS_ERR(frame))
1104                 return PTR_ERR(frame);
1105
1106         /* Add '.' and '..' from the htree header */
1107         if (!start_hash && !start_minor_hash) {
1108                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1109                 tmp_str.name = de->name;
1110                 tmp_str.len = de->name_len;
1111                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1112                                               de, &tmp_str);
1113                 if (err != 0)
1114                         goto errout;
1115                 count++;
1116         }
1117         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1118                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1119                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1120                 tmp_str.name = de->name;
1121                 tmp_str.len = de->name_len;
1122                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1123                                               de, &tmp_str);
1124                 if (err != 0)
1125                         goto errout;
1126                 count++;
1127         }
1128
1129         while (1) {
1130                 if (fatal_signal_pending(current)) {
1131                         err = -ERESTARTSYS;
1132                         goto errout;
1133                 }
1134                 cond_resched();
1135                 block = dx_get_block(frame->at);
1136                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1137                                              start_hash, start_minor_hash);
1138                 if (ret < 0) {
1139                         err = ret;
1140                         goto errout;
1141                 }
1142                 count += ret;
1143                 hashval = ~0;
1144                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1145                                             frame, frames, &hashval);
1146                 *next_hash = hashval;
1147                 if (ret < 0) {
1148                         err = ret;
1149                         goto errout;
1150                 }
1151                 /*
1152                  * Stop if:  (a) there are no more entries, or
1153                  * (b) we have inserted at least one entry and the
1154                  * next hash value is not a continuation
1155                  */
1156                 if ((ret == 0) ||
1157                     (count && ((hashval & 1) == 0)))
1158                         break;
1159         }
1160         dx_release(frames);
1161         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1162                        "next hash: %x\n", count, *next_hash));
1163         return count;
1164 errout:
1165         dx_release(frames);
1166         return (err);
1167 }
1168
1169 static inline int search_dirblock(struct buffer_head *bh,
1170                                   struct inode *dir,
1171                                   struct ext4_filename *fname,
1172                                   unsigned int offset,
1173                                   struct ext4_dir_entry_2 **res_dir)
1174 {
1175         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1176                                fname, offset, res_dir);
1177 }
1178
1179 /*
1180  * Directory block splitting, compacting
1181  */
1182
1183 /*
1184  * Create map of hash values, offsets, and sizes, stored at end of block.
1185  * Returns number of entries mapped.
1186  */
1187 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1188                        unsigned blocksize, struct dx_hash_info *hinfo,
1189                        struct dx_map_entry *map_tail)
1190 {
1191         int count = 0;
1192         char *base = (char *) de;
1193         struct dx_hash_info h = *hinfo;
1194
1195         while ((char *) de < base + blocksize) {
1196                 if (de->name_len && de->inode) {
1197                         ext4fs_dirhash(de->name, de->name_len, &h);
1198                         map_tail--;
1199                         map_tail->hash = h.hash;
1200                         map_tail->offs = ((char *) de - base)>>2;
1201                         map_tail->size = le16_to_cpu(de->rec_len);
1202                         count++;
1203                         cond_resched();
1204                 }
1205                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1206                 de = ext4_next_entry(de, blocksize);
1207         }
1208         return count;
1209 }
1210
1211 /* Sort map by hash value */
1212 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1213 {
1214         struct dx_map_entry *p, *q, *top = map + count - 1;
1215         int more;
1216         /* Combsort until bubble sort doesn't suck */
1217         while (count > 2) {
1218                 count = count*10/13;
1219                 if (count - 9 < 2) /* 9, 10 -> 11 */
1220                         count = 11;
1221                 for (p = top, q = p - count; q >= map; p--, q--)
1222                         if (p->hash < q->hash)
1223                                 swap(*p, *q);
1224         }
1225         /* Garden variety bubble sort */
1226         do {
1227                 more = 0;
1228                 q = top;
1229                 while (q-- > map) {
1230                         if (q[1].hash >= q[0].hash)
1231                                 continue;
1232                         swap(*(q+1), *q);
1233                         more = 1;
1234                 }
1235         } while(more);
1236 }
1237
1238 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1239 {
1240         struct dx_entry *entries = frame->entries;
1241         struct dx_entry *old = frame->at, *new = old + 1;
1242         int count = dx_get_count(entries);
1243
1244         assert(count < dx_get_limit(entries));
1245         assert(old < entries + count);
1246         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1247         dx_set_hash(new, hash);
1248         dx_set_block(new, block);
1249         dx_set_count(entries, count + 1);
1250 }
1251
1252 /*
1253  * Test whether a directory entry matches the filename being searched for.
1254  *
1255  * Return: %true if the directory entry matches, otherwise %false.
1256  */
1257 static inline bool ext4_match(const struct ext4_filename *fname,
1258                               const struct ext4_dir_entry_2 *de)
1259 {
1260         struct fscrypt_name f;
1261
1262         if (!de->inode)
1263                 return false;
1264
1265         f.usr_fname = fname->usr_fname;
1266         f.disk_name = fname->disk_name;
1267 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1268         f.crypto_buf = fname->crypto_buf;
1269 #endif
1270         return fscrypt_match_name(&f, de->name, de->name_len);
1271 }
1272
1273 /*
1274  * Returns 0 if not found, -1 on failure, and 1 on success
1275  */
1276 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277                     struct inode *dir, struct ext4_filename *fname,
1278                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1279 {
1280         struct ext4_dir_entry_2 * de;
1281         char * dlimit;
1282         int de_len;
1283
1284         de = (struct ext4_dir_entry_2 *)search_buf;
1285         dlimit = search_buf + buf_size;
1286         while ((char *) de < dlimit) {
1287                 /* this code is executed quadratically often */
1288                 /* do minimal checking `by hand' */
1289                 if ((char *) de + de->name_len <= dlimit &&
1290                     ext4_match(fname, de)) {
1291                         /* found a match - just to be sure, do
1292                          * a full check */
1293                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1294                                                  bh->b_size, offset))
1295                                 return -1;
1296                         *res_dir = de;
1297                         return 1;
1298                 }
1299                 /* prevent looping on a bad block */
1300                 de_len = ext4_rec_len_from_disk(de->rec_len,
1301                                                 dir->i_sb->s_blocksize);
1302                 if (de_len <= 0)
1303                         return -1;
1304                 offset += de_len;
1305                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1306         }
1307         return 0;
1308 }
1309
1310 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1311                                struct ext4_dir_entry *de)
1312 {
1313         struct super_block *sb = dir->i_sb;
1314
1315         if (!is_dx(dir))
1316                 return 0;
1317         if (block == 0)
1318                 return 1;
1319         if (de->inode == 0 &&
1320             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1321                         sb->s_blocksize)
1322                 return 1;
1323         return 0;
1324 }
1325
1326 /*
1327  *      ext4_find_entry()
1328  *
1329  * finds an entry in the specified directory with the wanted name. It
1330  * returns the cache buffer in which the entry was found, and the entry
1331  * itself (as a parameter - res_dir). It does NOT read the inode of the
1332  * entry - you'll have to do that yourself if you want to.
1333  *
1334  * The returned buffer_head has ->b_count elevated.  The caller is expected
1335  * to brelse() it when appropriate.
1336  */
1337 static struct buffer_head * ext4_find_entry (struct inode *dir,
1338                                         const struct qstr *d_name,
1339                                         struct ext4_dir_entry_2 **res_dir,
1340                                         int *inlined)
1341 {
1342         struct super_block *sb;
1343         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1344         struct buffer_head *bh, *ret = NULL;
1345         ext4_lblk_t start, block;
1346         const u8 *name = d_name->name;
1347         size_t ra_max = 0;      /* Number of bh's in the readahead
1348                                    buffer, bh_use[] */
1349         size_t ra_ptr = 0;      /* Current index into readahead
1350                                    buffer */
1351         ext4_lblk_t  nblocks;
1352         int i, namelen, retval;
1353         struct ext4_filename fname;
1354
1355         *res_dir = NULL;
1356         sb = dir->i_sb;
1357         namelen = d_name->len;
1358         if (namelen > EXT4_NAME_LEN)
1359                 return NULL;
1360
1361         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1362         if (retval == -ENOENT)
1363                 return NULL;
1364         if (retval)
1365                 return ERR_PTR(retval);
1366
1367         if (ext4_has_inline_data(dir)) {
1368                 int has_inline_data = 1;
1369                 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1370                                              &has_inline_data);
1371                 if (has_inline_data) {
1372                         if (inlined)
1373                                 *inlined = 1;
1374                         goto cleanup_and_exit;
1375                 }
1376         }
1377
1378         if ((namelen <= 2) && (name[0] == '.') &&
1379             (name[1] == '.' || name[1] == '\0')) {
1380                 /*
1381                  * "." or ".." will only be in the first block
1382                  * NFS may look up ".."; "." should be handled by the VFS
1383                  */
1384                 block = start = 0;
1385                 nblocks = 1;
1386                 goto restart;
1387         }
1388         if (is_dx(dir)) {
1389                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1390                 /*
1391                  * On success, or if the error was file not found,
1392                  * return.  Otherwise, fall back to doing a search the
1393                  * old fashioned way.
1394                  */
1395                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1396                         goto cleanup_and_exit;
1397                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1398                                "falling back\n"));
1399         }
1400         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1401         start = EXT4_I(dir)->i_dir_start_lookup;
1402         if (start >= nblocks)
1403                 start = 0;
1404         block = start;
1405 restart:
1406         do {
1407                 /*
1408                  * We deal with the read-ahead logic here.
1409                  */
1410                 if (ra_ptr >= ra_max) {
1411                         /* Refill the readahead buffer */
1412                         ra_ptr = 0;
1413                         if (block < start)
1414                                 ra_max = start - block;
1415                         else
1416                                 ra_max = nblocks - block;
1417                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1418                         retval = ext4_bread_batch(dir, block, ra_max,
1419                                                   false /* wait */, bh_use);
1420                         if (retval) {
1421                                 ret = ERR_PTR(retval);
1422                                 ra_max = 0;
1423                                 goto cleanup_and_exit;
1424                         }
1425                 }
1426                 if ((bh = bh_use[ra_ptr++]) == NULL)
1427                         goto next;
1428                 wait_on_buffer(bh);
1429                 if (!buffer_uptodate(bh)) {
1430                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1431                                          (unsigned long) block);
1432                         brelse(bh);
1433                         ret = ERR_PTR(-EIO);
1434                         goto cleanup_and_exit;
1435                 }
1436                 if (!buffer_verified(bh) &&
1437                     !is_dx_internal_node(dir, block,
1438                                          (struct ext4_dir_entry *)bh->b_data) &&
1439                     !ext4_dirent_csum_verify(dir,
1440                                 (struct ext4_dir_entry *)bh->b_data)) {
1441                         EXT4_ERROR_INODE(dir, "checksumming directory "
1442                                          "block %lu", (unsigned long)block);
1443                         brelse(bh);
1444                         ret = ERR_PTR(-EFSBADCRC);
1445                         goto cleanup_and_exit;
1446                 }
1447                 set_buffer_verified(bh);
1448                 i = search_dirblock(bh, dir, &fname,
1449                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1450                 if (i == 1) {
1451                         EXT4_I(dir)->i_dir_start_lookup = block;
1452                         ret = bh;
1453                         goto cleanup_and_exit;
1454                 } else {
1455                         brelse(bh);
1456                         if (i < 0)
1457                                 goto cleanup_and_exit;
1458                 }
1459         next:
1460                 if (++block >= nblocks)
1461                         block = 0;
1462         } while (block != start);
1463
1464         /*
1465          * If the directory has grown while we were searching, then
1466          * search the last part of the directory before giving up.
1467          */
1468         block = nblocks;
1469         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1470         if (block < nblocks) {
1471                 start = 0;
1472                 goto restart;
1473         }
1474
1475 cleanup_and_exit:
1476         /* Clean up the read-ahead blocks */
1477         for (; ra_ptr < ra_max; ra_ptr++)
1478                 brelse(bh_use[ra_ptr]);
1479         ext4_fname_free_filename(&fname);
1480         return ret;
1481 }
1482
1483 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1484                         struct ext4_filename *fname,
1485                         struct ext4_dir_entry_2 **res_dir)
1486 {
1487         struct super_block * sb = dir->i_sb;
1488         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1489         struct buffer_head *bh;
1490         ext4_lblk_t block;
1491         int retval;
1492
1493 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1494         *res_dir = NULL;
1495 #endif
1496         frame = dx_probe(fname, dir, NULL, frames);
1497         if (IS_ERR(frame))
1498                 return (struct buffer_head *) frame;
1499         do {
1500                 block = dx_get_block(frame->at);
1501                 bh = ext4_read_dirblock(dir, block, DIRENT);
1502                 if (IS_ERR(bh))
1503                         goto errout;
1504
1505                 retval = search_dirblock(bh, dir, fname,
1506                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1507                                          res_dir);
1508                 if (retval == 1)
1509                         goto success;
1510                 brelse(bh);
1511                 if (retval == -1) {
1512                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1513                         goto errout;
1514                 }
1515
1516                 /* Check to see if we should continue to search */
1517                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1518                                                frames, NULL);
1519                 if (retval < 0) {
1520                         ext4_warning_inode(dir,
1521                                 "error %d reading directory index block",
1522                                 retval);
1523                         bh = ERR_PTR(retval);
1524                         goto errout;
1525                 }
1526         } while (retval == 1);
1527
1528         bh = NULL;
1529 errout:
1530         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1531 success:
1532         dx_release(frames);
1533         return bh;
1534 }
1535
1536 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1537 {
1538         struct inode *inode;
1539         struct ext4_dir_entry_2 *de;
1540         struct buffer_head *bh;
1541
1542         if (ext4_encrypted_inode(dir)) {
1543                 int res = fscrypt_get_encryption_info(dir);
1544
1545                 /*
1546                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1547                  * created while the directory was encrypted and we
1548                  * have access to the key.
1549                  */
1550                 if (fscrypt_has_encryption_key(dir))
1551                         fscrypt_set_encrypted_dentry(dentry);
1552                 fscrypt_set_d_op(dentry);
1553                 if (res && res != -ENOKEY)
1554                         return ERR_PTR(res);
1555         }
1556
1557        if (dentry->d_name.len > EXT4_NAME_LEN)
1558                return ERR_PTR(-ENAMETOOLONG);
1559
1560         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1561         if (IS_ERR(bh))
1562                 return (struct dentry *) bh;
1563         inode = NULL;
1564         if (bh) {
1565                 __u32 ino = le32_to_cpu(de->inode);
1566                 brelse(bh);
1567                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1568                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1569                         return ERR_PTR(-EFSCORRUPTED);
1570                 }
1571                 if (unlikely(ino == dir->i_ino)) {
1572                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1573                                          dentry);
1574                         return ERR_PTR(-EFSCORRUPTED);
1575                 }
1576                 inode = ext4_iget_normal(dir->i_sb, ino);
1577                 if (inode == ERR_PTR(-ESTALE)) {
1578                         EXT4_ERROR_INODE(dir,
1579                                          "deleted inode referenced: %u",
1580                                          ino);
1581                         return ERR_PTR(-EFSCORRUPTED);
1582                 }
1583                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1584                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1585                     !fscrypt_has_permitted_context(dir, inode)) {
1586                         ext4_warning(inode->i_sb,
1587                                      "Inconsistent encryption contexts: %lu/%lu",
1588                                      dir->i_ino, inode->i_ino);
1589                         iput(inode);
1590                         return ERR_PTR(-EPERM);
1591                 }
1592         }
1593         return d_splice_alias(inode, dentry);
1594 }
1595
1596
1597 struct dentry *ext4_get_parent(struct dentry *child)
1598 {
1599         __u32 ino;
1600         static const struct qstr dotdot = QSTR_INIT("..", 2);
1601         struct ext4_dir_entry_2 * de;
1602         struct buffer_head *bh;
1603
1604         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1605         if (IS_ERR(bh))
1606                 return (struct dentry *) bh;
1607         if (!bh)
1608                 return ERR_PTR(-ENOENT);
1609         ino = le32_to_cpu(de->inode);
1610         brelse(bh);
1611
1612         if (!ext4_valid_inum(child->d_sb, ino)) {
1613                 EXT4_ERROR_INODE(d_inode(child),
1614                                  "bad parent inode number: %u", ino);
1615                 return ERR_PTR(-EFSCORRUPTED);
1616         }
1617
1618         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1619 }
1620
1621 /*
1622  * Move count entries from end of map between two memory locations.
1623  * Returns pointer to last entry moved.
1624  */
1625 static struct ext4_dir_entry_2 *
1626 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1627                 unsigned blocksize)
1628 {
1629         unsigned rec_len = 0;
1630
1631         while (count--) {
1632                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1633                                                 (from + (map->offs<<2));
1634                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1635                 memcpy (to, de, rec_len);
1636                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1637                                 ext4_rec_len_to_disk(rec_len, blocksize);
1638                 de->inode = 0;
1639                 map++;
1640                 to += rec_len;
1641         }
1642         return (struct ext4_dir_entry_2 *) (to - rec_len);
1643 }
1644
1645 /*
1646  * Compact each dir entry in the range to the minimal rec_len.
1647  * Returns pointer to last entry in range.
1648  */
1649 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1650 {
1651         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1652         unsigned rec_len = 0;
1653
1654         prev = to = de;
1655         while ((char*)de < base + blocksize) {
1656                 next = ext4_next_entry(de, blocksize);
1657                 if (de->inode && de->name_len) {
1658                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1659                         if (de > to)
1660                                 memmove(to, de, rec_len);
1661                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1662                         prev = to;
1663                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1664                 }
1665                 de = next;
1666         }
1667         return prev;
1668 }
1669
1670 /*
1671  * Split a full leaf block to make room for a new dir entry.
1672  * Allocate a new block, and move entries so that they are approx. equally full.
1673  * Returns pointer to de in block into which the new entry will be inserted.
1674  */
1675 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1676                         struct buffer_head **bh,struct dx_frame *frame,
1677                         struct dx_hash_info *hinfo)
1678 {
1679         unsigned blocksize = dir->i_sb->s_blocksize;
1680         unsigned count, continued;
1681         struct buffer_head *bh2;
1682         ext4_lblk_t newblock;
1683         u32 hash2;
1684         struct dx_map_entry *map;
1685         char *data1 = (*bh)->b_data, *data2;
1686         unsigned split, move, size;
1687         struct ext4_dir_entry_2 *de = NULL, *de2;
1688         struct ext4_dir_entry_tail *t;
1689         int     csum_size = 0;
1690         int     err = 0, i;
1691
1692         if (ext4_has_metadata_csum(dir->i_sb))
1693                 csum_size = sizeof(struct ext4_dir_entry_tail);
1694
1695         bh2 = ext4_append(handle, dir, &newblock);
1696         if (IS_ERR(bh2)) {
1697                 brelse(*bh);
1698                 *bh = NULL;
1699                 return (struct ext4_dir_entry_2 *) bh2;
1700         }
1701
1702         BUFFER_TRACE(*bh, "get_write_access");
1703         err = ext4_journal_get_write_access(handle, *bh);
1704         if (err)
1705                 goto journal_error;
1706
1707         BUFFER_TRACE(frame->bh, "get_write_access");
1708         err = ext4_journal_get_write_access(handle, frame->bh);
1709         if (err)
1710                 goto journal_error;
1711
1712         data2 = bh2->b_data;
1713
1714         /* create map in the end of data2 block */
1715         map = (struct dx_map_entry *) (data2 + blocksize);
1716         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1717                              blocksize, hinfo, map);
1718         map -= count;
1719         dx_sort_map(map, count);
1720         /* Split the existing block in the middle, size-wise */
1721         size = 0;
1722         move = 0;
1723         for (i = count-1; i >= 0; i--) {
1724                 /* is more than half of this entry in 2nd half of the block? */
1725                 if (size + map[i].size/2 > blocksize/2)
1726                         break;
1727                 size += map[i].size;
1728                 move++;
1729         }
1730         /* map index at which we will split */
1731         split = count - move;
1732         hash2 = map[split].hash;
1733         continued = hash2 == map[split - 1].hash;
1734         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1735                         (unsigned long)dx_get_block(frame->at),
1736                                         hash2, split, count-split));
1737
1738         /* Fancy dance to stay within two buffers */
1739         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1740                               blocksize);
1741         de = dx_pack_dirents(data1, blocksize);
1742         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1743                                            (char *) de,
1744                                            blocksize);
1745         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1746                                             (char *) de2,
1747                                             blocksize);
1748         if (csum_size) {
1749                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1750                 initialize_dirent_tail(t, blocksize);
1751
1752                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1753                 initialize_dirent_tail(t, blocksize);
1754         }
1755
1756         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1757                         blocksize, 1));
1758         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1759                         blocksize, 1));
1760
1761         /* Which block gets the new entry? */
1762         if (hinfo->hash >= hash2) {
1763                 swap(*bh, bh2);
1764                 de = de2;
1765         }
1766         dx_insert_block(frame, hash2 + continued, newblock);
1767         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1768         if (err)
1769                 goto journal_error;
1770         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1771         if (err)
1772                 goto journal_error;
1773         brelse(bh2);
1774         dxtrace(dx_show_index("frame", frame->entries));
1775         return de;
1776
1777 journal_error:
1778         brelse(*bh);
1779         brelse(bh2);
1780         *bh = NULL;
1781         ext4_std_error(dir->i_sb, err);
1782         return ERR_PTR(err);
1783 }
1784
1785 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1786                       struct buffer_head *bh,
1787                       void *buf, int buf_size,
1788                       struct ext4_filename *fname,
1789                       struct ext4_dir_entry_2 **dest_de)
1790 {
1791         struct ext4_dir_entry_2 *de;
1792         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1793         int nlen, rlen;
1794         unsigned int offset = 0;
1795         char *top;
1796
1797         de = (struct ext4_dir_entry_2 *)buf;
1798         top = buf + buf_size - reclen;
1799         while ((char *) de <= top) {
1800                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1801                                          buf, buf_size, offset))
1802                         return -EFSCORRUPTED;
1803                 if (ext4_match(fname, de))
1804                         return -EEXIST;
1805                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1806                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1807                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1808                         break;
1809                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1810                 offset += rlen;
1811         }
1812         if ((char *) de > top)
1813                 return -ENOSPC;
1814
1815         *dest_de = de;
1816         return 0;
1817 }
1818
1819 void ext4_insert_dentry(struct inode *inode,
1820                         struct ext4_dir_entry_2 *de,
1821                         int buf_size,
1822                         struct ext4_filename *fname)
1823 {
1824
1825         int nlen, rlen;
1826
1827         nlen = EXT4_DIR_REC_LEN(de->name_len);
1828         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1829         if (de->inode) {
1830                 struct ext4_dir_entry_2 *de1 =
1831                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1832                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1833                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1834                 de = de1;
1835         }
1836         de->file_type = EXT4_FT_UNKNOWN;
1837         de->inode = cpu_to_le32(inode->i_ino);
1838         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1839         de->name_len = fname_len(fname);
1840         memcpy(de->name, fname_name(fname), fname_len(fname));
1841 }
1842
1843 /*
1844  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1845  * it points to a directory entry which is guaranteed to be large
1846  * enough for new directory entry.  If de is NULL, then
1847  * add_dirent_to_buf will attempt search the directory block for
1848  * space.  It will return -ENOSPC if no space is available, and -EIO
1849  * and -EEXIST if directory entry already exists.
1850  */
1851 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1852                              struct inode *dir,
1853                              struct inode *inode, struct ext4_dir_entry_2 *de,
1854                              struct buffer_head *bh)
1855 {
1856         unsigned int    blocksize = dir->i_sb->s_blocksize;
1857         int             csum_size = 0;
1858         int             err;
1859
1860         if (ext4_has_metadata_csum(inode->i_sb))
1861                 csum_size = sizeof(struct ext4_dir_entry_tail);
1862
1863         if (!de) {
1864                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1865                                         blocksize - csum_size, fname, &de);
1866                 if (err)
1867                         return err;
1868         }
1869         BUFFER_TRACE(bh, "get_write_access");
1870         err = ext4_journal_get_write_access(handle, bh);
1871         if (err) {
1872                 ext4_std_error(dir->i_sb, err);
1873                 return err;
1874         }
1875
1876         /* By now the buffer is marked for journaling */
1877         ext4_insert_dentry(inode, de, blocksize, fname);
1878
1879         /*
1880          * XXX shouldn't update any times until successful
1881          * completion of syscall, but too many callers depend
1882          * on this.
1883          *
1884          * XXX similarly, too many callers depend on
1885          * ext4_new_inode() setting the times, but error
1886          * recovery deletes the inode, so the worst that can
1887          * happen is that the times are slightly out of date
1888          * and/or different from the directory change time.
1889          */
1890         dir->i_mtime = dir->i_ctime = current_time(dir);
1891         ext4_update_dx_flag(dir);
1892         inode_inc_iversion(dir);
1893         ext4_mark_inode_dirty(handle, dir);
1894         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1895         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1896         if (err)
1897                 ext4_std_error(dir->i_sb, err);
1898         return 0;
1899 }
1900
1901 /*
1902  * This converts a one block unindexed directory to a 3 block indexed
1903  * directory, and adds the dentry to the indexed directory.
1904  */
1905 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1906                             struct inode *dir,
1907                             struct inode *inode, struct buffer_head *bh)
1908 {
1909         struct buffer_head *bh2;
1910         struct dx_root  *root;
1911         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1912         struct dx_entry *entries;
1913         struct ext4_dir_entry_2 *de, *de2;
1914         struct ext4_dir_entry_tail *t;
1915         char            *data1, *top;
1916         unsigned        len;
1917         int             retval;
1918         unsigned        blocksize;
1919         ext4_lblk_t  block;
1920         struct fake_dirent *fde;
1921         int csum_size = 0;
1922
1923         if (ext4_has_metadata_csum(inode->i_sb))
1924                 csum_size = sizeof(struct ext4_dir_entry_tail);
1925
1926         blocksize =  dir->i_sb->s_blocksize;
1927         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1928         BUFFER_TRACE(bh, "get_write_access");
1929         retval = ext4_journal_get_write_access(handle, bh);
1930         if (retval) {
1931                 ext4_std_error(dir->i_sb, retval);
1932                 brelse(bh);
1933                 return retval;
1934         }
1935         root = (struct dx_root *) bh->b_data;
1936
1937         /* The 0th block becomes the root, move the dirents out */
1938         fde = &root->dotdot;
1939         de = (struct ext4_dir_entry_2 *)((char *)fde +
1940                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1941         if ((char *) de >= (((char *) root) + blocksize)) {
1942                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1943                 brelse(bh);
1944                 return -EFSCORRUPTED;
1945         }
1946         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1947
1948         /* Allocate new block for the 0th block's dirents */
1949         bh2 = ext4_append(handle, dir, &block);
1950         if (IS_ERR(bh2)) {
1951                 brelse(bh);
1952                 return PTR_ERR(bh2);
1953         }
1954         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1955         data1 = bh2->b_data;
1956
1957         memcpy (data1, de, len);
1958         de = (struct ext4_dir_entry_2 *) data1;
1959         top = data1 + len;
1960         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1961                 de = de2;
1962         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1963                                            (char *) de,
1964                                            blocksize);
1965
1966         if (csum_size) {
1967                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1968                 initialize_dirent_tail(t, blocksize);
1969         }
1970
1971         /* Initialize the root; the dot dirents already exist */
1972         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1973         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1974                                            blocksize);
1975         memset (&root->info, 0, sizeof(root->info));
1976         root->info.info_length = sizeof(root->info);
1977         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1978         entries = root->entries;
1979         dx_set_block(entries, 1);
1980         dx_set_count(entries, 1);
1981         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1982
1983         /* Initialize as for dx_probe */
1984         fname->hinfo.hash_version = root->info.hash_version;
1985         if (fname->hinfo.hash_version <= DX_HASH_TEA)
1986                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1987         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1988         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1989
1990         memset(frames, 0, sizeof(frames));
1991         frame = frames;
1992         frame->entries = entries;
1993         frame->at = entries;
1994         frame->bh = bh;
1995
1996         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1997         if (retval)
1998                 goto out_frames;
1999         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2000         if (retval)
2001                 goto out_frames;
2002
2003         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2004         if (IS_ERR(de)) {
2005                 retval = PTR_ERR(de);
2006                 goto out_frames;
2007         }
2008
2009         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2010 out_frames:
2011         /*
2012          * Even if the block split failed, we have to properly write
2013          * out all the changes we did so far. Otherwise we can end up
2014          * with corrupted filesystem.
2015          */
2016         if (retval)
2017                 ext4_mark_inode_dirty(handle, dir);
2018         dx_release(frames);
2019         brelse(bh2);
2020         return retval;
2021 }
2022
2023 /*
2024  *      ext4_add_entry()
2025  *
2026  * adds a file entry to the specified directory, using the same
2027  * semantics as ext4_find_entry(). It returns NULL if it failed.
2028  *
2029  * NOTE!! The inode part of 'de' is left at 0 - which means you
2030  * may not sleep between calling this and putting something into
2031  * the entry, as someone else might have used it while you slept.
2032  */
2033 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2034                           struct inode *inode)
2035 {
2036         struct inode *dir = d_inode(dentry->d_parent);
2037         struct buffer_head *bh = NULL;
2038         struct ext4_dir_entry_2 *de;
2039         struct ext4_dir_entry_tail *t;
2040         struct super_block *sb;
2041         struct ext4_filename fname;
2042         int     retval;
2043         int     dx_fallback=0;
2044         unsigned blocksize;
2045         ext4_lblk_t block, blocks;
2046         int     csum_size = 0;
2047
2048         if (ext4_has_metadata_csum(inode->i_sb))
2049                 csum_size = sizeof(struct ext4_dir_entry_tail);
2050
2051         sb = dir->i_sb;
2052         blocksize = sb->s_blocksize;
2053         if (!dentry->d_name.len)
2054                 return -EINVAL;
2055
2056         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2057         if (retval)
2058                 return retval;
2059
2060         if (ext4_has_inline_data(dir)) {
2061                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2062                 if (retval < 0)
2063                         goto out;
2064                 if (retval == 1) {
2065                         retval = 0;
2066                         goto out;
2067                 }
2068         }
2069
2070         if (is_dx(dir)) {
2071                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2072                 if (!retval || (retval != ERR_BAD_DX_DIR))
2073                         goto out;
2074                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2075                 dx_fallback++;
2076                 ext4_mark_inode_dirty(handle, dir);
2077         }
2078         blocks = dir->i_size >> sb->s_blocksize_bits;
2079         for (block = 0; block < blocks; block++) {
2080                 bh = ext4_read_dirblock(dir, block, DIRENT);
2081                 if (IS_ERR(bh)) {
2082                         retval = PTR_ERR(bh);
2083                         bh = NULL;
2084                         goto out;
2085                 }
2086                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2087                                            NULL, bh);
2088                 if (retval != -ENOSPC)
2089                         goto out;
2090
2091                 if (blocks == 1 && !dx_fallback &&
2092                     ext4_has_feature_dir_index(sb)) {
2093                         retval = make_indexed_dir(handle, &fname, dir,
2094                                                   inode, bh);
2095                         bh = NULL; /* make_indexed_dir releases bh */
2096                         goto out;
2097                 }
2098                 brelse(bh);
2099         }
2100         bh = ext4_append(handle, dir, &block);
2101         if (IS_ERR(bh)) {
2102                 retval = PTR_ERR(bh);
2103                 bh = NULL;
2104                 goto out;
2105         }
2106         de = (struct ext4_dir_entry_2 *) bh->b_data;
2107         de->inode = 0;
2108         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2109
2110         if (csum_size) {
2111                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2112                 initialize_dirent_tail(t, blocksize);
2113         }
2114
2115         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2116 out:
2117         ext4_fname_free_filename(&fname);
2118         brelse(bh);
2119         if (retval == 0)
2120                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2121         return retval;
2122 }
2123
2124 /*
2125  * Returns 0 for success, or a negative error value
2126  */
2127 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2128                              struct inode *dir, struct inode *inode)
2129 {
2130         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2131         struct dx_entry *entries, *at;
2132         struct buffer_head *bh;
2133         struct super_block *sb = dir->i_sb;
2134         struct ext4_dir_entry_2 *de;
2135         int restart;
2136         int err;
2137
2138 again:
2139         restart = 0;
2140         frame = dx_probe(fname, dir, NULL, frames);
2141         if (IS_ERR(frame))
2142                 return PTR_ERR(frame);
2143         entries = frame->entries;
2144         at = frame->at;
2145         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2146         if (IS_ERR(bh)) {
2147                 err = PTR_ERR(bh);
2148                 bh = NULL;
2149                 goto cleanup;
2150         }
2151
2152         BUFFER_TRACE(bh, "get_write_access");
2153         err = ext4_journal_get_write_access(handle, bh);
2154         if (err)
2155                 goto journal_error;
2156
2157         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2158         if (err != -ENOSPC)
2159                 goto cleanup;
2160
2161         err = 0;
2162         /* Block full, should compress but for now just split */
2163         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2164                        dx_get_count(entries), dx_get_limit(entries)));
2165         /* Need to split index? */
2166         if (dx_get_count(entries) == dx_get_limit(entries)) {
2167                 ext4_lblk_t newblock;
2168                 int levels = frame - frames + 1;
2169                 unsigned int icount;
2170                 int add_level = 1;
2171                 struct dx_entry *entries2;
2172                 struct dx_node *node2;
2173                 struct buffer_head *bh2;
2174
2175                 while (frame > frames) {
2176                         if (dx_get_count((frame - 1)->entries) <
2177                             dx_get_limit((frame - 1)->entries)) {
2178                                 add_level = 0;
2179                                 break;
2180                         }
2181                         frame--; /* split higher index block */
2182                         at = frame->at;
2183                         entries = frame->entries;
2184                         restart = 1;
2185                 }
2186                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2187                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2188                                          "reach max htree level :%d",
2189                                          dir->i_ino, levels);
2190                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2191                                 ext4_warning(sb, "Large directory feature is "
2192                                                  "not enabled on this "
2193                                                  "filesystem");
2194                         }
2195                         err = -ENOSPC;
2196                         goto cleanup;
2197                 }
2198                 icount = dx_get_count(entries);
2199                 bh2 = ext4_append(handle, dir, &newblock);
2200                 if (IS_ERR(bh2)) {
2201                         err = PTR_ERR(bh2);
2202                         goto cleanup;
2203                 }
2204                 node2 = (struct dx_node *)(bh2->b_data);
2205                 entries2 = node2->entries;
2206                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2207                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2208                                                            sb->s_blocksize);
2209                 BUFFER_TRACE(frame->bh, "get_write_access");
2210                 err = ext4_journal_get_write_access(handle, frame->bh);
2211                 if (err)
2212                         goto journal_error;
2213                 if (!add_level) {
2214                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2215                         unsigned hash2 = dx_get_hash(entries + icount1);
2216                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2217                                        icount1, icount2));
2218
2219                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2220                         err = ext4_journal_get_write_access(handle,
2221                                                              (frame - 1)->bh);
2222                         if (err)
2223                                 goto journal_error;
2224
2225                         memcpy((char *) entries2, (char *) (entries + icount1),
2226                                icount2 * sizeof(struct dx_entry));
2227                         dx_set_count(entries, icount1);
2228                         dx_set_count(entries2, icount2);
2229                         dx_set_limit(entries2, dx_node_limit(dir));
2230
2231                         /* Which index block gets the new entry? */
2232                         if (at - entries >= icount1) {
2233                                 frame->at = at = at - entries - icount1 + entries2;
2234                                 frame->entries = entries = entries2;
2235                                 swap(frame->bh, bh2);
2236                         }
2237                         dx_insert_block((frame - 1), hash2, newblock);
2238                         dxtrace(dx_show_index("node", frame->entries));
2239                         dxtrace(dx_show_index("node",
2240                                ((struct dx_node *) bh2->b_data)->entries));
2241                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2242                         if (err)
2243                                 goto journal_error;
2244                         brelse (bh2);
2245                         err = ext4_handle_dirty_dx_node(handle, dir,
2246                                                    (frame - 1)->bh);
2247                         if (err)
2248                                 goto journal_error;
2249                         if (restart) {
2250                                 err = ext4_handle_dirty_dx_node(handle, dir,
2251                                                            frame->bh);
2252                                 goto journal_error;
2253                         }
2254                 } else {
2255                         struct dx_root *dxroot;
2256                         memcpy((char *) entries2, (char *) entries,
2257                                icount * sizeof(struct dx_entry));
2258                         dx_set_limit(entries2, dx_node_limit(dir));
2259
2260                         /* Set up root */
2261                         dx_set_count(entries, 1);
2262                         dx_set_block(entries + 0, newblock);
2263                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2264                         dxroot->info.indirect_levels += 1;
2265                         dxtrace(printk(KERN_DEBUG
2266                                        "Creating %d level index...\n",
2267                                        info->indirect_levels));
2268                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2269                         if (err)
2270                                 goto journal_error;
2271                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2272                         brelse(bh2);
2273                         restart = 1;
2274                         goto journal_error;
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); /* this is a no-op if err == 0 */
2287 cleanup:
2288         brelse(bh);
2289         dx_release(frames);
2290         /* @restart is true means htree-path has been changed, we need to
2291          * repeat dx_probe() to find out valid htree-path
2292          */
2293         if (restart && err == 0)
2294                 goto again;
2295         return err;
2296 }
2297
2298 /*
2299  * ext4_generic_delete_entry deletes a directory entry by merging it
2300  * with the previous entry
2301  */
2302 int ext4_generic_delete_entry(handle_t *handle,
2303                               struct inode *dir,
2304                               struct ext4_dir_entry_2 *de_del,
2305                               struct buffer_head *bh,
2306                               void *entry_buf,
2307                               int buf_size,
2308                               int csum_size)
2309 {
2310         struct ext4_dir_entry_2 *de, *pde;
2311         unsigned int blocksize = dir->i_sb->s_blocksize;
2312         int i;
2313
2314         i = 0;
2315         pde = NULL;
2316         de = (struct ext4_dir_entry_2 *)entry_buf;
2317         while (i < buf_size - csum_size) {
2318                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2319                                          bh->b_data, bh->b_size, i))
2320                         return -EFSCORRUPTED;
2321                 if (de == de_del)  {
2322                         if (pde)
2323                                 pde->rec_len = ext4_rec_len_to_disk(
2324                                         ext4_rec_len_from_disk(pde->rec_len,
2325                                                                blocksize) +
2326                                         ext4_rec_len_from_disk(de->rec_len,
2327                                                                blocksize),
2328                                         blocksize);
2329                         else
2330                                 de->inode = 0;
2331                         inode_inc_iversion(dir);
2332                         return 0;
2333                 }
2334                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2335                 pde = de;
2336                 de = ext4_next_entry(de, blocksize);
2337         }
2338         return -ENOENT;
2339 }
2340
2341 static int ext4_delete_entry(handle_t *handle,
2342                              struct inode *dir,
2343                              struct ext4_dir_entry_2 *de_del,
2344                              struct buffer_head *bh)
2345 {
2346         int err, csum_size = 0;
2347
2348         if (ext4_has_inline_data(dir)) {
2349                 int has_inline_data = 1;
2350                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2351                                                &has_inline_data);
2352                 if (has_inline_data)
2353                         return err;
2354         }
2355
2356         if (ext4_has_metadata_csum(dir->i_sb))
2357                 csum_size = sizeof(struct ext4_dir_entry_tail);
2358
2359         BUFFER_TRACE(bh, "get_write_access");
2360         err = ext4_journal_get_write_access(handle, bh);
2361         if (unlikely(err))
2362                 goto out;
2363
2364         err = ext4_generic_delete_entry(handle, dir, de_del,
2365                                         bh, bh->b_data,
2366                                         dir->i_sb->s_blocksize, csum_size);
2367         if (err)
2368                 goto out;
2369
2370         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2371         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2372         if (unlikely(err))
2373                 goto out;
2374
2375         return 0;
2376 out:
2377         if (err != -ENOENT)
2378                 ext4_std_error(dir->i_sb, err);
2379         return err;
2380 }
2381
2382 /*
2383  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2384  * since this indicates that nlinks count was previously 1 to avoid overflowing
2385  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2386  * that subdirectory link counts are not being maintained accurately.
2387  *
2388  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2389  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2390  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2391  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2392  */
2393 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2394 {
2395         inc_nlink(inode);
2396         if (is_dx(inode) &&
2397             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2398                 set_nlink(inode, 1);
2399 }
2400
2401 /*
2402  * If a directory had nlink == 1, then we should let it be 1. This indicates
2403  * directory has >EXT4_LINK_MAX subdirs.
2404  */
2405 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2406 {
2407         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2408                 drop_nlink(inode);
2409 }
2410
2411
2412 static int ext4_add_nondir(handle_t *handle,
2413                 struct dentry *dentry, struct inode *inode)
2414 {
2415         int err = ext4_add_entry(handle, dentry, inode);
2416         if (!err) {
2417                 ext4_mark_inode_dirty(handle, inode);
2418                 unlock_new_inode(inode);
2419                 d_instantiate(dentry, inode);
2420                 return 0;
2421         }
2422         drop_nlink(inode);
2423         unlock_new_inode(inode);
2424         iput(inode);
2425         return err;
2426 }
2427
2428 /*
2429  * By the time this is called, we already have created
2430  * the directory cache entry for the new file, but it
2431  * is so far negative - it has no inode.
2432  *
2433  * If the create succeeds, we fill in the inode information
2434  * with d_instantiate().
2435  */
2436 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2437                        bool excl)
2438 {
2439         handle_t *handle;
2440         struct inode *inode;
2441         int err, credits, retries = 0;
2442
2443         err = dquot_initialize(dir);
2444         if (err)
2445                 return err;
2446
2447         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2448                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2449 retry:
2450         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2451                                             NULL, EXT4_HT_DIR, credits);
2452         handle = ext4_journal_current_handle();
2453         err = PTR_ERR(inode);
2454         if (!IS_ERR(inode)) {
2455                 inode->i_op = &ext4_file_inode_operations;
2456                 inode->i_fop = &ext4_file_operations;
2457                 ext4_set_aops(inode);
2458                 err = ext4_add_nondir(handle, dentry, inode);
2459                 if (!err && IS_DIRSYNC(dir))
2460                         ext4_handle_sync(handle);
2461         }
2462         if (handle)
2463                 ext4_journal_stop(handle);
2464         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2465                 goto retry;
2466         return err;
2467 }
2468
2469 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2470                       umode_t mode, dev_t rdev)
2471 {
2472         handle_t *handle;
2473         struct inode *inode;
2474         int err, credits, retries = 0;
2475
2476         err = dquot_initialize(dir);
2477         if (err)
2478                 return err;
2479
2480         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2481                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2482 retry:
2483         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2484                                             NULL, EXT4_HT_DIR, credits);
2485         handle = ext4_journal_current_handle();
2486         err = PTR_ERR(inode);
2487         if (!IS_ERR(inode)) {
2488                 init_special_inode(inode, inode->i_mode, rdev);
2489                 inode->i_op = &ext4_special_inode_operations;
2490                 err = ext4_add_nondir(handle, dentry, inode);
2491                 if (!err && IS_DIRSYNC(dir))
2492                         ext4_handle_sync(handle);
2493         }
2494         if (handle)
2495                 ext4_journal_stop(handle);
2496         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2497                 goto retry;
2498         return err;
2499 }
2500
2501 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2502 {
2503         handle_t *handle;
2504         struct inode *inode;
2505         int err, retries = 0;
2506
2507         err = dquot_initialize(dir);
2508         if (err)
2509                 return err;
2510
2511 retry:
2512         inode = ext4_new_inode_start_handle(dir, mode,
2513                                             NULL, 0, NULL,
2514                                             EXT4_HT_DIR,
2515                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2516                           4 + EXT4_XATTR_TRANS_BLOCKS);
2517         handle = ext4_journal_current_handle();
2518         err = PTR_ERR(inode);
2519         if (!IS_ERR(inode)) {
2520                 inode->i_op = &ext4_file_inode_operations;
2521                 inode->i_fop = &ext4_file_operations;
2522                 ext4_set_aops(inode);
2523                 d_tmpfile(dentry, inode);
2524                 err = ext4_orphan_add(handle, inode);
2525                 if (err)
2526                         goto err_unlock_inode;
2527                 mark_inode_dirty(inode);
2528                 unlock_new_inode(inode);
2529         }
2530         if (handle)
2531                 ext4_journal_stop(handle);
2532         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2533                 goto retry;
2534         return err;
2535 err_unlock_inode:
2536         ext4_journal_stop(handle);
2537         unlock_new_inode(inode);
2538         return err;
2539 }
2540
2541 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2542                           struct ext4_dir_entry_2 *de,
2543                           int blocksize, int csum_size,
2544                           unsigned int parent_ino, int dotdot_real_len)
2545 {
2546         de->inode = cpu_to_le32(inode->i_ino);
2547         de->name_len = 1;
2548         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2549                                            blocksize);
2550         strcpy(de->name, ".");
2551         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2552
2553         de = ext4_next_entry(de, blocksize);
2554         de->inode = cpu_to_le32(parent_ino);
2555         de->name_len = 2;
2556         if (!dotdot_real_len)
2557                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2558                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2559                                         blocksize);
2560         else
2561                 de->rec_len = ext4_rec_len_to_disk(
2562                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2563         strcpy(de->name, "..");
2564         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2565
2566         return ext4_next_entry(de, blocksize);
2567 }
2568
2569 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2570                              struct inode *inode)
2571 {
2572         struct buffer_head *dir_block = NULL;
2573         struct ext4_dir_entry_2 *de;
2574         struct ext4_dir_entry_tail *t;
2575         ext4_lblk_t block = 0;
2576         unsigned int blocksize = dir->i_sb->s_blocksize;
2577         int csum_size = 0;
2578         int err;
2579
2580         if (ext4_has_metadata_csum(dir->i_sb))
2581                 csum_size = sizeof(struct ext4_dir_entry_tail);
2582
2583         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2584                 err = ext4_try_create_inline_dir(handle, dir, inode);
2585                 if (err < 0 && err != -ENOSPC)
2586                         goto out;
2587                 if (!err)
2588                         goto out;
2589         }
2590
2591         inode->i_size = 0;
2592         dir_block = ext4_append(handle, inode, &block);
2593         if (IS_ERR(dir_block))
2594                 return PTR_ERR(dir_block);
2595         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2596         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2597         set_nlink(inode, 2);
2598         if (csum_size) {
2599                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2600                 initialize_dirent_tail(t, blocksize);
2601         }
2602
2603         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2604         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2605         if (err)
2606                 goto out;
2607         set_buffer_verified(dir_block);
2608 out:
2609         brelse(dir_block);
2610         return err;
2611 }
2612
2613 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2614 {
2615         handle_t *handle;
2616         struct inode *inode;
2617         int err, credits, retries = 0;
2618
2619         if (EXT4_DIR_LINK_MAX(dir))
2620                 return -EMLINK;
2621
2622         err = dquot_initialize(dir);
2623         if (err)
2624                 return err;
2625
2626         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2627                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2628 retry:
2629         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2630                                             &dentry->d_name,
2631                                             0, NULL, EXT4_HT_DIR, credits);
2632         handle = ext4_journal_current_handle();
2633         err = PTR_ERR(inode);
2634         if (IS_ERR(inode))
2635                 goto out_stop;
2636
2637         inode->i_op = &ext4_dir_inode_operations;
2638         inode->i_fop = &ext4_dir_operations;
2639         err = ext4_init_new_dir(handle, dir, inode);
2640         if (err)
2641                 goto out_clear_inode;
2642         err = ext4_mark_inode_dirty(handle, inode);
2643         if (!err)
2644                 err = ext4_add_entry(handle, dentry, inode);
2645         if (err) {
2646 out_clear_inode:
2647                 clear_nlink(inode);
2648                 unlock_new_inode(inode);
2649                 ext4_mark_inode_dirty(handle, inode);
2650                 iput(inode);
2651                 goto out_stop;
2652         }
2653         ext4_inc_count(handle, dir);
2654         ext4_update_dx_flag(dir);
2655         err = ext4_mark_inode_dirty(handle, dir);
2656         if (err)
2657                 goto out_clear_inode;
2658         unlock_new_inode(inode);
2659         d_instantiate(dentry, inode);
2660         if (IS_DIRSYNC(dir))
2661                 ext4_handle_sync(handle);
2662
2663 out_stop:
2664         if (handle)
2665                 ext4_journal_stop(handle);
2666         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2667                 goto retry;
2668         return err;
2669 }
2670
2671 /*
2672  * routine to check that the specified directory is empty (for rmdir)
2673  */
2674 bool ext4_empty_dir(struct inode *inode)
2675 {
2676         unsigned int offset;
2677         struct buffer_head *bh;
2678         struct ext4_dir_entry_2 *de, *de1;
2679         struct super_block *sb;
2680
2681         if (ext4_has_inline_data(inode)) {
2682                 int has_inline_data = 1;
2683                 int ret;
2684
2685                 ret = empty_inline_dir(inode, &has_inline_data);
2686                 if (has_inline_data)
2687                         return ret;
2688         }
2689
2690         sb = inode->i_sb;
2691         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2692                 EXT4_ERROR_INODE(inode, "invalid size");
2693                 return true;
2694         }
2695         bh = ext4_read_dirblock(inode, 0, EITHER);
2696         if (IS_ERR(bh))
2697                 return true;
2698
2699         de = (struct ext4_dir_entry_2 *) bh->b_data;
2700         de1 = ext4_next_entry(de, sb->s_blocksize);
2701         if (le32_to_cpu(de->inode) != inode->i_ino ||
2702                         le32_to_cpu(de1->inode) == 0 ||
2703                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2704                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2705                 brelse(bh);
2706                 return true;
2707         }
2708         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2709                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2710         de = ext4_next_entry(de1, sb->s_blocksize);
2711         while (offset < inode->i_size) {
2712                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2713                         unsigned int lblock;
2714                         brelse(bh);
2715                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2716                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2717                         if (IS_ERR(bh))
2718                                 return true;
2719                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2720                 }
2721                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2722                                          bh->b_data, bh->b_size, offset)) {
2723                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2724                                                          sb->s_blocksize);
2725                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2726                         continue;
2727                 }
2728                 if (le32_to_cpu(de->inode)) {
2729                         brelse(bh);
2730                         return false;
2731                 }
2732                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2733                 de = ext4_next_entry(de, sb->s_blocksize);
2734         }
2735         brelse(bh);
2736         return true;
2737 }
2738
2739 /*
2740  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2741  * such inodes, starting at the superblock, in case we crash before the
2742  * file is closed/deleted, or in case the inode truncate spans multiple
2743  * transactions and the last transaction is not recovered after a crash.
2744  *
2745  * At filesystem recovery time, we walk this list deleting unlinked
2746  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2747  *
2748  * Orphan list manipulation functions must be called under i_mutex unless
2749  * we are just creating the inode or deleting it.
2750  */
2751 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2752 {
2753         struct super_block *sb = inode->i_sb;
2754         struct ext4_sb_info *sbi = EXT4_SB(sb);
2755         struct ext4_iloc iloc;
2756         int err = 0, rc;
2757         bool dirty = false;
2758
2759         if (!sbi->s_journal || is_bad_inode(inode))
2760                 return 0;
2761
2762         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2763                      !inode_is_locked(inode));
2764         /*
2765          * Exit early if inode already is on orphan list. This is a big speedup
2766          * since we don't have to contend on the global s_orphan_lock.
2767          */
2768         if (!list_empty(&EXT4_I(inode)->i_orphan))
2769                 return 0;
2770
2771         /*
2772          * Orphan handling is only valid for files with data blocks
2773          * being truncated, or files being unlinked. Note that we either
2774          * hold i_mutex, or the inode can not be referenced from outside,
2775          * so i_nlink should not be bumped due to race
2776          */
2777         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2778                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2779
2780         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2781         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2782         if (err)
2783                 goto out;
2784
2785         err = ext4_reserve_inode_write(handle, inode, &iloc);
2786         if (err)
2787                 goto out;
2788
2789         mutex_lock(&sbi->s_orphan_lock);
2790         /*
2791          * Due to previous errors inode may be already a part of on-disk
2792          * orphan list. If so skip on-disk list modification.
2793          */
2794         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2795             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2796                 /* Insert this inode at the head of the on-disk orphan list */
2797                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2798                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2799                 dirty = true;
2800         }
2801         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2802         mutex_unlock(&sbi->s_orphan_lock);
2803
2804         if (dirty) {
2805                 err = ext4_handle_dirty_super(handle, sb);
2806                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2807                 if (!err)
2808                         err = rc;
2809                 if (err) {
2810                         /*
2811                          * We have to remove inode from in-memory list if
2812                          * addition to on disk orphan list failed. Stray orphan
2813                          * list entries can cause panics at unmount time.
2814                          */
2815                         mutex_lock(&sbi->s_orphan_lock);
2816                         list_del_init(&EXT4_I(inode)->i_orphan);
2817                         mutex_unlock(&sbi->s_orphan_lock);
2818                 }
2819         }
2820         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2821         jbd_debug(4, "orphan inode %lu will point to %d\n",
2822                         inode->i_ino, NEXT_ORPHAN(inode));
2823 out:
2824         ext4_std_error(sb, err);
2825         return err;
2826 }
2827
2828 /*
2829  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2830  * of such inodes stored on disk, because it is finally being cleaned up.
2831  */
2832 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2833 {
2834         struct list_head *prev;
2835         struct ext4_inode_info *ei = EXT4_I(inode);
2836         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2837         __u32 ino_next;
2838         struct ext4_iloc iloc;
2839         int err = 0;
2840
2841         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2842                 return 0;
2843
2844         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2845                      !inode_is_locked(inode));
2846         /* Do this quick check before taking global s_orphan_lock. */
2847         if (list_empty(&ei->i_orphan))
2848                 return 0;
2849
2850         if (handle) {
2851                 /* Grab inode buffer early before taking global s_orphan_lock */
2852                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2853         }
2854
2855         mutex_lock(&sbi->s_orphan_lock);
2856         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2857
2858         prev = ei->i_orphan.prev;
2859         list_del_init(&ei->i_orphan);
2860
2861         /* If we're on an error path, we may not have a valid
2862          * transaction handle with which to update the orphan list on
2863          * disk, but we still need to remove the inode from the linked
2864          * list in memory. */
2865         if (!handle || err) {
2866                 mutex_unlock(&sbi->s_orphan_lock);
2867                 goto out_err;
2868         }
2869
2870         ino_next = NEXT_ORPHAN(inode);
2871         if (prev == &sbi->s_orphan) {
2872                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2873                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2874                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2875                 if (err) {
2876                         mutex_unlock(&sbi->s_orphan_lock);
2877                         goto out_brelse;
2878                 }
2879                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2880                 mutex_unlock(&sbi->s_orphan_lock);
2881                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2882         } else {
2883                 struct ext4_iloc iloc2;
2884                 struct inode *i_prev =
2885                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2886
2887                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2888                           i_prev->i_ino, ino_next);
2889                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2890                 if (err) {
2891                         mutex_unlock(&sbi->s_orphan_lock);
2892                         goto out_brelse;
2893                 }
2894                 NEXT_ORPHAN(i_prev) = ino_next;
2895                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2896                 mutex_unlock(&sbi->s_orphan_lock);
2897         }
2898         if (err)
2899                 goto out_brelse;
2900         NEXT_ORPHAN(inode) = 0;
2901         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2902 out_err:
2903         ext4_std_error(inode->i_sb, err);
2904         return err;
2905
2906 out_brelse:
2907         brelse(iloc.bh);
2908         goto out_err;
2909 }
2910
2911 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2912 {
2913         int retval;
2914         struct inode *inode;
2915         struct buffer_head *bh;
2916         struct ext4_dir_entry_2 *de;
2917         handle_t *handle = NULL;
2918
2919         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2920                 return -EIO;
2921
2922         /* Initialize quotas before so that eventual writes go in
2923          * separate transaction */
2924         retval = dquot_initialize(dir);
2925         if (retval)
2926                 return retval;
2927         retval = dquot_initialize(d_inode(dentry));
2928         if (retval)
2929                 return retval;
2930
2931         retval = -ENOENT;
2932         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2933         if (IS_ERR(bh))
2934                 return PTR_ERR(bh);
2935         if (!bh)
2936                 goto end_rmdir;
2937
2938         inode = d_inode(dentry);
2939
2940         retval = -EFSCORRUPTED;
2941         if (le32_to_cpu(de->inode) != inode->i_ino)
2942                 goto end_rmdir;
2943
2944         retval = -ENOTEMPTY;
2945         if (!ext4_empty_dir(inode))
2946                 goto end_rmdir;
2947
2948         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2949                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2950         if (IS_ERR(handle)) {
2951                 retval = PTR_ERR(handle);
2952                 handle = NULL;
2953                 goto end_rmdir;
2954         }
2955
2956         if (IS_DIRSYNC(dir))
2957                 ext4_handle_sync(handle);
2958
2959         retval = ext4_delete_entry(handle, dir, de, bh);
2960         if (retval)
2961                 goto end_rmdir;
2962         if (!EXT4_DIR_LINK_EMPTY(inode))
2963                 ext4_warning_inode(inode,
2964                              "empty directory '%.*s' has too many links (%u)",
2965                              dentry->d_name.len, dentry->d_name.name,
2966                              inode->i_nlink);
2967         inode->i_version++;
2968         clear_nlink(inode);
2969         /* There's no need to set i_disksize: the fact that i_nlink is
2970          * zero will ensure that the right thing happens during any
2971          * recovery. */
2972         inode->i_size = 0;
2973         ext4_orphan_add(handle, inode);
2974         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2975         ext4_mark_inode_dirty(handle, inode);
2976         ext4_dec_count(handle, dir);
2977         ext4_update_dx_flag(dir);
2978         ext4_mark_inode_dirty(handle, dir);
2979
2980 end_rmdir:
2981         brelse(bh);
2982         if (handle)
2983                 ext4_journal_stop(handle);
2984         return retval;
2985 }
2986
2987 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2988 {
2989         int retval;
2990         struct inode *inode;
2991         struct buffer_head *bh;
2992         struct ext4_dir_entry_2 *de;
2993         handle_t *handle = NULL;
2994
2995         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2996                 return -EIO;
2997
2998         trace_ext4_unlink_enter(dir, dentry);
2999         /* Initialize quotas before so that eventual writes go
3000          * in separate transaction */
3001         retval = dquot_initialize(dir);
3002         if (retval)
3003                 return retval;
3004         retval = dquot_initialize(d_inode(dentry));
3005         if (retval)
3006                 return retval;
3007
3008         retval = -ENOENT;
3009         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3010         if (IS_ERR(bh))
3011                 return PTR_ERR(bh);
3012         if (!bh)
3013                 goto end_unlink;
3014
3015         inode = d_inode(dentry);
3016
3017         retval = -EFSCORRUPTED;
3018         if (le32_to_cpu(de->inode) != inode->i_ino)
3019                 goto end_unlink;
3020
3021         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3022                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3023         if (IS_ERR(handle)) {
3024                 retval = PTR_ERR(handle);
3025                 handle = NULL;
3026                 goto end_unlink;
3027         }
3028
3029         if (IS_DIRSYNC(dir))
3030                 ext4_handle_sync(handle);
3031
3032         if (inode->i_nlink == 0) {
3033                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3034                                    dentry->d_name.len, dentry->d_name.name);
3035                 set_nlink(inode, 1);
3036         }
3037         retval = ext4_delete_entry(handle, dir, de, bh);
3038         if (retval)
3039                 goto end_unlink;
3040         dir->i_ctime = dir->i_mtime = current_time(dir);
3041         ext4_update_dx_flag(dir);
3042         ext4_mark_inode_dirty(handle, dir);
3043         drop_nlink(inode);
3044         if (!inode->i_nlink)
3045                 ext4_orphan_add(handle, inode);
3046         inode->i_ctime = current_time(inode);
3047         ext4_mark_inode_dirty(handle, inode);
3048
3049 end_unlink:
3050         brelse(bh);
3051         if (handle)
3052                 ext4_journal_stop(handle);
3053         trace_ext4_unlink_exit(dentry, retval);
3054         return retval;
3055 }
3056
3057 static int ext4_symlink(struct inode *dir,
3058                         struct dentry *dentry, const char *symname)
3059 {
3060         handle_t *handle;
3061         struct inode *inode;
3062         int err, len = strlen(symname);
3063         int credits;
3064         bool encryption_required;
3065         struct fscrypt_str disk_link;
3066         struct fscrypt_symlink_data *sd = NULL;
3067
3068         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3069                 return -EIO;
3070
3071         disk_link.len = len + 1;
3072         disk_link.name = (char *) symname;
3073
3074         encryption_required = (ext4_encrypted_inode(dir) ||
3075                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3076         if (encryption_required) {
3077                 err = fscrypt_get_encryption_info(dir);
3078                 if (err)
3079                         return err;
3080                 if (!fscrypt_has_encryption_key(dir))
3081                         return -ENOKEY;
3082                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3083                                  sizeof(struct fscrypt_symlink_data));
3084                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3085                 if (!sd)
3086                         return -ENOMEM;
3087         }
3088
3089         if (disk_link.len > dir->i_sb->s_blocksize) {
3090                 err = -ENAMETOOLONG;
3091                 goto err_free_sd;
3092         }
3093
3094         err = dquot_initialize(dir);
3095         if (err)
3096                 goto err_free_sd;
3097
3098         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3099                 /*
3100                  * For non-fast symlinks, we just allocate inode and put it on
3101                  * orphan list in the first transaction => we need bitmap,
3102                  * group descriptor, sb, inode block, quota blocks, and
3103                  * possibly selinux xattr blocks.
3104                  */
3105                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3106                           EXT4_XATTR_TRANS_BLOCKS;
3107         } else {
3108                 /*
3109                  * Fast symlink. We have to add entry to directory
3110                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3111                  * allocate new inode (bitmap, group descriptor, inode block,
3112                  * quota blocks, sb is already counted in previous macros).
3113                  */
3114                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3115                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3116         }
3117
3118         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3119                                             &dentry->d_name, 0, NULL,
3120                                             EXT4_HT_DIR, credits);
3121         handle = ext4_journal_current_handle();
3122         if (IS_ERR(inode)) {
3123                 if (handle)
3124                         ext4_journal_stop(handle);
3125                 err = PTR_ERR(inode);
3126                 goto err_free_sd;
3127         }
3128
3129         if (encryption_required) {
3130                 struct qstr istr;
3131                 struct fscrypt_str ostr =
3132                         FSTR_INIT(sd->encrypted_path, disk_link.len);
3133
3134                 istr.name = (const unsigned char *) symname;
3135                 istr.len = len;
3136                 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3137                 if (err)
3138                         goto err_drop_inode;
3139                 sd->len = cpu_to_le16(ostr.len);
3140                 disk_link.name = (char *) sd;
3141                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3142         }
3143
3144         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3145                 if (!encryption_required)
3146                         inode->i_op = &ext4_symlink_inode_operations;
3147                 inode_nohighmem(inode);
3148                 ext4_set_aops(inode);
3149                 /*
3150                  * We cannot call page_symlink() with transaction started
3151                  * because it calls into ext4_write_begin() which can wait
3152                  * for transaction commit if we are running out of space
3153                  * and thus we deadlock. So we have to stop transaction now
3154                  * and restart it when symlink contents is written.
3155                  *
3156                  * To keep fs consistent in case of crash, we have to put inode
3157                  * to orphan list in the mean time.
3158                  */
3159                 drop_nlink(inode);
3160                 err = ext4_orphan_add(handle, inode);
3161                 ext4_journal_stop(handle);
3162                 handle = NULL;
3163                 if (err)
3164                         goto err_drop_inode;
3165                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3166                 if (err)
3167                         goto err_drop_inode;
3168                 /*
3169                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3170                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3171                  */
3172                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3173                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3174                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3175                 if (IS_ERR(handle)) {
3176                         err = PTR_ERR(handle);
3177                         handle = NULL;
3178                         goto err_drop_inode;
3179                 }
3180                 set_nlink(inode, 1);
3181                 err = ext4_orphan_del(handle, inode);
3182                 if (err)
3183                         goto err_drop_inode;
3184         } else {
3185                 /* clear the extent format for fast symlink */
3186                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3187                 if (!encryption_required) {
3188                         inode->i_op = &ext4_fast_symlink_inode_operations;
3189                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3190                 }
3191                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3192                        disk_link.len);
3193                 inode->i_size = disk_link.len - 1;
3194         }
3195         EXT4_I(inode)->i_disksize = inode->i_size;
3196         err = ext4_add_nondir(handle, dentry, inode);
3197         if (!err && IS_DIRSYNC(dir))
3198                 ext4_handle_sync(handle);
3199
3200         if (handle)
3201                 ext4_journal_stop(handle);
3202         kfree(sd);
3203         return err;
3204 err_drop_inode:
3205         if (handle)
3206                 ext4_journal_stop(handle);
3207         clear_nlink(inode);
3208         unlock_new_inode(inode);
3209         iput(inode);
3210 err_free_sd:
3211         kfree(sd);
3212         return err;
3213 }
3214
3215 static int ext4_link(struct dentry *old_dentry,
3216                      struct inode *dir, struct dentry *dentry)
3217 {
3218         handle_t *handle;
3219         struct inode *inode = d_inode(old_dentry);
3220         int err, retries = 0;
3221
3222         if (inode->i_nlink >= EXT4_LINK_MAX)
3223                 return -EMLINK;
3224         if (ext4_encrypted_inode(dir) &&
3225                         !fscrypt_has_permitted_context(dir, inode))
3226                 return -EPERM;
3227
3228        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3229            (!projid_eq(EXT4_I(dir)->i_projid,
3230                        EXT4_I(old_dentry->d_inode)->i_projid)))
3231                 return -EXDEV;
3232
3233         err = dquot_initialize(dir);
3234         if (err)
3235                 return err;
3236
3237 retry:
3238         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3239                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3240                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3241         if (IS_ERR(handle))
3242                 return PTR_ERR(handle);
3243
3244         if (IS_DIRSYNC(dir))
3245                 ext4_handle_sync(handle);
3246
3247         inode->i_ctime = current_time(inode);
3248         ext4_inc_count(handle, inode);
3249         ihold(inode);
3250
3251         err = ext4_add_entry(handle, dentry, inode);
3252         if (!err) {
3253                 ext4_mark_inode_dirty(handle, inode);
3254                 /* this can happen only for tmpfile being
3255                  * linked the first time
3256                  */
3257                 if (inode->i_nlink == 1)
3258                         ext4_orphan_del(handle, inode);
3259                 d_instantiate(dentry, inode);
3260         } else {
3261                 drop_nlink(inode);
3262                 iput(inode);
3263         }
3264         ext4_journal_stop(handle);
3265         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3266                 goto retry;
3267         return err;
3268 }
3269
3270
3271 /*
3272  * Try to find buffer head where contains the parent block.
3273  * It should be the inode block if it is inlined or the 1st block
3274  * if it is a normal dir.
3275  */
3276 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3277                                         struct inode *inode,
3278                                         int *retval,
3279                                         struct ext4_dir_entry_2 **parent_de,
3280                                         int *inlined)
3281 {
3282         struct buffer_head *bh;
3283
3284         if (!ext4_has_inline_data(inode)) {
3285                 bh = ext4_read_dirblock(inode, 0, EITHER);
3286                 if (IS_ERR(bh)) {
3287                         *retval = PTR_ERR(bh);
3288                         return NULL;
3289                 }
3290                 *parent_de = ext4_next_entry(
3291                                         (struct ext4_dir_entry_2 *)bh->b_data,
3292                                         inode->i_sb->s_blocksize);
3293                 return bh;
3294         }
3295
3296         *inlined = 1;
3297         return ext4_get_first_inline_block(inode, parent_de, retval);
3298 }
3299
3300 struct ext4_renament {
3301         struct inode *dir;
3302         struct dentry *dentry;
3303         struct inode *inode;
3304         bool is_dir;
3305         int dir_nlink_delta;
3306
3307         /* entry for "dentry" */
3308         struct buffer_head *bh;
3309         struct ext4_dir_entry_2 *de;
3310         int inlined;
3311
3312         /* entry for ".." in inode if it's a directory */
3313         struct buffer_head *dir_bh;
3314         struct ext4_dir_entry_2 *parent_de;
3315         int dir_inlined;
3316 };
3317
3318 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3319 {
3320         int retval;
3321
3322         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3323                                               &retval, &ent->parent_de,
3324                                               &ent->dir_inlined);
3325         if (!ent->dir_bh)
3326                 return retval;
3327         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3328                 return -EFSCORRUPTED;
3329         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3330         return ext4_journal_get_write_access(handle, ent->dir_bh);
3331 }
3332
3333 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3334                                   unsigned dir_ino)
3335 {
3336         int retval;
3337
3338         ent->parent_de->inode = cpu_to_le32(dir_ino);
3339         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3340         if (!ent->dir_inlined) {
3341                 if (is_dx(ent->inode)) {
3342                         retval = ext4_handle_dirty_dx_node(handle,
3343                                                            ent->inode,
3344                                                            ent->dir_bh);
3345                 } else {
3346                         retval = ext4_handle_dirty_dirent_node(handle,
3347                                                                ent->inode,
3348                                                                ent->dir_bh);
3349                 }
3350         } else {
3351                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3352         }
3353         if (retval) {
3354                 ext4_std_error(ent->dir->i_sb, retval);
3355                 return retval;
3356         }
3357         return 0;
3358 }
3359
3360 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3361                        unsigned ino, unsigned file_type)
3362 {
3363         int retval;
3364
3365         BUFFER_TRACE(ent->bh, "get write access");
3366         retval = ext4_journal_get_write_access(handle, ent->bh);
3367         if (retval)
3368                 return retval;
3369         ent->de->inode = cpu_to_le32(ino);
3370         if (ext4_has_feature_filetype(ent->dir->i_sb))
3371                 ent->de->file_type = file_type;
3372         ent->dir->i_version++;
3373         ent->dir->i_ctime = ent->dir->i_mtime =
3374                 current_time(ent->dir);
3375         ext4_mark_inode_dirty(handle, ent->dir);
3376         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3377         if (!ent->inlined) {
3378                 retval = ext4_handle_dirty_dirent_node(handle,
3379                                                        ent->dir, ent->bh);
3380                 if (unlikely(retval)) {
3381                         ext4_std_error(ent->dir->i_sb, retval);
3382                         return retval;
3383                 }
3384         }
3385         brelse(ent->bh);
3386         ent->bh = NULL;
3387
3388         return 0;
3389 }
3390
3391 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3392                                   const struct qstr *d_name)
3393 {
3394         int retval = -ENOENT;
3395         struct buffer_head *bh;
3396         struct ext4_dir_entry_2 *de;
3397
3398         bh = ext4_find_entry(dir, d_name, &de, NULL);
3399         if (IS_ERR(bh))
3400                 return PTR_ERR(bh);
3401         if (bh) {
3402                 retval = ext4_delete_entry(handle, dir, de, bh);
3403                 brelse(bh);
3404         }
3405         return retval;
3406 }
3407
3408 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3409                                int force_reread)
3410 {
3411         int retval;
3412         /*
3413          * ent->de could have moved from under us during htree split, so make
3414          * sure that we are deleting the right entry.  We might also be pointing
3415          * to a stale entry in the unused part of ent->bh so just checking inum
3416          * and the name isn't enough.
3417          */
3418         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3419             ent->de->name_len != ent->dentry->d_name.len ||
3420             strncmp(ent->de->name, ent->dentry->d_name.name,
3421                     ent->de->name_len) ||
3422             force_reread) {
3423                 retval = ext4_find_delete_entry(handle, ent->dir,
3424                                                 &ent->dentry->d_name);
3425         } else {
3426                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3427                 if (retval == -ENOENT) {
3428                         retval = ext4_find_delete_entry(handle, ent->dir,
3429                                                         &ent->dentry->d_name);
3430                 }
3431         }
3432
3433         if (retval) {
3434                 ext4_warning_inode(ent->dir,
3435                                    "Deleting old file: nlink %d, error=%d",
3436                                    ent->dir->i_nlink, retval);
3437         }
3438 }
3439
3440 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3441 {
3442         if (ent->dir_nlink_delta) {
3443                 if (ent->dir_nlink_delta == -1)
3444                         ext4_dec_count(handle, ent->dir);
3445                 else
3446                         ext4_inc_count(handle, ent->dir);
3447                 ext4_mark_inode_dirty(handle, ent->dir);
3448         }
3449 }
3450
3451 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3452                                               int credits, handle_t **h)
3453 {
3454         struct inode *wh;
3455         handle_t *handle;
3456         int retries = 0;
3457
3458         /*
3459          * for inode block, sb block, group summaries,
3460          * and inode bitmap
3461          */
3462         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3463                     EXT4_XATTR_TRANS_BLOCKS + 4);
3464 retry:
3465         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3466                                          &ent->dentry->d_name, 0, NULL,
3467                                          EXT4_HT_DIR, credits);
3468
3469         handle = ext4_journal_current_handle();
3470         if (IS_ERR(wh)) {
3471                 if (handle)
3472                         ext4_journal_stop(handle);
3473                 if (PTR_ERR(wh) == -ENOSPC &&
3474                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3475                         goto retry;
3476         } else {
3477                 *h = handle;
3478                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3479                 wh->i_op = &ext4_special_inode_operations;
3480         }
3481         return wh;
3482 }
3483
3484 /*
3485  * Anybody can rename anything with this: the permission checks are left to the
3486  * higher-level routines.
3487  *
3488  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3489  * while new_{dentry,inode) refers to the destination dentry/inode
3490  * This comes from rename(const char *oldpath, const char *newpath)
3491  */
3492 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3493                        struct inode *new_dir, struct dentry *new_dentry,
3494                        unsigned int flags)
3495 {
3496         handle_t *handle = NULL;
3497         struct ext4_renament old = {
3498                 .dir = old_dir,
3499                 .dentry = old_dentry,
3500                 .inode = d_inode(old_dentry),
3501         };
3502         struct ext4_renament new = {
3503                 .dir = new_dir,
3504                 .dentry = new_dentry,
3505                 .inode = d_inode(new_dentry),
3506         };
3507         int force_reread;
3508         int retval;
3509         struct inode *whiteout = NULL;
3510         int credits;
3511         u8 old_file_type;
3512
3513         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3514             (!projid_eq(EXT4_I(new_dir)->i_projid,
3515                         EXT4_I(old_dentry->d_inode)->i_projid)))
3516                 return -EXDEV;
3517
3518         if ((ext4_encrypted_inode(old_dir) &&
3519              !fscrypt_has_encryption_key(old_dir)) ||
3520             (ext4_encrypted_inode(new_dir) &&
3521              !fscrypt_has_encryption_key(new_dir)))
3522                 return -ENOKEY;
3523
3524         retval = dquot_initialize(old.dir);
3525         if (retval)
3526                 return retval;
3527         retval = dquot_initialize(new.dir);
3528         if (retval)
3529                 return retval;
3530
3531         /* Initialize quotas before so that eventual writes go
3532          * in separate transaction */
3533         if (new.inode) {
3534                 retval = dquot_initialize(new.inode);
3535                 if (retval)
3536                         return retval;
3537         }
3538
3539         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3540         if (IS_ERR(old.bh))
3541                 return PTR_ERR(old.bh);
3542         /*
3543          *  Check for inode number is _not_ due to possible IO errors.
3544          *  We might rmdir the source, keep it as pwd of some process
3545          *  and merrily kill the link to whatever was created under the
3546          *  same name. Goodbye sticky bit ;-<
3547          */
3548         retval = -ENOENT;
3549         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3550                 goto end_rename;
3551
3552         if ((old.dir != new.dir) &&
3553             ext4_encrypted_inode(new.dir) &&
3554             !fscrypt_has_permitted_context(new.dir, old.inode)) {
3555                 retval = -EPERM;
3556                 goto end_rename;
3557         }
3558
3559         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3560                                  &new.de, &new.inlined);
3561         if (IS_ERR(new.bh)) {
3562                 retval = PTR_ERR(new.bh);
3563                 new.bh = NULL;
3564                 goto end_rename;
3565         }
3566         if (new.bh) {
3567                 if (!new.inode) {
3568                         brelse(new.bh);
3569                         new.bh = NULL;
3570                 }
3571         }
3572         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3573                 ext4_alloc_da_blocks(old.inode);
3574
3575         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3576                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3577         if (!(flags & RENAME_WHITEOUT)) {
3578                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3579                 if (IS_ERR(handle)) {
3580                         retval = PTR_ERR(handle);
3581                         handle = NULL;
3582                         goto end_rename;
3583                 }
3584         } else {
3585                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3586                 if (IS_ERR(whiteout)) {
3587                         retval = PTR_ERR(whiteout);
3588                         whiteout = NULL;
3589                         goto end_rename;
3590                 }
3591         }
3592
3593         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3594                 ext4_handle_sync(handle);
3595
3596         if (S_ISDIR(old.inode->i_mode)) {
3597                 if (new.inode) {
3598                         retval = -ENOTEMPTY;
3599                         if (!ext4_empty_dir(new.inode))
3600                                 goto end_rename;
3601                 } else {
3602                         retval = -EMLINK;
3603                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3604                                 goto end_rename;
3605                 }
3606                 retval = ext4_rename_dir_prepare(handle, &old);
3607                 if (retval)
3608                         goto end_rename;
3609         }
3610         /*
3611          * If we're renaming a file within an inline_data dir and adding or
3612          * setting the new dirent causes a conversion from inline_data to
3613          * extents/blockmap, we need to force the dirent delete code to
3614          * re-read the directory, or else we end up trying to delete a dirent
3615          * from what is now the extent tree root (or a block map).
3616          */
3617         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3618                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3619
3620         old_file_type = old.de->file_type;
3621         if (whiteout) {
3622                 /*
3623                  * Do this before adding a new entry, so the old entry is sure
3624                  * to be still pointing to the valid old entry.
3625                  */
3626                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3627                                      EXT4_FT_CHRDEV);
3628                 if (retval)
3629                         goto end_rename;
3630                 ext4_mark_inode_dirty(handle, whiteout);
3631         }
3632         if (!new.bh) {
3633                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3634                 if (retval)
3635                         goto end_rename;
3636         } else {
3637                 retval = ext4_setent(handle, &new,
3638                                      old.inode->i_ino, old_file_type);
3639                 if (retval)
3640                         goto end_rename;
3641         }
3642         if (force_reread)
3643                 force_reread = !ext4_test_inode_flag(new.dir,
3644                                                      EXT4_INODE_INLINE_DATA);
3645
3646         /*
3647          * Like most other Unix systems, set the ctime for inodes on a
3648          * rename.
3649          */
3650         old.inode->i_ctime = current_time(old.inode);
3651         ext4_mark_inode_dirty(handle, old.inode);
3652
3653         if (!whiteout) {
3654                 /*
3655                  * ok, that's it
3656                  */
3657                 ext4_rename_delete(handle, &old, force_reread);
3658         }
3659
3660         if (new.inode) {
3661                 ext4_dec_count(handle, new.inode);
3662                 new.inode->i_ctime = current_time(new.inode);
3663         }
3664         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3665         ext4_update_dx_flag(old.dir);
3666         if (old.dir_bh) {
3667                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3668                 if (retval)
3669                         goto end_rename;
3670
3671                 ext4_dec_count(handle, old.dir);
3672                 if (new.inode) {
3673                         /* checked ext4_empty_dir above, can't have another
3674                          * parent, ext4_dec_count() won't work for many-linked
3675                          * dirs */
3676                         clear_nlink(new.inode);
3677                 } else {
3678                         ext4_inc_count(handle, new.dir);
3679                         ext4_update_dx_flag(new.dir);
3680                         ext4_mark_inode_dirty(handle, new.dir);
3681                 }
3682         }
3683         ext4_mark_inode_dirty(handle, old.dir);
3684         if (new.inode) {
3685                 ext4_mark_inode_dirty(handle, new.inode);
3686                 if (!new.inode->i_nlink)
3687                         ext4_orphan_add(handle, new.inode);
3688         }
3689         retval = 0;
3690
3691 end_rename:
3692         brelse(old.dir_bh);
3693         brelse(old.bh);
3694         brelse(new.bh);
3695         if (whiteout) {
3696                 if (retval)
3697                         drop_nlink(whiteout);
3698                 unlock_new_inode(whiteout);
3699                 iput(whiteout);
3700         }
3701         if (handle)
3702                 ext4_journal_stop(handle);
3703         return retval;
3704 }
3705
3706 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3707                              struct inode *new_dir, struct dentry *new_dentry)
3708 {
3709         handle_t *handle = NULL;
3710         struct ext4_renament old = {
3711                 .dir = old_dir,
3712                 .dentry = old_dentry,
3713                 .inode = d_inode(old_dentry),
3714         };
3715         struct ext4_renament new = {
3716                 .dir = new_dir,
3717                 .dentry = new_dentry,
3718                 .inode = d_inode(new_dentry),
3719         };
3720         u8 new_file_type;
3721         int retval;
3722         struct timespec ctime;
3723
3724         if ((ext4_encrypted_inode(old_dir) &&
3725              !fscrypt_has_encryption_key(old_dir)) ||
3726             (ext4_encrypted_inode(new_dir) &&
3727              !fscrypt_has_encryption_key(new_dir)))
3728                 return -ENOKEY;
3729
3730         if ((ext4_encrypted_inode(old_dir) ||
3731              ext4_encrypted_inode(new_dir)) &&
3732             (old_dir != new_dir) &&
3733             (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3734              !fscrypt_has_permitted_context(old_dir, new.inode)))
3735                 return -EPERM;
3736
3737         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3738              !projid_eq(EXT4_I(new_dir)->i_projid,
3739                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3740             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3741              !projid_eq(EXT4_I(old_dir)->i_projid,
3742                         EXT4_I(new_dentry->d_inode)->i_projid)))
3743                 return -EXDEV;
3744
3745         retval = dquot_initialize(old.dir);
3746         if (retval)
3747                 return retval;
3748         retval = dquot_initialize(new.dir);
3749         if (retval)
3750                 return retval;
3751
3752         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3753                                  &old.de, &old.inlined);
3754         if (IS_ERR(old.bh))
3755                 return PTR_ERR(old.bh);
3756         /*
3757          *  Check for inode number is _not_ due to possible IO errors.
3758          *  We might rmdir the source, keep it as pwd of some process
3759          *  and merrily kill the link to whatever was created under the
3760          *  same name. Goodbye sticky bit ;-<
3761          */
3762         retval = -ENOENT;
3763         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3764                 goto end_rename;
3765
3766         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3767                                  &new.de, &new.inlined);
3768         if (IS_ERR(new.bh)) {
3769                 retval = PTR_ERR(new.bh);
3770                 new.bh = NULL;
3771                 goto end_rename;
3772         }
3773
3774         /* RENAME_EXCHANGE case: old *and* new must both exist */
3775         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3776                 goto end_rename;
3777
3778         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3779                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3780                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3781         if (IS_ERR(handle)) {
3782                 retval = PTR_ERR(handle);
3783                 handle = NULL;
3784                 goto end_rename;
3785         }
3786
3787         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3788                 ext4_handle_sync(handle);
3789
3790         if (S_ISDIR(old.inode->i_mode)) {
3791                 old.is_dir = true;
3792                 retval = ext4_rename_dir_prepare(handle, &old);
3793                 if (retval)
3794                         goto end_rename;
3795         }
3796         if (S_ISDIR(new.inode->i_mode)) {
3797                 new.is_dir = true;
3798                 retval = ext4_rename_dir_prepare(handle, &new);
3799                 if (retval)
3800                         goto end_rename;
3801         }
3802
3803         /*
3804          * Other than the special case of overwriting a directory, parents'
3805          * nlink only needs to be modified if this is a cross directory rename.
3806          */
3807         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3808                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3809                 new.dir_nlink_delta = -old.dir_nlink_delta;
3810                 retval = -EMLINK;
3811                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3812                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3813                         goto end_rename;
3814         }
3815
3816         new_file_type = new.de->file_type;
3817         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3818         if (retval)
3819                 goto end_rename;
3820
3821         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3822         if (retval)
3823                 goto end_rename;
3824
3825         /*
3826          * Like most other Unix systems, set the ctime for inodes on a
3827          * rename.
3828          */
3829         ctime = current_time(old.inode);
3830         old.inode->i_ctime = ctime;
3831         new.inode->i_ctime = ctime;
3832         ext4_mark_inode_dirty(handle, old.inode);
3833         ext4_mark_inode_dirty(handle, new.inode);
3834
3835         if (old.dir_bh) {
3836                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3837                 if (retval)
3838                         goto end_rename;
3839         }
3840         if (new.dir_bh) {
3841                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3842                 if (retval)
3843                         goto end_rename;
3844         }
3845         ext4_update_dir_count(handle, &old);
3846         ext4_update_dir_count(handle, &new);
3847         retval = 0;
3848
3849 end_rename:
3850         brelse(old.dir_bh);
3851         brelse(new.dir_bh);
3852         brelse(old.bh);
3853         brelse(new.bh);
3854         if (handle)
3855                 ext4_journal_stop(handle);
3856         return retval;
3857 }
3858
3859 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3860                         struct inode *new_dir, struct dentry *new_dentry,
3861                         unsigned int flags)
3862 {
3863         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3864                 return -EIO;
3865
3866         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3867                 return -EINVAL;
3868
3869         if (flags & RENAME_EXCHANGE) {
3870                 return ext4_cross_rename(old_dir, old_dentry,
3871                                          new_dir, new_dentry);
3872         }
3873
3874         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3875 }
3876
3877 /*
3878  * directories can handle most operations...
3879  */
3880 const struct inode_operations ext4_dir_inode_operations = {
3881         .create         = ext4_create,
3882         .lookup         = ext4_lookup,
3883         .link           = ext4_link,
3884         .unlink         = ext4_unlink,
3885         .symlink        = ext4_symlink,
3886         .mkdir          = ext4_mkdir,
3887         .rmdir          = ext4_rmdir,
3888         .mknod          = ext4_mknod,
3889         .tmpfile        = ext4_tmpfile,
3890         .rename         = ext4_rename2,
3891         .setattr        = ext4_setattr,
3892         .getattr        = ext4_getattr,
3893         .listxattr      = ext4_listxattr,
3894         .get_acl        = ext4_get_acl,
3895         .set_acl        = ext4_set_acl,
3896         .fiemap         = ext4_fiemap,
3897 };
3898
3899 const struct inode_operations ext4_special_inode_operations = {
3900         .setattr        = ext4_setattr,
3901         .getattr        = ext4_getattr,
3902         .listxattr      = ext4_listxattr,
3903         .get_acl        = ext4_get_acl,
3904         .set_acl        = ext4_set_acl,
3905 };